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);
144 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 printk("Attempt to release TCP socket in state %d %p\n",
149 if (!sock_flag(sk, SOCK_DEAD)) {
150 printk("Attempt to release alive inet socket %p\n", sk);
154 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
155 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
156 BUG_TRAP(!sk->sk_wmem_queued);
157 BUG_TRAP(!sk->sk_forward_alloc);
160 dst_release(sk->sk_dst_cache);
161 sk_refcnt_debug_dec(sk);
165 * The routines beyond this point handle the behaviour of an AF_INET
166 * socket object. Mostly it punts to the subprotocols of IP to do
171 * Automatically bind an unbound socket.
174 static int inet_autobind(struct sock *sk)
176 struct inet_sock *inet;
177 /* We may need to bind the socket. */
181 if (sk->sk_prot->get_port(sk, 0)) {
185 inet->sport = htons(inet->num);
192 * Move a socket into listening state.
194 int inet_listen(struct socket *sock, int backlog)
196 struct sock *sk = sock->sk;
197 unsigned char old_state;
203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
206 old_state = sk->sk_state;
207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
213 if (old_state != TCP_LISTEN) {
214 err = inet_csk_listen_start(sk, backlog);
218 sk->sk_max_ack_backlog = backlog;
226 u32 inet_ehash_secret __read_mostly;
227 EXPORT_SYMBOL(inet_ehash_secret);
230 * inet_ehash_secret must be set exactly once
231 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
233 void build_ehash_secret(void)
237 get_random_bytes(&rnd, sizeof(rnd));
239 spin_lock_bh(&inetsw_lock);
240 if (!inet_ehash_secret)
241 inet_ehash_secret = rnd;
242 spin_unlock_bh(&inetsw_lock);
244 EXPORT_SYMBOL(build_ehash_secret);
246 static inline int inet_netns_ok(struct net *net, int protocol)
249 struct net_protocol *ipprot;
251 if (net == &init_net)
254 hash = protocol & (MAX_INET_PROTOS - 1);
255 ipprot = rcu_dereference(inet_protos[hash]);
260 return ipprot->netns_ok;
264 * Create an inet socket.
267 static int inet_create(struct net *net, struct socket *sock, int protocol)
271 struct inet_protosw *answer;
272 struct inet_sock *inet;
273 struct proto *answer_prot;
274 unsigned char answer_flags;
275 char answer_no_check;
276 int try_loading_module = 0;
279 if (sock->type != SOCK_RAW &&
280 sock->type != SOCK_DGRAM &&
282 build_ehash_secret();
284 sock->state = SS_UNCONNECTED;
286 /* Look for the requested type/protocol pair. */
289 err = -ESOCKTNOSUPPORT;
291 list_for_each_rcu(p, &inetsw[sock->type]) {
292 answer = list_entry(p, struct inet_protosw, list);
294 /* Check the non-wild match. */
295 if (protocol == answer->protocol) {
296 if (protocol != IPPROTO_IP)
299 /* Check for the two wild cases. */
300 if (IPPROTO_IP == protocol) {
301 protocol = answer->protocol;
304 if (IPPROTO_IP == answer->protocol)
307 err = -EPROTONOSUPPORT;
311 if (unlikely(answer == NULL)) {
312 if (try_loading_module < 2) {
315 * Be more specific, e.g. net-pf-2-proto-132-type-1
316 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
318 if (++try_loading_module == 1)
319 request_module("net-pf-%d-proto-%d-type-%d",
320 PF_INET, protocol, sock->type);
322 * Fall back to generic, e.g. net-pf-2-proto-132
323 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
326 request_module("net-pf-%d-proto-%d",
328 goto lookup_protocol;
334 if (answer->capability > 0 && !capable(answer->capability))
338 if (!inet_netns_ok(net, protocol))
341 sock->ops = answer->ops;
342 answer_prot = answer->prot;
343 answer_no_check = answer->no_check;
344 answer_flags = answer->flags;
347 BUG_TRAP(answer_prot->slab != NULL);
350 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
355 sk->sk_no_check = answer_no_check;
356 if (INET_PROTOSW_REUSE & answer_flags)
360 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
362 if (SOCK_RAW == sock->type) {
363 inet->num = protocol;
364 if (IPPROTO_RAW == protocol)
368 if (ipv4_config.no_pmtu_disc)
369 inet->pmtudisc = IP_PMTUDISC_DONT;
371 inet->pmtudisc = IP_PMTUDISC_WANT;
375 sock_init_data(sock, sk);
377 sk->sk_destruct = inet_sock_destruct;
378 sk->sk_family = PF_INET;
379 sk->sk_protocol = protocol;
380 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
386 inet->mc_list = NULL;
388 sk_refcnt_debug_inc(sk);
391 /* It assumes that any protocol which allows
392 * the user to assign a number at socket
393 * creation time automatically
396 inet->sport = htons(inet->num);
397 /* Add to protocol hash chains. */
398 sk->sk_prot->hash(sk);
401 if (sk->sk_prot->init) {
402 err = sk->sk_prot->init(sk);
404 sk_common_release(sk);
415 * The peer socket should always be NULL (or else). When we call this
416 * function we are destroying the object and from then on nobody
417 * should refer to it.
419 int inet_release(struct socket *sock)
421 struct sock *sk = sock->sk;
426 /* Applications forget to leave groups before exiting */
427 ip_mc_drop_socket(sk);
429 /* If linger is set, we don't return until the close
430 * is complete. Otherwise we return immediately. The
431 * actually closing is done the same either way.
433 * If the close is due to the process exiting, we never
437 if (sock_flag(sk, SOCK_LINGER) &&
438 !(current->flags & PF_EXITING))
439 timeout = sk->sk_lingertime;
441 sk->sk_prot->close(sk, timeout);
446 /* It is off by default, see below. */
447 int sysctl_ip_nonlocal_bind __read_mostly;
449 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
451 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
452 struct sock *sk = sock->sk;
453 struct inet_sock *inet = inet_sk(sk);
458 /* If the socket has its own bind function then use it. (RAW) */
459 if (sk->sk_prot->bind) {
460 err = sk->sk_prot->bind(sk, uaddr, addr_len);
464 if (addr_len < sizeof(struct sockaddr_in))
467 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
469 /* Not specified by any standard per-se, however it breaks too
470 * many applications when removed. It is unfortunate since
471 * allowing applications to make a non-local bind solves
472 * several problems with systems using dynamic addressing.
473 * (ie. your servers still start up even if your ISDN link
474 * is temporarily down)
476 err = -EADDRNOTAVAIL;
477 if (!sysctl_ip_nonlocal_bind &&
479 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
480 chk_addr_ret != RTN_LOCAL &&
481 chk_addr_ret != RTN_MULTICAST &&
482 chk_addr_ret != RTN_BROADCAST)
485 snum = ntohs(addr->sin_port);
487 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
490 /* We keep a pair of addresses. rcv_saddr is the one
491 * used by hash lookups, and saddr is used for transmit.
493 * In the BSD API these are the same except where it
494 * would be illegal to use them (multicast/broadcast) in
495 * which case the sending device address is used.
499 /* Check these errors (active socket, double bind). */
501 if (sk->sk_state != TCP_CLOSE || inet->num)
502 goto out_release_sock;
504 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
505 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
506 inet->saddr = 0; /* Use device */
508 /* Make sure we are allowed to bind here. */
509 if (sk->sk_prot->get_port(sk, snum)) {
510 inet->saddr = inet->rcv_saddr = 0;
512 goto out_release_sock;
516 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
518 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
519 inet->sport = htons(inet->num);
530 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
531 int addr_len, int flags)
533 struct sock *sk = sock->sk;
535 if (uaddr->sa_family == AF_UNSPEC)
536 return sk->sk_prot->disconnect(sk, flags);
538 if (!inet_sk(sk)->num && inet_autobind(sk))
540 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
543 static long inet_wait_for_connect(struct sock *sk, long timeo)
547 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
549 /* Basic assumption: if someone sets sk->sk_err, he _must_
550 * change state of the socket from TCP_SYN_*.
551 * Connect() does not allow to get error notifications
552 * without closing the socket.
554 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
556 timeo = schedule_timeout(timeo);
558 if (signal_pending(current) || !timeo)
560 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
562 finish_wait(sk->sk_sleep, &wait);
567 * Connect to a remote host. There is regrettably still a little
568 * TCP 'magic' in here.
570 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
571 int addr_len, int flags)
573 struct sock *sk = sock->sk;
579 if (uaddr->sa_family == AF_UNSPEC) {
580 err = sk->sk_prot->disconnect(sk, flags);
581 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
585 switch (sock->state) {
594 /* Fall out of switch with err, set for this state */
598 if (sk->sk_state != TCP_CLOSE)
601 err = sk->sk_prot->connect(sk, uaddr, addr_len);
605 sock->state = SS_CONNECTING;
607 /* Just entered SS_CONNECTING state; the only
608 * difference is that return value in non-blocking
609 * case is EINPROGRESS, rather than EALREADY.
615 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
617 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
618 /* Error code is set above */
619 if (!timeo || !inet_wait_for_connect(sk, timeo))
622 err = sock_intr_errno(timeo);
623 if (signal_pending(current))
627 /* Connection was closed by RST, timeout, ICMP error
628 * or another process disconnected us.
630 if (sk->sk_state == TCP_CLOSE)
633 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
634 * and error was received after socket entered established state.
635 * Hence, it is handled normally after connect() return successfully.
638 sock->state = SS_CONNECTED;
645 err = sock_error(sk) ? : -ECONNABORTED;
646 sock->state = SS_UNCONNECTED;
647 if (sk->sk_prot->disconnect(sk, flags))
648 sock->state = SS_DISCONNECTING;
653 * Accept a pending connection. The TCP layer now gives BSD semantics.
656 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
658 struct sock *sk1 = sock->sk;
660 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
667 BUG_TRAP((1 << sk2->sk_state) &
668 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
670 sock_graft(sk2, newsock);
672 newsock->state = SS_CONNECTED;
681 * This does both peername and sockname.
683 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
684 int *uaddr_len, int peer)
686 struct sock *sk = sock->sk;
687 struct inet_sock *inet = inet_sk(sk);
688 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
690 sin->sin_family = AF_INET;
693 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
696 sin->sin_port = inet->dport;
697 sin->sin_addr.s_addr = inet->daddr;
699 __be32 addr = inet->rcv_saddr;
702 sin->sin_port = inet->sport;
703 sin->sin_addr.s_addr = addr;
705 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
706 *uaddr_len = sizeof(*sin);
710 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
713 struct sock *sk = sock->sk;
715 /* We may need to bind the socket. */
716 if (!inet_sk(sk)->num && inet_autobind(sk))
719 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
723 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
725 struct sock *sk = sock->sk;
727 /* We may need to bind the socket. */
728 if (!inet_sk(sk)->num && inet_autobind(sk))
731 if (sk->sk_prot->sendpage)
732 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
733 return sock_no_sendpage(sock, page, offset, size, flags);
737 int inet_shutdown(struct socket *sock, int how)
739 struct sock *sk = sock->sk;
742 /* This should really check to make sure
743 * the socket is a TCP socket. (WHY AC...)
745 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
748 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
752 if (sock->state == SS_CONNECTING) {
753 if ((1 << sk->sk_state) &
754 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
755 sock->state = SS_DISCONNECTING;
757 sock->state = SS_CONNECTED;
760 switch (sk->sk_state) {
763 /* Hack to wake up other listeners, who can poll for
764 POLLHUP, even on eg. unconnected UDP sockets -- RR */
766 sk->sk_shutdown |= how;
767 if (sk->sk_prot->shutdown)
768 sk->sk_prot->shutdown(sk, how);
771 /* Remaining two branches are temporary solution for missing
772 * close() in multithreaded environment. It is _not_ a good idea,
773 * but we have no choice until close() is repaired at VFS level.
776 if (!(how & RCV_SHUTDOWN))
780 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
781 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
785 /* Wake up anyone sleeping in poll. */
786 sk->sk_state_change(sk);
792 * ioctl() calls you can issue on an INET socket. Most of these are
793 * device configuration and stuff and very rarely used. Some ioctls
794 * pass on to the socket itself.
796 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
797 * loads the devconfigure module does its configuring and unloads it.
798 * There's a good 20K of config code hanging around the kernel.
801 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
803 struct sock *sk = sock->sk;
805 struct net *net = sock_net(sk);
809 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
812 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
817 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
822 err = arp_ioctl(net, cmd, (void __user *)arg);
835 err = devinet_ioctl(net, cmd, (void __user *)arg);
838 if (sk->sk_prot->ioctl)
839 err = sk->sk_prot->ioctl(sk, cmd, arg);
847 const struct proto_ops inet_stream_ops = {
849 .owner = THIS_MODULE,
850 .release = inet_release,
852 .connect = inet_stream_connect,
853 .socketpair = sock_no_socketpair,
854 .accept = inet_accept,
855 .getname = inet_getname,
858 .listen = inet_listen,
859 .shutdown = inet_shutdown,
860 .setsockopt = sock_common_setsockopt,
861 .getsockopt = sock_common_getsockopt,
862 .sendmsg = tcp_sendmsg,
863 .recvmsg = sock_common_recvmsg,
864 .mmap = sock_no_mmap,
865 .sendpage = tcp_sendpage,
866 .splice_read = tcp_splice_read,
868 .compat_setsockopt = compat_sock_common_setsockopt,
869 .compat_getsockopt = compat_sock_common_getsockopt,
873 const struct proto_ops inet_dgram_ops = {
875 .owner = THIS_MODULE,
876 .release = inet_release,
878 .connect = inet_dgram_connect,
879 .socketpair = sock_no_socketpair,
880 .accept = sock_no_accept,
881 .getname = inet_getname,
884 .listen = sock_no_listen,
885 .shutdown = inet_shutdown,
886 .setsockopt = sock_common_setsockopt,
887 .getsockopt = sock_common_getsockopt,
888 .sendmsg = inet_sendmsg,
889 .recvmsg = sock_common_recvmsg,
890 .mmap = sock_no_mmap,
891 .sendpage = inet_sendpage,
893 .compat_setsockopt = compat_sock_common_setsockopt,
894 .compat_getsockopt = compat_sock_common_getsockopt,
899 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
902 static const struct proto_ops inet_sockraw_ops = {
904 .owner = THIS_MODULE,
905 .release = inet_release,
907 .connect = inet_dgram_connect,
908 .socketpair = sock_no_socketpair,
909 .accept = sock_no_accept,
910 .getname = inet_getname,
911 .poll = datagram_poll,
913 .listen = sock_no_listen,
914 .shutdown = inet_shutdown,
915 .setsockopt = sock_common_setsockopt,
916 .getsockopt = sock_common_getsockopt,
917 .sendmsg = inet_sendmsg,
918 .recvmsg = sock_common_recvmsg,
919 .mmap = sock_no_mmap,
920 .sendpage = inet_sendpage,
922 .compat_setsockopt = compat_sock_common_setsockopt,
923 .compat_getsockopt = compat_sock_common_getsockopt,
927 static struct net_proto_family inet_family_ops = {
929 .create = inet_create,
930 .owner = THIS_MODULE,
933 /* Upon startup we insert all the elements in inetsw_array[] into
934 * the linked list inetsw.
936 static struct inet_protosw inetsw_array[] =
940 .protocol = IPPROTO_TCP,
942 .ops = &inet_stream_ops,
945 .flags = INET_PROTOSW_PERMANENT |
951 .protocol = IPPROTO_UDP,
953 .ops = &inet_dgram_ops,
955 .no_check = UDP_CSUM_DEFAULT,
956 .flags = INET_PROTOSW_PERMANENT,
962 .protocol = IPPROTO_IP, /* wild card */
964 .ops = &inet_sockraw_ops,
965 .capability = CAP_NET_RAW,
966 .no_check = UDP_CSUM_DEFAULT,
967 .flags = INET_PROTOSW_REUSE,
971 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
973 void inet_register_protosw(struct inet_protosw *p)
975 struct list_head *lh;
976 struct inet_protosw *answer;
977 int protocol = p->protocol;
978 struct list_head *last_perm;
980 spin_lock_bh(&inetsw_lock);
982 if (p->type >= SOCK_MAX)
985 /* If we are trying to override a permanent protocol, bail. */
987 last_perm = &inetsw[p->type];
988 list_for_each(lh, &inetsw[p->type]) {
989 answer = list_entry(lh, struct inet_protosw, list);
991 /* Check only the non-wild match. */
992 if (INET_PROTOSW_PERMANENT & answer->flags) {
993 if (protocol == answer->protocol)
1003 /* Add the new entry after the last permanent entry if any, so that
1004 * the new entry does not override a permanent entry when matched with
1005 * a wild-card protocol. But it is allowed to override any existing
1006 * non-permanent entry. This means that when we remove this entry, the
1007 * system automatically returns to the old behavior.
1009 list_add_rcu(&p->list, last_perm);
1011 spin_unlock_bh(&inetsw_lock);
1018 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1024 "Ignoring attempt to register invalid socket type %d.\n",
1029 void inet_unregister_protosw(struct inet_protosw *p)
1031 if (INET_PROTOSW_PERMANENT & p->flags) {
1033 "Attempt to unregister permanent protocol %d.\n",
1036 spin_lock_bh(&inetsw_lock);
1037 list_del_rcu(&p->list);
1038 spin_unlock_bh(&inetsw_lock);
1045 * Shall we try to damage output packets if routing dev changes?
1048 int sysctl_ip_dynaddr __read_mostly;
1050 static int inet_sk_reselect_saddr(struct sock *sk)
1052 struct inet_sock *inet = inet_sk(sk);
1055 __be32 old_saddr = inet->saddr;
1057 __be32 daddr = inet->daddr;
1059 if (inet->opt && inet->opt->srr)
1060 daddr = inet->opt->faddr;
1062 /* Query new route. */
1063 err = ip_route_connect(&rt, daddr, 0,
1065 sk->sk_bound_dev_if,
1067 inet->sport, inet->dport, sk, 0);
1071 sk_setup_caps(sk, &rt->u.dst);
1073 new_saddr = rt->rt_src;
1075 if (new_saddr == old_saddr)
1078 if (sysctl_ip_dynaddr > 1) {
1079 printk(KERN_INFO "%s(): shifting inet->"
1080 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1083 NIPQUAD(new_saddr));
1086 inet->saddr = inet->rcv_saddr = new_saddr;
1089 * XXX The only one ugly spot where we need to
1090 * XXX really change the sockets identity after
1091 * XXX it has entered the hashes. -DaveM
1093 * Besides that, it does not check for connection
1094 * uniqueness. Wait for troubles.
1096 __sk_prot_rehash(sk);
1100 int inet_sk_rebuild_header(struct sock *sk)
1102 struct inet_sock *inet = inet_sk(sk);
1103 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1107 /* Route is OK, nothing to do. */
1112 daddr = inet->daddr;
1113 if (inet->opt && inet->opt->srr)
1114 daddr = inet->opt->faddr;
1117 .oif = sk->sk_bound_dev_if,
1121 .saddr = inet->saddr,
1122 .tos = RT_CONN_FLAGS(sk),
1125 .proto = sk->sk_protocol,
1128 .sport = inet->sport,
1129 .dport = inet->dport,
1134 security_sk_classify_flow(sk, &fl);
1135 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1138 sk_setup_caps(sk, &rt->u.dst);
1140 /* Routing failed... */
1141 sk->sk_route_caps = 0;
1143 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1144 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1146 if (!sysctl_ip_dynaddr ||
1147 sk->sk_state != TCP_SYN_SENT ||
1148 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1149 (err = inet_sk_reselect_saddr(sk)) != 0)
1150 sk->sk_err_soft = -err;
1156 EXPORT_SYMBOL(inet_sk_rebuild_header);
1158 static int inet_gso_send_check(struct sk_buff *skb)
1161 struct net_protocol *ops;
1166 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1171 if (ihl < sizeof(*iph))
1174 if (unlikely(!pskb_may_pull(skb, ihl)))
1177 __skb_pull(skb, ihl);
1178 skb_reset_transport_header(skb);
1180 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1181 err = -EPROTONOSUPPORT;
1184 ops = rcu_dereference(inet_protos[proto]);
1185 if (likely(ops && ops->gso_send_check))
1186 err = ops->gso_send_check(skb);
1193 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1195 struct sk_buff *segs = ERR_PTR(-EINVAL);
1197 struct net_protocol *ops;
1202 if (!(features & NETIF_F_V4_CSUM))
1203 features &= ~NETIF_F_SG;
1205 if (unlikely(skb_shinfo(skb)->gso_type &
1213 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1218 if (ihl < sizeof(*iph))
1221 if (unlikely(!pskb_may_pull(skb, ihl)))
1224 __skb_pull(skb, ihl);
1225 skb_reset_transport_header(skb);
1227 id = ntohs(iph->id);
1228 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1229 segs = ERR_PTR(-EPROTONOSUPPORT);
1232 ops = rcu_dereference(inet_protos[proto]);
1233 if (likely(ops && ops->gso_segment))
1234 segs = ops->gso_segment(skb, features);
1237 if (!segs || unlikely(IS_ERR(segs)))
1243 iph->id = htons(id++);
1244 iph->tot_len = htons(skb->len - skb->mac_len);
1246 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1247 } while ((skb = skb->next));
1253 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1254 unsigned short type, unsigned char protocol,
1257 struct socket *sock;
1258 int rc = sock_create_kern(family, type, protocol, &sock);
1262 (*sk)->sk_allocation = GFP_ATOMIC;
1264 * Unhash it so that IP input processing does not even see it,
1265 * we do not wish this socket to see incoming packets.
1267 (*sk)->sk_prot->unhash(*sk);
1269 sk_change_net(*sk, net);
1274 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1276 unsigned long snmp_fold_field(void *mib[], int offt)
1278 unsigned long res = 0;
1281 for_each_possible_cpu(i) {
1282 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1283 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1287 EXPORT_SYMBOL_GPL(snmp_fold_field);
1289 int snmp_mib_init(void *ptr[2], size_t mibsize)
1291 BUG_ON(ptr == NULL);
1292 ptr[0] = __alloc_percpu(mibsize);
1295 ptr[1] = __alloc_percpu(mibsize);
1300 free_percpu(ptr[0]);
1305 EXPORT_SYMBOL_GPL(snmp_mib_init);
1307 void snmp_mib_free(void *ptr[2])
1309 BUG_ON(ptr == NULL);
1310 free_percpu(ptr[0]);
1311 free_percpu(ptr[1]);
1312 ptr[0] = ptr[1] = NULL;
1314 EXPORT_SYMBOL_GPL(snmp_mib_free);
1316 #ifdef CONFIG_IP_MULTICAST
1317 static struct net_protocol igmp_protocol = {
1318 .handler = igmp_rcv,
1322 static struct net_protocol tcp_protocol = {
1323 .handler = tcp_v4_rcv,
1324 .err_handler = tcp_v4_err,
1325 .gso_send_check = tcp_v4_gso_send_check,
1326 .gso_segment = tcp_tso_segment,
1331 static struct net_protocol udp_protocol = {
1333 .err_handler = udp_err,
1338 static struct net_protocol icmp_protocol = {
1339 .handler = icmp_rcv,
1344 static int __init init_ipv4_mibs(void)
1346 if (snmp_mib_init((void **)net_statistics,
1347 sizeof(struct linux_mib)) < 0)
1349 if (snmp_mib_init((void **)ip_statistics,
1350 sizeof(struct ipstats_mib)) < 0)
1352 if (snmp_mib_init((void **)icmp_statistics,
1353 sizeof(struct icmp_mib)) < 0)
1355 if (snmp_mib_init((void **)icmpmsg_statistics,
1356 sizeof(struct icmpmsg_mib)) < 0)
1357 goto err_icmpmsg_mib;
1358 if (snmp_mib_init((void **)tcp_statistics,
1359 sizeof(struct tcp_mib)) < 0)
1361 if (snmp_mib_init((void **)udp_statistics,
1362 sizeof(struct udp_mib)) < 0)
1364 if (snmp_mib_init((void **)udplite_statistics,
1365 sizeof(struct udp_mib)) < 0)
1366 goto err_udplite_mib;
1373 snmp_mib_free((void **)udp_statistics);
1375 snmp_mib_free((void **)tcp_statistics);
1377 snmp_mib_free((void **)icmpmsg_statistics);
1379 snmp_mib_free((void **)icmp_statistics);
1381 snmp_mib_free((void **)ip_statistics);
1383 snmp_mib_free((void **)net_statistics);
1388 static int ipv4_proc_init(void);
1391 * IP protocol layer initialiser
1394 static struct packet_type ip_packet_type = {
1395 .type = __constant_htons(ETH_P_IP),
1397 .gso_send_check = inet_gso_send_check,
1398 .gso_segment = inet_gso_segment,
1401 static int __init inet_init(void)
1403 struct sk_buff *dummy_skb;
1404 struct inet_protosw *q;
1405 struct list_head *r;
1408 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1410 rc = proto_register(&tcp_prot, 1);
1414 rc = proto_register(&udp_prot, 1);
1416 goto out_unregister_tcp_proto;
1418 rc = proto_register(&raw_prot, 1);
1420 goto out_unregister_udp_proto;
1423 * Tell SOCKET that we are alive...
1426 (void)sock_register(&inet_family_ops);
1429 * Add all the base protocols.
1432 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1433 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1434 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1435 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1436 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1437 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1438 #ifdef CONFIG_IP_MULTICAST
1439 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1440 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1443 /* Register the socket-side information for inet_create. */
1444 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1447 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1448 inet_register_protosw(q);
1451 * Set the ARP module up
1457 * Set the IP module up
1464 /* Setup TCP slab cache for open requests. */
1467 /* Setup UDP memory threshold */
1470 /* Add UDP-Lite (RFC 3828) */
1471 udplite4_register();
1474 * Set the ICMP layer up
1477 if (icmp_init() < 0)
1478 panic("Failed to create the ICMP control socket.\n");
1481 * Initialise the multicast router
1483 #if defined(CONFIG_IP_MROUTE)
1487 * Initialise per-cpu ipv4 mibs
1490 if (init_ipv4_mibs())
1491 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1497 dev_add_pack(&ip_packet_type);
1502 out_unregister_udp_proto:
1503 proto_unregister(&udp_prot);
1504 out_unregister_tcp_proto:
1505 proto_unregister(&tcp_prot);
1509 fs_initcall(inet_init);
1511 /* ------------------------------------------------------------------------ */
1513 #ifdef CONFIG_PROC_FS
1514 static int __init ipv4_proc_init(void)
1518 if (raw_proc_init())
1520 if (tcp4_proc_init())
1522 if (udp4_proc_init())
1524 if (ip_misc_proc_init())
1539 #else /* CONFIG_PROC_FS */
1540 static int __init ipv4_proc_init(void)
1544 #endif /* CONFIG_PROC_FS */
1546 MODULE_ALIAS_NETPROTO(PF_INET);
1548 EXPORT_SYMBOL(inet_accept);
1549 EXPORT_SYMBOL(inet_bind);
1550 EXPORT_SYMBOL(inet_dgram_connect);
1551 EXPORT_SYMBOL(inet_dgram_ops);
1552 EXPORT_SYMBOL(inet_getname);
1553 EXPORT_SYMBOL(inet_ioctl);
1554 EXPORT_SYMBOL(inet_listen);
1555 EXPORT_SYMBOL(inet_register_protosw);
1556 EXPORT_SYMBOL(inet_release);
1557 EXPORT_SYMBOL(inet_sendmsg);
1558 EXPORT_SYMBOL(inet_shutdown);
1559 EXPORT_SYMBOL(inet_sock_destruct);
1560 EXPORT_SYMBOL(inet_stream_connect);
1561 EXPORT_SYMBOL(inet_stream_ops);
1562 EXPORT_SYMBOL(inet_unregister_protosw);
1563 EXPORT_SYMBOL(net_statistics);
1564 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);