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);
161 dst_release(sk->sk_dst_cache);
162 #ifdef INET_REFCNT_DEBUG
163 atomic_dec(&inet_sock_nr);
164 printk(KERN_DEBUG "INET socket %p released, %d are still alive\n",
165 sk, atomic_read(&inet_sock_nr));
169 void inet_sock_release(struct sock *sk)
171 if (sk->sk_prot->destroy)
172 sk->sk_prot->destroy(sk);
174 /* Observation: when inet_sock_release is called, processes have
175 * no access to socket. But net still has.
176 * Step one, detach it from networking:
178 * A. Remove from hash tables.
181 sk->sk_prot->unhash(sk);
183 /* In this point socket cannot receive new packets,
184 * but it is possible that some packets are in flight
185 * because some CPU runs receiver and did hash table lookup
186 * before we unhashed socket. They will achieve receive queue
187 * and will be purged by socket destructor.
189 * Also we still have packets pending on receive
190 * queue and probably, our own packets waiting in device queues.
191 * sock_destroy will drain receive queue, but transmitted
192 * packets will delay socket destruction until the last reference
198 xfrm_sk_free_policy(sk);
200 #ifdef INET_REFCNT_DEBUG
201 if (atomic_read(&sk->sk_refcnt) != 1)
202 printk(KERN_DEBUG "Destruction inet %p delayed, c=%d\n",
203 sk, atomic_read(&sk->sk_refcnt));
210 * The routines beyond this point handle the behaviour of an AF_INET
211 * socket object. Mostly it punts to the subprotocols of IP to do
216 * Set socket options on an inet socket.
218 int inet_setsockopt(struct socket *sock, int level, int optname,
219 char *optval, int optlen)
221 struct sock *sk = sock->sk;
223 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
227 * Get a socket option on an AF_INET socket.
229 * FIX: POSIX 1003.1g is very ambiguous here. It states that
230 * asynchronous errors should be reported by getsockopt. We assume
231 * this means if you specify SO_ERROR (otherwise whats the point of it).
234 int inet_getsockopt(struct socket *sock, int level, int optname,
235 char *optval, int *optlen)
237 struct sock *sk = sock->sk;
239 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
243 * Automatically bind an unbound socket.
246 static int inet_autobind(struct sock *sk)
248 struct inet_opt *inet;
249 /* We may need to bind the socket. */
253 if (sk->sk_prot->get_port(sk, 0)) {
257 inet->sport = htons(inet->num);
264 * Move a socket into listening state.
266 int inet_listen(struct socket *sock, int backlog)
268 struct sock *sk = sock->sk;
269 unsigned char old_state;
275 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
278 old_state = sk->sk_state;
279 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
282 /* Really, if the socket is already in listen state
283 * we can only allow the backlog to be adjusted.
285 if (old_state != TCP_LISTEN) {
286 err = tcp_listen_start(sk);
290 sk->sk_max_ack_backlog = backlog;
298 static __inline__ kmem_cache_t *inet_sk_slab(int protocol)
300 kmem_cache_t* rc = tcp_sk_cachep;
302 if (protocol == IPPROTO_UDP)
304 else if (protocol == IPPROTO_RAW)
309 static __inline__ int inet_sk_size(int protocol)
311 int rc = sizeof(struct tcp_sock);
313 if (protocol == IPPROTO_UDP)
314 rc = sizeof(struct udp_sock);
315 else if (protocol == IPPROTO_RAW)
316 rc = sizeof(struct raw_sock);
321 * Create an inet socket.
324 static int inet_create(struct socket *sock, int protocol)
328 struct inet_protosw *answer;
329 struct inet_opt *inet;
332 sock->state = SS_UNCONNECTED;
333 sk = sk_alloc(PF_INET, GFP_KERNEL, inet_sk_size(protocol),
334 inet_sk_slab(protocol));
338 /* Look for the requested type/protocol pair. */
341 list_for_each_rcu(p, &inetsw[sock->type]) {
342 answer = list_entry(p, struct inet_protosw, list);
344 /* Check the non-wild match. */
345 if (protocol == answer->protocol) {
346 if (protocol != IPPROTO_IP)
349 /* Check for the two wild cases. */
350 if (IPPROTO_IP == protocol) {
351 protocol = answer->protocol;
354 if (IPPROTO_IP == answer->protocol)
360 err = -ESOCKTNOSUPPORT;
364 if (answer->capability > 0 && !capable(answer->capability))
366 err = -EPROTONOSUPPORT;
370 sock->ops = answer->ops;
371 sk->sk_prot = answer->prot;
372 sk->sk_no_check = answer->no_check;
373 if (INET_PROTOSW_REUSE & answer->flags)
379 if (SOCK_RAW == sock->type) {
380 inet->num = protocol;
381 if (IPPROTO_RAW == protocol)
385 if (ipv4_config.no_pmtu_disc)
386 inet->pmtudisc = IP_PMTUDISC_DONT;
388 inet->pmtudisc = IP_PMTUDISC_WANT;
392 sock_init_data(sock, sk);
393 sk_set_owner(sk, THIS_MODULE);
395 sk->sk_destruct = inet_sock_destruct;
397 sk->sk_family = PF_INET;
398 sk->sk_protocol = protocol;
399 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
405 inet->mc_list = NULL;
407 #ifdef INET_REFCNT_DEBUG
408 atomic_inc(&inet_sock_nr);
412 /* It assumes that any protocol which allows
413 * the user to assign a number at socket
414 * creation time automatically
417 inet->sport = htons(inet->num);
418 /* Add to protocol hash chains. */
419 sk->sk_prot->hash(sk);
422 if (sk->sk_prot->init) {
423 err = sk->sk_prot->init(sk);
425 inet_sock_release(sk);
437 * The peer socket should always be NULL (or else). When we call this
438 * function we are destroying the object and from then on nobody
439 * should refer to it.
441 int inet_release(struct socket *sock)
443 struct sock *sk = sock->sk;
448 /* Applications forget to leave groups before exiting */
449 ip_mc_drop_socket(sk);
451 /* If linger is set, we don't return until the close
452 * is complete. Otherwise we return immediately. The
453 * actually closing is done the same either way.
455 * If the close is due to the process exiting, we never
459 if (sock_flag(sk, SOCK_LINGER) &&
460 !(current->flags & PF_EXITING))
461 timeout = sk->sk_lingertime;
463 sk->sk_prot->close(sk, timeout);
468 /* It is off by default, see below. */
469 int sysctl_ip_nonlocal_bind;
471 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
473 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
474 struct sock *sk = sock->sk;
475 struct inet_opt *inet = inet_sk(sk);
480 /* If the socket has its own bind function then use it. (RAW) */
481 if (sk->sk_prot->bind) {
482 err = sk->sk_prot->bind(sk, uaddr, addr_len);
486 if (addr_len < sizeof(struct sockaddr_in))
489 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
491 /* Not specified by any standard per-se, however it breaks too
492 * many applications when removed. It is unfortunate since
493 * allowing applications to make a non-local bind solves
494 * several problems with systems using dynamic addressing.
495 * (ie. your servers still start up even if your ISDN link
496 * is temporarily down)
498 err = -EADDRNOTAVAIL;
499 if (!sysctl_ip_nonlocal_bind &&
501 addr->sin_addr.s_addr != INADDR_ANY &&
502 chk_addr_ret != RTN_LOCAL &&
503 chk_addr_ret != RTN_MULTICAST &&
504 chk_addr_ret != RTN_BROADCAST)
507 snum = ntohs(addr->sin_port);
509 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
512 /* We keep a pair of addresses. rcv_saddr is the one
513 * used by hash lookups, and saddr is used for transmit.
515 * In the BSD API these are the same except where it
516 * would be illegal to use them (multicast/broadcast) in
517 * which case the sending device address is used.
521 /* Check these errors (active socket, double bind). */
523 if (sk->sk_state != TCP_CLOSE || inet->num)
524 goto out_release_sock;
526 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
527 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
528 inet->saddr = 0; /* Use device */
530 /* Make sure we are allowed to bind here. */
531 if (sk->sk_prot->get_port(sk, snum)) {
532 inet->saddr = inet->rcv_saddr = 0;
534 goto out_release_sock;
538 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
540 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
541 inet->sport = htons(inet->num);
552 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
553 int addr_len, int flags)
555 struct sock *sk = sock->sk;
557 if (uaddr->sa_family == AF_UNSPEC)
558 return sk->sk_prot->disconnect(sk, flags);
560 if (!inet_sk(sk)->num && inet_autobind(sk))
562 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
565 static long inet_wait_for_connect(struct sock *sk, long timeo)
569 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
571 /* Basic assumption: if someone sets sk->sk_err, he _must_
572 * change state of the socket from TCP_SYN_*.
573 * Connect() does not allow to get error notifications
574 * without closing the socket.
576 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
578 timeo = schedule_timeout(timeo);
580 if (signal_pending(current) || !timeo)
582 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
584 finish_wait(sk->sk_sleep, &wait);
589 * Connect to a remote host. There is regrettably still a little
590 * TCP 'magic' in here.
592 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
593 int addr_len, int flags)
595 struct sock *sk = sock->sk;
601 if (uaddr->sa_family == AF_UNSPEC) {
602 err = sk->sk_prot->disconnect(sk, flags);
603 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
607 switch (sock->state) {
616 /* Fall out of switch with err, set for this state */
620 if (sk->sk_state != TCP_CLOSE)
623 err = sk->sk_prot->connect(sk, uaddr, addr_len);
627 sock->state = SS_CONNECTING;
629 /* Just entered SS_CONNECTING state; the only
630 * difference is that return value in non-blocking
631 * case is EINPROGRESS, rather than EALREADY.
637 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
639 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
640 /* Error code is set above */
641 if (!timeo || !inet_wait_for_connect(sk, timeo))
644 err = sock_intr_errno(timeo);
645 if (signal_pending(current))
649 /* Connection was closed by RST, timeout, ICMP error
650 * or another process disconnected us.
652 if (sk->sk_state == TCP_CLOSE)
655 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
656 * and error was received after socket entered established state.
657 * Hence, it is handled normally after connect() return successfully.
660 sock->state = SS_CONNECTED;
667 err = sock_error(sk) ? : -ECONNABORTED;
668 sock->state = SS_UNCONNECTED;
669 if (sk->sk_prot->disconnect(sk, flags))
670 sock->state = SS_DISCONNECTING;
675 * Accept a pending connection. The TCP layer now gives BSD semantics.
678 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
680 struct sock *sk1 = sock->sk;
682 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
689 BUG_TRAP((1 << sk2->sk_state) &
690 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
692 sock_graft(sk2, newsock);
694 newsock->state = SS_CONNECTED;
703 * This does both peername and sockname.
705 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
706 int *uaddr_len, int peer)
708 struct sock *sk = sock->sk;
709 struct inet_opt *inet = inet_sk(sk);
710 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
712 sin->sin_family = AF_INET;
715 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
718 sin->sin_port = inet->dport;
719 sin->sin_addr.s_addr = inet->daddr;
721 __u32 addr = inet->rcv_saddr;
724 sin->sin_port = inet->sport;
725 sin->sin_addr.s_addr = addr;
727 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
728 *uaddr_len = sizeof(*sin);
733 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
734 size_t size, int flags)
736 struct sock *sk = sock->sk;
740 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
741 flags & ~MSG_DONTWAIT, &addr_len);
743 msg->msg_namelen = addr_len;
748 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
751 struct sock *sk = sock->sk;
753 /* We may need to bind the socket. */
754 if (!inet_sk(sk)->num && inet_autobind(sk))
757 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
761 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
763 struct sock *sk = sock->sk;
765 /* We may need to bind the socket. */
766 if (!inet_sk(sk)->num && inet_autobind(sk))
769 if (sk->sk_prot->sendpage)
770 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
771 return sock_no_sendpage(sock, page, offset, size, flags);
775 int inet_shutdown(struct socket *sock, int how)
777 struct sock *sk = sock->sk;
780 /* This should really check to make sure
781 * the socket is a TCP socket. (WHY AC...)
783 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
786 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
790 if (sock->state == SS_CONNECTING) {
791 if ((1 << sk->sk_state) &
792 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
793 sock->state = SS_DISCONNECTING;
795 sock->state = SS_CONNECTED;
798 switch (sk->sk_state) {
801 /* Hack to wake up other listeners, who can poll for
802 POLLHUP, even on eg. unconnected UDP sockets -- RR */
804 sk->sk_shutdown |= how;
805 if (sk->sk_prot->shutdown)
806 sk->sk_prot->shutdown(sk, how);
809 /* Remaining two branches are temporary solution for missing
810 * close() in multithreaded environment. It is _not_ a good idea,
811 * but we have no choice until close() is repaired at VFS level.
814 if (!(how & RCV_SHUTDOWN))
818 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
819 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
823 /* Wake up anyone sleeping in poll. */
824 sk->sk_state_change(sk);
830 * ioctl() calls you can issue on an INET socket. Most of these are
831 * device configuration and stuff and very rarely used. Some ioctls
832 * pass on to the socket itself.
834 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
835 * loads the devconfigure module does its configuring and unloads it.
836 * There's a good 20K of config code hanging around the kernel.
839 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
841 struct sock *sk = sock->sk;
846 err = sock_get_timestamp(sk, (struct timeval *)arg);
851 err = ip_rt_ioctl(cmd, (void *)arg);
856 err = arp_ioctl(cmd, (void *)arg);
869 err = devinet_ioctl(cmd, (void *)arg);
872 if (!sk->sk_prot->ioctl ||
873 (err = sk->sk_prot->ioctl(sk, cmd, arg)) ==
875 err = dev_ioctl(cmd, (void *)arg);
881 struct proto_ops inet_stream_ops = {
883 .owner = THIS_MODULE,
884 .release = inet_release,
886 .connect = inet_stream_connect,
887 .socketpair = sock_no_socketpair,
888 .accept = inet_accept,
889 .getname = inet_getname,
892 .listen = inet_listen,
893 .shutdown = inet_shutdown,
894 .setsockopt = inet_setsockopt,
895 .getsockopt = inet_getsockopt,
896 .sendmsg = inet_sendmsg,
897 .recvmsg = inet_recvmsg,
898 .mmap = sock_no_mmap,
899 .sendpage = tcp_sendpage
902 struct proto_ops inet_dgram_ops = {
904 .owner = THIS_MODULE,
905 .release = inet_release,
907 .connect = inet_dgram_connect,
908 .socketpair = sock_no_socketpair,
909 .accept = sock_no_accept,
910 .getname = inet_getname,
911 .poll = datagram_poll,
913 .listen = sock_no_listen,
914 .shutdown = inet_shutdown,
915 .setsockopt = inet_setsockopt,
916 .getsockopt = inet_getsockopt,
917 .sendmsg = inet_sendmsg,
918 .recvmsg = inet_recvmsg,
919 .mmap = sock_no_mmap,
920 .sendpage = inet_sendpage,
923 struct net_proto_family inet_family_ops = {
925 .create = inet_create,
926 .owner = THIS_MODULE,
930 extern void tcp_init(void);
931 extern void tcp_v4_init(struct net_proto_family *);
933 /* Upon startup we insert all the elements in inetsw_array[] into
934 * the linked list inetsw.
936 static struct inet_protosw inetsw_array[] =
940 .protocol = IPPROTO_TCP,
942 .ops = &inet_stream_ops,
945 .flags = INET_PROTOSW_PERMANENT,
950 .protocol = IPPROTO_UDP,
952 .ops = &inet_dgram_ops,
954 .no_check = UDP_CSUM_DEFAULT,
955 .flags = INET_PROTOSW_PERMANENT,
961 .protocol = IPPROTO_IP, /* wild card */
963 .ops = &inet_dgram_ops,
964 .capability = CAP_NET_RAW,
965 .no_check = UDP_CSUM_DEFAULT,
966 .flags = INET_PROTOSW_REUSE,
970 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
972 void inet_register_protosw(struct inet_protosw *p)
974 struct list_head *lh;
975 struct inet_protosw *answer;
976 int protocol = p->protocol;
977 struct list_head *last_perm;
979 spin_lock_bh(&inetsw_lock);
981 if (p->type > SOCK_MAX)
984 /* If we are trying to override a permanent protocol, bail. */
986 last_perm = &inetsw[p->type];
987 list_for_each(lh, &inetsw[p->type]) {
988 answer = list_entry(lh, struct inet_protosw, list);
990 /* Check only the non-wild match. */
991 if (INET_PROTOSW_PERMANENT & answer->flags) {
992 if (protocol == answer->protocol)
1002 /* Add the new entry after the last permanent entry if any, so that
1003 * the new entry does not override a permanent entry when matched with
1004 * a wild-card protocol. But it is allowed to override any existing
1005 * non-permanent entry. This means that when we remove this entry, the
1006 * system automatically returns to the old behavior.
1008 list_add_rcu(&p->list, last_perm);
1010 spin_unlock_bh(&inetsw_lock);
1017 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1023 "Ignoring attempt to register invalid socket type %d.\n",
1028 void inet_unregister_protosw(struct inet_protosw *p)
1030 if (INET_PROTOSW_PERMANENT & p->flags) {
1032 "Attempt to unregister permanent protocol %d.\n",
1035 spin_lock_bh(&inetsw_lock);
1036 list_del_rcu(&p->list);
1037 spin_unlock_bh(&inetsw_lock);
1043 #ifdef CONFIG_IP_MULTICAST
1044 static struct inet_protocol igmp_protocol = {
1045 .handler = igmp_rcv,
1049 static struct inet_protocol tcp_protocol = {
1050 .handler = tcp_v4_rcv,
1051 .err_handler = tcp_v4_err,
1055 static struct inet_protocol udp_protocol = {
1057 .err_handler = udp_err,
1061 static struct inet_protocol icmp_protocol = {
1062 .handler = icmp_rcv,
1065 static int __init init_ipv4_mibs(void)
1067 net_statistics[0] = alloc_percpu(struct linux_mib);
1068 net_statistics[1] = alloc_percpu(struct linux_mib);
1069 ip_statistics[0] = alloc_percpu(struct ip_mib);
1070 ip_statistics[1] = alloc_percpu(struct ip_mib);
1071 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1072 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1073 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1074 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1075 udp_statistics[0] = alloc_percpu(struct udp_mib);
1076 udp_statistics[1] = alloc_percpu(struct udp_mib);
1078 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1079 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1080 && udp_statistics[0] && udp_statistics[1]))
1083 (void) tcp_mib_init();
1088 int ipv4_proc_init(void);
1089 extern void ipfrag_init(void);
1091 static int __init inet_init(void)
1093 struct sk_buff *dummy_skb;
1094 struct inet_protosw *q;
1095 struct list_head *r;
1097 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb)) {
1098 printk(KERN_CRIT "%s: panic\n", __FUNCTION__);
1102 tcp_sk_cachep = kmem_cache_create("tcp_sock",
1103 sizeof(struct tcp_sock), 0,
1104 SLAB_HWCACHE_ALIGN, 0, 0);
1105 udp_sk_cachep = kmem_cache_create("udp_sock",
1106 sizeof(struct udp_sock), 0,
1107 SLAB_HWCACHE_ALIGN, 0, 0);
1108 raw4_sk_cachep = kmem_cache_create("raw4_sock",
1109 sizeof(struct raw_sock), 0,
1110 SLAB_HWCACHE_ALIGN, 0, 0);
1111 if (!tcp_sk_cachep || !udp_sk_cachep || !raw4_sk_cachep)
1113 "inet_init: Can't create protocol sock SLAB caches!\n");
1115 * Tell SOCKET that we are alive...
1118 (void)sock_register(&inet_family_ops);
1121 * Add all the base protocols.
1124 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1125 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1126 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1127 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1128 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1129 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1130 #ifdef CONFIG_IP_MULTICAST
1131 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1132 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1135 /* Register the socket-side information for inet_create. */
1136 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1139 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1140 inet_register_protosw(q);
1143 * Set the ARP module up
1149 * Set the IP module up
1154 tcp_v4_init(&inet_family_ops);
1156 /* Setup TCP slab cache for open requests. */
1161 * Set the ICMP layer up
1164 icmp_init(&inet_family_ops);
1167 * Initialise the multicast router
1169 #if defined(CONFIG_IP_MROUTE)
1173 * Initialise per-cpu ipv4 mibs
1176 if(init_ipv4_mibs())
1177 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1186 module_init(inet_init);
1188 /* ------------------------------------------------------------------------ */
1190 #ifdef CONFIG_PROC_FS
1191 extern int fib_proc_init(void);
1192 extern void fib_proc_exit(void);
1193 extern int ip_misc_proc_init(void);
1194 extern int raw_proc_init(void);
1195 extern void raw_proc_exit(void);
1196 extern int tcp4_proc_init(void);
1197 extern void tcp4_proc_exit(void);
1198 extern int udp4_proc_init(void);
1199 extern void udp4_proc_exit(void);
1201 int __init ipv4_proc_init(void)
1205 if (raw_proc_init())
1207 if (tcp4_proc_init())
1209 if (udp4_proc_init())
1211 if (fib_proc_init())
1213 if (ip_misc_proc_init())
1230 #else /* CONFIG_PROC_FS */
1231 int __init ipv4_proc_init(void)
1235 #endif /* CONFIG_PROC_FS */
1237 MODULE_ALIAS_NETPROTO(PF_INET);
1239 EXPORT_SYMBOL(inet_accept);
1240 EXPORT_SYMBOL(inet_bind);
1241 EXPORT_SYMBOL(inet_dgram_connect);
1242 EXPORT_SYMBOL(inet_dgram_ops);
1243 EXPORT_SYMBOL(inet_family_ops);
1244 EXPORT_SYMBOL(inet_getname);
1245 EXPORT_SYMBOL(inet_getsockopt);
1246 EXPORT_SYMBOL(inet_ioctl);
1247 EXPORT_SYMBOL(inet_listen);
1248 EXPORT_SYMBOL(inet_recvmsg);
1249 EXPORT_SYMBOL(inet_register_protosw);
1250 EXPORT_SYMBOL(inet_release);
1251 EXPORT_SYMBOL(inet_sendmsg);
1252 EXPORT_SYMBOL(inet_setsockopt);
1253 EXPORT_SYMBOL(inet_shutdown);
1254 EXPORT_SYMBOL(inet_sock_destruct);
1255 EXPORT_SYMBOL(inet_sock_release);
1256 EXPORT_SYMBOL(inet_stream_connect);
1257 EXPORT_SYMBOL(inet_stream_ops);
1258 EXPORT_SYMBOL(inet_unregister_protosw);
1259 EXPORT_SYMBOL(net_statistics);
1260 EXPORT_SYMBOL(tcp_protocol);
1261 EXPORT_SYMBOL(udp_protocol);
1263 #ifdef INET_REFCNT_DEBUG
1264 EXPORT_SYMBOL(inet_sock_nr);