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;
113 EXPORT_SYMBOL(udp_statistics);
115 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
116 EXPORT_SYMBOL(udp_stats_in6);
118 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
119 DEFINE_RWLOCK(udp_hash_lock);
121 static inline int __udp_lib_lport_inuse(__u16 num,
122 const struct hlist_head udptable[])
125 struct hlist_node *node;
127 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)])
128 if (sk->sk_hash == num)
134 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
136 * @sk: socket struct in question
137 * @snum: port number to look up
138 * @udptable: hash list table, must be of UDP_HTABLE_SIZE
139 * @saddr_comp: AF-dependent comparison of bound local IP addresses
141 int __udp_lib_get_port(struct sock *sk, unsigned short snum,
142 struct hlist_head udptable[],
143 int (*saddr_comp)(const struct sock *sk1,
144 const struct sock *sk2 ) )
146 struct hlist_node *node;
147 struct hlist_head *head;
151 write_lock_bh(&udp_hash_lock);
154 int i, low, high, remaining;
155 unsigned rover, best, best_size_so_far;
157 inet_get_local_port_range(&low, &high);
158 remaining = (high - low) + 1;
160 best_size_so_far = UINT_MAX;
161 best = rover = net_random() % remaining + low;
163 /* 1st pass: look for empty (or shortest) hash chain */
164 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
167 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)];
168 if (hlist_empty(head))
171 sk_for_each(sk2, node, head) {
172 if (++size >= best_size_so_far)
175 best_size_so_far = size;
178 /* fold back if end of range */
180 rover = low + ((rover - low)
181 & (UDP_HTABLE_SIZE - 1));
186 /* 2nd pass: find hole in shortest hash chain */
188 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
189 if (! __udp_lib_lport_inuse(rover, udptable))
191 rover += UDP_HTABLE_SIZE;
193 rover = low + ((rover - low)
194 & (UDP_HTABLE_SIZE - 1));
198 /* All ports in use! */
204 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
206 sk_for_each(sk2, node, head)
207 if (sk2->sk_hash == snum &&
209 (!sk2->sk_reuse || !sk->sk_reuse) &&
210 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
211 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
212 (*saddr_comp)(sk, sk2) )
216 inet_sk(sk)->num = snum;
218 if (sk_unhashed(sk)) {
219 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)];
220 sk_add_node(sk, head);
221 sock_prot_inc_use(sk->sk_prot);
225 write_unlock_bh(&udp_hash_lock);
229 int udp_get_port(struct sock *sk, unsigned short snum,
230 int (*scmp)(const struct sock *, const struct sock *))
232 return __udp_lib_get_port(sk, snum, udp_hash, scmp);
235 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
237 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
239 return ( !ipv6_only_sock(sk2) &&
240 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
241 inet1->rcv_saddr == inet2->rcv_saddr ));
244 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum)
246 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal);
249 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
250 * harder than this. -DaveM
252 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport,
253 __be32 daddr, __be16 dport,
254 int dif, struct hlist_head udptable[])
256 struct sock *sk, *result = NULL;
257 struct hlist_node *node;
258 unsigned short hnum = ntohs(dport);
261 read_lock(&udp_hash_lock);
262 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) {
263 struct inet_sock *inet = inet_sk(sk);
265 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) {
266 int score = (sk->sk_family == PF_INET ? 1 : 0);
267 if (inet->rcv_saddr) {
268 if (inet->rcv_saddr != daddr)
273 if (inet->daddr != saddr)
278 if (inet->dport != sport)
282 if (sk->sk_bound_dev_if) {
283 if (sk->sk_bound_dev_if != dif)
290 } else if (score > badness) {
298 read_unlock(&udp_hash_lock);
302 static inline struct sock *udp_v4_mcast_next(struct sock *sk,
303 __be16 loc_port, __be32 loc_addr,
304 __be16 rmt_port, __be32 rmt_addr,
307 struct hlist_node *node;
309 unsigned short hnum = ntohs(loc_port);
311 sk_for_each_from(s, node) {
312 struct inet_sock *inet = inet_sk(s);
314 if (s->sk_hash != hnum ||
315 (inet->daddr && inet->daddr != rmt_addr) ||
316 (inet->dport != rmt_port && inet->dport) ||
317 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
319 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
321 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
331 * This routine is called by the ICMP module when it gets some
332 * sort of error condition. If err < 0 then the socket should
333 * be closed and the error returned to the user. If err > 0
334 * it's just the icmp type << 8 | icmp code.
335 * Header points to the ip header of the error packet. We move
336 * on past this. Then (as it used to claim before adjustment)
337 * header points to the first 8 bytes of the udp header. We need
338 * to find the appropriate port.
341 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
343 struct inet_sock *inet;
344 struct iphdr *iph = (struct iphdr*)skb->data;
345 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
346 const int type = icmp_hdr(skb)->type;
347 const int code = icmp_hdr(skb)->code;
352 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source,
353 skb->dev->ifindex, udptable );
355 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS);
356 return; /* No socket for error */
365 case ICMP_TIME_EXCEEDED:
368 case ICMP_SOURCE_QUENCH:
370 case ICMP_PARAMETERPROB:
374 case ICMP_DEST_UNREACH:
375 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
376 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
384 if (code <= NR_ICMP_UNREACH) {
385 harderr = icmp_err_convert[code].fatal;
386 err = icmp_err_convert[code].errno;
392 * RFC1122: OK. Passes ICMP errors back to application, as per
395 if (!inet->recverr) {
396 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
399 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
402 sk->sk_error_report(sk);
407 void udp_err(struct sk_buff *skb, u32 info)
409 return __udp4_lib_err(skb, info, udp_hash);
413 * Throw away all pending data and cancel the corking. Socket is locked.
415 static void udp_flush_pending_frames(struct sock *sk)
417 struct udp_sock *up = udp_sk(sk);
422 ip_flush_pending_frames(sk);
427 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
428 * @sk: socket we are sending on
429 * @skb: sk_buff containing the filled-in UDP header
430 * (checksum field must be zeroed out)
432 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
433 __be32 src, __be32 dst, int len )
436 struct udphdr *uh = udp_hdr(skb);
439 if (skb_queue_len(&sk->sk_write_queue) == 1) {
441 * Only one fragment on the socket.
443 skb->csum_start = skb_transport_header(skb) - skb->head;
444 skb->csum_offset = offsetof(struct udphdr, check);
445 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
448 * HW-checksum won't work as there are two or more
449 * fragments on the socket so that all csums of sk_buffs
452 offset = skb_transport_offset(skb);
453 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
455 skb->ip_summed = CHECKSUM_NONE;
457 skb_queue_walk(&sk->sk_write_queue, skb) {
458 csum = csum_add(csum, skb->csum);
461 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
463 uh->check = CSUM_MANGLED_0;
468 * Push out all pending data as one UDP datagram. Socket is locked.
470 static int udp_push_pending_frames(struct sock *sk)
472 struct udp_sock *up = udp_sk(sk);
473 struct inet_sock *inet = inet_sk(sk);
474 struct flowi *fl = &inet->cork.fl;
478 int is_udplite = IS_UDPLITE(sk);
481 /* Grab the skbuff where UDP header space exists. */
482 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
486 * Create a UDP header
489 uh->source = fl->fl_ip_sport;
490 uh->dest = fl->fl_ip_dport;
491 uh->len = htons(up->len);
494 if (is_udplite) /* UDP-Lite */
495 csum = udplite_csum_outgoing(sk, skb);
497 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
499 skb->ip_summed = CHECKSUM_NONE;
502 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
504 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
507 } else /* `normal' UDP */
508 csum = udp_csum_outgoing(sk, skb);
510 /* add protocol-dependent pseudo-header */
511 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
512 sk->sk_protocol, csum );
514 uh->check = CSUM_MANGLED_0;
517 err = ip_push_pending_frames(sk);
522 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite);
526 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
529 struct inet_sock *inet = inet_sk(sk);
530 struct udp_sock *up = udp_sk(sk);
532 struct ipcm_cookie ipc;
533 struct rtable *rt = NULL;
536 __be32 daddr, faddr, saddr;
539 int err, is_udplite = IS_UDPLITE(sk);
540 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
541 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
550 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
557 * There are pending frames.
558 * The socket lock must be held while it's corked.
561 if (likely(up->pending)) {
562 if (unlikely(up->pending != AF_INET)) {
570 ulen += sizeof(struct udphdr);
573 * Get and verify the address.
576 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
577 if (msg->msg_namelen < sizeof(*usin))
579 if (usin->sin_family != AF_INET) {
580 if (usin->sin_family != AF_UNSPEC)
581 return -EAFNOSUPPORT;
584 daddr = usin->sin_addr.s_addr;
585 dport = usin->sin_port;
589 if (sk->sk_state != TCP_ESTABLISHED)
590 return -EDESTADDRREQ;
593 /* Open fast path for connected socket.
594 Route will not be used, if at least one option is set.
598 ipc.addr = inet->saddr;
600 ipc.oif = sk->sk_bound_dev_if;
601 if (msg->msg_controllen) {
602 err = ip_cmsg_send(msg, &ipc);
613 ipc.addr = faddr = daddr;
615 if (ipc.opt && ipc.opt->srr) {
618 faddr = ipc.opt->faddr;
621 tos = RT_TOS(inet->tos);
622 if (sock_flag(sk, SOCK_LOCALROUTE) ||
623 (msg->msg_flags & MSG_DONTROUTE) ||
624 (ipc.opt && ipc.opt->is_strictroute)) {
629 if (ipv4_is_multicast(daddr)) {
631 ipc.oif = inet->mc_index;
633 saddr = inet->mc_addr;
638 rt = (struct rtable*)sk_dst_check(sk, 0);
641 struct flowi fl = { .oif = ipc.oif,
646 .proto = sk->sk_protocol,
648 { .sport = inet->sport,
649 .dport = dport } } };
650 security_sk_classify_flow(sk, &fl);
651 err = ip_route_output_flow(&rt, &fl, sk, 1);
653 if (err == -ENETUNREACH)
654 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES);
659 if ((rt->rt_flags & RTCF_BROADCAST) &&
660 !sock_flag(sk, SOCK_BROADCAST))
663 sk_dst_set(sk, dst_clone(&rt->u.dst));
666 if (msg->msg_flags&MSG_CONFIRM)
672 daddr = ipc.addr = rt->rt_dst;
675 if (unlikely(up->pending)) {
676 /* The socket is already corked while preparing it. */
677 /* ... which is an evident application bug. --ANK */
680 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
685 * Now cork the socket to pend data.
687 inet->cork.fl.fl4_dst = daddr;
688 inet->cork.fl.fl_ip_dport = dport;
689 inet->cork.fl.fl4_src = saddr;
690 inet->cork.fl.fl_ip_sport = inet->sport;
691 up->pending = AF_INET;
695 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
696 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
697 sizeof(struct udphdr), &ipc, rt,
698 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
700 udp_flush_pending_frames(sk);
702 err = udp_push_pending_frames(sk);
703 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
714 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
715 * ENOBUFS might not be good (it's not tunable per se), but otherwise
716 * we don't have a good statistic (IpOutDiscards but it can be too many
717 * things). We could add another new stat but at least for now that
718 * seems like overkill.
720 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
721 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite);
726 dst_confirm(&rt->u.dst);
727 if (!(msg->msg_flags&MSG_PROBE) || len)
728 goto back_from_confirm;
733 int udp_sendpage(struct sock *sk, struct page *page, int offset,
734 size_t size, int flags)
736 struct udp_sock *up = udp_sk(sk);
740 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
742 /* Call udp_sendmsg to specify destination address which
743 * sendpage interface can't pass.
744 * This will succeed only when the socket is connected.
746 ret = udp_sendmsg(NULL, sk, &msg, 0);
753 if (unlikely(!up->pending)) {
756 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
760 ret = ip_append_page(sk, page, offset, size, flags);
761 if (ret == -EOPNOTSUPP) {
763 return sock_no_sendpage(sk->sk_socket, page, offset,
767 udp_flush_pending_frames(sk);
772 if (!(up->corkflag || (flags&MSG_MORE)))
773 ret = udp_push_pending_frames(sk);
782 * IOCTL requests applicable to the UDP protocol
785 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
790 int amount = atomic_read(&sk->sk_wmem_alloc);
791 return put_user(amount, (int __user *)arg);
797 unsigned long amount;
800 spin_lock_bh(&sk->sk_receive_queue.lock);
801 skb = skb_peek(&sk->sk_receive_queue);
804 * We will only return the amount
805 * of this packet since that is all
808 amount = skb->len - sizeof(struct udphdr);
810 spin_unlock_bh(&sk->sk_receive_queue.lock);
811 return put_user(amount, (int __user *)arg);
822 * This should be easy, if there is something there we
823 * return it, otherwise we block.
826 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
827 size_t len, int noblock, int flags, int *addr_len)
829 struct inet_sock *inet = inet_sk(sk);
830 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
832 unsigned int ulen, copied;
835 int is_udplite = IS_UDPLITE(sk);
838 * Check any passed addresses
841 *addr_len=sizeof(*sin);
843 if (flags & MSG_ERRQUEUE)
844 return ip_recv_error(sk, msg, len);
847 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
852 ulen = skb->len - sizeof(struct udphdr);
856 else if (copied < ulen)
857 msg->msg_flags |= MSG_TRUNC;
860 * If checksum is needed at all, try to do it while copying the
861 * data. If the data is truncated, or if we only want a partial
862 * coverage checksum (UDP-Lite), do it before the copy.
865 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
866 if (udp_lib_checksum_complete(skb))
870 if (skb_csum_unnecessary(skb))
871 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
872 msg->msg_iov, copied );
874 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
884 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite);
886 sock_recv_timestamp(msg, sk, skb);
888 /* Copy the address. */
891 sin->sin_family = AF_INET;
892 sin->sin_port = udp_hdr(skb)->source;
893 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
894 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
896 if (inet->cmsg_flags)
897 ip_cmsg_recv(msg, skb);
900 if (flags & MSG_TRUNC)
904 skb_free_datagram(sk, skb);
909 if (!skb_kill_datagram(sk, skb, flags))
910 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite);
918 int udp_disconnect(struct sock *sk, int flags)
920 struct inet_sock *inet = inet_sk(sk);
922 * 1003.1g - break association.
925 sk->sk_state = TCP_CLOSE;
928 sk->sk_bound_dev_if = 0;
929 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
930 inet_reset_saddr(sk);
932 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
933 sk->sk_prot->unhash(sk);
943 * >0: "udp encap" protocol resubmission
945 * Note that in the success and error cases, the skb is assumed to
946 * have either been requeued or freed.
948 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
950 struct udp_sock *up = udp_sk(sk);
952 int is_udplite = IS_UDPLITE(sk);
955 * Charge it to the socket, dropping if the queue is full.
957 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
961 if (up->encap_type) {
963 * This is an encapsulation socket so pass the skb to
964 * the socket's udp_encap_rcv() hook. Otherwise, just
965 * fall through and pass this up the UDP socket.
966 * up->encap_rcv() returns the following value:
967 * =0 if skb was successfully passed to the encap
968 * handler or was discarded by it.
969 * >0 if skb should be passed on to UDP.
970 * <0 if skb should be resubmitted as proto -N
973 /* if we're overly short, let UDP handle it */
974 if (skb->len > sizeof(struct udphdr) &&
975 up->encap_rcv != NULL) {
978 ret = (*up->encap_rcv)(sk, skb);
980 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS,
986 /* FALLTHROUGH -- it's a UDP Packet */
990 * UDP-Lite specific tests, ignored on UDP sockets
992 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
995 * MIB statistics other than incrementing the error count are
996 * disabled for the following two types of errors: these depend
997 * on the application settings, not on the functioning of the
998 * protocol stack as such.
1000 * RFC 3828 here recommends (sec 3.3): "There should also be a
1001 * way ... to ... at least let the receiving application block
1002 * delivery of packets with coverage values less than a value
1003 * provided by the application."
1005 if (up->pcrlen == 0) { /* full coverage was set */
1006 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1007 "%d while full coverage %d requested\n",
1008 UDP_SKB_CB(skb)->cscov, skb->len);
1011 /* The next case involves violating the min. coverage requested
1012 * by the receiver. This is subtle: if receiver wants x and x is
1013 * greater than the buffersize/MTU then receiver will complain
1014 * that it wants x while sender emits packets of smaller size y.
1015 * Therefore the above ...()->partial_cov statement is essential.
1017 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1018 LIMIT_NETDEBUG(KERN_WARNING
1019 "UDPLITE: coverage %d too small, need min %d\n",
1020 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1025 if (sk->sk_filter) {
1026 if (udp_lib_checksum_complete(skb))
1030 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) {
1031 /* Note that an ENOMEM error is charged twice */
1033 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite);
1040 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite);
1046 * Multicasts and broadcasts go to each listener.
1048 * Note: called only from the BH handler context,
1049 * so we don't need to lock the hashes.
1051 static int __udp4_lib_mcast_deliver(struct sk_buff *skb,
1053 __be32 saddr, __be32 daddr,
1054 struct hlist_head udptable[])
1059 read_lock(&udp_hash_lock);
1060 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]);
1061 dif = skb->dev->ifindex;
1062 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
1064 struct sock *sknext = NULL;
1067 struct sk_buff *skb1 = skb;
1069 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr,
1070 uh->source, saddr, dif);
1072 skb1 = skb_clone(skb, GFP_ATOMIC);
1075 int ret = udp_queue_rcv_skb(sk, skb1);
1077 /* we should probably re-process instead
1078 * of dropping packets here. */
1085 read_unlock(&udp_hash_lock);
1089 /* Initialize UDP checksum. If exited with zero value (success),
1090 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1091 * Otherwise, csum completion requires chacksumming packet body,
1092 * including udp header and folding it to skb->csum.
1094 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1097 const struct iphdr *iph;
1100 UDP_SKB_CB(skb)->partial_cov = 0;
1101 UDP_SKB_CB(skb)->cscov = skb->len;
1103 if (proto == IPPROTO_UDPLITE) {
1104 err = udplite_checksum_init(skb, uh);
1110 if (uh->check == 0) {
1111 skb->ip_summed = CHECKSUM_UNNECESSARY;
1112 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1113 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1115 skb->ip_summed = CHECKSUM_UNNECESSARY;
1117 if (!skb_csum_unnecessary(skb))
1118 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1119 skb->len, proto, 0);
1120 /* Probably, we should checksum udp header (it should be in cache
1121 * in any case) and data in tiny packets (< rx copybreak).
1128 * All we need to do is get the socket, and then do a checksum.
1131 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1135 struct udphdr *uh = udp_hdr(skb);
1136 unsigned short ulen;
1137 struct rtable *rt = (struct rtable*)skb->dst;
1138 __be32 saddr = ip_hdr(skb)->saddr;
1139 __be32 daddr = ip_hdr(skb)->daddr;
1142 * Validate the packet.
1144 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1145 goto drop; /* No space for header. */
1147 ulen = ntohs(uh->len);
1148 if (ulen > skb->len)
1151 if (proto == IPPROTO_UDP) {
1152 /* UDP validates ulen. */
1153 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1158 if (udp4_csum_init(skb, uh, proto))
1161 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1162 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable);
1164 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest,
1165 inet_iif(skb), udptable);
1168 int ret = udp_queue_rcv_skb(sk, skb);
1171 /* a return value > 0 means to resubmit the input, but
1172 * it wants the return to be -protocol, or 0
1179 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1183 /* No socket. Drop packet silently, if checksum is wrong */
1184 if (udp_lib_checksum_complete(skb))
1187 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1188 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1191 * Hmm. We got an UDP packet to a port to which we
1192 * don't wanna listen. Ignore it.
1198 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1199 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1210 * RFC1122: OK. Discards the bad packet silently (as far as
1211 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1213 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1214 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1221 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1226 int udp_rcv(struct sk_buff *skb)
1228 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1231 int udp_destroy_sock(struct sock *sk)
1234 udp_flush_pending_frames(sk);
1240 * Socket option code for UDP
1242 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1243 char __user *optval, int optlen,
1244 int (*push_pending_frames)(struct sock *))
1246 struct udp_sock *up = udp_sk(sk);
1249 int is_udplite = IS_UDPLITE(sk);
1251 if (optlen<sizeof(int))
1254 if (get_user(val, (int __user *)optval))
1264 (*push_pending_frames)(sk);
1272 case UDP_ENCAP_ESPINUDP:
1273 case UDP_ENCAP_ESPINUDP_NON_IKE:
1274 up->encap_rcv = xfrm4_udp_encap_rcv;
1276 case UDP_ENCAP_L2TPINUDP:
1277 up->encap_type = val;
1286 * UDP-Lite's partial checksum coverage (RFC 3828).
1288 /* The sender sets actual checksum coverage length via this option.
1289 * The case coverage > packet length is handled by send module. */
1290 case UDPLITE_SEND_CSCOV:
1291 if (!is_udplite) /* Disable the option on UDP sockets */
1292 return -ENOPROTOOPT;
1293 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1296 up->pcflag |= UDPLITE_SEND_CC;
1299 /* The receiver specifies a minimum checksum coverage value. To make
1300 * sense, this should be set to at least 8 (as done below). If zero is
1301 * used, this again means full checksum coverage. */
1302 case UDPLITE_RECV_CSCOV:
1303 if (!is_udplite) /* Disable the option on UDP sockets */
1304 return -ENOPROTOOPT;
1305 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1308 up->pcflag |= UDPLITE_RECV_CC;
1319 int udp_setsockopt(struct sock *sk, int level, int optname,
1320 char __user *optval, int optlen)
1322 if (level == SOL_UDP || level == SOL_UDPLITE)
1323 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1324 udp_push_pending_frames);
1325 return ip_setsockopt(sk, level, optname, optval, optlen);
1328 #ifdef CONFIG_COMPAT
1329 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1330 char __user *optval, int optlen)
1332 if (level == SOL_UDP || level == SOL_UDPLITE)
1333 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1334 udp_push_pending_frames);
1335 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1339 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1340 char __user *optval, int __user *optlen)
1342 struct udp_sock *up = udp_sk(sk);
1345 if (get_user(len,optlen))
1348 len = min_t(unsigned int, len, sizeof(int));
1359 val = up->encap_type;
1362 /* The following two cannot be changed on UDP sockets, the return is
1363 * always 0 (which corresponds to the full checksum coverage of UDP). */
1364 case UDPLITE_SEND_CSCOV:
1368 case UDPLITE_RECV_CSCOV:
1373 return -ENOPROTOOPT;
1376 if (put_user(len, optlen))
1378 if (copy_to_user(optval, &val,len))
1383 int udp_getsockopt(struct sock *sk, int level, int optname,
1384 char __user *optval, int __user *optlen)
1386 if (level == SOL_UDP || level == SOL_UDPLITE)
1387 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1388 return ip_getsockopt(sk, level, optname, optval, optlen);
1391 #ifdef CONFIG_COMPAT
1392 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1393 char __user *optval, int __user *optlen)
1395 if (level == SOL_UDP || level == SOL_UDPLITE)
1396 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1397 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1401 * udp_poll - wait for a UDP event.
1402 * @file - file struct
1404 * @wait - poll table
1406 * This is same as datagram poll, except for the special case of
1407 * blocking sockets. If application is using a blocking fd
1408 * and a packet with checksum error is in the queue;
1409 * then it could get return from select indicating data available
1410 * but then block when reading it. Add special case code
1411 * to work around these arguably broken applications.
1413 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1415 unsigned int mask = datagram_poll(file, sock, wait);
1416 struct sock *sk = sock->sk;
1417 int is_lite = IS_UDPLITE(sk);
1419 /* Check for false positives due to checksum errors */
1420 if ( (mask & POLLRDNORM) &&
1421 !(file->f_flags & O_NONBLOCK) &&
1422 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1423 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1424 struct sk_buff *skb;
1426 spin_lock_bh(&rcvq->lock);
1427 while ((skb = skb_peek(rcvq)) != NULL &&
1428 udp_lib_checksum_complete(skb)) {
1429 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite);
1430 __skb_unlink(skb, rcvq);
1433 spin_unlock_bh(&rcvq->lock);
1435 /* nothing to see, move along */
1437 mask &= ~(POLLIN | POLLRDNORM);
1444 DEFINE_PROTO_INUSE(udp)
1446 struct proto udp_prot = {
1448 .owner = THIS_MODULE,
1449 .close = udp_lib_close,
1450 .connect = ip4_datagram_connect,
1451 .disconnect = udp_disconnect,
1453 .destroy = udp_destroy_sock,
1454 .setsockopt = udp_setsockopt,
1455 .getsockopt = udp_getsockopt,
1456 .sendmsg = udp_sendmsg,
1457 .recvmsg = udp_recvmsg,
1458 .sendpage = udp_sendpage,
1459 .backlog_rcv = udp_queue_rcv_skb,
1460 .hash = udp_lib_hash,
1461 .unhash = udp_lib_unhash,
1462 .get_port = udp_v4_get_port,
1463 .obj_size = sizeof(struct udp_sock),
1464 #ifdef CONFIG_COMPAT
1465 .compat_setsockopt = compat_udp_setsockopt,
1466 .compat_getsockopt = compat_udp_getsockopt,
1468 REF_PROTO_INUSE(udp)
1471 /* ------------------------------------------------------------------------ */
1472 #ifdef CONFIG_PROC_FS
1474 static struct sock *udp_get_first(struct seq_file *seq)
1477 struct udp_iter_state *state = seq->private;
1479 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1480 struct hlist_node *node;
1481 sk_for_each(sk, node, state->hashtable + state->bucket) {
1482 if (sk->sk_family == state->family)
1491 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1493 struct udp_iter_state *state = seq->private;
1499 } while (sk && sk->sk_family != state->family);
1501 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1502 sk = sk_head(state->hashtable + state->bucket);
1508 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1510 struct sock *sk = udp_get_first(seq);
1513 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1515 return pos ? NULL : sk;
1518 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1520 read_lock(&udp_hash_lock);
1521 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1;
1524 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1529 sk = udp_get_idx(seq, 0);
1531 sk = udp_get_next(seq, v);
1537 static void udp_seq_stop(struct seq_file *seq, void *v)
1539 read_unlock(&udp_hash_lock);
1542 static int udp_seq_open(struct inode *inode, struct file *file)
1544 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1545 struct seq_file *seq;
1547 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL);
1551 s->family = afinfo->family;
1552 s->hashtable = afinfo->hashtable;
1553 s->seq_ops.start = udp_seq_start;
1554 s->seq_ops.next = udp_seq_next;
1555 s->seq_ops.show = afinfo->seq_show;
1556 s->seq_ops.stop = udp_seq_stop;
1558 rc = seq_open(file, &s->seq_ops);
1562 seq = file->private_data;
1571 /* ------------------------------------------------------------------------ */
1572 int udp_proc_register(struct udp_seq_afinfo *afinfo)
1574 struct proc_dir_entry *p;
1579 afinfo->seq_fops->owner = afinfo->owner;
1580 afinfo->seq_fops->open = udp_seq_open;
1581 afinfo->seq_fops->read = seq_read;
1582 afinfo->seq_fops->llseek = seq_lseek;
1583 afinfo->seq_fops->release = seq_release_private;
1585 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops);
1593 void udp_proc_unregister(struct udp_seq_afinfo *afinfo)
1597 proc_net_remove(&init_net, afinfo->name);
1598 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops));
1601 /* ------------------------------------------------------------------------ */
1602 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket)
1604 struct inet_sock *inet = inet_sk(sp);
1605 __be32 dest = inet->daddr;
1606 __be32 src = inet->rcv_saddr;
1607 __u16 destp = ntohs(inet->dport);
1608 __u16 srcp = ntohs(inet->sport);
1610 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1611 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1612 bucket, src, srcp, dest, destp, sp->sk_state,
1613 atomic_read(&sp->sk_wmem_alloc),
1614 atomic_read(&sp->sk_rmem_alloc),
1615 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1616 atomic_read(&sp->sk_refcnt), sp);
1619 int udp4_seq_show(struct seq_file *seq, void *v)
1621 if (v == SEQ_START_TOKEN)
1622 seq_printf(seq, "%-127s\n",
1623 " sl local_address rem_address st tx_queue "
1624 "rx_queue tr tm->when retrnsmt uid timeout "
1628 struct udp_iter_state *state = seq->private;
1630 udp4_format_sock(v, tmpbuf, state->bucket);
1631 seq_printf(seq, "%-127s\n", tmpbuf);
1636 /* ------------------------------------------------------------------------ */
1637 static struct file_operations udp4_seq_fops;
1638 static struct udp_seq_afinfo udp4_seq_afinfo = {
1639 .owner = THIS_MODULE,
1642 .hashtable = udp_hash,
1643 .seq_show = udp4_seq_show,
1644 .seq_fops = &udp4_seq_fops,
1647 int __init udp4_proc_init(void)
1649 return udp_proc_register(&udp4_seq_afinfo);
1652 void udp4_proc_exit(void)
1654 udp_proc_unregister(&udp4_seq_afinfo);
1656 #endif /* CONFIG_PROC_FS */
1658 EXPORT_SYMBOL(udp_disconnect);
1659 EXPORT_SYMBOL(udp_hash);
1660 EXPORT_SYMBOL(udp_hash_lock);
1661 EXPORT_SYMBOL(udp_ioctl);
1662 EXPORT_SYMBOL(udp_get_port);
1663 EXPORT_SYMBOL(udp_prot);
1664 EXPORT_SYMBOL(udp_sendmsg);
1665 EXPORT_SYMBOL(udp_lib_getsockopt);
1666 EXPORT_SYMBOL(udp_lib_setsockopt);
1667 EXPORT_SYMBOL(udp_poll);
1669 #ifdef CONFIG_PROC_FS
1670 EXPORT_SYMBOL(udp_proc_register);
1671 EXPORT_SYMBOL(udp_proc_unregister);