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>
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 /* Per protocol sock slabcache */
125 kmem_cache_t *tcp_sk_cachep;
126 static kmem_cache_t *udp_sk_cachep;
127 static kmem_cache_t *raw4_sk_cachep;
129 /* The inetsw table contains everything that inet_create needs to
130 * build a new socket.
132 static struct list_head inetsw[SOCK_MAX];
133 static spinlock_t inetsw_lock = SPIN_LOCK_UNLOCKED;
135 /* New destruction routine */
137 void inet_sock_destruct(struct sock *sk)
139 struct inet_opt *inet = inet_sk(sk);
141 __skb_queue_purge(&sk->sk_receive_queue);
142 __skb_queue_purge(&sk->sk_error_queue);
144 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 printk("Attempt to release TCP socket in state %d %p\n",
149 if (!sock_flag(sk, SOCK_DEAD)) {
150 printk("Attempt to release alive inet socket %p\n", sk);
154 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
155 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
156 BUG_TRAP(!sk->sk_wmem_queued);
157 BUG_TRAP(!sk->sk_forward_alloc);
162 BUG_ON(sk->sk_nx_info);
163 BUG_ON(sk->sk_vx_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;
234 static __inline__ kmem_cache_t *inet_sk_slab(int protocol)
236 kmem_cache_t* rc = tcp_sk_cachep;
238 if (protocol == IPPROTO_UDP)
240 else if (protocol == IPPROTO_RAW)
245 static __inline__ int inet_sk_size(int protocol)
247 int rc = sizeof(struct tcp_sock);
249 if (protocol == IPPROTO_UDP)
250 rc = sizeof(struct udp_sock);
251 else if (protocol == IPPROTO_RAW)
252 rc = sizeof(struct raw_sock);
257 * Create an inet socket.
260 static int inet_create(struct socket *sock, int protocol)
264 struct inet_protosw *answer;
265 struct inet_opt *inet;
268 sock->state = SS_UNCONNECTED;
269 sk = sk_alloc(PF_INET, GFP_KERNEL, inet_sk_size(protocol),
270 inet_sk_slab(protocol));
274 /* Look for the requested type/protocol pair. */
277 list_for_each_rcu(p, &inetsw[sock->type]) {
278 answer = list_entry(p, struct inet_protosw, list);
280 /* Check the non-wild match. */
281 if (protocol == answer->protocol) {
282 if (protocol != IPPROTO_IP)
285 /* Check for the two wild cases. */
286 if (IPPROTO_IP == protocol) {
287 protocol = answer->protocol;
290 if (IPPROTO_IP == answer->protocol)
296 err = -ESOCKTNOSUPPORT;
300 if (answer->capability > 0 && !capable(answer->capability))
302 err = -EPROTONOSUPPORT;
306 sock->ops = answer->ops;
307 sk->sk_prot = answer->prot;
308 sk->sk_no_check = answer->no_check;
309 if (INET_PROTOSW_REUSE & answer->flags)
315 if (SOCK_RAW == sock->type) {
316 inet->num = protocol;
317 if (IPPROTO_RAW == protocol)
321 if (ipv4_config.no_pmtu_disc)
322 inet->pmtudisc = IP_PMTUDISC_DONT;
324 inet->pmtudisc = IP_PMTUDISC_WANT;
328 sock_init_data(sock, sk);
329 sk_set_owner(sk, THIS_MODULE);
331 sk->sk_destruct = inet_sock_destruct;
333 sk->sk_family = PF_INET;
334 sk->sk_protocol = protocol;
335 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
337 set_vx_info(&sk->sk_vx_info, current->vx_info);
338 sk->sk_xid = vx_current_xid();
339 set_nx_info(&sk->sk_nx_info, current->nx_info);
340 sk->sk_nid = nx_current_nid();
346 inet->mc_list = NULL;
348 #ifdef INET_REFCNT_DEBUG
349 atomic_inc(&inet_sock_nr);
353 /* It assumes that any protocol which allows
354 * the user to assign a number at socket
355 * creation time automatically
358 inet->sport = htons(inet->num);
359 /* Add to protocol hash chains. */
360 sk->sk_prot->hash(sk);
363 if (sk->sk_prot->init) {
364 err = sk->sk_prot->init(sk);
366 sk_common_release(sk);
378 * The peer socket should always be NULL (or else). When we call this
379 * function we are destroying the object and from then on nobody
380 * should refer to it.
382 int inet_release(struct socket *sock)
384 struct sock *sk = sock->sk;
389 /* Applications forget to leave groups before exiting */
390 ip_mc_drop_socket(sk);
392 /* If linger is set, we don't return until the close
393 * is complete. Otherwise we return immediately. The
394 * actually closing is done the same either way.
396 * If the close is due to the process exiting, we never
400 if (sock_flag(sk, SOCK_LINGER) &&
401 !(current->flags & PF_EXITING))
402 timeout = sk->sk_lingertime;
404 clr_vx_info(&sk->sk_vx_info);
405 clr_nx_info(&sk->sk_nx_info);
406 sk->sk_prot->close(sk, timeout);
411 /* It is off by default, see below. */
412 int sysctl_ip_nonlocal_bind;
414 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
416 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
417 struct sock *sk = sock->sk;
418 struct inet_opt *inet = inet_sk(sk);
422 __u32 s_addr; /* Address used for validation */
424 __u32 s_addr2 = 0xffffffffl; /* Optional address of the socket */
425 struct nx_info *nxi = sk->sk_nx_info;
427 /* If the socket has its own bind function then use it. (RAW) */
428 if (sk->sk_prot->bind) {
429 err = sk->sk_prot->bind(sk, uaddr, addr_len);
433 if (addr_len < sizeof(struct sockaddr_in))
436 s_addr = s_addr1 = addr->sin_addr.s_addr;
437 nxdprintk("inet_bind(%p) %p,%p;%lx\n",
438 sk, nx_info, sk->sk_socket,
439 (sk->sk_socket?sk->sk_socket->flags:0));
441 __u32 v4_bcast = nxi->v4_bcast;
442 __u32 ipv4root = nxi->ipv4[0];
443 int nbipv4 = nxi->nbipv4;
452 } else if (s_addr == 0x0100007f) {
453 s_addr = s_addr1 = ipv4root;
454 } else if (s_addr != v4_bcast) {
456 for (i=0; i<nbipv4; i++) {
457 if (s_addr == nxi->ipv4[i])
461 return -EADDRNOTAVAIL;
465 chk_addr_ret = inet_addr_type(s_addr);
467 /* Not specified by any standard per-se, however it breaks too
468 * many applications when removed. It is unfortunate since
469 * allowing applications to make a non-local bind solves
470 * several problems with systems using dynamic addressing.
471 * (ie. your servers still start up even if your ISDN link
472 * is temporarily down)
474 err = -EADDRNOTAVAIL;
475 if (!sysctl_ip_nonlocal_bind &&
477 s_addr != INADDR_ANY &&
478 chk_addr_ret != RTN_LOCAL &&
479 chk_addr_ret != RTN_MULTICAST &&
480 chk_addr_ret != RTN_BROADCAST)
483 snum = ntohs(addr->sin_port);
485 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
488 /* We keep a pair of addresses. rcv_saddr is the one
489 * used by hash lookups, and saddr is used for transmit.
491 * In the BSD API these are the same except where it
492 * would be illegal to use them (multicast/broadcast) in
493 * which case the sending device address is used.
497 /* Check these errors (active socket, double bind). */
499 if (sk->sk_state != TCP_CLOSE || inet->num)
500 goto out_release_sock;
502 inet->rcv_saddr = inet->saddr = s_addr1;
503 inet->rcv_saddr2 = s_addr2;
504 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
505 inet->saddr = 0; /* Use device */
507 /* Make sure we are allowed to bind here. */
508 if (sk->sk_prot->get_port(sk, snum)) {
509 inet->saddr = inet->rcv_saddr = 0;
511 goto out_release_sock;
515 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
517 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
518 inet->sport = htons(inet->num);
529 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
530 int addr_len, int flags)
532 struct sock *sk = sock->sk;
534 if (uaddr->sa_family == AF_UNSPEC)
535 return sk->sk_prot->disconnect(sk, flags);
537 if (!inet_sk(sk)->num && inet_autobind(sk))
539 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
542 static long inet_wait_for_connect(struct sock *sk, long timeo)
546 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
548 /* Basic assumption: if someone sets sk->sk_err, he _must_
549 * change state of the socket from TCP_SYN_*.
550 * Connect() does not allow to get error notifications
551 * without closing the socket.
553 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
555 timeo = schedule_timeout(timeo);
557 if (signal_pending(current) || !timeo)
559 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
561 finish_wait(sk->sk_sleep, &wait);
566 * Connect to a remote host. There is regrettably still a little
567 * TCP 'magic' in here.
569 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
570 int addr_len, int flags)
572 struct sock *sk = sock->sk;
578 if (uaddr->sa_family == AF_UNSPEC) {
579 err = sk->sk_prot->disconnect(sk, flags);
580 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
584 switch (sock->state) {
593 /* Fall out of switch with err, set for this state */
597 if (sk->sk_state != TCP_CLOSE)
600 err = sk->sk_prot->connect(sk, uaddr, addr_len);
604 sock->state = SS_CONNECTING;
606 /* Just entered SS_CONNECTING state; the only
607 * difference is that return value in non-blocking
608 * case is EINPROGRESS, rather than EALREADY.
614 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
616 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
617 /* Error code is set above */
618 if (!timeo || !inet_wait_for_connect(sk, timeo))
621 err = sock_intr_errno(timeo);
622 if (signal_pending(current))
626 /* Connection was closed by RST, timeout, ICMP error
627 * or another process disconnected us.
629 if (sk->sk_state == TCP_CLOSE)
632 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
633 * and error was received after socket entered established state.
634 * Hence, it is handled normally after connect() return successfully.
637 sock->state = SS_CONNECTED;
644 err = sock_error(sk) ? : -ECONNABORTED;
645 sock->state = SS_UNCONNECTED;
646 if (sk->sk_prot->disconnect(sk, flags))
647 sock->state = SS_DISCONNECTING;
652 * Accept a pending connection. The TCP layer now gives BSD semantics.
655 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
657 struct sock *sk1 = sock->sk;
659 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
666 BUG_TRAP((1 << sk2->sk_state) &
667 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
669 sock_graft(sk2, newsock);
671 newsock->state = SS_CONNECTED;
680 * This does both peername and sockname.
682 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
683 int *uaddr_len, int peer)
685 struct sock *sk = sock->sk;
686 struct inet_opt *inet = inet_sk(sk);
687 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
689 sin->sin_family = AF_INET;
692 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
695 sin->sin_port = inet->dport;
696 sin->sin_addr.s_addr = inet->daddr;
698 __u32 addr = inet->rcv_saddr;
701 sin->sin_port = inet->sport;
702 sin->sin_addr.s_addr = addr;
704 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
705 *uaddr_len = sizeof(*sin);
709 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
712 struct sock *sk = sock->sk;
714 /* We may need to bind the socket. */
715 if (!inet_sk(sk)->num && inet_autobind(sk))
718 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
722 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
724 struct sock *sk = sock->sk;
726 /* We may need to bind the socket. */
727 if (!inet_sk(sk)->num && inet_autobind(sk))
730 if (sk->sk_prot->sendpage)
731 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
732 return sock_no_sendpage(sock, page, offset, size, flags);
736 int inet_shutdown(struct socket *sock, int how)
738 struct sock *sk = sock->sk;
741 /* This should really check to make sure
742 * the socket is a TCP socket. (WHY AC...)
744 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
747 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
751 if (sock->state == SS_CONNECTING) {
752 if ((1 << sk->sk_state) &
753 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
754 sock->state = SS_DISCONNECTING;
756 sock->state = SS_CONNECTED;
759 switch (sk->sk_state) {
762 /* Hack to wake up other listeners, who can poll for
763 POLLHUP, even on eg. unconnected UDP sockets -- RR */
765 sk->sk_shutdown |= how;
766 if (sk->sk_prot->shutdown)
767 sk->sk_prot->shutdown(sk, how);
770 /* Remaining two branches are temporary solution for missing
771 * close() in multithreaded environment. It is _not_ a good idea,
772 * but we have no choice until close() is repaired at VFS level.
775 if (!(how & RCV_SHUTDOWN))
779 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
780 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
784 /* Wake up anyone sleeping in poll. */
785 sk->sk_state_change(sk);
791 * ioctl() calls you can issue on an INET socket. Most of these are
792 * device configuration and stuff and very rarely used. Some ioctls
793 * pass on to the socket itself.
795 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
796 * loads the devconfigure module does its configuring and unloads it.
797 * There's a good 20K of config code hanging around the kernel.
800 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
802 struct sock *sk = sock->sk;
807 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
812 err = ip_rt_ioctl(cmd, (void __user *)arg);
817 err = arp_ioctl(cmd, (void __user *)arg);
830 err = devinet_ioctl(cmd, (void __user *)arg);
833 if (!sk->sk_prot->ioctl ||
834 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
836 err = dev_ioctl(cmd, (void __user *)arg);
842 struct proto_ops inet_stream_ops = {
844 .owner = THIS_MODULE,
845 .release = inet_release,
847 .connect = inet_stream_connect,
848 .socketpair = sock_no_socketpair,
849 .accept = inet_accept,
850 .getname = inet_getname,
853 .listen = inet_listen,
854 .shutdown = inet_shutdown,
855 .setsockopt = sock_common_setsockopt,
856 .getsockopt = sock_common_getsockopt,
857 .sendmsg = inet_sendmsg,
858 .recvmsg = sock_common_recvmsg,
859 .mmap = sock_no_mmap,
860 .sendpage = tcp_sendpage
863 struct proto_ops inet_dgram_ops = {
865 .owner = THIS_MODULE,
866 .release = inet_release,
868 .connect = inet_dgram_connect,
869 .socketpair = sock_no_socketpair,
870 .accept = sock_no_accept,
871 .getname = inet_getname,
872 .poll = datagram_poll,
874 .listen = sock_no_listen,
875 .shutdown = inet_shutdown,
876 .setsockopt = sock_common_setsockopt,
877 .getsockopt = sock_common_getsockopt,
878 .sendmsg = inet_sendmsg,
879 .recvmsg = sock_common_recvmsg,
880 .mmap = sock_no_mmap,
881 .sendpage = inet_sendpage,
884 struct net_proto_family inet_family_ops = {
886 .create = inet_create,
887 .owner = THIS_MODULE,
891 extern void tcp_init(void);
892 extern void tcp_v4_init(struct net_proto_family *);
894 /* Upon startup we insert all the elements in inetsw_array[] into
895 * the linked list inetsw.
897 static struct inet_protosw inetsw_array[] =
901 .protocol = IPPROTO_TCP,
903 .ops = &inet_stream_ops,
906 .flags = INET_PROTOSW_PERMANENT,
911 .protocol = IPPROTO_UDP,
913 .ops = &inet_dgram_ops,
915 .no_check = UDP_CSUM_DEFAULT,
916 .flags = INET_PROTOSW_PERMANENT,
922 .protocol = IPPROTO_IP, /* wild card */
924 .ops = &inet_dgram_ops,
925 .capability = CAP_NET_RAW,
926 .no_check = UDP_CSUM_DEFAULT,
927 .flags = INET_PROTOSW_REUSE,
931 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
933 void inet_register_protosw(struct inet_protosw *p)
935 struct list_head *lh;
936 struct inet_protosw *answer;
937 int protocol = p->protocol;
938 struct list_head *last_perm;
940 spin_lock_bh(&inetsw_lock);
942 if (p->type >= SOCK_MAX)
945 /* If we are trying to override a permanent protocol, bail. */
947 last_perm = &inetsw[p->type];
948 list_for_each(lh, &inetsw[p->type]) {
949 answer = list_entry(lh, struct inet_protosw, list);
951 /* Check only the non-wild match. */
952 if (INET_PROTOSW_PERMANENT & answer->flags) {
953 if (protocol == answer->protocol)
963 /* Add the new entry after the last permanent entry if any, so that
964 * the new entry does not override a permanent entry when matched with
965 * a wild-card protocol. But it is allowed to override any existing
966 * non-permanent entry. This means that when we remove this entry, the
967 * system automatically returns to the old behavior.
969 list_add_rcu(&p->list, last_perm);
971 spin_unlock_bh(&inetsw_lock);
978 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
984 "Ignoring attempt to register invalid socket type %d.\n",
989 void inet_unregister_protosw(struct inet_protosw *p)
991 if (INET_PROTOSW_PERMANENT & p->flags) {
993 "Attempt to unregister permanent protocol %d.\n",
996 spin_lock_bh(&inetsw_lock);
997 list_del_rcu(&p->list);
998 spin_unlock_bh(&inetsw_lock);
1004 #ifdef CONFIG_IP_MULTICAST
1005 static struct net_protocol igmp_protocol = {
1006 .handler = igmp_rcv,
1010 static struct net_protocol tcp_protocol = {
1011 .handler = tcp_v4_rcv,
1012 .err_handler = tcp_v4_err,
1016 static struct net_protocol udp_protocol = {
1018 .err_handler = udp_err,
1022 static struct net_protocol icmp_protocol = {
1023 .handler = icmp_rcv,
1026 static int __init init_ipv4_mibs(void)
1028 net_statistics[0] = alloc_percpu(struct linux_mib);
1029 net_statistics[1] = alloc_percpu(struct linux_mib);
1030 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1031 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1032 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1033 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1034 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1035 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1036 udp_statistics[0] = alloc_percpu(struct udp_mib);
1037 udp_statistics[1] = alloc_percpu(struct udp_mib);
1039 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1040 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1041 && udp_statistics[0] && udp_statistics[1]))
1044 (void) tcp_mib_init();
1049 int ipv4_proc_init(void);
1050 extern void ipfrag_init(void);
1052 static int __init inet_init(void)
1054 struct sk_buff *dummy_skb;
1055 struct inet_protosw *q;
1056 struct list_head *r;
1058 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1059 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1063 tcp_sk_cachep = kmem_cache_create("tcp_sock",
1064 sizeof(struct tcp_sock), 0,
1065 SLAB_HWCACHE_ALIGN, NULL, NULL);
1066 udp_sk_cachep = kmem_cache_create("udp_sock",
1067 sizeof(struct udp_sock), 0,
1068 SLAB_HWCACHE_ALIGN, NULL, NULL);
1069 raw4_sk_cachep = kmem_cache_create("raw4_sock",
1070 sizeof(struct raw_sock), 0,
1071 SLAB_HWCACHE_ALIGN, NULL, NULL);
1072 if (!tcp_sk_cachep || !udp_sk_cachep || !raw4_sk_cachep)
1074 "inet_init: Can't create protocol sock SLAB caches!\n");
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"); ;
1147 module_init(inet_init);
1149 /* ------------------------------------------------------------------------ */
1151 #ifdef CONFIG_PROC_FS
1152 extern int fib_proc_init(void);
1153 extern void fib_proc_exit(void);
1154 extern int ip_misc_proc_init(void);
1155 extern int raw_proc_init(void);
1156 extern void raw_proc_exit(void);
1157 extern int tcp4_proc_init(void);
1158 extern void tcp4_proc_exit(void);
1159 extern int udp4_proc_init(void);
1160 extern void udp4_proc_exit(void);
1162 int __init ipv4_proc_init(void)
1166 if (raw_proc_init())
1168 if (tcp4_proc_init())
1170 if (udp4_proc_init())
1172 if (fib_proc_init())
1174 if (ip_misc_proc_init())
1191 #else /* CONFIG_PROC_FS */
1192 int __init ipv4_proc_init(void)
1196 #endif /* CONFIG_PROC_FS */
1198 MODULE_ALIAS_NETPROTO(PF_INET);
1200 EXPORT_SYMBOL(inet_accept);
1201 EXPORT_SYMBOL(inet_bind);
1202 EXPORT_SYMBOL(inet_dgram_connect);
1203 EXPORT_SYMBOL(inet_dgram_ops);
1204 EXPORT_SYMBOL(inet_family_ops);
1205 EXPORT_SYMBOL(inet_getname);
1206 EXPORT_SYMBOL(inet_ioctl);
1207 EXPORT_SYMBOL(inet_listen);
1208 EXPORT_SYMBOL(inet_register_protosw);
1209 EXPORT_SYMBOL(inet_release);
1210 EXPORT_SYMBOL(inet_sendmsg);
1211 EXPORT_SYMBOL(inet_shutdown);
1212 EXPORT_SYMBOL(inet_sock_destruct);
1213 EXPORT_SYMBOL(inet_stream_connect);
1214 EXPORT_SYMBOL(inet_stream_ops);
1215 EXPORT_SYMBOL(inet_unregister_protosw);
1216 EXPORT_SYMBOL(net_statistics);
1217 EXPORT_SYMBOL(tcp_protocol);
1218 EXPORT_SYMBOL(udp_protocol);
1220 #ifdef INET_REFCNT_DEBUG
1221 EXPORT_SYMBOL(inet_sock_nr);