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 #include <net/net_namespace.h>
114 #ifdef CONFIG_IP_MROUTE
115 #include <linux/mroute.h>
118 extern void ip_mc_drop_socket(struct sock *sk);
120 /* The inetsw table contains everything that inet_create needs to
121 * build a new socket.
123 static struct list_head inetsw[SOCK_MAX];
124 static DEFINE_SPINLOCK(inetsw_lock);
126 struct ipv4_config ipv4_config;
128 EXPORT_SYMBOL(ipv4_config);
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);
141 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
142 printk("Attempt to release TCP socket in state %d %p\n",
146 if (!sock_flag(sk, SOCK_DEAD)) {
147 printk("Attempt to release alive inet socket %p\n", sk);
151 BUG_TRAP(!atomic_read(&sk->sk_rmem_alloc));
152 BUG_TRAP(!atomic_read(&sk->sk_wmem_alloc));
153 BUG_TRAP(!sk->sk_wmem_queued);
154 BUG_TRAP(!sk->sk_forward_alloc);
157 dst_release(sk->sk_dst_cache);
158 sk_refcnt_debug_dec(sk);
162 * The routines beyond this point handle the behaviour of an AF_INET
163 * socket object. Mostly it punts to the subprotocols of IP to do
168 * Automatically bind an unbound socket.
171 static int inet_autobind(struct sock *sk)
173 struct inet_sock *inet;
174 /* We may need to bind the socket. */
178 if (sk->sk_prot->get_port(sk, 0)) {
182 inet->sport = htons(inet->num);
189 * Move a socket into listening state.
191 int inet_listen(struct socket *sock, int backlog)
193 struct sock *sk = sock->sk;
194 unsigned char old_state;
200 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
203 old_state = sk->sk_state;
204 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
207 /* Really, if the socket is already in listen state
208 * we can only allow the backlog to be adjusted.
210 if (old_state != TCP_LISTEN) {
211 err = inet_csk_listen_start(sk, backlog);
215 sk->sk_max_ack_backlog = backlog;
223 u32 inet_ehash_secret __read_mostly;
224 EXPORT_SYMBOL(inet_ehash_secret);
227 * inet_ehash_secret must be set exactly once
228 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
230 void build_ehash_secret(void)
234 get_random_bytes(&rnd, sizeof(rnd));
236 spin_lock_bh(&inetsw_lock);
237 if (!inet_ehash_secret)
238 inet_ehash_secret = rnd;
239 spin_unlock_bh(&inetsw_lock);
241 EXPORT_SYMBOL(build_ehash_secret);
243 static inline int inet_netns_ok(struct net *net, int protocol)
246 struct net_protocol *ipprot;
248 if (net == &init_net)
251 hash = protocol & (MAX_INET_PROTOS - 1);
252 ipprot = rcu_dereference(inet_protos[hash]);
257 return ipprot->netns_ok;
261 * Create an inet socket.
264 static int inet_create(struct net *net, struct socket *sock, int protocol)
268 struct inet_protosw *answer;
269 struct inet_sock *inet;
270 struct proto *answer_prot;
271 unsigned char answer_flags;
272 char answer_no_check;
273 int try_loading_module = 0;
276 if (sock->type != SOCK_RAW &&
277 sock->type != SOCK_DGRAM &&
279 build_ehash_secret();
281 sock->state = SS_UNCONNECTED;
283 /* Look for the requested type/protocol pair. */
286 err = -ESOCKTNOSUPPORT;
288 list_for_each_rcu(p, &inetsw[sock->type]) {
289 answer = list_entry(p, struct inet_protosw, list);
291 /* Check the non-wild match. */
292 if (protocol == answer->protocol) {
293 if (protocol != IPPROTO_IP)
296 /* Check for the two wild cases. */
297 if (IPPROTO_IP == protocol) {
298 protocol = answer->protocol;
301 if (IPPROTO_IP == answer->protocol)
304 err = -EPROTONOSUPPORT;
308 if (unlikely(answer == NULL)) {
309 if (try_loading_module < 2) {
312 * Be more specific, e.g. net-pf-2-proto-132-type-1
313 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
315 if (++try_loading_module == 1)
316 request_module("net-pf-%d-proto-%d-type-%d",
317 PF_INET, protocol, sock->type);
319 * Fall back to generic, e.g. net-pf-2-proto-132
320 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
323 request_module("net-pf-%d-proto-%d",
325 goto lookup_protocol;
331 if (answer->capability > 0 && !capable(answer->capability))
335 if (!inet_netns_ok(net, protocol))
338 sock->ops = answer->ops;
339 answer_prot = answer->prot;
340 answer_no_check = answer->no_check;
341 answer_flags = answer->flags;
344 BUG_TRAP(answer_prot->slab != NULL);
347 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
352 sk->sk_no_check = answer_no_check;
353 if (INET_PROTOSW_REUSE & answer_flags)
357 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
359 if (SOCK_RAW == sock->type) {
360 inet->num = protocol;
361 if (IPPROTO_RAW == protocol)
365 if (ipv4_config.no_pmtu_disc)
366 inet->pmtudisc = IP_PMTUDISC_DONT;
368 inet->pmtudisc = IP_PMTUDISC_WANT;
372 sock_init_data(sock, sk);
374 sk->sk_destruct = inet_sock_destruct;
375 sk->sk_family = PF_INET;
376 sk->sk_protocol = protocol;
377 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
383 inet->mc_list = NULL;
385 sk_refcnt_debug_inc(sk);
388 /* It assumes that any protocol which allows
389 * the user to assign a number at socket
390 * creation time automatically
393 inet->sport = htons(inet->num);
394 /* Add to protocol hash chains. */
395 sk->sk_prot->hash(sk);
398 if (sk->sk_prot->init) {
399 err = sk->sk_prot->init(sk);
401 sk_common_release(sk);
412 * The peer socket should always be NULL (or else). When we call this
413 * function we are destroying the object and from then on nobody
414 * should refer to it.
416 int inet_release(struct socket *sock)
418 struct sock *sk = sock->sk;
423 /* Applications forget to leave groups before exiting */
424 ip_mc_drop_socket(sk);
426 /* If linger is set, we don't return until the close
427 * is complete. Otherwise we return immediately. The
428 * actually closing is done the same either way.
430 * If the close is due to the process exiting, we never
434 if (sock_flag(sk, SOCK_LINGER) &&
435 !(current->flags & PF_EXITING))
436 timeout = sk->sk_lingertime;
438 sk->sk_prot->close(sk, timeout);
443 /* It is off by default, see below. */
444 int sysctl_ip_nonlocal_bind __read_mostly;
446 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
448 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
449 struct sock *sk = sock->sk;
450 struct inet_sock *inet = inet_sk(sk);
455 /* If the socket has its own bind function then use it. (RAW) */
456 if (sk->sk_prot->bind) {
457 err = sk->sk_prot->bind(sk, uaddr, addr_len);
461 if (addr_len < sizeof(struct sockaddr_in))
464 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
466 /* Not specified by any standard per-se, however it breaks too
467 * many applications when removed. It is unfortunate since
468 * allowing applications to make a non-local bind solves
469 * several problems with systems using dynamic addressing.
470 * (ie. your servers still start up even if your ISDN link
471 * is temporarily down)
473 err = -EADDRNOTAVAIL;
474 if (!sysctl_ip_nonlocal_bind &&
476 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
477 chk_addr_ret != RTN_LOCAL &&
478 chk_addr_ret != RTN_MULTICAST &&
479 chk_addr_ret != RTN_BROADCAST)
482 snum = ntohs(addr->sin_port);
484 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
487 /* We keep a pair of addresses. rcv_saddr is the one
488 * used by hash lookups, and saddr is used for transmit.
490 * In the BSD API these are the same except where it
491 * would be illegal to use them (multicast/broadcast) in
492 * which case the sending device address is used.
496 /* Check these errors (active socket, double bind). */
498 if (sk->sk_state != TCP_CLOSE || inet->num)
499 goto out_release_sock;
501 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
502 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
503 inet->saddr = 0; /* Use device */
505 /* Make sure we are allowed to bind here. */
506 if (sk->sk_prot->get_port(sk, snum)) {
507 inet->saddr = inet->rcv_saddr = 0;
509 goto out_release_sock;
513 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
515 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
516 inet->sport = htons(inet->num);
527 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
528 int addr_len, int flags)
530 struct sock *sk = sock->sk;
532 if (uaddr->sa_family == AF_UNSPEC)
533 return sk->sk_prot->disconnect(sk, flags);
535 if (!inet_sk(sk)->num && inet_autobind(sk))
537 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
540 static long inet_wait_for_connect(struct sock *sk, long timeo)
544 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
546 /* Basic assumption: if someone sets sk->sk_err, he _must_
547 * change state of the socket from TCP_SYN_*.
548 * Connect() does not allow to get error notifications
549 * without closing the socket.
551 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
553 timeo = schedule_timeout(timeo);
555 if (signal_pending(current) || !timeo)
557 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
559 finish_wait(sk->sk_sleep, &wait);
564 * Connect to a remote host. There is regrettably still a little
565 * TCP 'magic' in here.
567 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
568 int addr_len, int flags)
570 struct sock *sk = sock->sk;
576 if (uaddr->sa_family == AF_UNSPEC) {
577 err = sk->sk_prot->disconnect(sk, flags);
578 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
582 switch (sock->state) {
591 /* Fall out of switch with err, set for this state */
595 if (sk->sk_state != TCP_CLOSE)
598 err = sk->sk_prot->connect(sk, uaddr, addr_len);
602 sock->state = SS_CONNECTING;
604 /* Just entered SS_CONNECTING state; the only
605 * difference is that return value in non-blocking
606 * case is EINPROGRESS, rather than EALREADY.
612 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
614 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
615 /* Error code is set above */
616 if (!timeo || !inet_wait_for_connect(sk, timeo))
619 err = sock_intr_errno(timeo);
620 if (signal_pending(current))
624 /* Connection was closed by RST, timeout, ICMP error
625 * or another process disconnected us.
627 if (sk->sk_state == TCP_CLOSE)
630 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
631 * and error was received after socket entered established state.
632 * Hence, it is handled normally after connect() return successfully.
635 sock->state = SS_CONNECTED;
642 err = sock_error(sk) ? : -ECONNABORTED;
643 sock->state = SS_UNCONNECTED;
644 if (sk->sk_prot->disconnect(sk, flags))
645 sock->state = SS_DISCONNECTING;
650 * Accept a pending connection. The TCP layer now gives BSD semantics.
653 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
655 struct sock *sk1 = sock->sk;
657 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
664 BUG_TRAP((1 << sk2->sk_state) &
665 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE));
667 sock_graft(sk2, newsock);
669 newsock->state = SS_CONNECTED;
678 * This does both peername and sockname.
680 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
681 int *uaddr_len, int peer)
683 struct sock *sk = sock->sk;
684 struct inet_sock *inet = inet_sk(sk);
685 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
687 sin->sin_family = AF_INET;
690 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
693 sin->sin_port = inet->dport;
694 sin->sin_addr.s_addr = inet->daddr;
696 __be32 addr = inet->rcv_saddr;
699 sin->sin_port = inet->sport;
700 sin->sin_addr.s_addr = addr;
702 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
703 *uaddr_len = sizeof(*sin);
707 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
710 struct sock *sk = sock->sk;
712 /* We may need to bind the socket. */
713 if (!inet_sk(sk)->num && inet_autobind(sk))
716 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
720 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
722 struct sock *sk = sock->sk;
724 /* We may need to bind the socket. */
725 if (!inet_sk(sk)->num && inet_autobind(sk))
728 if (sk->sk_prot->sendpage)
729 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
730 return sock_no_sendpage(sock, page, offset, size, flags);
734 int inet_shutdown(struct socket *sock, int how)
736 struct sock *sk = sock->sk;
739 /* This should really check to make sure
740 * the socket is a TCP socket. (WHY AC...)
742 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
745 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
749 if (sock->state == SS_CONNECTING) {
750 if ((1 << sk->sk_state) &
751 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
752 sock->state = SS_DISCONNECTING;
754 sock->state = SS_CONNECTED;
757 switch (sk->sk_state) {
760 /* Hack to wake up other listeners, who can poll for
761 POLLHUP, even on eg. unconnected UDP sockets -- RR */
763 sk->sk_shutdown |= how;
764 if (sk->sk_prot->shutdown)
765 sk->sk_prot->shutdown(sk, how);
768 /* Remaining two branches are temporary solution for missing
769 * close() in multithreaded environment. It is _not_ a good idea,
770 * but we have no choice until close() is repaired at VFS level.
773 if (!(how & RCV_SHUTDOWN))
777 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
778 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
782 /* Wake up anyone sleeping in poll. */
783 sk->sk_state_change(sk);
789 * ioctl() calls you can issue on an INET socket. Most of these are
790 * device configuration and stuff and very rarely used. Some ioctls
791 * pass on to the socket itself.
793 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
794 * loads the devconfigure module does its configuring and unloads it.
795 * There's a good 20K of config code hanging around the kernel.
798 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
800 struct sock *sk = sock->sk;
802 struct net *net = sock_net(sk);
806 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
809 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
814 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
819 err = arp_ioctl(net, cmd, (void __user *)arg);
832 err = devinet_ioctl(net, cmd, (void __user *)arg);
835 if (sk->sk_prot->ioctl)
836 err = sk->sk_prot->ioctl(sk, cmd, arg);
844 const struct proto_ops inet_stream_ops = {
846 .owner = THIS_MODULE,
847 .release = inet_release,
849 .connect = inet_stream_connect,
850 .socketpair = sock_no_socketpair,
851 .accept = inet_accept,
852 .getname = inet_getname,
855 .listen = inet_listen,
856 .shutdown = inet_shutdown,
857 .setsockopt = sock_common_setsockopt,
858 .getsockopt = sock_common_getsockopt,
859 .sendmsg = tcp_sendmsg,
860 .recvmsg = sock_common_recvmsg,
861 .mmap = sock_no_mmap,
862 .sendpage = tcp_sendpage,
863 .splice_read = tcp_splice_read,
865 .compat_setsockopt = compat_sock_common_setsockopt,
866 .compat_getsockopt = compat_sock_common_getsockopt,
870 const struct proto_ops inet_dgram_ops = {
872 .owner = THIS_MODULE,
873 .release = inet_release,
875 .connect = inet_dgram_connect,
876 .socketpair = sock_no_socketpair,
877 .accept = sock_no_accept,
878 .getname = inet_getname,
881 .listen = sock_no_listen,
882 .shutdown = inet_shutdown,
883 .setsockopt = sock_common_setsockopt,
884 .getsockopt = sock_common_getsockopt,
885 .sendmsg = inet_sendmsg,
886 .recvmsg = sock_common_recvmsg,
887 .mmap = sock_no_mmap,
888 .sendpage = inet_sendpage,
890 .compat_setsockopt = compat_sock_common_setsockopt,
891 .compat_getsockopt = compat_sock_common_getsockopt,
896 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
899 static const struct proto_ops inet_sockraw_ops = {
901 .owner = THIS_MODULE,
902 .release = inet_release,
904 .connect = inet_dgram_connect,
905 .socketpair = sock_no_socketpair,
906 .accept = sock_no_accept,
907 .getname = inet_getname,
908 .poll = datagram_poll,
910 .listen = sock_no_listen,
911 .shutdown = inet_shutdown,
912 .setsockopt = sock_common_setsockopt,
913 .getsockopt = sock_common_getsockopt,
914 .sendmsg = inet_sendmsg,
915 .recvmsg = sock_common_recvmsg,
916 .mmap = sock_no_mmap,
917 .sendpage = inet_sendpage,
919 .compat_setsockopt = compat_sock_common_setsockopt,
920 .compat_getsockopt = compat_sock_common_getsockopt,
924 static struct net_proto_family inet_family_ops = {
926 .create = inet_create,
927 .owner = THIS_MODULE,
930 /* Upon startup we insert all the elements in inetsw_array[] into
931 * the linked list inetsw.
933 static struct inet_protosw inetsw_array[] =
937 .protocol = IPPROTO_TCP,
939 .ops = &inet_stream_ops,
942 .flags = INET_PROTOSW_PERMANENT |
948 .protocol = IPPROTO_UDP,
950 .ops = &inet_dgram_ops,
952 .no_check = UDP_CSUM_DEFAULT,
953 .flags = INET_PROTOSW_PERMANENT,
959 .protocol = IPPROTO_IP, /* wild card */
961 .ops = &inet_sockraw_ops,
962 .capability = CAP_NET_RAW,
963 .no_check = UDP_CSUM_DEFAULT,
964 .flags = INET_PROTOSW_REUSE,
968 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
970 void inet_register_protosw(struct inet_protosw *p)
972 struct list_head *lh;
973 struct inet_protosw *answer;
974 int protocol = p->protocol;
975 struct list_head *last_perm;
977 spin_lock_bh(&inetsw_lock);
979 if (p->type >= SOCK_MAX)
982 /* If we are trying to override a permanent protocol, bail. */
984 last_perm = &inetsw[p->type];
985 list_for_each(lh, &inetsw[p->type]) {
986 answer = list_entry(lh, struct inet_protosw, list);
988 /* Check only the non-wild match. */
989 if (INET_PROTOSW_PERMANENT & answer->flags) {
990 if (protocol == answer->protocol)
1000 /* Add the new entry after the last permanent entry if any, so that
1001 * the new entry does not override a permanent entry when matched with
1002 * a wild-card protocol. But it is allowed to override any existing
1003 * non-permanent entry. This means that when we remove this entry, the
1004 * system automatically returns to the old behavior.
1006 list_add_rcu(&p->list, last_perm);
1008 spin_unlock_bh(&inetsw_lock);
1015 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1021 "Ignoring attempt to register invalid socket type %d.\n",
1026 void inet_unregister_protosw(struct inet_protosw *p)
1028 if (INET_PROTOSW_PERMANENT & p->flags) {
1030 "Attempt to unregister permanent protocol %d.\n",
1033 spin_lock_bh(&inetsw_lock);
1034 list_del_rcu(&p->list);
1035 spin_unlock_bh(&inetsw_lock);
1042 * Shall we try to damage output packets if routing dev changes?
1045 int sysctl_ip_dynaddr __read_mostly;
1047 static int inet_sk_reselect_saddr(struct sock *sk)
1049 struct inet_sock *inet = inet_sk(sk);
1052 __be32 old_saddr = inet->saddr;
1054 __be32 daddr = inet->daddr;
1056 if (inet->opt && inet->opt->srr)
1057 daddr = inet->opt->faddr;
1059 /* Query new route. */
1060 err = ip_route_connect(&rt, daddr, 0,
1062 sk->sk_bound_dev_if,
1064 inet->sport, inet->dport, sk, 0);
1068 sk_setup_caps(sk, &rt->u.dst);
1070 new_saddr = rt->rt_src;
1072 if (new_saddr == old_saddr)
1075 if (sysctl_ip_dynaddr > 1) {
1076 printk(KERN_INFO "%s(): shifting inet->"
1077 "saddr from " NIPQUAD_FMT " to " NIPQUAD_FMT "\n",
1080 NIPQUAD(new_saddr));
1083 inet->saddr = inet->rcv_saddr = new_saddr;
1086 * XXX The only one ugly spot where we need to
1087 * XXX really change the sockets identity after
1088 * XXX it has entered the hashes. -DaveM
1090 * Besides that, it does not check for connection
1091 * uniqueness. Wait for troubles.
1093 __sk_prot_rehash(sk);
1097 int inet_sk_rebuild_header(struct sock *sk)
1099 struct inet_sock *inet = inet_sk(sk);
1100 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1104 /* Route is OK, nothing to do. */
1109 daddr = inet->daddr;
1110 if (inet->opt && inet->opt->srr)
1111 daddr = inet->opt->faddr;
1114 .oif = sk->sk_bound_dev_if,
1118 .saddr = inet->saddr,
1119 .tos = RT_CONN_FLAGS(sk),
1122 .proto = sk->sk_protocol,
1125 .sport = inet->sport,
1126 .dport = inet->dport,
1131 security_sk_classify_flow(sk, &fl);
1132 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1135 sk_setup_caps(sk, &rt->u.dst);
1137 /* Routing failed... */
1138 sk->sk_route_caps = 0;
1140 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1141 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1143 if (!sysctl_ip_dynaddr ||
1144 sk->sk_state != TCP_SYN_SENT ||
1145 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1146 (err = inet_sk_reselect_saddr(sk)) != 0)
1147 sk->sk_err_soft = -err;
1153 EXPORT_SYMBOL(inet_sk_rebuild_header);
1155 static int inet_gso_send_check(struct sk_buff *skb)
1158 struct net_protocol *ops;
1163 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1168 if (ihl < sizeof(*iph))
1171 if (unlikely(!pskb_may_pull(skb, ihl)))
1174 __skb_pull(skb, ihl);
1175 skb_reset_transport_header(skb);
1177 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1178 err = -EPROTONOSUPPORT;
1181 ops = rcu_dereference(inet_protos[proto]);
1182 if (likely(ops && ops->gso_send_check))
1183 err = ops->gso_send_check(skb);
1190 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1192 struct sk_buff *segs = ERR_PTR(-EINVAL);
1194 struct net_protocol *ops;
1199 if (!(features & NETIF_F_V4_CSUM))
1200 features &= ~NETIF_F_SG;
1202 if (unlikely(skb_shinfo(skb)->gso_type &
1210 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1215 if (ihl < sizeof(*iph))
1218 if (unlikely(!pskb_may_pull(skb, ihl)))
1221 __skb_pull(skb, ihl);
1222 skb_reset_transport_header(skb);
1224 id = ntohs(iph->id);
1225 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1226 segs = ERR_PTR(-EPROTONOSUPPORT);
1229 ops = rcu_dereference(inet_protos[proto]);
1230 if (likely(ops && ops->gso_segment))
1231 segs = ops->gso_segment(skb, features);
1234 if (!segs || IS_ERR(segs))
1240 iph->id = htons(id++);
1241 iph->tot_len = htons(skb->len - skb->mac_len);
1243 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1244 } while ((skb = skb->next));
1250 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1251 unsigned short type, unsigned char protocol,
1254 struct socket *sock;
1255 int rc = sock_create_kern(family, type, protocol, &sock);
1259 (*sk)->sk_allocation = GFP_ATOMIC;
1261 * Unhash it so that IP input processing does not even see it,
1262 * we do not wish this socket to see incoming packets.
1264 (*sk)->sk_prot->unhash(*sk);
1266 sk_change_net(*sk, net);
1271 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1273 unsigned long snmp_fold_field(void *mib[], int offt)
1275 unsigned long res = 0;
1278 for_each_possible_cpu(i) {
1279 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1280 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1284 EXPORT_SYMBOL_GPL(snmp_fold_field);
1286 int snmp_mib_init(void *ptr[2], size_t mibsize)
1288 BUG_ON(ptr == NULL);
1289 ptr[0] = __alloc_percpu(mibsize);
1292 ptr[1] = __alloc_percpu(mibsize);
1297 free_percpu(ptr[0]);
1302 EXPORT_SYMBOL_GPL(snmp_mib_init);
1304 void snmp_mib_free(void *ptr[2])
1306 BUG_ON(ptr == NULL);
1307 free_percpu(ptr[0]);
1308 free_percpu(ptr[1]);
1309 ptr[0] = ptr[1] = NULL;
1311 EXPORT_SYMBOL_GPL(snmp_mib_free);
1313 #ifdef CONFIG_IP_MULTICAST
1314 static struct net_protocol igmp_protocol = {
1315 .handler = igmp_rcv,
1319 static struct net_protocol tcp_protocol = {
1320 .handler = tcp_v4_rcv,
1321 .err_handler = tcp_v4_err,
1322 .gso_send_check = tcp_v4_gso_send_check,
1323 .gso_segment = tcp_tso_segment,
1328 static struct net_protocol udp_protocol = {
1330 .err_handler = udp_err,
1335 static struct net_protocol icmp_protocol = {
1336 .handler = icmp_rcv,
1341 static __net_init int ipv4_mib_init_net(struct net *net)
1343 if (snmp_mib_init((void **)net->mib.tcp_statistics,
1344 sizeof(struct tcp_mib)) < 0)
1346 if (snmp_mib_init((void **)net->mib.ip_statistics,
1347 sizeof(struct ipstats_mib)) < 0)
1349 if (snmp_mib_init((void **)net->mib.net_statistics,
1350 sizeof(struct linux_mib)) < 0)
1352 if (snmp_mib_init((void **)net->mib.udp_statistics,
1353 sizeof(struct udp_mib)) < 0)
1355 if (snmp_mib_init((void **)net->mib.udplite_statistics,
1356 sizeof(struct udp_mib)) < 0)
1357 goto err_udplite_mib;
1358 if (snmp_mib_init((void **)net->mib.icmp_statistics,
1359 sizeof(struct icmp_mib)) < 0)
1361 if (snmp_mib_init((void **)net->mib.icmpmsg_statistics,
1362 sizeof(struct icmpmsg_mib)) < 0)
1363 goto err_icmpmsg_mib;
1369 snmp_mib_free((void **)net->mib.icmp_statistics);
1371 snmp_mib_free((void **)net->mib.udplite_statistics);
1373 snmp_mib_free((void **)net->mib.udp_statistics);
1375 snmp_mib_free((void **)net->mib.net_statistics);
1377 snmp_mib_free((void **)net->mib.ip_statistics);
1379 snmp_mib_free((void **)net->mib.tcp_statistics);
1384 static __net_exit void ipv4_mib_exit_net(struct net *net)
1386 snmp_mib_free((void **)net->mib.icmpmsg_statistics);
1387 snmp_mib_free((void **)net->mib.icmp_statistics);
1388 snmp_mib_free((void **)net->mib.udplite_statistics);
1389 snmp_mib_free((void **)net->mib.udp_statistics);
1390 snmp_mib_free((void **)net->mib.net_statistics);
1391 snmp_mib_free((void **)net->mib.ip_statistics);
1392 snmp_mib_free((void **)net->mib.tcp_statistics);
1395 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1396 .init = ipv4_mib_init_net,
1397 .exit = ipv4_mib_exit_net,
1400 static int __init init_ipv4_mibs(void)
1402 return register_pernet_subsys(&ipv4_mib_ops);
1405 static int ipv4_proc_init(void);
1408 * IP protocol layer initialiser
1411 static struct packet_type ip_packet_type = {
1412 .type = __constant_htons(ETH_P_IP),
1414 .gso_send_check = inet_gso_send_check,
1415 .gso_segment = inet_gso_segment,
1418 static int __init inet_init(void)
1420 struct sk_buff *dummy_skb;
1421 struct inet_protosw *q;
1422 struct list_head *r;
1425 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1427 rc = proto_register(&tcp_prot, 1);
1431 rc = proto_register(&udp_prot, 1);
1433 goto out_unregister_tcp_proto;
1435 rc = proto_register(&raw_prot, 1);
1437 goto out_unregister_udp_proto;
1440 * Tell SOCKET that we are alive...
1443 (void)sock_register(&inet_family_ops);
1446 * Add all the base protocols.
1449 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1450 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1451 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1452 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1453 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1454 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1455 #ifdef CONFIG_IP_MULTICAST
1456 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1457 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1460 /* Register the socket-side information for inet_create. */
1461 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1464 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1465 inet_register_protosw(q);
1468 * Set the ARP module up
1474 * Set the IP module up
1481 /* Setup TCP slab cache for open requests. */
1484 /* Setup UDP memory threshold */
1487 /* Add UDP-Lite (RFC 3828) */
1488 udplite4_register();
1491 * Set the ICMP layer up
1494 if (icmp_init() < 0)
1495 panic("Failed to create the ICMP control socket.\n");
1498 * Initialise the multicast router
1500 #if defined(CONFIG_IP_MROUTE)
1502 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1505 * Initialise per-cpu ipv4 mibs
1508 if (init_ipv4_mibs())
1509 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1515 dev_add_pack(&ip_packet_type);
1520 out_unregister_udp_proto:
1521 proto_unregister(&udp_prot);
1522 out_unregister_tcp_proto:
1523 proto_unregister(&tcp_prot);
1527 fs_initcall(inet_init);
1529 /* ------------------------------------------------------------------------ */
1531 #ifdef CONFIG_PROC_FS
1532 static int __init ipv4_proc_init(void)
1536 if (raw_proc_init())
1538 if (tcp4_proc_init())
1540 if (udp4_proc_init())
1542 if (ip_misc_proc_init())
1557 #else /* CONFIG_PROC_FS */
1558 static int __init ipv4_proc_init(void)
1562 #endif /* CONFIG_PROC_FS */
1564 MODULE_ALIAS_NETPROTO(PF_INET);
1566 EXPORT_SYMBOL(inet_accept);
1567 EXPORT_SYMBOL(inet_bind);
1568 EXPORT_SYMBOL(inet_dgram_connect);
1569 EXPORT_SYMBOL(inet_dgram_ops);
1570 EXPORT_SYMBOL(inet_getname);
1571 EXPORT_SYMBOL(inet_ioctl);
1572 EXPORT_SYMBOL(inet_listen);
1573 EXPORT_SYMBOL(inet_register_protosw);
1574 EXPORT_SYMBOL(inet_release);
1575 EXPORT_SYMBOL(inet_sendmsg);
1576 EXPORT_SYMBOL(inet_shutdown);
1577 EXPORT_SYMBOL(inet_sock_destruct);
1578 EXPORT_SYMBOL(inet_stream_connect);
1579 EXPORT_SYMBOL(inet_stream_ops);
1580 EXPORT_SYMBOL(inet_unregister_protosw);
1581 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);