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 $
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
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/config.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
75 #include <linux/kernel.h>
76 #include <linux/major.h>
77 #include <linux/module.h>
78 #include <linux/sched.h>
79 #include <linux/timer.h>
80 #include <linux/string.h>
81 #include <linux/sockios.h>
82 #include <linux/net.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/netdevice.h>
99 #include <net/protocol.h>
101 #include <net/route.h>
102 #include <net/ip_fib.h>
105 #include <linux/skbuff.h>
106 #include <net/sock.h>
108 #include <net/icmp.h>
109 #include <net/ipip.h>
110 #include <net/inet_common.h>
111 #include <net/xfrm.h>
112 #ifdef CONFIG_IP_MROUTE
113 #include <linux/mroute.h>
115 #include <linux/vs_limit.h>
117 DEFINE_SNMP_STAT(struct linux_mib, net_statistics);
119 #ifdef INET_REFCNT_DEBUG
120 atomic_t inet_sock_nr;
123 extern void ip_mc_drop_socket(struct sock *sk);
125 /* The inetsw table contains everything that inet_create needs to
126 * build a new socket.
128 static struct list_head inetsw[SOCK_MAX];
129 static spinlock_t inetsw_lock = SPIN_LOCK_UNLOCKED;
131 /* New destruction routine */
133 void inet_sock_destruct(struct sock *sk)
135 struct inet_opt *inet = inet_sk(sk);
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
140 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
141 printk("Attempt to release TCP socket in state %d %p\n",
145 if (!sock_flag(sk, SOCK_DEAD)) {
146 printk("Attempt to release alive inet socket %p\n", sk);
150 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
151 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
152 BUG_TRAP(!sk->sk_wmem_queued);
153 BUG_TRAP(!sk->sk_forward_alloc);
159 clr_vx_info(&sk->sk_vx_info);
161 clr_nx_info(&sk->sk_nx_info);
164 dst_release(sk->sk_dst_cache);
165 #ifdef INET_REFCNT_DEBUG
166 atomic_dec(&inet_sock_nr);
167 printk(KERN_DEBUG "INET socket %p released, %d are still alive\n",
168 sk, atomic_read(&inet_sock_nr));
173 * The routines beyond this point handle the behaviour of an AF_INET
174 * socket object. Mostly it punts to the subprotocols of IP to do
179 * Automatically bind an unbound socket.
182 static int inet_autobind(struct sock *sk)
184 struct inet_opt *inet;
185 /* We may need to bind the socket. */
189 if (sk->sk_prot->get_port(sk, 0)) {
193 inet->sport = htons(inet->num);
200 * Move a socket into listening state.
202 int inet_listen(struct socket *sock, int backlog)
204 struct sock *sk = sock->sk;
205 unsigned char old_state;
211 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
214 old_state = sk->sk_state;
215 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
218 /* Really, if the socket is already in listen state
219 * we can only allow the backlog to be adjusted.
221 if (old_state != TCP_LISTEN) {
222 err = tcp_listen_start(sk);
226 sk->sk_max_ack_backlog = backlog;
235 * Create an inet socket.
238 static int inet_create(struct socket *sock, int protocol)
242 struct inet_protosw *answer;
243 struct inet_opt *inet;
244 struct proto *answer_prot;
245 unsigned char answer_flags;
246 char answer_no_check;
249 sock->state = SS_UNCONNECTED;
251 /* Look for the requested type/protocol pair. */
254 list_for_each_rcu(p, &inetsw[sock->type]) {
255 answer = list_entry(p, struct inet_protosw, list);
257 /* Check the non-wild match. */
258 if (protocol == answer->protocol) {
259 if (protocol != IPPROTO_IP)
262 /* Check for the two wild cases. */
263 if (IPPROTO_IP == protocol) {
264 protocol = answer->protocol;
267 if (IPPROTO_IP == answer->protocol)
273 err = -ESOCKTNOSUPPORT;
277 if ((protocol == IPPROTO_ICMP) && vx_ccaps(VXC_RAW_ICMP))
279 if (answer->capability > 0 && !capable(answer->capability))
282 err = -EPROTONOSUPPORT;
286 sock->ops = answer->ops;
287 answer_prot = answer->prot;
288 answer_no_check = answer->no_check;
289 answer_flags = answer->flags;
292 BUG_TRAP(answer_prot->slab != NULL);
295 sk = sk_alloc(PF_INET, GFP_KERNEL,
296 answer_prot->slab_obj_size,
302 sk->sk_prot = answer_prot;
303 sk->sk_no_check = answer_no_check;
304 if (INET_PROTOSW_REUSE & answer_flags)
309 if (SOCK_RAW == sock->type) {
310 inet->num = protocol;
311 if (IPPROTO_RAW == protocol)
315 if (ipv4_config.no_pmtu_disc)
316 inet->pmtudisc = IP_PMTUDISC_DONT;
318 inet->pmtudisc = IP_PMTUDISC_WANT;
322 sock_init_data(sock, sk);
323 sk_set_owner(sk, sk->sk_prot->owner);
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;
330 set_vx_info(&sk->sk_vx_info, current->vx_info);
331 sk->sk_xid = vx_current_xid();
333 set_nx_info(&sk->sk_nx_info, current->nx_info);
334 sk->sk_nid = nx_current_nid();
340 inet->mc_list = NULL;
342 #ifdef INET_REFCNT_DEBUG
343 atomic_inc(&inet_sock_nr);
347 /* It assumes that any protocol which allows
348 * the user to assign a number at socket
349 * creation time automatically
352 inet->sport = htons(inet->num);
353 /* Add to protocol hash chains. */
354 sk->sk_prot->hash(sk);
357 if (sk->sk_prot->init) {
358 err = sk->sk_prot->init(sk);
360 sk_common_release(sk);
371 * The peer socket should always be NULL (or else). When we call this
372 * function we are destroying the object and from then on nobody
373 * should refer to it.
375 int inet_release(struct socket *sock)
377 struct sock *sk = sock->sk;
382 /* Applications forget to leave groups before exiting */
383 ip_mc_drop_socket(sk);
385 /* If linger is set, we don't return until the close
386 * is complete. Otherwise we return immediately. The
387 * actually closing is done the same either way.
389 * If the close is due to the process exiting, we never
393 if (sock_flag(sk, SOCK_LINGER) &&
394 !(current->flags & PF_EXITING))
395 timeout = sk->sk_lingertime;
398 clr_vx_info(&sk->sk_vx_info);
400 clr_nx_info(&sk->sk_nx_info);
402 sk->sk_prot->close(sk, timeout);
407 /* It is off by default, see below. */
408 int sysctl_ip_nonlocal_bind;
410 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
412 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
413 struct sock *sk = sock->sk;
414 struct inet_opt *inet = inet_sk(sk);
419 /* If the socket has its own bind function then use it. (RAW) */
420 if (sk->sk_prot->bind) {
421 err = sk->sk_prot->bind(sk, uaddr, addr_len);
425 if (addr_len < sizeof(struct sockaddr_in))
428 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
430 /* Not specified by any standard per-se, however it breaks too
431 * many applications when removed. It is unfortunate since
432 * allowing applications to make a non-local bind solves
433 * several problems with systems using dynamic addressing.
434 * (ie. your servers still start up even if your ISDN link
435 * is temporarily down)
437 err = -EADDRNOTAVAIL;
438 if (!sysctl_ip_nonlocal_bind &&
440 addr->sin_addr.s_addr != INADDR_ANY &&
441 chk_addr_ret != RTN_LOCAL &&
442 chk_addr_ret != RTN_MULTICAST &&
443 chk_addr_ret != RTN_BROADCAST)
446 snum = ntohs(addr->sin_port);
448 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
451 /* We keep a pair of addresses. rcv_saddr is the one
452 * used by hash lookups, and saddr is used for transmit.
454 * In the BSD API these are the same except where it
455 * would be illegal to use them (multicast/broadcast) in
456 * which case the sending device address is used.
460 /* Check these errors (active socket, double bind). */
462 if (sk->sk_state != TCP_CLOSE || inet->num)
463 goto out_release_sock;
465 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
466 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
467 inet->saddr = 0; /* Use device */
469 /* Make sure we are allowed to bind here. */
470 if (sk->sk_prot->get_port(sk, snum)) {
471 inet->saddr = inet->rcv_saddr = 0;
473 goto out_release_sock;
477 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
479 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
480 inet->sport = htons(inet->num);
491 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
492 int addr_len, int flags)
494 struct sock *sk = sock->sk;
496 if (uaddr->sa_family == AF_UNSPEC)
497 return sk->sk_prot->disconnect(sk, flags);
499 if (!inet_sk(sk)->num && inet_autobind(sk))
501 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
504 static long inet_wait_for_connect(struct sock *sk, long timeo)
508 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
510 /* Basic assumption: if someone sets sk->sk_err, he _must_
511 * change state of the socket from TCP_SYN_*.
512 * Connect() does not allow to get error notifications
513 * without closing the socket.
515 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
517 timeo = schedule_timeout(timeo);
519 if (signal_pending(current) || !timeo)
521 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
523 finish_wait(sk->sk_sleep, &wait);
528 * Connect to a remote host. There is regrettably still a little
529 * TCP 'magic' in here.
531 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
532 int addr_len, int flags)
534 struct sock *sk = sock->sk;
540 if (uaddr->sa_family == AF_UNSPEC) {
541 err = sk->sk_prot->disconnect(sk, flags);
542 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
546 switch (sock->state) {
555 /* Fall out of switch with err, set for this state */
559 if (sk->sk_state != TCP_CLOSE)
562 err = sk->sk_prot->connect(sk, uaddr, addr_len);
566 sock->state = SS_CONNECTING;
568 /* Just entered SS_CONNECTING state; the only
569 * difference is that return value in non-blocking
570 * case is EINPROGRESS, rather than EALREADY.
576 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
578 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
579 /* Error code is set above */
580 if (!timeo || !inet_wait_for_connect(sk, timeo))
583 err = sock_intr_errno(timeo);
584 if (signal_pending(current))
588 /* Connection was closed by RST, timeout, ICMP error
589 * or another process disconnected us.
591 if (sk->sk_state == TCP_CLOSE)
594 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
595 * and error was received after socket entered established state.
596 * Hence, it is handled normally after connect() return successfully.
599 sock->state = SS_CONNECTED;
606 err = sock_error(sk) ? : -ECONNABORTED;
607 sock->state = SS_UNCONNECTED;
608 if (sk->sk_prot->disconnect(sk, flags))
609 sock->state = SS_DISCONNECTING;
614 * Accept a pending connection. The TCP layer now gives BSD semantics.
617 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
619 struct sock *sk1 = sock->sk;
621 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
628 BUG_TRAP((1 << sk2->sk_state) &
629 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
631 sock_graft(sk2, newsock);
633 newsock->state = SS_CONNECTED;
642 * This does both peername and sockname.
644 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
645 int *uaddr_len, int peer)
647 struct sock *sk = sock->sk;
648 struct inet_opt *inet = inet_sk(sk);
649 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
651 sin->sin_family = AF_INET;
654 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
657 sin->sin_port = inet->dport;
658 sin->sin_addr.s_addr = inet->daddr;
660 __u32 addr = inet->rcv_saddr;
663 sin->sin_port = inet->sport;
664 sin->sin_addr.s_addr = addr;
666 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
667 *uaddr_len = sizeof(*sin);
671 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
674 struct sock *sk = sock->sk;
676 /* We may need to bind the socket. */
677 if (!inet_sk(sk)->num && inet_autobind(sk))
680 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
684 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
686 struct sock *sk = sock->sk;
688 /* We may need to bind the socket. */
689 if (!inet_sk(sk)->num && inet_autobind(sk))
692 if (sk->sk_prot->sendpage)
693 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
694 return sock_no_sendpage(sock, page, offset, size, flags);
698 int inet_shutdown(struct socket *sock, int how)
700 struct sock *sk = sock->sk;
703 /* This should really check to make sure
704 * the socket is a TCP socket. (WHY AC...)
706 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
709 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
713 if (sock->state == SS_CONNECTING) {
714 if ((1 << sk->sk_state) &
715 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
716 sock->state = SS_DISCONNECTING;
718 sock->state = SS_CONNECTED;
721 switch (sk->sk_state) {
724 /* Hack to wake up other listeners, who can poll for
725 POLLHUP, even on eg. unconnected UDP sockets -- RR */
727 sk->sk_shutdown |= how;
728 if (sk->sk_prot->shutdown)
729 sk->sk_prot->shutdown(sk, how);
732 /* Remaining two branches are temporary solution for missing
733 * close() in multithreaded environment. It is _not_ a good idea,
734 * but we have no choice until close() is repaired at VFS level.
737 if (!(how & RCV_SHUTDOWN))
741 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
742 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
746 /* Wake up anyone sleeping in poll. */
747 sk->sk_state_change(sk);
753 * ioctl() calls you can issue on an INET socket. Most of these are
754 * device configuration and stuff and very rarely used. Some ioctls
755 * pass on to the socket itself.
757 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
758 * loads the devconfigure module does its configuring and unloads it.
759 * There's a good 20K of config code hanging around the kernel.
762 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
764 struct sock *sk = sock->sk;
769 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
774 err = ip_rt_ioctl(cmd, (void __user *)arg);
779 err = arp_ioctl(cmd, (void __user *)arg);
792 err = devinet_ioctl(cmd, (void __user *)arg);
795 if (!sk->sk_prot->ioctl ||
796 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
798 err = dev_ioctl(cmd, (void __user *)arg);
804 struct proto_ops inet_stream_ops = {
806 .owner = THIS_MODULE,
807 .release = inet_release,
809 .connect = inet_stream_connect,
810 .socketpair = sock_no_socketpair,
811 .accept = inet_accept,
812 .getname = inet_getname,
815 .listen = inet_listen,
816 .shutdown = inet_shutdown,
817 .setsockopt = sock_common_setsockopt,
818 .getsockopt = sock_common_getsockopt,
819 .sendmsg = inet_sendmsg,
820 .recvmsg = sock_common_recvmsg,
821 .mmap = sock_no_mmap,
822 .sendpage = tcp_sendpage
825 struct proto_ops inet_dgram_ops = {
827 .owner = THIS_MODULE,
828 .release = inet_release,
830 .connect = inet_dgram_connect,
831 .socketpair = sock_no_socketpair,
832 .accept = sock_no_accept,
833 .getname = inet_getname,
836 .listen = sock_no_listen,
837 .shutdown = inet_shutdown,
838 .setsockopt = sock_common_setsockopt,
839 .getsockopt = sock_common_getsockopt,
840 .sendmsg = inet_sendmsg,
841 .recvmsg = sock_common_recvmsg,
842 .mmap = sock_no_mmap,
843 .sendpage = inet_sendpage,
847 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
850 static struct proto_ops inet_sockraw_ops = {
852 .owner = THIS_MODULE,
853 .release = inet_release,
855 .connect = inet_dgram_connect,
856 .socketpair = sock_no_socketpair,
857 .accept = sock_no_accept,
858 .getname = inet_getname,
859 .poll = datagram_poll,
861 .listen = sock_no_listen,
862 .shutdown = inet_shutdown,
863 .setsockopt = sock_common_setsockopt,
864 .getsockopt = sock_common_getsockopt,
865 .sendmsg = inet_sendmsg,
866 .recvmsg = sock_common_recvmsg,
867 .mmap = sock_no_mmap,
868 .sendpage = inet_sendpage,
871 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
873 EXPORT_SYMBOL(vnet_active);
874 struct net_proto_family inet_family_ops;
875 EXPORT_SYMBOL(inet_family_ops);
879 struct net_proto_family inet_family_ops = {
881 .create = inet_create,
882 .owner = THIS_MODULE,
886 extern void tcp_init(void);
887 extern void tcp_v4_init(struct net_proto_family *);
889 /* Upon startup we insert all the elements in inetsw_array[] into
890 * the linked list inetsw.
892 static struct inet_protosw inetsw_array[] =
896 .protocol = IPPROTO_TCP,
898 .ops = &inet_stream_ops,
901 .flags = INET_PROTOSW_PERMANENT,
906 .protocol = IPPROTO_UDP,
908 .ops = &inet_dgram_ops,
910 .no_check = UDP_CSUM_DEFAULT,
911 .flags = INET_PROTOSW_PERMANENT,
917 .protocol = IPPROTO_IP, /* wild card */
919 .ops = &inet_sockraw_ops,
920 .capability = CAP_NET_RAW,
921 .no_check = UDP_CSUM_DEFAULT,
922 .flags = INET_PROTOSW_REUSE,
926 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
928 void inet_register_protosw(struct inet_protosw *p)
930 struct list_head *lh;
931 struct inet_protosw *answer;
932 int protocol = p->protocol;
933 struct list_head *last_perm;
935 spin_lock_bh(&inetsw_lock);
937 if (p->type >= SOCK_MAX)
940 /* If we are trying to override a permanent protocol, bail. */
942 last_perm = &inetsw[p->type];
943 list_for_each(lh, &inetsw[p->type]) {
944 answer = list_entry(lh, struct inet_protosw, list);
946 /* Check only the non-wild match. */
947 if (INET_PROTOSW_PERMANENT & answer->flags) {
948 if (protocol == answer->protocol)
958 /* Add the new entry after the last permanent entry if any, so that
959 * the new entry does not override a permanent entry when matched with
960 * a wild-card protocol. But it is allowed to override any existing
961 * non-permanent entry. This means that when we remove this entry, the
962 * system automatically returns to the old behavior.
964 list_add_rcu(&p->list, last_perm);
966 spin_unlock_bh(&inetsw_lock);
973 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
979 "Ignoring attempt to register invalid socket type %d.\n",
984 void inet_unregister_protosw(struct inet_protosw *p)
986 if (INET_PROTOSW_PERMANENT & p->flags) {
988 "Attempt to unregister permanent protocol %d.\n",
991 spin_lock_bh(&inetsw_lock);
992 list_del_rcu(&p->list);
993 spin_unlock_bh(&inetsw_lock);
999 #ifdef CONFIG_IP_MULTICAST
1000 static struct net_protocol igmp_protocol = {
1001 .handler = igmp_rcv,
1005 static struct net_protocol tcp_protocol = {
1006 .handler = tcp_v4_rcv,
1007 .err_handler = tcp_v4_err,
1011 static struct net_protocol udp_protocol = {
1013 .err_handler = udp_err,
1017 static struct net_protocol icmp_protocol = {
1018 .handler = icmp_rcv,
1021 static int __init init_ipv4_mibs(void)
1023 net_statistics[0] = alloc_percpu(struct linux_mib);
1024 net_statistics[1] = alloc_percpu(struct linux_mib);
1025 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1026 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1027 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1028 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1029 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1030 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1031 udp_statistics[0] = alloc_percpu(struct udp_mib);
1032 udp_statistics[1] = alloc_percpu(struct udp_mib);
1034 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1035 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1036 && udp_statistics[0] && udp_statistics[1]))
1039 (void) tcp_mib_init();
1044 int ipv4_proc_init(void);
1045 extern void ipfrag_init(void);
1047 static int __init inet_init(void)
1049 struct sk_buff *dummy_skb;
1050 struct inet_protosw *q;
1051 struct list_head *r;
1054 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1055 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1059 rc = sk_alloc_slab(&tcp_prot, "tcp_sock");
1061 sk_alloc_slab_error(&tcp_prot);
1064 rc = sk_alloc_slab(&udp_prot, "udp_sock");
1066 sk_alloc_slab_error(&udp_prot);
1067 goto out_tcp_free_slab;
1069 rc = sk_alloc_slab(&raw_prot, "raw_sock");
1071 sk_alloc_slab_error(&raw_prot);
1072 goto out_udp_free_slab;
1076 * Tell SOCKET that we are alive...
1079 (void)sock_register(&inet_family_ops);
1082 * Add all the base protocols.
1085 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1086 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1087 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1088 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1089 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1090 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1091 #ifdef CONFIG_IP_MULTICAST
1092 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1093 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1096 /* Register the socket-side information for inet_create. */
1097 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1100 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1101 inet_register_protosw(q);
1104 * Set the ARP module up
1110 * Set the IP module up
1115 tcp_v4_init(&inet_family_ops);
1117 /* Setup TCP slab cache for open requests. */
1122 * Set the ICMP layer up
1125 icmp_init(&inet_family_ops);
1128 * Initialise the multicast router
1130 #if defined(CONFIG_IP_MROUTE)
1134 * Initialise per-cpu ipv4 mibs
1137 if(init_ipv4_mibs())
1138 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1148 sk_free_slab(&tcp_prot);
1150 sk_free_slab(&udp_prot);
1154 module_init(inet_init);
1156 /* ------------------------------------------------------------------------ */
1158 #ifdef CONFIG_PROC_FS
1159 extern int fib_proc_init(void);
1160 extern void fib_proc_exit(void);
1161 extern int ip_misc_proc_init(void);
1162 extern int raw_proc_init(void);
1163 extern void raw_proc_exit(void);
1164 extern int tcp4_proc_init(void);
1165 extern void tcp4_proc_exit(void);
1166 extern int udp4_proc_init(void);
1167 extern void udp4_proc_exit(void);
1169 int __init ipv4_proc_init(void)
1173 if (raw_proc_init())
1175 if (tcp4_proc_init())
1177 if (udp4_proc_init())
1179 if (fib_proc_init())
1181 if (ip_misc_proc_init())
1198 #else /* CONFIG_PROC_FS */
1199 int __init ipv4_proc_init(void)
1203 #endif /* CONFIG_PROC_FS */
1205 MODULE_ALIAS_NETPROTO(PF_INET);
1207 EXPORT_SYMBOL(inet_accept);
1208 EXPORT_SYMBOL(inet_bind);
1209 EXPORT_SYMBOL(inet_dgram_connect);
1210 EXPORT_SYMBOL(inet_dgram_ops);
1211 EXPORT_SYMBOL(inet_getname);
1212 EXPORT_SYMBOL(inet_ioctl);
1213 EXPORT_SYMBOL(inet_listen);
1214 EXPORT_SYMBOL(inet_register_protosw);
1215 EXPORT_SYMBOL(inet_release);
1216 EXPORT_SYMBOL(inet_sendmsg);
1217 EXPORT_SYMBOL(inet_shutdown);
1218 EXPORT_SYMBOL(inet_sock_destruct);
1219 EXPORT_SYMBOL(inet_stream_connect);
1220 EXPORT_SYMBOL(inet_stream_ops);
1221 EXPORT_SYMBOL(inet_unregister_protosw);
1222 EXPORT_SYMBOL(net_statistics);
1224 #ifdef INET_REFCNT_DEBUG
1225 EXPORT_SYMBOL(inet_sock_nr);