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/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/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/fcntl.h>
84 #include <linux/interrupt.h>
85 #include <linux/stat.h>
86 #include <linux/init.h>
87 #include <linux/poll.h>
88 #include <linux/netfilter_ipv4.h>
90 #include <asm/uaccess.h>
91 #include <asm/system.h>
93 #include <linux/smp_lock.h>
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/netdevice.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/ip_fib.h>
104 #include <linux/skbuff.h>
105 #include <net/sock.h>
107 #include <net/icmp.h>
108 #include <net/ipip.h>
109 #include <net/inet_common.h>
110 #include <net/xfrm.h>
111 #ifdef CONFIG_IP_MROUTE
112 #include <linux/mroute.h>
114 #include <linux/vs_limit.h>
116 DEFINE_SNMP_STAT(struct linux_mib, net_statistics);
118 #ifdef INET_REFCNT_DEBUG
119 atomic_t inet_sock_nr;
122 extern void ip_mc_drop_socket(struct sock *sk);
124 /* The inetsw table contains everything that inet_create needs to
125 * build a new socket.
127 static struct list_head inetsw[SOCK_MAX];
128 static DEFINE_SPINLOCK(inetsw_lock);
130 /* New destruction routine */
132 void inet_sock_destruct(struct sock *sk)
134 struct inet_sock *inet = inet_sk(sk);
136 __skb_queue_purge(&sk->sk_receive_queue);
137 __skb_queue_purge(&sk->sk_error_queue);
139 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
140 printk("Attempt to release TCP socket in state %d %p\n",
144 if (!sock_flag(sk, SOCK_DEAD)) {
145 printk("Attempt to release alive inet socket %p\n", sk);
149 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
150 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
151 BUG_TRAP(!sk->sk_wmem_queued);
152 BUG_TRAP(!sk->sk_forward_alloc);
156 dst_release(sk->sk_dst_cache);
157 #ifdef INET_REFCNT_DEBUG
158 atomic_dec(&inet_sock_nr);
159 printk(KERN_DEBUG "INET socket %p released, %d are still alive\n",
160 sk, atomic_read(&inet_sock_nr));
165 * The routines beyond this point handle the behaviour of an AF_INET
166 * socket object. Mostly it punts to the subprotocols of IP to do
171 * Automatically bind an unbound socket.
174 static int inet_autobind(struct sock *sk)
176 struct inet_sock *inet;
177 /* We may need to bind the socket. */
181 if (sk->sk_prot->get_port(sk, 0)) {
185 inet->sport = htons(inet->num);
192 * Move a socket into listening state.
194 int inet_listen(struct socket *sock, int backlog)
196 struct sock *sk = sock->sk;
197 unsigned char old_state;
203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
206 old_state = sk->sk_state;
207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
213 if (old_state != TCP_LISTEN) {
214 err = tcp_listen_start(sk);
218 sk->sk_max_ack_backlog = backlog;
227 * Create an inet socket.
230 static int inet_create(struct socket *sock, int protocol)
234 struct inet_protosw *answer;
235 struct inet_sock *inet;
236 struct proto *answer_prot;
237 unsigned char answer_flags;
238 char answer_no_check;
241 sock->state = SS_UNCONNECTED;
243 /* Look for the requested type/protocol pair. */
246 list_for_each_rcu(p, &inetsw[sock->type]) {
247 answer = list_entry(p, struct inet_protosw, list);
249 /* Check the non-wild match. */
250 if (protocol == answer->protocol) {
251 if (protocol != IPPROTO_IP)
254 /* Check for the two wild cases. */
255 if (IPPROTO_IP == protocol) {
256 protocol = answer->protocol;
259 if (IPPROTO_IP == answer->protocol)
265 err = -ESOCKTNOSUPPORT;
269 if ((protocol == IPPROTO_ICMP) && vx_ccaps(VXC_RAW_ICMP))
271 if (answer->capability > 0 && !capable(answer->capability))
274 err = -EPROTONOSUPPORT;
278 sock->ops = answer->ops;
279 answer_prot = answer->prot;
280 answer_no_check = answer->no_check;
281 answer_flags = answer->flags;
284 BUG_TRAP(answer_prot->slab != NULL);
287 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
292 sk->sk_no_check = answer_no_check;
293 if (INET_PROTOSW_REUSE & answer_flags)
298 if (SOCK_RAW == sock->type) {
299 inet->num = protocol;
300 if (IPPROTO_RAW == protocol)
304 if (ipv4_config.no_pmtu_disc)
305 inet->pmtudisc = IP_PMTUDISC_DONT;
307 inet->pmtudisc = IP_PMTUDISC_WANT;
311 sock_init_data(sock, sk);
313 sk->sk_destruct = inet_sock_destruct;
314 sk->sk_family = PF_INET;
315 sk->sk_protocol = protocol;
316 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
322 inet->mc_list = NULL;
324 #ifdef INET_REFCNT_DEBUG
325 atomic_inc(&inet_sock_nr);
329 /* It assumes that any protocol which allows
330 * the user to assign a number at socket
331 * creation time automatically
334 inet->sport = htons(inet->num);
335 /* Add to protocol hash chains. */
336 sk->sk_prot->hash(sk);
339 if (sk->sk_prot->init) {
340 err = sk->sk_prot->init(sk);
342 sk_common_release(sk);
353 * The peer socket should always be NULL (or else). When we call this
354 * function we are destroying the object and from then on nobody
355 * should refer to it.
357 int inet_release(struct socket *sock)
359 struct sock *sk = sock->sk;
364 /* Applications forget to leave groups before exiting */
365 ip_mc_drop_socket(sk);
367 /* If linger is set, we don't return until the close
368 * is complete. Otherwise we return immediately. The
369 * actually closing is done the same either way.
371 * If the close is due to the process exiting, we never
375 if (sock_flag(sk, SOCK_LINGER) &&
376 !(current->flags & PF_EXITING))
377 timeout = sk->sk_lingertime;
379 sk->sk_prot->close(sk, timeout);
384 /* It is off by default, see below. */
385 int sysctl_ip_nonlocal_bind;
387 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
389 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
390 struct sock *sk = sock->sk;
391 struct inet_sock *inet = inet_sk(sk);
395 __u32 s_addr; /* Address used for validation */
396 __u32 s_addr1; /* Address used for socket */
397 __u32 s_addr2; /* Broadcast address for the socket */
398 struct nx_info *nxi = sk->sk_nx_info;
400 /* If the socket has its own bind function then use it. (RAW) */
401 if (sk->sk_prot->bind) {
402 err = sk->sk_prot->bind(sk, uaddr, addr_len);
406 if (addr_len < sizeof(struct sockaddr_in))
409 s_addr = addr->sin_addr.s_addr;
411 s_addr2 = 0xffffffffl;
413 vxdprintk(VXD_CBIT(net, 3),
414 "inet_bind(%p)* %p,%p;%lx %d.%d.%d.%d",
415 sk, sk->sk_nx_info, sk->sk_socket,
416 (sk->sk_socket?sk->sk_socket->flags:0),
419 __u32 v4_bcast = nxi->v4_bcast;
420 __u32 ipv4root = nxi->ipv4[0];
421 int nbipv4 = nxi->nbipv4;
424 /* bind to any for 1-n */
426 s_addr1 = (nbipv4 > 1) ? 0 : s_addr;
428 } else if (s_addr == 0x0100007f) {
429 /* rewrite localhost to ipv4root */
432 } else if (s_addr != v4_bcast) {
433 /* normal address bind */
434 if (!addr_in_nx_info(nxi, s_addr))
435 return -EADDRNOTAVAIL;
438 chk_addr_ret = inet_addr_type(s_addr);
440 vxdprintk(VXD_CBIT(net, 3),
441 "inet_bind(%p) %d.%d.%d.%d, %d.%d.%d.%d, %d.%d.%d.%d",
442 sk, VXD_QUAD(s_addr), VXD_QUAD(s_addr1), VXD_QUAD(s_addr2));
444 /* Not specified by any standard per-se, however it breaks too
445 * many applications when removed. It is unfortunate since
446 * allowing applications to make a non-local bind solves
447 * several problems with systems using dynamic addressing.
448 * (ie. your servers still start up even if your ISDN link
449 * is temporarily down)
451 err = -EADDRNOTAVAIL;
452 if (!sysctl_ip_nonlocal_bind &&
454 s_addr != INADDR_ANY &&
455 chk_addr_ret != RTN_LOCAL &&
456 chk_addr_ret != RTN_MULTICAST &&
457 chk_addr_ret != RTN_BROADCAST)
460 snum = ntohs(addr->sin_port);
462 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
465 /* We keep a pair of addresses. rcv_saddr is the one
466 * used by hash lookups, and saddr is used for transmit.
468 * In the BSD API these are the same except where it
469 * would be illegal to use them (multicast/broadcast) in
470 * which case the sending device address is used.
474 /* Check these errors (active socket, double bind). */
476 if (sk->sk_state != TCP_CLOSE || inet->num)
477 goto out_release_sock;
479 inet->rcv_saddr = inet->saddr = s_addr1;
480 inet->rcv_saddr2 = s_addr2;
481 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
482 inet->saddr = 0; /* Use device */
484 /* Make sure we are allowed to bind here. */
485 if (sk->sk_prot->get_port(sk, snum)) {
486 inet->saddr = inet->rcv_saddr = 0;
488 goto out_release_sock;
492 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
494 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
495 inet->sport = htons(inet->num);
506 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
507 int addr_len, int flags)
509 struct sock *sk = sock->sk;
511 if (uaddr->sa_family == AF_UNSPEC)
512 return sk->sk_prot->disconnect(sk, flags);
514 if (!inet_sk(sk)->num && inet_autobind(sk))
516 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
519 static long inet_wait_for_connect(struct sock *sk, long timeo)
523 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
525 /* Basic assumption: if someone sets sk->sk_err, he _must_
526 * change state of the socket from TCP_SYN_*.
527 * Connect() does not allow to get error notifications
528 * without closing the socket.
530 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
532 timeo = schedule_timeout(timeo);
534 if (signal_pending(current) || !timeo)
536 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
538 finish_wait(sk->sk_sleep, &wait);
543 * Connect to a remote host. There is regrettably still a little
544 * TCP 'magic' in here.
546 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
547 int addr_len, int flags)
549 struct sock *sk = sock->sk;
555 if (uaddr->sa_family == AF_UNSPEC) {
556 err = sk->sk_prot->disconnect(sk, flags);
557 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
561 switch (sock->state) {
570 /* Fall out of switch with err, set for this state */
574 if (sk->sk_state != TCP_CLOSE)
577 err = sk->sk_prot->connect(sk, uaddr, addr_len);
581 sock->state = SS_CONNECTING;
583 /* Just entered SS_CONNECTING state; the only
584 * difference is that return value in non-blocking
585 * case is EINPROGRESS, rather than EALREADY.
591 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
593 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
594 /* Error code is set above */
595 if (!timeo || !inet_wait_for_connect(sk, timeo))
598 err = sock_intr_errno(timeo);
599 if (signal_pending(current))
603 /* Connection was closed by RST, timeout, ICMP error
604 * or another process disconnected us.
606 if (sk->sk_state == TCP_CLOSE)
609 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
610 * and error was received after socket entered established state.
611 * Hence, it is handled normally after connect() return successfully.
614 sock->state = SS_CONNECTED;
621 err = sock_error(sk) ? : -ECONNABORTED;
622 sock->state = SS_UNCONNECTED;
623 if (sk->sk_prot->disconnect(sk, flags))
624 sock->state = SS_DISCONNECTING;
629 * Accept a pending connection. The TCP layer now gives BSD semantics.
632 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
634 struct sock *sk1 = sock->sk;
636 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
643 BUG_TRAP((1 << sk2->sk_state) &
644 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
646 sock_graft(sk2, newsock);
648 newsock->state = SS_CONNECTED;
657 * This does both peername and sockname.
659 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
660 int *uaddr_len, int peer)
662 struct sock *sk = sock->sk;
663 struct inet_sock *inet = inet_sk(sk);
664 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
666 sin->sin_family = AF_INET;
669 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
672 sin->sin_port = inet->dport;
673 sin->sin_addr.s_addr = inet->daddr;
675 __u32 addr = inet->rcv_saddr;
678 sin->sin_port = inet->sport;
679 sin->sin_addr.s_addr = addr;
681 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
682 *uaddr_len = sizeof(*sin);
686 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
689 struct sock *sk = sock->sk;
691 /* We may need to bind the socket. */
692 if (!inet_sk(sk)->num && inet_autobind(sk))
695 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
699 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
701 struct sock *sk = sock->sk;
703 /* We may need to bind the socket. */
704 if (!inet_sk(sk)->num && inet_autobind(sk))
707 if (sk->sk_prot->sendpage)
708 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
709 return sock_no_sendpage(sock, page, offset, size, flags);
713 int inet_shutdown(struct socket *sock, int how)
715 struct sock *sk = sock->sk;
718 /* This should really check to make sure
719 * the socket is a TCP socket. (WHY AC...)
721 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
724 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
728 if (sock->state == SS_CONNECTING) {
729 if ((1 << sk->sk_state) &
730 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
731 sock->state = SS_DISCONNECTING;
733 sock->state = SS_CONNECTED;
736 switch (sk->sk_state) {
739 /* Hack to wake up other listeners, who can poll for
740 POLLHUP, even on eg. unconnected UDP sockets -- RR */
742 sk->sk_shutdown |= how;
743 if (sk->sk_prot->shutdown)
744 sk->sk_prot->shutdown(sk, how);
747 /* Remaining two branches are temporary solution for missing
748 * close() in multithreaded environment. It is _not_ a good idea,
749 * but we have no choice until close() is repaired at VFS level.
752 if (!(how & RCV_SHUTDOWN))
756 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
757 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
761 /* Wake up anyone sleeping in poll. */
762 sk->sk_state_change(sk);
768 * ioctl() calls you can issue on an INET socket. Most of these are
769 * device configuration and stuff and very rarely used. Some ioctls
770 * pass on to the socket itself.
772 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
773 * loads the devconfigure module does its configuring and unloads it.
774 * There's a good 20K of config code hanging around the kernel.
777 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
779 struct sock *sk = sock->sk;
784 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
789 err = ip_rt_ioctl(cmd, (void __user *)arg);
794 err = arp_ioctl(cmd, (void __user *)arg);
807 err = devinet_ioctl(cmd, (void __user *)arg);
810 if (!sk->sk_prot->ioctl ||
811 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
813 err = dev_ioctl(cmd, (void __user *)arg);
819 struct proto_ops inet_stream_ops = {
821 .owner = THIS_MODULE,
822 .release = inet_release,
824 .connect = inet_stream_connect,
825 .socketpair = sock_no_socketpair,
826 .accept = inet_accept,
827 .getname = inet_getname,
830 .listen = inet_listen,
831 .shutdown = inet_shutdown,
832 .setsockopt = sock_common_setsockopt,
833 .getsockopt = sock_common_getsockopt,
834 .sendmsg = inet_sendmsg,
835 .recvmsg = sock_common_recvmsg,
836 .mmap = sock_no_mmap,
837 .sendpage = tcp_sendpage
840 struct proto_ops inet_dgram_ops = {
842 .owner = THIS_MODULE,
843 .release = inet_release,
845 .connect = inet_dgram_connect,
846 .socketpair = sock_no_socketpair,
847 .accept = sock_no_accept,
848 .getname = inet_getname,
851 .listen = sock_no_listen,
852 .shutdown = inet_shutdown,
853 .setsockopt = sock_common_setsockopt,
854 .getsockopt = sock_common_getsockopt,
855 .sendmsg = inet_sendmsg,
856 .recvmsg = sock_common_recvmsg,
857 .mmap = sock_no_mmap,
858 .sendpage = inet_sendpage,
862 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
865 static struct proto_ops inet_sockraw_ops = {
867 .owner = THIS_MODULE,
868 .release = inet_release,
870 .connect = inet_dgram_connect,
871 .socketpair = sock_no_socketpair,
872 .accept = sock_no_accept,
873 .getname = inet_getname,
874 .poll = datagram_poll,
876 .listen = sock_no_listen,
877 .shutdown = inet_shutdown,
878 .setsockopt = sock_common_setsockopt,
879 .getsockopt = sock_common_getsockopt,
880 .sendmsg = inet_sendmsg,
881 .recvmsg = sock_common_recvmsg,
882 .mmap = sock_no_mmap,
883 .sendpage = inet_sendpage,
886 static struct net_proto_family inet_family_ops = {
888 .create = inet_create,
889 .owner = THIS_MODULE,
893 extern void tcp_init(void);
894 extern void tcp_v4_init(struct net_proto_family *);
896 /* Upon startup we insert all the elements in inetsw_array[] into
897 * the linked list inetsw.
899 static struct inet_protosw inetsw_array[] =
903 .protocol = IPPROTO_TCP,
905 .ops = &inet_stream_ops,
908 .flags = INET_PROTOSW_PERMANENT,
913 .protocol = IPPROTO_UDP,
915 .ops = &inet_dgram_ops,
917 .no_check = UDP_CSUM_DEFAULT,
918 .flags = INET_PROTOSW_PERMANENT,
924 .protocol = IPPROTO_IP, /* wild card */
926 .ops = &inet_sockraw_ops,
927 .capability = CAP_NET_RAW,
928 .no_check = UDP_CSUM_DEFAULT,
929 .flags = INET_PROTOSW_REUSE,
933 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
935 void inet_register_protosw(struct inet_protosw *p)
937 struct list_head *lh;
938 struct inet_protosw *answer;
939 int protocol = p->protocol;
940 struct list_head *last_perm;
942 spin_lock_bh(&inetsw_lock);
944 if (p->type >= SOCK_MAX)
947 /* If we are trying to override a permanent protocol, bail. */
949 last_perm = &inetsw[p->type];
950 list_for_each(lh, &inetsw[p->type]) {
951 answer = list_entry(lh, struct inet_protosw, list);
953 /* Check only the non-wild match. */
954 if (INET_PROTOSW_PERMANENT & answer->flags) {
955 if (protocol == answer->protocol)
965 /* Add the new entry after the last permanent entry if any, so that
966 * the new entry does not override a permanent entry when matched with
967 * a wild-card protocol. But it is allowed to override any existing
968 * non-permanent entry. This means that when we remove this entry, the
969 * system automatically returns to the old behavior.
971 list_add_rcu(&p->list, last_perm);
973 spin_unlock_bh(&inetsw_lock);
980 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
986 "Ignoring attempt to register invalid socket type %d.\n",
991 void inet_unregister_protosw(struct inet_protosw *p)
993 if (INET_PROTOSW_PERMANENT & p->flags) {
995 "Attempt to unregister permanent protocol %d.\n",
998 spin_lock_bh(&inetsw_lock);
999 list_del_rcu(&p->list);
1000 spin_unlock_bh(&inetsw_lock);
1006 #ifdef CONFIG_IP_MULTICAST
1007 static struct net_protocol igmp_protocol = {
1008 .handler = igmp_rcv,
1012 static struct net_protocol tcp_protocol = {
1013 .handler = tcp_v4_rcv,
1014 .err_handler = tcp_v4_err,
1018 static struct net_protocol udp_protocol = {
1020 .err_handler = udp_err,
1024 static struct net_protocol icmp_protocol = {
1025 .handler = icmp_rcv,
1028 static int __init init_ipv4_mibs(void)
1030 net_statistics[0] = alloc_percpu(struct linux_mib);
1031 net_statistics[1] = alloc_percpu(struct linux_mib);
1032 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1033 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1034 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1035 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1036 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1037 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1038 udp_statistics[0] = alloc_percpu(struct udp_mib);
1039 udp_statistics[1] = alloc_percpu(struct udp_mib);
1041 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1042 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1043 && udp_statistics[0] && udp_statistics[1]))
1046 (void) tcp_mib_init();
1051 static int ipv4_proc_init(void);
1052 extern void ipfrag_init(void);
1054 static int __init inet_init(void)
1056 struct sk_buff *dummy_skb;
1057 struct inet_protosw *q;
1058 struct list_head *r;
1061 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1062 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1066 rc = proto_register(&tcp_prot, 1);
1070 rc = proto_register(&udp_prot, 1);
1072 goto out_unregister_tcp_proto;
1074 rc = proto_register(&raw_prot, 1);
1076 goto out_unregister_udp_proto;
1079 * Tell SOCKET that we are alive...
1082 (void)sock_register(&inet_family_ops);
1085 * Add all the base protocols.
1088 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1089 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1090 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1091 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1092 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1093 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1094 #ifdef CONFIG_IP_MULTICAST
1095 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1096 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1099 /* Register the socket-side information for inet_create. */
1100 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1103 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1104 inet_register_protosw(q);
1107 * Set the ARP module up
1113 * Set the IP module up
1118 tcp_v4_init(&inet_family_ops);
1120 /* Setup TCP slab cache for open requests. */
1125 * Set the ICMP layer up
1128 icmp_init(&inet_family_ops);
1131 * Initialise the multicast router
1133 #if defined(CONFIG_IP_MROUTE)
1137 * Initialise per-cpu ipv4 mibs
1140 if(init_ipv4_mibs())
1141 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1150 out_unregister_tcp_proto:
1151 proto_unregister(&tcp_prot);
1152 out_unregister_udp_proto:
1153 proto_unregister(&udp_prot);
1157 module_init(inet_init);
1159 /* ------------------------------------------------------------------------ */
1161 #ifdef CONFIG_PROC_FS
1162 extern int fib_proc_init(void);
1163 extern void fib_proc_exit(void);
1164 extern int ip_misc_proc_init(void);
1165 extern int raw_proc_init(void);
1166 extern void raw_proc_exit(void);
1167 extern int tcp4_proc_init(void);
1168 extern void tcp4_proc_exit(void);
1169 extern int udp4_proc_init(void);
1170 extern void udp4_proc_exit(void);
1172 static int __init ipv4_proc_init(void)
1176 if (raw_proc_init())
1178 if (tcp4_proc_init())
1180 if (udp4_proc_init())
1182 if (fib_proc_init())
1184 if (ip_misc_proc_init())
1201 #else /* CONFIG_PROC_FS */
1202 static int __init ipv4_proc_init(void)
1206 #endif /* CONFIG_PROC_FS */
1208 MODULE_ALIAS_NETPROTO(PF_INET);
1210 EXPORT_SYMBOL(inet_accept);
1211 EXPORT_SYMBOL(inet_bind);
1212 EXPORT_SYMBOL(inet_dgram_connect);
1213 EXPORT_SYMBOL(inet_dgram_ops);
1214 EXPORT_SYMBOL(inet_getname);
1215 EXPORT_SYMBOL(inet_ioctl);
1216 EXPORT_SYMBOL(inet_listen);
1217 EXPORT_SYMBOL(inet_register_protosw);
1218 EXPORT_SYMBOL(inet_release);
1219 EXPORT_SYMBOL(inet_sendmsg);
1220 EXPORT_SYMBOL(inet_shutdown);
1221 EXPORT_SYMBOL(inet_sock_destruct);
1222 EXPORT_SYMBOL(inet_stream_connect);
1223 EXPORT_SYMBOL(inet_stream_ops);
1224 EXPORT_SYMBOL(inet_unregister_protosw);
1225 EXPORT_SYMBOL(net_statistics);
1226 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
1228 #ifdef INET_REFCNT_DEBUG
1229 EXPORT_SYMBOL(inet_sock_nr);