2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * The User Datagram Protocol (UDP).
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no>
11 * Alan Cox, <Alan.Cox@linux.org>
12 * Hirokazu Takahashi, <taka@valinux.co.jp>
15 * Alan Cox : verify_area() calls
16 * Alan Cox : stopped close while in use off icmp
17 * messages. Not a fix but a botch that
18 * for udp at least is 'valid'.
19 * Alan Cox : Fixed icmp handling properly
20 * Alan Cox : Correct error for oversized datagrams
21 * Alan Cox : Tidied select() semantics.
22 * Alan Cox : udp_err() fixed properly, also now
23 * select and read wake correctly on errors
24 * Alan Cox : udp_send verify_area moved to avoid mem leak
25 * Alan Cox : UDP can count its memory
26 * Alan Cox : send to an unknown connection causes
27 * an ECONNREFUSED off the icmp, but
29 * Alan Cox : Switched to new sk_buff handlers. No more backlog!
30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK
31 * bug no longer crashes it.
32 * Fred Van Kempen : Net2e support for sk->broadcast.
33 * Alan Cox : Uses skb_free_datagram
34 * Alan Cox : Added get/set sockopt support.
35 * Alan Cox : Broadcasting without option set returns EACCES.
36 * Alan Cox : No wakeup calls. Instead we now use the callbacks.
37 * Alan Cox : Use ip_tos and ip_ttl
38 * Alan Cox : SNMP Mibs
39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support.
40 * Matt Dillon : UDP length checks.
41 * Alan Cox : Smarter af_inet used properly.
42 * Alan Cox : Use new kernel side addressing.
43 * Alan Cox : Incorrect return on truncated datagram receive.
44 * Arnt Gulbrandsen : New udp_send and stuff
45 * Alan Cox : Cache last socket
46 * Alan Cox : Route cache
47 * Jon Peatfield : Minor efficiency fix to sendto().
48 * Mike Shaver : RFC1122 checks.
49 * Alan Cox : Nonblocking error fix.
50 * Willy Konynenberg : Transparent proxying support.
51 * Mike McLagan : Routing by source
52 * David S. Miller : New socket lookup architecture.
53 * Last socket cache retained as it
54 * does have a high hit rate.
55 * Olaf Kirch : Don't linearise iovec on sendmsg.
56 * Andi Kleen : Some cleanups, cache destination entry
58 * Vitaly E. Lavrov : Transparent proxy revived after year coma.
59 * Melvin Smith : Check msg_name not msg_namelen in sendto(),
60 * return ENOTCONN for unconnected sockets (POSIX)
61 * Janos Farkas : don't deliver multi/broadcasts to a different
62 * bound-to-device socket
63 * Hirokazu Takahashi : HW checksumming for outgoing UDP
65 * Hirokazu Takahashi : sendfile() on UDP works now.
66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file
67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which
68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind
69 * a single port at the same time.
70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
71 * James Chapman : Add L2TP encapsulation type.
74 * This program is free software; you can redistribute it and/or
75 * modify it under the terms of the GNU General Public License
76 * as published by the Free Software Foundation; either version
77 * 2 of the License, or (at your option) any later version.
80 #include <asm/system.h>
81 #include <asm/uaccess.h>
82 #include <asm/ioctls.h>
83 #include <linux/bootmem.h>
84 #include <linux/types.h>
85 #include <linux/fcntl.h>
86 #include <linux/module.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/igmp.h>
91 #include <linux/errno.h>
92 #include <linux/timer.h>
94 #include <linux/inet.h>
95 #include <linux/netdevice.h>
96 #include <net/tcp_states.h>
97 #include <linux/skbuff.h>
98 #include <linux/proc_fs.h>
99 #include <linux/seq_file.h>
100 #include <net/net_namespace.h>
101 #include <net/icmp.h>
102 #include <net/route.h>
103 #include <net/checksum.h>
104 #include <net/xfrm.h>
105 #include "udp_impl.h"
108 * Snmp MIB for the UDP layer
111 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly;
112 EXPORT_SYMBOL(udp_stats_in6);
114 struct hlist_head udp_hash[UDP_HTABLE_SIZE];
115 DEFINE_RWLOCK(udp_hash_lock);
117 int sysctl_udp_mem[3] __read_mostly;
118 int sysctl_udp_rmem_min __read_mostly;
119 int sysctl_udp_wmem_min __read_mostly;
121 EXPORT_SYMBOL(sysctl_udp_mem);
122 EXPORT_SYMBOL(sysctl_udp_rmem_min);
123 EXPORT_SYMBOL(sysctl_udp_wmem_min);
125 atomic_t udp_memory_allocated;
126 EXPORT_SYMBOL(udp_memory_allocated);
128 static inline int __udp_lib_lport_inuse(struct net *net, __u16 num,
129 const struct hlist_head udptable[])
132 struct hlist_node *node;
134 sk_for_each(sk, node, &udptable[udp_hashfn(net, num)])
135 if (net_eq(sock_net(sk), net) && sk->sk_hash == num)
141 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6
143 * @sk: socket struct in question
144 * @snum: port number to look up
145 * @saddr_comp: AF-dependent comparison of bound local IP addresses
147 int udp_lib_get_port(struct sock *sk, unsigned short snum,
148 int (*saddr_comp)(const struct sock *sk1,
149 const struct sock *sk2 ) )
151 struct hlist_head *udptable = sk->sk_prot->h.udp_hash;
152 struct hlist_node *node;
153 struct hlist_head *head;
156 struct net *net = sock_net(sk);
158 write_lock_bh(&udp_hash_lock);
161 int i, low, high, remaining;
162 unsigned rover, best, best_size_so_far;
164 inet_get_local_port_range(&low, &high);
165 remaining = (high - low) + 1;
167 best_size_so_far = UINT_MAX;
168 best = rover = net_random() % remaining + low;
170 /* 1st pass: look for empty (or shortest) hash chain */
171 for (i = 0; i < UDP_HTABLE_SIZE; i++) {
174 head = &udptable[udp_hashfn(net, rover)];
175 if (hlist_empty(head))
178 sk_for_each(sk2, node, head) {
179 if (++size >= best_size_so_far)
182 best_size_so_far = size;
185 /* fold back if end of range */
187 rover = low + ((rover - low)
188 & (UDP_HTABLE_SIZE - 1));
193 /* 2nd pass: find hole in shortest hash chain */
195 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) {
196 if (! __udp_lib_lport_inuse(net, rover, udptable))
198 rover += UDP_HTABLE_SIZE;
200 rover = low + ((rover - low)
201 & (UDP_HTABLE_SIZE - 1));
205 /* All ports in use! */
211 head = &udptable[udp_hashfn(net, snum)];
213 sk_for_each(sk2, node, head)
214 if (sk2->sk_hash == snum &&
216 net_eq(sock_net(sk2), net) &&
217 (!sk2->sk_reuse || !sk->sk_reuse) &&
218 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if
219 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
220 (*saddr_comp)(sk, sk2) )
224 inet_sk(sk)->num = snum;
226 if (sk_unhashed(sk)) {
227 head = &udptable[udp_hashfn(net, snum)];
228 sk_add_node(sk, head);
229 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
233 write_unlock_bh(&udp_hash_lock);
237 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
239 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
241 return ( !ipv6_only_sock(sk2) &&
242 (!inet1->rcv_saddr || !inet2->rcv_saddr ||
243 inet1->rcv_saddr == inet2->rcv_saddr ));
246 int udp_v4_get_port(struct sock *sk, unsigned short snum)
248 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal);
251 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try
252 * harder than this. -DaveM
254 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
255 __be16 sport, __be32 daddr, __be16 dport,
256 int dif, struct hlist_head udptable[])
258 struct sock *sk, *result = NULL;
259 struct hlist_node *node;
260 unsigned short hnum = ntohs(dport);
263 read_lock(&udp_hash_lock);
264 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) {
265 struct inet_sock *inet = inet_sk(sk);
267 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum &&
268 !ipv6_only_sock(sk)) {
269 int score = (sk->sk_family == PF_INET ? 1 : 0);
270 if (inet->rcv_saddr) {
271 if (inet->rcv_saddr != daddr)
276 if (inet->daddr != saddr)
281 if (inet->dport != sport)
285 if (sk->sk_bound_dev_if) {
286 if (sk->sk_bound_dev_if != dif)
293 } else if (score > badness) {
301 read_unlock(&udp_hash_lock);
305 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
306 __be16 loc_port, __be32 loc_addr,
307 __be16 rmt_port, __be32 rmt_addr,
310 struct hlist_node *node;
312 unsigned short hnum = ntohs(loc_port);
314 sk_for_each_from(s, node) {
315 struct inet_sock *inet = inet_sk(s);
317 if (!net_eq(sock_net(s), net) ||
318 s->sk_hash != hnum ||
319 (inet->daddr && inet->daddr != rmt_addr) ||
320 (inet->dport != rmt_port && inet->dport) ||
321 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) ||
323 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
325 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
335 * This routine is called by the ICMP module when it gets some
336 * sort of error condition. If err < 0 then the socket should
337 * be closed and the error returned to the user. If err > 0
338 * it's just the icmp type << 8 | icmp code.
339 * Header points to the ip header of the error packet. We move
340 * on past this. Then (as it used to claim before adjustment)
341 * header points to the first 8 bytes of the udp header. We need
342 * to find the appropriate port.
345 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[])
347 struct inet_sock *inet;
348 struct iphdr *iph = (struct iphdr*)skb->data;
349 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
350 const int type = icmp_hdr(skb)->type;
351 const int code = icmp_hdr(skb)->code;
355 struct net *net = dev_net(skb->dev);
357 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
358 iph->saddr, uh->source, skb->dev->ifindex, udptable);
360 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
361 return; /* No socket for error */
370 case ICMP_TIME_EXCEEDED:
373 case ICMP_SOURCE_QUENCH:
375 case ICMP_PARAMETERPROB:
379 case ICMP_DEST_UNREACH:
380 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
381 if (inet->pmtudisc != IP_PMTUDISC_DONT) {
389 if (code <= NR_ICMP_UNREACH) {
390 harderr = icmp_err_convert[code].fatal;
391 err = icmp_err_convert[code].errno;
397 * RFC1122: OK. Passes ICMP errors back to application, as per
400 if (!inet->recverr) {
401 if (!harderr || sk->sk_state != TCP_ESTABLISHED)
404 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
407 sk->sk_error_report(sk);
412 void udp_err(struct sk_buff *skb, u32 info)
414 __udp4_lib_err(skb, info, udp_hash);
418 * Throw away all pending data and cancel the corking. Socket is locked.
420 void udp_flush_pending_frames(struct sock *sk)
422 struct udp_sock *up = udp_sk(sk);
427 ip_flush_pending_frames(sk);
430 EXPORT_SYMBOL(udp_flush_pending_frames);
433 * udp4_hwcsum_outgoing - handle outgoing HW checksumming
434 * @sk: socket we are sending on
435 * @skb: sk_buff containing the filled-in UDP header
436 * (checksum field must be zeroed out)
438 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb,
439 __be32 src, __be32 dst, int len )
442 struct udphdr *uh = udp_hdr(skb);
445 if (skb_queue_len(&sk->sk_write_queue) == 1) {
447 * Only one fragment on the socket.
449 skb->csum_start = skb_transport_header(skb) - skb->head;
450 skb->csum_offset = offsetof(struct udphdr, check);
451 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0);
454 * HW-checksum won't work as there are two or more
455 * fragments on the socket so that all csums of sk_buffs
458 offset = skb_transport_offset(skb);
459 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0);
461 skb->ip_summed = CHECKSUM_NONE;
463 skb_queue_walk(&sk->sk_write_queue, skb) {
464 csum = csum_add(csum, skb->csum);
467 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
469 uh->check = CSUM_MANGLED_0;
474 * Push out all pending data as one UDP datagram. Socket is locked.
476 static int udp_push_pending_frames(struct sock *sk)
478 struct udp_sock *up = udp_sk(sk);
479 struct inet_sock *inet = inet_sk(sk);
480 struct flowi *fl = &inet->cork.fl;
484 int is_udplite = IS_UDPLITE(sk);
487 /* Grab the skbuff where UDP header space exists. */
488 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL)
492 * Create a UDP header
495 uh->source = fl->fl_ip_sport;
496 uh->dest = fl->fl_ip_dport;
497 uh->len = htons(up->len);
500 if (is_udplite) /* UDP-Lite */
501 csum = udplite_csum_outgoing(sk, skb);
503 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */
505 skb->ip_summed = CHECKSUM_NONE;
508 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
510 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len);
513 } else /* `normal' UDP */
514 csum = udp_csum_outgoing(sk, skb);
516 /* add protocol-dependent pseudo-header */
517 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len,
518 sk->sk_protocol, csum );
520 uh->check = CSUM_MANGLED_0;
523 err = ip_push_pending_frames(sk);
528 UDP_INC_STATS_USER(sock_net(sk),
529 UDP_MIB_OUTDATAGRAMS, is_udplite);
533 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
536 struct inet_sock *inet = inet_sk(sk);
537 struct udp_sock *up = udp_sk(sk);
539 struct ipcm_cookie ipc;
540 struct rtable *rt = NULL;
543 __be32 daddr, faddr, saddr;
546 int err, is_udplite = IS_UDPLITE(sk);
547 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
548 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
557 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */
564 * There are pending frames.
565 * The socket lock must be held while it's corked.
568 if (likely(up->pending)) {
569 if (unlikely(up->pending != AF_INET)) {
577 ulen += sizeof(struct udphdr);
580 * Get and verify the address.
583 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
584 if (msg->msg_namelen < sizeof(*usin))
586 if (usin->sin_family != AF_INET) {
587 if (usin->sin_family != AF_UNSPEC)
588 return -EAFNOSUPPORT;
591 daddr = usin->sin_addr.s_addr;
592 dport = usin->sin_port;
596 if (sk->sk_state != TCP_ESTABLISHED)
597 return -EDESTADDRREQ;
600 /* Open fast path for connected socket.
601 Route will not be used, if at least one option is set.
605 ipc.addr = inet->saddr;
607 ipc.oif = sk->sk_bound_dev_if;
608 if (msg->msg_controllen) {
609 err = ip_cmsg_send(sock_net(sk), msg, &ipc);
620 ipc.addr = faddr = daddr;
622 if (ipc.opt && ipc.opt->srr) {
625 faddr = ipc.opt->faddr;
628 tos = RT_TOS(inet->tos);
629 if (sock_flag(sk, SOCK_LOCALROUTE) ||
630 (msg->msg_flags & MSG_DONTROUTE) ||
631 (ipc.opt && ipc.opt->is_strictroute)) {
636 if (ipv4_is_multicast(daddr)) {
638 ipc.oif = inet->mc_index;
640 saddr = inet->mc_addr;
645 rt = (struct rtable*)sk_dst_check(sk, 0);
648 struct flowi fl = { .oif = ipc.oif,
653 .proto = sk->sk_protocol,
655 { .sport = inet->sport,
656 .dport = dport } } };
657 struct net *net = sock_net(sk);
659 security_sk_classify_flow(sk, &fl);
660 err = ip_route_output_flow(net, &rt, &fl, sk, 1);
662 if (err == -ENETUNREACH)
663 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
668 if ((rt->rt_flags & RTCF_BROADCAST) &&
669 !sock_flag(sk, SOCK_BROADCAST))
672 sk_dst_set(sk, dst_clone(&rt->u.dst));
675 if (msg->msg_flags&MSG_CONFIRM)
681 daddr = ipc.addr = rt->rt_dst;
684 if (unlikely(up->pending)) {
685 /* The socket is already corked while preparing it. */
686 /* ... which is an evident application bug. --ANK */
689 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n");
694 * Now cork the socket to pend data.
696 inet->cork.fl.fl4_dst = daddr;
697 inet->cork.fl.fl_ip_dport = dport;
698 inet->cork.fl.fl4_src = saddr;
699 inet->cork.fl.fl_ip_sport = inet->sport;
700 up->pending = AF_INET;
704 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
705 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen,
706 sizeof(struct udphdr), &ipc, rt,
707 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
709 udp_flush_pending_frames(sk);
711 err = udp_push_pending_frames(sk);
712 else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
723 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting
724 * ENOBUFS might not be good (it's not tunable per se), but otherwise
725 * we don't have a good statistic (IpOutDiscards but it can be too many
726 * things). We could add another new stat but at least for now that
727 * seems like overkill.
729 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
730 UDP_INC_STATS_USER(sock_net(sk),
731 UDP_MIB_SNDBUFERRORS, is_udplite);
736 dst_confirm(&rt->u.dst);
737 if (!(msg->msg_flags&MSG_PROBE) || len)
738 goto back_from_confirm;
743 int udp_sendpage(struct sock *sk, struct page *page, int offset,
744 size_t size, int flags)
746 struct udp_sock *up = udp_sk(sk);
750 struct msghdr msg = { .msg_flags = flags|MSG_MORE };
752 /* Call udp_sendmsg to specify destination address which
753 * sendpage interface can't pass.
754 * This will succeed only when the socket is connected.
756 ret = udp_sendmsg(NULL, sk, &msg, 0);
763 if (unlikely(!up->pending)) {
766 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n");
770 ret = ip_append_page(sk, page, offset, size, flags);
771 if (ret == -EOPNOTSUPP) {
773 return sock_no_sendpage(sk->sk_socket, page, offset,
777 udp_flush_pending_frames(sk);
782 if (!(up->corkflag || (flags&MSG_MORE)))
783 ret = udp_push_pending_frames(sk);
792 * IOCTL requests applicable to the UDP protocol
795 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
800 int amount = atomic_read(&sk->sk_wmem_alloc);
801 return put_user(amount, (int __user *)arg);
807 unsigned long amount;
810 spin_lock_bh(&sk->sk_receive_queue.lock);
811 skb = skb_peek(&sk->sk_receive_queue);
814 * We will only return the amount
815 * of this packet since that is all
818 amount = skb->len - sizeof(struct udphdr);
820 spin_unlock_bh(&sk->sk_receive_queue.lock);
821 return put_user(amount, (int __user *)arg);
832 * This should be easy, if there is something there we
833 * return it, otherwise we block.
836 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
837 size_t len, int noblock, int flags, int *addr_len)
839 struct inet_sock *inet = inet_sk(sk);
840 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
842 unsigned int ulen, copied;
845 int is_udplite = IS_UDPLITE(sk);
848 * Check any passed addresses
851 *addr_len=sizeof(*sin);
853 if (flags & MSG_ERRQUEUE)
854 return ip_recv_error(sk, msg, len);
857 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
862 ulen = skb->len - sizeof(struct udphdr);
866 else if (copied < ulen)
867 msg->msg_flags |= MSG_TRUNC;
870 * If checksum is needed at all, try to do it while copying the
871 * data. If the data is truncated, or if we only want a partial
872 * coverage checksum (UDP-Lite), do it before the copy.
875 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
876 if (udp_lib_checksum_complete(skb))
880 if (skb_csum_unnecessary(skb))
881 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
882 msg->msg_iov, copied );
884 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
894 UDP_INC_STATS_USER(sock_net(sk),
895 UDP_MIB_INDATAGRAMS, is_udplite);
897 sock_recv_timestamp(msg, sk, skb);
899 /* Copy the address. */
902 sin->sin_family = AF_INET;
903 sin->sin_port = udp_hdr(skb)->source;
904 sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
905 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
907 if (inet->cmsg_flags)
908 ip_cmsg_recv(msg, skb);
911 if (flags & MSG_TRUNC)
916 skb_free_datagram(sk, skb);
923 if (!skb_kill_datagram(sk, skb, flags))
924 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
933 int udp_disconnect(struct sock *sk, int flags)
935 struct inet_sock *inet = inet_sk(sk);
937 * 1003.1g - break association.
940 sk->sk_state = TCP_CLOSE;
943 sk->sk_bound_dev_if = 0;
944 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
945 inet_reset_saddr(sk);
947 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
948 sk->sk_prot->unhash(sk);
955 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
957 int is_udplite = IS_UDPLITE(sk);
960 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) {
961 /* Note that an ENOMEM error is charged twice */
963 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
971 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
979 * >0: "udp encap" protocol resubmission
981 * Note that in the success and error cases, the skb is assumed to
982 * have either been requeued or freed.
984 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
986 struct udp_sock *up = udp_sk(sk);
988 int is_udplite = IS_UDPLITE(sk);
991 * Charge it to the socket, dropping if the queue is full.
993 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
997 if (up->encap_type) {
999 * This is an encapsulation socket so pass the skb to
1000 * the socket's udp_encap_rcv() hook. Otherwise, just
1001 * fall through and pass this up the UDP socket.
1002 * up->encap_rcv() returns the following value:
1003 * =0 if skb was successfully passed to the encap
1004 * handler or was discarded by it.
1005 * >0 if skb should be passed on to UDP.
1006 * <0 if skb should be resubmitted as proto -N
1009 /* if we're overly short, let UDP handle it */
1010 if (skb->len > sizeof(struct udphdr) &&
1011 up->encap_rcv != NULL) {
1014 ret = (*up->encap_rcv)(sk, skb);
1016 UDP_INC_STATS_BH(sock_net(sk),
1017 UDP_MIB_INDATAGRAMS,
1023 /* FALLTHROUGH -- it's a UDP Packet */
1027 * UDP-Lite specific tests, ignored on UDP sockets
1029 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) {
1032 * MIB statistics other than incrementing the error count are
1033 * disabled for the following two types of errors: these depend
1034 * on the application settings, not on the functioning of the
1035 * protocol stack as such.
1037 * RFC 3828 here recommends (sec 3.3): "There should also be a
1038 * way ... to ... at least let the receiving application block
1039 * delivery of packets with coverage values less than a value
1040 * provided by the application."
1042 if (up->pcrlen == 0) { /* full coverage was set */
1043 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage "
1044 "%d while full coverage %d requested\n",
1045 UDP_SKB_CB(skb)->cscov, skb->len);
1048 /* The next case involves violating the min. coverage requested
1049 * by the receiver. This is subtle: if receiver wants x and x is
1050 * greater than the buffersize/MTU then receiver will complain
1051 * that it wants x while sender emits packets of smaller size y.
1052 * Therefore the above ...()->partial_cov statement is essential.
1054 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) {
1055 LIMIT_NETDEBUG(KERN_WARNING
1056 "UDPLITE: coverage %d too small, need min %d\n",
1057 UDP_SKB_CB(skb)->cscov, up->pcrlen);
1062 if (sk->sk_filter) {
1063 if (udp_lib_checksum_complete(skb))
1070 if (!sock_owned_by_user(sk))
1071 rc = __udp_queue_rcv_skb(sk, skb);
1073 sk_add_backlog(sk, skb);
1079 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1085 * Multicasts and broadcasts go to each listener.
1087 * Note: called only from the BH handler context,
1088 * so we don't need to lock the hashes.
1090 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1092 __be32 saddr, __be32 daddr,
1093 struct hlist_head udptable[])
1098 read_lock(&udp_hash_lock);
1099 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]);
1100 dif = skb->dev->ifindex;
1101 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1103 struct sock *sknext = NULL;
1106 struct sk_buff *skb1 = skb;
1108 sknext = udp_v4_mcast_next(net, sk_next(sk), uh->dest,
1109 daddr, uh->source, saddr,
1112 skb1 = skb_clone(skb, GFP_ATOMIC);
1115 int ret = udp_queue_rcv_skb(sk, skb1);
1117 /* we should probably re-process instead
1118 * of dropping packets here. */
1125 read_unlock(&udp_hash_lock);
1129 /* Initialize UDP checksum. If exited with zero value (success),
1130 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1131 * Otherwise, csum completion requires chacksumming packet body,
1132 * including udp header and folding it to skb->csum.
1134 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1137 const struct iphdr *iph;
1140 UDP_SKB_CB(skb)->partial_cov = 0;
1141 UDP_SKB_CB(skb)->cscov = skb->len;
1143 if (proto == IPPROTO_UDPLITE) {
1144 err = udplite_checksum_init(skb, uh);
1150 if (uh->check == 0) {
1151 skb->ip_summed = CHECKSUM_UNNECESSARY;
1152 } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1153 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1155 skb->ip_summed = CHECKSUM_UNNECESSARY;
1157 if (!skb_csum_unnecessary(skb))
1158 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1159 skb->len, proto, 0);
1160 /* Probably, we should checksum udp header (it should be in cache
1161 * in any case) and data in tiny packets (< rx copybreak).
1168 * All we need to do is get the socket, and then do a checksum.
1171 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[],
1175 struct udphdr *uh = udp_hdr(skb);
1176 unsigned short ulen;
1177 struct rtable *rt = (struct rtable*)skb->dst;
1178 __be32 saddr = ip_hdr(skb)->saddr;
1179 __be32 daddr = ip_hdr(skb)->daddr;
1180 struct net *net = dev_net(skb->dev);
1183 * Validate the packet.
1185 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1186 goto drop; /* No space for header. */
1188 ulen = ntohs(uh->len);
1189 if (ulen > skb->len)
1192 if (proto == IPPROTO_UDP) {
1193 /* UDP validates ulen. */
1194 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1199 if (udp4_csum_init(skb, uh, proto))
1202 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1203 return __udp4_lib_mcast_deliver(net, skb, uh,
1204 saddr, daddr, udptable);
1206 sk = __udp4_lib_lookup(net, saddr, uh->source, daddr,
1207 uh->dest, inet_iif(skb), udptable);
1210 int ret = udp_queue_rcv_skb(sk, skb);
1213 /* a return value > 0 means to resubmit the input, but
1214 * it wants the return to be -protocol, or 0
1221 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1225 /* No socket. Drop packet silently, if checksum is wrong */
1226 if (udp_lib_checksum_complete(skb))
1229 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1230 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1233 * Hmm. We got an UDP packet to a port to which we
1234 * don't wanna listen. Ignore it.
1240 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n",
1241 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1252 * RFC1122: OK. Discards the bad packet silently (as far as
1253 * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1255 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n",
1256 proto == IPPROTO_UDPLITE ? "-Lite" : "",
1263 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1268 int udp_rcv(struct sk_buff *skb)
1270 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP);
1273 void udp_destroy_sock(struct sock *sk)
1276 udp_flush_pending_frames(sk);
1281 * Socket option code for UDP
1283 int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1284 char __user *optval, int optlen,
1285 int (*push_pending_frames)(struct sock *))
1287 struct udp_sock *up = udp_sk(sk);
1290 int is_udplite = IS_UDPLITE(sk);
1292 if (optlen<sizeof(int))
1295 if (get_user(val, (int __user *)optval))
1305 (*push_pending_frames)(sk);
1313 case UDP_ENCAP_ESPINUDP:
1314 case UDP_ENCAP_ESPINUDP_NON_IKE:
1315 up->encap_rcv = xfrm4_udp_encap_rcv;
1317 case UDP_ENCAP_L2TPINUDP:
1318 up->encap_type = val;
1327 * UDP-Lite's partial checksum coverage (RFC 3828).
1329 /* The sender sets actual checksum coverage length via this option.
1330 * The case coverage > packet length is handled by send module. */
1331 case UDPLITE_SEND_CSCOV:
1332 if (!is_udplite) /* Disable the option on UDP sockets */
1333 return -ENOPROTOOPT;
1334 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1336 else if (val > USHORT_MAX)
1339 up->pcflag |= UDPLITE_SEND_CC;
1342 /* The receiver specifies a minimum checksum coverage value. To make
1343 * sense, this should be set to at least 8 (as done below). If zero is
1344 * used, this again means full checksum coverage. */
1345 case UDPLITE_RECV_CSCOV:
1346 if (!is_udplite) /* Disable the option on UDP sockets */
1347 return -ENOPROTOOPT;
1348 if (val != 0 && val < 8) /* Avoid silly minimal values. */
1350 else if (val > USHORT_MAX)
1353 up->pcflag |= UDPLITE_RECV_CC;
1364 int udp_setsockopt(struct sock *sk, int level, int optname,
1365 char __user *optval, int optlen)
1367 if (level == SOL_UDP || level == SOL_UDPLITE)
1368 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1369 udp_push_pending_frames);
1370 return ip_setsockopt(sk, level, optname, optval, optlen);
1373 #ifdef CONFIG_COMPAT
1374 int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1375 char __user *optval, int optlen)
1377 if (level == SOL_UDP || level == SOL_UDPLITE)
1378 return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1379 udp_push_pending_frames);
1380 return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1384 int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1385 char __user *optval, int __user *optlen)
1387 struct udp_sock *up = udp_sk(sk);
1390 if (get_user(len,optlen))
1393 len = min_t(unsigned int, len, sizeof(int));
1404 val = up->encap_type;
1407 /* The following two cannot be changed on UDP sockets, the return is
1408 * always 0 (which corresponds to the full checksum coverage of UDP). */
1409 case UDPLITE_SEND_CSCOV:
1413 case UDPLITE_RECV_CSCOV:
1418 return -ENOPROTOOPT;
1421 if (put_user(len, optlen))
1423 if (copy_to_user(optval, &val,len))
1428 int udp_getsockopt(struct sock *sk, int level, int optname,
1429 char __user *optval, int __user *optlen)
1431 if (level == SOL_UDP || level == SOL_UDPLITE)
1432 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1433 return ip_getsockopt(sk, level, optname, optval, optlen);
1436 #ifdef CONFIG_COMPAT
1437 int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1438 char __user *optval, int __user *optlen)
1440 if (level == SOL_UDP || level == SOL_UDPLITE)
1441 return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1442 return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1446 * udp_poll - wait for a UDP event.
1447 * @file - file struct
1449 * @wait - poll table
1451 * This is same as datagram poll, except for the special case of
1452 * blocking sockets. If application is using a blocking fd
1453 * and a packet with checksum error is in the queue;
1454 * then it could get return from select indicating data available
1455 * but then block when reading it. Add special case code
1456 * to work around these arguably broken applications.
1458 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1460 unsigned int mask = datagram_poll(file, sock, wait);
1461 struct sock *sk = sock->sk;
1462 int is_lite = IS_UDPLITE(sk);
1464 /* Check for false positives due to checksum errors */
1465 if ( (mask & POLLRDNORM) &&
1466 !(file->f_flags & O_NONBLOCK) &&
1467 !(sk->sk_shutdown & RCV_SHUTDOWN)){
1468 struct sk_buff_head *rcvq = &sk->sk_receive_queue;
1469 struct sk_buff *skb;
1471 spin_lock_bh(&rcvq->lock);
1472 while ((skb = skb_peek(rcvq)) != NULL &&
1473 udp_lib_checksum_complete(skb)) {
1474 UDP_INC_STATS_BH(sock_net(sk),
1475 UDP_MIB_INERRORS, is_lite);
1476 __skb_unlink(skb, rcvq);
1479 spin_unlock_bh(&rcvq->lock);
1481 /* nothing to see, move along */
1483 mask &= ~(POLLIN | POLLRDNORM);
1490 struct proto udp_prot = {
1492 .owner = THIS_MODULE,
1493 .close = udp_lib_close,
1494 .connect = ip4_datagram_connect,
1495 .disconnect = udp_disconnect,
1497 .destroy = udp_destroy_sock,
1498 .setsockopt = udp_setsockopt,
1499 .getsockopt = udp_getsockopt,
1500 .sendmsg = udp_sendmsg,
1501 .recvmsg = udp_recvmsg,
1502 .sendpage = udp_sendpage,
1503 .backlog_rcv = __udp_queue_rcv_skb,
1504 .hash = udp_lib_hash,
1505 .unhash = udp_lib_unhash,
1506 .get_port = udp_v4_get_port,
1507 .memory_allocated = &udp_memory_allocated,
1508 .sysctl_mem = sysctl_udp_mem,
1509 .sysctl_wmem = &sysctl_udp_wmem_min,
1510 .sysctl_rmem = &sysctl_udp_rmem_min,
1511 .obj_size = sizeof(struct udp_sock),
1512 .h.udp_hash = udp_hash,
1513 #ifdef CONFIG_COMPAT
1514 .compat_setsockopt = compat_udp_setsockopt,
1515 .compat_getsockopt = compat_udp_getsockopt,
1519 /* ------------------------------------------------------------------------ */
1520 #ifdef CONFIG_PROC_FS
1522 static struct sock *udp_get_first(struct seq_file *seq)
1525 struct udp_iter_state *state = seq->private;
1526 struct net *net = seq_file_net(seq);
1528 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) {
1529 struct hlist_node *node;
1530 sk_for_each(sk, node, state->hashtable + state->bucket) {
1531 if (!net_eq(sock_net(sk), net))
1533 if (sk->sk_family == state->family)
1542 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
1544 struct udp_iter_state *state = seq->private;
1545 struct net *net = seq_file_net(seq);
1551 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
1553 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) {
1554 sk = sk_head(state->hashtable + state->bucket);
1560 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
1562 struct sock *sk = udp_get_first(seq);
1565 while (pos && (sk = udp_get_next(seq, sk)) != NULL)
1567 return pos ? NULL : sk;
1570 static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
1571 __acquires(udp_hash_lock)
1573 read_lock(&udp_hash_lock);
1574 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
1577 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1581 if (v == SEQ_START_TOKEN)
1582 sk = udp_get_idx(seq, 0);
1584 sk = udp_get_next(seq, v);
1590 static void udp_seq_stop(struct seq_file *seq, void *v)
1591 __releases(udp_hash_lock)
1593 read_unlock(&udp_hash_lock);
1596 static int udp_seq_open(struct inode *inode, struct file *file)
1598 struct udp_seq_afinfo *afinfo = PDE(inode)->data;
1599 struct udp_iter_state *s;
1602 err = seq_open_net(inode, file, &afinfo->seq_ops,
1603 sizeof(struct udp_iter_state));
1607 s = ((struct seq_file *)file->private_data)->private;
1608 s->family = afinfo->family;
1609 s->hashtable = afinfo->hashtable;
1613 /* ------------------------------------------------------------------------ */
1614 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
1616 struct proc_dir_entry *p;
1619 afinfo->seq_fops.open = udp_seq_open;
1620 afinfo->seq_fops.read = seq_read;
1621 afinfo->seq_fops.llseek = seq_lseek;
1622 afinfo->seq_fops.release = seq_release_net;
1624 afinfo->seq_ops.start = udp_seq_start;
1625 afinfo->seq_ops.next = udp_seq_next;
1626 afinfo->seq_ops.stop = udp_seq_stop;
1628 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
1629 &afinfo->seq_fops, afinfo);
1635 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
1637 proc_net_remove(net, afinfo->name);
1640 /* ------------------------------------------------------------------------ */
1641 static void udp4_format_sock(struct sock *sp, struct seq_file *f,
1642 int bucket, int *len)
1644 struct inet_sock *inet = inet_sk(sp);
1645 __be32 dest = inet->daddr;
1646 __be32 src = inet->rcv_saddr;
1647 __u16 destp = ntohs(inet->dport);
1648 __u16 srcp = ntohs(inet->sport);
1650 seq_printf(f, "%4d: %08X:%04X %08X:%04X"
1651 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n",
1652 bucket, src, srcp, dest, destp, sp->sk_state,
1653 atomic_read(&sp->sk_wmem_alloc),
1654 atomic_read(&sp->sk_rmem_alloc),
1655 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
1656 atomic_read(&sp->sk_refcnt), sp,
1657 atomic_read(&sp->sk_drops), len);
1660 int udp4_seq_show(struct seq_file *seq, void *v)
1662 if (v == SEQ_START_TOKEN)
1663 seq_printf(seq, "%-127s\n",
1664 " sl local_address rem_address st tx_queue "
1665 "rx_queue tr tm->when retrnsmt uid timeout "
1666 "inode ref pointer drops");
1668 struct udp_iter_state *state = seq->private;
1671 udp4_format_sock(v, seq, state->bucket, &len);
1672 seq_printf(seq, "%*s\n", 127 - len ,"");
1677 /* ------------------------------------------------------------------------ */
1678 static struct udp_seq_afinfo udp4_seq_afinfo = {
1681 .hashtable = udp_hash,
1683 .owner = THIS_MODULE,
1686 .show = udp4_seq_show,
1690 static int udp4_proc_init_net(struct net *net)
1692 return udp_proc_register(net, &udp4_seq_afinfo);
1695 static void udp4_proc_exit_net(struct net *net)
1697 udp_proc_unregister(net, &udp4_seq_afinfo);
1700 static struct pernet_operations udp4_net_ops = {
1701 .init = udp4_proc_init_net,
1702 .exit = udp4_proc_exit_net,
1705 int __init udp4_proc_init(void)
1707 return register_pernet_subsys(&udp4_net_ops);
1710 void udp4_proc_exit(void)
1712 unregister_pernet_subsys(&udp4_net_ops);
1714 #endif /* CONFIG_PROC_FS */
1716 void __init udp_init(void)
1718 unsigned long limit;
1720 /* Set the pressure threshold up by the same strategy of TCP. It is a
1721 * fraction of global memory that is up to 1/2 at 256 MB, decreasing
1722 * toward zero with the amount of memory, with a floor of 128 pages.
1724 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT);
1725 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11);
1726 limit = max(limit, 128UL);
1727 sysctl_udp_mem[0] = limit / 4 * 3;
1728 sysctl_udp_mem[1] = limit;
1729 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
1731 sysctl_udp_rmem_min = SK_MEM_QUANTUM;
1732 sysctl_udp_wmem_min = SK_MEM_QUANTUM;
1735 EXPORT_SYMBOL(udp_disconnect);
1736 EXPORT_SYMBOL(udp_hash);
1737 EXPORT_SYMBOL(udp_hash_lock);
1738 EXPORT_SYMBOL(udp_ioctl);
1739 EXPORT_SYMBOL(udp_prot);
1740 EXPORT_SYMBOL(udp_sendmsg);
1741 EXPORT_SYMBOL(udp_lib_getsockopt);
1742 EXPORT_SYMBOL(udp_lib_setsockopt);
1743 EXPORT_SYMBOL(udp_poll);
1744 EXPORT_SYMBOL(udp_lib_get_port);
1746 #ifdef CONFIG_PROC_FS
1747 EXPORT_SYMBOL(udp_proc_register);
1748 EXPORT_SYMBOL(udp_proc_unregister);