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 /* New destruction routine */
131 void inet_sock_destruct(struct sock *sk)
133 struct inet_sock *inet = inet_sk(sk);
135 __skb_queue_purge(&sk->sk_receive_queue);
136 __skb_queue_purge(&sk->sk_error_queue);
138 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
139 printk("Attempt to release TCP socket in state %d %p\n",
143 if (!sock_flag(sk, SOCK_DEAD)) {
144 printk("Attempt to release alive inet socket %p\n", sk);
148 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
149 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
150 BUG_TRAP(!sk->sk_wmem_queued);
151 BUG_TRAP(!sk->sk_forward_alloc);
154 dst_release(sk->sk_dst_cache);
155 sk_refcnt_debug_dec(sk);
159 * The routines beyond this point handle the behaviour of an AF_INET
160 * socket object. Mostly it punts to the subprotocols of IP to do
165 * Automatically bind an unbound socket.
168 static int inet_autobind(struct sock *sk)
170 struct inet_sock *inet;
171 /* We may need to bind the socket. */
175 if (sk->sk_prot->get_port(sk, 0)) {
179 inet->sport = htons(inet->num);
186 * Move a socket into listening state.
188 int inet_listen(struct socket *sock, int backlog)
190 struct sock *sk = sock->sk;
191 unsigned char old_state;
197 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
200 old_state = sk->sk_state;
201 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
204 /* Really, if the socket is already in listen state
205 * we can only allow the backlog to be adjusted.
207 if (old_state != TCP_LISTEN) {
208 err = inet_csk_listen_start(sk, backlog);
212 sk->sk_max_ack_backlog = backlog;
220 u32 inet_ehash_secret __read_mostly;
221 EXPORT_SYMBOL(inet_ehash_secret);
224 * inet_ehash_secret must be set exactly once
225 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
227 void build_ehash_secret(void)
231 get_random_bytes(&rnd, sizeof(rnd));
233 spin_lock_bh(&inetsw_lock);
234 if (!inet_ehash_secret)
235 inet_ehash_secret = rnd;
236 spin_unlock_bh(&inetsw_lock);
238 EXPORT_SYMBOL(build_ehash_secret);
241 * Create an inet socket.
244 static int inet_create(struct net *net, struct socket *sock, int protocol)
248 struct inet_protosw *answer;
249 struct inet_sock *inet;
250 struct proto *answer_prot;
251 unsigned char answer_flags;
252 char answer_no_check;
253 int try_loading_module = 0;
256 if (net != &init_net)
257 return -EAFNOSUPPORT;
259 if (sock->type != SOCK_RAW &&
260 sock->type != SOCK_DGRAM &&
262 build_ehash_secret();
264 sock->state = SS_UNCONNECTED;
266 /* Look for the requested type/protocol pair. */
269 err = -ESOCKTNOSUPPORT;
271 list_for_each_rcu(p, &inetsw[sock->type]) {
272 answer = list_entry(p, struct inet_protosw, list);
274 /* Check the non-wild match. */
275 if (protocol == answer->protocol) {
276 if (protocol != IPPROTO_IP)
279 /* Check for the two wild cases. */
280 if (IPPROTO_IP == protocol) {
281 protocol = answer->protocol;
284 if (IPPROTO_IP == answer->protocol)
287 err = -EPROTONOSUPPORT;
291 if (unlikely(answer == NULL)) {
292 if (try_loading_module < 2) {
295 * Be more specific, e.g. net-pf-2-proto-132-type-1
296 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
298 if (++try_loading_module == 1)
299 request_module("net-pf-%d-proto-%d-type-%d",
300 PF_INET, protocol, sock->type);
302 * Fall back to generic, e.g. net-pf-2-proto-132
303 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
306 request_module("net-pf-%d-proto-%d",
308 goto lookup_protocol;
314 if (answer->capability > 0 && !capable(answer->capability))
317 sock->ops = answer->ops;
318 answer_prot = answer->prot;
319 answer_no_check = answer->no_check;
320 answer_flags = answer->flags;
323 BUG_TRAP(answer_prot->slab != NULL);
326 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, 1);
331 sk->sk_no_check = answer_no_check;
332 if (INET_PROTOSW_REUSE & answer_flags)
336 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
338 if (SOCK_RAW == sock->type) {
339 inet->num = protocol;
340 if (IPPROTO_RAW == protocol)
344 if (ipv4_config.no_pmtu_disc)
345 inet->pmtudisc = IP_PMTUDISC_DONT;
347 inet->pmtudisc = IP_PMTUDISC_WANT;
351 sock_init_data(sock, sk);
353 sk->sk_destruct = inet_sock_destruct;
354 sk->sk_family = PF_INET;
355 sk->sk_protocol = protocol;
356 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
362 inet->mc_list = NULL;
364 sk_refcnt_debug_inc(sk);
367 /* It assumes that any protocol which allows
368 * the user to assign a number at socket
369 * creation time automatically
372 inet->sport = htons(inet->num);
373 /* Add to protocol hash chains. */
374 sk->sk_prot->hash(sk);
377 if (sk->sk_prot->init) {
378 err = sk->sk_prot->init(sk);
380 sk_common_release(sk);
391 * The peer socket should always be NULL (or else). When we call this
392 * function we are destroying the object and from then on nobody
393 * should refer to it.
395 int inet_release(struct socket *sock)
397 struct sock *sk = sock->sk;
402 /* Applications forget to leave groups before exiting */
403 ip_mc_drop_socket(sk);
405 /* If linger is set, we don't return until the close
406 * is complete. Otherwise we return immediately. The
407 * actually closing is done the same either way.
409 * If the close is due to the process exiting, we never
413 if (sock_flag(sk, SOCK_LINGER) &&
414 !(current->flags & PF_EXITING))
415 timeout = sk->sk_lingertime;
417 sk->sk_prot->close(sk, timeout);
422 /* It is off by default, see below. */
423 int sysctl_ip_nonlocal_bind __read_mostly;
425 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
427 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
428 struct sock *sk = sock->sk;
429 struct inet_sock *inet = inet_sk(sk);
434 /* If the socket has its own bind function then use it. (RAW) */
435 if (sk->sk_prot->bind) {
436 err = sk->sk_prot->bind(sk, uaddr, addr_len);
440 if (addr_len < sizeof(struct sockaddr_in))
443 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
445 /* Not specified by any standard per-se, however it breaks too
446 * many applications when removed. It is unfortunate since
447 * allowing applications to make a non-local bind solves
448 * several problems with systems using dynamic addressing.
449 * (ie. your servers still start up even if your ISDN link
450 * is temporarily down)
452 err = -EADDRNOTAVAIL;
453 if (!sysctl_ip_nonlocal_bind &&
455 addr->sin_addr.s_addr != INADDR_ANY &&
456 chk_addr_ret != RTN_LOCAL &&
457 chk_addr_ret != RTN_MULTICAST &&
458 chk_addr_ret != RTN_BROADCAST)
461 snum = ntohs(addr->sin_port);
463 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
466 /* We keep a pair of addresses. rcv_saddr is the one
467 * used by hash lookups, and saddr is used for transmit.
469 * In the BSD API these are the same except where it
470 * would be illegal to use them (multicast/broadcast) in
471 * which case the sending device address is used.
475 /* Check these errors (active socket, double bind). */
477 if (sk->sk_state != TCP_CLOSE || inet->num)
478 goto out_release_sock;
480 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
481 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
482 inet->saddr = 0; /* Use device */
484 /* Make sure we are allowed to bind here. */
485 if (sk->sk_prot->get_port(sk, snum)) {
486 inet->saddr = inet->rcv_saddr = 0;
488 goto out_release_sock;
492 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
494 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
495 inet->sport = htons(inet->num);
506 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
507 int addr_len, int flags)
509 struct sock *sk = sock->sk;
511 if (uaddr->sa_family == AF_UNSPEC)
512 return sk->sk_prot->disconnect(sk, flags);
514 if (!inet_sk(sk)->num && inet_autobind(sk))
516 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
519 static long inet_wait_for_connect(struct sock *sk, long timeo)
523 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
525 /* Basic assumption: if someone sets sk->sk_err, he _must_
526 * change state of the socket from TCP_SYN_*.
527 * Connect() does not allow to get error notifications
528 * without closing the socket.
530 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
532 timeo = schedule_timeout(timeo);
534 if (signal_pending(current) || !timeo)
536 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
538 finish_wait(sk->sk_sleep, &wait);
543 * Connect to a remote host. There is regrettably still a little
544 * TCP 'magic' in here.
546 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
547 int addr_len, int flags)
549 struct sock *sk = sock->sk;
555 if (uaddr->sa_family == AF_UNSPEC) {
556 err = sk->sk_prot->disconnect(sk, flags);
557 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
561 switch (sock->state) {
570 /* Fall out of switch with err, set for this state */
574 if (sk->sk_state != TCP_CLOSE)
577 err = sk->sk_prot->connect(sk, uaddr, addr_len);
581 sock->state = SS_CONNECTING;
583 /* Just entered SS_CONNECTING state; the only
584 * difference is that return value in non-blocking
585 * case is EINPROGRESS, rather than EALREADY.
591 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
593 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
594 /* Error code is set above */
595 if (!timeo || !inet_wait_for_connect(sk, timeo))
598 err = sock_intr_errno(timeo);
599 if (signal_pending(current))
603 /* Connection was closed by RST, timeout, ICMP error
604 * or another process disconnected us.
606 if (sk->sk_state == TCP_CLOSE)
609 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
610 * and error was received after socket entered established state.
611 * Hence, it is handled normally after connect() return successfully.
614 sock->state = SS_CONNECTED;
621 err = sock_error(sk) ? : -ECONNABORTED;
622 sock->state = SS_UNCONNECTED;
623 if (sk->sk_prot->disconnect(sk, flags))
624 sock->state = SS_DISCONNECTING;
629 * Accept a pending connection. The TCP layer now gives BSD semantics.
632 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
634 struct sock *sk1 = sock->sk;
636 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
643 BUG_TRAP((1 << sk2->sk_state) &
644 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
646 sock_graft(sk2, newsock);
648 newsock->state = SS_CONNECTED;
657 * This does both peername and sockname.
659 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
660 int *uaddr_len, int peer)
662 struct sock *sk = sock->sk;
663 struct inet_sock *inet = inet_sk(sk);
664 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
666 sin->sin_family = AF_INET;
669 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
672 sin->sin_port = inet->dport;
673 sin->sin_addr.s_addr = inet->daddr;
675 __be32 addr = inet->rcv_saddr;
678 sin->sin_port = inet->sport;
679 sin->sin_addr.s_addr = addr;
681 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
682 *uaddr_len = sizeof(*sin);
686 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
689 struct sock *sk = sock->sk;
691 /* We may need to bind the socket. */
692 if (!inet_sk(sk)->num && inet_autobind(sk))
695 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
699 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
701 struct sock *sk = sock->sk;
703 /* We may need to bind the socket. */
704 if (!inet_sk(sk)->num && inet_autobind(sk))
707 if (sk->sk_prot->sendpage)
708 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
709 return sock_no_sendpage(sock, page, offset, size, flags);
713 int inet_shutdown(struct socket *sock, int how)
715 struct sock *sk = sock->sk;
718 /* This should really check to make sure
719 * the socket is a TCP socket. (WHY AC...)
721 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
724 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
728 if (sock->state == SS_CONNECTING) {
729 if ((1 << sk->sk_state) &
730 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
731 sock->state = SS_DISCONNECTING;
733 sock->state = SS_CONNECTED;
736 switch (sk->sk_state) {
739 /* Hack to wake up other listeners, who can poll for
740 POLLHUP, even on eg. unconnected UDP sockets -- RR */
742 sk->sk_shutdown |= how;
743 if (sk->sk_prot->shutdown)
744 sk->sk_prot->shutdown(sk, how);
747 /* Remaining two branches are temporary solution for missing
748 * close() in multithreaded environment. It is _not_ a good idea,
749 * but we have no choice until close() is repaired at VFS level.
752 if (!(how & RCV_SHUTDOWN))
756 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
757 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
761 /* Wake up anyone sleeping in poll. */
762 sk->sk_state_change(sk);
768 * ioctl() calls you can issue on an INET socket. Most of these are
769 * device configuration and stuff and very rarely used. Some ioctls
770 * pass on to the socket itself.
772 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
773 * loads the devconfigure module does its configuring and unloads it.
774 * There's a good 20K of config code hanging around the kernel.
777 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
779 struct sock *sk = sock->sk;
784 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
787 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
792 err = ip_rt_ioctl(cmd, (void __user *)arg);
797 err = arp_ioctl(cmd, (void __user *)arg);
810 err = devinet_ioctl(cmd, (void __user *)arg);
813 if (sk->sk_prot->ioctl)
814 err = sk->sk_prot->ioctl(sk, cmd, arg);
822 const struct proto_ops inet_stream_ops = {
824 .owner = THIS_MODULE,
825 .release = inet_release,
827 .connect = inet_stream_connect,
828 .socketpair = sock_no_socketpair,
829 .accept = inet_accept,
830 .getname = inet_getname,
833 .listen = inet_listen,
834 .shutdown = inet_shutdown,
835 .setsockopt = sock_common_setsockopt,
836 .getsockopt = sock_common_getsockopt,
837 .sendmsg = tcp_sendmsg,
838 .recvmsg = sock_common_recvmsg,
839 .mmap = sock_no_mmap,
840 .sendpage = tcp_sendpage,
842 .compat_setsockopt = compat_sock_common_setsockopt,
843 .compat_getsockopt = compat_sock_common_getsockopt,
847 const struct proto_ops inet_dgram_ops = {
849 .owner = THIS_MODULE,
850 .release = inet_release,
852 .connect = inet_dgram_connect,
853 .socketpair = sock_no_socketpair,
854 .accept = sock_no_accept,
855 .getname = inet_getname,
858 .listen = sock_no_listen,
859 .shutdown = inet_shutdown,
860 .setsockopt = sock_common_setsockopt,
861 .getsockopt = sock_common_getsockopt,
862 .sendmsg = inet_sendmsg,
863 .recvmsg = sock_common_recvmsg,
864 .mmap = sock_no_mmap,
865 .sendpage = inet_sendpage,
867 .compat_setsockopt = compat_sock_common_setsockopt,
868 .compat_getsockopt = compat_sock_common_getsockopt,
873 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
876 static const struct proto_ops inet_sockraw_ops = {
878 .owner = THIS_MODULE,
879 .release = inet_release,
881 .connect = inet_dgram_connect,
882 .socketpair = sock_no_socketpair,
883 .accept = sock_no_accept,
884 .getname = inet_getname,
885 .poll = datagram_poll,
887 .listen = sock_no_listen,
888 .shutdown = inet_shutdown,
889 .setsockopt = sock_common_setsockopt,
890 .getsockopt = sock_common_getsockopt,
891 .sendmsg = inet_sendmsg,
892 .recvmsg = sock_common_recvmsg,
893 .mmap = sock_no_mmap,
894 .sendpage = inet_sendpage,
896 .compat_setsockopt = compat_sock_common_setsockopt,
897 .compat_getsockopt = compat_sock_common_getsockopt,
901 static struct net_proto_family inet_family_ops = {
903 .create = inet_create,
904 .owner = THIS_MODULE,
907 /* Upon startup we insert all the elements in inetsw_array[] into
908 * the linked list inetsw.
910 static struct inet_protosw inetsw_array[] =
914 .protocol = IPPROTO_TCP,
916 .ops = &inet_stream_ops,
919 .flags = INET_PROTOSW_PERMANENT |
925 .protocol = IPPROTO_UDP,
927 .ops = &inet_dgram_ops,
929 .no_check = UDP_CSUM_DEFAULT,
930 .flags = INET_PROTOSW_PERMANENT,
936 .protocol = IPPROTO_IP, /* wild card */
938 .ops = &inet_sockraw_ops,
939 .capability = CAP_NET_RAW,
940 .no_check = UDP_CSUM_DEFAULT,
941 .flags = INET_PROTOSW_REUSE,
945 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
947 void inet_register_protosw(struct inet_protosw *p)
949 struct list_head *lh;
950 struct inet_protosw *answer;
951 int protocol = p->protocol;
952 struct list_head *last_perm;
954 spin_lock_bh(&inetsw_lock);
956 if (p->type >= SOCK_MAX)
959 /* If we are trying to override a permanent protocol, bail. */
961 last_perm = &inetsw[p->type];
962 list_for_each(lh, &inetsw[p->type]) {
963 answer = list_entry(lh, struct inet_protosw, list);
965 /* Check only the non-wild match. */
966 if (INET_PROTOSW_PERMANENT & answer->flags) {
967 if (protocol == answer->protocol)
977 /* Add the new entry after the last permanent entry if any, so that
978 * the new entry does not override a permanent entry when matched with
979 * a wild-card protocol. But it is allowed to override any existing
980 * non-permanent entry. This means that when we remove this entry, the
981 * system automatically returns to the old behavior.
983 list_add_rcu(&p->list, last_perm);
985 spin_unlock_bh(&inetsw_lock);
992 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
998 "Ignoring attempt to register invalid socket type %d.\n",
1003 void inet_unregister_protosw(struct inet_protosw *p)
1005 if (INET_PROTOSW_PERMANENT & p->flags) {
1007 "Attempt to unregister permanent protocol %d.\n",
1010 spin_lock_bh(&inetsw_lock);
1011 list_del_rcu(&p->list);
1012 spin_unlock_bh(&inetsw_lock);
1019 * Shall we try to damage output packets if routing dev changes?
1022 int sysctl_ip_dynaddr __read_mostly;
1024 static int inet_sk_reselect_saddr(struct sock *sk)
1026 struct inet_sock *inet = inet_sk(sk);
1029 __be32 old_saddr = inet->saddr;
1031 __be32 daddr = inet->daddr;
1033 if (inet->opt && inet->opt->srr)
1034 daddr = inet->opt->faddr;
1036 /* Query new route. */
1037 err = ip_route_connect(&rt, daddr, 0,
1039 sk->sk_bound_dev_if,
1041 inet->sport, inet->dport, sk, 0);
1045 sk_setup_caps(sk, &rt->u.dst);
1047 new_saddr = rt->rt_src;
1049 if (new_saddr == old_saddr)
1052 if (sysctl_ip_dynaddr > 1) {
1053 printk(KERN_INFO "%s(): shifting inet->"
1054 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1057 NIPQUAD(new_saddr));
1060 inet->saddr = inet->rcv_saddr = new_saddr;
1063 * XXX The only one ugly spot where we need to
1064 * XXX really change the sockets identity after
1065 * XXX it has entered the hashes. -DaveM
1067 * Besides that, it does not check for connection
1068 * uniqueness. Wait for troubles.
1070 __sk_prot_rehash(sk);
1074 int inet_sk_rebuild_header(struct sock *sk)
1076 struct inet_sock *inet = inet_sk(sk);
1077 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1081 /* Route is OK, nothing to do. */
1086 daddr = inet->daddr;
1087 if (inet->opt && inet->opt->srr)
1088 daddr = inet->opt->faddr;
1091 .oif = sk->sk_bound_dev_if,
1095 .saddr = inet->saddr,
1096 .tos = RT_CONN_FLAGS(sk),
1099 .proto = sk->sk_protocol,
1102 .sport = inet->sport,
1103 .dport = inet->dport,
1108 security_sk_classify_flow(sk, &fl);
1109 err = ip_route_output_flow(&rt, &fl, sk, 0);
1112 sk_setup_caps(sk, &rt->u.dst);
1114 /* Routing failed... */
1115 sk->sk_route_caps = 0;
1117 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1118 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1120 if (!sysctl_ip_dynaddr ||
1121 sk->sk_state != TCP_SYN_SENT ||
1122 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1123 (err = inet_sk_reselect_saddr(sk)) != 0)
1124 sk->sk_err_soft = -err;
1130 EXPORT_SYMBOL(inet_sk_rebuild_header);
1132 static int inet_gso_send_check(struct sk_buff *skb)
1135 struct net_protocol *ops;
1140 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1145 if (ihl < sizeof(*iph))
1148 if (unlikely(!pskb_may_pull(skb, ihl)))
1151 __skb_pull(skb, ihl);
1152 skb_reset_transport_header(skb);
1154 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1155 err = -EPROTONOSUPPORT;
1158 ops = rcu_dereference(inet_protos[proto]);
1159 if (likely(ops && ops->gso_send_check))
1160 err = ops->gso_send_check(skb);
1167 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1169 struct sk_buff *segs = ERR_PTR(-EINVAL);
1171 struct net_protocol *ops;
1176 if (!(features & NETIF_F_V4_CSUM))
1177 features &= ~NETIF_F_SG;
1179 if (unlikely(skb_shinfo(skb)->gso_type &
1187 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1192 if (ihl < sizeof(*iph))
1195 if (unlikely(!pskb_may_pull(skb, ihl)))
1198 __skb_pull(skb, ihl);
1199 skb_reset_transport_header(skb);
1201 id = ntohs(iph->id);
1202 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1203 segs = ERR_PTR(-EPROTONOSUPPORT);
1206 ops = rcu_dereference(inet_protos[proto]);
1207 if (likely(ops && ops->gso_segment))
1208 segs = ops->gso_segment(skb, features);
1211 if (!segs || unlikely(IS_ERR(segs)))
1217 iph->id = htons(id++);
1218 iph->tot_len = htons(skb->len - skb->mac_len);
1220 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1221 } while ((skb = skb->next));
1227 unsigned long snmp_fold_field(void *mib[], int offt)
1229 unsigned long res = 0;
1232 for_each_possible_cpu(i) {
1233 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1234 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1238 EXPORT_SYMBOL_GPL(snmp_fold_field);
1240 int snmp_mib_init(void *ptr[2], size_t mibsize, size_t mibalign)
1242 BUG_ON(ptr == NULL);
1243 ptr[0] = __alloc_percpu(mibsize);
1246 ptr[1] = __alloc_percpu(mibsize);
1251 free_percpu(ptr[0]);
1256 EXPORT_SYMBOL_GPL(snmp_mib_init);
1258 void snmp_mib_free(void *ptr[2])
1260 BUG_ON(ptr == NULL);
1261 free_percpu(ptr[0]);
1262 free_percpu(ptr[1]);
1263 ptr[0] = ptr[1] = NULL;
1265 EXPORT_SYMBOL_GPL(snmp_mib_free);
1267 #ifdef CONFIG_IP_MULTICAST
1268 static struct net_protocol igmp_protocol = {
1269 .handler = igmp_rcv,
1273 static struct net_protocol tcp_protocol = {
1274 .handler = tcp_v4_rcv,
1275 .err_handler = tcp_v4_err,
1276 .gso_send_check = tcp_v4_gso_send_check,
1277 .gso_segment = tcp_tso_segment,
1281 static struct net_protocol udp_protocol = {
1283 .err_handler = udp_err,
1287 static struct net_protocol icmp_protocol = {
1288 .handler = icmp_rcv,
1291 static int __init init_ipv4_mibs(void)
1293 if (snmp_mib_init((void **)net_statistics,
1294 sizeof(struct linux_mib),
1295 __alignof__(struct linux_mib)) < 0)
1297 if (snmp_mib_init((void **)ip_statistics,
1298 sizeof(struct ipstats_mib),
1299 __alignof__(struct ipstats_mib)) < 0)
1301 if (snmp_mib_init((void **)icmp_statistics,
1302 sizeof(struct icmp_mib),
1303 __alignof__(struct icmp_mib)) < 0)
1305 if (snmp_mib_init((void **)icmpmsg_statistics,
1306 sizeof(struct icmpmsg_mib),
1307 __alignof__(struct icmpmsg_mib)) < 0)
1308 goto err_icmpmsg_mib;
1309 if (snmp_mib_init((void **)tcp_statistics,
1310 sizeof(struct tcp_mib),
1311 __alignof__(struct tcp_mib)) < 0)
1313 if (snmp_mib_init((void **)udp_statistics,
1314 sizeof(struct udp_mib),
1315 __alignof__(struct udp_mib)) < 0)
1317 if (snmp_mib_init((void **)udplite_statistics,
1318 sizeof(struct udp_mib),
1319 __alignof__(struct udp_mib)) < 0)
1320 goto err_udplite_mib;
1327 snmp_mib_free((void **)udp_statistics);
1329 snmp_mib_free((void **)tcp_statistics);
1331 snmp_mib_free((void **)icmpmsg_statistics);
1333 snmp_mib_free((void **)icmp_statistics);
1335 snmp_mib_free((void **)ip_statistics);
1337 snmp_mib_free((void **)net_statistics);
1342 static int ipv4_proc_init(void);
1345 * IP protocol layer initialiser
1348 static struct packet_type ip_packet_type = {
1349 .type = __constant_htons(ETH_P_IP),
1351 .gso_send_check = inet_gso_send_check,
1352 .gso_segment = inet_gso_segment,
1355 static int __init inet_init(void)
1357 struct sk_buff *dummy_skb;
1358 struct inet_protosw *q;
1359 struct list_head *r;
1362 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1364 rc = proto_register(&tcp_prot, 1);
1368 rc = proto_register(&udp_prot, 1);
1370 goto out_unregister_tcp_proto;
1372 rc = proto_register(&raw_prot, 1);
1374 goto out_unregister_udp_proto;
1377 * Tell SOCKET that we are alive...
1380 (void)sock_register(&inet_family_ops);
1383 * Add all the base protocols.
1386 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1387 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1388 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1389 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1390 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1391 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1392 #ifdef CONFIG_IP_MULTICAST
1393 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1394 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1397 /* Register the socket-side information for inet_create. */
1398 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1401 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1402 inet_register_protosw(q);
1405 * Set the ARP module up
1411 * Set the IP module up
1416 tcp_v4_init(&inet_family_ops);
1418 /* Setup TCP slab cache for open requests. */
1421 /* Add UDP-Lite (RFC 3828) */
1422 udplite4_register();
1425 * Set the ICMP layer up
1428 icmp_init(&inet_family_ops);
1431 * Initialise the multicast router
1433 #if defined(CONFIG_IP_MROUTE)
1437 * Initialise per-cpu ipv4 mibs
1440 if (init_ipv4_mibs())
1441 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1447 dev_add_pack(&ip_packet_type);
1452 out_unregister_udp_proto:
1453 proto_unregister(&udp_prot);
1454 out_unregister_tcp_proto:
1455 proto_unregister(&tcp_prot);
1459 fs_initcall(inet_init);
1461 /* ------------------------------------------------------------------------ */
1463 #ifdef CONFIG_PROC_FS
1464 static int __init ipv4_proc_init(void)
1468 if (raw_proc_init())
1470 if (tcp4_proc_init())
1472 if (udp4_proc_init())
1474 if (fib_proc_init())
1476 if (ip_misc_proc_init())
1493 #else /* CONFIG_PROC_FS */
1494 static int __init ipv4_proc_init(void)
1498 #endif /* CONFIG_PROC_FS */
1500 MODULE_ALIAS_NETPROTO(PF_INET);
1502 EXPORT_SYMBOL(inet_accept);
1503 EXPORT_SYMBOL(inet_bind);
1504 EXPORT_SYMBOL(inet_dgram_connect);
1505 EXPORT_SYMBOL(inet_dgram_ops);
1506 EXPORT_SYMBOL(inet_getname);
1507 EXPORT_SYMBOL(inet_ioctl);
1508 EXPORT_SYMBOL(inet_listen);
1509 EXPORT_SYMBOL(inet_register_protosw);
1510 EXPORT_SYMBOL(inet_release);
1511 EXPORT_SYMBOL(inet_sendmsg);
1512 EXPORT_SYMBOL(inet_shutdown);
1513 EXPORT_SYMBOL(inet_sock_destruct);
1514 EXPORT_SYMBOL(inet_stream_connect);
1515 EXPORT_SYMBOL(inet_stream_ops);
1516 EXPORT_SYMBOL(inet_unregister_protosw);
1517 EXPORT_SYMBOL(net_statistics);
1518 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);