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 __read_mostly;
222 EXPORT_SYMBOL(inet_ehash_secret);
225 * inet_ehash_secret must be set exactly once
226 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
228 void build_ehash_secret(void)
232 get_random_bytes(&rnd, sizeof(rnd));
234 spin_lock_bh(&inetsw_lock);
235 if (!inet_ehash_secret)
236 inet_ehash_secret = rnd;
237 spin_unlock_bh(&inetsw_lock);
239 EXPORT_SYMBOL(build_ehash_secret);
242 * Create an inet socket.
245 static int inet_create(struct socket *sock, int protocol)
249 struct inet_protosw *answer;
250 struct inet_sock *inet;
251 struct proto *answer_prot;
252 unsigned char answer_flags;
253 char answer_no_check;
254 int try_loading_module = 0;
257 if (sock->type != SOCK_RAW &&
258 sock->type != SOCK_DGRAM &&
260 build_ehash_secret();
262 sock->state = SS_UNCONNECTED;
264 /* Look for the requested type/protocol pair. */
267 err = -ESOCKTNOSUPPORT;
269 list_for_each_rcu(p, &inetsw[sock->type]) {
270 answer = list_entry(p, struct inet_protosw, list);
272 /* Check the non-wild match. */
273 if (protocol == answer->protocol) {
274 if (protocol != IPPROTO_IP)
277 /* Check for the two wild cases. */
278 if (IPPROTO_IP == protocol) {
279 protocol = answer->protocol;
282 if (IPPROTO_IP == answer->protocol)
285 err = -EPROTONOSUPPORT;
289 if (unlikely(answer == NULL)) {
290 if (try_loading_module < 2) {
293 * Be more specific, e.g. net-pf-2-proto-132-type-1
294 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
296 if (++try_loading_module == 1)
297 request_module("net-pf-%d-proto-%d-type-%d",
298 PF_INET, protocol, sock->type);
300 * Fall back to generic, e.g. net-pf-2-proto-132
301 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
304 request_module("net-pf-%d-proto-%d",
306 goto lookup_protocol;
312 if (answer->capability > 0 && !capable(answer->capability))
315 sock->ops = answer->ops;
316 answer_prot = answer->prot;
317 answer_no_check = answer->no_check;
318 answer_flags = answer->flags;
321 BUG_TRAP(answer_prot->slab != NULL);
324 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
329 sk->sk_no_check = answer_no_check;
330 if (INET_PROTOSW_REUSE & answer_flags)
334 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
336 if (SOCK_RAW == sock->type) {
337 inet->num = protocol;
338 if (IPPROTO_RAW == protocol)
342 if (ipv4_config.no_pmtu_disc)
343 inet->pmtudisc = IP_PMTUDISC_DONT;
345 inet->pmtudisc = IP_PMTUDISC_WANT;
349 sock_init_data(sock, sk);
351 sk->sk_destruct = inet_sock_destruct;
352 sk->sk_family = PF_INET;
353 sk->sk_protocol = protocol;
354 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
360 inet->mc_list = NULL;
362 sk_refcnt_debug_inc(sk);
365 /* It assumes that any protocol which allows
366 * the user to assign a number at socket
367 * creation time automatically
370 inet->sport = htons(inet->num);
371 /* Add to protocol hash chains. */
372 sk->sk_prot->hash(sk);
375 if (sk->sk_prot->init) {
376 err = sk->sk_prot->init(sk);
378 sk_common_release(sk);
389 * The peer socket should always be NULL (or else). When we call this
390 * function we are destroying the object and from then on nobody
391 * should refer to it.
393 int inet_release(struct socket *sock)
395 struct sock *sk = sock->sk;
400 /* Applications forget to leave groups before exiting */
401 ip_mc_drop_socket(sk);
403 /* If linger is set, we don't return until the close
404 * is complete. Otherwise we return immediately. The
405 * actually closing is done the same either way.
407 * If the close is due to the process exiting, we never
411 if (sock_flag(sk, SOCK_LINGER) &&
412 !(current->flags & PF_EXITING))
413 timeout = sk->sk_lingertime;
415 sk->sk_prot->close(sk, timeout);
420 /* It is off by default, see below. */
421 int sysctl_ip_nonlocal_bind __read_mostly;
423 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
425 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
426 struct sock *sk = sock->sk;
427 struct inet_sock *inet = inet_sk(sk);
432 /* If the socket has its own bind function then use it. (RAW) */
433 if (sk->sk_prot->bind) {
434 err = sk->sk_prot->bind(sk, uaddr, addr_len);
438 if (addr_len < sizeof(struct sockaddr_in))
441 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
443 /* Not specified by any standard per-se, however it breaks too
444 * many applications when removed. It is unfortunate since
445 * allowing applications to make a non-local bind solves
446 * several problems with systems using dynamic addressing.
447 * (ie. your servers still start up even if your ISDN link
448 * is temporarily down)
450 err = -EADDRNOTAVAIL;
451 if (!sysctl_ip_nonlocal_bind &&
453 addr->sin_addr.s_addr != INADDR_ANY &&
454 chk_addr_ret != RTN_LOCAL &&
455 chk_addr_ret != RTN_MULTICAST &&
456 chk_addr_ret != RTN_BROADCAST)
459 snum = ntohs(addr->sin_port);
461 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
464 /* We keep a pair of addresses. rcv_saddr is the one
465 * used by hash lookups, and saddr is used for transmit.
467 * In the BSD API these are the same except where it
468 * would be illegal to use them (multicast/broadcast) in
469 * which case the sending device address is used.
473 /* Check these errors (active socket, double bind). */
475 if (sk->sk_state != TCP_CLOSE || inet->num)
476 goto out_release_sock;
478 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
479 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
480 inet->saddr = 0; /* Use device */
482 /* Make sure we are allowed to bind here. */
483 if (sk->sk_prot->get_port(sk, snum)) {
484 inet->saddr = inet->rcv_saddr = 0;
486 goto out_release_sock;
490 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
492 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
493 inet->sport = htons(inet->num);
504 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
505 int addr_len, int flags)
507 struct sock *sk = sock->sk;
509 if (uaddr->sa_family == AF_UNSPEC)
510 return sk->sk_prot->disconnect(sk, flags);
512 if (!inet_sk(sk)->num && inet_autobind(sk))
514 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
517 static long inet_wait_for_connect(struct sock *sk, long timeo)
521 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
523 /* Basic assumption: if someone sets sk->sk_err, he _must_
524 * change state of the socket from TCP_SYN_*.
525 * Connect() does not allow to get error notifications
526 * without closing the socket.
528 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
530 timeo = schedule_timeout(timeo);
532 if (signal_pending(current) || !timeo)
534 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
536 finish_wait(sk->sk_sleep, &wait);
541 * Connect to a remote host. There is regrettably still a little
542 * TCP 'magic' in here.
544 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
545 int addr_len, int flags)
547 struct sock *sk = sock->sk;
553 if (uaddr->sa_family == AF_UNSPEC) {
554 err = sk->sk_prot->disconnect(sk, flags);
555 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
559 switch (sock->state) {
568 /* Fall out of switch with err, set for this state */
572 if (sk->sk_state != TCP_CLOSE)
575 err = sk->sk_prot->connect(sk, uaddr, addr_len);
579 sock->state = SS_CONNECTING;
581 /* Just entered SS_CONNECTING state; the only
582 * difference is that return value in non-blocking
583 * case is EINPROGRESS, rather than EALREADY.
589 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
591 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
592 /* Error code is set above */
593 if (!timeo || !inet_wait_for_connect(sk, timeo))
596 err = sock_intr_errno(timeo);
597 if (signal_pending(current))
601 /* Connection was closed by RST, timeout, ICMP error
602 * or another process disconnected us.
604 if (sk->sk_state == TCP_CLOSE)
607 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
608 * and error was received after socket entered established state.
609 * Hence, it is handled normally after connect() return successfully.
612 sock->state = SS_CONNECTED;
619 err = sock_error(sk) ? : -ECONNABORTED;
620 sock->state = SS_UNCONNECTED;
621 if (sk->sk_prot->disconnect(sk, flags))
622 sock->state = SS_DISCONNECTING;
627 * Accept a pending connection. The TCP layer now gives BSD semantics.
630 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
632 struct sock *sk1 = sock->sk;
634 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
641 BUG_TRAP((1 << sk2->sk_state) &
642 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
644 sock_graft(sk2, newsock);
646 newsock->state = SS_CONNECTED;
655 * This does both peername and sockname.
657 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
658 int *uaddr_len, int peer)
660 struct sock *sk = sock->sk;
661 struct inet_sock *inet = inet_sk(sk);
662 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
664 sin->sin_family = AF_INET;
667 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
670 sin->sin_port = inet->dport;
671 sin->sin_addr.s_addr = inet->daddr;
673 __be32 addr = inet->rcv_saddr;
676 sin->sin_port = inet->sport;
677 sin->sin_addr.s_addr = addr;
679 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
680 *uaddr_len = sizeof(*sin);
684 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
687 struct sock *sk = sock->sk;
689 /* We may need to bind the socket. */
690 if (!inet_sk(sk)->num && inet_autobind(sk))
693 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
697 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
699 struct sock *sk = sock->sk;
701 /* We may need to bind the socket. */
702 if (!inet_sk(sk)->num && inet_autobind(sk))
705 if (sk->sk_prot->sendpage)
706 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
707 return sock_no_sendpage(sock, page, offset, size, flags);
711 int inet_shutdown(struct socket *sock, int how)
713 struct sock *sk = sock->sk;
716 /* This should really check to make sure
717 * the socket is a TCP socket. (WHY AC...)
719 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
722 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
726 if (sock->state == SS_CONNECTING) {
727 if ((1 << sk->sk_state) &
728 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
729 sock->state = SS_DISCONNECTING;
731 sock->state = SS_CONNECTED;
734 switch (sk->sk_state) {
737 /* Hack to wake up other listeners, who can poll for
738 POLLHUP, even on eg. unconnected UDP sockets -- RR */
740 sk->sk_shutdown |= how;
741 if (sk->sk_prot->shutdown)
742 sk->sk_prot->shutdown(sk, how);
745 /* Remaining two branches are temporary solution for missing
746 * close() in multithreaded environment. It is _not_ a good idea,
747 * but we have no choice until close() is repaired at VFS level.
750 if (!(how & RCV_SHUTDOWN))
754 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
755 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
759 /* Wake up anyone sleeping in poll. */
760 sk->sk_state_change(sk);
766 * ioctl() calls you can issue on an INET socket. Most of these are
767 * device configuration and stuff and very rarely used. Some ioctls
768 * pass on to the socket itself.
770 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
771 * loads the devconfigure module does its configuring and unloads it.
772 * There's a good 20K of config code hanging around the kernel.
775 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
777 struct sock *sk = sock->sk;
782 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
785 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
790 err = ip_rt_ioctl(cmd, (void __user *)arg);
795 err = arp_ioctl(cmd, (void __user *)arg);
808 err = devinet_ioctl(cmd, (void __user *)arg);
811 if (sk->sk_prot->ioctl)
812 err = sk->sk_prot->ioctl(sk, cmd, arg);
820 const struct proto_ops inet_stream_ops = {
822 .owner = THIS_MODULE,
823 .release = inet_release,
825 .connect = inet_stream_connect,
826 .socketpair = sock_no_socketpair,
827 .accept = inet_accept,
828 .getname = inet_getname,
831 .listen = inet_listen,
832 .shutdown = inet_shutdown,
833 .setsockopt = sock_common_setsockopt,
834 .getsockopt = sock_common_getsockopt,
835 .sendmsg = inet_sendmsg,
836 .recvmsg = sock_common_recvmsg,
837 .mmap = sock_no_mmap,
838 .sendpage = tcp_sendpage,
840 .compat_setsockopt = compat_sock_common_setsockopt,
841 .compat_getsockopt = compat_sock_common_getsockopt,
845 const struct proto_ops inet_dgram_ops = {
847 .owner = THIS_MODULE,
848 .release = inet_release,
850 .connect = inet_dgram_connect,
851 .socketpair = sock_no_socketpair,
852 .accept = sock_no_accept,
853 .getname = inet_getname,
856 .listen = sock_no_listen,
857 .shutdown = inet_shutdown,
858 .setsockopt = sock_common_setsockopt,
859 .getsockopt = sock_common_getsockopt,
860 .sendmsg = inet_sendmsg,
861 .recvmsg = sock_common_recvmsg,
862 .mmap = sock_no_mmap,
863 .sendpage = inet_sendpage,
865 .compat_setsockopt = compat_sock_common_setsockopt,
866 .compat_getsockopt = compat_sock_common_getsockopt,
871 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
874 static const struct proto_ops inet_sockraw_ops = {
876 .owner = THIS_MODULE,
877 .release = inet_release,
879 .connect = inet_dgram_connect,
880 .socketpair = sock_no_socketpair,
881 .accept = sock_no_accept,
882 .getname = inet_getname,
883 .poll = datagram_poll,
885 .listen = sock_no_listen,
886 .shutdown = inet_shutdown,
887 .setsockopt = sock_common_setsockopt,
888 .getsockopt = sock_common_getsockopt,
889 .sendmsg = inet_sendmsg,
890 .recvmsg = sock_common_recvmsg,
891 .mmap = sock_no_mmap,
892 .sendpage = inet_sendpage,
894 .compat_setsockopt = compat_sock_common_setsockopt,
895 .compat_getsockopt = compat_sock_common_getsockopt,
899 static struct net_proto_family inet_family_ops = {
901 .create = inet_create,
902 .owner = THIS_MODULE,
905 /* Upon startup we insert all the elements in inetsw_array[] into
906 * the linked list inetsw.
908 static struct inet_protosw inetsw_array[] =
912 .protocol = IPPROTO_TCP,
914 .ops = &inet_stream_ops,
917 .flags = INET_PROTOSW_PERMANENT |
923 .protocol = IPPROTO_UDP,
925 .ops = &inet_dgram_ops,
927 .no_check = UDP_CSUM_DEFAULT,
928 .flags = INET_PROTOSW_PERMANENT,
934 .protocol = IPPROTO_IP, /* wild card */
936 .ops = &inet_sockraw_ops,
937 .capability = CAP_NET_RAW,
938 .no_check = UDP_CSUM_DEFAULT,
939 .flags = INET_PROTOSW_REUSE,
943 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
945 void inet_register_protosw(struct inet_protosw *p)
947 struct list_head *lh;
948 struct inet_protosw *answer;
949 int protocol = p->protocol;
950 struct list_head *last_perm;
952 spin_lock_bh(&inetsw_lock);
954 if (p->type >= SOCK_MAX)
957 /* If we are trying to override a permanent protocol, bail. */
959 last_perm = &inetsw[p->type];
960 list_for_each(lh, &inetsw[p->type]) {
961 answer = list_entry(lh, struct inet_protosw, list);
963 /* Check only the non-wild match. */
964 if (INET_PROTOSW_PERMANENT & answer->flags) {
965 if (protocol == answer->protocol)
975 /* Add the new entry after the last permanent entry if any, so that
976 * the new entry does not override a permanent entry when matched with
977 * a wild-card protocol. But it is allowed to override any existing
978 * non-permanent entry. This means that when we remove this entry, the
979 * system automatically returns to the old behavior.
981 list_add_rcu(&p->list, last_perm);
983 spin_unlock_bh(&inetsw_lock);
990 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
996 "Ignoring attempt to register invalid socket type %d.\n",
1001 void inet_unregister_protosw(struct inet_protosw *p)
1003 if (INET_PROTOSW_PERMANENT & p->flags) {
1005 "Attempt to unregister permanent protocol %d.\n",
1008 spin_lock_bh(&inetsw_lock);
1009 list_del_rcu(&p->list);
1010 spin_unlock_bh(&inetsw_lock);
1017 * Shall we try to damage output packets if routing dev changes?
1020 int sysctl_ip_dynaddr __read_mostly;
1022 static int inet_sk_reselect_saddr(struct sock *sk)
1024 struct inet_sock *inet = inet_sk(sk);
1027 __be32 old_saddr = inet->saddr;
1029 __be32 daddr = inet->daddr;
1031 if (inet->opt && inet->opt->srr)
1032 daddr = inet->opt->faddr;
1034 /* Query new route. */
1035 err = ip_route_connect(&rt, daddr, 0,
1037 sk->sk_bound_dev_if,
1039 inet->sport, inet->dport, sk, 0);
1043 sk_setup_caps(sk, &rt->u.dst);
1045 new_saddr = rt->rt_src;
1047 if (new_saddr == old_saddr)
1050 if (sysctl_ip_dynaddr > 1) {
1051 printk(KERN_INFO "%s(): shifting inet->"
1052 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1055 NIPQUAD(new_saddr));
1058 inet->saddr = inet->rcv_saddr = new_saddr;
1061 * XXX The only one ugly spot where we need to
1062 * XXX really change the sockets identity after
1063 * XXX it has entered the hashes. -DaveM
1065 * Besides that, it does not check for connection
1066 * uniqueness. Wait for troubles.
1068 __sk_prot_rehash(sk);
1072 int inet_sk_rebuild_header(struct sock *sk)
1074 struct inet_sock *inet = inet_sk(sk);
1075 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1079 /* Route is OK, nothing to do. */
1084 daddr = inet->daddr;
1085 if (inet->opt && inet->opt->srr)
1086 daddr = inet->opt->faddr;
1089 .oif = sk->sk_bound_dev_if,
1093 .saddr = inet->saddr,
1094 .tos = RT_CONN_FLAGS(sk),
1097 .proto = sk->sk_protocol,
1100 .sport = inet->sport,
1101 .dport = inet->dport,
1106 security_sk_classify_flow(sk, &fl);
1107 err = ip_route_output_flow(&rt, &fl, sk, 0);
1110 sk_setup_caps(sk, &rt->u.dst);
1112 /* Routing failed... */
1113 sk->sk_route_caps = 0;
1115 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1116 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1118 if (!sysctl_ip_dynaddr ||
1119 sk->sk_state != TCP_SYN_SENT ||
1120 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1121 (err = inet_sk_reselect_saddr(sk)) != 0)
1122 sk->sk_err_soft = -err;
1128 EXPORT_SYMBOL(inet_sk_rebuild_header);
1130 static int inet_gso_send_check(struct sk_buff *skb)
1133 struct net_protocol *ops;
1138 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1143 if (ihl < sizeof(*iph))
1146 if (unlikely(!pskb_may_pull(skb, ihl)))
1149 __skb_pull(skb, ihl);
1150 skb_reset_transport_header(skb);
1152 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1153 err = -EPROTONOSUPPORT;
1156 ops = rcu_dereference(inet_protos[proto]);
1157 if (likely(ops && ops->gso_send_check))
1158 err = ops->gso_send_check(skb);
1165 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1167 struct sk_buff *segs = ERR_PTR(-EINVAL);
1169 struct net_protocol *ops;
1174 if (unlikely(skb_shinfo(skb)->gso_type &
1182 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1187 if (ihl < sizeof(*iph))
1190 if (unlikely(!pskb_may_pull(skb, ihl)))
1193 __skb_pull(skb, ihl);
1194 skb_reset_transport_header(skb);
1196 id = ntohs(iph->id);
1197 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1198 segs = ERR_PTR(-EPROTONOSUPPORT);
1201 ops = rcu_dereference(inet_protos[proto]);
1202 if (likely(ops && ops->gso_segment))
1203 segs = ops->gso_segment(skb, features);
1206 if (!segs || unlikely(IS_ERR(segs)))
1212 iph->id = htons(id++);
1213 iph->tot_len = htons(skb->len - skb->mac_len);
1215 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1216 } while ((skb = skb->next));
1222 unsigned long snmp_fold_field(void *mib[], int offt)
1224 unsigned long res = 0;
1227 for_each_possible_cpu(i) {
1228 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1229 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1233 EXPORT_SYMBOL_GPL(snmp_fold_field);
1235 int snmp_mib_init(void *ptr[2], size_t mibsize, size_t mibalign)
1237 BUG_ON(ptr == NULL);
1238 ptr[0] = __alloc_percpu(mibsize);
1241 ptr[1] = __alloc_percpu(mibsize);
1246 free_percpu(ptr[0]);
1251 EXPORT_SYMBOL_GPL(snmp_mib_init);
1253 void snmp_mib_free(void *ptr[2])
1255 BUG_ON(ptr == NULL);
1256 free_percpu(ptr[0]);
1257 free_percpu(ptr[1]);
1258 ptr[0] = ptr[1] = NULL;
1260 EXPORT_SYMBOL_GPL(snmp_mib_free);
1262 #ifdef CONFIG_IP_MULTICAST
1263 static struct net_protocol igmp_protocol = {
1264 .handler = igmp_rcv,
1268 static struct net_protocol tcp_protocol = {
1269 .handler = tcp_v4_rcv,
1270 .err_handler = tcp_v4_err,
1271 .gso_send_check = tcp_v4_gso_send_check,
1272 .gso_segment = tcp_tso_segment,
1276 static struct net_protocol udp_protocol = {
1278 .err_handler = udp_err,
1282 static struct net_protocol icmp_protocol = {
1283 .handler = icmp_rcv,
1286 static int __init init_ipv4_mibs(void)
1288 if (snmp_mib_init((void **)net_statistics,
1289 sizeof(struct linux_mib),
1290 __alignof__(struct linux_mib)) < 0)
1292 if (snmp_mib_init((void **)ip_statistics,
1293 sizeof(struct ipstats_mib),
1294 __alignof__(struct ipstats_mib)) < 0)
1296 if (snmp_mib_init((void **)icmp_statistics,
1297 sizeof(struct icmp_mib),
1298 __alignof__(struct icmp_mib)) < 0)
1300 if (snmp_mib_init((void **)tcp_statistics,
1301 sizeof(struct tcp_mib),
1302 __alignof__(struct tcp_mib)) < 0)
1304 if (snmp_mib_init((void **)udp_statistics,
1305 sizeof(struct udp_mib),
1306 __alignof__(struct udp_mib)) < 0)
1308 if (snmp_mib_init((void **)udplite_statistics,
1309 sizeof(struct udp_mib),
1310 __alignof__(struct udp_mib)) < 0)
1311 goto err_udplite_mib;
1318 snmp_mib_free((void **)udp_statistics);
1320 snmp_mib_free((void **)tcp_statistics);
1322 snmp_mib_free((void **)icmp_statistics);
1324 snmp_mib_free((void **)ip_statistics);
1326 snmp_mib_free((void **)net_statistics);
1331 static int ipv4_proc_init(void);
1334 * IP protocol layer initialiser
1337 static struct packet_type ip_packet_type = {
1338 .type = __constant_htons(ETH_P_IP),
1340 .gso_send_check = inet_gso_send_check,
1341 .gso_segment = inet_gso_segment,
1344 static int __init inet_init(void)
1346 struct sk_buff *dummy_skb;
1347 struct inet_protosw *q;
1348 struct list_head *r;
1351 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1353 rc = proto_register(&tcp_prot, 1);
1357 rc = proto_register(&udp_prot, 1);
1359 goto out_unregister_tcp_proto;
1361 rc = proto_register(&raw_prot, 1);
1363 goto out_unregister_udp_proto;
1366 * Tell SOCKET that we are alive...
1369 (void)sock_register(&inet_family_ops);
1372 * Add all the base protocols.
1375 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1376 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1377 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1378 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1379 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1380 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1381 #ifdef CONFIG_IP_MULTICAST
1382 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1383 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1386 /* Register the socket-side information for inet_create. */
1387 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1390 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1391 inet_register_protosw(q);
1394 * Set the ARP module up
1400 * Set the IP module up
1405 tcp_v4_init(&inet_family_ops);
1407 /* Setup TCP slab cache for open requests. */
1410 /* Add UDP-Lite (RFC 3828) */
1411 udplite4_register();
1414 * Set the ICMP layer up
1417 icmp_init(&inet_family_ops);
1420 * Initialise the multicast router
1422 #if defined(CONFIG_IP_MROUTE)
1426 * Initialise per-cpu ipv4 mibs
1429 if (init_ipv4_mibs())
1430 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1436 dev_add_pack(&ip_packet_type);
1441 out_unregister_udp_proto:
1442 proto_unregister(&udp_prot);
1443 out_unregister_tcp_proto:
1444 proto_unregister(&tcp_prot);
1448 fs_initcall(inet_init);
1450 /* ------------------------------------------------------------------------ */
1452 #ifdef CONFIG_PROC_FS
1453 static int __init ipv4_proc_init(void)
1457 if (raw_proc_init())
1459 if (tcp4_proc_init())
1461 if (udp4_proc_init())
1463 if (fib_proc_init())
1465 if (ip_misc_proc_init())
1482 #else /* CONFIG_PROC_FS */
1483 static int __init ipv4_proc_init(void)
1487 #endif /* CONFIG_PROC_FS */
1489 MODULE_ALIAS_NETPROTO(PF_INET);
1491 EXPORT_SYMBOL(inet_accept);
1492 EXPORT_SYMBOL(inet_bind);
1493 EXPORT_SYMBOL(inet_dgram_connect);
1494 EXPORT_SYMBOL(inet_dgram_ops);
1495 EXPORT_SYMBOL(inet_getname);
1496 EXPORT_SYMBOL(inet_ioctl);
1497 EXPORT_SYMBOL(inet_listen);
1498 EXPORT_SYMBOL(inet_register_protosw);
1499 EXPORT_SYMBOL(inet_release);
1500 EXPORT_SYMBOL(inet_sendmsg);
1501 EXPORT_SYMBOL(inet_shutdown);
1502 EXPORT_SYMBOL(inet_sock_destruct);
1503 EXPORT_SYMBOL(inet_stream_connect);
1504 EXPORT_SYMBOL(inet_stream_ops);
1505 EXPORT_SYMBOL(inet_unregister_protosw);
1506 EXPORT_SYMBOL(net_statistics);
1507 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);