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>
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
94 #include <linux/smp_lock.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>
118 #include <linux/vs_limit.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;
222 * Create an inet socket.
225 static int inet_create(struct socket *sock, int protocol)
229 struct inet_protosw *answer;
230 struct inet_sock *inet;
231 struct proto *answer_prot;
232 unsigned char answer_flags;
233 char answer_no_check;
234 int try_loading_module = 0;
237 sock->state = SS_UNCONNECTED;
239 /* Look for the requested type/protocol pair. */
242 err = -ESOCKTNOSUPPORT;
244 list_for_each_rcu(p, &inetsw[sock->type]) {
245 answer = list_entry(p, struct inet_protosw, list);
247 /* Check the non-wild match. */
248 if (protocol == answer->protocol) {
249 if (protocol != IPPROTO_IP)
252 /* Check for the two wild cases. */
253 if (IPPROTO_IP == protocol) {
254 protocol = answer->protocol;
257 if (IPPROTO_IP == answer->protocol)
260 err = -EPROTONOSUPPORT;
264 if (unlikely(answer == NULL)) {
265 if (try_loading_module < 2) {
268 * Be more specific, e.g. net-pf-2-proto-132-type-1
269 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
271 if (++try_loading_module == 1)
272 request_module("net-pf-%d-proto-%d-type-%d",
273 PF_INET, protocol, sock->type);
275 * Fall back to generic, e.g. net-pf-2-proto-132
276 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
279 request_module("net-pf-%d-proto-%d",
281 goto lookup_protocol;
287 if ((protocol == IPPROTO_ICMP) && vx_ccaps(VXC_RAW_ICMP))
289 if (answer->capability > 0 && !capable(answer->capability))
292 sock->ops = answer->ops;
293 answer_prot = answer->prot;
294 answer_no_check = answer->no_check;
295 answer_flags = answer->flags;
298 BUG_TRAP(answer_prot->slab != NULL);
301 sk = sk_alloc(PF_INET, GFP_KERNEL, answer_prot, 1);
306 sk->sk_no_check = answer_no_check;
307 if (INET_PROTOSW_REUSE & answer_flags)
311 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
313 if (SOCK_RAW == sock->type) {
314 inet->num = protocol;
315 if (IPPROTO_RAW == protocol)
319 if (ipv4_config.no_pmtu_disc)
320 inet->pmtudisc = IP_PMTUDISC_DONT;
322 inet->pmtudisc = IP_PMTUDISC_WANT;
326 sock_init_data(sock, sk);
328 sk->sk_destruct = inet_sock_destruct;
329 sk->sk_family = PF_INET;
330 sk->sk_protocol = protocol;
331 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
337 inet->mc_list = NULL;
339 sk_refcnt_debug_inc(sk);
342 /* It assumes that any protocol which allows
343 * the user to assign a number at socket
344 * creation time automatically
347 inet->sport = htons(inet->num);
348 /* Add to protocol hash chains. */
349 sk->sk_prot->hash(sk);
352 if (sk->sk_prot->init) {
353 err = sk->sk_prot->init(sk);
355 sk_common_release(sk);
366 * The peer socket should always be NULL (or else). When we call this
367 * function we are destroying the object and from then on nobody
368 * should refer to it.
370 int inet_release(struct socket *sock)
372 struct sock *sk = sock->sk;
377 /* Applications forget to leave groups before exiting */
378 ip_mc_drop_socket(sk);
380 /* If linger is set, we don't return until the close
381 * is complete. Otherwise we return immediately. The
382 * actually closing is done the same either way.
384 * If the close is due to the process exiting, we never
388 if (sock_flag(sk, SOCK_LINGER) &&
389 !(current->flags & PF_EXITING))
390 timeout = sk->sk_lingertime;
392 sk->sk_prot->close(sk, timeout);
397 /* It is off by default, see below. */
398 int sysctl_ip_nonlocal_bind __read_mostly;
400 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
402 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
403 struct sock *sk = sock->sk;
404 struct inet_sock *inet = inet_sk(sk);
408 __u32 s_addr; /* Address used for validation */
409 __u32 s_addr1; /* Address used for socket */
410 __u32 s_addr2; /* Broadcast address for the socket */
411 struct nx_info *nxi = sk->sk_nx_info;
413 /* If the socket has its own bind function then use it. (RAW) */
414 if (sk->sk_prot->bind) {
415 err = sk->sk_prot->bind(sk, uaddr, addr_len);
419 if (addr_len < sizeof(struct sockaddr_in))
422 s_addr = addr->sin_addr.s_addr;
424 s_addr2 = 0xffffffffl;
426 vxdprintk(VXD_CBIT(net, 3),
427 "inet_bind(%p)* %p,%p;%lx %d.%d.%d.%d",
428 sk, sk->sk_nx_info, sk->sk_socket,
429 (sk->sk_socket?sk->sk_socket->flags:0),
432 __u32 v4_bcast = nxi->v4_bcast;
433 __u32 ipv4root = nxi->ipv4[0];
434 int nbipv4 = nxi->nbipv4;
437 /* bind to any for 1-n */
439 s_addr1 = (nbipv4 > 1) ? 0 : s_addr;
441 } else if (s_addr == IPI_LOOPBACK) {
442 /* rewrite localhost to ipv4root */
445 } else if (s_addr != v4_bcast) {
446 /* normal address bind */
447 if (!addr_in_nx_info(nxi, s_addr))
448 return -EADDRNOTAVAIL;
451 chk_addr_ret = inet_addr_type(s_addr);
453 vxdprintk(VXD_CBIT(net, 3),
454 "inet_bind(%p) %d.%d.%d.%d, %d.%d.%d.%d, %d.%d.%d.%d",
455 sk, VXD_QUAD(s_addr), VXD_QUAD(s_addr1), VXD_QUAD(s_addr2));
457 /* Not specified by any standard per-se, however it breaks too
458 * many applications when removed. It is unfortunate since
459 * allowing applications to make a non-local bind solves
460 * several problems with systems using dynamic addressing.
461 * (ie. your servers still start up even if your ISDN link
462 * is temporarily down)
464 err = -EADDRNOTAVAIL;
465 if (!sysctl_ip_nonlocal_bind &&
467 s_addr != INADDR_ANY &&
468 chk_addr_ret != RTN_LOCAL &&
469 chk_addr_ret != RTN_MULTICAST &&
470 chk_addr_ret != RTN_BROADCAST)
473 snum = ntohs(addr->sin_port);
475 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
478 /* We keep a pair of addresses. rcv_saddr is the one
479 * used by hash lookups, and saddr is used for transmit.
481 * In the BSD API these are the same except where it
482 * would be illegal to use them (multicast/broadcast) in
483 * which case the sending device address is used.
487 /* Check these errors (active socket, double bind). */
489 if (sk->sk_state != TCP_CLOSE || inet->num)
490 goto out_release_sock;
492 inet->rcv_saddr = inet->saddr = s_addr1;
493 inet->rcv_saddr2 = s_addr2;
494 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
495 inet->saddr = 0; /* Use device */
497 /* Make sure we are allowed to bind here. */
498 if (sk->sk_prot->get_port(sk, snum)) {
499 inet->saddr = inet->rcv_saddr = 0;
501 goto out_release_sock;
505 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
507 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
508 inet->sport = htons(inet->num);
519 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
520 int addr_len, int flags)
522 struct sock *sk = sock->sk;
524 if (uaddr->sa_family == AF_UNSPEC)
525 return sk->sk_prot->disconnect(sk, flags);
527 if (!inet_sk(sk)->num && inet_autobind(sk))
529 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
532 static long inet_wait_for_connect(struct sock *sk, long timeo)
536 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
538 /* Basic assumption: if someone sets sk->sk_err, he _must_
539 * change state of the socket from TCP_SYN_*.
540 * Connect() does not allow to get error notifications
541 * without closing the socket.
543 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
545 timeo = schedule_timeout(timeo);
547 if (signal_pending(current) || !timeo)
549 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
551 finish_wait(sk->sk_sleep, &wait);
556 * Connect to a remote host. There is regrettably still a little
557 * TCP 'magic' in here.
559 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
560 int addr_len, int flags)
562 struct sock *sk = sock->sk;
568 if (uaddr->sa_family == AF_UNSPEC) {
569 err = sk->sk_prot->disconnect(sk, flags);
570 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
574 switch (sock->state) {
583 /* Fall out of switch with err, set for this state */
587 if (sk->sk_state != TCP_CLOSE)
590 err = sk->sk_prot->connect(sk, uaddr, addr_len);
594 sock->state = SS_CONNECTING;
596 /* Just entered SS_CONNECTING state; the only
597 * difference is that return value in non-blocking
598 * case is EINPROGRESS, rather than EALREADY.
604 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
606 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
607 /* Error code is set above */
608 if (!timeo || !inet_wait_for_connect(sk, timeo))
611 err = sock_intr_errno(timeo);
612 if (signal_pending(current))
616 /* Connection was closed by RST, timeout, ICMP error
617 * or another process disconnected us.
619 if (sk->sk_state == TCP_CLOSE)
622 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
623 * and error was received after socket entered established state.
624 * Hence, it is handled normally after connect() return successfully.
627 sock->state = SS_CONNECTED;
634 err = sock_error(sk) ? : -ECONNABORTED;
635 sock->state = SS_UNCONNECTED;
636 if (sk->sk_prot->disconnect(sk, flags))
637 sock->state = SS_DISCONNECTING;
642 * Accept a pending connection. The TCP layer now gives BSD semantics.
645 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
647 struct sock *sk1 = sock->sk;
649 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
656 BUG_TRAP((1 << sk2->sk_state) &
657 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
659 sock_graft(sk2, newsock);
661 newsock->state = SS_CONNECTED;
670 * This does both peername and sockname.
672 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
673 int *uaddr_len, int peer)
675 struct sock *sk = sock->sk;
676 struct inet_sock *inet = inet_sk(sk);
677 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
679 sin->sin_family = AF_INET;
682 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
685 sin->sin_port = inet->dport;
686 sin->sin_addr.s_addr = inet->daddr;
688 __be32 addr = inet->rcv_saddr;
691 sin->sin_port = inet->sport;
692 sin->sin_addr.s_addr = addr;
694 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
695 *uaddr_len = sizeof(*sin);
699 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
702 struct sock *sk = sock->sk;
704 /* We may need to bind the socket. */
705 if (!inet_sk(sk)->num && inet_autobind(sk))
708 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
712 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
714 struct sock *sk = sock->sk;
716 /* We may need to bind the socket. */
717 if (!inet_sk(sk)->num && inet_autobind(sk))
720 if (sk->sk_prot->sendpage)
721 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
722 return sock_no_sendpage(sock, page, offset, size, flags);
726 int inet_shutdown(struct socket *sock, int how)
728 struct sock *sk = sock->sk;
731 /* This should really check to make sure
732 * the socket is a TCP socket. (WHY AC...)
734 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
737 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
741 if (sock->state == SS_CONNECTING) {
742 if ((1 << sk->sk_state) &
743 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
744 sock->state = SS_DISCONNECTING;
746 sock->state = SS_CONNECTED;
749 switch (sk->sk_state) {
752 /* Hack to wake up other listeners, who can poll for
753 POLLHUP, even on eg. unconnected UDP sockets -- RR */
755 sk->sk_shutdown |= how;
756 if (sk->sk_prot->shutdown)
757 sk->sk_prot->shutdown(sk, how);
760 /* Remaining two branches are temporary solution for missing
761 * close() in multithreaded environment. It is _not_ a good idea,
762 * but we have no choice until close() is repaired at VFS level.
765 if (!(how & RCV_SHUTDOWN))
769 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
770 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
774 /* Wake up anyone sleeping in poll. */
775 sk->sk_state_change(sk);
781 * ioctl() calls you can issue on an INET socket. Most of these are
782 * device configuration and stuff and very rarely used. Some ioctls
783 * pass on to the socket itself.
785 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
786 * loads the devconfigure module does its configuring and unloads it.
787 * There's a good 20K of config code hanging around the kernel.
790 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
792 struct sock *sk = sock->sk;
797 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
802 err = ip_rt_ioctl(cmd, (void __user *)arg);
807 err = arp_ioctl(cmd, (void __user *)arg);
820 err = devinet_ioctl(cmd, (void __user *)arg);
823 if (sk->sk_prot->ioctl)
824 err = sk->sk_prot->ioctl(sk, cmd, arg);
832 const struct proto_ops inet_stream_ops = {
834 .owner = THIS_MODULE,
835 .release = inet_release,
837 .connect = inet_stream_connect,
838 .socketpair = sock_no_socketpair,
839 .accept = inet_accept,
840 .getname = inet_getname,
843 .listen = inet_listen,
844 .shutdown = inet_shutdown,
845 .setsockopt = sock_common_setsockopt,
846 .getsockopt = sock_common_getsockopt,
847 .sendmsg = inet_sendmsg,
848 .recvmsg = sock_common_recvmsg,
849 .mmap = sock_no_mmap,
850 .sendpage = tcp_sendpage,
852 .compat_setsockopt = compat_sock_common_setsockopt,
853 .compat_getsockopt = compat_sock_common_getsockopt,
857 const struct proto_ops inet_dgram_ops = {
859 .owner = THIS_MODULE,
860 .release = inet_release,
862 .connect = inet_dgram_connect,
863 .socketpair = sock_no_socketpair,
864 .accept = sock_no_accept,
865 .getname = inet_getname,
868 .listen = sock_no_listen,
869 .shutdown = inet_shutdown,
870 .setsockopt = sock_common_setsockopt,
871 .getsockopt = sock_common_getsockopt,
872 .sendmsg = inet_sendmsg,
873 .recvmsg = sock_common_recvmsg,
874 .mmap = sock_no_mmap,
875 .sendpage = inet_sendpage,
877 .compat_setsockopt = compat_sock_common_setsockopt,
878 .compat_getsockopt = compat_sock_common_getsockopt,
883 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
886 static const struct proto_ops inet_sockraw_ops = {
888 .owner = THIS_MODULE,
889 .release = inet_release,
891 .connect = inet_dgram_connect,
892 .socketpair = sock_no_socketpair,
893 .accept = sock_no_accept,
894 .getname = inet_getname,
895 .poll = datagram_poll,
897 .listen = sock_no_listen,
898 .shutdown = inet_shutdown,
899 .setsockopt = sock_common_setsockopt,
900 .getsockopt = sock_common_getsockopt,
901 .sendmsg = inet_sendmsg,
902 .recvmsg = sock_common_recvmsg,
903 .mmap = sock_no_mmap,
904 .sendpage = inet_sendpage,
906 .compat_setsockopt = compat_sock_common_setsockopt,
907 .compat_getsockopt = compat_sock_common_getsockopt,
911 static struct net_proto_family inet_family_ops = {
913 .create = inet_create,
914 .owner = THIS_MODULE,
917 /* Upon startup we insert all the elements in inetsw_array[] into
918 * the linked list inetsw.
920 static struct inet_protosw inetsw_array[] =
924 .protocol = IPPROTO_TCP,
926 .ops = &inet_stream_ops,
929 .flags = INET_PROTOSW_PERMANENT |
935 .protocol = IPPROTO_UDP,
937 .ops = &inet_dgram_ops,
939 .no_check = UDP_CSUM_DEFAULT,
940 .flags = INET_PROTOSW_PERMANENT,
946 .protocol = IPPROTO_IP, /* wild card */
948 .ops = &inet_sockraw_ops,
949 .capability = CAP_NET_RAW,
950 .no_check = UDP_CSUM_DEFAULT,
951 .flags = INET_PROTOSW_REUSE,
955 #define INETSW_ARRAY_LEN (sizeof(inetsw_array) / sizeof(struct inet_protosw))
957 void inet_register_protosw(struct inet_protosw *p)
959 struct list_head *lh;
960 struct inet_protosw *answer;
961 int protocol = p->protocol;
962 struct list_head *last_perm;
964 spin_lock_bh(&inetsw_lock);
966 if (p->type >= SOCK_MAX)
969 /* If we are trying to override a permanent protocol, bail. */
971 last_perm = &inetsw[p->type];
972 list_for_each(lh, &inetsw[p->type]) {
973 answer = list_entry(lh, struct inet_protosw, list);
975 /* Check only the non-wild match. */
976 if (INET_PROTOSW_PERMANENT & answer->flags) {
977 if (protocol == answer->protocol)
987 /* Add the new entry after the last permanent entry if any, so that
988 * the new entry does not override a permanent entry when matched with
989 * a wild-card protocol. But it is allowed to override any existing
990 * non-permanent entry. This means that when we remove this entry, the
991 * system automatically returns to the old behavior.
993 list_add_rcu(&p->list, last_perm);
995 spin_unlock_bh(&inetsw_lock);
1002 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1008 "Ignoring attempt to register invalid socket type %d.\n",
1013 void inet_unregister_protosw(struct inet_protosw *p)
1015 if (INET_PROTOSW_PERMANENT & p->flags) {
1017 "Attempt to unregister permanent protocol %d.\n",
1020 spin_lock_bh(&inetsw_lock);
1021 list_del_rcu(&p->list);
1022 spin_unlock_bh(&inetsw_lock);
1029 * Shall we try to damage output packets if routing dev changes?
1032 int sysctl_ip_dynaddr __read_mostly;
1034 static int inet_sk_reselect_saddr(struct sock *sk)
1036 struct inet_sock *inet = inet_sk(sk);
1039 __be32 old_saddr = inet->saddr;
1041 __be32 daddr = inet->daddr;
1043 if (inet->opt && inet->opt->srr)
1044 daddr = inet->opt->faddr;
1046 /* Query new route. */
1047 err = ip_route_connect(&rt, daddr, 0,
1049 sk->sk_bound_dev_if,
1051 inet->sport, inet->dport, sk);
1055 sk_setup_caps(sk, &rt->u.dst);
1057 new_saddr = rt->rt_src;
1059 if (new_saddr == old_saddr)
1062 if (sysctl_ip_dynaddr > 1) {
1063 printk(KERN_INFO "%s(): shifting inet->"
1064 "saddr from %d.%d.%d.%d to %d.%d.%d.%d\n",
1067 NIPQUAD(new_saddr));
1070 inet->saddr = inet->rcv_saddr = new_saddr;
1073 * XXX The only one ugly spot where we need to
1074 * XXX really change the sockets identity after
1075 * XXX it has entered the hashes. -DaveM
1077 * Besides that, it does not check for connection
1078 * uniqueness. Wait for troubles.
1080 __sk_prot_rehash(sk);
1084 int inet_sk_rebuild_header(struct sock *sk)
1086 struct inet_sock *inet = inet_sk(sk);
1087 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1091 /* Route is OK, nothing to do. */
1096 daddr = inet->daddr;
1097 if (inet->opt && inet->opt->srr)
1098 daddr = inet->opt->faddr;
1101 .oif = sk->sk_bound_dev_if,
1105 .saddr = inet->saddr,
1106 .tos = RT_CONN_FLAGS(sk),
1109 .proto = sk->sk_protocol,
1112 .sport = inet->sport,
1113 .dport = inet->dport,
1118 security_sk_classify_flow(sk, &fl);
1119 err = ip_route_output_flow(&rt, &fl, sk, 0);
1122 sk_setup_caps(sk, &rt->u.dst);
1124 /* Routing failed... */
1125 sk->sk_route_caps = 0;
1127 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1128 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1130 if (!sysctl_ip_dynaddr ||
1131 sk->sk_state != TCP_SYN_SENT ||
1132 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1133 (err = inet_sk_reselect_saddr(sk)) != 0)
1134 sk->sk_err_soft = -err;
1140 EXPORT_SYMBOL(inet_sk_rebuild_header);
1142 static int inet_gso_send_check(struct sk_buff *skb)
1145 struct net_protocol *ops;
1150 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1155 if (ihl < sizeof(*iph))
1158 if (unlikely(!pskb_may_pull(skb, ihl)))
1161 skb->h.raw = __skb_pull(skb, ihl);
1163 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1164 err = -EPROTONOSUPPORT;
1167 ops = rcu_dereference(inet_protos[proto]);
1168 if (likely(ops && ops->gso_send_check))
1169 err = ops->gso_send_check(skb);
1176 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1178 struct sk_buff *segs = ERR_PTR(-EINVAL);
1180 struct net_protocol *ops;
1185 if (unlikely(skb_shinfo(skb)->gso_type &
1193 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1198 if (ihl < sizeof(*iph))
1201 if (unlikely(!pskb_may_pull(skb, ihl)))
1204 skb->h.raw = __skb_pull(skb, ihl);
1206 id = ntohs(iph->id);
1207 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1208 segs = ERR_PTR(-EPROTONOSUPPORT);
1211 ops = rcu_dereference(inet_protos[proto]);
1212 if (likely(ops && ops->gso_segment))
1213 segs = ops->gso_segment(skb, features);
1216 if (!segs || unlikely(IS_ERR(segs)))
1222 iph->id = htons(id++);
1223 iph->tot_len = htons(skb->len - skb->mac_len);
1225 iph->check = ip_fast_csum(skb->nh.raw, iph->ihl);
1226 } while ((skb = skb->next));
1232 #ifdef CONFIG_IP_MULTICAST
1233 static struct net_protocol igmp_protocol = {
1234 .handler = igmp_rcv,
1238 static struct net_protocol tcp_protocol = {
1239 .handler = tcp_v4_rcv,
1240 .err_handler = tcp_v4_err,
1241 .gso_send_check = tcp_v4_gso_send_check,
1242 .gso_segment = tcp_tso_segment,
1246 static struct net_protocol udp_protocol = {
1248 .err_handler = udp_err,
1252 static struct net_protocol icmp_protocol = {
1253 .handler = icmp_rcv,
1256 static int __init init_ipv4_mibs(void)
1258 net_statistics[0] = alloc_percpu(struct linux_mib);
1259 net_statistics[1] = alloc_percpu(struct linux_mib);
1260 ip_statistics[0] = alloc_percpu(struct ipstats_mib);
1261 ip_statistics[1] = alloc_percpu(struct ipstats_mib);
1262 icmp_statistics[0] = alloc_percpu(struct icmp_mib);
1263 icmp_statistics[1] = alloc_percpu(struct icmp_mib);
1264 tcp_statistics[0] = alloc_percpu(struct tcp_mib);
1265 tcp_statistics[1] = alloc_percpu(struct tcp_mib);
1266 udp_statistics[0] = alloc_percpu(struct udp_mib);
1267 udp_statistics[1] = alloc_percpu(struct udp_mib);
1268 udplite_statistics[0] = alloc_percpu(struct udp_mib);
1269 udplite_statistics[1] = alloc_percpu(struct udp_mib);
1271 (net_statistics[0] && net_statistics[1] && ip_statistics[0]
1272 && ip_statistics[1] && tcp_statistics[0] && tcp_statistics[1]
1273 && udp_statistics[0] && udp_statistics[1]
1274 && udplite_statistics[0] && udplite_statistics[1] ) )
1277 (void) tcp_mib_init();
1282 static int ipv4_proc_init(void);
1285 * IP protocol layer initialiser
1288 static struct packet_type ip_packet_type = {
1289 .type = __constant_htons(ETH_P_IP),
1291 .gso_send_check = inet_gso_send_check,
1292 .gso_segment = inet_gso_segment,
1295 static int __init inet_init(void)
1297 struct sk_buff *dummy_skb;
1298 struct inet_protosw *q;
1299 struct list_head *r;
1302 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1304 rc = proto_register(&tcp_prot, 1);
1308 rc = proto_register(&udp_prot, 1);
1310 goto out_unregister_tcp_proto;
1312 rc = proto_register(&raw_prot, 1);
1314 goto out_unregister_udp_proto;
1317 * Tell SOCKET that we are alive...
1320 (void)sock_register(&inet_family_ops);
1323 * Add all the base protocols.
1326 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1327 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1328 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1329 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1330 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1331 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1332 #ifdef CONFIG_IP_MULTICAST
1333 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1334 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1337 /* Register the socket-side information for inet_create. */
1338 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1341 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1342 inet_register_protosw(q);
1345 * Set the ARP module up
1351 * Set the IP module up
1356 tcp_v4_init(&inet_family_ops);
1358 /* Setup TCP slab cache for open requests. */
1361 /* Add UDP-Lite (RFC 3828) */
1362 udplite4_register();
1365 * Set the ICMP layer up
1368 icmp_init(&inet_family_ops);
1371 * Initialise the multicast router
1373 #if defined(CONFIG_IP_MROUTE)
1377 * Initialise per-cpu ipv4 mibs
1380 if(init_ipv4_mibs())
1381 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n"); ;
1387 dev_add_pack(&ip_packet_type);
1392 out_unregister_udp_proto:
1393 proto_unregister(&udp_prot);
1394 out_unregister_tcp_proto:
1395 proto_unregister(&tcp_prot);
1399 fs_initcall(inet_init);
1401 /* ------------------------------------------------------------------------ */
1403 #ifdef CONFIG_PROC_FS
1404 static int __init ipv4_proc_init(void)
1408 if (raw_proc_init())
1410 if (tcp4_proc_init())
1412 if (udp4_proc_init())
1414 if (fib_proc_init())
1416 if (ip_misc_proc_init())
1433 #else /* CONFIG_PROC_FS */
1434 static int __init ipv4_proc_init(void)
1438 #endif /* CONFIG_PROC_FS */
1440 MODULE_ALIAS_NETPROTO(PF_INET);
1442 EXPORT_SYMBOL(inet_accept);
1443 EXPORT_SYMBOL(inet_bind);
1444 EXPORT_SYMBOL(inet_dgram_connect);
1445 EXPORT_SYMBOL(inet_dgram_ops);
1446 EXPORT_SYMBOL(inet_getname);
1447 EXPORT_SYMBOL(inet_ioctl);
1448 EXPORT_SYMBOL(inet_listen);
1449 EXPORT_SYMBOL(inet_register_protosw);
1450 EXPORT_SYMBOL(inet_release);
1451 EXPORT_SYMBOL(inet_sendmsg);
1452 EXPORT_SYMBOL(inet_shutdown);
1453 EXPORT_SYMBOL(inet_sock_destruct);
1454 EXPORT_SYMBOL(inet_stream_connect);
1455 EXPORT_SYMBOL(inet_stream_ops);
1456 EXPORT_SYMBOL(inet_unregister_protosw);
1457 EXPORT_SYMBOL(net_statistics);
1458 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);