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 i, low, high, remaining;
159 unsigned rover, best, best_size_so_far;
161 inet_get_local_port_range(&low, &high);
162 remaining = (high - low) + 1;
164 best_size_so_far = UINT_MAX;
165 best = rover = net_random() % remaining + low;
167 /* 1st pass: look for empty (or shortest) hash chain */
168 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
171 head = &udptable[udp_hashfn(net, rover)];
172 if (hlist_empty(head))
175 sk_for_each(sk2, node, head) {
176 if (++size >= best_size_so_far)
179 best_size_so_far = size;
182 /* fold back if end of range */
184 rover = low + ((rover - low)
185 & (UDP_HTABLE_SIZE - 1));
190 /* 2nd pass: find hole in shortest hash chain */
192 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
193 if (! __udp_lib_lport_inuse(net, rover, udptable))
195 rover += UDP_HTABLE_SIZE;
197 rover = low + ((rover - low)
198 & (UDP_HTABLE_SIZE - 1));
202 /* All ports in use! */
208 head = &udptable[udp_hashfn(net, snum)];
210 sk_for_each(sk2, node, head)
211 if (sk2->sk_hash == snum &&
213 net_eq(sock_net(sk2), net) &&
214 (!sk2->sk_reuse || !sk->sk_reuse) &&
215 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
216 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
217 (*saddr_comp)(sk, sk2) )
221 inet_sk(sk)->num = snum;
223 if (sk_unhashed(sk)) {
224 head = &udptable[udp_hashfn(net, snum)];
225 sk_add_node(sk, head);
226 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
230 write_unlock_bh(&udp_hash_lock);
234 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
236 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
238 return ( !ipv6_only_sock(sk2) &&
239 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
240 inet1->rcv_saddr == inet2->rcv_saddr ));
243 int udp_v4_get_port(struct sock *sk, unsigned short snum)
245 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
248 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
249 * harder than this. -DaveM
251 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
252 __be16 sport, __be32 daddr, __be16 dport,
253 int dif, struct hlist_head udptable[])
255 struct sock *sk, *result = NULL;
256 struct hlist_node *node;
257 unsigned short hnum = ntohs(dport);
260 read_lock(&udp_hash_lock);
261 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) {
262 struct inet_sock *inet = inet_sk(sk);
264 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
265 !ipv6_only_sock(sk)) {
266 int score = (sk->sk_family == PF_INET ? 1 : 0);
267 if (inet->rcv_saddr) {
268 if (inet->rcv_saddr != daddr)
273 if (inet->daddr != saddr)
278 if (inet->dport != sport)
282 if (sk->sk_bound_dev_if) {
283 if (sk->sk_bound_dev_if != dif)
290 } else if (score > badness) {
298 read_unlock(&udp_hash_lock);
302 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
303 __be16 sport, __be16 dport,
304 struct hlist_head udptable[])
307 const struct iphdr *iph = ip_hdr(skb);
309 if (unlikely(sk = skb_steal_sock(skb)))
312 return __udp4_lib_lookup(dev_net(skb->dst->dev), iph->saddr, sport,
313 iph->daddr, dport, inet_iif(skb),
317 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
318 __be32 daddr, __be16 dport, int dif)
320 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, udp_hash);
322 EXPORT_SYMBOL_GPL(udp4_lib_lookup);
324 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
325 __be16 loc_port, __be32 loc_addr,
326 __be16 rmt_port, __be32 rmt_addr,
329 struct hlist_node *node;
331 unsigned short hnum = ntohs(loc_port);
333 sk_for_each_from(s, node) {
334 struct inet_sock *inet = inet_sk(s);
336 if (s->sk_hash != hnum ||
337 (inet->daddr && inet->daddr != rmt_addr) ||
338 (inet->dport != rmt_port && inet->dport) ||
339 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
341 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
343 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
353 * This routine is called by the ICMP module when it gets some
354 * sort of error condition. If err < 0 then the socket should
355 * be closed and the error returned to the user. If err > 0
356 * it's just the icmp type << 8 | icmp code.
357 * Header points to the ip header of the error packet. We move
358 * on past this. Then (as it used to claim before adjustment)
359 * header points to the first 8 bytes of the udp header. We need
360 * to find the appropriate port.
363 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
365 struct inet_sock *inet;
366 struct iphdr *iph = (struct iphdr*)skb->data;
367 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
368 const int type = icmp_hdr(skb)->type;
369 const int code = icmp_hdr(skb)->code;
373 struct net *net = dev_net(skb->dev);
375 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
376 iph->saddr, uh->source, skb->dev->ifindex, udptable);
378 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
379 return; /* No socket for error */
388 case ICMP_TIME_EXCEEDED:
391 case ICMP_SOURCE_QUENCH:
393 case ICMP_PARAMETERPROB:
397 case ICMP_DEST_UNREACH:
398 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
399 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
407 if (code <= NR_ICMP_UNREACH) {
408 harderr = icmp_err_convert[code].fatal;
409 err = icmp_err_convert[code].errno;
415 * RFC1122: OK. Passes ICMP errors back to application, as per
418 if (!inet->recverr) {
419 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
422 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
425 sk->sk_error_report(sk);
430 void udp_err(struct sk_buff *skb, u32 info)
432 __udp4_lib_err(skb, info, udp_hash);
436 * Throw away all pending data and cancel the corking. Socket is locked.
438 void udp_flush_pending_frames(struct sock *sk)
440 struct udp_sock *up = udp_sk(sk);
445 ip_flush_pending_frames(sk);
448 EXPORT_SYMBOL(udp_flush_pending_frames);
451 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
452 * @sk: socket we are sending on
453 * @skb: sk_buff containing the filled-in UDP header
454 * (checksum field must be zeroed out)
456 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
457 __be32 src, __be32 dst, int len )
460 struct udphdr *uh = udp_hdr(skb);
463 if (skb_queue_len(&sk->sk_write_queue) == 1) {
465 * Only one fragment on the socket.
467 skb->csum_start = skb_transport_header(skb) - skb->head;
468 skb->csum_offset = offsetof(struct udphdr, check);
469 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
472 * HW-checksum won't work as there are two or more
473 * fragments on the socket so that all csums of sk_buffs
476 offset = skb_transport_offset(skb);
477 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
479 skb->ip_summed = CHECKSUM_NONE;
481 skb_queue_walk(&sk->sk_write_queue, skb) {
482 csum = csum_add(csum, skb->csum);
485 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
487 uh->check = CSUM_MANGLED_0;
492 * Push out all pending data as one UDP datagram. Socket is locked.
494 static int udp_push_pending_frames(struct sock *sk)
496 struct udp_sock *up = udp_sk(sk);
497 struct inet_sock *inet = inet_sk(sk);
498 struct flowi *fl = &inet->cork.fl;
502 int is_udplite = IS_UDPLITE(sk);
505 /* Grab the skbuff where UDP header space exists. */
506 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
510 * Create a UDP header
513 uh->source = fl->fl_ip_sport;
514 uh->dest = fl->fl_ip_dport;
515 uh->len = htons(up->len);
518 if (is_udplite) /* UDP-Lite */
519 csum = udplite_csum_outgoing(sk, skb);
521 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
523 skb->ip_summed = CHECKSUM_NONE;
526 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
528 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
531 } else /* `normal' UDP */
532 csum = udp_csum_outgoing(sk, skb);
534 /* add protocol-dependent pseudo-header */
535 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
536 sk->sk_protocol, csum );
538 uh->check = CSUM_MANGLED_0;
541 err = ip_push_pending_frames(sk);
546 UDP_INC_STATS_USER(sock_net(sk),
547 UDP_MIB_OUTDATAGRAMS, is_udplite);
551 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
554 struct inet_sock *inet = inet_sk(sk);
555 struct udp_sock *up = udp_sk(sk);
557 struct ipcm_cookie ipc;
558 struct rtable *rt = NULL;
561 __be32 daddr, faddr, saddr;
564 int err, is_udplite = IS_UDPLITE(sk);
565 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
566 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
575 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
582 * There are pending frames.
583 * The socket lock must be held while it's corked.
586 if (likely(up->pending)) {
587 if (unlikely(up->pending != AF_INET)) {
595 ulen += sizeof(struct udphdr);
598 * Get and verify the address.
601 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
602 if (msg->msg_namelen < sizeof(*usin))
604 if (usin->sin_family != AF_INET) {
605 if (usin->sin_family != AF_UNSPEC)
606 return -EAFNOSUPPORT;
609 daddr = usin->sin_addr.s_addr;
610 dport = usin->sin_port;
614 if (sk->sk_state != TCP_ESTABLISHED)
615 return -EDESTADDRREQ;
618 /* Open fast path for connected socket.
619 Route will not be used, if at least one option is set.
623 ipc.addr = inet->saddr;
625 ipc.oif = sk->sk_bound_dev_if;
626 if (msg->msg_controllen) {
627 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
638 ipc.addr = faddr = daddr;
640 if (ipc.opt && ipc.opt->srr) {
643 faddr = ipc.opt->faddr;
646 tos = RT_TOS(inet->tos);
647 if (sock_flag(sk, SOCK_LOCALROUTE) ||
648 (msg->msg_flags & MSG_DONTROUTE) ||
649 (ipc.opt && ipc.opt->is_strictroute)) {
654 if (ipv4_is_multicast(daddr)) {
656 ipc.oif = inet->mc_index;
658 saddr = inet->mc_addr;
663 rt = (struct rtable*)sk_dst_check(sk, 0);
666 struct flowi fl = { .oif = ipc.oif,
671 .proto = sk->sk_protocol,
673 { .sport = inet->sport,
674 .dport = dport } } };
675 struct net *net = sock_net(sk);
677 security_sk_classify_flow(sk, &fl);
678 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
680 if (err == -ENETUNREACH)
681 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
686 if ((rt->rt_flags & RTCF_BROADCAST) &&
687 !sock_flag(sk, SOCK_BROADCAST))
690 sk_dst_set(sk, dst_clone(&rt->u.dst));
693 if (msg->msg_flags&MSG_CONFIRM)
699 daddr = ipc.addr = rt->rt_dst;
702 if (unlikely(up->pending)) {
703 /* The socket is already corked while preparing it. */
704 /* ... which is an evident application bug. --ANK */
707 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
712 * Now cork the socket to pend data.
714 inet->cork.fl.fl4_dst = daddr;
715 inet->cork.fl.fl_ip_dport = dport;
716 inet->cork.fl.fl4_src = saddr;
717 inet->cork.fl.fl_ip_sport = inet->sport;
718 up->pending = AF_INET;
722 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
723 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
724 sizeof(struct udphdr), &ipc, rt,
725 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
727 udp_flush_pending_frames(sk);
729 err = udp_push_pending_frames(sk);
730 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
741 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
742 * ENOBUFS might not be good (it's not tunable per se), but otherwise
743 * we don't have a good statistic (IpOutDiscards but it can be too many
744 * things). We could add another new stat but at least for now that
745 * seems like overkill.
747 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
748 UDP_INC_STATS_USER(sock_net(sk),
749 UDP_MIB_SNDBUFERRORS, is_udplite);
754 dst_confirm(&rt->u.dst);
755 if (!(msg->msg_flags&MSG_PROBE) || len)
756 goto back_from_confirm;
761 int udp_sendpage(struct sock *sk, struct page *page, int offset,
762 size_t size, int flags)
764 struct udp_sock *up = udp_sk(sk);
768 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
770 /* Call udp_sendmsg to specify destination address which
771 * sendpage interface can't pass.
772 * This will succeed only when the socket is connected.
774 ret = udp_sendmsg(NULL, sk, &msg, 0);
781 if (unlikely(!up->pending)) {
784 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
788 ret = ip_append_page(sk, page, offset, size, flags);
789 if (ret == -EOPNOTSUPP) {
791 return sock_no_sendpage(sk->sk_socket, page, offset,
795 udp_flush_pending_frames(sk);
800 if (!(up->corkflag || (flags&MSG_MORE)))
801 ret = udp_push_pending_frames(sk);
810 * IOCTL requests applicable to the UDP protocol
813 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
818 int amount = atomic_read(&sk->sk_wmem_alloc);
819 return put_user(amount, (int __user *)arg);
825 unsigned long amount;
828 spin_lock_bh(&sk->sk_receive_queue.lock);
829 skb = skb_peek(&sk->sk_receive_queue);
832 * We will only return the amount
833 * of this packet since that is all
836 amount = skb->len - sizeof(struct udphdr);
838 spin_unlock_bh(&sk->sk_receive_queue.lock);
839 return put_user(amount, (int __user *)arg);
850 * This should be easy, if there is something there we
851 * return it, otherwise we block.
854 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
855 size_t len, int noblock, int flags, int *addr_len)
857 struct inet_sock *inet = inet_sk(sk);
858 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
860 unsigned int ulen, copied;
863 int is_udplite = IS_UDPLITE(sk);
866 * Check any passed addresses
869 *addr_len=sizeof(*sin);
871 if (flags & MSG_ERRQUEUE)
872 return ip_recv_error(sk, msg, len);
875 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
880 ulen = skb->len - sizeof(struct udphdr);
884 else if (copied < ulen)
885 msg->msg_flags |= MSG_TRUNC;
888 * If checksum is needed at all, try to do it while copying the
889 * data. If the data is truncated, or if we only want a partial
890 * coverage checksum (UDP-Lite), do it before the copy.
893 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
894 if (udp_lib_checksum_complete(skb))
898 if (skb_csum_unnecessary(skb))
899 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
900 msg->msg_iov, copied );
902 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
912 UDP_INC_STATS_USER(sock_net(sk),
913 UDP_MIB_INDATAGRAMS, is_udplite);
915 sock_recv_timestamp(msg, sk, skb);
917 /* Copy the address. */
920 sin->sin_family = AF_INET;
921 sin->sin_port = udp_hdr(skb)->source;
922 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
923 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
925 if (inet->cmsg_flags)
926 ip_cmsg_recv(msg, skb);
929 if (flags & MSG_TRUNC)
934 skb_free_datagram(sk, skb);
941 if (!skb_kill_datagram(sk, skb, flags))
942 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
951 int udp_disconnect(struct sock *sk, int flags)
953 struct inet_sock *inet = inet_sk(sk);
955 * 1003.1g - break association.
958 sk->sk_state = TCP_CLOSE;
961 sk->sk_bound_dev_if = 0;
962 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
963 inet_reset_saddr(sk);
965 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
966 sk->sk_prot->unhash(sk);
973 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
975 int is_udplite = IS_UDPLITE(sk);
978 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
979 /* Note that an ENOMEM error is charged twice */
981 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
989 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
997 * >0: "udp encap" protocol resubmission
999 * Note that in the success and error cases, the skb is assumed to
1000 * have either been requeued or freed.
1002 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1004 struct udp_sock *up = udp_sk(sk);
1006 int is_udplite = IS_UDPLITE(sk);
1009 * Charge it to the socket, dropping if the queue is full.
1011 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1015 if (up->encap_type) {
1017 * This is an encapsulation socket so pass the skb to
1018 * the socket's udp_encap_rcv() hook. Otherwise, just
1019 * fall through and pass this up the UDP socket.
1020 * up->encap_rcv() returns the following value:
1021 * =0 if skb was successfully passed to the encap
1022 * handler or was discarded by it.
1023 * >0 if skb should be passed on to UDP.
1024 * <0 if skb should be resubmitted as proto -N
1027 /* if we're overly short, let UDP handle it */
1028 if (skb->len > sizeof(struct udphdr) &&
1029 up->encap_rcv != NULL) {
1032 ret = (*up->encap_rcv)(sk, skb);
1034 UDP_INC_STATS_BH(sock_net(sk),
1035 UDP_MIB_INDATAGRAMS,
1041 /* FALLTHROUGH -- it's a UDP Packet */
1045 * UDP-Lite specific tests, ignored on UDP sockets
1047 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1050 * MIB statistics other than incrementing the error count are
1051 * disabled for the following two types of errors: these depend
1052 * on the application settings, not on the functioning of the
1053 * protocol stack as such.
1055 * RFC 3828 here recommends (sec 3.3): "There should also be a
1056 * way ... to ... at least let the receiving application block
1057 * delivery of packets with coverage values less than a value
1058 * provided by the application."
1060 if (up->pcrlen == 0) { /* full coverage was set */
1061 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1062 "%d while full coverage %d requested\n",
1063 UDP_SKB_CB(skb)->cscov, skb->len);
1066 /* The next case involves violating the min. coverage requested
1067 * by the receiver. This is subtle: if receiver wants x and x is
1068 * greater than the buffersize/MTU then receiver will complain
1069 * that it wants x while sender emits packets of smaller size y.
1070 * Therefore the above ...()->partial_cov statement is essential.
1072 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1073 LIMIT_NETDEBUG(KERN_WARNING
1074 "UDPLITE: coverage %d too small, need min %d\n",
1075 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1080 if (sk->sk_filter) {
1081 if (udp_lib_checksum_complete(skb))
1088 if (!sock_owned_by_user(sk))
1089 rc = __udp_queue_rcv_skb(sk, skb);
1091 sk_add_backlog(sk, skb);
1097 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1103 * Multicasts and broadcasts go to each listener.
1105 * Note: called only from the BH handler context,
1106 * so we don't need to lock the hashes.
1108 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1110 __be32 saddr, __be32 daddr,
1111 struct hlist_head udptable[])
1116 read_lock(&udp_hash_lock);
1117 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
1118 dif = skb->dev->ifindex;
1119 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1121 struct sock *sknext = NULL;
1124 struct sk_buff *skb1 = skb;
1126 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1127 uh->source, saddr, dif);
1129 skb1 = skb_clone(skb, GFP_ATOMIC);
1132 int ret = udp_queue_rcv_skb(sk, skb1);
1134 /* we should probably re-process instead
1135 * of dropping packets here. */
1142 read_unlock(&udp_hash_lock);
1146 /* Initialize UDP checksum. If exited with zero value (success),
1147 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1148 * Otherwise, csum completion requires chacksumming packet body,
1149 * including udp header and folding it to skb->csum.
1151 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1154 const struct iphdr *iph;
1157 UDP_SKB_CB(skb)->partial_cov = 0;
1158 UDP_SKB_CB(skb)->cscov = skb->len;
1160 if (proto == IPPROTO_UDPLITE) {
1161 err = udplite_checksum_init(skb, uh);
1167 if (uh->check == 0) {
1168 skb->ip_summed = CHECKSUM_UNNECESSARY;
1169 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1170 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1172 skb->ip_summed = CHECKSUM_UNNECESSARY;
1174 if (!skb_csum_unnecessary(skb))
1175 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1176 skb->len, proto, 0);
1177 /* Probably, we should checksum udp header (it should be in cache
1178 * in any case) and data in tiny packets (< rx copybreak).
1185 * All we need to do is get the socket, and then do a checksum.
1188 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1192 struct udphdr *uh = udp_hdr(skb);
1193 unsigned short ulen;
1194 struct rtable *rt = (struct rtable*)skb->dst;
1195 __be32 saddr = ip_hdr(skb)->saddr;
1196 __be32 daddr = ip_hdr(skb)->daddr;
1197 struct net *net = dev_net(skb->dev);
1200 * Validate the packet.
1202 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1203 goto drop; /* No space for header. */
1205 ulen = ntohs(uh->len);
1206 if (ulen > skb->len)
1209 if (proto == IPPROTO_UDP) {
1210 /* UDP validates ulen. */
1211 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1216 if (udp4_csum_init(skb, uh, proto))
1219 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1220 return __udp4_lib_mcast_deliver(net, skb, uh,
1221 saddr, daddr, udptable);
1223 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1226 int ret = udp_queue_rcv_skb(sk, skb);
1229 /* a return value > 0 means to resubmit the input, but
1230 * it wants the return to be -protocol, or 0
1237 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1241 /* No socket. Drop packet silently, if checksum is wrong */
1242 if (udp_lib_checksum_complete(skb))
1245 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1246 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1249 * Hmm. We got an UDP packet to a port to which we
1250 * don't wanna listen. Ignore it.
1256 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1257 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1268 * RFC1122: OK. Discards the bad packet silently (as far as
1269 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1271 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1272 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1279 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1284 int udp_rcv(struct sk_buff *skb)
1286 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1289 void udp_destroy_sock(struct sock *sk)
1292 udp_flush_pending_frames(sk);
1297 * Socket option code for UDP
1299 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1300 char __user *optval, int optlen,
1301 int (*push_pending_frames)(struct sock *))
1303 struct udp_sock *up = udp_sk(sk);
1306 int is_udplite = IS_UDPLITE(sk);
1308 if (optlen<sizeof(int))
1311 if (get_user(val, (int __user *)optval))
1321 (*push_pending_frames)(sk);
1329 case UDP_ENCAP_ESPINUDP:
1330 case UDP_ENCAP_ESPINUDP_NON_IKE:
1331 up->encap_rcv = xfrm4_udp_encap_rcv;
1333 case UDP_ENCAP_L2TPINUDP:
1334 up->encap_type = val;
1343 * UDP-Lite's partial checksum coverage (RFC 3828).
1345 /* The sender sets actual checksum coverage length via this option.
1346 * The case coverage > packet length is handled by send module. */
1347 case UDPLITE_SEND_CSCOV:
1348 if (!is_udplite) /* Disable the option on UDP sockets */
1349 return -ENOPROTOOPT;
1350 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1352 else if (val > USHORT_MAX)
1355 up->pcflag |= UDPLITE_SEND_CC;
1358 /* The receiver specifies a minimum checksum coverage value. To make
1359 * sense, this should be set to at least 8 (as done below). If zero is
1360 * used, this again means full checksum coverage. */
1361 case UDPLITE_RECV_CSCOV:
1362 if (!is_udplite) /* Disable the option on UDP sockets */
1363 return -ENOPROTOOPT;
1364 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1366 else if (val > USHORT_MAX)
1369 up->pcflag |= UDPLITE_RECV_CC;
1380 int udp_setsockopt(struct sock *sk, int level, int optname,
1381 char __user *optval, int optlen)
1383 if (level == SOL_UDP || level == SOL_UDPLITE)
1384 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1385 udp_push_pending_frames);
1386 return ip_setsockopt(sk, level, optname, optval, optlen);
1389 #ifdef CONFIG_COMPAT
1390 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1391 char __user *optval, int optlen)
1393 if (level == SOL_UDP || level == SOL_UDPLITE)
1394 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1395 udp_push_pending_frames);
1396 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1400 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1401 char __user *optval, int __user *optlen)
1403 struct udp_sock *up = udp_sk(sk);
1406 if (get_user(len,optlen))
1409 len = min_t(unsigned int, len, sizeof(int));
1420 val = up->encap_type;
1423 /* The following two cannot be changed on UDP sockets, the return is
1424 * always 0 (which corresponds to the full checksum coverage of UDP). */
1425 case UDPLITE_SEND_CSCOV:
1429 case UDPLITE_RECV_CSCOV:
1434 return -ENOPROTOOPT;
1437 if (put_user(len, optlen))
1439 if (copy_to_user(optval, &val,len))
1444 int udp_getsockopt(struct sock *sk, int level, int optname,
1445 char __user *optval, int __user *optlen)
1447 if (level == SOL_UDP || level == SOL_UDPLITE)
1448 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1449 return ip_getsockopt(sk, level, optname, optval, optlen);
1452 #ifdef CONFIG_COMPAT
1453 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1454 char __user *optval, int __user *optlen)
1456 if (level == SOL_UDP || level == SOL_UDPLITE)
1457 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1458 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1462 * udp_poll - wait for a UDP event.
1463 * @file - file struct
1465 * @wait - poll table
1467 * This is same as datagram poll, except for the special case of
1468 * blocking sockets. If application is using a blocking fd
1469 * and a packet with checksum error is in the queue;
1470 * then it could get return from select indicating data available
1471 * but then block when reading it. Add special case code
1472 * to work around these arguably broken applications.
1474 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1476 unsigned int mask = datagram_poll(file, sock, wait);
1477 struct sock *sk = sock->sk;
1478 int is_lite = IS_UDPLITE(sk);
1480 /* Check for false positives due to checksum errors */
1481 if ( (mask & POLLRDNORM) &&
1482 !(file->f_flags & O_NONBLOCK) &&
1483 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1484 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1485 struct sk_buff *skb;
1487 spin_lock_bh(&rcvq->lock);
1488 while ((skb = skb_peek(rcvq)) != NULL &&
1489 udp_lib_checksum_complete(skb)) {
1490 UDP_INC_STATS_BH(sock_net(sk),
1491 UDP_MIB_INERRORS, is_lite);
1492 __skb_unlink(skb, rcvq);
1495 spin_unlock_bh(&rcvq->lock);
1497 /* nothing to see, move along */
1499 mask &= ~(POLLIN | POLLRDNORM);
1506 struct proto udp_prot = {
1508 .owner = THIS_MODULE,
1509 .close = udp_lib_close,
1510 .connect = ip4_datagram_connect,
1511 .disconnect = udp_disconnect,
1513 .destroy = udp_destroy_sock,
1514 .setsockopt = udp_setsockopt,
1515 .getsockopt = udp_getsockopt,
1516 .sendmsg = udp_sendmsg,
1517 .recvmsg = udp_recvmsg,
1518 .sendpage = udp_sendpage,
1519 .backlog_rcv = __udp_queue_rcv_skb,
1520 .hash = udp_lib_hash,
1521 .unhash = udp_lib_unhash,
1522 .get_port = udp_v4_get_port,
1523 .memory_allocated = &udp_memory_allocated,
1524 .sysctl_mem = sysctl_udp_mem,
1525 .sysctl_wmem = &sysctl_udp_wmem_min,
1526 .sysctl_rmem = &sysctl_udp_rmem_min,
1527 .obj_size = sizeof(struct udp_sock),
1528 .h.udp_hash = udp_hash,
1529 #ifdef CONFIG_COMPAT
1530 .compat_setsockopt = compat_udp_setsockopt,
1531 .compat_getsockopt = compat_udp_getsockopt,
1535 /* ------------------------------------------------------------------------ */
1536 #ifdef CONFIG_PROC_FS
1538 static struct sock *udp_get_first(struct seq_file *seq)
1541 struct udp_iter_state *state = seq->private;
1542 struct net *net = seq_file_net(seq);
1544 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1545 struct hlist_node *node;
1546 sk_for_each(sk, node, state->hashtable + state->bucket) {
1547 if (!net_eq(sock_net(sk), net))
1549 if (sk->sk_family == state->family)
1558 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1560 struct udp_iter_state *state = seq->private;
1561 struct net *net = seq_file_net(seq);
1567 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1569 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1570 sk = sk_head(state->hashtable + state->bucket);
1576 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1578 struct sock *sk = udp_get_first(seq);
1581 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1583 return pos ? NULL : sk;
1586 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1587 __acquires(udp_hash_lock)
1589 read_lock(&udp_hash_lock);
1590 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1593 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1597 if (v == SEQ_START_TOKEN)
1598 sk = udp_get_idx(seq, 0);
1600 sk = udp_get_next(seq, v);
1606 static void udp_seq_stop(struct seq_file *seq, void *v)
1607 __releases(udp_hash_lock)
1609 read_unlock(&udp_hash_lock);
1612 static int udp_seq_open(struct inode *inode, struct file *file)
1614 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1615 struct udp_iter_state *s;
1618 err = seq_open_net(inode, file, &afinfo->seq_ops,
1619 sizeof(struct udp_iter_state));
1623 s = ((struct seq_file *)file->private_data)->private;
1624 s->family = afinfo->family;
1625 s->hashtable = afinfo->hashtable;
1629 /* ------------------------------------------------------------------------ */
1630 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1632 struct proc_dir_entry *p;
1635 afinfo->seq_fops.open = udp_seq_open;
1636 afinfo->seq_fops.read = seq_read;
1637 afinfo->seq_fops.llseek = seq_lseek;
1638 afinfo->seq_fops.release = seq_release_net;
1640 afinfo->seq_ops.start = udp_seq_start;
1641 afinfo->seq_ops.next = udp_seq_next;
1642 afinfo->seq_ops.stop = udp_seq_stop;
1644 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1645 &afinfo->seq_fops, afinfo);
1651 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1653 proc_net_remove(net, afinfo->name);
1656 /* ------------------------------------------------------------------------ */
1657 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1658 int bucket, int *len)
1660 struct inet_sock *inet = inet_sk(sp);
1661 __be32 dest = inet->daddr;
1662 __be32 src = inet->rcv_saddr;
1663 __u16 destp = ntohs(inet->dport);
1664 __u16 srcp = ntohs(inet->sport);
1666 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1667 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1668 bucket, src, srcp, dest, destp, sp->sk_state,
1669 atomic_read(&sp->sk_wmem_alloc),
1670 atomic_read(&sp->sk_rmem_alloc),
1671 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1672 atomic_read(&sp->sk_refcnt), sp,
1673 atomic_read(&sp->sk_drops), len);
1676 int udp4_seq_show(struct seq_file *seq, void *v)
1678 if (v == SEQ_START_TOKEN)
1679 seq_printf(seq, "%-127s\n",
1680 " sl local_address rem_address st tx_queue "
1681 "rx_queue tr tm->when retrnsmt uid timeout "
1682 "inode ref pointer drops");
1684 struct udp_iter_state *state = seq->private;
1687 udp4_format_sock(v, seq, state->bucket, &len);
1688 seq_printf(seq, "%*s\n", 127 - len ,"");
1693 /* ------------------------------------------------------------------------ */
1694 static struct udp_seq_afinfo udp4_seq_afinfo = {
1697 .hashtable = udp_hash,
1699 .owner = THIS_MODULE,
1702 .show = udp4_seq_show,
1706 static int udp4_proc_init_net(struct net *net)
1708 return udp_proc_register(net, &udp4_seq_afinfo);
1711 static void udp4_proc_exit_net(struct net *net)
1713 udp_proc_unregister(net, &udp4_seq_afinfo);
1716 static struct pernet_operations udp4_net_ops = {
1717 .init = udp4_proc_init_net,
1718 .exit = udp4_proc_exit_net,
1721 int __init udp4_proc_init(void)
1723 return register_pernet_subsys(&udp4_net_ops);
1726 void udp4_proc_exit(void)
1728 unregister_pernet_subsys(&udp4_net_ops);
1730 #endif /* CONFIG_PROC_FS */
1732 void __init udp_init(void)
1734 unsigned long limit;
1736 /* Set the pressure threshold up by the same strategy of TCP. It is a
1737 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1738 * toward zero with the amount of memory, with a floor of 128 pages.
1740 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1741 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1742 limit = max(limit, 128UL);
1743 sysctl_udp_mem[0] = limit / 4 * 3;
1744 sysctl_udp_mem[1] = limit;
1745 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1747 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1748 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1751 EXPORT_SYMBOL(udp_disconnect);
1752 EXPORT_SYMBOL(udp_hash);
1753 EXPORT_SYMBOL(udp_hash_lock);
1754 EXPORT_SYMBOL(udp_ioctl);
1755 EXPORT_SYMBOL(udp_prot);
1756 EXPORT_SYMBOL(udp_sendmsg);
1757 EXPORT_SYMBOL(udp_lib_getsockopt);
1758 EXPORT_SYMBOL(udp_lib_setsockopt);
1759 EXPORT_SYMBOL(udp_poll);
1760 EXPORT_SYMBOL(udp_lib_get_port);
1762 #ifdef CONFIG_PROC_FS
1763 EXPORT_SYMBOL(udp_proc_register);
1764 EXPORT_SYMBOL(udp_proc_unregister);