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/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
100 #include <net/protocol.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
107 #include <net/udplite.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
111 #include <net/icmp.h>
112 #include <net/ipip.h>
113 #include <net/inet_common.h>
114 #include <net/xfrm.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
119 DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
121 extern void ip_mc_drop_socket(struct sock *sk);
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
129 struct ipv4_config ipv4_config;
131 EXPORT_SYMBOL(ipv4_config);
133 /* New destruction routine */
135 void inet_sock_destruct(struct sock *sk)
137 struct inet_sock *inet = inet_sk(sk);
139 __skb_queue_purge(&sk->sk_receive_queue);
140 __skb_queue_purge(&sk->sk_error_queue);
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 printk("Attempt to release TCP socket in state %d %p\n",
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 printk("Attempt to release alive inet socket %p\n", sk);
152 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
153 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
154 BUG_TRAP(!sk->sk_wmem_queued);
155 BUG_TRAP(!sk->sk_forward_alloc);
158 dst_release(sk->sk_dst_cache);
159 sk_refcnt_debug_dec(sk);
163 * The routines beyond this point handle the behaviour of an AF_INET
164 * socket object. Mostly it punts to the subprotocols of IP to do
169 * Automatically bind an unbound socket.
172 static int inet_autobind(struct sock *sk)
174 struct inet_sock *inet;
175 /* We may need to bind the socket. */
179 if (sk->sk_prot->get_port(sk, 0)) {
183 inet->sport = htons(inet->num);
190 * Move a socket into listening state.
192 int inet_listen(struct socket *sock, int backlog)
194 struct sock *sk = sock->sk;
195 unsigned char old_state;
201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 old_state = sk->sk_state;
205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 /* Really, if the socket is already in listen state
209 * we can only allow the backlog to be adjusted.
211 if (old_state != TCP_LISTEN) {
212 err = inet_csk_listen_start(sk, backlog);
216 sk->sk_max_ack_backlog = backlog;
224 u32 inet_ehash_secret __read_mostly;
225 EXPORT_SYMBOL(inet_ehash_secret);
228 * inet_ehash_secret must be set exactly once
229 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
231 void build_ehash_secret(void)
235 get_random_bytes(&rnd, sizeof(rnd));
237 spin_lock_bh(&inetsw_lock);
238 if (!inet_ehash_secret)
239 inet_ehash_secret = rnd;
240 spin_unlock_bh(&inetsw_lock);
242 EXPORT_SYMBOL(build_ehash_secret);
245 * Create an inet socket.
248 static int inet_create(struct net *net, struct socket *sock, int protocol)
252 struct inet_protosw *answer;
253 struct inet_sock *inet;
254 struct proto *answer_prot;
255 unsigned char answer_flags;
256 char answer_no_check;
257 int try_loading_module = 0;
260 if (net != &init_net)
261 return -EAFNOSUPPORT;
263 if (sock->type != SOCK_RAW &&
264 sock->type != SOCK_DGRAM &&
266 build_ehash_secret();
268 sock->state = SS_UNCONNECTED;
270 /* Look for the requested type/protocol pair. */
273 err = -ESOCKTNOSUPPORT;
275 list_for_each_rcu(p, &inetsw[sock->type]) {
276 answer = list_entry(p, struct inet_protosw, list);
278 /* Check the non-wild match. */
279 if (protocol == answer->protocol) {
280 if (protocol != IPPROTO_IP)
283 /* Check for the two wild cases. */
284 if (IPPROTO_IP == protocol) {
285 protocol = answer->protocol;
288 if (IPPROTO_IP == answer->protocol)
291 err = -EPROTONOSUPPORT;
295 if (unlikely(answer == NULL)) {
296 if (try_loading_module < 2) {
299 * Be more specific, e.g. net-pf-2-proto-132-type-1
300 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
302 if (++try_loading_module == 1)
303 request_module("net-pf-%d-proto-%d-type-%d",
304 PF_INET, protocol, sock->type);
306 * Fall back to generic, e.g. net-pf-2-proto-132
307 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
310 request_module("net-pf-%d-proto-%d",
312 goto lookup_protocol;
318 if (answer->capability > 0 && !capable(answer->capability))
321 sock->ops = answer->ops;
322 answer_prot = answer->prot;
323 answer_no_check = answer->no_check;
324 answer_flags = answer->flags;
327 BUG_TRAP(answer_prot->slab != NULL);
330 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
335 sk->sk_no_check = answer_no_check;
336 if (INET_PROTOSW_REUSE & answer_flags)
340 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
342 if (SOCK_RAW == sock->type) {
343 inet->num = protocol;
344 if (IPPROTO_RAW == protocol)
348 if (ipv4_config.no_pmtu_disc)
349 inet->pmtudisc = IP_PMTUDISC_DONT;
351 inet->pmtudisc = IP_PMTUDISC_WANT;
355 sock_init_data(sock, sk);
357 sk->sk_destruct = inet_sock_destruct;
358 sk->sk_family = PF_INET;
359 sk->sk_protocol = protocol;
360 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
366 inet->mc_list = NULL;
368 sk_refcnt_debug_inc(sk);
371 /* It assumes that any protocol which allows
372 * the user to assign a number at socket
373 * creation time automatically
376 inet->sport = htons(inet->num);
377 /* Add to protocol hash chains. */
378 sk->sk_prot->hash(sk);
381 if (sk->sk_prot->init) {
382 err = sk->sk_prot->init(sk);
384 sk_common_release(sk);
395 * The peer socket should always be NULL (or else). When we call this
396 * function we are destroying the object and from then on nobody
397 * should refer to it.
399 int inet_release(struct socket *sock)
401 struct sock *sk = sock->sk;
406 /* Applications forget to leave groups before exiting */
407 ip_mc_drop_socket(sk);
409 /* If linger is set, we don't return until the close
410 * is complete. Otherwise we return immediately. The
411 * actually closing is done the same either way.
413 * If the close is due to the process exiting, we never
417 if (sock_flag(sk, SOCK_LINGER) &&
418 !(current->flags & PF_EXITING))
419 timeout = sk->sk_lingertime;
421 sk->sk_prot->close(sk, timeout);
426 /* It is off by default, see below. */
427 int sysctl_ip_nonlocal_bind __read_mostly;
429 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
431 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
432 struct sock *sk = sock->sk;
433 struct inet_sock *inet = inet_sk(sk);
438 /* If the socket has its own bind function then use it. (RAW) */
439 if (sk->sk_prot->bind) {
440 err = sk->sk_prot->bind(sk, uaddr, addr_len);
444 if (addr_len < sizeof(struct sockaddr_in))
447 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
449 /* Not specified by any standard per-se, however it breaks too
450 * many applications when removed. It is unfortunate since
451 * allowing applications to make a non-local bind solves
452 * several problems with systems using dynamic addressing.
453 * (ie. your servers still start up even if your ISDN link
454 * is temporarily down)
456 err = -EADDRNOTAVAIL;
457 if (!sysctl_ip_nonlocal_bind &&
459 addr->sin_addr.s_addr != INADDR_ANY &&
460 chk_addr_ret != RTN_LOCAL &&
461 chk_addr_ret != RTN_MULTICAST &&
462 chk_addr_ret != RTN_BROADCAST)
465 snum = ntohs(addr->sin_port);
467 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
470 /* We keep a pair of addresses. rcv_saddr is the one
471 * used by hash lookups, and saddr is used for transmit.
473 * In the BSD API these are the same except where it
474 * would be illegal to use them (multicast/broadcast) in
475 * which case the sending device address is used.
479 /* Check these errors (active socket, double bind). */
481 if (sk->sk_state != TCP_CLOSE || inet->num)
482 goto out_release_sock;
484 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
485 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
486 inet->saddr = 0; /* Use device */
488 /* Make sure we are allowed to bind here. */
489 if (sk->sk_prot->get_port(sk, snum)) {
490 inet->saddr = inet->rcv_saddr = 0;
492 goto out_release_sock;
496 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
498 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
499 inet->sport = htons(inet->num);
510 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
511 int addr_len, int flags)
513 struct sock *sk = sock->sk;
515 if (uaddr->sa_family == AF_UNSPEC)
516 return sk->sk_prot->disconnect(sk, flags);
518 if (!inet_sk(sk)->num && inet_autobind(sk))
520 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
523 static long inet_wait_for_connect(struct sock *sk, long timeo)
527 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
529 /* Basic assumption: if someone sets sk->sk_err, he _must_
530 * change state of the socket from TCP_SYN_*.
531 * Connect() does not allow to get error notifications
532 * without closing the socket.
534 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
536 timeo = schedule_timeout(timeo);
538 if (signal_pending(current) || !timeo)
540 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
542 finish_wait(sk->sk_sleep, &wait);
547 * Connect to a remote host. There is regrettably still a little
548 * TCP 'magic' in here.
550 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
551 int addr_len, int flags)
553 struct sock *sk = sock->sk;
559 if (uaddr->sa_family == AF_UNSPEC) {
560 err = sk->sk_prot->disconnect(sk, flags);
561 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
565 switch (sock->state) {
574 /* Fall out of switch with err, set for this state */
578 if (sk->sk_state != TCP_CLOSE)
581 err = sk->sk_prot->connect(sk, uaddr, addr_len);
585 sock->state = SS_CONNECTING;
587 /* Just entered SS_CONNECTING state; the only
588 * difference is that return value in non-blocking
589 * case is EINPROGRESS, rather than EALREADY.
595 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
597 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
598 /* Error code is set above */
599 if (!timeo || !inet_wait_for_connect(sk, timeo))
602 err = sock_intr_errno(timeo);
603 if (signal_pending(current))
607 /* Connection was closed by RST, timeout, ICMP error
608 * or another process disconnected us.
610 if (sk->sk_state == TCP_CLOSE)
613 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
614 * and error was received after socket entered established state.
615 * Hence, it is handled normally after connect() return successfully.
618 sock->state = SS_CONNECTED;
625 err = sock_error(sk) ? : -ECONNABORTED;
626 sock->state = SS_UNCONNECTED;
627 if (sk->sk_prot->disconnect(sk, flags))
628 sock->state = SS_DISCONNECTING;
633 * Accept a pending connection. The TCP layer now gives BSD semantics.
636 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
638 struct sock *sk1 = sock->sk;
640 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
647 BUG_TRAP((1 << sk2->sk_state) &
648 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
650 sock_graft(sk2, newsock);
652 newsock->state = SS_CONNECTED;
661 * This does both peername and sockname.
663 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
664 int *uaddr_len, int peer)
666 struct sock *sk = sock->sk;
667 struct inet_sock *inet = inet_sk(sk);
668 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
670 sin->sin_family = AF_INET;
673 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
676 sin->sin_port = inet->dport;
677 sin->sin_addr.s_addr = inet->daddr;
679 __be32 addr = inet->rcv_saddr;
682 sin->sin_port = inet->sport;
683 sin->sin_addr.s_addr = addr;
685 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
686 *uaddr_len = sizeof(*sin);
690 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
693 struct sock *sk = sock->sk;
695 /* We may need to bind the socket. */
696 if (!inet_sk(sk)->num && inet_autobind(sk))
699 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
703 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
705 struct sock *sk = sock->sk;
707 /* We may need to bind the socket. */
708 if (!inet_sk(sk)->num && inet_autobind(sk))
711 if (sk->sk_prot->sendpage)
712 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
713 return sock_no_sendpage(sock, page, offset, size, flags);
717 int inet_shutdown(struct socket *sock, int how)
719 struct sock *sk = sock->sk;
722 /* This should really check to make sure
723 * the socket is a TCP socket. (WHY AC...)
725 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
728 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
732 if (sock->state == SS_CONNECTING) {
733 if ((1 << sk->sk_state) &
734 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
735 sock->state = SS_DISCONNECTING;
737 sock->state = SS_CONNECTED;
740 switch (sk->sk_state) {
743 /* Hack to wake up other listeners, who can poll for
744 POLLHUP, even on eg. unconnected UDP sockets -- RR */
746 sk->sk_shutdown |= how;
747 if (sk->sk_prot->shutdown)
748 sk->sk_prot->shutdown(sk, how);
751 /* Remaining two branches are temporary solution for missing
752 * close() in multithreaded environment. It is _not_ a good idea,
753 * but we have no choice until close() is repaired at VFS level.
756 if (!(how & RCV_SHUTDOWN))
760 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
761 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
765 /* Wake up anyone sleeping in poll. */
766 sk->sk_state_change(sk);
772 * ioctl() calls you can issue on an INET socket. Most of these are
773 * device configuration and stuff and very rarely used. Some ioctls
774 * pass on to the socket itself.
776 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
777 * loads the devconfigure module does its configuring and unloads it.
778 * There's a good 20K of config code hanging around the kernel.
781 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
783 struct sock *sk = sock->sk;
788 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
791 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
796 err = ip_rt_ioctl(cmd, (void __user *)arg);
801 err = arp_ioctl(sk->sk_net, cmd, (void __user *)arg);
814 err = devinet_ioctl(cmd, (void __user *)arg);
817 if (sk->sk_prot->ioctl)
818 err = sk->sk_prot->ioctl(sk, cmd, arg);
826 const struct proto_ops inet_stream_ops = {
828 .owner = THIS_MODULE,
829 .release = inet_release,
831 .connect = inet_stream_connect,
832 .socketpair = sock_no_socketpair,
833 .accept = inet_accept,
834 .getname = inet_getname,
837 .listen = inet_listen,
838 .shutdown = inet_shutdown,
839 .setsockopt = sock_common_setsockopt,
840 .getsockopt = sock_common_getsockopt,
841 .sendmsg = tcp_sendmsg,
842 .recvmsg = sock_common_recvmsg,
843 .mmap = sock_no_mmap,
844 .sendpage = tcp_sendpage,
845 .splice_read = tcp_splice_read,
847 .compat_setsockopt = compat_sock_common_setsockopt,
848 .compat_getsockopt = compat_sock_common_getsockopt,
852 const struct proto_ops inet_dgram_ops = {
854 .owner = THIS_MODULE,
855 .release = inet_release,
857 .connect = inet_dgram_connect,
858 .socketpair = sock_no_socketpair,
859 .accept = sock_no_accept,
860 .getname = inet_getname,
863 .listen = sock_no_listen,
864 .shutdown = inet_shutdown,
865 .setsockopt = sock_common_setsockopt,
866 .getsockopt = sock_common_getsockopt,
867 .sendmsg = inet_sendmsg,
868 .recvmsg = sock_common_recvmsg,
869 .mmap = sock_no_mmap,
870 .sendpage = inet_sendpage,
872 .compat_setsockopt = compat_sock_common_setsockopt,
873 .compat_getsockopt = compat_sock_common_getsockopt,
878 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
881 static const struct proto_ops inet_sockraw_ops = {
883 .owner = THIS_MODULE,
884 .release = inet_release,
886 .connect = inet_dgram_connect,
887 .socketpair = sock_no_socketpair,
888 .accept = sock_no_accept,
889 .getname = inet_getname,
890 .poll = datagram_poll,
892 .listen = sock_no_listen,
893 .shutdown = inet_shutdown,
894 .setsockopt = sock_common_setsockopt,
895 .getsockopt = sock_common_getsockopt,
896 .sendmsg = inet_sendmsg,
897 .recvmsg = sock_common_recvmsg,
898 .mmap = sock_no_mmap,
899 .sendpage = inet_sendpage,
901 .compat_setsockopt = compat_sock_common_setsockopt,
902 .compat_getsockopt = compat_sock_common_getsockopt,
906 static struct net_proto_family inet_family_ops = {
908 .create = inet_create,
909 .owner = THIS_MODULE,
912 /* Upon startup we insert all the elements in inetsw_array[] into
913 * the linked list inetsw.
915 static struct inet_protosw inetsw_array[] =
919 .protocol = IPPROTO_TCP,
921 .ops = &inet_stream_ops,
924 .flags = INET_PROTOSW_PERMANENT |
930 .protocol = IPPROTO_UDP,
932 .ops = &inet_dgram_ops,
934 .no_check = UDP_CSUM_DEFAULT,
935 .flags = INET_PROTOSW_PERMANENT,
941 .protocol = IPPROTO_IP, /* wild card */
943 .ops = &inet_sockraw_ops,
944 .capability = CAP_NET_RAW,
945 .no_check = UDP_CSUM_DEFAULT,
946 .flags = INET_PROTOSW_REUSE,
950 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
952 void inet_register_protosw(struct inet_protosw *p)
954 struct list_head *lh;
955 struct inet_protosw *answer;
956 int protocol = p->protocol;
957 struct list_head *last_perm;
959 spin_lock_bh(&inetsw_lock);
961 if (p->type >= SOCK_MAX)
964 /* If we are trying to override a permanent protocol, bail. */
966 last_perm = &inetsw[p->type];
967 list_for_each(lh, &inetsw[p->type]) {
968 answer = list_entry(lh, struct inet_protosw, list);
970 /* Check only the non-wild match. */
971 if (INET_PROTOSW_PERMANENT & answer->flags) {
972 if (protocol == answer->protocol)
982 /* Add the new entry after the last permanent entry if any, so that
983 * the new entry does not override a permanent entry when matched with
984 * a wild-card protocol. But it is allowed to override any existing
985 * non-permanent entry. This means that when we remove this entry, the
986 * system automatically returns to the old behavior.
988 list_add_rcu(&p->list, last_perm);
990 spin_unlock_bh(&inetsw_lock);
997 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1003 "Ignoring attempt to register invalid socket type %d.\n",
1008 void inet_unregister_protosw(struct inet_protosw *p)
1010 if (INET_PROTOSW_PERMANENT & p->flags) {
1012 "Attempt to unregister permanent protocol %d.\n",
1015 spin_lock_bh(&inetsw_lock);
1016 list_del_rcu(&p->list);
1017 spin_unlock_bh(&inetsw_lock);
1024 * Shall we try to damage output packets if routing dev changes?
1027 int sysctl_ip_dynaddr __read_mostly;
1029 static int inet_sk_reselect_saddr(struct sock *sk)
1031 struct inet_sock *inet = inet_sk(sk);
1034 __be32 old_saddr = inet->saddr;
1036 __be32 daddr = inet->daddr;
1038 if (inet->opt && inet->opt->srr)
1039 daddr = inet->opt->faddr;
1041 /* Query new route. */
1042 err = ip_route_connect(&rt, daddr, 0,
1044 sk->sk_bound_dev_if,
1046 inet->sport, inet->dport, sk, 0);
1050 sk_setup_caps(sk, &rt->u.dst);
1052 new_saddr = rt->rt_src;
1054 if (new_saddr == old_saddr)
1057 if (sysctl_ip_dynaddr > 1) {
1058 printk(KERN_INFO "%s(): shifting inet->"
1059 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1062 NIPQUAD(new_saddr));
1065 inet->saddr = inet->rcv_saddr = new_saddr;
1068 * XXX The only one ugly spot where we need to
1069 * XXX really change the sockets identity after
1070 * XXX it has entered the hashes. -DaveM
1072 * Besides that, it does not check for connection
1073 * uniqueness. Wait for troubles.
1075 __sk_prot_rehash(sk);
1079 int inet_sk_rebuild_header(struct sock *sk)
1081 struct inet_sock *inet = inet_sk(sk);
1082 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1086 /* Route is OK, nothing to do. */
1091 daddr = inet->daddr;
1092 if (inet->opt && inet->opt->srr)
1093 daddr = inet->opt->faddr;
1096 .oif = sk->sk_bound_dev_if,
1100 .saddr = inet->saddr,
1101 .tos = RT_CONN_FLAGS(sk),
1104 .proto = sk->sk_protocol,
1107 .sport = inet->sport,
1108 .dport = inet->dport,
1113 security_sk_classify_flow(sk, &fl);
1114 err = ip_route_output_flow(&rt, &fl, sk, 0);
1117 sk_setup_caps(sk, &rt->u.dst);
1119 /* Routing failed... */
1120 sk->sk_route_caps = 0;
1122 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1123 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1125 if (!sysctl_ip_dynaddr ||
1126 sk->sk_state != TCP_SYN_SENT ||
1127 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1128 (err = inet_sk_reselect_saddr(sk)) != 0)
1129 sk->sk_err_soft = -err;
1135 EXPORT_SYMBOL(inet_sk_rebuild_header);
1137 static int inet_gso_send_check(struct sk_buff *skb)
1140 struct net_protocol *ops;
1145 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1150 if (ihl < sizeof(*iph))
1153 if (unlikely(!pskb_may_pull(skb, ihl)))
1156 __skb_pull(skb, ihl);
1157 skb_reset_transport_header(skb);
1159 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1160 err = -EPROTONOSUPPORT;
1163 ops = rcu_dereference(inet_protos[proto]);
1164 if (likely(ops && ops->gso_send_check))
1165 err = ops->gso_send_check(skb);
1172 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1174 struct sk_buff *segs = ERR_PTR(-EINVAL);
1176 struct net_protocol *ops;
1181 if (!(features & NETIF_F_V4_CSUM))
1182 features &= ~NETIF_F_SG;
1184 if (unlikely(skb_shinfo(skb)->gso_type &
1192 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1197 if (ihl < sizeof(*iph))
1200 if (unlikely(!pskb_may_pull(skb, ihl)))
1203 __skb_pull(skb, ihl);
1204 skb_reset_transport_header(skb);
1206 id = ntohs(iph->id);
1207 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1208 segs = ERR_PTR(-EPROTONOSUPPORT);
1211 ops = rcu_dereference(inet_protos[proto]);
1212 if (likely(ops && ops->gso_segment))
1213 segs = ops->gso_segment(skb, features);
1216 if (!segs || unlikely(IS_ERR(segs)))
1222 iph->id = htons(id++);
1223 iph->tot_len = htons(skb->len - skb->mac_len);
1225 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1226 } while ((skb = skb->next));
1232 unsigned long snmp_fold_field(void *mib[], int offt)
1234 unsigned long res = 0;
1237 for_each_possible_cpu(i) {
1238 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1239 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1243 EXPORT_SYMBOL_GPL(snmp_fold_field);
1245 int snmp_mib_init(void *ptr[2], size_t mibsize)
1247 BUG_ON(ptr == NULL);
1248 ptr[0] = __alloc_percpu(mibsize);
1251 ptr[1] = __alloc_percpu(mibsize);
1256 free_percpu(ptr[0]);
1261 EXPORT_SYMBOL_GPL(snmp_mib_init);
1263 void snmp_mib_free(void *ptr[2])
1265 BUG_ON(ptr == NULL);
1266 free_percpu(ptr[0]);
1267 free_percpu(ptr[1]);
1268 ptr[0] = ptr[1] = NULL;
1270 EXPORT_SYMBOL_GPL(snmp_mib_free);
1272 #ifdef CONFIG_IP_MULTICAST
1273 static struct net_protocol igmp_protocol = {
1274 .handler = igmp_rcv,
1278 static struct net_protocol tcp_protocol = {
1279 .handler = tcp_v4_rcv,
1280 .err_handler = tcp_v4_err,
1281 .gso_send_check = tcp_v4_gso_send_check,
1282 .gso_segment = tcp_tso_segment,
1286 static struct net_protocol udp_protocol = {
1288 .err_handler = udp_err,
1292 static struct net_protocol icmp_protocol = {
1293 .handler = icmp_rcv,
1297 static int __init init_ipv4_mibs(void)
1299 if (snmp_mib_init((void **)net_statistics,
1300 sizeof(struct linux_mib)) < 0)
1302 if (snmp_mib_init((void **)ip_statistics,
1303 sizeof(struct ipstats_mib)) < 0)
1305 if (snmp_mib_init((void **)icmp_statistics,
1306 sizeof(struct icmp_mib)) < 0)
1308 if (snmp_mib_init((void **)icmpmsg_statistics,
1309 sizeof(struct icmpmsg_mib)) < 0)
1310 goto err_icmpmsg_mib;
1311 if (snmp_mib_init((void **)tcp_statistics,
1312 sizeof(struct tcp_mib)) < 0)
1314 if (snmp_mib_init((void **)udp_statistics,
1315 sizeof(struct udp_mib)) < 0)
1317 if (snmp_mib_init((void **)udplite_statistics,
1318 sizeof(struct udp_mib)) < 0)
1319 goto err_udplite_mib;
1326 snmp_mib_free((void **)udp_statistics);
1328 snmp_mib_free((void **)tcp_statistics);
1330 snmp_mib_free((void **)icmpmsg_statistics);
1332 snmp_mib_free((void **)icmp_statistics);
1334 snmp_mib_free((void **)ip_statistics);
1336 snmp_mib_free((void **)net_statistics);
1341 static int ipv4_proc_init(void);
1344 * IP protocol layer initialiser
1347 static struct packet_type ip_packet_type = {
1348 .type = __constant_htons(ETH_P_IP),
1350 .gso_send_check = inet_gso_send_check,
1351 .gso_segment = inet_gso_segment,
1354 static int __init inet_init(void)
1356 struct sk_buff *dummy_skb;
1357 struct inet_protosw *q;
1358 struct list_head *r;
1361 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1363 rc = proto_register(&tcp_prot, 1);
1367 rc = proto_register(&udp_prot, 1);
1369 goto out_unregister_tcp_proto;
1371 rc = proto_register(&raw_prot, 1);
1373 goto out_unregister_udp_proto;
1376 * Tell SOCKET that we are alive...
1379 (void)sock_register(&inet_family_ops);
1382 * Add all the base protocols.
1385 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1386 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1387 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1388 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1389 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1390 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1391 #ifdef CONFIG_IP_MULTICAST
1392 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1393 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1396 /* Register the socket-side information for inet_create. */
1397 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1400 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1401 inet_register_protosw(q);
1404 * Set the ARP module up
1410 * Set the IP module up
1415 tcp_v4_init(&inet_family_ops);
1417 /* Setup TCP slab cache for open requests. */
1420 /* Add UDP-Lite (RFC 3828) */
1421 udplite4_register();
1424 * Set the ICMP layer up
1427 icmp_init(&inet_family_ops);
1430 * Initialise the multicast router
1432 #if defined(CONFIG_IP_MROUTE)
1436 * Initialise per-cpu ipv4 mibs
1439 if (init_ipv4_mibs())
1440 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1446 dev_add_pack(&ip_packet_type);
1451 out_unregister_udp_proto:
1452 proto_unregister(&udp_prot);
1453 out_unregister_tcp_proto:
1454 proto_unregister(&tcp_prot);
1458 fs_initcall(inet_init);
1460 /* ------------------------------------------------------------------------ */
1462 #ifdef CONFIG_PROC_FS
1463 static int __init ipv4_proc_init(void)
1467 if (raw_proc_init())
1469 if (tcp4_proc_init())
1471 if (udp4_proc_init())
1473 if (fib_proc_init())
1475 if (ip_misc_proc_init())
1492 #else /* CONFIG_PROC_FS */
1493 static int __init ipv4_proc_init(void)
1497 #endif /* CONFIG_PROC_FS */
1499 MODULE_ALIAS_NETPROTO(PF_INET);
1501 EXPORT_SYMBOL(inet_accept);
1502 EXPORT_SYMBOL(inet_bind);
1503 EXPORT_SYMBOL(inet_dgram_connect);
1504 EXPORT_SYMBOL(inet_dgram_ops);
1505 EXPORT_SYMBOL(inet_getname);
1506 EXPORT_SYMBOL(inet_ioctl);
1507 EXPORT_SYMBOL(inet_listen);
1508 EXPORT_SYMBOL(inet_register_protosw);
1509 EXPORT_SYMBOL(inet_release);
1510 EXPORT_SYMBOL(inet_sendmsg);
1511 EXPORT_SYMBOL(inet_shutdown);
1512 EXPORT_SYMBOL(inet_sock_destruct);
1513 EXPORT_SYMBOL(inet_stream_connect);
1514 EXPORT_SYMBOL(inet_stream_ops);
1515 EXPORT_SYMBOL(inet_unregister_protosw);
1516 EXPORT_SYMBOL(net_statistics);
1517 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);