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 * PF_INET protocol family socket handler.
8 * Version: $Id: af_inet.c,v 1.137 2002/02/01 22:01:03 davem Exp $
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
15 * Changes (see also sock.c)
18 * Karl Knutson : Socket protocol table
19 * A.N.Kuznetsov : Socket death error in accept().
20 * John Richardson : Fix non blocking error in connect()
21 * so sockets that fail to connect
22 * don't return -EINPROGRESS.
23 * Alan Cox : Asynchronous I/O support
24 * Alan Cox : Keep correct socket pointer on sock
27 * Alan Cox : Semantics of SO_LINGER aren't state
28 * moved to close when you look carefully.
29 * With this fixed and the accept bug fixed
30 * some RPC stuff seems happier.
31 * Niibe Yutaka : 4.4BSD style write async I/O
33 * Tony Gale : Fixed reuse semantics.
34 * Alan Cox : bind() shouldn't abort existing but dead
35 * sockets. Stops FTP netin:.. I hope.
36 * Alan Cox : bind() works correctly for RAW sockets.
37 * Note that FreeBSD at least was broken
38 * in this respect so be careful with
39 * compatibility tests...
40 * Alan Cox : routing cache support
41 * Alan Cox : memzero the socket structure for
43 * Matt Day : nonblock connect error handler
44 * Alan Cox : Allow large numbers of pending sockets
45 * (eg for big web sites), but only if
46 * specifically application requested.
47 * Alan Cox : New buffering throughout IP. Used
49 * Alan Cox : New buffering now used smartly.
50 * Alan Cox : BSD rather than common sense
51 * interpretation of listen.
52 * Germano Caronni : Assorted small races.
53 * Alan Cox : sendmsg/recvmsg basic support.
54 * Alan Cox : Only sendmsg/recvmsg now supported.
55 * Alan Cox : Locked down bind (see security list).
56 * Alan Cox : Loosened bind a little.
57 * Mike McLagan : ADD/DEL DLCI Ioctls
58 * Willy Konynenberg : Transparent proxying support.
59 * David S. Miller : New socket lookup architecture.
60 * Some other random speedups.
61 * Cyrus Durgin : Cleaned up file for kmod hacks.
62 * Andi Kleen : Fix inet_stream_connect TCP race.
64 * This program is free software; you can redistribute it and/or
65 * modify it under the terms of the GNU General Public License
66 * as published by the Free Software Foundation; either version
67 * 2 of the License, or (at your option) any later version.
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
94 #include <linux/smp_lock.h>
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
100 #include <net/protocol.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
107 #include <net/udplite.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
111 #include <net/icmp.h>
112 #include <net/ipip.h>
113 #include <net/inet_common.h>
114 #include <net/xfrm.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
119 DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
121 extern void ip_mc_drop_socket(struct sock *sk);
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
129 /* New destruction routine */
131 void inet_sock_destruct(struct sock *sk)
133 struct inet_sock *inet = inet_sk(sk);
135 __skb_queue_purge(&sk->sk_receive_queue);
136 __skb_queue_purge(&sk->sk_error_queue);
138 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
139 printk("Attempt to release TCP socket in state %d %p\n",
143 if (!sock_flag(sk, SOCK_DEAD)) {
144 printk("Attempt to release alive inet socket %p\n", sk);
148 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
149 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
150 BUG_TRAP(!sk->sk_wmem_queued);
151 BUG_TRAP(!sk->sk_forward_alloc);
154 dst_release(sk->sk_dst_cache);
155 sk_refcnt_debug_dec(sk);
159 * The routines beyond this point handle the behaviour of an AF_INET
160 * socket object. Mostly it punts to the subprotocols of IP to do
165 * Automatically bind an unbound socket.
168 static int inet_autobind(struct sock *sk)
170 struct inet_sock *inet;
171 /* We may need to bind the socket. */
175 if (sk->sk_prot->get_port(sk, 0)) {
179 inet->sport = htons(inet->num);
186 * Move a socket into listening state.
188 int inet_listen(struct socket *sock, int backlog)
190 struct sock *sk = sock->sk;
191 unsigned char old_state;
197 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
200 old_state = sk->sk_state;
201 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
204 /* Really, if the socket is already in listen state
205 * we can only allow the backlog to be adjusted.
207 if (old_state != TCP_LISTEN) {
208 err = inet_csk_listen_start(sk, backlog);
212 sk->sk_max_ack_backlog = backlog;
221 * Create an inet socket.
224 static int inet_create(struct socket *sock, int protocol)
228 struct inet_protosw *answer;
229 struct inet_sock *inet;
230 struct proto *answer_prot;
231 unsigned char answer_flags;
232 char answer_no_check;
233 int try_loading_module = 0;
236 sock->state = SS_UNCONNECTED;
238 /* Look for the requested type/protocol pair. */
241 err = -ESOCKTNOSUPPORT;
243 list_for_each_rcu(p, &inetsw[sock->type]) {
244 answer = list_entry(p, struct inet_protosw, list);
246 /* Check the non-wild match. */
247 if (protocol == answer->protocol) {
248 if (protocol != IPPROTO_IP)
251 /* Check for the two wild cases. */
252 if (IPPROTO_IP == protocol) {
253 protocol = answer->protocol;
256 if (IPPROTO_IP == answer->protocol)
259 err = -EPROTONOSUPPORT;
263 if (unlikely(answer == NULL)) {
264 if (try_loading_module < 2) {
267 * Be more specific, e.g. net-pf-2-proto-132-type-1
268 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
270 if (++try_loading_module == 1)
271 request_module("net-pf-%d-proto-%d-type-%d",
272 PF_INET, protocol, sock->type);
274 * Fall back to generic, e.g. net-pf-2-proto-132
275 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
278 request_module("net-pf-%d-proto-%d",
280 goto lookup_protocol;
286 if (answer->capability > 0 && !capable(answer->capability))
289 sock->ops = answer->ops;
290 answer_prot = answer->prot;
291 answer_no_check = answer->no_check;
292 answer_flags = answer->flags;
295 BUG_TRAP(answer_prot->slab != NULL);
298 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
303 sk->sk_no_check = answer_no_check;
304 if (INET_PROTOSW_REUSE & answer_flags)
308 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
310 if (SOCK_RAW == sock->type) {
311 inet->num = protocol;
312 if (IPPROTO_RAW == protocol)
316 if (ipv4_config.no_pmtu_disc)
317 inet->pmtudisc = IP_PMTUDISC_DONT;
319 inet->pmtudisc = IP_PMTUDISC_WANT;
323 sock_init_data(sock, sk);
325 sk->sk_destruct = inet_sock_destruct;
326 sk->sk_family = PF_INET;
327 sk->sk_protocol = protocol;
328 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
334 inet->mc_list = NULL;
336 sk_refcnt_debug_inc(sk);
339 /* It assumes that any protocol which allows
340 * the user to assign a number at socket
341 * creation time automatically
344 inet->sport = htons(inet->num);
345 /* Add to protocol hash chains. */
346 sk->sk_prot->hash(sk);
349 if (sk->sk_prot->init) {
350 err = sk->sk_prot->init(sk);
352 sk_common_release(sk);
363 * The peer socket should always be NULL (or else). When we call this
364 * function we are destroying the object and from then on nobody
365 * should refer to it.
367 int inet_release(struct socket *sock)
369 struct sock *sk = sock->sk;
374 /* Applications forget to leave groups before exiting */
375 ip_mc_drop_socket(sk);
377 /* If linger is set, we don't return until the close
378 * is complete. Otherwise we return immediately. The
379 * actually closing is done the same either way.
381 * If the close is due to the process exiting, we never
385 if (sock_flag(sk, SOCK_LINGER) &&
386 !(current->flags & PF_EXITING))
387 timeout = sk->sk_lingertime;
389 sk->sk_prot->close(sk, timeout);
394 /* It is off by default, see below. */
395 int sysctl_ip_nonlocal_bind __read_mostly;
397 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
399 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
400 struct sock *sk = sock->sk;
401 struct inet_sock *inet = inet_sk(sk);
406 /* If the socket has its own bind function then use it. (RAW) */
407 if (sk->sk_prot->bind) {
408 err = sk->sk_prot->bind(sk, uaddr, addr_len);
412 if (addr_len < sizeof(struct sockaddr_in))
415 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
417 /* Not specified by any standard per-se, however it breaks too
418 * many applications when removed. It is unfortunate since
419 * allowing applications to make a non-local bind solves
420 * several problems with systems using dynamic addressing.
421 * (ie. your servers still start up even if your ISDN link
422 * is temporarily down)
424 err = -EADDRNOTAVAIL;
425 if (!sysctl_ip_nonlocal_bind &&
427 addr->sin_addr.s_addr != INADDR_ANY &&
428 chk_addr_ret != RTN_LOCAL &&
429 chk_addr_ret != RTN_MULTICAST &&
430 chk_addr_ret != RTN_BROADCAST)
433 snum = ntohs(addr->sin_port);
435 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
438 /* We keep a pair of addresses. rcv_saddr is the one
439 * used by hash lookups, and saddr is used for transmit.
441 * In the BSD API these are the same except where it
442 * would be illegal to use them (multicast/broadcast) in
443 * which case the sending device address is used.
447 /* Check these errors (active socket, double bind). */
449 if (sk->sk_state != TCP_CLOSE || inet->num)
450 goto out_release_sock;
452 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
453 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
454 inet->saddr = 0; /* Use device */
456 /* Make sure we are allowed to bind here. */
457 if (sk->sk_prot->get_port(sk, snum)) {
458 inet->saddr = inet->rcv_saddr = 0;
460 goto out_release_sock;
464 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
466 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
467 inet->sport = htons(inet->num);
478 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
479 int addr_len, int flags)
481 struct sock *sk = sock->sk;
483 if (uaddr->sa_family == AF_UNSPEC)
484 return sk->sk_prot->disconnect(sk, flags);
486 if (!inet_sk(sk)->num && inet_autobind(sk))
488 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
491 static long inet_wait_for_connect(struct sock *sk, long timeo)
495 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
497 /* Basic assumption: if someone sets sk->sk_err, he _must_
498 * change state of the socket from TCP_SYN_*.
499 * Connect() does not allow to get error notifications
500 * without closing the socket.
502 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
504 timeo = schedule_timeout(timeo);
506 if (signal_pending(current) || !timeo)
508 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
510 finish_wait(sk->sk_sleep, &wait);
515 * Connect to a remote host. There is regrettably still a little
516 * TCP 'magic' in here.
518 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
519 int addr_len, int flags)
521 struct sock *sk = sock->sk;
527 if (uaddr->sa_family == AF_UNSPEC) {
528 err = sk->sk_prot->disconnect(sk, flags);
529 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
533 switch (sock->state) {
542 /* Fall out of switch with err, set for this state */
546 if (sk->sk_state != TCP_CLOSE)
549 err = sk->sk_prot->connect(sk, uaddr, addr_len);
553 sock->state = SS_CONNECTING;
555 /* Just entered SS_CONNECTING state; the only
556 * difference is that return value in non-blocking
557 * case is EINPROGRESS, rather than EALREADY.
563 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
565 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
566 /* Error code is set above */
567 if (!timeo || !inet_wait_for_connect(sk, timeo))
570 err = sock_intr_errno(timeo);
571 if (signal_pending(current))
575 /* Connection was closed by RST, timeout, ICMP error
576 * or another process disconnected us.
578 if (sk->sk_state == TCP_CLOSE)
581 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
582 * and error was received after socket entered established state.
583 * Hence, it is handled normally after connect() return successfully.
586 sock->state = SS_CONNECTED;
593 err = sock_error(sk) ? : -ECONNABORTED;
594 sock->state = SS_UNCONNECTED;
595 if (sk->sk_prot->disconnect(sk, flags))
596 sock->state = SS_DISCONNECTING;
601 * Accept a pending connection. The TCP layer now gives BSD semantics.
604 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
606 struct sock *sk1 = sock->sk;
608 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
615 BUG_TRAP((1 << sk2->sk_state) &
616 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
618 sock_graft(sk2, newsock);
620 newsock->state = SS_CONNECTED;
629 * This does both peername and sockname.
631 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
632 int *uaddr_len, int peer)
634 struct sock *sk = sock->sk;
635 struct inet_sock *inet = inet_sk(sk);
636 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
638 sin->sin_family = AF_INET;
641 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
644 sin->sin_port = inet->dport;
645 sin->sin_addr.s_addr = inet->daddr;
647 __be32 addr = inet->rcv_saddr;
650 sin->sin_port = inet->sport;
651 sin->sin_addr.s_addr = addr;
653 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
654 *uaddr_len = sizeof(*sin);
658 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
661 struct sock *sk = sock->sk;
663 /* We may need to bind the socket. */
664 if (!inet_sk(sk)->num && inet_autobind(sk))
667 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
671 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
673 struct sock *sk = sock->sk;
675 /* We may need to bind the socket. */
676 if (!inet_sk(sk)->num && inet_autobind(sk))
679 if (sk->sk_prot->sendpage)
680 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
681 return sock_no_sendpage(sock, page, offset, size, flags);
685 int inet_shutdown(struct socket *sock, int how)
687 struct sock *sk = sock->sk;
690 /* This should really check to make sure
691 * the socket is a TCP socket. (WHY AC...)
693 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
696 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
700 if (sock->state == SS_CONNECTING) {
701 if ((1 << sk->sk_state) &
702 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
703 sock->state = SS_DISCONNECTING;
705 sock->state = SS_CONNECTED;
708 switch (sk->sk_state) {
711 /* Hack to wake up other listeners, who can poll for
712 POLLHUP, even on eg. unconnected UDP sockets -- RR */
714 sk->sk_shutdown |= how;
715 if (sk->sk_prot->shutdown)
716 sk->sk_prot->shutdown(sk, how);
719 /* Remaining two branches are temporary solution for missing
720 * close() in multithreaded environment. It is _not_ a good idea,
721 * but we have no choice until close() is repaired at VFS level.
724 if (!(how & RCV_SHUTDOWN))
728 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
729 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
733 /* Wake up anyone sleeping in poll. */
734 sk->sk_state_change(sk);
740 * ioctl() calls you can issue on an INET socket. Most of these are
741 * device configuration and stuff and very rarely used. Some ioctls
742 * pass on to the socket itself.
744 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
745 * loads the devconfigure module does its configuring and unloads it.
746 * There's a good 20K of config code hanging around the kernel.
749 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
751 struct sock *sk = sock->sk;
756 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
761 err = ip_rt_ioctl(cmd, (void __user *)arg);
766 err = arp_ioctl(cmd, (void __user *)arg);
779 err = devinet_ioctl(cmd, (void __user *)arg);
782 if (sk->sk_prot->ioctl)
783 err = sk->sk_prot->ioctl(sk, cmd, arg);
791 const struct proto_ops inet_stream_ops = {
793 .owner = THIS_MODULE,
794 .release = inet_release,
796 .connect = inet_stream_connect,
797 .socketpair = sock_no_socketpair,
798 .accept = inet_accept,
799 .getname = inet_getname,
802 .listen = inet_listen,
803 .shutdown = inet_shutdown,
804 .setsockopt = sock_common_setsockopt,
805 .getsockopt = sock_common_getsockopt,
806 .sendmsg = inet_sendmsg,
807 .recvmsg = sock_common_recvmsg,
808 .mmap = sock_no_mmap,
809 .sendpage = tcp_sendpage,
811 .compat_setsockopt = compat_sock_common_setsockopt,
812 .compat_getsockopt = compat_sock_common_getsockopt,
816 const struct proto_ops inet_dgram_ops = {
818 .owner = THIS_MODULE,
819 .release = inet_release,
821 .connect = inet_dgram_connect,
822 .socketpair = sock_no_socketpair,
823 .accept = sock_no_accept,
824 .getname = inet_getname,
827 .listen = sock_no_listen,
828 .shutdown = inet_shutdown,
829 .setsockopt = sock_common_setsockopt,
830 .getsockopt = sock_common_getsockopt,
831 .sendmsg = inet_sendmsg,
832 .recvmsg = sock_common_recvmsg,
833 .mmap = sock_no_mmap,
834 .sendpage = inet_sendpage,
836 .compat_setsockopt = compat_sock_common_setsockopt,
837 .compat_getsockopt = compat_sock_common_getsockopt,
842 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
845 static const struct proto_ops inet_sockraw_ops = {
847 .owner = THIS_MODULE,
848 .release = inet_release,
850 .connect = inet_dgram_connect,
851 .socketpair = sock_no_socketpair,
852 .accept = sock_no_accept,
853 .getname = inet_getname,
854 .poll = datagram_poll,
856 .listen = sock_no_listen,
857 .shutdown = inet_shutdown,
858 .setsockopt = sock_common_setsockopt,
859 .getsockopt = sock_common_getsockopt,
860 .sendmsg = inet_sendmsg,
861 .recvmsg = sock_common_recvmsg,
862 .mmap = sock_no_mmap,
863 .sendpage = inet_sendpage,
865 .compat_setsockopt = compat_sock_common_setsockopt,
866 .compat_getsockopt = compat_sock_common_getsockopt,
870 static struct net_proto_family inet_family_ops = {
872 .create = inet_create,
873 .owner = THIS_MODULE,
876 /* Upon startup we insert all the elements in inetsw_array[] into
877 * the linked list inetsw.
879 static struct inet_protosw inetsw_array[] =
883 .protocol = IPPROTO_TCP,
885 .ops = &inet_stream_ops,
888 .flags = INET_PROTOSW_PERMANENT |
894 .protocol = IPPROTO_UDP,
896 .ops = &inet_dgram_ops,
898 .no_check = UDP_CSUM_DEFAULT,
899 .flags = INET_PROTOSW_PERMANENT,
905 .protocol = IPPROTO_IP, /* wild card */
907 .ops = &inet_sockraw_ops,
908 .capability = CAP_NET_RAW,
909 .no_check = UDP_CSUM_DEFAULT,
910 .flags = INET_PROTOSW_REUSE,
914 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
916 void inet_register_protosw(struct inet_protosw *p)
918 struct list_head *lh;
919 struct inet_protosw *answer;
920 int protocol = p->protocol;
921 struct list_head *last_perm;
923 spin_lock_bh(&inetsw_lock);
925 if (p->type >= SOCK_MAX)
928 /* If we are trying to override a permanent protocol, bail. */
930 last_perm = &inetsw[p->type];
931 list_for_each(lh, &inetsw[p->type]) {
932 answer = list_entry(lh, struct inet_protosw, list);
934 /* Check only the non-wild match. */
935 if (INET_PROTOSW_PERMANENT & answer->flags) {
936 if (protocol == answer->protocol)
946 /* Add the new entry after the last permanent entry if any, so that
947 * the new entry does not override a permanent entry when matched with
948 * a wild-card protocol. But it is allowed to override any existing
949 * non-permanent entry. This means that when we remove this entry, the
950 * system automatically returns to the old behavior.
952 list_add_rcu(&p->list, last_perm);
954 spin_unlock_bh(&inetsw_lock);
961 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
967 "Ignoring attempt to register invalid socket type %d.\n",
972 void inet_unregister_protosw(struct inet_protosw *p)
974 if (INET_PROTOSW_PERMANENT & p->flags) {
976 "Attempt to unregister permanent protocol %d.\n",
979 spin_lock_bh(&inetsw_lock);
980 list_del_rcu(&p->list);
981 spin_unlock_bh(&inetsw_lock);
988 * Shall we try to damage output packets if routing dev changes?
991 int sysctl_ip_dynaddr __read_mostly;
993 static int inet_sk_reselect_saddr(struct sock *sk)
995 struct inet_sock *inet = inet_sk(sk);
998 __be32 old_saddr = inet->saddr;
1000 __be32 daddr = inet->daddr;
1002 if (inet->opt && inet->opt->srr)
1003 daddr = inet->opt->faddr;
1005 /* Query new route. */
1006 err = ip_route_connect(&rt, daddr, 0,
1008 sk->sk_bound_dev_if,
1010 inet->sport, inet->dport, sk, 0);
1014 sk_setup_caps(sk, &rt->u.dst);
1016 new_saddr = rt->rt_src;
1018 if (new_saddr == old_saddr)
1021 if (sysctl_ip_dynaddr > 1) {
1022 printk(KERN_INFO "%s(): shifting inet->"
1023 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1026 NIPQUAD(new_saddr));
1029 inet->saddr = inet->rcv_saddr = new_saddr;
1032 * XXX The only one ugly spot where we need to
1033 * XXX really change the sockets identity after
1034 * XXX it has entered the hashes. -DaveM
1036 * Besides that, it does not check for connection
1037 * uniqueness. Wait for troubles.
1039 __sk_prot_rehash(sk);
1043 int inet_sk_rebuild_header(struct sock *sk)
1045 struct inet_sock *inet = inet_sk(sk);
1046 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1050 /* Route is OK, nothing to do. */
1055 daddr = inet->daddr;
1056 if (inet->opt && inet->opt->srr)
1057 daddr = inet->opt->faddr;
1060 .oif = sk->sk_bound_dev_if,
1064 .saddr = inet->saddr,
1065 .tos = RT_CONN_FLAGS(sk),
1068 .proto = sk->sk_protocol,
1071 .sport = inet->sport,
1072 .dport = inet->dport,
1077 security_sk_classify_flow(sk, &fl);
1078 err = ip_route_output_flow(&rt, &fl, sk, 0);
1081 sk_setup_caps(sk, &rt->u.dst);
1083 /* Routing failed... */
1084 sk->sk_route_caps = 0;
1086 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1087 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1089 if (!sysctl_ip_dynaddr ||
1090 sk->sk_state != TCP_SYN_SENT ||
1091 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1092 (err = inet_sk_reselect_saddr(sk)) != 0)
1093 sk->sk_err_soft = -err;
1099 EXPORT_SYMBOL(inet_sk_rebuild_header);
1101 static int inet_gso_send_check(struct sk_buff *skb)
1104 struct net_protocol *ops;
1109 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1114 if (ihl < sizeof(*iph))
1117 if (unlikely(!pskb_may_pull(skb, ihl)))
1120 skb->h.raw = __skb_pull(skb, ihl);
1122 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1123 err = -EPROTONOSUPPORT;
1126 ops = rcu_dereference(inet_protos[proto]);
1127 if (likely(ops && ops->gso_send_check))
1128 err = ops->gso_send_check(skb);
1135 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1137 struct sk_buff *segs = ERR_PTR(-EINVAL);
1139 struct net_protocol *ops;
1144 if (unlikely(skb_shinfo(skb)->gso_type &
1152 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1157 if (ihl < sizeof(*iph))
1160 if (unlikely(!pskb_may_pull(skb, ihl)))
1163 skb->h.raw = __skb_pull(skb, ihl);
1165 id = ntohs(iph->id);
1166 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1167 segs = ERR_PTR(-EPROTONOSUPPORT);
1170 ops = rcu_dereference(inet_protos[proto]);
1171 if (likely(ops && ops->gso_segment))
1172 segs = ops->gso_segment(skb, features);
1175 if (!segs || unlikely(IS_ERR(segs)))
1181 iph->id = htons(id++);
1182 iph->tot_len = htons(skb->len - skb->mac_len);
1184 iph->check = ip_fast_csum(skb->nh.raw, iph->ihl);
1185 } while ((skb = skb->next));
1191 #ifdef CONFIG_IP_MULTICAST
1192 static struct net_protocol igmp_protocol = {
1193 .handler = igmp_rcv,
1197 static struct net_protocol tcp_protocol = {
1198 .handler = tcp_v4_rcv,
1199 .err_handler = tcp_v4_err,
1200 .gso_send_check = tcp_v4_gso_send_check,
1201 .gso_segment = tcp_tso_segment,
1205 static struct net_protocol udp_protocol = {
1207 .err_handler = udp_err,
1211 static struct net_protocol icmp_protocol = {
1212 .handler = icmp_rcv,
1215 static int __init init_ipv4_mibs(void)
1217 net_statistics[0] = alloc_percpu(struct linux_mib);
1218 net_statistics[1] = alloc_percpu(struct linux_mib);
1219 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1220 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1221 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1222 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1223 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1224 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1225 udp_statistics[0] = alloc_percpu(struct udp_mib);
1226 udp_statistics[1] = alloc_percpu(struct udp_mib);
1227 udplite_statistics[0] = alloc_percpu(struct udp_mib);
1228 udplite_statistics[1] = alloc_percpu(struct udp_mib);
1230 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1231 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1232 && udp_statistics[0] && udp_statistics[1]
1233 && udplite_statistics[0] && udplite_statistics[1] ) )
1236 (void) tcp_mib_init();
1241 static int ipv4_proc_init(void);
1244 * IP protocol layer initialiser
1247 static struct packet_type ip_packet_type = {
1248 .type = __constant_htons(ETH_P_IP),
1250 .gso_send_check = inet_gso_send_check,
1251 .gso_segment = inet_gso_segment,
1254 static int __init inet_init(void)
1256 struct sk_buff *dummy_skb;
1257 struct inet_protosw *q;
1258 struct list_head *r;
1261 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1263 rc = proto_register(&tcp_prot, 1);
1267 rc = proto_register(&udp_prot, 1);
1269 goto out_unregister_tcp_proto;
1271 rc = proto_register(&raw_prot, 1);
1273 goto out_unregister_udp_proto;
1276 * Tell SOCKET that we are alive...
1279 (void)sock_register(&inet_family_ops);
1282 * Add all the base protocols.
1285 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1286 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1287 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1288 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1289 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1290 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1291 #ifdef CONFIG_IP_MULTICAST
1292 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1293 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1296 /* Register the socket-side information for inet_create. */
1297 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1300 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1301 inet_register_protosw(q);
1304 * Set the ARP module up
1310 * Set the IP module up
1315 tcp_v4_init(&inet_family_ops);
1317 /* Setup TCP slab cache for open requests. */
1320 /* Add UDP-Lite (RFC 3828) */
1321 udplite4_register();
1324 * Set the ICMP layer up
1327 icmp_init(&inet_family_ops);
1330 * Initialise the multicast router
1332 #if defined(CONFIG_IP_MROUTE)
1336 * Initialise per-cpu ipv4 mibs
1339 if(init_ipv4_mibs())
1340 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1346 dev_add_pack(&ip_packet_type);
1351 out_unregister_udp_proto:
1352 proto_unregister(&udp_prot);
1353 out_unregister_tcp_proto:
1354 proto_unregister(&tcp_prot);
1358 fs_initcall(inet_init);
1360 /* ------------------------------------------------------------------------ */
1362 #ifdef CONFIG_PROC_FS
1363 static int __init ipv4_proc_init(void)
1367 if (raw_proc_init())
1369 if (tcp4_proc_init())
1371 if (udp4_proc_init())
1373 if (fib_proc_init())
1375 if (ip_misc_proc_init())
1392 #else /* CONFIG_PROC_FS */
1393 static int __init ipv4_proc_init(void)
1397 #endif /* CONFIG_PROC_FS */
1399 MODULE_ALIAS_NETPROTO(PF_INET);
1401 EXPORT_SYMBOL(inet_accept);
1402 EXPORT_SYMBOL(inet_bind);
1403 EXPORT_SYMBOL(inet_dgram_connect);
1404 EXPORT_SYMBOL(inet_dgram_ops);
1405 EXPORT_SYMBOL(inet_getname);
1406 EXPORT_SYMBOL(inet_ioctl);
1407 EXPORT_SYMBOL(inet_listen);
1408 EXPORT_SYMBOL(inet_register_protosw);
1409 EXPORT_SYMBOL(inet_release);
1410 EXPORT_SYMBOL(inet_sendmsg);
1411 EXPORT_SYMBOL(inet_shutdown);
1412 EXPORT_SYMBOL(inet_sock_destruct);
1413 EXPORT_SYMBOL(inet_stream_connect);
1414 EXPORT_SYMBOL(inet_stream_ops);
1415 EXPORT_SYMBOL(inet_unregister_protosw);
1416 EXPORT_SYMBOL(net_statistics);
1417 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);