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>
97 #include <net/checksum.h>
99 #include <net/protocol.h>
101 #include <net/route.h>
102 #include <net/ip_fib.h>
103 #include <net/inet_connection_sock.h>
106 #include <net/udplite.h>
107 #include <linux/skbuff.h>
108 #include <net/sock.h>
110 #include <net/icmp.h>
111 #include <net/ipip.h>
112 #include <net/inet_common.h>
113 #include <net/xfrm.h>
114 #include <net/net_namespace.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
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;
127 EXPORT_SYMBOL(ipv4_config);
129 /* New destruction routine */
131 void inet_sock_destruct(struct sock *sk)
133 struct inet_sock *inet = inet_sk(sk);
135 __skb_queue_purge(&sk->sk_receive_queue);
136 __skb_queue_purge(&sk->sk_error_queue);
140 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
141 pr_err("Attempt to release TCP socket in state %d %p\n",
145 if (!sock_flag(sk, SOCK_DEAD)) {
146 pr_err("Attempt to release alive inet socket %p\n", sk);
150 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
151 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
152 WARN_ON(sk->sk_wmem_queued);
153 WARN_ON(sk->sk_forward_alloc);
156 dst_release(sk->sk_dst_cache);
157 sk_refcnt_debug_dec(sk);
159 EXPORT_SYMBOL(inet_sock_destruct);
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. */
177 if (!inet->inet_num) {
178 if (sk->sk_prot->get_port(sk, 0)) {
182 inet->inet_sport = htons(inet->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;
222 EXPORT_SYMBOL(inet_listen);
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 const struct net_protocol *ipprot;
249 if (net_eq(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 (unlikely(!inet_ehash_secret))
278 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
279 build_ehash_secret();
281 sock->state = SS_UNCONNECTED;
283 /* Look for the requested type/protocol pair. */
285 err = -ESOCKTNOSUPPORT;
287 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
290 /* Check the non-wild match. */
291 if (protocol == answer->protocol) {
292 if (protocol != IPPROTO_IP)
295 /* Check for the two wild cases. */
296 if (IPPROTO_IP == protocol) {
297 protocol = answer->protocol;
300 if (IPPROTO_IP == answer->protocol)
303 err = -EPROTONOSUPPORT;
307 if (try_loading_module < 2) {
310 * Be more specific, e.g. net-pf-2-proto-132-type-1
311 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
313 if (++try_loading_module == 1)
314 request_module("net-pf-%d-proto-%d-type-%d",
315 PF_INET, protocol, sock->type);
317 * Fall back to generic, e.g. net-pf-2-proto-132
318 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
321 request_module("net-pf-%d-proto-%d",
323 goto lookup_protocol;
329 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
333 if (!inet_netns_ok(net, protocol))
336 sock->ops = answer->ops;
337 answer_prot = answer->prot;
338 answer_no_check = answer->no_check;
339 answer_flags = answer->flags;
342 WARN_ON(answer_prot->slab == NULL);
345 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
350 sk->sk_no_check = answer_no_check;
351 if (INET_PROTOSW_REUSE & answer_flags)
355 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
357 if (SOCK_RAW == sock->type) {
358 inet->inet_num = protocol;
359 if (IPPROTO_RAW == protocol)
363 if (ipv4_config.no_pmtu_disc)
364 inet->pmtudisc = IP_PMTUDISC_DONT;
366 inet->pmtudisc = IP_PMTUDISC_WANT;
370 sock_init_data(sock, sk);
372 sk->sk_destruct = inet_sock_destruct;
373 sk->sk_protocol = protocol;
374 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
381 inet->mc_list = NULL;
383 sk_refcnt_debug_inc(sk);
385 if (inet->inet_num) {
386 /* It assumes that any protocol which allows
387 * the user to assign a number at socket
388 * creation time automatically
391 inet->inet_sport = htons(inet->inet_num);
392 /* Add to protocol hash chains. */
393 sk->sk_prot->hash(sk);
396 if (sk->sk_prot->init) {
397 err = sk->sk_prot->init(sk);
399 sk_common_release(sk);
410 * The peer socket should always be NULL (or else). When we call this
411 * function we are destroying the object and from then on nobody
412 * should refer to it.
414 int inet_release(struct socket *sock)
416 struct sock *sk = sock->sk;
421 /* Applications forget to leave groups before exiting */
422 ip_mc_drop_socket(sk);
424 /* If linger is set, we don't return until the close
425 * is complete. Otherwise we return immediately. The
426 * actually closing is done the same either way.
428 * If the close is due to the process exiting, we never
432 if (sock_flag(sk, SOCK_LINGER) &&
433 !(current->flags & PF_EXITING))
434 timeout = sk->sk_lingertime;
436 sk->sk_prot->close(sk, timeout);
440 EXPORT_SYMBOL(inet_release);
442 /* It is off by default, see below. */
443 int sysctl_ip_nonlocal_bind __read_mostly;
444 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
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 &&
475 !(inet->freebind || inet->transparent) &&
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->inet_num)
499 goto out_release_sock;
501 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
502 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
503 inet->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->inet_saddr = inet->inet_rcv_saddr = 0;
509 goto out_release_sock;
512 if (inet->inet_rcv_saddr)
513 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
515 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
516 inet->inet_sport = htons(inet->inet_num);
517 inet->inet_daddr = 0;
518 inet->inet_dport = 0;
526 EXPORT_SYMBOL(inet_bind);
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)->inet_num && inet_autobind(sk))
538 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
540 EXPORT_SYMBOL(inet_dgram_connect);
542 static long inet_wait_for_connect(struct sock *sk, long timeo)
546 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
548 /* Basic assumption: if someone sets sk->sk_err, he _must_
549 * change state of the socket from TCP_SYN_*.
550 * Connect() does not allow to get error notifications
551 * without closing the socket.
553 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
555 timeo = schedule_timeout(timeo);
557 if (signal_pending(current) || !timeo)
559 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
561 finish_wait(sk->sk_sleep, &wait);
566 * Connect to a remote host. There is regrettably still a little
567 * TCP 'magic' in here.
569 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
570 int addr_len, int flags)
572 struct sock *sk = sock->sk;
578 if (uaddr->sa_family == AF_UNSPEC) {
579 err = sk->sk_prot->disconnect(sk, flags);
580 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
584 switch (sock->state) {
593 /* Fall out of switch with err, set for this state */
597 if (sk->sk_state != TCP_CLOSE)
600 err = sk->sk_prot->connect(sk, uaddr, addr_len);
604 sock->state = SS_CONNECTING;
606 /* Just entered SS_CONNECTING state; the only
607 * difference is that return value in non-blocking
608 * case is EINPROGRESS, rather than EALREADY.
614 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
616 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
617 /* Error code is set above */
618 if (!timeo || !inet_wait_for_connect(sk, timeo))
621 err = sock_intr_errno(timeo);
622 if (signal_pending(current))
626 /* Connection was closed by RST, timeout, ICMP error
627 * or another process disconnected us.
629 if (sk->sk_state == TCP_CLOSE)
632 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
633 * and error was received after socket entered established state.
634 * Hence, it is handled normally after connect() return successfully.
637 sock->state = SS_CONNECTED;
644 err = sock_error(sk) ? : -ECONNABORTED;
645 sock->state = SS_UNCONNECTED;
646 if (sk->sk_prot->disconnect(sk, flags))
647 sock->state = SS_DISCONNECTING;
650 EXPORT_SYMBOL(inet_stream_connect);
653 * Accept a pending connection. The TCP layer now gives BSD semantics.
656 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
658 struct sock *sk1 = sock->sk;
660 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
667 WARN_ON(!((1 << sk2->sk_state) &
668 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
670 sock_graft(sk2, newsock);
672 newsock->state = SS_CONNECTED;
678 EXPORT_SYMBOL(inet_accept);
682 * This does both peername and sockname.
684 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
685 int *uaddr_len, int peer)
687 struct sock *sk = sock->sk;
688 struct inet_sock *inet = inet_sk(sk);
689 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
691 sin->sin_family = AF_INET;
693 if (!inet->inet_dport ||
694 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
697 sin->sin_port = inet->inet_dport;
698 sin->sin_addr.s_addr = inet->inet_daddr;
700 __be32 addr = inet->inet_rcv_saddr;
702 addr = inet->inet_saddr;
703 sin->sin_port = inet->inet_sport;
704 sin->sin_addr.s_addr = addr;
706 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
707 *uaddr_len = sizeof(*sin);
710 EXPORT_SYMBOL(inet_getname);
712 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
715 struct sock *sk = sock->sk;
717 /* We may need to bind the socket. */
718 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
721 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
723 EXPORT_SYMBOL(inet_sendmsg);
726 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
727 size_t size, int flags)
729 struct sock *sk = sock->sk;
731 /* We may need to bind the socket. */
732 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
735 if (sk->sk_prot->sendpage)
736 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
737 return sock_no_sendpage(sock, page, offset, size, flags);
741 int inet_shutdown(struct socket *sock, int how)
743 struct sock *sk = sock->sk;
746 /* This should really check to make sure
747 * the socket is a TCP socket. (WHY AC...)
749 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
752 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
756 if (sock->state == SS_CONNECTING) {
757 if ((1 << sk->sk_state) &
758 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
759 sock->state = SS_DISCONNECTING;
761 sock->state = SS_CONNECTED;
764 switch (sk->sk_state) {
767 /* Hack to wake up other listeners, who can poll for
768 POLLHUP, even on eg. unconnected UDP sockets -- RR */
770 sk->sk_shutdown |= how;
771 if (sk->sk_prot->shutdown)
772 sk->sk_prot->shutdown(sk, how);
775 /* Remaining two branches are temporary solution for missing
776 * close() in multithreaded environment. It is _not_ a good idea,
777 * but we have no choice until close() is repaired at VFS level.
780 if (!(how & RCV_SHUTDOWN))
784 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
785 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
789 /* Wake up anyone sleeping in poll. */
790 sk->sk_state_change(sk);
794 EXPORT_SYMBOL(inet_shutdown);
797 * ioctl() calls you can issue on an INET socket. Most of these are
798 * device configuration and stuff and very rarely used. Some ioctls
799 * pass on to the socket itself.
801 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
802 * loads the devconfigure module does its configuring and unloads it.
803 * There's a good 20K of config code hanging around the kernel.
806 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
808 struct sock *sk = sock->sk;
810 struct net *net = sock_net(sk);
814 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
817 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
822 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
827 err = arp_ioctl(net, cmd, (void __user *)arg);
840 err = devinet_ioctl(net, cmd, (void __user *)arg);
843 if (sk->sk_prot->ioctl)
844 err = sk->sk_prot->ioctl(sk, cmd, arg);
851 EXPORT_SYMBOL(inet_ioctl);
853 const struct proto_ops inet_stream_ops = {
855 .owner = THIS_MODULE,
856 .release = inet_release,
858 .connect = inet_stream_connect,
859 .socketpair = sock_no_socketpair,
860 .accept = inet_accept,
861 .getname = inet_getname,
864 .listen = inet_listen,
865 .shutdown = inet_shutdown,
866 .setsockopt = sock_common_setsockopt,
867 .getsockopt = sock_common_getsockopt,
868 .sendmsg = tcp_sendmsg,
869 .recvmsg = sock_common_recvmsg,
870 .mmap = sock_no_mmap,
871 .sendpage = tcp_sendpage,
872 .splice_read = tcp_splice_read,
874 .compat_setsockopt = compat_sock_common_setsockopt,
875 .compat_getsockopt = compat_sock_common_getsockopt,
878 EXPORT_SYMBOL(inet_stream_ops);
880 const struct proto_ops inet_dgram_ops = {
882 .owner = THIS_MODULE,
883 .release = inet_release,
885 .connect = inet_dgram_connect,
886 .socketpair = sock_no_socketpair,
887 .accept = sock_no_accept,
888 .getname = inet_getname,
891 .listen = sock_no_listen,
892 .shutdown = inet_shutdown,
893 .setsockopt = sock_common_setsockopt,
894 .getsockopt = sock_common_getsockopt,
895 .sendmsg = inet_sendmsg,
896 .recvmsg = sock_common_recvmsg,
897 .mmap = sock_no_mmap,
898 .sendpage = inet_sendpage,
900 .compat_setsockopt = compat_sock_common_setsockopt,
901 .compat_getsockopt = compat_sock_common_getsockopt,
904 EXPORT_SYMBOL(inet_dgram_ops);
907 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
910 static const struct proto_ops inet_sockraw_ops = {
912 .owner = THIS_MODULE,
913 .release = inet_release,
915 .connect = inet_dgram_connect,
916 .socketpair = sock_no_socketpair,
917 .accept = sock_no_accept,
918 .getname = inet_getname,
919 .poll = datagram_poll,
921 .listen = sock_no_listen,
922 .shutdown = inet_shutdown,
923 .setsockopt = sock_common_setsockopt,
924 .getsockopt = sock_common_getsockopt,
925 .sendmsg = inet_sendmsg,
926 .recvmsg = sock_common_recvmsg,
927 .mmap = sock_no_mmap,
928 .sendpage = inet_sendpage,
930 .compat_setsockopt = compat_sock_common_setsockopt,
931 .compat_getsockopt = compat_sock_common_getsockopt,
935 static const struct net_proto_family inet_family_ops = {
937 .create = inet_create,
938 .owner = THIS_MODULE,
941 /* Upon startup we insert all the elements in inetsw_array[] into
942 * the linked list inetsw.
944 static struct inet_protosw inetsw_array[] =
948 .protocol = IPPROTO_TCP,
950 .ops = &inet_stream_ops,
952 .flags = INET_PROTOSW_PERMANENT |
958 .protocol = IPPROTO_UDP,
960 .ops = &inet_dgram_ops,
961 .no_check = UDP_CSUM_DEFAULT,
962 .flags = INET_PROTOSW_PERMANENT,
968 .protocol = IPPROTO_IP, /* wild card */
970 .ops = &inet_sockraw_ops,
971 .no_check = UDP_CSUM_DEFAULT,
972 .flags = INET_PROTOSW_REUSE,
976 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
978 void inet_register_protosw(struct inet_protosw *p)
980 struct list_head *lh;
981 struct inet_protosw *answer;
982 int protocol = p->protocol;
983 struct list_head *last_perm;
985 spin_lock_bh(&inetsw_lock);
987 if (p->type >= SOCK_MAX)
990 /* If we are trying to override a permanent protocol, bail. */
992 last_perm = &inetsw[p->type];
993 list_for_each(lh, &inetsw[p->type]) {
994 answer = list_entry(lh, struct inet_protosw, list);
996 /* Check only the non-wild match. */
997 if (INET_PROTOSW_PERMANENT & answer->flags) {
998 if (protocol == answer->protocol)
1008 /* Add the new entry after the last permanent entry if any, so that
1009 * the new entry does not override a permanent entry when matched with
1010 * a wild-card protocol. But it is allowed to override any existing
1011 * non-permanent entry. This means that when we remove this entry, the
1012 * system automatically returns to the old behavior.
1014 list_add_rcu(&p->list, last_perm);
1016 spin_unlock_bh(&inetsw_lock);
1021 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1027 "Ignoring attempt to register invalid socket type %d.\n",
1031 EXPORT_SYMBOL(inet_register_protosw);
1033 void inet_unregister_protosw(struct inet_protosw *p)
1035 if (INET_PROTOSW_PERMANENT & p->flags) {
1037 "Attempt to unregister permanent protocol %d.\n",
1040 spin_lock_bh(&inetsw_lock);
1041 list_del_rcu(&p->list);
1042 spin_unlock_bh(&inetsw_lock);
1047 EXPORT_SYMBOL(inet_unregister_protosw);
1050 * Shall we try to damage output packets if routing dev changes?
1053 int sysctl_ip_dynaddr __read_mostly;
1055 static int inet_sk_reselect_saddr(struct sock *sk)
1057 struct inet_sock *inet = inet_sk(sk);
1060 __be32 old_saddr = inet->inet_saddr;
1062 __be32 daddr = inet->inet_daddr;
1064 if (inet->opt && inet->opt->srr)
1065 daddr = inet->opt->faddr;
1067 /* Query new route. */
1068 err = ip_route_connect(&rt, daddr, 0,
1070 sk->sk_bound_dev_if,
1072 inet->inet_sport, inet->inet_dport, sk, 0);
1076 sk_setup_caps(sk, &rt->u.dst);
1078 new_saddr = rt->rt_src;
1080 if (new_saddr == old_saddr)
1083 if (sysctl_ip_dynaddr > 1) {
1084 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1085 __func__, &old_saddr, &new_saddr);
1088 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1091 * XXX The only one ugly spot where we need to
1092 * XXX really change the sockets identity after
1093 * XXX it has entered the hashes. -DaveM
1095 * Besides that, it does not check for connection
1096 * uniqueness. Wait for troubles.
1098 __sk_prot_rehash(sk);
1102 int inet_sk_rebuild_header(struct sock *sk)
1104 struct inet_sock *inet = inet_sk(sk);
1105 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1109 /* Route is OK, nothing to do. */
1114 daddr = inet->inet_daddr;
1115 if (inet->opt && inet->opt->srr)
1116 daddr = inet->opt->faddr;
1119 .oif = sk->sk_bound_dev_if,
1120 .mark = sk->sk_mark,
1124 .saddr = inet->inet_saddr,
1125 .tos = RT_CONN_FLAGS(sk),
1128 .proto = sk->sk_protocol,
1129 .flags = inet_sk_flowi_flags(sk),
1132 .sport = inet->inet_sport,
1133 .dport = inet->inet_dport,
1138 security_sk_classify_flow(sk, &fl);
1139 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1142 sk_setup_caps(sk, &rt->u.dst);
1144 /* Routing failed... */
1145 sk->sk_route_caps = 0;
1147 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1148 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1150 if (!sysctl_ip_dynaddr ||
1151 sk->sk_state != TCP_SYN_SENT ||
1152 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1153 (err = inet_sk_reselect_saddr(sk)) != 0)
1154 sk->sk_err_soft = -err;
1159 EXPORT_SYMBOL(inet_sk_rebuild_header);
1161 static int inet_gso_send_check(struct sk_buff *skb)
1164 const struct net_protocol *ops;
1169 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1174 if (ihl < sizeof(*iph))
1177 if (unlikely(!pskb_may_pull(skb, ihl)))
1180 __skb_pull(skb, ihl);
1181 skb_reset_transport_header(skb);
1183 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1184 err = -EPROTONOSUPPORT;
1187 ops = rcu_dereference(inet_protos[proto]);
1188 if (likely(ops && ops->gso_send_check))
1189 err = ops->gso_send_check(skb);
1196 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1198 struct sk_buff *segs = ERR_PTR(-EINVAL);
1200 const struct net_protocol *ops;
1204 unsigned int offset = 0;
1206 if (!(features & NETIF_F_V4_CSUM))
1207 features &= ~NETIF_F_SG;
1209 if (unlikely(skb_shinfo(skb)->gso_type &
1217 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1222 if (ihl < sizeof(*iph))
1225 if (unlikely(!pskb_may_pull(skb, ihl)))
1228 __skb_pull(skb, ihl);
1229 skb_reset_transport_header(skb);
1231 id = ntohs(iph->id);
1232 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1233 segs = ERR_PTR(-EPROTONOSUPPORT);
1236 ops = rcu_dereference(inet_protos[proto]);
1237 if (likely(ops && ops->gso_segment))
1238 segs = ops->gso_segment(skb, features);
1241 if (!segs || IS_ERR(segs))
1247 if (proto == IPPROTO_UDP) {
1248 iph->id = htons(id);
1249 iph->frag_off = htons(offset >> 3);
1250 if (skb->next != NULL)
1251 iph->frag_off |= htons(IP_MF);
1252 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1254 iph->id = htons(id++);
1255 iph->tot_len = htons(skb->len - skb->mac_len);
1257 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1258 } while ((skb = skb->next));
1264 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1265 struct sk_buff *skb)
1267 const struct net_protocol *ops;
1268 struct sk_buff **pp = NULL;
1277 off = skb_gro_offset(skb);
1278 hlen = off + sizeof(*iph);
1279 iph = skb_gro_header_fast(skb, off);
1280 if (skb_gro_header_hard(skb, hlen)) {
1281 iph = skb_gro_header_slow(skb, hlen, off);
1286 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1289 ops = rcu_dereference(inet_protos[proto]);
1290 if (!ops || !ops->gro_receive)
1293 if (*(u8 *)iph != 0x45)
1296 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1299 id = ntohl(*(u32 *)&iph->id);
1300 flush = (u16)((ntohl(*(u32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1303 for (p = *head; p; p = p->next) {
1306 if (!NAPI_GRO_CB(p)->same_flow)
1311 if ((iph->protocol ^ iph2->protocol) |
1312 (iph->tos ^ iph2->tos) |
1313 (iph->saddr ^ iph2->saddr) |
1314 (iph->daddr ^ iph2->daddr)) {
1315 NAPI_GRO_CB(p)->same_flow = 0;
1319 /* All fields must match except length and checksum. */
1320 NAPI_GRO_CB(p)->flush |=
1321 (iph->ttl ^ iph2->ttl) |
1322 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1324 NAPI_GRO_CB(p)->flush |= flush;
1327 NAPI_GRO_CB(skb)->flush |= flush;
1328 skb_gro_pull(skb, sizeof(*iph));
1329 skb_set_transport_header(skb, skb_gro_offset(skb));
1331 pp = ops->gro_receive(head, skb);
1337 NAPI_GRO_CB(skb)->flush |= flush;
1342 static int inet_gro_complete(struct sk_buff *skb)
1344 const struct net_protocol *ops;
1345 struct iphdr *iph = ip_hdr(skb);
1346 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1348 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1350 csum_replace2(&iph->check, iph->tot_len, newlen);
1351 iph->tot_len = newlen;
1354 ops = rcu_dereference(inet_protos[proto]);
1355 if (WARN_ON(!ops || !ops->gro_complete))
1358 err = ops->gro_complete(skb);
1366 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1367 unsigned short type, unsigned char protocol,
1370 struct socket *sock;
1371 int rc = sock_create_kern(family, type, protocol, &sock);
1375 (*sk)->sk_allocation = GFP_ATOMIC;
1377 * Unhash it so that IP input processing does not even see it,
1378 * we do not wish this socket to see incoming packets.
1380 (*sk)->sk_prot->unhash(*sk);
1382 sk_change_net(*sk, net);
1386 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1388 unsigned long snmp_fold_field(void *mib[], int offt)
1390 unsigned long res = 0;
1393 for_each_possible_cpu(i) {
1394 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1395 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1399 EXPORT_SYMBOL_GPL(snmp_fold_field);
1401 int snmp_mib_init(void *ptr[2], size_t mibsize)
1403 BUG_ON(ptr == NULL);
1404 ptr[0] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1407 ptr[1] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1412 free_percpu(ptr[0]);
1417 EXPORT_SYMBOL_GPL(snmp_mib_init);
1419 void snmp_mib_free(void *ptr[2])
1421 BUG_ON(ptr == NULL);
1422 free_percpu(ptr[0]);
1423 free_percpu(ptr[1]);
1424 ptr[0] = ptr[1] = NULL;
1426 EXPORT_SYMBOL_GPL(snmp_mib_free);
1428 #ifdef CONFIG_IP_MULTICAST
1429 static const struct net_protocol igmp_protocol = {
1430 .handler = igmp_rcv,
1435 static const struct net_protocol tcp_protocol = {
1436 .handler = tcp_v4_rcv,
1437 .err_handler = tcp_v4_err,
1438 .gso_send_check = tcp_v4_gso_send_check,
1439 .gso_segment = tcp_tso_segment,
1440 .gro_receive = tcp4_gro_receive,
1441 .gro_complete = tcp4_gro_complete,
1446 static const struct net_protocol udp_protocol = {
1448 .err_handler = udp_err,
1449 .gso_send_check = udp4_ufo_send_check,
1450 .gso_segment = udp4_ufo_fragment,
1455 static const struct net_protocol icmp_protocol = {
1456 .handler = icmp_rcv,
1461 static __net_init int ipv4_mib_init_net(struct net *net)
1463 if (snmp_mib_init((void **)net->mib.tcp_statistics,
1464 sizeof(struct tcp_mib)) < 0)
1466 if (snmp_mib_init((void **)net->mib.ip_statistics,
1467 sizeof(struct ipstats_mib)) < 0)
1469 if (snmp_mib_init((void **)net->mib.net_statistics,
1470 sizeof(struct linux_mib)) < 0)
1472 if (snmp_mib_init((void **)net->mib.udp_statistics,
1473 sizeof(struct udp_mib)) < 0)
1475 if (snmp_mib_init((void **)net->mib.udplite_statistics,
1476 sizeof(struct udp_mib)) < 0)
1477 goto err_udplite_mib;
1478 if (snmp_mib_init((void **)net->mib.icmp_statistics,
1479 sizeof(struct icmp_mib)) < 0)
1481 if (snmp_mib_init((void **)net->mib.icmpmsg_statistics,
1482 sizeof(struct icmpmsg_mib)) < 0)
1483 goto err_icmpmsg_mib;
1489 snmp_mib_free((void **)net->mib.icmp_statistics);
1491 snmp_mib_free((void **)net->mib.udplite_statistics);
1493 snmp_mib_free((void **)net->mib.udp_statistics);
1495 snmp_mib_free((void **)net->mib.net_statistics);
1497 snmp_mib_free((void **)net->mib.ip_statistics);
1499 snmp_mib_free((void **)net->mib.tcp_statistics);
1504 static __net_exit void ipv4_mib_exit_net(struct net *net)
1506 snmp_mib_free((void **)net->mib.icmpmsg_statistics);
1507 snmp_mib_free((void **)net->mib.icmp_statistics);
1508 snmp_mib_free((void **)net->mib.udplite_statistics);
1509 snmp_mib_free((void **)net->mib.udp_statistics);
1510 snmp_mib_free((void **)net->mib.net_statistics);
1511 snmp_mib_free((void **)net->mib.ip_statistics);
1512 snmp_mib_free((void **)net->mib.tcp_statistics);
1515 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1516 .init = ipv4_mib_init_net,
1517 .exit = ipv4_mib_exit_net,
1520 static int __init init_ipv4_mibs(void)
1522 return register_pernet_subsys(&ipv4_mib_ops);
1525 static int ipv4_proc_init(void);
1528 * IP protocol layer initialiser
1531 static struct packet_type ip_packet_type __read_mostly = {
1532 .type = cpu_to_be16(ETH_P_IP),
1534 .gso_send_check = inet_gso_send_check,
1535 .gso_segment = inet_gso_segment,
1536 .gro_receive = inet_gro_receive,
1537 .gro_complete = inet_gro_complete,
1540 static int __init inet_init(void)
1542 struct sk_buff *dummy_skb;
1543 struct inet_protosw *q;
1544 struct list_head *r;
1547 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1549 rc = proto_register(&tcp_prot, 1);
1553 rc = proto_register(&udp_prot, 1);
1555 goto out_unregister_tcp_proto;
1557 rc = proto_register(&raw_prot, 1);
1559 goto out_unregister_udp_proto;
1562 * Tell SOCKET that we are alive...
1565 (void)sock_register(&inet_family_ops);
1567 #ifdef CONFIG_SYSCTL
1568 ip_static_sysctl_init();
1572 * Add all the base protocols.
1575 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1576 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1577 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1578 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1579 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1580 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1581 #ifdef CONFIG_IP_MULTICAST
1582 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1583 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1586 /* Register the socket-side information for inet_create. */
1587 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1590 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1591 inet_register_protosw(q);
1594 * Set the ARP module up
1600 * Set the IP module up
1607 /* Setup TCP slab cache for open requests. */
1610 /* Setup UDP memory threshold */
1613 /* Add UDP-Lite (RFC 3828) */
1614 udplite4_register();
1617 * Set the ICMP layer up
1620 if (icmp_init() < 0)
1621 panic("Failed to create the ICMP control socket.\n");
1624 * Initialise the multicast router
1626 #if defined(CONFIG_IP_MROUTE)
1628 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1631 * Initialise per-cpu ipv4 mibs
1634 if (init_ipv4_mibs())
1635 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1641 dev_add_pack(&ip_packet_type);
1646 out_unregister_udp_proto:
1647 proto_unregister(&udp_prot);
1648 out_unregister_tcp_proto:
1649 proto_unregister(&tcp_prot);
1653 fs_initcall(inet_init);
1655 /* ------------------------------------------------------------------------ */
1657 #ifdef CONFIG_PROC_FS
1658 static int __init ipv4_proc_init(void)
1662 if (raw_proc_init())
1664 if (tcp4_proc_init())
1666 if (udp4_proc_init())
1668 if (ip_misc_proc_init())
1683 #else /* CONFIG_PROC_FS */
1684 static int __init ipv4_proc_init(void)
1688 #endif /* CONFIG_PROC_FS */
1690 MODULE_ALIAS_NETPROTO(PF_INET);