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.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
13 * Changes (see also sock.c)
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
68 #include <linux/err.h>
69 #include <linux/errno.h>
70 #include <linux/types.h>
71 #include <linux/socket.h>
73 #include <linux/kernel.h>
74 #include <linux/module.h>
75 #include <linux/sched.h>
76 #include <linux/timer.h>
77 #include <linux/string.h>
78 #include <linux/sockios.h>
79 #include <linux/net.h>
80 #include <linux/capability.h>
81 #include <linux/fcntl.h>
83 #include <linux/interrupt.h>
84 #include <linux/stat.h>
85 #include <linux/init.h>
86 #include <linux/poll.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
90 #include <asm/uaccess.h>
91 #include <asm/system.h>
93 #include <linux/inet.h>
94 #include <linux/igmp.h>
95 #include <linux/inetdevice.h>
96 #include <linux/netdevice.h>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/ip_fib.h>
102 #include <net/inet_connection_sock.h>
105 #include <net/udplite.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
109 #include <net/icmp.h>
110 #include <net/ipip.h>
111 #include <net/inet_common.h>
112 #include <net/xfrm.h>
113 #ifdef CONFIG_IP_MROUTE
114 #include <linux/mroute.h>
117 DEFINE_SNMP_STAT(struct linux_mib, net_statistics) __read_mostly;
119 extern void ip_mc_drop_socket(struct sock *sk);
121 /* The inetsw table contains everything that inet_create needs to
122 * build a new socket.
124 static struct list_head inetsw[SOCK_MAX];
125 static DEFINE_SPINLOCK(inetsw_lock);
127 struct ipv4_config ipv4_config;
129 EXPORT_SYMBOL(ipv4_config);
131 /* New destruction routine */
133 void inet_sock_destruct(struct sock *sk)
135 struct inet_sock *inet = inet_sk(sk);
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 printk("Attempt to release TCP socket in state %d %p\n",
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 printk("Attempt to release alive inet socket %p\n", sk);
152 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
153 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
154 BUG_TRAP(!sk->sk_wmem_queued);
155 BUG_TRAP(!sk->sk_forward_alloc);
158 dst_release(sk->sk_dst_cache);
159 sk_refcnt_debug_dec(sk);
163 * The routines beyond this point handle the behaviour of an AF_INET
164 * socket object. Mostly it punts to the subprotocols of IP to do
169 * Automatically bind an unbound socket.
172 static int inet_autobind(struct sock *sk)
174 struct inet_sock *inet;
175 /* We may need to bind the socket. */
179 if (sk->sk_prot->get_port(sk, 0)) {
183 inet->sport = htons(inet->num);
190 * Move a socket into listening state.
192 int inet_listen(struct socket *sock, int backlog)
194 struct sock *sk = sock->sk;
195 unsigned char old_state;
201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 old_state = sk->sk_state;
205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 /* Really, if the socket is already in listen state
209 * we can only allow the backlog to be adjusted.
211 if (old_state != TCP_LISTEN) {
212 err = inet_csk_listen_start(sk, backlog);
216 sk->sk_max_ack_backlog = backlog;
224 u32 inet_ehash_secret __read_mostly;
225 EXPORT_SYMBOL(inet_ehash_secret);
228 * inet_ehash_secret must be set exactly once
229 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
231 void build_ehash_secret(void)
235 get_random_bytes(&rnd, sizeof(rnd));
237 spin_lock_bh(&inetsw_lock);
238 if (!inet_ehash_secret)
239 inet_ehash_secret = rnd;
240 spin_unlock_bh(&inetsw_lock);
242 EXPORT_SYMBOL(build_ehash_secret);
244 static inline int inet_netns_ok(struct net *net, int protocol)
247 struct net_protocol *ipprot;
249 if (net == &init_net)
252 hash = protocol & (MAX_INET_PROTOS - 1);
253 ipprot = rcu_dereference(inet_protos[hash]);
258 return ipprot->netns_ok;
262 * Create an inet socket.
265 static int inet_create(struct net *net, struct socket *sock, int protocol)
269 struct inet_protosw *answer;
270 struct inet_sock *inet;
271 struct proto *answer_prot;
272 unsigned char answer_flags;
273 char answer_no_check;
274 int try_loading_module = 0;
277 if (sock->type != SOCK_RAW &&
278 sock->type != SOCK_DGRAM &&
280 build_ehash_secret();
282 sock->state = SS_UNCONNECTED;
284 /* Look for the requested type/protocol pair. */
287 err = -ESOCKTNOSUPPORT;
289 list_for_each_rcu(p, &inetsw[sock->type]) {
290 answer = list_entry(p, struct inet_protosw, list);
292 /* Check the non-wild match. */
293 if (protocol == answer->protocol) {
294 if (protocol != IPPROTO_IP)
297 /* Check for the two wild cases. */
298 if (IPPROTO_IP == protocol) {
299 protocol = answer->protocol;
302 if (IPPROTO_IP == answer->protocol)
305 err = -EPROTONOSUPPORT;
309 if (unlikely(answer == NULL)) {
310 if (try_loading_module < 2) {
313 * Be more specific, e.g. net-pf-2-proto-132-type-1
314 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
316 if (++try_loading_module == 1)
317 request_module("net-pf-%d-proto-%d-type-%d",
318 PF_INET, protocol, sock->type);
320 * Fall back to generic, e.g. net-pf-2-proto-132
321 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
324 request_module("net-pf-%d-proto-%d",
326 goto lookup_protocol;
332 if (answer->capability > 0 && !capable(answer->capability))
336 if (!inet_netns_ok(net, protocol))
339 sock->ops = answer->ops;
340 answer_prot = answer->prot;
341 answer_no_check = answer->no_check;
342 answer_flags = answer->flags;
345 BUG_TRAP(answer_prot->slab != NULL);
348 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
353 sk->sk_no_check = answer_no_check;
354 if (INET_PROTOSW_REUSE & answer_flags)
358 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
360 if (SOCK_RAW == sock->type) {
361 inet->num = protocol;
362 if (IPPROTO_RAW == protocol)
366 if (ipv4_config.no_pmtu_disc)
367 inet->pmtudisc = IP_PMTUDISC_DONT;
369 inet->pmtudisc = IP_PMTUDISC_WANT;
373 sock_init_data(sock, sk);
375 sk->sk_destruct = inet_sock_destruct;
376 sk->sk_family = PF_INET;
377 sk->sk_protocol = protocol;
378 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
384 inet->mc_list = NULL;
386 sk_refcnt_debug_inc(sk);
389 /* It assumes that any protocol which allows
390 * the user to assign a number at socket
391 * creation time automatically
394 inet->sport = htons(inet->num);
395 /* Add to protocol hash chains. */
396 sk->sk_prot->hash(sk);
399 if (sk->sk_prot->init) {
400 err = sk->sk_prot->init(sk);
402 sk_common_release(sk);
413 * The peer socket should always be NULL (or else). When we call this
414 * function we are destroying the object and from then on nobody
415 * should refer to it.
417 int inet_release(struct socket *sock)
419 struct sock *sk = sock->sk;
424 /* Applications forget to leave groups before exiting */
425 ip_mc_drop_socket(sk);
427 /* If linger is set, we don't return until the close
428 * is complete. Otherwise we return immediately. The
429 * actually closing is done the same either way.
431 * If the close is due to the process exiting, we never
435 if (sock_flag(sk, SOCK_LINGER) &&
436 !(current->flags & PF_EXITING))
437 timeout = sk->sk_lingertime;
439 sk->sk_prot->close(sk, timeout);
444 /* It is off by default, see below. */
445 int sysctl_ip_nonlocal_bind __read_mostly;
447 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
449 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
450 struct sock *sk = sock->sk;
451 struct inet_sock *inet = inet_sk(sk);
456 /* If the socket has its own bind function then use it. (RAW) */
457 if (sk->sk_prot->bind) {
458 err = sk->sk_prot->bind(sk, uaddr, addr_len);
462 if (addr_len < sizeof(struct sockaddr_in))
465 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
467 /* Not specified by any standard per-se, however it breaks too
468 * many applications when removed. It is unfortunate since
469 * allowing applications to make a non-local bind solves
470 * several problems with systems using dynamic addressing.
471 * (ie. your servers still start up even if your ISDN link
472 * is temporarily down)
474 err = -EADDRNOTAVAIL;
475 if (!sysctl_ip_nonlocal_bind &&
477 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
478 chk_addr_ret != RTN_LOCAL &&
479 chk_addr_ret != RTN_MULTICAST &&
480 chk_addr_ret != RTN_BROADCAST)
483 snum = ntohs(addr->sin_port);
485 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
488 /* We keep a pair of addresses. rcv_saddr is the one
489 * used by hash lookups, and saddr is used for transmit.
491 * In the BSD API these are the same except where it
492 * would be illegal to use them (multicast/broadcast) in
493 * which case the sending device address is used.
497 /* Check these errors (active socket, double bind). */
499 if (sk->sk_state != TCP_CLOSE || inet->num)
500 goto out_release_sock;
502 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
503 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
504 inet->saddr = 0; /* Use device */
506 /* Make sure we are allowed to bind here. */
507 if (sk->sk_prot->get_port(sk, snum)) {
508 inet->saddr = inet->rcv_saddr = 0;
510 goto out_release_sock;
514 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
516 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
517 inet->sport = htons(inet->num);
528 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
529 int addr_len, int flags)
531 struct sock *sk = sock->sk;
533 if (uaddr->sa_family == AF_UNSPEC)
534 return sk->sk_prot->disconnect(sk, flags);
536 if (!inet_sk(sk)->num && inet_autobind(sk))
538 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
541 static long inet_wait_for_connect(struct sock *sk, long timeo)
545 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
547 /* Basic assumption: if someone sets sk->sk_err, he _must_
548 * change state of the socket from TCP_SYN_*.
549 * Connect() does not allow to get error notifications
550 * without closing the socket.
552 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
554 timeo = schedule_timeout(timeo);
556 if (signal_pending(current) || !timeo)
558 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
560 finish_wait(sk->sk_sleep, &wait);
565 * Connect to a remote host. There is regrettably still a little
566 * TCP 'magic' in here.
568 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
569 int addr_len, int flags)
571 struct sock *sk = sock->sk;
577 if (uaddr->sa_family == AF_UNSPEC) {
578 err = sk->sk_prot->disconnect(sk, flags);
579 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
583 switch (sock->state) {
592 /* Fall out of switch with err, set for this state */
596 if (sk->sk_state != TCP_CLOSE)
599 err = sk->sk_prot->connect(sk, uaddr, addr_len);
603 sock->state = SS_CONNECTING;
605 /* Just entered SS_CONNECTING state; the only
606 * difference is that return value in non-blocking
607 * case is EINPROGRESS, rather than EALREADY.
613 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
615 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
616 /* Error code is set above */
617 if (!timeo || !inet_wait_for_connect(sk, timeo))
620 err = sock_intr_errno(timeo);
621 if (signal_pending(current))
625 /* Connection was closed by RST, timeout, ICMP error
626 * or another process disconnected us.
628 if (sk->sk_state == TCP_CLOSE)
631 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
632 * and error was received after socket entered established state.
633 * Hence, it is handled normally after connect() return successfully.
636 sock->state = SS_CONNECTED;
643 err = sock_error(sk) ? : -ECONNABORTED;
644 sock->state = SS_UNCONNECTED;
645 if (sk->sk_prot->disconnect(sk, flags))
646 sock->state = SS_DISCONNECTING;
651 * Accept a pending connection. The TCP layer now gives BSD semantics.
654 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
656 struct sock *sk1 = sock->sk;
658 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
665 BUG_TRAP((1 << sk2->sk_state) &
666 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
668 sock_graft(sk2, newsock);
670 newsock->state = SS_CONNECTED;
679 * This does both peername and sockname.
681 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
682 int *uaddr_len, int peer)
684 struct sock *sk = sock->sk;
685 struct inet_sock *inet = inet_sk(sk);
686 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
688 sin->sin_family = AF_INET;
691 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
694 sin->sin_port = inet->dport;
695 sin->sin_addr.s_addr = inet->daddr;
697 __be32 addr = inet->rcv_saddr;
700 sin->sin_port = inet->sport;
701 sin->sin_addr.s_addr = addr;
703 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
704 *uaddr_len = sizeof(*sin);
708 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
711 struct sock *sk = sock->sk;
713 /* We may need to bind the socket. */
714 if (!inet_sk(sk)->num && inet_autobind(sk))
717 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
721 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
723 struct sock *sk = sock->sk;
725 /* We may need to bind the socket. */
726 if (!inet_sk(sk)->num && inet_autobind(sk))
729 if (sk->sk_prot->sendpage)
730 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
731 return sock_no_sendpage(sock, page, offset, size, flags);
735 int inet_shutdown(struct socket *sock, int how)
737 struct sock *sk = sock->sk;
740 /* This should really check to make sure
741 * the socket is a TCP socket. (WHY AC...)
743 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
746 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
750 if (sock->state == SS_CONNECTING) {
751 if ((1 << sk->sk_state) &
752 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
753 sock->state = SS_DISCONNECTING;
755 sock->state = SS_CONNECTED;
758 switch (sk->sk_state) {
761 /* Hack to wake up other listeners, who can poll for
762 POLLHUP, even on eg. unconnected UDP sockets -- RR */
764 sk->sk_shutdown |= how;
765 if (sk->sk_prot->shutdown)
766 sk->sk_prot->shutdown(sk, how);
769 /* Remaining two branches are temporary solution for missing
770 * close() in multithreaded environment. It is _not_ a good idea,
771 * but we have no choice until close() is repaired at VFS level.
774 if (!(how & RCV_SHUTDOWN))
778 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
779 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
783 /* Wake up anyone sleeping in poll. */
784 sk->sk_state_change(sk);
790 * ioctl() calls you can issue on an INET socket. Most of these are
791 * device configuration and stuff and very rarely used. Some ioctls
792 * pass on to the socket itself.
794 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
795 * loads the devconfigure module does its configuring and unloads it.
796 * There's a good 20K of config code hanging around the kernel.
799 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
801 struct sock *sk = sock->sk;
803 struct net *net = sock_net(sk);
807 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
810 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
815 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
820 err = arp_ioctl(net, cmd, (void __user *)arg);
833 err = devinet_ioctl(net, cmd, (void __user *)arg);
836 if (sk->sk_prot->ioctl)
837 err = sk->sk_prot->ioctl(sk, cmd, arg);
845 const struct proto_ops inet_stream_ops = {
847 .owner = THIS_MODULE,
848 .release = inet_release,
850 .connect = inet_stream_connect,
851 .socketpair = sock_no_socketpair,
852 .accept = inet_accept,
853 .getname = inet_getname,
856 .listen = inet_listen,
857 .shutdown = inet_shutdown,
858 .setsockopt = sock_common_setsockopt,
859 .getsockopt = sock_common_getsockopt,
860 .sendmsg = tcp_sendmsg,
861 .recvmsg = sock_common_recvmsg,
862 .mmap = sock_no_mmap,
863 .sendpage = tcp_sendpage,
864 .splice_read = tcp_splice_read,
866 .compat_setsockopt = compat_sock_common_setsockopt,
867 .compat_getsockopt = compat_sock_common_getsockopt,
871 const struct proto_ops inet_dgram_ops = {
873 .owner = THIS_MODULE,
874 .release = inet_release,
876 .connect = inet_dgram_connect,
877 .socketpair = sock_no_socketpair,
878 .accept = sock_no_accept,
879 .getname = inet_getname,
882 .listen = sock_no_listen,
883 .shutdown = inet_shutdown,
884 .setsockopt = sock_common_setsockopt,
885 .getsockopt = sock_common_getsockopt,
886 .sendmsg = inet_sendmsg,
887 .recvmsg = sock_common_recvmsg,
888 .mmap = sock_no_mmap,
889 .sendpage = inet_sendpage,
891 .compat_setsockopt = compat_sock_common_setsockopt,
892 .compat_getsockopt = compat_sock_common_getsockopt,
897 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
900 static const struct proto_ops inet_sockraw_ops = {
902 .owner = THIS_MODULE,
903 .release = inet_release,
905 .connect = inet_dgram_connect,
906 .socketpair = sock_no_socketpair,
907 .accept = sock_no_accept,
908 .getname = inet_getname,
909 .poll = datagram_poll,
911 .listen = sock_no_listen,
912 .shutdown = inet_shutdown,
913 .setsockopt = sock_common_setsockopt,
914 .getsockopt = sock_common_getsockopt,
915 .sendmsg = inet_sendmsg,
916 .recvmsg = sock_common_recvmsg,
917 .mmap = sock_no_mmap,
918 .sendpage = inet_sendpage,
920 .compat_setsockopt = compat_sock_common_setsockopt,
921 .compat_getsockopt = compat_sock_common_getsockopt,
925 static struct net_proto_family inet_family_ops = {
927 .create = inet_create,
928 .owner = THIS_MODULE,
931 /* Upon startup we insert all the elements in inetsw_array[] into
932 * the linked list inetsw.
934 static struct inet_protosw inetsw_array[] =
938 .protocol = IPPROTO_TCP,
940 .ops = &inet_stream_ops,
943 .flags = INET_PROTOSW_PERMANENT |
949 .protocol = IPPROTO_UDP,
951 .ops = &inet_dgram_ops,
953 .no_check = UDP_CSUM_DEFAULT,
954 .flags = INET_PROTOSW_PERMANENT,
960 .protocol = IPPROTO_IP, /* wild card */
962 .ops = &inet_sockraw_ops,
963 .capability = CAP_NET_RAW,
964 .no_check = UDP_CSUM_DEFAULT,
965 .flags = INET_PROTOSW_REUSE,
969 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
971 void inet_register_protosw(struct inet_protosw *p)
973 struct list_head *lh;
974 struct inet_protosw *answer;
975 int protocol = p->protocol;
976 struct list_head *last_perm;
978 spin_lock_bh(&inetsw_lock);
980 if (p->type >= SOCK_MAX)
983 /* If we are trying to override a permanent protocol, bail. */
985 last_perm = &inetsw[p->type];
986 list_for_each(lh, &inetsw[p->type]) {
987 answer = list_entry(lh, struct inet_protosw, list);
989 /* Check only the non-wild match. */
990 if (INET_PROTOSW_PERMANENT & answer->flags) {
991 if (protocol == answer->protocol)
1001 /* Add the new entry after the last permanent entry if any, so that
1002 * the new entry does not override a permanent entry when matched with
1003 * a wild-card protocol. But it is allowed to override any existing
1004 * non-permanent entry. This means that when we remove this entry, the
1005 * system automatically returns to the old behavior.
1007 list_add_rcu(&p->list, last_perm);
1009 spin_unlock_bh(&inetsw_lock);
1016 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1022 "Ignoring attempt to register invalid socket type %d.\n",
1027 void inet_unregister_protosw(struct inet_protosw *p)
1029 if (INET_PROTOSW_PERMANENT & p->flags) {
1031 "Attempt to unregister permanent protocol %d.\n",
1034 spin_lock_bh(&inetsw_lock);
1035 list_del_rcu(&p->list);
1036 spin_unlock_bh(&inetsw_lock);
1043 * Shall we try to damage output packets if routing dev changes?
1046 int sysctl_ip_dynaddr __read_mostly;
1048 static int inet_sk_reselect_saddr(struct sock *sk)
1050 struct inet_sock *inet = inet_sk(sk);
1053 __be32 old_saddr = inet->saddr;
1055 __be32 daddr = inet->daddr;
1057 if (inet->opt && inet->opt->srr)
1058 daddr = inet->opt->faddr;
1060 /* Query new route. */
1061 err = ip_route_connect(&rt, daddr, 0,
1063 sk->sk_bound_dev_if,
1065 inet->sport, inet->dport, sk, 0);
1069 sk_setup_caps(sk, &rt->u.dst);
1071 new_saddr = rt->rt_src;
1073 if (new_saddr == old_saddr)
1076 if (sysctl_ip_dynaddr > 1) {
1077 printk(KERN_INFO "%s(): shifting inet->"
1078 "saddr from " NIPQUAD_FMT " to " NIPQUAD_FMT "\n",
1081 NIPQUAD(new_saddr));
1084 inet->saddr = inet->rcv_saddr = new_saddr;
1087 * XXX The only one ugly spot where we need to
1088 * XXX really change the sockets identity after
1089 * XXX it has entered the hashes. -DaveM
1091 * Besides that, it does not check for connection
1092 * uniqueness. Wait for troubles.
1094 __sk_prot_rehash(sk);
1098 int inet_sk_rebuild_header(struct sock *sk)
1100 struct inet_sock *inet = inet_sk(sk);
1101 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1105 /* Route is OK, nothing to do. */
1110 daddr = inet->daddr;
1111 if (inet->opt && inet->opt->srr)
1112 daddr = inet->opt->faddr;
1115 .oif = sk->sk_bound_dev_if,
1119 .saddr = inet->saddr,
1120 .tos = RT_CONN_FLAGS(sk),
1123 .proto = sk->sk_protocol,
1126 .sport = inet->sport,
1127 .dport = inet->dport,
1132 security_sk_classify_flow(sk, &fl);
1133 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1136 sk_setup_caps(sk, &rt->u.dst);
1138 /* Routing failed... */
1139 sk->sk_route_caps = 0;
1141 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1142 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1144 if (!sysctl_ip_dynaddr ||
1145 sk->sk_state != TCP_SYN_SENT ||
1146 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1147 (err = inet_sk_reselect_saddr(sk)) != 0)
1148 sk->sk_err_soft = -err;
1154 EXPORT_SYMBOL(inet_sk_rebuild_header);
1156 static int inet_gso_send_check(struct sk_buff *skb)
1159 struct net_protocol *ops;
1164 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1169 if (ihl < sizeof(*iph))
1172 if (unlikely(!pskb_may_pull(skb, ihl)))
1175 __skb_pull(skb, ihl);
1176 skb_reset_transport_header(skb);
1178 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1179 err = -EPROTONOSUPPORT;
1182 ops = rcu_dereference(inet_protos[proto]);
1183 if (likely(ops && ops->gso_send_check))
1184 err = ops->gso_send_check(skb);
1191 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1193 struct sk_buff *segs = ERR_PTR(-EINVAL);
1195 struct net_protocol *ops;
1200 if (!(features & NETIF_F_V4_CSUM))
1201 features &= ~NETIF_F_SG;
1203 if (unlikely(skb_shinfo(skb)->gso_type &
1211 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1216 if (ihl < sizeof(*iph))
1219 if (unlikely(!pskb_may_pull(skb, ihl)))
1222 __skb_pull(skb, ihl);
1223 skb_reset_transport_header(skb);
1225 id = ntohs(iph->id);
1226 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1227 segs = ERR_PTR(-EPROTONOSUPPORT);
1230 ops = rcu_dereference(inet_protos[proto]);
1231 if (likely(ops && ops->gso_segment))
1232 segs = ops->gso_segment(skb, features);
1235 if (!segs || IS_ERR(segs))
1241 iph->id = htons(id++);
1242 iph->tot_len = htons(skb->len - skb->mac_len);
1244 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1245 } while ((skb = skb->next));
1251 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1252 unsigned short type, unsigned char protocol,
1255 struct socket *sock;
1256 int rc = sock_create_kern(family, type, protocol, &sock);
1260 (*sk)->sk_allocation = GFP_ATOMIC;
1262 * Unhash it so that IP input processing does not even see it,
1263 * we do not wish this socket to see incoming packets.
1265 (*sk)->sk_prot->unhash(*sk);
1267 sk_change_net(*sk, net);
1272 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1274 unsigned long snmp_fold_field(void *mib[], int offt)
1276 unsigned long res = 0;
1279 for_each_possible_cpu(i) {
1280 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1281 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1285 EXPORT_SYMBOL_GPL(snmp_fold_field);
1287 int snmp_mib_init(void *ptr[2], size_t mibsize)
1289 BUG_ON(ptr == NULL);
1290 ptr[0] = __alloc_percpu(mibsize);
1293 ptr[1] = __alloc_percpu(mibsize);
1298 free_percpu(ptr[0]);
1303 EXPORT_SYMBOL_GPL(snmp_mib_init);
1305 void snmp_mib_free(void *ptr[2])
1307 BUG_ON(ptr == NULL);
1308 free_percpu(ptr[0]);
1309 free_percpu(ptr[1]);
1310 ptr[0] = ptr[1] = NULL;
1312 EXPORT_SYMBOL_GPL(snmp_mib_free);
1314 #ifdef CONFIG_IP_MULTICAST
1315 static struct net_protocol igmp_protocol = {
1316 .handler = igmp_rcv,
1320 static struct net_protocol tcp_protocol = {
1321 .handler = tcp_v4_rcv,
1322 .err_handler = tcp_v4_err,
1323 .gso_send_check = tcp_v4_gso_send_check,
1324 .gso_segment = tcp_tso_segment,
1329 static struct net_protocol udp_protocol = {
1331 .err_handler = udp_err,
1336 static struct net_protocol icmp_protocol = {
1337 .handler = icmp_rcv,
1342 static int __init init_ipv4_mibs(void)
1344 if (snmp_mib_init((void **)net_statistics,
1345 sizeof(struct linux_mib)) < 0)
1347 if (snmp_mib_init((void **)ip_statistics,
1348 sizeof(struct ipstats_mib)) < 0)
1350 if (snmp_mib_init((void **)icmp_statistics,
1351 sizeof(struct icmp_mib)) < 0)
1353 if (snmp_mib_init((void **)icmpmsg_statistics,
1354 sizeof(struct icmpmsg_mib)) < 0)
1355 goto err_icmpmsg_mib;
1356 if (snmp_mib_init((void **)tcp_statistics,
1357 sizeof(struct tcp_mib)) < 0)
1359 if (snmp_mib_init((void **)udp_statistics,
1360 sizeof(struct udp_mib)) < 0)
1362 if (snmp_mib_init((void **)udplite_statistics,
1363 sizeof(struct udp_mib)) < 0)
1364 goto err_udplite_mib;
1371 snmp_mib_free((void **)udp_statistics);
1373 snmp_mib_free((void **)tcp_statistics);
1375 snmp_mib_free((void **)icmpmsg_statistics);
1377 snmp_mib_free((void **)icmp_statistics);
1379 snmp_mib_free((void **)ip_statistics);
1381 snmp_mib_free((void **)net_statistics);
1386 static int ipv4_proc_init(void);
1389 * IP protocol layer initialiser
1392 static struct packet_type ip_packet_type = {
1393 .type = __constant_htons(ETH_P_IP),
1395 .gso_send_check = inet_gso_send_check,
1396 .gso_segment = inet_gso_segment,
1399 static int __init inet_init(void)
1401 struct sk_buff *dummy_skb;
1402 struct inet_protosw *q;
1403 struct list_head *r;
1406 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1408 rc = proto_register(&tcp_prot, 1);
1412 rc = proto_register(&udp_prot, 1);
1414 goto out_unregister_tcp_proto;
1416 rc = proto_register(&raw_prot, 1);
1418 goto out_unregister_udp_proto;
1421 * Tell SOCKET that we are alive...
1424 (void)sock_register(&inet_family_ops);
1427 * Add all the base protocols.
1430 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1431 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1432 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1433 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1434 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1435 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1436 #ifdef CONFIG_IP_MULTICAST
1437 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1438 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1441 /* Register the socket-side information for inet_create. */
1442 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1445 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1446 inet_register_protosw(q);
1449 * Set the ARP module up
1455 * Set the IP module up
1462 /* Setup TCP slab cache for open requests. */
1465 /* Setup UDP memory threshold */
1468 /* Add UDP-Lite (RFC 3828) */
1469 udplite4_register();
1472 * Set the ICMP layer up
1475 if (icmp_init() < 0)
1476 panic("Failed to create the ICMP control socket.\n");
1479 * Initialise the multicast router
1481 #if defined(CONFIG_IP_MROUTE)
1483 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1486 * Initialise per-cpu ipv4 mibs
1489 if (init_ipv4_mibs())
1490 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1496 dev_add_pack(&ip_packet_type);
1501 out_unregister_udp_proto:
1502 proto_unregister(&udp_prot);
1503 out_unregister_tcp_proto:
1504 proto_unregister(&tcp_prot);
1508 fs_initcall(inet_init);
1510 /* ------------------------------------------------------------------------ */
1512 #ifdef CONFIG_PROC_FS
1513 static int __init ipv4_proc_init(void)
1517 if (raw_proc_init())
1519 if (tcp4_proc_init())
1521 if (udp4_proc_init())
1523 if (ip_misc_proc_init())
1538 #else /* CONFIG_PROC_FS */
1539 static int __init ipv4_proc_init(void)
1543 #endif /* CONFIG_PROC_FS */
1545 MODULE_ALIAS_NETPROTO(PF_INET);
1547 EXPORT_SYMBOL(inet_accept);
1548 EXPORT_SYMBOL(inet_bind);
1549 EXPORT_SYMBOL(inet_dgram_connect);
1550 EXPORT_SYMBOL(inet_dgram_ops);
1551 EXPORT_SYMBOL(inet_getname);
1552 EXPORT_SYMBOL(inet_ioctl);
1553 EXPORT_SYMBOL(inet_listen);
1554 EXPORT_SYMBOL(inet_register_protosw);
1555 EXPORT_SYMBOL(inet_release);
1556 EXPORT_SYMBOL(inet_sendmsg);
1557 EXPORT_SYMBOL(inet_shutdown);
1558 EXPORT_SYMBOL(inet_sock_destruct);
1559 EXPORT_SYMBOL(inet_stream_connect);
1560 EXPORT_SYMBOL(inet_stream_ops);
1561 EXPORT_SYMBOL(inet_unregister_protosw);
1562 EXPORT_SYMBOL(net_statistics);
1563 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);