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);
157 #warning MEF: figure out whether the following vserver net code is required by PlanetLab
160 clr_vx_info(&sk->sk_vx_info);
162 clr_nx_info(&sk->sk_nx_info);
166 dst_release(sk->sk_dst_cache);
167 #ifdef INET_REFCNT_DEBUG
168 atomic_dec(&inet_sock_nr);
169 printk(KERN_DEBUG "INET socket %p released, %d are still alive\n",
170 sk, atomic_read(&inet_sock_nr));
175 * The routines beyond this point handle the behaviour of an AF_INET
176 * socket object. Mostly it punts to the subprotocols of IP to do
181 * Automatically bind an unbound socket.
184 static int inet_autobind(struct sock *sk)
186 struct inet_sock *inet;
187 /* We may need to bind the socket. */
191 if (sk->sk_prot->get_port(sk, 0)) {
195 inet->sport = htons(inet->num);
202 * Move a socket into listening state.
204 int inet_listen(struct socket *sock, int backlog)
206 struct sock *sk = sock->sk;
207 unsigned char old_state;
213 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
216 old_state = sk->sk_state;
217 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
220 /* Really, if the socket is already in listen state
221 * we can only allow the backlog to be adjusted.
223 if (old_state != TCP_LISTEN) {
224 err = tcp_listen_start(sk);
228 sk->sk_max_ack_backlog = backlog;
237 * Create an inet socket.
240 static int inet_create(struct socket *sock, int protocol)
244 struct inet_protosw *answer;
245 struct inet_sock *inet;
246 struct proto *answer_prot;
247 unsigned char answer_flags;
248 char answer_no_check;
251 sock->state = SS_UNCONNECTED;
253 /* Look for the requested type/protocol pair. */
256 list_for_each_rcu(p, &inetsw[sock->type]) {
257 answer = list_entry(p, struct inet_protosw, list);
259 /* Check the non-wild match. */
260 if (protocol == answer->protocol) {
261 if (protocol != IPPROTO_IP)
264 /* Check for the two wild cases. */
265 if (IPPROTO_IP == protocol) {
266 protocol = answer->protocol;
269 if (IPPROTO_IP == answer->protocol)
275 err = -ESOCKTNOSUPPORT;
279 if ((protocol == IPPROTO_ICMP) && vx_ccaps(VXC_RAW_ICMP))
281 if (answer->capability > 0 && !capable(answer->capability))
284 err = -EPROTONOSUPPORT;
288 sock->ops = answer->ops;
289 answer_prot = answer->prot;
290 answer_no_check = answer->no_check;
291 answer_flags = answer->flags;
294 BUG_TRAP(answer_prot->slab != NULL);
297 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
302 sk->sk_no_check = answer_no_check;
303 if (INET_PROTOSW_REUSE & answer_flags)
308 if (SOCK_RAW == sock->type) {
309 inet->num = protocol;
310 if (IPPROTO_RAW == protocol)
314 if (ipv4_config.no_pmtu_disc)
315 inet->pmtudisc = IP_PMTUDISC_DONT;
317 inet->pmtudisc = IP_PMTUDISC_WANT;
321 sock_init_data(sock, sk);
323 sk->sk_destruct = inet_sock_destruct;
324 sk->sk_family = PF_INET;
325 sk->sk_protocol = protocol;
326 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
328 #warning MEF: figure out whether the following vserver net code is required by PlanetLab
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();
341 inet->mc_list = NULL;
343 #ifdef INET_REFCNT_DEBUG
344 atomic_inc(&inet_sock_nr);
348 /* It assumes that any protocol which allows
349 * the user to assign a number at socket
350 * creation time automatically
353 inet->sport = htons(inet->num);
354 /* Add to protocol hash chains. */
355 sk->sk_prot->hash(sk);
358 if (sk->sk_prot->init) {
359 err = sk->sk_prot->init(sk);
361 sk_common_release(sk);
372 * The peer socket should always be NULL (or else). When we call this
373 * function we are destroying the object and from then on nobody
374 * should refer to it.
376 int inet_release(struct socket *sock)
378 struct sock *sk = sock->sk;
383 /* Applications forget to leave groups before exiting */
384 ip_mc_drop_socket(sk);
386 /* If linger is set, we don't return until the close
387 * is complete. Otherwise we return immediately. The
388 * actually closing is done the same either way.
390 * If the close is due to the process exiting, we never
394 if (sock_flag(sk, SOCK_LINGER) &&
395 !(current->flags & PF_EXITING))
396 timeout = sk->sk_lingertime;
399 #warning MEF: figure out whether the following vserver net code is required by PlanetLab
402 clr_vx_info(&sk->sk_vx_info);
404 clr_nx_info(&sk->sk_nx_info);
407 sk->sk_prot->close(sk, timeout);
412 /* It is off by default, see below. */
413 int sysctl_ip_nonlocal_bind;
415 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
417 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
418 struct sock *sk = sock->sk;
419 struct inet_sock *inet = inet_sk(sk);
423 __u32 s_addr; /* Address used for validation */
424 __u32 s_addr1; /* Address used for socket */
425 __u32 s_addr2; /* Broadcast address for the socket */
426 struct nx_info *nxi = sk->sk_nx_info;
428 /* If the socket has its own bind function then use it. (RAW) */
429 if (sk->sk_prot->bind) {
430 err = sk->sk_prot->bind(sk, uaddr, addr_len);
434 if (addr_len < sizeof(struct sockaddr_in))
437 s_addr = addr->sin_addr.s_addr;
439 s_addr2 = 0xffffffffl;
441 vxdprintk(VXD_CBIT(net, 3),
442 "inet_bind(%p)* %p,%p;%lx %d.%d.%d.%d",
443 sk, sk->sk_nx_info, sk->sk_socket,
444 (sk->sk_socket?sk->sk_socket->flags:0),
447 __u32 v4_bcast = nxi->v4_bcast;
448 __u32 ipv4root = nxi->ipv4[0];
449 int nbipv4 = nxi->nbipv4;
452 /* bind to any for 1-n */
454 s_addr1 = (nbipv4 > 1) ? 0 : s_addr;
456 } else if (s_addr == 0x0100007f) {
457 /* rewrite localhost to ipv4root */
460 } else if (s_addr != v4_bcast) {
461 /* normal address bind */
462 if (!addr_in_nx_info(nxi, s_addr))
463 return -EADDRNOTAVAIL;
466 chk_addr_ret = inet_addr_type(s_addr);
468 vxdprintk(VXD_CBIT(net, 3),
469 "inet_bind(%p) %d.%d.%d.%d, %d.%d.%d.%d, %d.%d.%d.%d",
470 sk, VXD_QUAD(s_addr), VXD_QUAD(s_addr1), VXD_QUAD(s_addr2));
472 /* Not specified by any standard per-se, however it breaks too
473 * many applications when removed. It is unfortunate since
474 * allowing applications to make a non-local bind solves
475 * several problems with systems using dynamic addressing.
476 * (ie. your servers still start up even if your ISDN link
477 * is temporarily down)
479 err = -EADDRNOTAVAIL;
480 if (!sysctl_ip_nonlocal_bind &&
482 s_addr != INADDR_ANY &&
483 chk_addr_ret != RTN_LOCAL &&
484 chk_addr_ret != RTN_MULTICAST &&
485 chk_addr_ret != RTN_BROADCAST)
488 snum = ntohs(addr->sin_port);
490 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
493 /* We keep a pair of addresses. rcv_saddr is the one
494 * used by hash lookups, and saddr is used for transmit.
496 * In the BSD API these are the same except where it
497 * would be illegal to use them (multicast/broadcast) in
498 * which case the sending device address is used.
502 /* Check these errors (active socket, double bind). */
504 if (sk->sk_state != TCP_CLOSE || inet->num)
505 goto out_release_sock;
507 inet->rcv_saddr = inet->saddr = s_addr1;
508 inet->rcv_saddr2 = s_addr2;
509 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
510 inet->saddr = 0; /* Use device */
512 /* Make sure we are allowed to bind here. */
513 if (sk->sk_prot->get_port(sk, snum)) {
514 inet->saddr = inet->rcv_saddr = 0;
516 goto out_release_sock;
520 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
522 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
523 inet->sport = htons(inet->num);
534 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
535 int addr_len, int flags)
537 struct sock *sk = sock->sk;
539 if (uaddr->sa_family == AF_UNSPEC)
540 return sk->sk_prot->disconnect(sk, flags);
542 if (!inet_sk(sk)->num && inet_autobind(sk))
544 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
547 static long inet_wait_for_connect(struct sock *sk, long timeo)
551 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
553 /* Basic assumption: if someone sets sk->sk_err, he _must_
554 * change state of the socket from TCP_SYN_*.
555 * Connect() does not allow to get error notifications
556 * without closing the socket.
558 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
560 timeo = schedule_timeout(timeo);
562 if (signal_pending(current) || !timeo)
564 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
566 finish_wait(sk->sk_sleep, &wait);
571 * Connect to a remote host. There is regrettably still a little
572 * TCP 'magic' in here.
574 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
575 int addr_len, int flags)
577 struct sock *sk = sock->sk;
583 if (uaddr->sa_family == AF_UNSPEC) {
584 err = sk->sk_prot->disconnect(sk, flags);
585 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
589 switch (sock->state) {
598 /* Fall out of switch with err, set for this state */
602 if (sk->sk_state != TCP_CLOSE)
605 err = sk->sk_prot->connect(sk, uaddr, addr_len);
609 sock->state = SS_CONNECTING;
611 /* Just entered SS_CONNECTING state; the only
612 * difference is that return value in non-blocking
613 * case is EINPROGRESS, rather than EALREADY.
619 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
621 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
622 /* Error code is set above */
623 if (!timeo || !inet_wait_for_connect(sk, timeo))
626 err = sock_intr_errno(timeo);
627 if (signal_pending(current))
631 /* Connection was closed by RST, timeout, ICMP error
632 * or another process disconnected us.
634 if (sk->sk_state == TCP_CLOSE)
637 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
638 * and error was received after socket entered established state.
639 * Hence, it is handled normally after connect() return successfully.
642 sock->state = SS_CONNECTED;
649 err = sock_error(sk) ? : -ECONNABORTED;
650 sock->state = SS_UNCONNECTED;
651 if (sk->sk_prot->disconnect(sk, flags))
652 sock->state = SS_DISCONNECTING;
657 * Accept a pending connection. The TCP layer now gives BSD semantics.
660 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
662 struct sock *sk1 = sock->sk;
664 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
671 BUG_TRAP((1 << sk2->sk_state) &
672 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
674 sock_graft(sk2, newsock);
676 newsock->state = SS_CONNECTED;
685 * This does both peername and sockname.
687 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
688 int *uaddr_len, int peer)
690 struct sock *sk = sock->sk;
691 struct inet_sock *inet = inet_sk(sk);
692 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
694 sin->sin_family = AF_INET;
697 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
700 sin->sin_port = inet->dport;
701 sin->sin_addr.s_addr = inet->daddr;
703 __u32 addr = inet->rcv_saddr;
706 sin->sin_port = inet->sport;
707 sin->sin_addr.s_addr = addr;
709 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
710 *uaddr_len = sizeof(*sin);
714 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
717 struct sock *sk = sock->sk;
719 /* We may need to bind the socket. */
720 if (!inet_sk(sk)->num && inet_autobind(sk))
723 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
727 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
729 struct sock *sk = sock->sk;
731 /* We may need to bind the socket. */
732 if (!inet_sk(sk)->num && inet_autobind(sk))
735 if (sk->sk_prot->sendpage)
736 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
737 return sock_no_sendpage(sock, page, offset, size, flags);
741 int inet_shutdown(struct socket *sock, int how)
743 struct sock *sk = sock->sk;
746 /* This should really check to make sure
747 * the socket is a TCP socket. (WHY AC...)
749 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
752 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
756 if (sock->state == SS_CONNECTING) {
757 if ((1 << sk->sk_state) &
758 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
759 sock->state = SS_DISCONNECTING;
761 sock->state = SS_CONNECTED;
764 switch (sk->sk_state) {
767 /* Hack to wake up other listeners, who can poll for
768 POLLHUP, even on eg. unconnected UDP sockets -- RR */
770 sk->sk_shutdown |= how;
771 if (sk->sk_prot->shutdown)
772 sk->sk_prot->shutdown(sk, how);
775 /* Remaining two branches are temporary solution for missing
776 * close() in multithreaded environment. It is _not_ a good idea,
777 * but we have no choice until close() is repaired at VFS level.
780 if (!(how & RCV_SHUTDOWN))
784 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
785 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
789 /* Wake up anyone sleeping in poll. */
790 sk->sk_state_change(sk);
796 * ioctl() calls you can issue on an INET socket. Most of these are
797 * device configuration and stuff and very rarely used. Some ioctls
798 * pass on to the socket itself.
800 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
801 * loads the devconfigure module does its configuring and unloads it.
802 * There's a good 20K of config code hanging around the kernel.
805 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
807 struct sock *sk = sock->sk;
812 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
817 err = ip_rt_ioctl(cmd, (void __user *)arg);
822 err = arp_ioctl(cmd, (void __user *)arg);
835 err = devinet_ioctl(cmd, (void __user *)arg);
838 if (!sk->sk_prot->ioctl ||
839 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
841 err = dev_ioctl(cmd, (void __user *)arg);
847 struct proto_ops inet_stream_ops = {
849 .owner = THIS_MODULE,
850 .release = inet_release,
852 .connect = inet_stream_connect,
853 .socketpair = sock_no_socketpair,
854 .accept = inet_accept,
855 .getname = inet_getname,
858 .listen = inet_listen,
859 .shutdown = inet_shutdown,
860 .setsockopt = sock_common_setsockopt,
861 .getsockopt = sock_common_getsockopt,
862 .sendmsg = inet_sendmsg,
863 .recvmsg = sock_common_recvmsg,
864 .mmap = sock_no_mmap,
865 .sendpage = tcp_sendpage
868 struct proto_ops inet_dgram_ops = {
870 .owner = THIS_MODULE,
871 .release = inet_release,
873 .connect = inet_dgram_connect,
874 .socketpair = sock_no_socketpair,
875 .accept = sock_no_accept,
876 .getname = inet_getname,
879 .listen = sock_no_listen,
880 .shutdown = inet_shutdown,
881 .setsockopt = sock_common_setsockopt,
882 .getsockopt = sock_common_getsockopt,
883 .sendmsg = inet_sendmsg,
884 .recvmsg = sock_common_recvmsg,
885 .mmap = sock_no_mmap,
886 .sendpage = inet_sendpage,
890 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
893 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
894 struct proto_ops inet_sockraw_ops;
895 EXPORT_SYMBOL(inet_sockraw_ops);
899 struct proto_ops inet_sockraw_ops = {
901 .owner = THIS_MODULE,
902 .release = inet_release,
904 .connect = inet_dgram_connect,
905 .socketpair = sock_no_socketpair,
906 .accept = sock_no_accept,
907 .getname = inet_getname,
908 .poll = datagram_poll,
910 .listen = sock_no_listen,
911 .shutdown = inet_shutdown,
912 .setsockopt = sock_common_setsockopt,
913 .getsockopt = sock_common_getsockopt,
914 .sendmsg = inet_sendmsg,
915 .recvmsg = sock_common_recvmsg,
916 .mmap = sock_no_mmap,
917 .sendpage = inet_sendpage,
920 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
922 EXPORT_SYMBOL(vnet_active);
923 struct net_proto_family inet_family_ops;
924 EXPORT_SYMBOL(inet_family_ops);
928 struct net_proto_family inet_family_ops = {
930 .create = inet_create,
931 .owner = THIS_MODULE,
935 extern void tcp_init(void);
936 extern void tcp_v4_init(struct net_proto_family *);
938 /* Upon startup we insert all the elements in inetsw_array[] into
939 * the linked list inetsw.
941 static struct inet_protosw inetsw_array[] =
945 .protocol = IPPROTO_TCP,
947 .ops = &inet_stream_ops,
950 .flags = INET_PROTOSW_PERMANENT,
955 .protocol = IPPROTO_UDP,
957 .ops = &inet_dgram_ops,
959 .no_check = UDP_CSUM_DEFAULT,
960 .flags = INET_PROTOSW_PERMANENT,
966 .protocol = IPPROTO_IP, /* wild card */
968 .ops = &inet_sockraw_ops,
969 .capability = CAP_NET_RAW,
970 .no_check = UDP_CSUM_DEFAULT,
971 .flags = INET_PROTOSW_REUSE,
975 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
977 void inet_register_protosw(struct inet_protosw *p)
979 struct list_head *lh;
980 struct inet_protosw *answer;
981 int protocol = p->protocol;
982 struct list_head *last_perm;
984 spin_lock_bh(&inetsw_lock);
986 if (p->type >= SOCK_MAX)
989 /* If we are trying to override a permanent protocol, bail. */
991 last_perm = &inetsw[p->type];
992 list_for_each(lh, &inetsw[p->type]) {
993 answer = list_entry(lh, struct inet_protosw, list);
995 /* Check only the non-wild match. */
996 if (INET_PROTOSW_PERMANENT & answer->flags) {
997 if (protocol == answer->protocol)
1007 /* Add the new entry after the last permanent entry if any, so that
1008 * the new entry does not override a permanent entry when matched with
1009 * a wild-card protocol. But it is allowed to override any existing
1010 * non-permanent entry. This means that when we remove this entry, the
1011 * system automatically returns to the old behavior.
1013 list_add_rcu(&p->list, last_perm);
1015 spin_unlock_bh(&inetsw_lock);
1022 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1028 "Ignoring attempt to register invalid socket type %d.\n",
1033 void inet_unregister_protosw(struct inet_protosw *p)
1035 if (INET_PROTOSW_PERMANENT & p->flags) {
1037 "Attempt to unregister permanent protocol %d.\n",
1040 spin_lock_bh(&inetsw_lock);
1041 list_del_rcu(&p->list);
1042 spin_unlock_bh(&inetsw_lock);
1048 #ifdef CONFIG_IP_MULTICAST
1049 static struct net_protocol igmp_protocol = {
1050 .handler = igmp_rcv,
1054 static struct net_protocol tcp_protocol = {
1055 .handler = tcp_v4_rcv,
1056 .err_handler = tcp_v4_err,
1060 static struct net_protocol udp_protocol = {
1062 .err_handler = udp_err,
1066 static struct net_protocol icmp_protocol = {
1067 .handler = icmp_rcv,
1070 static int __init init_ipv4_mibs(void)
1072 net_statistics[0] = alloc_percpu(struct linux_mib);
1073 net_statistics[1] = alloc_percpu(struct linux_mib);
1074 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1075 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1076 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1077 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1078 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1079 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1080 udp_statistics[0] = alloc_percpu(struct udp_mib);
1081 udp_statistics[1] = alloc_percpu(struct udp_mib);
1083 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1084 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1085 && udp_statistics[0] && udp_statistics[1]))
1088 (void) tcp_mib_init();
1093 static int ipv4_proc_init(void);
1094 extern void ipfrag_init(void);
1096 static int __init inet_init(void)
1098 struct sk_buff *dummy_skb;
1099 struct inet_protosw *q;
1100 struct list_head *r;
1103 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1104 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1108 rc = proto_register(&tcp_prot, 1);
1112 rc = proto_register(&udp_prot, 1);
1114 goto out_unregister_tcp_proto;
1116 rc = proto_register(&raw_prot, 1);
1118 goto out_unregister_udp_proto;
1121 * Tell SOCKET that we are alive...
1124 (void)sock_register(&inet_family_ops);
1127 * Add all the base protocols.
1130 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1131 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1132 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1133 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1134 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1135 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1136 #ifdef CONFIG_IP_MULTICAST
1137 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1138 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1141 /* Register the socket-side information for inet_create. */
1142 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1145 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1146 inet_register_protosw(q);
1149 * Set the ARP module up
1155 * Set the IP module up
1160 tcp_v4_init(&inet_family_ops);
1162 /* Setup TCP slab cache for open requests. */
1167 * Set the ICMP layer up
1170 icmp_init(&inet_family_ops);
1173 * Initialise the multicast router
1175 #if defined(CONFIG_IP_MROUTE)
1179 * Initialise per-cpu ipv4 mibs
1182 if(init_ipv4_mibs())
1183 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1192 out_unregister_tcp_proto:
1193 proto_unregister(&tcp_prot);
1194 out_unregister_udp_proto:
1195 proto_unregister(&udp_prot);
1199 module_init(inet_init);
1201 /* ------------------------------------------------------------------------ */
1203 #ifdef CONFIG_PROC_FS
1204 extern int fib_proc_init(void);
1205 extern void fib_proc_exit(void);
1206 extern int ip_misc_proc_init(void);
1207 extern int raw_proc_init(void);
1208 extern void raw_proc_exit(void);
1209 extern int tcp4_proc_init(void);
1210 extern void tcp4_proc_exit(void);
1211 extern int udp4_proc_init(void);
1212 extern void udp4_proc_exit(void);
1214 static int __init ipv4_proc_init(void)
1218 if (raw_proc_init())
1220 if (tcp4_proc_init())
1222 if (udp4_proc_init())
1224 if (fib_proc_init())
1226 if (ip_misc_proc_init())
1243 #else /* CONFIG_PROC_FS */
1244 static int __init ipv4_proc_init(void)
1248 #endif /* CONFIG_PROC_FS */
1250 MODULE_ALIAS_NETPROTO(PF_INET);
1252 EXPORT_SYMBOL(inet_accept);
1253 EXPORT_SYMBOL(inet_bind);
1254 EXPORT_SYMBOL(inet_dgram_connect);
1255 EXPORT_SYMBOL(inet_dgram_ops);
1256 EXPORT_SYMBOL(inet_getname);
1257 EXPORT_SYMBOL(inet_ioctl);
1258 EXPORT_SYMBOL(inet_listen);
1259 EXPORT_SYMBOL(inet_register_protosw);
1260 EXPORT_SYMBOL(inet_release);
1261 EXPORT_SYMBOL(inet_sendmsg);
1262 EXPORT_SYMBOL(inet_shutdown);
1263 EXPORT_SYMBOL(inet_sock_destruct);
1264 EXPORT_SYMBOL(inet_stream_connect);
1265 EXPORT_SYMBOL(inet_stream_ops);
1266 EXPORT_SYMBOL(inet_unregister_protosw);
1267 EXPORT_SYMBOL(net_statistics);
1268 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
1270 #ifdef INET_REFCNT_DEBUG
1271 EXPORT_SYMBOL(inet_sock_nr);