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 DEFINE_SPINLOCK(inetsw_lock);
131 /* New destruction routine */
133 void inet_sock_destruct(struct sock *sk)
135 struct inet_sock *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);
157 dst_release(sk->sk_dst_cache);
158 #ifdef INET_REFCNT_DEBUG
159 atomic_dec(&inet_sock_nr);
160 printk(KERN_DEBUG "INET socket %p released, %d are still alive\n",
161 sk, atomic_read(&inet_sock_nr));
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
172 * Automatically bind an unbound socket.
175 static int inet_autobind(struct sock *sk)
177 struct inet_sock *inet;
178 /* We may need to bind the socket. */
182 if (sk->sk_prot->get_port(sk, 0)) {
186 inet->sport = htons(inet->num);
193 * Move a socket into listening state.
195 int inet_listen(struct socket *sock, int backlog)
197 struct sock *sk = sock->sk;
198 unsigned char old_state;
204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
207 old_state = sk->sk_state;
208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
211 /* Really, if the socket is already in listen state
212 * we can only allow the backlog to be adjusted.
214 if (old_state != TCP_LISTEN) {
215 err = tcp_listen_start(sk);
219 sk->sk_max_ack_backlog = backlog;
228 * Create an inet socket.
231 static int inet_create(struct socket *sock, int protocol)
235 struct inet_protosw *answer;
236 struct inet_sock *inet;
237 struct proto *answer_prot;
238 unsigned char answer_flags;
239 char answer_no_check;
242 sock->state = SS_UNCONNECTED;
244 /* Look for the requested type/protocol pair. */
247 list_for_each_rcu(p, &inetsw[sock->type]) {
248 answer = list_entry(p, struct inet_protosw, list);
250 /* Check the non-wild match. */
251 if (protocol == answer->protocol) {
252 if (protocol != IPPROTO_IP)
255 /* Check for the two wild cases. */
256 if (IPPROTO_IP == protocol) {
257 protocol = answer->protocol;
260 if (IPPROTO_IP == answer->protocol)
266 err = -ESOCKTNOSUPPORT;
270 if ((protocol == IPPROTO_ICMP) && vx_ccaps(VXC_RAW_ICMP))
272 if (answer->capability > 0 && !capable(answer->capability))
275 err = -EPROTONOSUPPORT;
279 sock->ops = answer->ops;
280 answer_prot = answer->prot;
281 answer_no_check = answer->no_check;
282 answer_flags = answer->flags;
285 BUG_TRAP(answer_prot->slab != NULL);
288 sk = sk_alloc(PF_INET, GFP_KERNEL,
289 answer_prot->slab_obj_size,
295 sk->sk_prot = answer_prot;
296 sk->sk_no_check = answer_no_check;
297 if (INET_PROTOSW_REUSE & answer_flags)
302 if (SOCK_RAW == sock->type) {
303 inet->num = protocol;
304 if (IPPROTO_RAW == protocol)
308 if (ipv4_config.no_pmtu_disc)
309 inet->pmtudisc = IP_PMTUDISC_DONT;
311 inet->pmtudisc = IP_PMTUDISC_WANT;
315 sock_init_data(sock, sk);
316 sk_set_owner(sk, sk->sk_prot->owner);
318 sk->sk_destruct = inet_sock_destruct;
319 sk->sk_family = PF_INET;
320 sk->sk_protocol = protocol;
321 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
327 inet->mc_list = NULL;
329 #ifdef INET_REFCNT_DEBUG
330 atomic_inc(&inet_sock_nr);
334 /* It assumes that any protocol which allows
335 * the user to assign a number at socket
336 * creation time automatically
339 inet->sport = htons(inet->num);
340 /* Add to protocol hash chains. */
341 sk->sk_prot->hash(sk);
344 if (sk->sk_prot->init) {
345 err = sk->sk_prot->init(sk);
347 sk_common_release(sk);
358 * The peer socket should always be NULL (or else). When we call this
359 * function we are destroying the object and from then on nobody
360 * should refer to it.
362 int inet_release(struct socket *sock)
364 struct sock *sk = sock->sk;
369 /* Applications forget to leave groups before exiting */
370 ip_mc_drop_socket(sk);
372 /* If linger is set, we don't return until the close
373 * is complete. Otherwise we return immediately. The
374 * actually closing is done the same either way.
376 * If the close is due to the process exiting, we never
380 if (sock_flag(sk, SOCK_LINGER) &&
381 !(current->flags & PF_EXITING))
382 timeout = sk->sk_lingertime;
384 sk->sk_prot->close(sk, timeout);
389 /* It is off by default, see below. */
390 int sysctl_ip_nonlocal_bind;
392 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
394 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
395 struct sock *sk = sock->sk;
396 struct inet_sock *inet = inet_sk(sk);
400 __u32 s_addr; /* Address used for validation */
401 __u32 s_addr1; /* Address used for socket */
402 __u32 s_addr2; /* Broadcast address for the socket */
403 struct nx_info *nxi = sk->sk_nx_info;
405 /* If the socket has its own bind function then use it. (RAW) */
406 if (sk->sk_prot->bind) {
407 err = sk->sk_prot->bind(sk, uaddr, addr_len);
411 if (addr_len < sizeof(struct sockaddr_in))
414 s_addr = addr->sin_addr.s_addr;
416 s_addr2 = 0xffffffffl;
418 vxdprintk(VXD_CBIT(net, 3),
419 "inet_bind(%p)* %p,%p;%lx %d.%d.%d.%d",
420 sk, sk->sk_nx_info, sk->sk_socket,
421 (sk->sk_socket?sk->sk_socket->flags:0),
424 __u32 v4_bcast = nxi->v4_bcast;
425 __u32 ipv4root = nxi->ipv4[0];
426 int nbipv4 = nxi->nbipv4;
429 /* bind to any for 1-n */
431 s_addr1 = (nbipv4 > 1) ? 0 : s_addr;
433 } else if (s_addr == 0x0100007f) {
434 /* rewrite localhost to ipv4root */
437 } else if (s_addr != v4_bcast) {
438 /* normal address bind */
439 if (!addr_in_nx_info(nxi, s_addr))
440 return -EADDRNOTAVAIL;
443 chk_addr_ret = inet_addr_type(s_addr);
445 vxdprintk(VXD_CBIT(net, 3),
446 "inet_bind(%p) %d.%d.%d.%d, %d.%d.%d.%d, %d.%d.%d.%d",
447 sk, VXD_QUAD(s_addr), VXD_QUAD(s_addr1), VXD_QUAD(s_addr2));
449 /* Not specified by any standard per-se, however it breaks too
450 * many applications when removed. It is unfortunate since
451 * allowing applications to make a non-local bind solves
452 * several problems with systems using dynamic addressing.
453 * (ie. your servers still start up even if your ISDN link
454 * is temporarily down)
456 err = -EADDRNOTAVAIL;
457 if (!sysctl_ip_nonlocal_bind &&
459 s_addr != INADDR_ANY &&
460 chk_addr_ret != RTN_LOCAL &&
461 chk_addr_ret != RTN_MULTICAST &&
462 chk_addr_ret != RTN_BROADCAST)
465 snum = ntohs(addr->sin_port);
467 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
470 /* We keep a pair of addresses. rcv_saddr is the one
471 * used by hash lookups, and saddr is used for transmit.
473 * In the BSD API these are the same except where it
474 * would be illegal to use them (multicast/broadcast) in
475 * which case the sending device address is used.
479 /* Check these errors (active socket, double bind). */
481 if (sk->sk_state != TCP_CLOSE || inet->num)
482 goto out_release_sock;
484 inet->rcv_saddr = inet->saddr = s_addr1;
485 inet->rcv_saddr2 = s_addr2;
486 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
487 inet->saddr = 0; /* Use device */
489 /* Make sure we are allowed to bind here. */
490 if (sk->sk_prot->get_port(sk, snum)) {
491 inet->saddr = inet->rcv_saddr = 0;
493 goto out_release_sock;
497 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
499 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
500 inet->sport = htons(inet->num);
511 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
512 int addr_len, int flags)
514 struct sock *sk = sock->sk;
516 if (uaddr->sa_family == AF_UNSPEC)
517 return sk->sk_prot->disconnect(sk, flags);
519 if (!inet_sk(sk)->num && inet_autobind(sk))
521 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
524 static long inet_wait_for_connect(struct sock *sk, long timeo)
528 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
530 /* Basic assumption: if someone sets sk->sk_err, he _must_
531 * change state of the socket from TCP_SYN_*.
532 * Connect() does not allow to get error notifications
533 * without closing the socket.
535 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
537 timeo = schedule_timeout(timeo);
539 if (signal_pending(current) || !timeo)
541 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
543 finish_wait(sk->sk_sleep, &wait);
548 * Connect to a remote host. There is regrettably still a little
549 * TCP 'magic' in here.
551 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
552 int addr_len, int flags)
554 struct sock *sk = sock->sk;
560 if (uaddr->sa_family == AF_UNSPEC) {
561 err = sk->sk_prot->disconnect(sk, flags);
562 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
566 switch (sock->state) {
575 /* Fall out of switch with err, set for this state */
579 if (sk->sk_state != TCP_CLOSE)
582 err = sk->sk_prot->connect(sk, uaddr, addr_len);
586 sock->state = SS_CONNECTING;
588 /* Just entered SS_CONNECTING state; the only
589 * difference is that return value in non-blocking
590 * case is EINPROGRESS, rather than EALREADY.
596 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
598 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
599 /* Error code is set above */
600 if (!timeo || !inet_wait_for_connect(sk, timeo))
603 err = sock_intr_errno(timeo);
604 if (signal_pending(current))
608 /* Connection was closed by RST, timeout, ICMP error
609 * or another process disconnected us.
611 if (sk->sk_state == TCP_CLOSE)
614 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
615 * and error was received after socket entered established state.
616 * Hence, it is handled normally after connect() return successfully.
619 sock->state = SS_CONNECTED;
626 err = sock_error(sk) ? : -ECONNABORTED;
627 sock->state = SS_UNCONNECTED;
628 if (sk->sk_prot->disconnect(sk, flags))
629 sock->state = SS_DISCONNECTING;
634 * Accept a pending connection. The TCP layer now gives BSD semantics.
637 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
639 struct sock *sk1 = sock->sk;
641 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
648 BUG_TRAP((1 << sk2->sk_state) &
649 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
651 sock_graft(sk2, newsock);
653 newsock->state = SS_CONNECTED;
662 * This does both peername and sockname.
664 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
665 int *uaddr_len, int peer)
667 struct sock *sk = sock->sk;
668 struct inet_sock *inet = inet_sk(sk);
669 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
671 sin->sin_family = AF_INET;
674 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
677 sin->sin_port = inet->dport;
678 sin->sin_addr.s_addr = inet->daddr;
680 __u32 addr = inet->rcv_saddr;
683 sin->sin_port = inet->sport;
684 sin->sin_addr.s_addr = addr;
686 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
687 *uaddr_len = sizeof(*sin);
691 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
694 struct sock *sk = sock->sk;
696 /* We may need to bind the socket. */
697 if (!inet_sk(sk)->num && inet_autobind(sk))
700 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
704 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
706 struct sock *sk = sock->sk;
708 /* We may need to bind the socket. */
709 if (!inet_sk(sk)->num && inet_autobind(sk))
712 if (sk->sk_prot->sendpage)
713 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
714 return sock_no_sendpage(sock, page, offset, size, flags);
718 int inet_shutdown(struct socket *sock, int how)
720 struct sock *sk = sock->sk;
723 /* This should really check to make sure
724 * the socket is a TCP socket. (WHY AC...)
726 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
729 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
733 if (sock->state == SS_CONNECTING) {
734 if ((1 << sk->sk_state) &
735 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
736 sock->state = SS_DISCONNECTING;
738 sock->state = SS_CONNECTED;
741 switch (sk->sk_state) {
744 /* Hack to wake up other listeners, who can poll for
745 POLLHUP, even on eg. unconnected UDP sockets -- RR */
747 sk->sk_shutdown |= how;
748 if (sk->sk_prot->shutdown)
749 sk->sk_prot->shutdown(sk, how);
752 /* Remaining two branches are temporary solution for missing
753 * close() in multithreaded environment. It is _not_ a good idea,
754 * but we have no choice until close() is repaired at VFS level.
757 if (!(how & RCV_SHUTDOWN))
761 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
762 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
766 /* Wake up anyone sleeping in poll. */
767 sk->sk_state_change(sk);
773 * ioctl() calls you can issue on an INET socket. Most of these are
774 * device configuration and stuff and very rarely used. Some ioctls
775 * pass on to the socket itself.
777 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
778 * loads the devconfigure module does its configuring and unloads it.
779 * There's a good 20K of config code hanging around the kernel.
782 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
784 struct sock *sk = sock->sk;
789 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
794 err = ip_rt_ioctl(cmd, (void __user *)arg);
799 err = arp_ioctl(cmd, (void __user *)arg);
812 err = devinet_ioctl(cmd, (void __user *)arg);
815 if (!sk->sk_prot->ioctl ||
816 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
818 err = dev_ioctl(cmd, (void __user *)arg);
824 struct proto_ops inet_stream_ops = {
826 .owner = THIS_MODULE,
827 .release = inet_release,
829 .connect = inet_stream_connect,
830 .socketpair = sock_no_socketpair,
831 .accept = inet_accept,
832 .getname = inet_getname,
835 .listen = inet_listen,
836 .shutdown = inet_shutdown,
837 .setsockopt = sock_common_setsockopt,
838 .getsockopt = sock_common_getsockopt,
839 .sendmsg = inet_sendmsg,
840 .recvmsg = sock_common_recvmsg,
841 .mmap = sock_no_mmap,
842 .sendpage = tcp_sendpage
845 struct proto_ops inet_dgram_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,
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,
867 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
870 static struct proto_ops inet_sockraw_ops = {
872 .owner = THIS_MODULE,
873 .release = inet_release,
875 .connect = inet_dgram_connect,
876 .socketpair = sock_no_socketpair,
877 .accept = sock_no_accept,
878 .getname = inet_getname,
879 .poll = datagram_poll,
881 .listen = sock_no_listen,
882 .shutdown = inet_shutdown,
883 .setsockopt = sock_common_setsockopt,
884 .getsockopt = sock_common_getsockopt,
885 .sendmsg = inet_sendmsg,
886 .recvmsg = sock_common_recvmsg,
887 .mmap = sock_no_mmap,
888 .sendpage = inet_sendpage,
891 static struct net_proto_family inet_family_ops = {
893 .create = inet_create,
894 .owner = THIS_MODULE,
898 extern void tcp_init(void);
899 extern void tcp_v4_init(struct net_proto_family *);
901 /* Upon startup we insert all the elements in inetsw_array[] into
902 * the linked list inetsw.
904 static struct inet_protosw inetsw_array[] =
908 .protocol = IPPROTO_TCP,
910 .ops = &inet_stream_ops,
913 .flags = INET_PROTOSW_PERMANENT,
918 .protocol = IPPROTO_UDP,
920 .ops = &inet_dgram_ops,
922 .no_check = UDP_CSUM_DEFAULT,
923 .flags = INET_PROTOSW_PERMANENT,
929 .protocol = IPPROTO_IP, /* wild card */
931 .ops = &inet_sockraw_ops,
932 .capability = CAP_NET_RAW,
933 .no_check = UDP_CSUM_DEFAULT,
934 .flags = INET_PROTOSW_REUSE,
938 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
940 void inet_register_protosw(struct inet_protosw *p)
942 struct list_head *lh;
943 struct inet_protosw *answer;
944 int protocol = p->protocol;
945 struct list_head *last_perm;
947 spin_lock_bh(&inetsw_lock);
949 if (p->type >= SOCK_MAX)
952 /* If we are trying to override a permanent protocol, bail. */
954 last_perm = &inetsw[p->type];
955 list_for_each(lh, &inetsw[p->type]) {
956 answer = list_entry(lh, struct inet_protosw, list);
958 /* Check only the non-wild match. */
959 if (INET_PROTOSW_PERMANENT & answer->flags) {
960 if (protocol == answer->protocol)
970 /* Add the new entry after the last permanent entry if any, so that
971 * the new entry does not override a permanent entry when matched with
972 * a wild-card protocol. But it is allowed to override any existing
973 * non-permanent entry. This means that when we remove this entry, the
974 * system automatically returns to the old behavior.
976 list_add_rcu(&p->list, last_perm);
978 spin_unlock_bh(&inetsw_lock);
985 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
991 "Ignoring attempt to register invalid socket type %d.\n",
996 void inet_unregister_protosw(struct inet_protosw *p)
998 if (INET_PROTOSW_PERMANENT & p->flags) {
1000 "Attempt to unregister permanent protocol %d.\n",
1003 spin_lock_bh(&inetsw_lock);
1004 list_del_rcu(&p->list);
1005 spin_unlock_bh(&inetsw_lock);
1011 #ifdef CONFIG_IP_MULTICAST
1012 static struct net_protocol igmp_protocol = {
1013 .handler = igmp_rcv,
1017 static struct net_protocol tcp_protocol = {
1018 .handler = tcp_v4_rcv,
1019 .err_handler = tcp_v4_err,
1023 static struct net_protocol udp_protocol = {
1025 .err_handler = udp_err,
1029 static struct net_protocol icmp_protocol = {
1030 .handler = icmp_rcv,
1033 static int __init init_ipv4_mibs(void)
1035 net_statistics[0] = alloc_percpu(struct linux_mib);
1036 net_statistics[1] = alloc_percpu(struct linux_mib);
1037 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1038 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1039 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1040 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1041 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1042 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1043 udp_statistics[0] = alloc_percpu(struct udp_mib);
1044 udp_statistics[1] = alloc_percpu(struct udp_mib);
1046 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1047 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1048 && udp_statistics[0] && udp_statistics[1]))
1051 (void) tcp_mib_init();
1056 static int ipv4_proc_init(void);
1057 extern void ipfrag_init(void);
1059 static int __init inet_init(void)
1061 struct sk_buff *dummy_skb;
1062 struct inet_protosw *q;
1063 struct list_head *r;
1066 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1067 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1071 rc = sk_alloc_slab(&tcp_prot, "tcp_sock");
1073 sk_alloc_slab_error(&tcp_prot);
1076 rc = sk_alloc_slab(&udp_prot, "udp_sock");
1078 sk_alloc_slab_error(&udp_prot);
1079 goto out_tcp_free_slab;
1081 rc = sk_alloc_slab(&raw_prot, "raw_sock");
1083 sk_alloc_slab_error(&raw_prot);
1084 goto out_udp_free_slab;
1088 * Tell SOCKET that we are alive...
1091 (void)sock_register(&inet_family_ops);
1094 * Add all the base protocols.
1097 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1098 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1099 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1100 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1101 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1102 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1103 #ifdef CONFIG_IP_MULTICAST
1104 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1105 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1108 /* Register the socket-side information for inet_create. */
1109 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1112 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1113 inet_register_protosw(q);
1116 * Set the ARP module up
1122 * Set the IP module up
1127 tcp_v4_init(&inet_family_ops);
1129 /* Setup TCP slab cache for open requests. */
1134 * Set the ICMP layer up
1137 icmp_init(&inet_family_ops);
1140 * Initialise the multicast router
1142 #if defined(CONFIG_IP_MROUTE)
1146 * Initialise per-cpu ipv4 mibs
1149 if(init_ipv4_mibs())
1150 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1160 sk_free_slab(&tcp_prot);
1162 sk_free_slab(&udp_prot);
1166 module_init(inet_init);
1168 /* ------------------------------------------------------------------------ */
1170 #ifdef CONFIG_PROC_FS
1171 extern int fib_proc_init(void);
1172 extern void fib_proc_exit(void);
1173 extern int ip_misc_proc_init(void);
1174 extern int raw_proc_init(void);
1175 extern void raw_proc_exit(void);
1176 extern int tcp4_proc_init(void);
1177 extern void tcp4_proc_exit(void);
1178 extern int udp4_proc_init(void);
1179 extern void udp4_proc_exit(void);
1181 static int __init ipv4_proc_init(void)
1185 if (raw_proc_init())
1187 if (tcp4_proc_init())
1189 if (udp4_proc_init())
1191 if (fib_proc_init())
1193 if (ip_misc_proc_init())
1210 #else /* CONFIG_PROC_FS */
1211 static int __init ipv4_proc_init(void)
1215 #endif /* CONFIG_PROC_FS */
1217 MODULE_ALIAS_NETPROTO(PF_INET);
1219 EXPORT_SYMBOL(inet_accept);
1220 EXPORT_SYMBOL(inet_bind);
1221 EXPORT_SYMBOL(inet_dgram_connect);
1222 EXPORT_SYMBOL(inet_dgram_ops);
1223 EXPORT_SYMBOL(inet_getname);
1224 EXPORT_SYMBOL(inet_ioctl);
1225 EXPORT_SYMBOL(inet_listen);
1226 EXPORT_SYMBOL(inet_register_protosw);
1227 EXPORT_SYMBOL(inet_release);
1228 EXPORT_SYMBOL(inet_sendmsg);
1229 EXPORT_SYMBOL(inet_shutdown);
1230 EXPORT_SYMBOL(inet_sock_destruct);
1231 EXPORT_SYMBOL(inet_stream_connect);
1232 EXPORT_SYMBOL(inet_stream_ops);
1233 EXPORT_SYMBOL(inet_unregister_protosw);
1234 EXPORT_SYMBOL(net_statistics);
1236 #ifdef INET_REFCNT_DEBUG
1237 EXPORT_SYMBOL(inet_sock_nr);