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 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
152 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
153 WARN_ON(sk->sk_wmem_queued);
154 WARN_ON(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_eq(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)
267 struct inet_protosw *answer;
268 struct inet_sock *inet;
269 struct proto *answer_prot;
270 unsigned char answer_flags;
271 char answer_no_check;
272 int try_loading_module = 0;
275 if (unlikely(!inet_ehash_secret))
276 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
277 build_ehash_secret();
279 sock->state = SS_UNCONNECTED;
281 /* Look for the requested type/protocol pair. */
283 err = -ESOCKTNOSUPPORT;
285 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
288 /* Check the non-wild match. */
289 if (protocol == answer->protocol) {
290 if (protocol != IPPROTO_IP)
293 /* Check for the two wild cases. */
294 if (IPPROTO_IP == protocol) {
295 protocol = answer->protocol;
298 if (IPPROTO_IP == answer->protocol)
301 err = -EPROTONOSUPPORT;
305 if (try_loading_module < 2) {
308 * Be more specific, e.g. net-pf-2-proto-132-type-1
309 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
311 if (++try_loading_module == 1)
312 request_module("net-pf-%d-proto-%d-type-%d",
313 PF_INET, protocol, sock->type);
315 * Fall back to generic, e.g. net-pf-2-proto-132
316 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
319 request_module("net-pf-%d-proto-%d",
321 goto lookup_protocol;
327 if (answer->capability > 0 && !capable(answer->capability))
331 if (!inet_netns_ok(net, protocol))
334 sock->ops = answer->ops;
335 answer_prot = answer->prot;
336 answer_no_check = answer->no_check;
337 answer_flags = answer->flags;
340 WARN_ON(answer_prot->slab == NULL);
343 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
348 sk->sk_no_check = answer_no_check;
349 if (INET_PROTOSW_REUSE & answer_flags)
353 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
355 if (SOCK_RAW == sock->type) {
356 inet->num = protocol;
357 if (IPPROTO_RAW == protocol)
361 if (ipv4_config.no_pmtu_disc)
362 inet->pmtudisc = IP_PMTUDISC_DONT;
364 inet->pmtudisc = IP_PMTUDISC_WANT;
368 sock_init_data(sock, sk);
370 sk->sk_destruct = inet_sock_destruct;
371 sk->sk_family = PF_INET;
372 sk->sk_protocol = protocol;
373 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
379 inet->mc_list = NULL;
381 sk_refcnt_debug_inc(sk);
384 /* It assumes that any protocol which allows
385 * the user to assign a number at socket
386 * creation time automatically
389 inet->sport = htons(inet->num);
390 /* Add to protocol hash chains. */
391 sk->sk_prot->hash(sk);
394 if (sk->sk_prot->init) {
395 err = sk->sk_prot->init(sk);
397 sk_common_release(sk);
408 * The peer socket should always be NULL (or else). When we call this
409 * function we are destroying the object and from then on nobody
410 * should refer to it.
412 int inet_release(struct socket *sock)
414 struct sock *sk = sock->sk;
419 /* Applications forget to leave groups before exiting */
420 ip_mc_drop_socket(sk);
422 /* If linger is set, we don't return until the close
423 * is complete. Otherwise we return immediately. The
424 * actually closing is done the same either way.
426 * If the close is due to the process exiting, we never
430 if (sock_flag(sk, SOCK_LINGER) &&
431 !(current->flags & PF_EXITING))
432 timeout = sk->sk_lingertime;
434 sk->sk_prot->close(sk, timeout);
439 /* It is off by default, see below. */
440 int sysctl_ip_nonlocal_bind __read_mostly;
442 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
444 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
445 struct sock *sk = sock->sk;
446 struct inet_sock *inet = inet_sk(sk);
451 /* If the socket has its own bind function then use it. (RAW) */
452 if (sk->sk_prot->bind) {
453 err = sk->sk_prot->bind(sk, uaddr, addr_len);
457 if (addr_len < sizeof(struct sockaddr_in))
460 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
462 /* Not specified by any standard per-se, however it breaks too
463 * many applications when removed. It is unfortunate since
464 * allowing applications to make a non-local bind solves
465 * several problems with systems using dynamic addressing.
466 * (ie. your servers still start up even if your ISDN link
467 * is temporarily down)
469 err = -EADDRNOTAVAIL;
470 if (!sysctl_ip_nonlocal_bind &&
471 !(inet->freebind || inet->transparent) &&
472 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
473 chk_addr_ret != RTN_LOCAL &&
474 chk_addr_ret != RTN_MULTICAST &&
475 chk_addr_ret != RTN_BROADCAST)
478 snum = ntohs(addr->sin_port);
480 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
483 /* We keep a pair of addresses. rcv_saddr is the one
484 * used by hash lookups, and saddr is used for transmit.
486 * In the BSD API these are the same except where it
487 * would be illegal to use them (multicast/broadcast) in
488 * which case the sending device address is used.
492 /* Check these errors (active socket, double bind). */
494 if (sk->sk_state != TCP_CLOSE || inet->num)
495 goto out_release_sock;
497 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
498 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
499 inet->saddr = 0; /* Use device */
501 /* Make sure we are allowed to bind here. */
502 if (sk->sk_prot->get_port(sk, snum)) {
503 inet->saddr = inet->rcv_saddr = 0;
505 goto out_release_sock;
509 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
511 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
512 inet->sport = htons(inet->num);
523 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
524 int addr_len, int flags)
526 struct sock *sk = sock->sk;
528 if (uaddr->sa_family == AF_UNSPEC)
529 return sk->sk_prot->disconnect(sk, flags);
531 if (!inet_sk(sk)->num && inet_autobind(sk))
533 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
536 static long inet_wait_for_connect(struct sock *sk, long timeo)
540 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
542 /* Basic assumption: if someone sets sk->sk_err, he _must_
543 * change state of the socket from TCP_SYN_*.
544 * Connect() does not allow to get error notifications
545 * without closing the socket.
547 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
549 timeo = schedule_timeout(timeo);
551 if (signal_pending(current) || !timeo)
553 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
555 finish_wait(sk->sk_sleep, &wait);
560 * Connect to a remote host. There is regrettably still a little
561 * TCP 'magic' in here.
563 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
564 int addr_len, int flags)
566 struct sock *sk = sock->sk;
572 if (uaddr->sa_family == AF_UNSPEC) {
573 err = sk->sk_prot->disconnect(sk, flags);
574 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
578 switch (sock->state) {
587 /* Fall out of switch with err, set for this state */
591 if (sk->sk_state != TCP_CLOSE)
594 err = sk->sk_prot->connect(sk, uaddr, addr_len);
598 sock->state = SS_CONNECTING;
600 /* Just entered SS_CONNECTING state; the only
601 * difference is that return value in non-blocking
602 * case is EINPROGRESS, rather than EALREADY.
608 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
610 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
611 /* Error code is set above */
612 if (!timeo || !inet_wait_for_connect(sk, timeo))
615 err = sock_intr_errno(timeo);
616 if (signal_pending(current))
620 /* Connection was closed by RST, timeout, ICMP error
621 * or another process disconnected us.
623 if (sk->sk_state == TCP_CLOSE)
626 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
627 * and error was received after socket entered established state.
628 * Hence, it is handled normally after connect() return successfully.
631 sock->state = SS_CONNECTED;
638 err = sock_error(sk) ? : -ECONNABORTED;
639 sock->state = SS_UNCONNECTED;
640 if (sk->sk_prot->disconnect(sk, flags))
641 sock->state = SS_DISCONNECTING;
646 * Accept a pending connection. The TCP layer now gives BSD semantics.
649 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
651 struct sock *sk1 = sock->sk;
653 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
660 WARN_ON(!((1 << sk2->sk_state) &
661 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
663 sock_graft(sk2, newsock);
665 newsock->state = SS_CONNECTED;
674 * This does both peername and sockname.
676 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
677 int *uaddr_len, int peer)
679 struct sock *sk = sock->sk;
680 struct inet_sock *inet = inet_sk(sk);
681 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
683 sin->sin_family = AF_INET;
686 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
689 sin->sin_port = inet->dport;
690 sin->sin_addr.s_addr = inet->daddr;
692 __be32 addr = inet->rcv_saddr;
695 sin->sin_port = inet->sport;
696 sin->sin_addr.s_addr = addr;
698 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
699 *uaddr_len = sizeof(*sin);
703 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
706 struct sock *sk = sock->sk;
708 /* We may need to bind the socket. */
709 if (!inet_sk(sk)->num && inet_autobind(sk))
712 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
716 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
718 struct sock *sk = sock->sk;
720 /* We may need to bind the socket. */
721 if (!inet_sk(sk)->num && inet_autobind(sk))
724 if (sk->sk_prot->sendpage)
725 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
726 return sock_no_sendpage(sock, page, offset, size, flags);
730 int inet_shutdown(struct socket *sock, int how)
732 struct sock *sk = sock->sk;
735 /* This should really check to make sure
736 * the socket is a TCP socket. (WHY AC...)
738 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
741 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
745 if (sock->state == SS_CONNECTING) {
746 if ((1 << sk->sk_state) &
747 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
748 sock->state = SS_DISCONNECTING;
750 sock->state = SS_CONNECTED;
753 switch (sk->sk_state) {
756 /* Hack to wake up other listeners, who can poll for
757 POLLHUP, even on eg. unconnected UDP sockets -- RR */
759 sk->sk_shutdown |= how;
760 if (sk->sk_prot->shutdown)
761 sk->sk_prot->shutdown(sk, how);
764 /* Remaining two branches are temporary solution for missing
765 * close() in multithreaded environment. It is _not_ a good idea,
766 * but we have no choice until close() is repaired at VFS level.
769 if (!(how & RCV_SHUTDOWN))
773 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
774 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
778 /* Wake up anyone sleeping in poll. */
779 sk->sk_state_change(sk);
785 * ioctl() calls you can issue on an INET socket. Most of these are
786 * device configuration and stuff and very rarely used. Some ioctls
787 * pass on to the socket itself.
789 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
790 * loads the devconfigure module does its configuring and unloads it.
791 * There's a good 20K of config code hanging around the kernel.
794 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
796 struct sock *sk = sock->sk;
798 struct net *net = sock_net(sk);
802 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
805 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
810 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
815 err = arp_ioctl(net, cmd, (void __user *)arg);
828 err = devinet_ioctl(net, cmd, (void __user *)arg);
831 if (sk->sk_prot->ioctl)
832 err = sk->sk_prot->ioctl(sk, cmd, arg);
840 const struct proto_ops inet_stream_ops = {
842 .owner = THIS_MODULE,
843 .release = inet_release,
845 .connect = inet_stream_connect,
846 .socketpair = sock_no_socketpair,
847 .accept = inet_accept,
848 .getname = inet_getname,
851 .listen = inet_listen,
852 .shutdown = inet_shutdown,
853 .setsockopt = sock_common_setsockopt,
854 .getsockopt = sock_common_getsockopt,
855 .sendmsg = tcp_sendmsg,
856 .recvmsg = sock_common_recvmsg,
857 .mmap = sock_no_mmap,
858 .sendpage = tcp_sendpage,
859 .splice_read = tcp_splice_read,
861 .compat_setsockopt = compat_sock_common_setsockopt,
862 .compat_getsockopt = compat_sock_common_getsockopt,
866 const struct proto_ops inet_dgram_ops = {
868 .owner = THIS_MODULE,
869 .release = inet_release,
871 .connect = inet_dgram_connect,
872 .socketpair = sock_no_socketpair,
873 .accept = sock_no_accept,
874 .getname = inet_getname,
877 .listen = sock_no_listen,
878 .shutdown = inet_shutdown,
879 .setsockopt = sock_common_setsockopt,
880 .getsockopt = sock_common_getsockopt,
881 .sendmsg = inet_sendmsg,
882 .recvmsg = sock_common_recvmsg,
883 .mmap = sock_no_mmap,
884 .sendpage = inet_sendpage,
886 .compat_setsockopt = compat_sock_common_setsockopt,
887 .compat_getsockopt = compat_sock_common_getsockopt,
892 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
895 static const struct proto_ops inet_sockraw_ops = {
897 .owner = THIS_MODULE,
898 .release = inet_release,
900 .connect = inet_dgram_connect,
901 .socketpair = sock_no_socketpair,
902 .accept = sock_no_accept,
903 .getname = inet_getname,
904 .poll = datagram_poll,
906 .listen = sock_no_listen,
907 .shutdown = inet_shutdown,
908 .setsockopt = sock_common_setsockopt,
909 .getsockopt = sock_common_getsockopt,
910 .sendmsg = inet_sendmsg,
911 .recvmsg = sock_common_recvmsg,
912 .mmap = sock_no_mmap,
913 .sendpage = inet_sendpage,
915 .compat_setsockopt = compat_sock_common_setsockopt,
916 .compat_getsockopt = compat_sock_common_getsockopt,
920 static struct net_proto_family inet_family_ops = {
922 .create = inet_create,
923 .owner = THIS_MODULE,
926 /* Upon startup we insert all the elements in inetsw_array[] into
927 * the linked list inetsw.
929 static struct inet_protosw inetsw_array[] =
933 .protocol = IPPROTO_TCP,
935 .ops = &inet_stream_ops,
938 .flags = INET_PROTOSW_PERMANENT |
944 .protocol = IPPROTO_UDP,
946 .ops = &inet_dgram_ops,
948 .no_check = UDP_CSUM_DEFAULT,
949 .flags = INET_PROTOSW_PERMANENT,
955 .protocol = IPPROTO_IP, /* wild card */
957 .ops = &inet_sockraw_ops,
958 .capability = CAP_NET_RAW,
959 .no_check = UDP_CSUM_DEFAULT,
960 .flags = INET_PROTOSW_REUSE,
964 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
966 void inet_register_protosw(struct inet_protosw *p)
968 struct list_head *lh;
969 struct inet_protosw *answer;
970 int protocol = p->protocol;
971 struct list_head *last_perm;
973 spin_lock_bh(&inetsw_lock);
975 if (p->type >= SOCK_MAX)
978 /* If we are trying to override a permanent protocol, bail. */
980 last_perm = &inetsw[p->type];
981 list_for_each(lh, &inetsw[p->type]) {
982 answer = list_entry(lh, struct inet_protosw, list);
984 /* Check only the non-wild match. */
985 if (INET_PROTOSW_PERMANENT & answer->flags) {
986 if (protocol == answer->protocol)
996 /* Add the new entry after the last permanent entry if any, so that
997 * the new entry does not override a permanent entry when matched with
998 * a wild-card protocol. But it is allowed to override any existing
999 * non-permanent entry. This means that when we remove this entry, the
1000 * system automatically returns to the old behavior.
1002 list_add_rcu(&p->list, last_perm);
1004 spin_unlock_bh(&inetsw_lock);
1011 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1017 "Ignoring attempt to register invalid socket type %d.\n",
1022 void inet_unregister_protosw(struct inet_protosw *p)
1024 if (INET_PROTOSW_PERMANENT & p->flags) {
1026 "Attempt to unregister permanent protocol %d.\n",
1029 spin_lock_bh(&inetsw_lock);
1030 list_del_rcu(&p->list);
1031 spin_unlock_bh(&inetsw_lock);
1038 * Shall we try to damage output packets if routing dev changes?
1041 int sysctl_ip_dynaddr __read_mostly;
1043 static int inet_sk_reselect_saddr(struct sock *sk)
1045 struct inet_sock *inet = inet_sk(sk);
1048 __be32 old_saddr = inet->saddr;
1050 __be32 daddr = inet->daddr;
1052 if (inet->opt && inet->opt->srr)
1053 daddr = inet->opt->faddr;
1055 /* Query new route. */
1056 err = ip_route_connect(&rt, daddr, 0,
1058 sk->sk_bound_dev_if,
1060 inet->sport, inet->dport, sk, 0);
1064 sk_setup_caps(sk, &rt->u.dst);
1066 new_saddr = rt->rt_src;
1068 if (new_saddr == old_saddr)
1071 if (sysctl_ip_dynaddr > 1) {
1072 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1073 __func__, &old_saddr, &new_saddr);
1076 inet->saddr = inet->rcv_saddr = new_saddr;
1079 * XXX The only one ugly spot where we need to
1080 * XXX really change the sockets identity after
1081 * XXX it has entered the hashes. -DaveM
1083 * Besides that, it does not check for connection
1084 * uniqueness. Wait for troubles.
1086 __sk_prot_rehash(sk);
1090 int inet_sk_rebuild_header(struct sock *sk)
1092 struct inet_sock *inet = inet_sk(sk);
1093 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1097 /* Route is OK, nothing to do. */
1102 daddr = inet->daddr;
1103 if (inet->opt && inet->opt->srr)
1104 daddr = inet->opt->faddr;
1107 .oif = sk->sk_bound_dev_if,
1111 .saddr = inet->saddr,
1112 .tos = RT_CONN_FLAGS(sk),
1115 .proto = sk->sk_protocol,
1116 .flags = inet_sk_flowi_flags(sk),
1119 .sport = inet->sport,
1120 .dport = inet->dport,
1125 security_sk_classify_flow(sk, &fl);
1126 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1129 sk_setup_caps(sk, &rt->u.dst);
1131 /* Routing failed... */
1132 sk->sk_route_caps = 0;
1134 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1135 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1137 if (!sysctl_ip_dynaddr ||
1138 sk->sk_state != TCP_SYN_SENT ||
1139 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1140 (err = inet_sk_reselect_saddr(sk)) != 0)
1141 sk->sk_err_soft = -err;
1147 EXPORT_SYMBOL(inet_sk_rebuild_header);
1149 static int inet_gso_send_check(struct sk_buff *skb)
1152 struct net_protocol *ops;
1157 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1162 if (ihl < sizeof(*iph))
1165 if (unlikely(!pskb_may_pull(skb, ihl)))
1168 __skb_pull(skb, ihl);
1169 skb_reset_transport_header(skb);
1171 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1172 err = -EPROTONOSUPPORT;
1175 ops = rcu_dereference(inet_protos[proto]);
1176 if (likely(ops && ops->gso_send_check))
1177 err = ops->gso_send_check(skb);
1184 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1186 struct sk_buff *segs = ERR_PTR(-EINVAL);
1188 struct net_protocol *ops;
1193 if (!(features & NETIF_F_V4_CSUM))
1194 features &= ~NETIF_F_SG;
1196 if (unlikely(skb_shinfo(skb)->gso_type &
1204 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1209 if (ihl < sizeof(*iph))
1212 if (unlikely(!pskb_may_pull(skb, ihl)))
1215 __skb_pull(skb, ihl);
1216 skb_reset_transport_header(skb);
1218 id = ntohs(iph->id);
1219 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1220 segs = ERR_PTR(-EPROTONOSUPPORT);
1223 ops = rcu_dereference(inet_protos[proto]);
1224 if (likely(ops && ops->gso_segment))
1225 segs = ops->gso_segment(skb, features);
1228 if (!segs || IS_ERR(segs))
1234 iph->id = htons(id++);
1235 iph->tot_len = htons(skb->len - skb->mac_len);
1237 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1238 } while ((skb = skb->next));
1244 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1245 unsigned short type, unsigned char protocol,
1248 struct socket *sock;
1249 int rc = sock_create_kern(family, type, protocol, &sock);
1253 (*sk)->sk_allocation = GFP_ATOMIC;
1255 * Unhash it so that IP input processing does not even see it,
1256 * we do not wish this socket to see incoming packets.
1258 (*sk)->sk_prot->unhash(*sk);
1260 sk_change_net(*sk, net);
1265 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1267 unsigned long snmp_fold_field(void *mib[], int offt)
1269 unsigned long res = 0;
1272 for_each_possible_cpu(i) {
1273 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1274 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1278 EXPORT_SYMBOL_GPL(snmp_fold_field);
1280 int snmp_mib_init(void *ptr[2], size_t mibsize)
1282 BUG_ON(ptr == NULL);
1283 ptr[0] = __alloc_percpu(mibsize);
1286 ptr[1] = __alloc_percpu(mibsize);
1291 free_percpu(ptr[0]);
1296 EXPORT_SYMBOL_GPL(snmp_mib_init);
1298 void snmp_mib_free(void *ptr[2])
1300 BUG_ON(ptr == NULL);
1301 free_percpu(ptr[0]);
1302 free_percpu(ptr[1]);
1303 ptr[0] = ptr[1] = NULL;
1305 EXPORT_SYMBOL_GPL(snmp_mib_free);
1307 #ifdef CONFIG_IP_MULTICAST
1308 static struct net_protocol igmp_protocol = {
1309 .handler = igmp_rcv,
1313 static struct net_protocol tcp_protocol = {
1314 .handler = tcp_v4_rcv,
1315 .err_handler = tcp_v4_err,
1316 .gso_send_check = tcp_v4_gso_send_check,
1317 .gso_segment = tcp_tso_segment,
1322 static struct net_protocol udp_protocol = {
1324 .err_handler = udp_err,
1329 static struct net_protocol icmp_protocol = {
1330 .handler = icmp_rcv,
1335 static __net_init int ipv4_mib_init_net(struct net *net)
1337 if (snmp_mib_init((void **)net->mib.tcp_statistics,
1338 sizeof(struct tcp_mib)) < 0)
1340 if (snmp_mib_init((void **)net->mib.ip_statistics,
1341 sizeof(struct ipstats_mib)) < 0)
1343 if (snmp_mib_init((void **)net->mib.net_statistics,
1344 sizeof(struct linux_mib)) < 0)
1346 if (snmp_mib_init((void **)net->mib.udp_statistics,
1347 sizeof(struct udp_mib)) < 0)
1349 if (snmp_mib_init((void **)net->mib.udplite_statistics,
1350 sizeof(struct udp_mib)) < 0)
1351 goto err_udplite_mib;
1352 if (snmp_mib_init((void **)net->mib.icmp_statistics,
1353 sizeof(struct icmp_mib)) < 0)
1355 if (snmp_mib_init((void **)net->mib.icmpmsg_statistics,
1356 sizeof(struct icmpmsg_mib)) < 0)
1357 goto err_icmpmsg_mib;
1363 snmp_mib_free((void **)net->mib.icmp_statistics);
1365 snmp_mib_free((void **)net->mib.udplite_statistics);
1367 snmp_mib_free((void **)net->mib.udp_statistics);
1369 snmp_mib_free((void **)net->mib.net_statistics);
1371 snmp_mib_free((void **)net->mib.ip_statistics);
1373 snmp_mib_free((void **)net->mib.tcp_statistics);
1378 static __net_exit void ipv4_mib_exit_net(struct net *net)
1380 snmp_mib_free((void **)net->mib.icmpmsg_statistics);
1381 snmp_mib_free((void **)net->mib.icmp_statistics);
1382 snmp_mib_free((void **)net->mib.udplite_statistics);
1383 snmp_mib_free((void **)net->mib.udp_statistics);
1384 snmp_mib_free((void **)net->mib.net_statistics);
1385 snmp_mib_free((void **)net->mib.ip_statistics);
1386 snmp_mib_free((void **)net->mib.tcp_statistics);
1389 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1390 .init = ipv4_mib_init_net,
1391 .exit = ipv4_mib_exit_net,
1394 static int __init init_ipv4_mibs(void)
1396 return register_pernet_subsys(&ipv4_mib_ops);
1399 static int ipv4_proc_init(void);
1402 * IP protocol layer initialiser
1405 static struct packet_type ip_packet_type = {
1406 .type = __constant_htons(ETH_P_IP),
1408 .gso_send_check = inet_gso_send_check,
1409 .gso_segment = inet_gso_segment,
1412 static int __init inet_init(void)
1414 struct sk_buff *dummy_skb;
1415 struct inet_protosw *q;
1416 struct list_head *r;
1419 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1421 rc = proto_register(&tcp_prot, 1);
1425 rc = proto_register(&udp_prot, 1);
1427 goto out_unregister_tcp_proto;
1429 rc = proto_register(&raw_prot, 1);
1431 goto out_unregister_udp_proto;
1434 * Tell SOCKET that we are alive...
1437 (void)sock_register(&inet_family_ops);
1439 #ifdef CONFIG_SYSCTL
1440 ip_static_sysctl_init();
1444 * Add all the base protocols.
1447 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1448 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1449 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1450 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1451 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1452 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1453 #ifdef CONFIG_IP_MULTICAST
1454 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1455 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1458 /* Register the socket-side information for inet_create. */
1459 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1462 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1463 inet_register_protosw(q);
1466 * Set the ARP module up
1472 * Set the IP module up
1479 /* Setup TCP slab cache for open requests. */
1482 /* Setup UDP memory threshold */
1485 /* Add UDP-Lite (RFC 3828) */
1486 udplite4_register();
1489 * Set the ICMP layer up
1492 if (icmp_init() < 0)
1493 panic("Failed to create the ICMP control socket.\n");
1496 * Initialise the multicast router
1498 #if defined(CONFIG_IP_MROUTE)
1500 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1503 * Initialise per-cpu ipv4 mibs
1506 if (init_ipv4_mibs())
1507 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1513 dev_add_pack(&ip_packet_type);
1518 out_unregister_udp_proto:
1519 proto_unregister(&udp_prot);
1520 out_unregister_tcp_proto:
1521 proto_unregister(&tcp_prot);
1525 fs_initcall(inet_init);
1527 /* ------------------------------------------------------------------------ */
1529 #ifdef CONFIG_PROC_FS
1530 static int __init ipv4_proc_init(void)
1534 if (raw_proc_init())
1536 if (tcp4_proc_init())
1538 if (udp4_proc_init())
1540 if (ip_misc_proc_init())
1555 #else /* CONFIG_PROC_FS */
1556 static int __init ipv4_proc_init(void)
1560 #endif /* CONFIG_PROC_FS */
1562 MODULE_ALIAS_NETPROTO(PF_INET);
1564 EXPORT_SYMBOL(inet_accept);
1565 EXPORT_SYMBOL(inet_bind);
1566 EXPORT_SYMBOL(inet_dgram_connect);
1567 EXPORT_SYMBOL(inet_dgram_ops);
1568 EXPORT_SYMBOL(inet_getname);
1569 EXPORT_SYMBOL(inet_ioctl);
1570 EXPORT_SYMBOL(inet_listen);
1571 EXPORT_SYMBOL(inet_register_protosw);
1572 EXPORT_SYMBOL(inet_release);
1573 EXPORT_SYMBOL(inet_sendmsg);
1574 EXPORT_SYMBOL(inet_shutdown);
1575 EXPORT_SYMBOL(inet_sock_destruct);
1576 EXPORT_SYMBOL(inet_stream_connect);
1577 EXPORT_SYMBOL(inet_stream_ops);
1578 EXPORT_SYMBOL(inet_unregister_protosw);
1579 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);