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
75 * This program is free software; you can redistribute it and/or
76 * modify it under the terms of the GNU General Public License
77 * as published by the Free Software Foundation; either version
78 * 2 of the License, or (at your option) any later version.
81 #include <asm/system.h>
82 #include <asm/uaccess.h>
83 #include <asm/ioctls.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/icmp.h>
101 #include <net/route.h>
102 #include <net/checksum.h>
103 #include <net/xfrm.h>
104 #include "udp_impl.h"
107 * Snmp MIB for the UDP layer
110 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly;
112 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
113 DEFINE_RWLOCK(udp_hash_lock);
115 static int udp_port_rover;
117 static inline int __udp_lib_lport_inuse(__u16 num, struct hlist_head udptable[])
120 struct hlist_node *node;
122 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
123 if (sk->sk_hash == num)
129 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
131 * @sk: socket struct in question
132 * @snum: port number to look up
133 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
134 * @port_rover: pointer to record of last unallocated port
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[], int *port_rover,
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);
149 int best_size_so_far, best, result, i;
151 if (*port_rover > sysctl_local_port_range[1] ||
152 *port_rover < sysctl_local_port_range[0])
153 *port_rover = sysctl_local_port_range[0];
154 best_size_so_far = 32767;
155 best = result = *port_rover;
156 for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) {
159 head = &udptable[result & (UDP_HTABLE_SIZE - 1)];
160 if (hlist_empty(head)) {
161 if (result > sysctl_local_port_range[1])
162 result = sysctl_local_port_range[0] +
163 ((result - sysctl_local_port_range[0]) &
164 (UDP_HTABLE_SIZE - 1));
168 sk_for_each(sk2, node, head) {
169 if (++size >= best_size_so_far)
172 best_size_so_far = size;
178 for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result += UDP_HTABLE_SIZE) {
179 if (result > sysctl_local_port_range[1])
180 result = sysctl_local_port_range[0]
181 + ((result - sysctl_local_port_range[0]) &
182 (UDP_HTABLE_SIZE - 1));
183 if (! __udp_lib_lport_inuse(result, udptable))
186 if (i >= (1 << 16) / UDP_HTABLE_SIZE)
189 *port_rover = snum = result;
191 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
193 sk_for_each(sk2, node, head)
194 if (sk2->sk_hash == snum &&
196 (!sk2->sk_reuse || !sk->sk_reuse) &&
197 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
198 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
199 (*saddr_comp)(sk, sk2) )
202 inet_sk(sk)->num = snum;
204 if (sk_unhashed(sk)) {
205 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
206 sk_add_node(sk, head);
207 sock_prot_inc_use(sk->sk_prot);
211 write_unlock_bh(&udp_hash_lock);
215 __inline__ int udp_get_port(struct sock *sk, unsigned short snum,
216 int (*scmp)(const struct sock *, const struct sock *))
218 return __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp);
221 inline int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
223 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
225 return ( !ipv6_only_sock(sk2) &&
226 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
227 inet1->rcv_saddr == inet2->rcv_saddr ));
230 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
232 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
235 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
236 * harder than this. -DaveM
238 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
239 __be32 daddr, __be16 dport,
240 int dif, struct hlist_head udptable[])
242 struct sock *sk, *result = NULL;
243 struct hlist_node *node;
244 unsigned short hnum = ntohs(dport);
247 read_lock(&udp_hash_lock);
248 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
249 struct inet_sock *inet = inet_sk(sk);
251 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
252 int score = (sk->sk_family == PF_INET ? 1 : 0);
253 if (inet->rcv_saddr) {
254 if (inet->rcv_saddr != daddr)
259 if (inet->daddr != saddr)
264 if (inet->dport != sport)
268 if (sk->sk_bound_dev_if) {
269 if (sk->sk_bound_dev_if != dif)
276 } else if(score > badness) {
284 read_unlock(&udp_hash_lock);
288 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
289 __be16 loc_port, __be32 loc_addr,
290 __be16 rmt_port, __be32 rmt_addr,
293 struct hlist_node *node;
295 unsigned short hnum = ntohs(loc_port);
297 sk_for_each_from(s, node) {
298 struct inet_sock *inet = inet_sk(s);
300 if (s->sk_hash != hnum ||
301 (inet->daddr && inet->daddr != rmt_addr) ||
302 (inet->dport != rmt_port && inet->dport) ||
303 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
305 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
307 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
317 * This routine is called by the ICMP module when it gets some
318 * sort of error condition. If err < 0 then the socket should
319 * be closed and the error returned to the user. If err > 0
320 * it's just the icmp type << 8 | icmp code.
321 * Header points to the ip header of the error packet. We move
322 * on past this. Then (as it used to claim before adjustment)
323 * header points to the first 8 bytes of the udp header. We need
324 * to find the appropriate port.
327 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
329 struct inet_sock *inet;
330 struct iphdr *iph = (struct iphdr*)skb->data;
331 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
332 int type = skb->h.icmph->type;
333 int code = skb->h.icmph->code;
338 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
339 skb->dev->ifindex, udptable );
341 ICMP_INC_STATS_BH(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 __inline__ void udp_err(struct sk_buff *skb, u32 info)
395 return __udp4_lib_err(skb, info, udp_hash);
399 * Throw away all pending data and cancel the corking. Socket is locked.
401 static void udp_flush_pending_frames(struct sock *sk)
403 struct udp_sock *up = udp_sk(sk);
408 ip_flush_pending_frames(sk);
413 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
414 * @sk: socket we are sending on
415 * @skb: sk_buff containing the filled-in UDP header
416 * (checksum field must be zeroed out)
418 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
419 __be32 src, __be32 dst, int len )
422 struct udphdr *uh = skb->h.uh;
425 if (skb_queue_len(&sk->sk_write_queue) == 1) {
427 * Only one fragment on the socket.
429 skb->csum_offset = offsetof(struct udphdr, check);
430 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
433 * HW-checksum won't work as there are two or more
434 * fragments on the socket so that all csums of sk_buffs
437 offset = skb->h.raw - skb->data;
438 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
440 skb->ip_summed = CHECKSUM_NONE;
442 skb_queue_walk(&sk->sk_write_queue, skb) {
443 csum = csum_add(csum, skb->csum);
446 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
448 uh->check = CSUM_MANGLED_0;
453 * Push out all pending data as one UDP datagram. Socket is locked.
455 static int udp_push_pending_frames(struct sock *sk)
457 struct udp_sock *up = udp_sk(sk);
458 struct inet_sock *inet = inet_sk(sk);
459 struct flowi *fl = &inet->cork.fl;
465 /* Grab the skbuff where UDP header space exists. */
466 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
470 * Create a UDP header
473 uh->source = fl->fl_ip_sport;
474 uh->dest = fl->fl_ip_dport;
475 uh->len = htons(up->len);
478 if (up->pcflag) /* UDP-Lite */
479 csum = udplite_csum_outgoing(sk, skb);
481 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
483 skb->ip_summed = CHECKSUM_NONE;
486 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
488 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
491 } else /* `normal' UDP */
492 csum = udp_csum_outgoing(sk, skb);
494 /* add protocol-dependent pseudo-header */
495 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
496 sk->sk_protocol, csum );
498 uh->check = CSUM_MANGLED_0;
501 err = ip_push_pending_frames(sk);
508 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
511 struct inet_sock *inet = inet_sk(sk);
512 struct udp_sock *up = udp_sk(sk);
514 struct ipcm_cookie ipc;
515 struct rtable *rt = NULL;
518 __be32 daddr, faddr, saddr;
521 int err, is_udplite = up->pcflag;
522 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
523 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
532 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
539 * There are pending frames.
540 * The socket lock must be held while it's corked.
543 if (likely(up->pending)) {
544 if (unlikely(up->pending != AF_INET)) {
552 ulen += sizeof(struct udphdr);
555 * Get and verify the address.
558 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
559 if (msg->msg_namelen < sizeof(*usin))
561 if (usin->sin_family != AF_INET) {
562 if (usin->sin_family != AF_UNSPEC)
563 return -EAFNOSUPPORT;
566 daddr = usin->sin_addr.s_addr;
567 dport = usin->sin_port;
571 if (sk->sk_state != TCP_ESTABLISHED)
572 return -EDESTADDRREQ;
575 /* Open fast path for connected socket.
576 Route will not be used, if at least one option is set.
580 ipc.addr = inet->saddr;
582 ipc.oif = sk->sk_bound_dev_if;
583 if (msg->msg_controllen) {
584 err = ip_cmsg_send(msg, &ipc);
595 ipc.addr = faddr = daddr;
597 if (ipc.opt && ipc.opt->srr) {
600 faddr = ipc.opt->faddr;
603 tos = RT_TOS(inet->tos);
604 if (sock_flag(sk, SOCK_LOCALROUTE) ||
605 (msg->msg_flags & MSG_DONTROUTE) ||
606 (ipc.opt && ipc.opt->is_strictroute)) {
611 if (MULTICAST(daddr)) {
613 ipc.oif = inet->mc_index;
615 saddr = inet->mc_addr;
620 rt = (struct rtable*)sk_dst_check(sk, 0);
623 struct flowi fl = { .oif = ipc.oif,
628 .proto = sk->sk_protocol,
630 { .sport = inet->sport,
631 .dport = dport } } };
632 security_sk_classify_flow(sk, &fl);
633 err = ip_route_output_flow(&rt, &fl, sk, 1);
638 if ((rt->rt_flags & RTCF_BROADCAST) &&
639 !sock_flag(sk, SOCK_BROADCAST))
642 sk_dst_set(sk, dst_clone(&rt->u.dst));
645 if (msg->msg_flags&MSG_CONFIRM)
651 daddr = ipc.addr = rt->rt_dst;
654 if (unlikely(up->pending)) {
655 /* The socket is already corked while preparing it. */
656 /* ... which is an evident application bug. --ANK */
659 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
664 * Now cork the socket to pend data.
666 inet->cork.fl.fl4_dst = daddr;
667 inet->cork.fl.fl_ip_dport = dport;
668 inet->cork.fl.fl4_src = saddr;
669 inet->cork.fl.fl_ip_sport = inet->sport;
670 up->pending = AF_INET;
674 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
675 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
676 sizeof(struct udphdr), &ipc, rt,
677 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
679 udp_flush_pending_frames(sk);
681 err = udp_push_pending_frames(sk);
682 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
691 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
695 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
696 * ENOBUFS might not be good (it's not tunable per se), but otherwise
697 * we don't have a good statistic (IpOutDiscards but it can be too many
698 * things). We could add another new stat but at least for now that
699 * seems like overkill.
701 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
702 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
707 dst_confirm(&rt->u.dst);
708 if (!(msg->msg_flags&MSG_PROBE) || len)
709 goto back_from_confirm;
714 int udp_sendpage(struct sock *sk, struct page *page, int offset,
715 size_t size, int flags)
717 struct udp_sock *up = udp_sk(sk);
721 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
723 /* Call udp_sendmsg to specify destination address which
724 * sendpage interface can't pass.
725 * This will succeed only when the socket is connected.
727 ret = udp_sendmsg(NULL, sk, &msg, 0);
734 if (unlikely(!up->pending)) {
737 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
741 ret = ip_append_page(sk, page, offset, size, flags);
742 if (ret == -EOPNOTSUPP) {
744 return sock_no_sendpage(sk->sk_socket, page, offset,
748 udp_flush_pending_frames(sk);
753 if (!(up->corkflag || (flags&MSG_MORE)))
754 ret = udp_push_pending_frames(sk);
763 * IOCTL requests applicable to the UDP protocol
766 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
772 int amount = atomic_read(&sk->sk_wmem_alloc);
773 return put_user(amount, (int __user *)arg);
779 unsigned long amount;
782 spin_lock_bh(&sk->sk_receive_queue.lock);
783 skb = skb_peek(&sk->sk_receive_queue);
786 * We will only return the amount
787 * of this packet since that is all
790 amount = skb->len - sizeof(struct udphdr);
792 spin_unlock_bh(&sk->sk_receive_queue.lock);
793 return put_user(amount, (int __user *)arg);
803 * This should be easy, if there is something there we
804 * return it, otherwise we block.
807 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
808 size_t len, int noblock, int flags, int *addr_len)
810 struct inet_sock *inet = inet_sk(sk);
811 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
813 unsigned int ulen, copied;
815 int is_udplite = IS_UDPLITE(sk);
818 * Check any passed addresses
821 *addr_len=sizeof(*sin);
823 if (flags & MSG_ERRQUEUE)
824 return ip_recv_error(sk, msg, len);
827 skb = skb_recv_datagram(sk, flags, noblock, &err);
831 ulen = skb->len - sizeof(struct udphdr);
835 else if (copied < ulen)
836 msg->msg_flags |= MSG_TRUNC;
839 * If checksum is needed at all, try to do it while copying the
840 * data. If the data is truncated, or if we only want a partial
841 * coverage checksum (UDP-Lite), do it before the copy.
844 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
845 if (udp_lib_checksum_complete(skb))
849 if (skb->ip_summed == CHECKSUM_UNNECESSARY)
850 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
851 msg->msg_iov, copied );
853 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
862 sock_recv_timestamp(msg, sk, skb);
864 /* Copy the address. */
867 sin->sin_family = AF_INET;
868 sin->sin_port = skb->h.uh->source;
869 sin->sin_addr.s_addr = skb->nh.iph->saddr;
870 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
872 if (inet->cmsg_flags)
873 ip_cmsg_recv(msg, skb);
876 if (flags & MSG_TRUNC)
880 skb_free_datagram(sk, skb);
885 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
887 skb_kill_datagram(sk, skb, flags);
895 int udp_disconnect(struct sock *sk, int flags)
897 struct inet_sock *inet = inet_sk(sk);
899 * 1003.1g - break association.
902 sk->sk_state = TCP_CLOSE;
905 sk->sk_bound_dev_if = 0;
906 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
907 inet_reset_saddr(sk);
909 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
910 sk->sk_prot->unhash(sk);
918 * 1 if the the UDP system should process it
919 * 0 if we should drop this packet
920 * -1 if it should get processed by xfrm4_rcv_encap
922 static int udp_encap_rcv(struct sock * sk, struct sk_buff *skb)
927 struct udp_sock *up = udp_sk(sk);
934 __u16 encap_type = up->encap_type;
936 /* if we're overly short, let UDP handle it */
937 len = skb->len - sizeof(struct udphdr);
941 /* if this is not encapsulated socket, then just return now */
945 /* If this is a paged skb, make sure we pull up
946 * whatever data we need to look at. */
947 if (!pskb_may_pull(skb, sizeof(struct udphdr) + min(len, 8)))
950 /* Now we can get the pointers */
952 udpdata = (__u8 *)uh + sizeof(struct udphdr);
953 udpdata32 = (__be32 *)udpdata;
955 switch (encap_type) {
957 case UDP_ENCAP_ESPINUDP:
958 /* Check if this is a keepalive packet. If so, eat it. */
959 if (len == 1 && udpdata[0] == 0xff) {
961 } else if (len > sizeof(struct ip_esp_hdr) && udpdata32[0] != 0 ) {
962 /* ESP Packet without Non-ESP header */
963 len = sizeof(struct udphdr);
965 /* Must be an IKE packet.. pass it through */
968 case UDP_ENCAP_ESPINUDP_NON_IKE:
969 /* Check if this is a keepalive packet. If so, eat it. */
970 if (len == 1 && udpdata[0] == 0xff) {
972 } else if (len > 2 * sizeof(u32) + sizeof(struct ip_esp_hdr) &&
973 udpdata32[0] == 0 && udpdata32[1] == 0) {
975 /* ESP Packet with Non-IKE marker */
976 len = sizeof(struct udphdr) + 2 * sizeof(u32);
978 /* Must be an IKE packet.. pass it through */
983 /* At this point we are sure that this is an ESPinUDP packet,
984 * so we need to remove 'len' bytes from the packet (the UDP
985 * header and optional ESP marker bytes) and then modify the
986 * protocol to ESP, and then call into the transform receiver.
988 if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
991 /* Now we can update and verify the packet length... */
993 iphlen = iph->ihl << 2;
994 iph->tot_len = htons(ntohs(iph->tot_len) - len);
995 if (skb->len < iphlen + len) {
996 /* packet is too small!?! */
1000 /* pull the data buffer up to the ESP header and set the
1001 * transport header to point to ESP. Keep UDP on the stack
1004 skb->h.raw = skb_pull(skb, len);
1006 /* modify the protocol (it's ESP!) */
1007 iph->protocol = IPPROTO_ESP;
1009 /* and let the caller know to send this into the ESP processor... */
1017 * >0: "udp encap" protocol resubmission
1019 * Note that in the success and error cases, the skb is assumed to
1020 * have either been requeued or freed.
1022 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1024 struct udp_sock *up = udp_sk(sk);
1028 * Charge it to the socket, dropping if the queue is full.
1030 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1034 if (up->encap_type) {
1036 * This is an encapsulation socket, so let's see if this is
1037 * an encapsulated packet.
1038 * If it's a keepalive packet, then just eat it.
1039 * If it's an encapsulateed packet, then pass it to the
1040 * IPsec xfrm input and return the response
1041 * appropriately. Otherwise, just fall through and
1042 * pass this up the UDP socket.
1046 ret = udp_encap_rcv(sk, skb);
1048 /* Eat the packet .. */
1053 /* process the ESP packet */
1054 ret = xfrm4_rcv_encap(skb, up->encap_type);
1055 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1058 /* FALLTHROUGH -- it's a UDP Packet */
1062 * UDP-Lite specific tests, ignored on UDP sockets
1064 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1067 * MIB statistics other than incrementing the error count are
1068 * disabled for the following two types of errors: these depend
1069 * on the application settings, not on the functioning of the
1070 * protocol stack as such.
1072 * RFC 3828 here recommends (sec 3.3): "There should also be a
1073 * way ... to ... at least let the receiving application block
1074 * delivery of packets with coverage values less than a value
1075 * provided by the application."
1077 if (up->pcrlen == 0) { /* full coverage was set */
1078 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1079 "%d while full coverage %d requested\n",
1080 UDP_SKB_CB(skb)->cscov, skb->len);
1083 /* The next case involves violating the min. coverage requested
1084 * by the receiver. This is subtle: if receiver wants x and x is
1085 * greater than the buffersize/MTU then receiver will complain
1086 * that it wants x while sender emits packets of smaller size y.
1087 * Therefore the above ...()->partial_cov statement is essential.
1089 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1090 LIMIT_NETDEBUG(KERN_WARNING
1091 "UDPLITE: coverage %d too small, need min %d\n",
1092 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1097 if (sk->sk_filter) {
1098 if (udp_lib_checksum_complete(skb))
1102 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1103 /* Note that an ENOMEM error is charged twice */
1105 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
1109 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1113 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
1119 * Multicasts and broadcasts go to each listener.
1121 * Note: called only from the BH handler context,
1122 * so we don't need to lock the hashes.
1124 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1126 __be32 saddr, __be32 daddr,
1127 struct hlist_head udptable[])
1132 read_lock(&udp_hash_lock);
1133 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1134 dif = skb->dev->ifindex;
1135 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1137 struct sock *sknext = NULL;
1140 struct sk_buff *skb1 = skb;
1142 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1143 uh->source, saddr, dif);
1145 skb1 = skb_clone(skb, GFP_ATOMIC);
1148 int ret = udp_queue_rcv_skb(sk, skb1);
1150 /* we should probably re-process instead
1151 * of dropping packets here. */
1158 read_unlock(&udp_hash_lock);
1162 /* Initialize UDP checksum. If exited with zero value (success),
1163 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1164 * Otherwise, csum completion requires chacksumming packet body,
1165 * including udp header and folding it to skb->csum.
1167 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1172 UDP_SKB_CB(skb)->partial_cov = 0;
1173 UDP_SKB_CB(skb)->cscov = skb->len;
1175 if (proto == IPPROTO_UDPLITE) {
1176 err = udplite_checksum_init(skb, uh);
1181 if (uh->check == 0) {
1182 skb->ip_summed = CHECKSUM_UNNECESSARY;
1183 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1184 if (!csum_tcpudp_magic(skb->nh.iph->saddr, skb->nh.iph->daddr,
1185 skb->len, proto, skb->csum))
1186 skb->ip_summed = CHECKSUM_UNNECESSARY;
1188 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
1189 skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr,
1191 skb->len, proto, 0);
1192 /* Probably, we should checksum udp header (it should be in cache
1193 * in any case) and data in tiny packets (< rx copybreak).
1200 * All we need to do is get the socket, and then do a checksum.
1203 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1207 struct udphdr *uh = skb->h.uh;
1208 unsigned short ulen;
1209 struct rtable *rt = (struct rtable*)skb->dst;
1210 __be32 saddr = skb->nh.iph->saddr;
1211 __be32 daddr = skb->nh.iph->daddr;
1214 * Validate the packet.
1216 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1217 goto drop; /* No space for header. */
1219 ulen = ntohs(uh->len);
1220 if (ulen > skb->len)
1223 if (proto == IPPROTO_UDP) {
1224 /* UDP validates ulen. */
1225 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1230 if (udp4_csum_init(skb, uh, proto))
1233 if(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1234 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1236 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1237 skb->dev->ifindex, udptable );
1240 int ret = udp_queue_rcv_skb(sk, skb);
1243 /* a return value > 0 means to resubmit the input, but
1244 * it wants the return to be -protocol, or 0
1251 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1255 /* No socket. Drop packet silently, if checksum is wrong */
1256 if (udp_lib_checksum_complete(skb))
1259 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1260 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1263 * Hmm. We got an UDP packet to a port to which we
1264 * don't wanna listen. Ignore it.
1270 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1271 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1282 * RFC1122: OK. Discards the bad packet silently (as far as
1283 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1285 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1286 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1293 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1298 __inline__ int udp_rcv(struct sk_buff *skb)
1300 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1303 int udp_destroy_sock(struct sock *sk)
1306 udp_flush_pending_frames(sk);
1312 * Socket option code for UDP
1314 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1315 char __user *optval, int optlen,
1316 int (*push_pending_frames)(struct sock *))
1318 struct udp_sock *up = udp_sk(sk);
1322 if(optlen<sizeof(int))
1325 if (get_user(val, (int __user *)optval))
1335 (*push_pending_frames)(sk);
1343 case UDP_ENCAP_ESPINUDP:
1344 case UDP_ENCAP_ESPINUDP_NON_IKE:
1345 up->encap_type = val;
1354 * UDP-Lite's partial checksum coverage (RFC 3828).
1356 /* The sender sets actual checksum coverage length via this option.
1357 * The case coverage > packet length is handled by send module. */
1358 case UDPLITE_SEND_CSCOV:
1359 if (!up->pcflag) /* Disable the option on UDP sockets */
1360 return -ENOPROTOOPT;
1361 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1364 up->pcflag |= UDPLITE_SEND_CC;
1367 /* The receiver specifies a minimum checksum coverage value. To make
1368 * sense, this should be set to at least 8 (as done below). If zero is
1369 * used, this again means full checksum coverage. */
1370 case UDPLITE_RECV_CSCOV:
1371 if (!up->pcflag) /* Disable the option on UDP sockets */
1372 return -ENOPROTOOPT;
1373 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1376 up->pcflag |= UDPLITE_RECV_CC;
1387 int udp_setsockopt(struct sock *sk, int level, int optname,
1388 char __user *optval, int optlen)
1390 if (level == SOL_UDP || level == SOL_UDPLITE)
1391 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1392 udp_push_pending_frames);
1393 return ip_setsockopt(sk, level, optname, optval, optlen);
1396 #ifdef CONFIG_COMPAT
1397 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1398 char __user *optval, int optlen)
1400 if (level == SOL_UDP || level == SOL_UDPLITE)
1401 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1402 udp_push_pending_frames);
1403 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1407 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1408 char __user *optval, int __user *optlen)
1410 struct udp_sock *up = udp_sk(sk);
1413 if(get_user(len,optlen))
1416 len = min_t(unsigned int, len, sizeof(int));
1427 val = up->encap_type;
1430 /* The following two cannot be changed on UDP sockets, the return is
1431 * always 0 (which corresponds to the full checksum coverage of UDP). */
1432 case UDPLITE_SEND_CSCOV:
1436 case UDPLITE_RECV_CSCOV:
1441 return -ENOPROTOOPT;
1444 if(put_user(len, optlen))
1446 if(copy_to_user(optval, &val,len))
1451 int udp_getsockopt(struct sock *sk, int level, int optname,
1452 char __user *optval, int __user *optlen)
1454 if (level == SOL_UDP || level == SOL_UDPLITE)
1455 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1456 return ip_getsockopt(sk, level, optname, optval, optlen);
1459 #ifdef CONFIG_COMPAT
1460 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1461 char __user *optval, int __user *optlen)
1463 if (level == SOL_UDP || level == SOL_UDPLITE)
1464 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1465 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1469 * udp_poll - wait for a UDP event.
1470 * @file - file struct
1472 * @wait - poll table
1474 * This is same as datagram poll, except for the special case of
1475 * blocking sockets. If application is using a blocking fd
1476 * and a packet with checksum error is in the queue;
1477 * then it could get return from select indicating data available
1478 * but then block when reading it. Add special case code
1479 * to work around these arguably broken applications.
1481 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1483 unsigned int mask = datagram_poll(file, sock, wait);
1484 struct sock *sk = sock->sk;
1485 int is_lite = IS_UDPLITE(sk);
1487 /* Check for false positives due to checksum errors */
1488 if ( (mask & POLLRDNORM) &&
1489 !(file->f_flags & O_NONBLOCK) &&
1490 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1491 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1492 struct sk_buff *skb;
1494 spin_lock_bh(&rcvq->lock);
1495 while ((skb = skb_peek(rcvq)) != NULL &&
1496 udp_lib_checksum_complete(skb)) {
1497 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1498 __skb_unlink(skb, rcvq);
1501 spin_unlock_bh(&rcvq->lock);
1503 /* nothing to see, move along */
1505 mask &= ~(POLLIN | POLLRDNORM);
1512 struct proto udp_prot = {
1514 .owner = THIS_MODULE,
1515 .close = udp_lib_close,
1516 .connect = ip4_datagram_connect,
1517 .disconnect = udp_disconnect,
1519 .destroy = udp_destroy_sock,
1520 .setsockopt = udp_setsockopt,
1521 .getsockopt = udp_getsockopt,
1522 .sendmsg = udp_sendmsg,
1523 .recvmsg = udp_recvmsg,
1524 .sendpage = udp_sendpage,
1525 .backlog_rcv = udp_queue_rcv_skb,
1526 .hash = udp_lib_hash,
1527 .unhash = udp_lib_unhash,
1528 .get_port = udp_v4_get_port,
1529 .obj_size = sizeof(struct udp_sock),
1530 #ifdef CONFIG_COMPAT
1531 .compat_setsockopt = compat_udp_setsockopt,
1532 .compat_getsockopt = compat_udp_getsockopt,
1536 /* ------------------------------------------------------------------------ */
1537 #ifdef CONFIG_PROC_FS
1539 static struct sock *udp_get_first(struct seq_file *seq)
1542 struct udp_iter_state *state = seq->private;
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 (sk->sk_family == state->family)
1556 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1558 struct udp_iter_state *state = seq->private;
1564 } while (sk && sk->sk_family != state->family);
1566 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1567 sk = sk_head(state->hashtable + state->bucket);
1573 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1575 struct sock *sk = udp_get_first(seq);
1578 while(pos && (sk = udp_get_next(seq, sk)) != NULL)
1580 return pos ? NULL : sk;
1583 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1585 read_lock(&udp_hash_lock);
1586 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1589 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1594 sk = udp_get_idx(seq, 0);
1596 sk = udp_get_next(seq, v);
1602 static void udp_seq_stop(struct seq_file *seq, void *v)
1604 read_unlock(&udp_hash_lock);
1607 static int udp_seq_open(struct inode *inode, struct file *file)
1609 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1610 struct seq_file *seq;
1612 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1616 s->family = afinfo->family;
1617 s->hashtable = afinfo->hashtable;
1618 s->seq_ops.start = udp_seq_start;
1619 s->seq_ops.next = udp_seq_next;
1620 s->seq_ops.show = afinfo->seq_show;
1621 s->seq_ops.stop = udp_seq_stop;
1623 rc = seq_open(file, &s->seq_ops);
1627 seq = file->private_data;
1636 /* ------------------------------------------------------------------------ */
1637 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1639 struct proc_dir_entry *p;
1644 afinfo->seq_fops->owner = afinfo->owner;
1645 afinfo->seq_fops->open = udp_seq_open;
1646 afinfo->seq_fops->read = seq_read;
1647 afinfo->seq_fops->llseek = seq_lseek;
1648 afinfo->seq_fops->release = seq_release_private;
1650 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1658 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1662 proc_net_remove(afinfo->name);
1663 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1666 /* ------------------------------------------------------------------------ */
1667 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1669 struct inet_sock *inet = inet_sk(sp);
1670 __be32 dest = inet->daddr;
1671 __be32 src = inet->rcv_saddr;
1672 __u16 destp = ntohs(inet->dport);
1673 __u16 srcp = ntohs(inet->sport);
1675 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1676 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1677 bucket, src, srcp, dest, destp, sp->sk_state,
1678 atomic_read(&sp->sk_wmem_alloc),
1679 atomic_read(&sp->sk_rmem_alloc),
1680 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1681 atomic_read(&sp->sk_refcnt), sp);
1684 int udp4_seq_show(struct seq_file *seq, void *v)
1686 if (v == SEQ_START_TOKEN)
1687 seq_printf(seq, "%-127s\n",
1688 " sl local_address rem_address st tx_queue "
1689 "rx_queue tr tm->when retrnsmt uid timeout "
1693 struct udp_iter_state *state = seq->private;
1695 udp4_format_sock(v, tmpbuf, state->bucket);
1696 seq_printf(seq, "%-127s\n", tmpbuf);
1701 /* ------------------------------------------------------------------------ */
1702 static struct file_operations udp4_seq_fops;
1703 static struct udp_seq_afinfo udp4_seq_afinfo = {
1704 .owner = THIS_MODULE,
1707 .hashtable = udp_hash,
1708 .seq_show = udp4_seq_show,
1709 .seq_fops = &udp4_seq_fops,
1712 int __init udp4_proc_init(void)
1714 return udp_proc_register(&udp4_seq_afinfo);
1717 void udp4_proc_exit(void)
1719 udp_proc_unregister(&udp4_seq_afinfo);
1721 #endif /* CONFIG_PROC_FS */
1723 EXPORT_SYMBOL(udp_disconnect);
1724 EXPORT_SYMBOL(udp_hash);
1725 EXPORT_SYMBOL(udp_hash_lock);
1726 EXPORT_SYMBOL(udp_ioctl);
1727 EXPORT_SYMBOL(udp_get_port);
1728 EXPORT_SYMBOL(udp_prot);
1729 EXPORT_SYMBOL(udp_sendmsg);
1730 EXPORT_SYMBOL(udp_lib_getsockopt);
1731 EXPORT_SYMBOL(udp_lib_setsockopt);
1732 EXPORT_SYMBOL(udp_poll);
1734 #ifdef CONFIG_PROC_FS
1735 EXPORT_SYMBOL(udp_proc_register);
1736 EXPORT_SYMBOL(udp_proc_unregister);