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/err.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/capability.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>
90 #include <linux/random.h>
92 #include <asm/uaccess.h>
93 #include <asm/system.h>
95 #include <linux/smp_lock.h>
96 #include <linux/inet.h>
97 #include <linux/igmp.h>
98 #include <linux/inetdevice.h>
99 #include <linux/netdevice.h>
101 #include <net/protocol.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
108 #include <net/udplite.h>
109 #include <linux/skbuff.h>
110 #include <net/sock.h>
112 #include <net/icmp.h>
113 #include <net/ipip.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #ifdef CONFIG_IP_MROUTE
117 #include <linux/mroute.h>
120 DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
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);
155 dst_release(sk->sk_dst_cache);
156 sk_refcnt_debug_dec(sk);
160 * The routines beyond this point handle the behaviour of an AF_INET
161 * socket object. Mostly it punts to the subprotocols of IP to do
166 * Automatically bind an unbound socket.
169 static int inet_autobind(struct sock *sk)
171 struct inet_sock *inet;
172 /* We may need to bind the socket. */
176 if (sk->sk_prot->get_port(sk, 0)) {
180 inet->sport = htons(inet->num);
187 * Move a socket into listening state.
189 int inet_listen(struct socket *sock, int backlog)
191 struct sock *sk = sock->sk;
192 unsigned char old_state;
198 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
201 old_state = sk->sk_state;
202 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
205 /* Really, if the socket is already in listen state
206 * we can only allow the backlog to be adjusted.
208 if (old_state != TCP_LISTEN) {
209 err = inet_csk_listen_start(sk, backlog);
213 sk->sk_max_ack_backlog = backlog;
221 u32 inet_ehash_secret;
222 EXPORT_SYMBOL(inet_ehash_secret);
224 void build_ehash_secret(void)
226 while (!inet_ehash_secret)
227 get_random_bytes(&inet_ehash_secret, 4);
229 EXPORT_SYMBOL(build_ehash_secret);
232 * Create an inet socket.
235 static int inet_create(struct socket *sock, int protocol)
239 struct inet_protosw *answer;
240 struct inet_sock *inet;
241 struct proto *answer_prot;
242 unsigned char answer_flags;
243 char answer_no_check;
244 int try_loading_module = 0;
247 if (sock->type != SOCK_RAW &&
248 sock->type != SOCK_DGRAM &&
250 build_ehash_secret();
252 sock->state = SS_UNCONNECTED;
254 /* Look for the requested type/protocol pair. */
257 err = -ESOCKTNOSUPPORT;
259 list_for_each_rcu(p, &inetsw[sock->type]) {
260 answer = list_entry(p, struct inet_protosw, list);
262 /* Check the non-wild match. */
263 if (protocol == answer->protocol) {
264 if (protocol != IPPROTO_IP)
267 /* Check for the two wild cases. */
268 if (IPPROTO_IP == protocol) {
269 protocol = answer->protocol;
272 if (IPPROTO_IP == answer->protocol)
275 err = -EPROTONOSUPPORT;
279 if (unlikely(answer == NULL)) {
280 if (try_loading_module < 2) {
283 * Be more specific, e.g. net-pf-2-proto-132-type-1
284 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
286 if (++try_loading_module == 1)
287 request_module("net-pf-%d-proto-%d-type-%d",
288 PF_INET, protocol, sock->type);
290 * Fall back to generic, e.g. net-pf-2-proto-132
291 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
294 request_module("net-pf-%d-proto-%d",
296 goto lookup_protocol;
302 if (answer->capability > 0 && !capable(answer->capability))
305 sock->ops = answer->ops;
306 answer_prot = answer->prot;
307 answer_no_check = answer->no_check;
308 answer_flags = answer->flags;
311 BUG_TRAP(answer_prot->slab != NULL);
314 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
319 sk->sk_no_check = answer_no_check;
320 if (INET_PROTOSW_REUSE & answer_flags)
324 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
326 if (SOCK_RAW == sock->type) {
327 inet->num = protocol;
328 if (IPPROTO_RAW == protocol)
332 if (ipv4_config.no_pmtu_disc)
333 inet->pmtudisc = IP_PMTUDISC_DONT;
335 inet->pmtudisc = IP_PMTUDISC_WANT;
339 sock_init_data(sock, sk);
341 sk->sk_destruct = inet_sock_destruct;
342 sk->sk_family = PF_INET;
343 sk->sk_protocol = protocol;
344 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
350 inet->mc_list = NULL;
352 sk_refcnt_debug_inc(sk);
355 /* It assumes that any protocol which allows
356 * the user to assign a number at socket
357 * creation time automatically
360 inet->sport = htons(inet->num);
361 /* Add to protocol hash chains. */
362 sk->sk_prot->hash(sk);
365 if (sk->sk_prot->init) {
366 err = sk->sk_prot->init(sk);
368 sk_common_release(sk);
379 * The peer socket should always be NULL (or else). When we call this
380 * function we are destroying the object and from then on nobody
381 * should refer to it.
383 int inet_release(struct socket *sock)
385 struct sock *sk = sock->sk;
390 /* Applications forget to leave groups before exiting */
391 ip_mc_drop_socket(sk);
393 /* If linger is set, we don't return until the close
394 * is complete. Otherwise we return immediately. The
395 * actually closing is done the same either way.
397 * If the close is due to the process exiting, we never
401 if (sock_flag(sk, SOCK_LINGER) &&
402 !(current->flags & PF_EXITING))
403 timeout = sk->sk_lingertime;
405 sk->sk_prot->close(sk, timeout);
410 /* It is off by default, see below. */
411 int sysctl_ip_nonlocal_bind __read_mostly;
413 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
415 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
416 struct sock *sk = sock->sk;
417 struct inet_sock *inet = inet_sk(sk);
422 /* If the socket has its own bind function then use it. (RAW) */
423 if (sk->sk_prot->bind) {
424 err = sk->sk_prot->bind(sk, uaddr, addr_len);
428 if (addr_len < sizeof(struct sockaddr_in))
431 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
433 /* Not specified by any standard per-se, however it breaks too
434 * many applications when removed. It is unfortunate since
435 * allowing applications to make a non-local bind solves
436 * several problems with systems using dynamic addressing.
437 * (ie. your servers still start up even if your ISDN link
438 * is temporarily down)
440 err = -EADDRNOTAVAIL;
441 if (!sysctl_ip_nonlocal_bind &&
443 addr->sin_addr.s_addr != INADDR_ANY &&
444 chk_addr_ret != RTN_LOCAL &&
445 chk_addr_ret != RTN_MULTICAST &&
446 chk_addr_ret != RTN_BROADCAST)
449 snum = ntohs(addr->sin_port);
451 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
454 /* We keep a pair of addresses. rcv_saddr is the one
455 * used by hash lookups, and saddr is used for transmit.
457 * In the BSD API these are the same except where it
458 * would be illegal to use them (multicast/broadcast) in
459 * which case the sending device address is used.
463 /* Check these errors (active socket, double bind). */
465 if (sk->sk_state != TCP_CLOSE || inet->num)
466 goto out_release_sock;
468 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
469 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
470 inet->saddr = 0; /* Use device */
472 /* Make sure we are allowed to bind here. */
473 if (sk->sk_prot->get_port(sk, snum)) {
474 inet->saddr = inet->rcv_saddr = 0;
476 goto out_release_sock;
480 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
482 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
483 inet->sport = htons(inet->num);
494 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
495 int addr_len, int flags)
497 struct sock *sk = sock->sk;
499 if (uaddr->sa_family == AF_UNSPEC)
500 return sk->sk_prot->disconnect(sk, flags);
502 if (!inet_sk(sk)->num && inet_autobind(sk))
504 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
507 static long inet_wait_for_connect(struct sock *sk, long timeo)
511 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
513 /* Basic assumption: if someone sets sk->sk_err, he _must_
514 * change state of the socket from TCP_SYN_*.
515 * Connect() does not allow to get error notifications
516 * without closing the socket.
518 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
520 timeo = schedule_timeout(timeo);
522 if (signal_pending(current) || !timeo)
524 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
526 finish_wait(sk->sk_sleep, &wait);
531 * Connect to a remote host. There is regrettably still a little
532 * TCP 'magic' in here.
534 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
535 int addr_len, int flags)
537 struct sock *sk = sock->sk;
543 if (uaddr->sa_family == AF_UNSPEC) {
544 err = sk->sk_prot->disconnect(sk, flags);
545 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
549 switch (sock->state) {
558 /* Fall out of switch with err, set for this state */
562 if (sk->sk_state != TCP_CLOSE)
565 err = sk->sk_prot->connect(sk, uaddr, addr_len);
569 sock->state = SS_CONNECTING;
571 /* Just entered SS_CONNECTING state; the only
572 * difference is that return value in non-blocking
573 * case is EINPROGRESS, rather than EALREADY.
579 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
581 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
582 /* Error code is set above */
583 if (!timeo || !inet_wait_for_connect(sk, timeo))
586 err = sock_intr_errno(timeo);
587 if (signal_pending(current))
591 /* Connection was closed by RST, timeout, ICMP error
592 * or another process disconnected us.
594 if (sk->sk_state == TCP_CLOSE)
597 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
598 * and error was received after socket entered established state.
599 * Hence, it is handled normally after connect() return successfully.
602 sock->state = SS_CONNECTED;
609 err = sock_error(sk) ? : -ECONNABORTED;
610 sock->state = SS_UNCONNECTED;
611 if (sk->sk_prot->disconnect(sk, flags))
612 sock->state = SS_DISCONNECTING;
617 * Accept a pending connection. The TCP layer now gives BSD semantics.
620 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
622 struct sock *sk1 = sock->sk;
624 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
631 BUG_TRAP((1 << sk2->sk_state) &
632 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
634 sock_graft(sk2, newsock);
636 newsock->state = SS_CONNECTED;
645 * This does both peername and sockname.
647 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
648 int *uaddr_len, int peer)
650 struct sock *sk = sock->sk;
651 struct inet_sock *inet = inet_sk(sk);
652 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
654 sin->sin_family = AF_INET;
657 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
660 sin->sin_port = inet->dport;
661 sin->sin_addr.s_addr = inet->daddr;
663 __be32 addr = inet->rcv_saddr;
666 sin->sin_port = inet->sport;
667 sin->sin_addr.s_addr = addr;
669 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
670 *uaddr_len = sizeof(*sin);
674 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
677 struct sock *sk = sock->sk;
679 /* We may need to bind the socket. */
680 if (!inet_sk(sk)->num && inet_autobind(sk))
683 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
687 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
689 struct sock *sk = sock->sk;
691 /* We may need to bind the socket. */
692 if (!inet_sk(sk)->num && inet_autobind(sk))
695 if (sk->sk_prot->sendpage)
696 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
697 return sock_no_sendpage(sock, page, offset, size, flags);
701 int inet_shutdown(struct socket *sock, int how)
703 struct sock *sk = sock->sk;
706 /* This should really check to make sure
707 * the socket is a TCP socket. (WHY AC...)
709 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
712 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
716 if (sock->state == SS_CONNECTING) {
717 if ((1 << sk->sk_state) &
718 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
719 sock->state = SS_DISCONNECTING;
721 sock->state = SS_CONNECTED;
724 switch (sk->sk_state) {
727 /* Hack to wake up other listeners, who can poll for
728 POLLHUP, even on eg. unconnected UDP sockets -- RR */
730 sk->sk_shutdown |= how;
731 if (sk->sk_prot->shutdown)
732 sk->sk_prot->shutdown(sk, how);
735 /* Remaining two branches are temporary solution for missing
736 * close() in multithreaded environment. It is _not_ a good idea,
737 * but we have no choice until close() is repaired at VFS level.
740 if (!(how & RCV_SHUTDOWN))
744 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
745 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
749 /* Wake up anyone sleeping in poll. */
750 sk->sk_state_change(sk);
756 * ioctl() calls you can issue on an INET socket. Most of these are
757 * device configuration and stuff and very rarely used. Some ioctls
758 * pass on to the socket itself.
760 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
761 * loads the devconfigure module does its configuring and unloads it.
762 * There's a good 20K of config code hanging around the kernel.
765 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
767 struct sock *sk = sock->sk;
772 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
775 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
780 err = ip_rt_ioctl(cmd, (void __user *)arg);
785 err = arp_ioctl(cmd, (void __user *)arg);
798 err = devinet_ioctl(cmd, (void __user *)arg);
801 if (sk->sk_prot->ioctl)
802 err = sk->sk_prot->ioctl(sk, cmd, arg);
810 const struct proto_ops inet_stream_ops = {
812 .owner = THIS_MODULE,
813 .release = inet_release,
815 .connect = inet_stream_connect,
816 .socketpair = sock_no_socketpair,
817 .accept = inet_accept,
818 .getname = inet_getname,
821 .listen = inet_listen,
822 .shutdown = inet_shutdown,
823 .setsockopt = sock_common_setsockopt,
824 .getsockopt = sock_common_getsockopt,
825 .sendmsg = inet_sendmsg,
826 .recvmsg = sock_common_recvmsg,
827 .mmap = sock_no_mmap,
828 .sendpage = tcp_sendpage,
830 .compat_setsockopt = compat_sock_common_setsockopt,
831 .compat_getsockopt = compat_sock_common_getsockopt,
835 const struct proto_ops inet_dgram_ops = {
837 .owner = THIS_MODULE,
838 .release = inet_release,
840 .connect = inet_dgram_connect,
841 .socketpair = sock_no_socketpair,
842 .accept = sock_no_accept,
843 .getname = inet_getname,
846 .listen = sock_no_listen,
847 .shutdown = inet_shutdown,
848 .setsockopt = sock_common_setsockopt,
849 .getsockopt = sock_common_getsockopt,
850 .sendmsg = inet_sendmsg,
851 .recvmsg = sock_common_recvmsg,
852 .mmap = sock_no_mmap,
853 .sendpage = inet_sendpage,
855 .compat_setsockopt = compat_sock_common_setsockopt,
856 .compat_getsockopt = compat_sock_common_getsockopt,
861 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
864 static const struct proto_ops inet_sockraw_ops = {
866 .owner = THIS_MODULE,
867 .release = inet_release,
869 .connect = inet_dgram_connect,
870 .socketpair = sock_no_socketpair,
871 .accept = sock_no_accept,
872 .getname = inet_getname,
873 .poll = datagram_poll,
875 .listen = sock_no_listen,
876 .shutdown = inet_shutdown,
877 .setsockopt = sock_common_setsockopt,
878 .getsockopt = sock_common_getsockopt,
879 .sendmsg = inet_sendmsg,
880 .recvmsg = sock_common_recvmsg,
881 .mmap = sock_no_mmap,
882 .sendpage = inet_sendpage,
884 .compat_setsockopt = compat_sock_common_setsockopt,
885 .compat_getsockopt = compat_sock_common_getsockopt,
889 static struct net_proto_family inet_family_ops = {
891 .create = inet_create,
892 .owner = THIS_MODULE,
895 /* Upon startup we insert all the elements in inetsw_array[] into
896 * the linked list inetsw.
898 static struct inet_protosw inetsw_array[] =
902 .protocol = IPPROTO_TCP,
904 .ops = &inet_stream_ops,
907 .flags = INET_PROTOSW_PERMANENT |
913 .protocol = IPPROTO_UDP,
915 .ops = &inet_dgram_ops,
917 .no_check = UDP_CSUM_DEFAULT,
918 .flags = INET_PROTOSW_PERMANENT,
924 .protocol = IPPROTO_IP, /* wild card */
926 .ops = &inet_sockraw_ops,
927 .capability = CAP_NET_RAW,
928 .no_check = UDP_CSUM_DEFAULT,
929 .flags = INET_PROTOSW_REUSE,
933 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
935 void inet_register_protosw(struct inet_protosw *p)
937 struct list_head *lh;
938 struct inet_protosw *answer;
939 int protocol = p->protocol;
940 struct list_head *last_perm;
942 spin_lock_bh(&inetsw_lock);
944 if (p->type >= SOCK_MAX)
947 /* If we are trying to override a permanent protocol, bail. */
949 last_perm = &inetsw[p->type];
950 list_for_each(lh, &inetsw[p->type]) {
951 answer = list_entry(lh, struct inet_protosw, list);
953 /* Check only the non-wild match. */
954 if (INET_PROTOSW_PERMANENT & answer->flags) {
955 if (protocol == answer->protocol)
965 /* Add the new entry after the last permanent entry if any, so that
966 * the new entry does not override a permanent entry when matched with
967 * a wild-card protocol. But it is allowed to override any existing
968 * non-permanent entry. This means that when we remove this entry, the
969 * system automatically returns to the old behavior.
971 list_add_rcu(&p->list, last_perm);
973 spin_unlock_bh(&inetsw_lock);
980 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
986 "Ignoring attempt to register invalid socket type %d.\n",
991 void inet_unregister_protosw(struct inet_protosw *p)
993 if (INET_PROTOSW_PERMANENT & p->flags) {
995 "Attempt to unregister permanent protocol %d.\n",
998 spin_lock_bh(&inetsw_lock);
999 list_del_rcu(&p->list);
1000 spin_unlock_bh(&inetsw_lock);
1007 * Shall we try to damage output packets if routing dev changes?
1010 int sysctl_ip_dynaddr __read_mostly;
1012 static int inet_sk_reselect_saddr(struct sock *sk)
1014 struct inet_sock *inet = inet_sk(sk);
1017 __be32 old_saddr = inet->saddr;
1019 __be32 daddr = inet->daddr;
1021 if (inet->opt && inet->opt->srr)
1022 daddr = inet->opt->faddr;
1024 /* Query new route. */
1025 err = ip_route_connect(&rt, daddr, 0,
1027 sk->sk_bound_dev_if,
1029 inet->sport, inet->dport, sk, 0);
1033 sk_setup_caps(sk, &rt->u.dst);
1035 new_saddr = rt->rt_src;
1037 if (new_saddr == old_saddr)
1040 if (sysctl_ip_dynaddr > 1) {
1041 printk(KERN_INFO "%s(): shifting inet->"
1042 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1045 NIPQUAD(new_saddr));
1048 inet->saddr = inet->rcv_saddr = new_saddr;
1051 * XXX The only one ugly spot where we need to
1052 * XXX really change the sockets identity after
1053 * XXX it has entered the hashes. -DaveM
1055 * Besides that, it does not check for connection
1056 * uniqueness. Wait for troubles.
1058 __sk_prot_rehash(sk);
1062 int inet_sk_rebuild_header(struct sock *sk)
1064 struct inet_sock *inet = inet_sk(sk);
1065 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1069 /* Route is OK, nothing to do. */
1074 daddr = inet->daddr;
1075 if (inet->opt && inet->opt->srr)
1076 daddr = inet->opt->faddr;
1079 .oif = sk->sk_bound_dev_if,
1083 .saddr = inet->saddr,
1084 .tos = RT_CONN_FLAGS(sk),
1087 .proto = sk->sk_protocol,
1090 .sport = inet->sport,
1091 .dport = inet->dport,
1096 security_sk_classify_flow(sk, &fl);
1097 err = ip_route_output_flow(&rt, &fl, sk, 0);
1100 sk_setup_caps(sk, &rt->u.dst);
1102 /* Routing failed... */
1103 sk->sk_route_caps = 0;
1105 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1106 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1108 if (!sysctl_ip_dynaddr ||
1109 sk->sk_state != TCP_SYN_SENT ||
1110 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1111 (err = inet_sk_reselect_saddr(sk)) != 0)
1112 sk->sk_err_soft = -err;
1118 EXPORT_SYMBOL(inet_sk_rebuild_header);
1120 static int inet_gso_send_check(struct sk_buff *skb)
1123 struct net_protocol *ops;
1128 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1133 if (ihl < sizeof(*iph))
1136 if (unlikely(!pskb_may_pull(skb, ihl)))
1139 __skb_pull(skb, ihl);
1140 skb_reset_transport_header(skb);
1142 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1143 err = -EPROTONOSUPPORT;
1146 ops = rcu_dereference(inet_protos[proto]);
1147 if (likely(ops && ops->gso_send_check))
1148 err = ops->gso_send_check(skb);
1155 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1157 struct sk_buff *segs = ERR_PTR(-EINVAL);
1159 struct net_protocol *ops;
1164 if (unlikely(skb_shinfo(skb)->gso_type &
1172 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1177 if (ihl < sizeof(*iph))
1180 if (unlikely(!pskb_may_pull(skb, ihl)))
1183 __skb_pull(skb, ihl);
1184 skb_reset_transport_header(skb);
1186 id = ntohs(iph->id);
1187 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1188 segs = ERR_PTR(-EPROTONOSUPPORT);
1191 ops = rcu_dereference(inet_protos[proto]);
1192 if (likely(ops && ops->gso_segment))
1193 segs = ops->gso_segment(skb, features);
1196 if (!segs || unlikely(IS_ERR(segs)))
1202 iph->id = htons(id++);
1203 iph->tot_len = htons(skb->len - skb->mac_len);
1205 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1206 } while ((skb = skb->next));
1212 #ifdef CONFIG_IP_MULTICAST
1213 static struct net_protocol igmp_protocol = {
1214 .handler = igmp_rcv,
1218 static struct net_protocol tcp_protocol = {
1219 .handler = tcp_v4_rcv,
1220 .err_handler = tcp_v4_err,
1221 .gso_send_check = tcp_v4_gso_send_check,
1222 .gso_segment = tcp_tso_segment,
1226 static struct net_protocol udp_protocol = {
1228 .err_handler = udp_err,
1232 static struct net_protocol icmp_protocol = {
1233 .handler = icmp_rcv,
1236 static int __init init_ipv4_mibs(void)
1238 net_statistics[0] = alloc_percpu(struct linux_mib);
1239 net_statistics[1] = alloc_percpu(struct linux_mib);
1240 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1241 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1242 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1243 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1244 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1245 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1246 udp_statistics[0] = alloc_percpu(struct udp_mib);
1247 udp_statistics[1] = alloc_percpu(struct udp_mib);
1248 udplite_statistics[0] = alloc_percpu(struct udp_mib);
1249 udplite_statistics[1] = alloc_percpu(struct udp_mib);
1251 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1252 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1253 && udp_statistics[0] && udp_statistics[1]
1254 && udplite_statistics[0] && udplite_statistics[1] ) )
1257 (void) tcp_mib_init();
1262 static int ipv4_proc_init(void);
1265 * IP protocol layer initialiser
1268 static struct packet_type ip_packet_type = {
1269 .type = __constant_htons(ETH_P_IP),
1271 .gso_send_check = inet_gso_send_check,
1272 .gso_segment = inet_gso_segment,
1275 static int __init inet_init(void)
1277 struct sk_buff *dummy_skb;
1278 struct inet_protosw *q;
1279 struct list_head *r;
1282 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1284 rc = proto_register(&tcp_prot, 1);
1288 rc = proto_register(&udp_prot, 1);
1290 goto out_unregister_tcp_proto;
1292 rc = proto_register(&raw_prot, 1);
1294 goto out_unregister_udp_proto;
1297 * Tell SOCKET that we are alive...
1300 (void)sock_register(&inet_family_ops);
1303 * Add all the base protocols.
1306 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1307 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1308 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1309 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1310 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1311 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1312 #ifdef CONFIG_IP_MULTICAST
1313 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1314 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1317 /* Register the socket-side information for inet_create. */
1318 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1321 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1322 inet_register_protosw(q);
1325 * Set the ARP module up
1331 * Set the IP module up
1336 tcp_v4_init(&inet_family_ops);
1338 /* Setup TCP slab cache for open requests. */
1341 /* Add UDP-Lite (RFC 3828) */
1342 udplite4_register();
1345 * Set the ICMP layer up
1348 icmp_init(&inet_family_ops);
1351 * Initialise the multicast router
1353 #if defined(CONFIG_IP_MROUTE)
1357 * Initialise per-cpu ipv4 mibs
1360 if (init_ipv4_mibs())
1361 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1367 dev_add_pack(&ip_packet_type);
1372 out_unregister_udp_proto:
1373 proto_unregister(&udp_prot);
1374 out_unregister_tcp_proto:
1375 proto_unregister(&tcp_prot);
1379 fs_initcall(inet_init);
1381 /* ------------------------------------------------------------------------ */
1383 #ifdef CONFIG_PROC_FS
1384 static int __init ipv4_proc_init(void)
1388 if (raw_proc_init())
1390 if (tcp4_proc_init())
1392 if (udp4_proc_init())
1394 if (fib_proc_init())
1396 if (ip_misc_proc_init())
1413 #else /* CONFIG_PROC_FS */
1414 static int __init ipv4_proc_init(void)
1418 #endif /* CONFIG_PROC_FS */
1420 MODULE_ALIAS_NETPROTO(PF_INET);
1422 EXPORT_SYMBOL(inet_accept);
1423 EXPORT_SYMBOL(inet_bind);
1424 EXPORT_SYMBOL(inet_dgram_connect);
1425 EXPORT_SYMBOL(inet_dgram_ops);
1426 EXPORT_SYMBOL(inet_getname);
1427 EXPORT_SYMBOL(inet_ioctl);
1428 EXPORT_SYMBOL(inet_listen);
1429 EXPORT_SYMBOL(inet_register_protosw);
1430 EXPORT_SYMBOL(inet_release);
1431 EXPORT_SYMBOL(inet_sendmsg);
1432 EXPORT_SYMBOL(inet_shutdown);
1433 EXPORT_SYMBOL(inet_sock_destruct);
1434 EXPORT_SYMBOL(inet_stream_connect);
1435 EXPORT_SYMBOL(inet_stream_ops);
1436 EXPORT_SYMBOL(inet_unregister_protosw);
1437 EXPORT_SYMBOL(net_statistics);
1438 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);