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).
8 * Version: $Id: udp.c,v 1.102 2002/02/01 22:01:04 davem Exp $
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
13 * Alan Cox, <Alan.Cox@linux.org>
14 * Hirokazu Takahashi, <taka@valinux.co.jp>
17 * Alan Cox : verify_area() calls
18 * Alan Cox : stopped close while in use off icmp
19 * messages. Not a fix but a botch that
20 * for udp at least is 'valid'.
21 * Alan Cox : Fixed icmp handling properly
22 * Alan Cox : Correct error for oversized datagrams
23 * Alan Cox : Tidied select() semantics.
24 * Alan Cox : udp_err() fixed properly, also now
25 * select and read wake correctly on errors
26 * Alan Cox : udp_send verify_area moved to avoid mem leak
27 * Alan Cox : UDP can count its memory
28 * Alan Cox : send to an unknown connection causes
29 * an ECONNREFUSED off the icmp, but
31 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
32 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
33 * bug no longer crashes it.
34 * Fred Van Kempen : Net2e support for sk->broadcast.
35 * Alan Cox : Uses skb_free_datagram
36 * Alan Cox : Added get/set sockopt support.
37 * Alan Cox : Broadcasting without option set returns EACCES.
38 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
39 * Alan Cox : Use ip_tos and ip_ttl
40 * Alan Cox : SNMP Mibs
41 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
42 * Matt Dillon : UDP length checks.
43 * Alan Cox : Smarter af_inet used properly.
44 * Alan Cox : Use new kernel side addressing.
45 * Alan Cox : Incorrect return on truncated datagram receive.
46 * Arnt Gulbrandsen : New udp_send and stuff
47 * Alan Cox : Cache last socket
48 * Alan Cox : Route cache
49 * Jon Peatfield : Minor efficiency fix to sendto().
50 * Mike Shaver : RFC1122 checks.
51 * Alan Cox : Nonblocking error fix.
52 * Willy Konynenberg : Transparent proxying support.
53 * Mike McLagan : Routing by source
54 * David S. Miller : New socket lookup architecture.
55 * Last socket cache retained as it
56 * does have a high hit rate.
57 * Olaf Kirch : Don't linearise iovec on sendmsg.
58 * Andi Kleen : Some cleanups, cache destination entry
60 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
61 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
62 * return ENOTCONN for unconnected sockets (POSIX)
63 * Janos Farkas : don't deliver multi/broadcasts to a different
64 * bound-to-device socket
65 * Hirokazu Takahashi : HW checksumming for outgoing UDP
67 * Hirokazu Takahashi : sendfile() on UDP works now.
68 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
69 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
70 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
71 * a single port at the same time.
72 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
73 * James Chapman : Add L2TP encapsulation type.
76 * This program is free software; you can redistribute it and/or
77 * modify it under the terms of the GNU General Public License
78 * as published by the Free Software Foundation; either version
79 * 2 of the License, or (at your option) any later version.
82 #include <asm/system.h>
83 #include <asm/uaccess.h>
84 #include <asm/ioctls.h>
85 #include <linux/types.h>
86 #include <linux/fcntl.h>
87 #include <linux/module.h>
88 #include <linux/socket.h>
89 #include <linux/sockios.h>
90 #include <linux/igmp.h>
92 #include <linux/errno.h>
93 #include <linux/timer.h>
95 #include <linux/inet.h>
96 #include <linux/netdevice.h>
97 #include <net/tcp_states.h>
98 #include <linux/skbuff.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <net/net_namespace.h>
102 #include <net/icmp.h>
103 #include <net/route.h>
104 #include <net/checksum.h>
105 #include <net/xfrm.h>
106 #include "udp_impl.h"
109 * Snmp MIB for the UDP layer
112 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
114 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
115 DEFINE_RWLOCK(udp_hash_lock);
117 static inline int __udp_lib_lport_inuse(__u16 num,
118 const struct hlist_head udptable[])
121 struct hlist_node *node;
123 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
124 if (sk->sk_hash == num)
130 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
132 * @sk: socket struct in question
133 * @snum: port number to look up
134 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
135 * @saddr_comp: AF-dependent comparison of bound local IP addresses
137 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
138 struct hlist_head udptable[],
139 int (*saddr_comp)(const struct sock *sk1,
140 const struct sock *sk2 ) )
142 struct hlist_node *node;
143 struct hlist_head *head;
147 write_lock_bh(&udp_hash_lock);
150 int i, low, high, remaining;
151 unsigned rover, best, best_size_so_far;
153 inet_get_local_port_range(&low, &high);
154 remaining = (high - low) + 1;
156 best_size_so_far = UINT_MAX;
157 best = rover = net_random() % remaining + low;
159 /* 1st pass: look for empty (or shortest) hash chain */
160 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
163 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)];
164 if (hlist_empty(head))
167 sk_for_each(sk2, node, head) {
168 if (++size >= best_size_so_far)
171 best_size_so_far = size;
174 /* fold back if end of range */
176 rover = low + ((rover - low)
177 & (UDP_HTABLE_SIZE - 1));
182 /* 2nd pass: find hole in shortest hash chain */
184 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
185 if (! __udp_lib_lport_inuse(rover, udptable))
187 rover += UDP_HTABLE_SIZE;
189 rover = low + ((rover - low)
190 & (UDP_HTABLE_SIZE - 1));
194 /* All ports in use! */
200 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
202 sk_for_each(sk2, node, head)
203 if (sk2->sk_hash == snum &&
205 (!sk2->sk_reuse || !sk->sk_reuse) &&
206 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
207 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
208 (*saddr_comp)(sk, sk2) )
212 inet_sk(sk)->num = snum;
214 if (sk_unhashed(sk)) {
215 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
216 sk_add_node(sk, head);
217 sock_prot_inc_use(sk->sk_prot);
221 write_unlock_bh(&udp_hash_lock);
225 int udp_get_port(struct sock *sk, unsigned short snum,
226 int (*scmp)(const struct sock *, const struct sock *))
228 return __udp_lib_get_port(sk, snum, udp_hash, scmp);
231 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
233 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
235 return ( !ipv6_only_sock(sk2) &&
236 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
237 inet1->rcv_saddr == inet2->rcv_saddr ));
240 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
242 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
245 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
246 * harder than this. -DaveM
248 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
249 __be32 daddr, __be16 dport,
250 int dif, struct hlist_head udptable[])
252 struct sock *sk, *result = NULL;
253 struct hlist_node *node;
254 unsigned short hnum = ntohs(dport);
257 read_lock(&udp_hash_lock);
258 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
259 struct inet_sock *inet = inet_sk(sk);
261 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
262 int score = (sk->sk_family == PF_INET ? 1 : 0);
263 if (inet->rcv_saddr) {
264 if (inet->rcv_saddr != daddr)
269 if (inet->daddr != saddr)
274 if (inet->dport != sport)
278 if (sk->sk_bound_dev_if) {
279 if (sk->sk_bound_dev_if != dif)
286 } else if (score > badness) {
294 read_unlock(&udp_hash_lock);
298 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
299 __be16 loc_port, __be32 loc_addr,
300 __be16 rmt_port, __be32 rmt_addr,
303 struct hlist_node *node;
305 unsigned short hnum = ntohs(loc_port);
307 sk_for_each_from(s, node) {
308 struct inet_sock *inet = inet_sk(s);
310 if (s->sk_hash != hnum ||
311 (inet->daddr && inet->daddr != rmt_addr) ||
312 (inet->dport != rmt_port && inet->dport) ||
313 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
315 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
317 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
327 * This routine is called by the ICMP module when it gets some
328 * sort of error condition. If err < 0 then the socket should
329 * be closed and the error returned to the user. If err > 0
330 * it's just the icmp type << 8 | icmp code.
331 * Header points to the ip header of the error packet. We move
332 * on past this. Then (as it used to claim before adjustment)
333 * header points to the first 8 bytes of the udp header. We need
334 * to find the appropriate port.
337 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
339 struct inet_sock *inet;
340 struct iphdr *iph = (struct iphdr*)skb->data;
341 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
342 const int type = icmp_hdr(skb)->type;
343 const int code = icmp_hdr(skb)->code;
348 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
349 skb->dev->ifindex, udptable );
351 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
352 return; /* No socket for error */
361 case ICMP_TIME_EXCEEDED:
364 case ICMP_SOURCE_QUENCH:
366 case ICMP_PARAMETERPROB:
370 case ICMP_DEST_UNREACH:
371 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
372 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
380 if (code <= NR_ICMP_UNREACH) {
381 harderr = icmp_err_convert[code].fatal;
382 err = icmp_err_convert[code].errno;
388 * RFC1122: OK. Passes ICMP errors back to application, as per
391 if (!inet->recverr) {
392 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
395 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
398 sk->sk_error_report(sk);
403 void udp_err(struct sk_buff *skb, u32 info)
405 return __udp4_lib_err(skb, info, udp_hash);
409 * Throw away all pending data and cancel the corking. Socket is locked.
411 static void udp_flush_pending_frames(struct sock *sk)
413 struct udp_sock *up = udp_sk(sk);
418 ip_flush_pending_frames(sk);
423 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
424 * @sk: socket we are sending on
425 * @skb: sk_buff containing the filled-in UDP header
426 * (checksum field must be zeroed out)
428 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
429 __be32 src, __be32 dst, int len )
432 struct udphdr *uh = udp_hdr(skb);
435 if (skb_queue_len(&sk->sk_write_queue) == 1) {
437 * Only one fragment on the socket.
439 skb->csum_start = skb_transport_header(skb) - skb->head;
440 skb->csum_offset = offsetof(struct udphdr, check);
441 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
444 * HW-checksum won't work as there are two or more
445 * fragments on the socket so that all csums of sk_buffs
448 offset = skb_transport_offset(skb);
449 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
451 skb->ip_summed = CHECKSUM_NONE;
453 skb_queue_walk(&sk->sk_write_queue, skb) {
454 csum = csum_add(csum, skb->csum);
457 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
459 uh->check = CSUM_MANGLED_0;
464 * Push out all pending data as one UDP datagram. Socket is locked.
466 static int udp_push_pending_frames(struct sock *sk)
468 struct udp_sock *up = udp_sk(sk);
469 struct inet_sock *inet = inet_sk(sk);
470 struct flowi *fl = &inet->cork.fl;
476 /* Grab the skbuff where UDP header space exists. */
477 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
481 * Create a UDP header
484 uh->source = fl->fl_ip_sport;
485 uh->dest = fl->fl_ip_dport;
486 uh->len = htons(up->len);
489 if (up->pcflag) /* UDP-Lite */
490 csum = udplite_csum_outgoing(sk, skb);
492 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
494 skb->ip_summed = CHECKSUM_NONE;
497 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
499 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
502 } else /* `normal' UDP */
503 csum = udp_csum_outgoing(sk, skb);
505 /* add protocol-dependent pseudo-header */
506 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
507 sk->sk_protocol, csum );
509 uh->check = CSUM_MANGLED_0;
512 err = ip_push_pending_frames(sk);
517 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, up->pcflag);
521 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
524 struct inet_sock *inet = inet_sk(sk);
525 struct udp_sock *up = udp_sk(sk);
527 struct ipcm_cookie ipc;
528 struct rtable *rt = NULL;
531 __be32 daddr, faddr, saddr;
534 int err, is_udplite = up->pcflag;
535 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
536 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
545 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
552 * There are pending frames.
553 * The socket lock must be held while it's corked.
556 if (likely(up->pending)) {
557 if (unlikely(up->pending != AF_INET)) {
565 ulen += sizeof(struct udphdr);
568 * Get and verify the address.
571 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
572 if (msg->msg_namelen < sizeof(*usin))
574 if (usin->sin_family != AF_INET) {
575 if (usin->sin_family != AF_UNSPEC)
576 return -EAFNOSUPPORT;
579 daddr = usin->sin_addr.s_addr;
580 dport = usin->sin_port;
584 if (sk->sk_state != TCP_ESTABLISHED)
585 return -EDESTADDRREQ;
588 /* Open fast path for connected socket.
589 Route will not be used, if at least one option is set.
593 ipc.addr = inet->saddr;
595 ipc.oif = sk->sk_bound_dev_if;
596 if (msg->msg_controllen) {
597 err = ip_cmsg_send(msg, &ipc);
608 ipc.addr = faddr = daddr;
610 if (ipc.opt && ipc.opt->srr) {
613 faddr = ipc.opt->faddr;
616 tos = RT_TOS(inet->tos);
617 if (sock_flag(sk, SOCK_LOCALROUTE) ||
618 (msg->msg_flags & MSG_DONTROUTE) ||
619 (ipc.opt && ipc.opt->is_strictroute)) {
624 if (MULTICAST(daddr)) {
626 ipc.oif = inet->mc_index;
628 saddr = inet->mc_addr;
633 rt = (struct rtable*)sk_dst_check(sk, 0);
636 struct flowi fl = { .oif = ipc.oif,
641 .proto = sk->sk_protocol,
643 { .sport = inet->sport,
644 .dport = dport } } };
645 security_sk_classify_flow(sk, &fl);
646 err = ip_route_output_flow(&rt, &fl, sk, 1);
648 if (err == -ENETUNREACH)
649 IP_INC_STATS_BH(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(UDP_MIB_SNDBUFERRORS, is_udplite);
721 dst_confirm(&rt->u.dst);
722 if (!(msg->msg_flags&MSG_PROBE) || len)
723 goto back_from_confirm;
728 int udp_sendpage(struct sock *sk, struct page *page, int offset,
729 size_t size, int flags)
731 struct udp_sock *up = udp_sk(sk);
735 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
737 /* Call udp_sendmsg to specify destination address which
738 * sendpage interface can't pass.
739 * This will succeed only when the socket is connected.
741 ret = udp_sendmsg(NULL, sk, &msg, 0);
748 if (unlikely(!up->pending)) {
751 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
755 ret = ip_append_page(sk, page, offset, size, flags);
756 if (ret == -EOPNOTSUPP) {
758 return sock_no_sendpage(sk->sk_socket, page, offset,
762 udp_flush_pending_frames(sk);
767 if (!(up->corkflag || (flags&MSG_MORE)))
768 ret = udp_push_pending_frames(sk);
777 * IOCTL requests applicable to the UDP protocol
780 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
785 int amount = atomic_read(&sk->sk_wmem_alloc);
786 return put_user(amount, (int __user *)arg);
792 unsigned long amount;
795 spin_lock_bh(&sk->sk_receive_queue.lock);
796 skb = skb_peek(&sk->sk_receive_queue);
799 * We will only return the amount
800 * of this packet since that is all
803 amount = skb->len - sizeof(struct udphdr);
805 spin_unlock_bh(&sk->sk_receive_queue.lock);
806 return put_user(amount, (int __user *)arg);
817 * This should be easy, if there is something there we
818 * return it, otherwise we block.
821 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
822 size_t len, int noblock, int flags, int *addr_len)
824 struct inet_sock *inet = inet_sk(sk);
825 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
827 unsigned int ulen, copied;
829 int is_udplite = IS_UDPLITE(sk);
832 * Check any passed addresses
835 *addr_len=sizeof(*sin);
837 if (flags & MSG_ERRQUEUE)
838 return ip_recv_error(sk, msg, len);
841 skb = skb_recv_datagram(sk, flags, noblock, &err);
845 ulen = skb->len - sizeof(struct udphdr);
849 else if (copied < ulen)
850 msg->msg_flags |= MSG_TRUNC;
853 * If checksum is needed at all, try to do it while copying the
854 * data. If the data is truncated, or if we only want a partial
855 * coverage checksum (UDP-Lite), do it before the copy.
858 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
859 if (udp_lib_checksum_complete(skb))
863 if (skb_csum_unnecessary(skb))
864 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
865 msg->msg_iov, copied );
867 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
876 sock_recv_timestamp(msg, sk, skb);
878 /* Copy the address. */
881 sin->sin_family = AF_INET;
882 sin->sin_port = udp_hdr(skb)->source;
883 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
884 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
886 if (inet->cmsg_flags)
887 ip_cmsg_recv(msg, skb);
890 if (flags & MSG_TRUNC)
894 skb_free_datagram(sk, skb);
899 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
901 skb_kill_datagram(sk, skb, flags);
909 int udp_disconnect(struct sock *sk, int flags)
911 struct inet_sock *inet = inet_sk(sk);
913 * 1003.1g - break association.
916 sk->sk_state = TCP_CLOSE;
919 sk->sk_bound_dev_if = 0;
920 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
921 inet_reset_saddr(sk);
923 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
924 sk->sk_prot->unhash(sk);
934 * >0: "udp encap" protocol resubmission
936 * Note that in the success and error cases, the skb is assumed to
937 * have either been requeued or freed.
939 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
941 struct udp_sock *up = udp_sk(sk);
945 * Charge it to the socket, dropping if the queue is full.
947 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
951 if (up->encap_type) {
953 * This is an encapsulation socket so pass the skb to
954 * the socket's udp_encap_rcv() hook. Otherwise, just
955 * fall through and pass this up the UDP socket.
956 * up->encap_rcv() returns the following value:
957 * =0 if skb was successfully passed to the encap
958 * handler or was discarded by it.
959 * >0 if skb should be passed on to UDP.
960 * <0 if skb should be resubmitted as proto -N
963 /* if we're overly short, let UDP handle it */
964 if (skb->len > sizeof(struct udphdr) &&
965 up->encap_rcv != NULL) {
968 ret = (*up->encap_rcv)(sk, skb);
970 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
975 /* FALLTHROUGH -- it's a UDP Packet */
979 * UDP-Lite specific tests, ignored on UDP sockets
981 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
984 * MIB statistics other than incrementing the error count are
985 * disabled for the following two types of errors: these depend
986 * on the application settings, not on the functioning of the
987 * protocol stack as such.
989 * RFC 3828 here recommends (sec 3.3): "There should also be a
990 * way ... to ... at least let the receiving application block
991 * delivery of packets with coverage values less than a value
992 * provided by the application."
994 if (up->pcrlen == 0) { /* full coverage was set */
995 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
996 "%d while full coverage %d requested\n",
997 UDP_SKB_CB(skb)->cscov, skb->len);
1000 /* The next case involves violating the min. coverage requested
1001 * by the receiver. This is subtle: if receiver wants x and x is
1002 * greater than the buffersize/MTU then receiver will complain
1003 * that it wants x while sender emits packets of smaller size y.
1004 * Therefore the above ...()->partial_cov statement is essential.
1006 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1007 LIMIT_NETDEBUG(KERN_WARNING
1008 "UDPLITE: coverage %d too small, need min %d\n",
1009 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1014 if (sk->sk_filter) {
1015 if (udp_lib_checksum_complete(skb))
1019 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1020 /* Note that an ENOMEM error is charged twice */
1022 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
1026 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1030 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
1036 * Multicasts and broadcasts go to each listener.
1038 * Note: called only from the BH handler context,
1039 * so we don't need to lock the hashes.
1041 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1043 __be32 saddr, __be32 daddr,
1044 struct hlist_head udptable[])
1049 read_lock(&udp_hash_lock);
1050 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1051 dif = skb->dev->ifindex;
1052 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1054 struct sock *sknext = NULL;
1057 struct sk_buff *skb1 = skb;
1059 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1060 uh->source, saddr, dif);
1062 skb1 = skb_clone(skb, GFP_ATOMIC);
1065 int ret = udp_queue_rcv_skb(sk, skb1);
1067 /* we should probably re-process instead
1068 * of dropping packets here. */
1075 read_unlock(&udp_hash_lock);
1079 /* Initialize UDP checksum. If exited with zero value (success),
1080 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1081 * Otherwise, csum completion requires chacksumming packet body,
1082 * including udp header and folding it to skb->csum.
1084 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1087 const struct iphdr *iph;
1090 UDP_SKB_CB(skb)->partial_cov = 0;
1091 UDP_SKB_CB(skb)->cscov = skb->len;
1093 if (proto == IPPROTO_UDPLITE) {
1094 err = udplite_checksum_init(skb, uh);
1100 if (uh->check == 0) {
1101 skb->ip_summed = CHECKSUM_UNNECESSARY;
1102 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1103 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1105 skb->ip_summed = CHECKSUM_UNNECESSARY;
1107 if (!skb_csum_unnecessary(skb))
1108 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1109 skb->len, proto, 0);
1110 /* Probably, we should checksum udp header (it should be in cache
1111 * in any case) and data in tiny packets (< rx copybreak).
1118 * All we need to do is get the socket, and then do a checksum.
1121 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1125 struct udphdr *uh = udp_hdr(skb);
1126 unsigned short ulen;
1127 struct rtable *rt = (struct rtable*)skb->dst;
1128 __be32 saddr = ip_hdr(skb)->saddr;
1129 __be32 daddr = ip_hdr(skb)->daddr;
1132 * Validate the packet.
1134 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1135 goto drop; /* No space for header. */
1137 ulen = ntohs(uh->len);
1138 if (ulen > skb->len)
1141 if (proto == IPPROTO_UDP) {
1142 /* UDP validates ulen. */
1143 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1148 if (udp4_csum_init(skb, uh, proto))
1151 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1152 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1154 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1155 inet_iif(skb), udptable);
1158 int ret = udp_queue_rcv_skb(sk, skb);
1161 /* a return value > 0 means to resubmit the input, but
1162 * it wants the return to be -protocol, or 0
1169 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1173 /* No socket. Drop packet silently, if checksum is wrong */
1174 if (udp_lib_checksum_complete(skb))
1177 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1178 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1181 * Hmm. We got an UDP packet to a port to which we
1182 * don't wanna listen. Ignore it.
1188 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1189 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1200 * RFC1122: OK. Discards the bad packet silently (as far as
1201 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1203 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1204 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1211 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1216 int udp_rcv(struct sk_buff *skb)
1218 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1221 int udp_destroy_sock(struct sock *sk)
1224 udp_flush_pending_frames(sk);
1230 * Socket option code for UDP
1232 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1233 char __user *optval, int optlen,
1234 int (*push_pending_frames)(struct sock *))
1236 struct udp_sock *up = udp_sk(sk);
1240 if (optlen<sizeof(int))
1243 if (get_user(val, (int __user *)optval))
1253 (*push_pending_frames)(sk);
1261 case UDP_ENCAP_ESPINUDP:
1262 case UDP_ENCAP_ESPINUDP_NON_IKE:
1263 up->encap_rcv = xfrm4_udp_encap_rcv;
1265 case UDP_ENCAP_L2TPINUDP:
1266 up->encap_type = val;
1275 * UDP-Lite's partial checksum coverage (RFC 3828).
1277 /* The sender sets actual checksum coverage length via this option.
1278 * The case coverage > packet length is handled by send module. */
1279 case UDPLITE_SEND_CSCOV:
1280 if (!up->pcflag) /* Disable the option on UDP sockets */
1281 return -ENOPROTOOPT;
1282 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1285 up->pcflag |= UDPLITE_SEND_CC;
1288 /* The receiver specifies a minimum checksum coverage value. To make
1289 * sense, this should be set to at least 8 (as done below). If zero is
1290 * used, this again means full checksum coverage. */
1291 case UDPLITE_RECV_CSCOV:
1292 if (!up->pcflag) /* Disable the option on UDP sockets */
1293 return -ENOPROTOOPT;
1294 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1297 up->pcflag |= UDPLITE_RECV_CC;
1308 int udp_setsockopt(struct sock *sk, int level, int optname,
1309 char __user *optval, int optlen)
1311 if (level == SOL_UDP || level == SOL_UDPLITE)
1312 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1313 udp_push_pending_frames);
1314 return ip_setsockopt(sk, level, optname, optval, optlen);
1317 #ifdef CONFIG_COMPAT
1318 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1319 char __user *optval, int optlen)
1321 if (level == SOL_UDP || level == SOL_UDPLITE)
1322 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1323 udp_push_pending_frames);
1324 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1328 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1329 char __user *optval, int __user *optlen)
1331 struct udp_sock *up = udp_sk(sk);
1334 if (get_user(len,optlen))
1337 len = min_t(unsigned int, len, sizeof(int));
1348 val = up->encap_type;
1351 /* The following two cannot be changed on UDP sockets, the return is
1352 * always 0 (which corresponds to the full checksum coverage of UDP). */
1353 case UDPLITE_SEND_CSCOV:
1357 case UDPLITE_RECV_CSCOV:
1362 return -ENOPROTOOPT;
1365 if (put_user(len, optlen))
1367 if (copy_to_user(optval, &val,len))
1372 int udp_getsockopt(struct sock *sk, int level, int optname,
1373 char __user *optval, int __user *optlen)
1375 if (level == SOL_UDP || level == SOL_UDPLITE)
1376 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1377 return ip_getsockopt(sk, level, optname, optval, optlen);
1380 #ifdef CONFIG_COMPAT
1381 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1382 char __user *optval, int __user *optlen)
1384 if (level == SOL_UDP || level == SOL_UDPLITE)
1385 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1386 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1390 * udp_poll - wait for a UDP event.
1391 * @file - file struct
1393 * @wait - poll table
1395 * This is same as datagram poll, except for the special case of
1396 * blocking sockets. If application is using a blocking fd
1397 * and a packet with checksum error is in the queue;
1398 * then it could get return from select indicating data available
1399 * but then block when reading it. Add special case code
1400 * to work around these arguably broken applications.
1402 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1404 unsigned int mask = datagram_poll(file, sock, wait);
1405 struct sock *sk = sock->sk;
1406 int is_lite = IS_UDPLITE(sk);
1408 /* Check for false positives due to checksum errors */
1409 if ( (mask & POLLRDNORM) &&
1410 !(file->f_flags & O_NONBLOCK) &&
1411 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1412 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1413 struct sk_buff *skb;
1415 spin_lock_bh(&rcvq->lock);
1416 while ((skb = skb_peek(rcvq)) != NULL &&
1417 udp_lib_checksum_complete(skb)) {
1418 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1419 __skb_unlink(skb, rcvq);
1422 spin_unlock_bh(&rcvq->lock);
1424 /* nothing to see, move along */
1426 mask &= ~(POLLIN | POLLRDNORM);
1433 struct proto udp_prot = {
1435 .owner = THIS_MODULE,
1436 .close = udp_lib_close,
1437 .connect = ip4_datagram_connect,
1438 .disconnect = udp_disconnect,
1440 .destroy = udp_destroy_sock,
1441 .setsockopt = udp_setsockopt,
1442 .getsockopt = udp_getsockopt,
1443 .sendmsg = udp_sendmsg,
1444 .recvmsg = udp_recvmsg,
1445 .sendpage = udp_sendpage,
1446 .backlog_rcv = udp_queue_rcv_skb,
1447 .hash = udp_lib_hash,
1448 .unhash = udp_lib_unhash,
1449 .get_port = udp_v4_get_port,
1450 .obj_size = sizeof(struct udp_sock),
1451 #ifdef CONFIG_COMPAT
1452 .compat_setsockopt = compat_udp_setsockopt,
1453 .compat_getsockopt = compat_udp_getsockopt,
1457 /* ------------------------------------------------------------------------ */
1458 #ifdef CONFIG_PROC_FS
1460 static struct sock *udp_get_first(struct seq_file *seq)
1463 struct udp_iter_state *state = seq->private;
1465 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1466 struct hlist_node *node;
1467 sk_for_each(sk, node, state->hashtable + state->bucket) {
1468 if (sk->sk_family == state->family)
1477 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1479 struct udp_iter_state *state = seq->private;
1485 } while (sk && sk->sk_family != state->family);
1487 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1488 sk = sk_head(state->hashtable + state->bucket);
1494 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1496 struct sock *sk = udp_get_first(seq);
1499 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1501 return pos ? NULL : sk;
1504 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1506 read_lock(&udp_hash_lock);
1507 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1510 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1515 sk = udp_get_idx(seq, 0);
1517 sk = udp_get_next(seq, v);
1523 static void udp_seq_stop(struct seq_file *seq, void *v)
1525 read_unlock(&udp_hash_lock);
1528 static int udp_seq_open(struct inode *inode, struct file *file)
1530 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1531 struct seq_file *seq;
1533 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1537 s->family = afinfo->family;
1538 s->hashtable = afinfo->hashtable;
1539 s->seq_ops.start = udp_seq_start;
1540 s->seq_ops.next = udp_seq_next;
1541 s->seq_ops.show = afinfo->seq_show;
1542 s->seq_ops.stop = udp_seq_stop;
1544 rc = seq_open(file, &s->seq_ops);
1548 seq = file->private_data;
1557 /* ------------------------------------------------------------------------ */
1558 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1560 struct proc_dir_entry *p;
1565 afinfo->seq_fops->owner = afinfo->owner;
1566 afinfo->seq_fops->open = udp_seq_open;
1567 afinfo->seq_fops->read = seq_read;
1568 afinfo->seq_fops->llseek = seq_lseek;
1569 afinfo->seq_fops->release = seq_release_private;
1571 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops);
1579 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1583 proc_net_remove(&init_net, afinfo->name);
1584 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1587 /* ------------------------------------------------------------------------ */
1588 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1590 struct inet_sock *inet = inet_sk(sp);
1591 __be32 dest = inet->daddr;
1592 __be32 src = inet->rcv_saddr;
1593 __u16 destp = ntohs(inet->dport);
1594 __u16 srcp = ntohs(inet->sport);
1596 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1597 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1598 bucket, src, srcp, dest, destp, sp->sk_state,
1599 atomic_read(&sp->sk_wmem_alloc),
1600 atomic_read(&sp->sk_rmem_alloc),
1601 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1602 atomic_read(&sp->sk_refcnt), sp);
1605 int udp4_seq_show(struct seq_file *seq, void *v)
1607 if (v == SEQ_START_TOKEN)
1608 seq_printf(seq, "%-127s\n",
1609 " sl local_address rem_address st tx_queue "
1610 "rx_queue tr tm->when retrnsmt uid timeout "
1614 struct udp_iter_state *state = seq->private;
1616 udp4_format_sock(v, tmpbuf, state->bucket);
1617 seq_printf(seq, "%-127s\n", tmpbuf);
1622 /* ------------------------------------------------------------------------ */
1623 static struct file_operations udp4_seq_fops;
1624 static struct udp_seq_afinfo udp4_seq_afinfo = {
1625 .owner = THIS_MODULE,
1628 .hashtable = udp_hash,
1629 .seq_show = udp4_seq_show,
1630 .seq_fops = &udp4_seq_fops,
1633 int __init udp4_proc_init(void)
1635 return udp_proc_register(&udp4_seq_afinfo);
1638 void udp4_proc_exit(void)
1640 udp_proc_unregister(&udp4_seq_afinfo);
1642 #endif /* CONFIG_PROC_FS */
1644 EXPORT_SYMBOL(udp_disconnect);
1645 EXPORT_SYMBOL(udp_hash);
1646 EXPORT_SYMBOL(udp_hash_lock);
1647 EXPORT_SYMBOL(udp_ioctl);
1648 EXPORT_SYMBOL(udp_get_port);
1649 EXPORT_SYMBOL(udp_prot);
1650 EXPORT_SYMBOL(udp_sendmsg);
1651 EXPORT_SYMBOL(udp_lib_getsockopt);
1652 EXPORT_SYMBOL(udp_lib_setsockopt);
1653 EXPORT_SYMBOL(udp_poll);
1655 #ifdef CONFIG_PROC_FS
1656 EXPORT_SYMBOL(udp_proc_register);
1657 EXPORT_SYMBOL(udp_proc_unregister);