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 (inet_sk(sk)->num == 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) {
170 best_size_so_far = size;
175 for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result += UDP_HTABLE_SIZE) {
176 if (result > sysctl_local_port_range[1])
177 result = sysctl_local_port_range[0]
178 + ((result - sysctl_local_port_range[0]) &
179 (UDP_HTABLE_SIZE - 1));
180 if (! __udp_lib_lport_inuse(result, udptable))
183 if (i >= (1 << 16) / UDP_HTABLE_SIZE)
186 *port_rover = snum = result;
188 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
190 sk_for_each(sk2, node, head)
191 if (inet_sk(sk2)->num == snum &&
193 (!sk2->sk_reuse || !sk->sk_reuse) &&
194 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
195 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
196 (*saddr_comp)(sk, sk2) )
199 inet_sk(sk)->num = snum;
200 if (sk_unhashed(sk)) {
201 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
202 sk_add_node(sk, head);
203 sock_prot_inc_use(sk->sk_prot);
207 write_unlock_bh(&udp_hash_lock);
211 __inline__ int udp_get_port(struct sock *sk, unsigned short snum,
212 int (*scmp)(const struct sock *, const struct sock *))
214 return __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp);
217 inline int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
219 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
221 return ( !ipv6_only_sock(sk2) &&
222 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
223 inet1->rcv_saddr == inet2->rcv_saddr ));
226 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
228 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
231 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
232 * harder than this. -DaveM
234 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
235 __be32 daddr, __be16 dport,
236 int dif, struct hlist_head udptable[])
238 struct sock *sk, *result = NULL;
239 struct hlist_node *node;
240 unsigned short hnum = ntohs(dport);
243 read_lock(&udp_hash_lock);
244 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
245 struct inet_sock *inet = inet_sk(sk);
247 if (inet->num == hnum && !ipv6_only_sock(sk)) {
248 int score = (sk->sk_family == PF_INET ? 1 : 0);
249 if (inet->rcv_saddr) {
250 if (inet->rcv_saddr != daddr)
255 if (inet->daddr != saddr)
260 if (inet->dport != sport)
264 if (sk->sk_bound_dev_if) {
265 if (sk->sk_bound_dev_if != dif)
272 } else if(score > badness) {
280 read_unlock(&udp_hash_lock);
284 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
285 __be16 loc_port, __be32 loc_addr,
286 __be16 rmt_port, __be32 rmt_addr,
289 struct hlist_node *node;
291 unsigned short hnum = ntohs(loc_port);
293 sk_for_each_from(s, node) {
294 struct inet_sock *inet = inet_sk(s);
296 if (inet->num != hnum ||
297 (inet->daddr && inet->daddr != rmt_addr) ||
298 (inet->dport != rmt_port && inet->dport) ||
299 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
301 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
303 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
313 * This routine is called by the ICMP module when it gets some
314 * sort of error condition. If err < 0 then the socket should
315 * be closed and the error returned to the user. If err > 0
316 * it's just the icmp type << 8 | icmp code.
317 * Header points to the ip header of the error packet. We move
318 * on past this. Then (as it used to claim before adjustment)
319 * header points to the first 8 bytes of the udp header. We need
320 * to find the appropriate port.
323 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
325 struct inet_sock *inet;
326 struct iphdr *iph = (struct iphdr*)skb->data;
327 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
328 int type = skb->h.icmph->type;
329 int code = skb->h.icmph->code;
334 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
335 skb->dev->ifindex, udptable );
337 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
338 return; /* No socket for error */
347 case ICMP_TIME_EXCEEDED:
350 case ICMP_SOURCE_QUENCH:
352 case ICMP_PARAMETERPROB:
356 case ICMP_DEST_UNREACH:
357 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
358 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
366 if (code <= NR_ICMP_UNREACH) {
367 harderr = icmp_err_convert[code].fatal;
368 err = icmp_err_convert[code].errno;
374 * RFC1122: OK. Passes ICMP errors back to application, as per
377 if (!inet->recverr) {
378 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
381 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
384 sk->sk_error_report(sk);
389 __inline__ void udp_err(struct sk_buff *skb, u32 info)
391 return __udp4_lib_err(skb, info, udp_hash);
395 * Throw away all pending data and cancel the corking. Socket is locked.
397 static void udp_flush_pending_frames(struct sock *sk)
399 struct udp_sock *up = udp_sk(sk);
404 ip_flush_pending_frames(sk);
409 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
410 * @sk: socket we are sending on
411 * @skb: sk_buff containing the filled-in UDP header
412 * (checksum field must be zeroed out)
414 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
415 __be32 src, __be32 dst, int len )
418 struct udphdr *uh = skb->h.uh;
421 if (skb_queue_len(&sk->sk_write_queue) == 1) {
423 * Only one fragment on the socket.
425 skb->csum_offset = offsetof(struct udphdr, check);
426 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
429 * HW-checksum won't work as there are two or more
430 * fragments on the socket so that all csums of sk_buffs
433 offset = skb->h.raw - skb->data;
434 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
436 skb->ip_summed = CHECKSUM_NONE;
438 skb_queue_walk(&sk->sk_write_queue, skb) {
439 csum = csum_add(csum, skb->csum);
442 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
444 uh->check = CSUM_MANGLED_0;
449 * Push out all pending data as one UDP datagram. Socket is locked.
451 static int udp_push_pending_frames(struct sock *sk)
453 struct udp_sock *up = udp_sk(sk);
454 struct inet_sock *inet = inet_sk(sk);
455 struct flowi *fl = &inet->cork.fl;
461 /* Grab the skbuff where UDP header space exists. */
462 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
466 * Create a UDP header
469 uh->source = fl->fl_ip_sport;
470 uh->dest = fl->fl_ip_dport;
471 uh->len = htons(up->len);
474 if (up->pcflag) /* UDP-Lite */
475 csum = udplite_csum_outgoing(sk, skb);
477 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
479 skb->ip_summed = CHECKSUM_NONE;
482 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
484 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
487 } else /* `normal' UDP */
488 csum = udp_csum_outgoing(sk, skb);
490 /* add protocol-dependent pseudo-header */
491 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
492 sk->sk_protocol, csum );
494 uh->check = CSUM_MANGLED_0;
497 err = ip_push_pending_frames(sk);
504 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
507 struct inet_sock *inet = inet_sk(sk);
508 struct udp_sock *up = udp_sk(sk);
510 struct ipcm_cookie ipc;
511 struct rtable *rt = NULL;
514 __be32 daddr, faddr, saddr;
517 int err, is_udplite = up->pcflag;
518 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
519 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
528 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
535 * There are pending frames.
536 * The socket lock must be held while it's corked.
539 if (likely(up->pending)) {
540 if (unlikely(up->pending != AF_INET)) {
548 ulen += sizeof(struct udphdr);
551 * Get and verify the address.
554 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
555 if (msg->msg_namelen < sizeof(*usin))
557 if (usin->sin_family != AF_INET) {
558 if (usin->sin_family != AF_UNSPEC)
559 return -EAFNOSUPPORT;
562 daddr = usin->sin_addr.s_addr;
563 dport = usin->sin_port;
567 if (sk->sk_state != TCP_ESTABLISHED)
568 return -EDESTADDRREQ;
571 /* Open fast path for connected socket.
572 Route will not be used, if at least one option is set.
576 ipc.addr = inet->saddr;
578 ipc.oif = sk->sk_bound_dev_if;
579 if (msg->msg_controllen) {
580 err = ip_cmsg_send(msg, &ipc);
591 ipc.addr = faddr = daddr;
593 if (ipc.opt && ipc.opt->srr) {
596 faddr = ipc.opt->faddr;
599 tos = RT_TOS(inet->tos);
600 if (sock_flag(sk, SOCK_LOCALROUTE) ||
601 (msg->msg_flags & MSG_DONTROUTE) ||
602 (ipc.opt && ipc.opt->is_strictroute)) {
607 if (MULTICAST(daddr)) {
609 ipc.oif = inet->mc_index;
611 saddr = inet->mc_addr;
616 rt = (struct rtable*)sk_dst_check(sk, 0);
619 struct flowi fl = { .oif = ipc.oif,
624 .proto = sk->sk_protocol,
626 { .sport = inet->sport,
627 .dport = dport } } };
628 security_sk_classify_flow(sk, &fl);
629 err = ip_route_output_flow(&rt, &fl, sk, !(msg->msg_flags&MSG_DONTWAIT));
634 if ((rt->rt_flags & RTCF_BROADCAST) &&
635 !sock_flag(sk, SOCK_BROADCAST))
638 sk_dst_set(sk, dst_clone(&rt->u.dst));
641 if (msg->msg_flags&MSG_CONFIRM)
647 daddr = ipc.addr = rt->rt_dst;
650 if (unlikely(up->pending)) {
651 /* The socket is already corked while preparing it. */
652 /* ... which is an evident application bug. --ANK */
655 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
660 * Now cork the socket to pend data.
662 inet->cork.fl.fl4_dst = daddr;
663 inet->cork.fl.fl_ip_dport = dport;
664 inet->cork.fl.fl4_src = saddr;
665 inet->cork.fl.fl_ip_sport = inet->sport;
666 up->pending = AF_INET;
670 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
671 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
672 sizeof(struct udphdr), &ipc, rt,
673 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
675 udp_flush_pending_frames(sk);
677 err = udp_push_pending_frames(sk);
678 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
687 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
691 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
692 * ENOBUFS might not be good (it's not tunable per se), but otherwise
693 * we don't have a good statistic (IpOutDiscards but it can be too many
694 * things). We could add another new stat but at least for now that
695 * seems like overkill.
697 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
698 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
703 dst_confirm(&rt->u.dst);
704 if (!(msg->msg_flags&MSG_PROBE) || len)
705 goto back_from_confirm;
710 int udp_sendpage(struct sock *sk, struct page *page, int offset,
711 size_t size, int flags)
713 struct udp_sock *up = udp_sk(sk);
717 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
719 /* Call udp_sendmsg to specify destination address which
720 * sendpage interface can't pass.
721 * This will succeed only when the socket is connected.
723 ret = udp_sendmsg(NULL, sk, &msg, 0);
730 if (unlikely(!up->pending)) {
733 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
737 ret = ip_append_page(sk, page, offset, size, flags);
738 if (ret == -EOPNOTSUPP) {
740 return sock_no_sendpage(sk->sk_socket, page, offset,
744 udp_flush_pending_frames(sk);
749 if (!(up->corkflag || (flags&MSG_MORE)))
750 ret = udp_push_pending_frames(sk);
759 * IOCTL requests applicable to the UDP protocol
762 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
768 int amount = atomic_read(&sk->sk_wmem_alloc);
769 return put_user(amount, (int __user *)arg);
775 unsigned long amount;
778 spin_lock_bh(&sk->sk_receive_queue.lock);
779 skb = skb_peek(&sk->sk_receive_queue);
782 * We will only return the amount
783 * of this packet since that is all
786 amount = skb->len - sizeof(struct udphdr);
788 spin_unlock_bh(&sk->sk_receive_queue.lock);
789 return put_user(amount, (int __user *)arg);
799 * This should be easy, if there is something there we
800 * return it, otherwise we block.
803 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
804 size_t len, int noblock, int flags, int *addr_len)
806 struct inet_sock *inet = inet_sk(sk);
807 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
809 int copied, err, copy_only, is_udplite = IS_UDPLITE(sk);
812 * Check any passed addresses
815 *addr_len=sizeof(*sin);
817 if (flags & MSG_ERRQUEUE)
818 return ip_recv_error(sk, msg, len);
821 skb = skb_recv_datagram(sk, flags, noblock, &err);
825 copied = skb->len - sizeof(struct udphdr);
828 msg->msg_flags |= MSG_TRUNC;
832 * Decide whether to checksum and/or copy data.
834 * UDP: checksum may have been computed in HW,
835 * (re-)compute it if message is truncated.
836 * UDP-Lite: always needs to checksum, no HW support.
838 copy_only = (skb->ip_summed==CHECKSUM_UNNECESSARY);
840 if (is_udplite || (!copy_only && msg->msg_flags&MSG_TRUNC)) {
841 if (__udp_lib_checksum_complete(skb))
847 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
848 msg->msg_iov, copied );
850 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
859 sock_recv_timestamp(msg, sk, skb);
861 /* Copy the address. */
864 sin->sin_family = AF_INET;
865 sin->sin_port = skb->h.uh->source;
866 sin->sin_addr.s_addr = skb->nh.iph->saddr;
867 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
869 if (inet->cmsg_flags)
870 ip_cmsg_recv(msg, skb);
873 if (flags & MSG_TRUNC)
874 err = skb->len - sizeof(struct udphdr);
877 skb_free_datagram(sk, skb);
882 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
884 skb_kill_datagram(sk, skb, flags);
892 int udp_disconnect(struct sock *sk, int flags)
894 struct inet_sock *inet = inet_sk(sk);
896 * 1003.1g - break association.
899 sk->sk_state = TCP_CLOSE;
902 sk->sk_bound_dev_if = 0;
903 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
904 inet_reset_saddr(sk);
906 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
907 sk->sk_prot->unhash(sk);
915 * 1 if the the UDP system should process it
916 * 0 if we should drop this packet
917 * -1 if it should get processed by xfrm4_rcv_encap
919 static int udp_encap_rcv(struct sock * sk, struct sk_buff *skb)
924 struct udp_sock *up = udp_sk(sk);
931 __u16 encap_type = up->encap_type;
933 /* if we're overly short, let UDP handle it */
934 len = skb->len - sizeof(struct udphdr);
938 /* if this is not encapsulated socket, then just return now */
942 /* If this is a paged skb, make sure we pull up
943 * whatever data we need to look at. */
944 if (!pskb_may_pull(skb, sizeof(struct udphdr) + min(len, 8)))
947 /* Now we can get the pointers */
949 udpdata = (__u8 *)uh + sizeof(struct udphdr);
950 udpdata32 = (__be32 *)udpdata;
952 switch (encap_type) {
954 case UDP_ENCAP_ESPINUDP:
955 /* Check if this is a keepalive packet. If so, eat it. */
956 if (len == 1 && udpdata[0] == 0xff) {
958 } else if (len > sizeof(struct ip_esp_hdr) && udpdata32[0] != 0 ) {
959 /* ESP Packet without Non-ESP header */
960 len = sizeof(struct udphdr);
962 /* Must be an IKE packet.. pass it through */
965 case UDP_ENCAP_ESPINUDP_NON_IKE:
966 /* Check if this is a keepalive packet. If so, eat it. */
967 if (len == 1 && udpdata[0] == 0xff) {
969 } else if (len > 2 * sizeof(u32) + sizeof(struct ip_esp_hdr) &&
970 udpdata32[0] == 0 && udpdata32[1] == 0) {
972 /* ESP Packet with Non-IKE marker */
973 len = sizeof(struct udphdr) + 2 * sizeof(u32);
975 /* Must be an IKE packet.. pass it through */
980 /* At this point we are sure that this is an ESPinUDP packet,
981 * so we need to remove 'len' bytes from the packet (the UDP
982 * header and optional ESP marker bytes) and then modify the
983 * protocol to ESP, and then call into the transform receiver.
985 if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
988 /* Now we can update and verify the packet length... */
990 iphlen = iph->ihl << 2;
991 iph->tot_len = htons(ntohs(iph->tot_len) - len);
992 if (skb->len < iphlen + len) {
993 /* packet is too small!?! */
997 /* pull the data buffer up to the ESP header and set the
998 * transport header to point to ESP. Keep UDP on the stack
1001 skb->h.raw = skb_pull(skb, len);
1003 /* modify the protocol (it's ESP!) */
1004 iph->protocol = IPPROTO_ESP;
1006 /* and let the caller know to send this into the ESP processor... */
1014 * >0: "udp encap" protocol resubmission
1016 * Note that in the success and error cases, the skb is assumed to
1017 * have either been requeued or freed.
1019 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
1021 struct udp_sock *up = udp_sk(sk);
1025 * Charge it to the socket, dropping if the queue is full.
1027 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1031 if (up->encap_type) {
1033 * This is an encapsulation socket, so let's see if this is
1034 * an encapsulated packet.
1035 * If it's a keepalive packet, then just eat it.
1036 * If it's an encapsulateed packet, then pass it to the
1037 * IPsec xfrm input and return the response
1038 * appropriately. Otherwise, just fall through and
1039 * pass this up the UDP socket.
1043 ret = udp_encap_rcv(sk, skb);
1045 /* Eat the packet .. */
1050 /* process the ESP packet */
1051 ret = xfrm4_rcv_encap(skb, up->encap_type);
1052 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1055 /* FALLTHROUGH -- it's a UDP Packet */
1059 * UDP-Lite specific tests, ignored on UDP sockets
1061 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1064 * MIB statistics other than incrementing the error count are
1065 * disabled for the following two types of errors: these depend
1066 * on the application settings, not on the functioning of the
1067 * protocol stack as such.
1069 * RFC 3828 here recommends (sec 3.3): "There should also be a
1070 * way ... to ... at least let the receiving application block
1071 * delivery of packets with coverage values less than a value
1072 * provided by the application."
1074 if (up->pcrlen == 0) { /* full coverage was set */
1075 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1076 "%d while full coverage %d requested\n",
1077 UDP_SKB_CB(skb)->cscov, skb->len);
1080 /* The next case involves violating the min. coverage requested
1081 * by the receiver. This is subtle: if receiver wants x and x is
1082 * greater than the buffersize/MTU then receiver will complain
1083 * that it wants x while sender emits packets of smaller size y.
1084 * Therefore the above ...()->partial_cov statement is essential.
1086 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1087 LIMIT_NETDEBUG(KERN_WARNING
1088 "UDPLITE: coverage %d too small, need min %d\n",
1089 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1094 if (sk->sk_filter && skb->ip_summed != CHECKSUM_UNNECESSARY) {
1095 if (__udp_lib_checksum_complete(skb))
1097 skb->ip_summed = CHECKSUM_UNNECESSARY;
1100 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1101 /* Note that an ENOMEM error is charged twice */
1103 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag);
1107 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag);
1111 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag);
1117 * Multicasts and broadcasts go to each listener.
1119 * Note: called only from the BH handler context,
1120 * so we don't need to lock the hashes.
1122 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1124 __be32 saddr, __be32 daddr,
1125 struct hlist_head udptable[])
1130 read_lock(&udp_hash_lock);
1131 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1132 dif = skb->dev->ifindex;
1133 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1135 struct sock *sknext = NULL;
1138 struct sk_buff *skb1 = skb;
1140 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1141 uh->source, saddr, dif);
1143 skb1 = skb_clone(skb, GFP_ATOMIC);
1146 int ret = udp_queue_rcv_skb(sk, skb1);
1148 /* we should probably re-process instead
1149 * of dropping packets here. */
1156 read_unlock(&udp_hash_lock);
1160 /* Initialize UDP checksum. If exited with zero value (success),
1161 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1162 * Otherwise, csum completion requires chacksumming packet body,
1163 * including udp header and folding it to skb->csum.
1165 static inline void udp4_csum_init(struct sk_buff *skb, struct udphdr *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(skb->nh.iph->saddr, skb->nh.iph->daddr,
1171 skb->len, IPPROTO_UDP, skb->csum ))
1172 skb->ip_summed = CHECKSUM_UNNECESSARY;
1174 if (skb->ip_summed != CHECKSUM_UNNECESSARY)
1175 skb->csum = csum_tcpudp_nofold(skb->nh.iph->saddr,
1177 skb->len, IPPROTO_UDP, 0);
1178 /* Probably, we should checksum udp header (it should be in cache
1179 * in any case) and data in tiny packets (< rx copybreak).
1182 /* UDP = UDP-Lite with a non-partial checksum coverage */
1183 UDP_SKB_CB(skb)->partial_cov = 0;
1187 * All we need to do is get the socket, and then do a checksum.
1190 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1194 struct udphdr *uh = skb->h.uh;
1195 unsigned short ulen;
1196 struct rtable *rt = (struct rtable*)skb->dst;
1197 __be32 saddr = skb->nh.iph->saddr;
1198 __be32 daddr = skb->nh.iph->daddr;
1201 * Validate the packet.
1203 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1204 goto drop; /* No space for header. */
1206 ulen = ntohs(uh->len);
1207 if (ulen > skb->len)
1210 if(! is_udplite ) { /* UDP validates ulen. */
1212 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1215 udp4_csum_init(skb, uh);
1217 } else { /* UDP-Lite validates cscov. */
1218 if (udplite4_csum_init(skb, uh))
1222 if(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1223 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1225 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1226 skb->dev->ifindex, udptable );
1229 int ret = udp_queue_rcv_skb(sk, skb);
1232 /* a return value > 0 means to resubmit the input, but
1233 * it wants the return to be -protocol, or 0
1240 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1244 /* No socket. Drop packet silently, if checksum is wrong */
1245 if (udp_lib_checksum_complete(skb))
1248 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, is_udplite);
1249 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1252 * Hmm. We got an UDP packet to a port to which we
1253 * don't wanna listen. Ignore it.
1259 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1260 is_udplite? "-Lite" : "",
1271 * RFC1122: OK. Discards the bad packet silently (as far as
1272 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1274 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1275 is_udplite? "-Lite" : "",
1282 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1287 __inline__ int udp_rcv(struct sk_buff *skb)
1289 return __udp4_lib_rcv(skb, udp_hash, 0);
1292 int udp_destroy_sock(struct sock *sk)
1295 udp_flush_pending_frames(sk);
1301 * Socket option code for UDP
1303 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1304 char __user *optval, int optlen,
1305 int (*push_pending_frames)(struct sock *))
1307 struct udp_sock *up = udp_sk(sk);
1311 if(optlen<sizeof(int))
1314 if (get_user(val, (int __user *)optval))
1324 (*push_pending_frames)(sk);
1332 case UDP_ENCAP_ESPINUDP:
1333 case UDP_ENCAP_ESPINUDP_NON_IKE:
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 (!up->pcflag) /* Disable the option on UDP sockets */
1349 return -ENOPROTOOPT;
1350 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1353 up->pcflag |= UDPLITE_SEND_CC;
1356 /* The receiver specifies a minimum checksum coverage value. To make
1357 * sense, this should be set to at least 8 (as done below). If zero is
1358 * used, this again means full checksum coverage. */
1359 case UDPLITE_RECV_CSCOV:
1360 if (!up->pcflag) /* Disable the option on UDP sockets */
1361 return -ENOPROTOOPT;
1362 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1365 up->pcflag |= UDPLITE_RECV_CC;
1376 int udp_setsockopt(struct sock *sk, int level, int optname,
1377 char __user *optval, int optlen)
1379 if (level == SOL_UDP || level == SOL_UDPLITE)
1380 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1381 udp_push_pending_frames);
1382 return ip_setsockopt(sk, level, optname, optval, optlen);
1385 #ifdef CONFIG_COMPAT
1386 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1387 char __user *optval, int optlen)
1389 if (level == SOL_UDP || level == SOL_UDPLITE)
1390 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1391 udp_push_pending_frames);
1392 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1396 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1397 char __user *optval, int __user *optlen)
1399 struct udp_sock *up = udp_sk(sk);
1402 if(get_user(len,optlen))
1405 len = min_t(unsigned int, len, sizeof(int));
1416 val = up->encap_type;
1419 /* The following two cannot be changed on UDP sockets, the return is
1420 * always 0 (which corresponds to the full checksum coverage of UDP). */
1421 case UDPLITE_SEND_CSCOV:
1425 case UDPLITE_RECV_CSCOV:
1430 return -ENOPROTOOPT;
1433 if(put_user(len, optlen))
1435 if(copy_to_user(optval, &val,len))
1440 int udp_getsockopt(struct sock *sk, int level, int optname,
1441 char __user *optval, int __user *optlen)
1443 if (level == SOL_UDP || level == SOL_UDPLITE)
1444 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1445 return ip_getsockopt(sk, level, optname, optval, optlen);
1448 #ifdef CONFIG_COMPAT
1449 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1450 char __user *optval, int __user *optlen)
1452 if (level == SOL_UDP || level == SOL_UDPLITE)
1453 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1454 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1458 * udp_poll - wait for a UDP event.
1459 * @file - file struct
1461 * @wait - poll table
1463 * This is same as datagram poll, except for the special case of
1464 * blocking sockets. If application is using a blocking fd
1465 * and a packet with checksum error is in the queue;
1466 * then it could get return from select indicating data available
1467 * but then block when reading it. Add special case code
1468 * to work around these arguably broken applications.
1470 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1472 unsigned int mask = datagram_poll(file, sock, wait);
1473 struct sock *sk = sock->sk;
1474 int is_lite = IS_UDPLITE(sk);
1476 /* Check for false positives due to checksum errors */
1477 if ( (mask & POLLRDNORM) &&
1478 !(file->f_flags & O_NONBLOCK) &&
1479 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1480 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1481 struct sk_buff *skb;
1483 spin_lock_bh(&rcvq->lock);
1484 while ((skb = skb_peek(rcvq)) != NULL) {
1485 if (udp_lib_checksum_complete(skb)) {
1486 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1487 __skb_unlink(skb, rcvq);
1490 skb->ip_summed = CHECKSUM_UNNECESSARY;
1494 spin_unlock_bh(&rcvq->lock);
1496 /* nothing to see, move along */
1498 mask &= ~(POLLIN | POLLRDNORM);
1505 struct proto udp_prot = {
1507 .owner = THIS_MODULE,
1508 .close = udp_lib_close,
1509 .connect = ip4_datagram_connect,
1510 .disconnect = udp_disconnect,
1512 .destroy = udp_destroy_sock,
1513 .setsockopt = udp_setsockopt,
1514 .getsockopt = udp_getsockopt,
1515 .sendmsg = udp_sendmsg,
1516 .recvmsg = udp_recvmsg,
1517 .sendpage = udp_sendpage,
1518 .backlog_rcv = udp_queue_rcv_skb,
1519 .hash = udp_lib_hash,
1520 .unhash = udp_lib_unhash,
1521 .get_port = udp_v4_get_port,
1522 .obj_size = sizeof(struct udp_sock),
1523 #ifdef CONFIG_COMPAT
1524 .compat_setsockopt = compat_udp_setsockopt,
1525 .compat_getsockopt = compat_udp_getsockopt,
1529 /* ------------------------------------------------------------------------ */
1530 #ifdef CONFIG_PROC_FS
1532 static struct sock *udp_get_first(struct seq_file *seq)
1535 struct udp_iter_state *state = seq->private;
1537 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1538 struct hlist_node *node;
1539 sk_for_each(sk, node, state->hashtable + state->bucket) {
1540 if (sk->sk_family == state->family)
1549 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1551 struct udp_iter_state *state = seq->private;
1557 } while (sk && sk->sk_family != state->family);
1559 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1560 sk = sk_head(state->hashtable + state->bucket);
1566 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1568 struct sock *sk = udp_get_first(seq);
1571 while(pos && (sk = udp_get_next(seq, sk)) != NULL)
1573 return pos ? NULL : sk;
1576 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1578 read_lock(&udp_hash_lock);
1579 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1582 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1587 sk = udp_get_idx(seq, 0);
1589 sk = udp_get_next(seq, v);
1595 static void udp_seq_stop(struct seq_file *seq, void *v)
1597 read_unlock(&udp_hash_lock);
1600 static int udp_seq_open(struct inode *inode, struct file *file)
1602 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1603 struct seq_file *seq;
1605 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1609 s->family = afinfo->family;
1610 s->hashtable = afinfo->hashtable;
1611 s->seq_ops.start = udp_seq_start;
1612 s->seq_ops.next = udp_seq_next;
1613 s->seq_ops.show = afinfo->seq_show;
1614 s->seq_ops.stop = udp_seq_stop;
1616 rc = seq_open(file, &s->seq_ops);
1620 seq = file->private_data;
1629 /* ------------------------------------------------------------------------ */
1630 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1632 struct proc_dir_entry *p;
1637 afinfo->seq_fops->owner = afinfo->owner;
1638 afinfo->seq_fops->open = udp_seq_open;
1639 afinfo->seq_fops->read = seq_read;
1640 afinfo->seq_fops->llseek = seq_lseek;
1641 afinfo->seq_fops->release = seq_release_private;
1643 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops);
1651 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1655 proc_net_remove(afinfo->name);
1656 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1659 /* ------------------------------------------------------------------------ */
1660 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1662 struct inet_sock *inet = inet_sk(sp);
1663 __be32 dest = inet->daddr;
1664 __be32 src = inet->rcv_saddr;
1665 __u16 destp = ntohs(inet->dport);
1666 __u16 srcp = ntohs(inet->sport);
1668 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1669 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1670 bucket, src, srcp, dest, destp, sp->sk_state,
1671 atomic_read(&sp->sk_wmem_alloc),
1672 atomic_read(&sp->sk_rmem_alloc),
1673 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1674 atomic_read(&sp->sk_refcnt), sp);
1677 int udp4_seq_show(struct seq_file *seq, void *v)
1679 if (v == SEQ_START_TOKEN)
1680 seq_printf(seq, "%-127s\n",
1681 " sl local_address rem_address st tx_queue "
1682 "rx_queue tr tm->when retrnsmt uid timeout "
1686 struct udp_iter_state *state = seq->private;
1688 udp4_format_sock(v, tmpbuf, state->bucket);
1689 seq_printf(seq, "%-127s\n", tmpbuf);
1694 /* ------------------------------------------------------------------------ */
1695 static struct file_operations udp4_seq_fops;
1696 static struct udp_seq_afinfo udp4_seq_afinfo = {
1697 .owner = THIS_MODULE,
1700 .hashtable = udp_hash,
1701 .seq_show = udp4_seq_show,
1702 .seq_fops = &udp4_seq_fops,
1705 int __init udp4_proc_init(void)
1707 return udp_proc_register(&udp4_seq_afinfo);
1710 void udp4_proc_exit(void)
1712 udp_proc_unregister(&udp4_seq_afinfo);
1714 #endif /* CONFIG_PROC_FS */
1716 EXPORT_SYMBOL(udp_disconnect);
1717 EXPORT_SYMBOL(udp_hash);
1718 EXPORT_SYMBOL(udp_hash_lock);
1719 EXPORT_SYMBOL(udp_ioctl);
1720 EXPORT_SYMBOL(udp_get_port);
1721 EXPORT_SYMBOL(udp_prot);
1722 EXPORT_SYMBOL(udp_sendmsg);
1723 EXPORT_SYMBOL(udp_lib_getsockopt);
1724 EXPORT_SYMBOL(udp_lib_setsockopt);
1725 EXPORT_SYMBOL(udp_poll);
1727 #ifdef CONFIG_PROC_FS
1728 EXPORT_SYMBOL(udp_proc_register);
1729 EXPORT_SYMBOL(udp_proc_unregister);