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.
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>
89 #include <linux/slab.h>
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/inetdevice.h>
97 #include <linux/netdevice.h>
98 #include <net/checksum.h>
100 #include <net/protocol.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
107 #include <net/udplite.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
111 #include <net/icmp.h>
112 #include <net/ipip.h>
113 #include <net/inet_common.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #ifdef CONFIG_IP_MROUTE
117 #include <linux/mroute.h>
121 /* The inetsw table contains everything that inet_create needs to
122 * build a new socket.
124 static struct list_head inetsw[SOCK_MAX];
125 static DEFINE_SPINLOCK(inetsw_lock);
127 struct ipv4_config ipv4_config;
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 pr_err("Attempt to release TCP socket in state %d %p\n",
146 if (!sock_flag(sk, SOCK_DEAD)) {
147 pr_err("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(rcu_dereference_check(sk->sk_dst_cache, 1));
158 sk_refcnt_debug_dec(sk);
160 EXPORT_SYMBOL(inet_sock_destruct);
163 * The routines beyond this point handle the behaviour of an AF_INET
164 * socket object. Mostly it punts to the subprotocols of IP to do
169 * Automatically bind an unbound socket.
172 static int inet_autobind(struct sock *sk)
174 struct inet_sock *inet;
175 /* We may need to bind the socket. */
178 if (!inet->inet_num) {
179 if (sk->sk_prot->get_port(sk, 0)) {
183 inet->inet_sport = htons(inet->inet_num);
190 * Move a socket into listening state.
192 int inet_listen(struct socket *sock, int backlog)
194 struct sock *sk = sock->sk;
195 unsigned char old_state;
201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 old_state = sk->sk_state;
205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 /* Really, if the socket is already in listen state
209 * we can only allow the backlog to be adjusted.
211 if (old_state != TCP_LISTEN) {
212 err = inet_csk_listen_start(sk, backlog);
216 sk->sk_max_ack_backlog = backlog;
223 EXPORT_SYMBOL(inet_listen);
225 u32 inet_ehash_secret __read_mostly;
226 EXPORT_SYMBOL(inet_ehash_secret);
229 * inet_ehash_secret must be set exactly once
231 void build_ehash_secret(void)
236 get_random_bytes(&rnd, sizeof(rnd));
239 cmpxchg(&inet_ehash_secret, 0, rnd);
241 EXPORT_SYMBOL(build_ehash_secret);
243 static inline int inet_netns_ok(struct net *net, int protocol)
246 const 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,
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 (unlikely(!inet_ehash_secret))
277 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
278 build_ehash_secret();
280 sock->state = SS_UNCONNECTED;
282 /* Look for the requested type/protocol pair. */
284 err = -ESOCKTNOSUPPORT;
286 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
289 /* Check the non-wild match. */
290 if (protocol == answer->protocol) {
291 if (protocol != IPPROTO_IP)
294 /* Check for the two wild cases. */
295 if (IPPROTO_IP == protocol) {
296 protocol = answer->protocol;
299 if (IPPROTO_IP == answer->protocol)
302 err = -EPROTONOSUPPORT;
306 if (try_loading_module < 2) {
309 * Be more specific, e.g. net-pf-2-proto-132-type-1
310 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
312 if (++try_loading_module == 1)
313 request_module("net-pf-%d-proto-%d-type-%d",
314 PF_INET, protocol, sock->type);
316 * Fall back to generic, e.g. net-pf-2-proto-132
317 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
320 request_module("net-pf-%d-proto-%d",
322 goto lookup_protocol;
328 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
332 if (!inet_netns_ok(net, protocol))
335 sock->ops = answer->ops;
336 answer_prot = answer->prot;
337 answer_no_check = answer->no_check;
338 answer_flags = answer->flags;
341 WARN_ON(answer_prot->slab == NULL);
344 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
349 sk->sk_no_check = answer_no_check;
350 if (INET_PROTOSW_REUSE & answer_flags)
354 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
358 if (SOCK_RAW == sock->type) {
359 inet->inet_num = protocol;
360 if (IPPROTO_RAW == protocol)
364 if (ipv4_config.no_pmtu_disc)
365 inet->pmtudisc = IP_PMTUDISC_DONT;
367 inet->pmtudisc = IP_PMTUDISC_WANT;
371 sock_init_data(sock, sk);
373 sk->sk_destruct = inet_sock_destruct;
374 sk->sk_protocol = protocol;
375 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
382 inet->mc_list = NULL;
384 sk_refcnt_debug_inc(sk);
386 if (inet->inet_num) {
387 /* It assumes that any protocol which allows
388 * the user to assign a number at socket
389 * creation time automatically
392 inet->inet_sport = htons(inet->inet_num);
393 /* Add to protocol hash chains. */
394 sk->sk_prot->hash(sk);
397 if (sk->sk_prot->init) {
398 err = sk->sk_prot->init(sk);
400 sk_common_release(sk);
411 * The peer socket should always be NULL (or else). When we call this
412 * function we are destroying the object and from then on nobody
413 * should refer to it.
415 int inet_release(struct socket *sock)
417 struct sock *sk = sock->sk;
422 sock_rps_reset_flow(sk);
424 /* Applications forget to leave groups before exiting */
425 ip_mc_drop_socket(sk);
427 /* If linger is set, we don't return until the close
428 * is complete. Otherwise we return immediately. The
429 * actually closing is done the same either way.
431 * If the close is due to the process exiting, we never
435 if (sock_flag(sk, SOCK_LINGER) &&
436 !(current->flags & PF_EXITING))
437 timeout = sk->sk_lingertime;
439 sk->sk_prot->close(sk, timeout);
443 EXPORT_SYMBOL(inet_release);
445 /* It is off by default, see below. */
446 int sysctl_ip_nonlocal_bind __read_mostly;
447 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
449 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
451 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
452 struct sock *sk = sock->sk;
453 struct inet_sock *inet = inet_sk(sk);
458 /* If the socket has its own bind function then use it. (RAW) */
459 if (sk->sk_prot->bind) {
460 err = sk->sk_prot->bind(sk, uaddr, addr_len);
464 if (addr_len < sizeof(struct sockaddr_in))
467 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
469 /* Not specified by any standard per-se, however it breaks too
470 * many applications when removed. It is unfortunate since
471 * allowing applications to make a non-local bind solves
472 * several problems with systems using dynamic addressing.
473 * (ie. your servers still start up even if your ISDN link
474 * is temporarily down)
476 err = -EADDRNOTAVAIL;
477 if (!sysctl_ip_nonlocal_bind &&
478 !(inet->freebind || inet->transparent) &&
479 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
480 chk_addr_ret != RTN_LOCAL &&
481 chk_addr_ret != RTN_MULTICAST &&
482 chk_addr_ret != RTN_BROADCAST)
485 snum = ntohs(addr->sin_port);
487 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
490 /* We keep a pair of addresses. rcv_saddr is the one
491 * used by hash lookups, and saddr is used for transmit.
493 * In the BSD API these are the same except where it
494 * would be illegal to use them (multicast/broadcast) in
495 * which case the sending device address is used.
499 /* Check these errors (active socket, double bind). */
501 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
502 goto out_release_sock;
504 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
505 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
506 inet->inet_saddr = 0; /* Use device */
508 /* Make sure we are allowed to bind here. */
509 if (sk->sk_prot->get_port(sk, snum)) {
510 inet->inet_saddr = inet->inet_rcv_saddr = 0;
512 goto out_release_sock;
515 if (inet->inet_rcv_saddr)
516 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
518 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
519 inet->inet_sport = htons(inet->inet_num);
520 inet->inet_daddr = 0;
521 inet->inet_dport = 0;
529 EXPORT_SYMBOL(inet_bind);
531 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
532 int addr_len, int flags)
534 struct sock *sk = sock->sk;
536 if (addr_len < sizeof(uaddr->sa_family))
538 if (uaddr->sa_family == AF_UNSPEC)
539 return sk->sk_prot->disconnect(sk, flags);
541 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
543 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
545 EXPORT_SYMBOL(inet_dgram_connect);
547 static long inet_wait_for_connect(struct sock *sk, long timeo)
551 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
553 /* Basic assumption: if someone sets sk->sk_err, he _must_
554 * change state of the socket from TCP_SYN_*.
555 * Connect() does not allow to get error notifications
556 * without closing the socket.
558 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
560 timeo = schedule_timeout(timeo);
562 if (signal_pending(current) || !timeo)
564 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
566 finish_wait(sk_sleep(sk), &wait);
571 * Connect to a remote host. There is regrettably still a little
572 * TCP 'magic' in here.
574 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
575 int addr_len, int flags)
577 struct sock *sk = sock->sk;
581 if (addr_len < sizeof(uaddr->sa_family))
586 if (uaddr->sa_family == AF_UNSPEC) {
587 err = sk->sk_prot->disconnect(sk, flags);
588 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
592 switch (sock->state) {
601 /* Fall out of switch with err, set for this state */
605 if (sk->sk_state != TCP_CLOSE)
608 err = sk->sk_prot->connect(sk, uaddr, addr_len);
612 sock->state = SS_CONNECTING;
614 /* Just entered SS_CONNECTING state; the only
615 * difference is that return value in non-blocking
616 * case is EINPROGRESS, rather than EALREADY.
622 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
624 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
625 /* Error code is set above */
626 if (!timeo || !inet_wait_for_connect(sk, timeo))
629 err = sock_intr_errno(timeo);
630 if (signal_pending(current))
634 /* Connection was closed by RST, timeout, ICMP error
635 * or another process disconnected us.
637 if (sk->sk_state == TCP_CLOSE)
640 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
641 * and error was received after socket entered established state.
642 * Hence, it is handled normally after connect() return successfully.
645 sock->state = SS_CONNECTED;
652 err = sock_error(sk) ? : -ECONNABORTED;
653 sock->state = SS_UNCONNECTED;
654 if (sk->sk_prot->disconnect(sk, flags))
655 sock->state = SS_DISCONNECTING;
658 EXPORT_SYMBOL(inet_stream_connect);
661 * Accept a pending connection. The TCP layer now gives BSD semantics.
664 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
666 struct sock *sk1 = sock->sk;
668 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
675 WARN_ON(!((1 << sk2->sk_state) &
676 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
678 sock_graft(sk2, newsock);
680 newsock->state = SS_CONNECTED;
686 EXPORT_SYMBOL(inet_accept);
690 * This does both peername and sockname.
692 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
693 int *uaddr_len, int peer)
695 struct sock *sk = sock->sk;
696 struct inet_sock *inet = inet_sk(sk);
697 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
699 sin->sin_family = AF_INET;
701 if (!inet->inet_dport ||
702 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
705 sin->sin_port = inet->inet_dport;
706 sin->sin_addr.s_addr = inet->inet_daddr;
708 __be32 addr = inet->inet_rcv_saddr;
710 addr = inet->inet_saddr;
711 sin->sin_port = inet->inet_sport;
712 sin->sin_addr.s_addr = addr;
714 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
715 *uaddr_len = sizeof(*sin);
718 EXPORT_SYMBOL(inet_getname);
720 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
723 struct sock *sk = sock->sk;
725 sock_rps_record_flow(sk);
727 /* We may need to bind the socket. */
728 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
732 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
734 EXPORT_SYMBOL(inet_sendmsg);
736 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
737 size_t size, int flags)
739 struct sock *sk = sock->sk;
741 sock_rps_record_flow(sk);
743 /* We may need to bind the socket. */
744 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
748 if (sk->sk_prot->sendpage)
749 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
750 return sock_no_sendpage(sock, page, offset, size, flags);
752 EXPORT_SYMBOL(inet_sendpage);
754 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
755 size_t size, int flags)
757 struct sock *sk = sock->sk;
761 sock_rps_record_flow(sk);
763 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
764 flags & ~MSG_DONTWAIT, &addr_len);
766 msg->msg_namelen = addr_len;
769 EXPORT_SYMBOL(inet_recvmsg);
771 int inet_shutdown(struct socket *sock, int how)
773 struct sock *sk = sock->sk;
776 /* This should really check to make sure
777 * the socket is a TCP socket. (WHY AC...)
779 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
782 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
786 if (sock->state == SS_CONNECTING) {
787 if ((1 << sk->sk_state) &
788 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
789 sock->state = SS_DISCONNECTING;
791 sock->state = SS_CONNECTED;
794 switch (sk->sk_state) {
797 /* Hack to wake up other listeners, who can poll for
798 POLLHUP, even on eg. unconnected UDP sockets -- RR */
800 sk->sk_shutdown |= how;
801 if (sk->sk_prot->shutdown)
802 sk->sk_prot->shutdown(sk, how);
805 /* Remaining two branches are temporary solution for missing
806 * close() in multithreaded environment. It is _not_ a good idea,
807 * but we have no choice until close() is repaired at VFS level.
810 if (!(how & RCV_SHUTDOWN))
814 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
815 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
819 /* Wake up anyone sleeping in poll. */
820 sk->sk_state_change(sk);
824 EXPORT_SYMBOL(inet_shutdown);
827 * ioctl() calls you can issue on an INET socket. Most of these are
828 * device configuration and stuff and very rarely used. Some ioctls
829 * pass on to the socket itself.
831 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
832 * loads the devconfigure module does its configuring and unloads it.
833 * There's a good 20K of config code hanging around the kernel.
836 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
838 struct sock *sk = sock->sk;
840 struct net *net = sock_net(sk);
844 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
847 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
852 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
857 err = arp_ioctl(net, cmd, (void __user *)arg);
870 err = devinet_ioctl(net, cmd, (void __user *)arg);
873 if (sk->sk_prot->ioctl)
874 err = sk->sk_prot->ioctl(sk, cmd, arg);
881 EXPORT_SYMBOL(inet_ioctl);
883 const struct proto_ops inet_stream_ops = {
885 .owner = THIS_MODULE,
886 .release = inet_release,
888 .connect = inet_stream_connect,
889 .socketpair = sock_no_socketpair,
890 .accept = inet_accept,
891 .getname = inet_getname,
894 .listen = inet_listen,
895 .shutdown = inet_shutdown,
896 .setsockopt = sock_common_setsockopt,
897 .getsockopt = sock_common_getsockopt,
898 .sendmsg = inet_sendmsg,
899 .recvmsg = inet_recvmsg,
900 .mmap = sock_no_mmap,
901 .sendpage = inet_sendpage,
902 .splice_read = tcp_splice_read,
904 .compat_setsockopt = compat_sock_common_setsockopt,
905 .compat_getsockopt = compat_sock_common_getsockopt,
908 EXPORT_SYMBOL(inet_stream_ops);
910 const struct proto_ops inet_dgram_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,
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 = inet_recvmsg,
927 .mmap = sock_no_mmap,
928 .sendpage = inet_sendpage,
930 .compat_setsockopt = compat_sock_common_setsockopt,
931 .compat_getsockopt = compat_sock_common_getsockopt,
934 EXPORT_SYMBOL(inet_dgram_ops);
937 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
940 static const struct proto_ops inet_sockraw_ops = {
942 .owner = THIS_MODULE,
943 .release = inet_release,
945 .connect = inet_dgram_connect,
946 .socketpair = sock_no_socketpair,
947 .accept = sock_no_accept,
948 .getname = inet_getname,
949 .poll = datagram_poll,
951 .listen = sock_no_listen,
952 .shutdown = inet_shutdown,
953 .setsockopt = sock_common_setsockopt,
954 .getsockopt = sock_common_getsockopt,
955 .sendmsg = inet_sendmsg,
956 .recvmsg = inet_recvmsg,
957 .mmap = sock_no_mmap,
958 .sendpage = inet_sendpage,
960 .compat_setsockopt = compat_sock_common_setsockopt,
961 .compat_getsockopt = compat_sock_common_getsockopt,
965 static const struct net_proto_family inet_family_ops = {
967 .create = inet_create,
968 .owner = THIS_MODULE,
971 /* Upon startup we insert all the elements in inetsw_array[] into
972 * the linked list inetsw.
974 static struct inet_protosw inetsw_array[] =
978 .protocol = IPPROTO_TCP,
980 .ops = &inet_stream_ops,
982 .flags = INET_PROTOSW_PERMANENT |
988 .protocol = IPPROTO_UDP,
990 .ops = &inet_dgram_ops,
991 .no_check = UDP_CSUM_DEFAULT,
992 .flags = INET_PROTOSW_PERMANENT,
998 .protocol = IPPROTO_IP, /* wild card */
1000 .ops = &inet_sockraw_ops,
1001 .no_check = UDP_CSUM_DEFAULT,
1002 .flags = INET_PROTOSW_REUSE,
1006 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1008 void inet_register_protosw(struct inet_protosw *p)
1010 struct list_head *lh;
1011 struct inet_protosw *answer;
1012 int protocol = p->protocol;
1013 struct list_head *last_perm;
1015 spin_lock_bh(&inetsw_lock);
1017 if (p->type >= SOCK_MAX)
1020 /* If we are trying to override a permanent protocol, bail. */
1022 last_perm = &inetsw[p->type];
1023 list_for_each(lh, &inetsw[p->type]) {
1024 answer = list_entry(lh, struct inet_protosw, list);
1026 /* Check only the non-wild match. */
1027 if (INET_PROTOSW_PERMANENT & answer->flags) {
1028 if (protocol == answer->protocol)
1038 /* Add the new entry after the last permanent entry if any, so that
1039 * the new entry does not override a permanent entry when matched with
1040 * a wild-card protocol. But it is allowed to override any existing
1041 * non-permanent entry. This means that when we remove this entry, the
1042 * system automatically returns to the old behavior.
1044 list_add_rcu(&p->list, last_perm);
1046 spin_unlock_bh(&inetsw_lock);
1051 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1057 "Ignoring attempt to register invalid socket type %d.\n",
1061 EXPORT_SYMBOL(inet_register_protosw);
1063 void inet_unregister_protosw(struct inet_protosw *p)
1065 if (INET_PROTOSW_PERMANENT & p->flags) {
1067 "Attempt to unregister permanent protocol %d.\n",
1070 spin_lock_bh(&inetsw_lock);
1071 list_del_rcu(&p->list);
1072 spin_unlock_bh(&inetsw_lock);
1077 EXPORT_SYMBOL(inet_unregister_protosw);
1080 * Shall we try to damage output packets if routing dev changes?
1083 int sysctl_ip_dynaddr __read_mostly;
1085 static int inet_sk_reselect_saddr(struct sock *sk)
1087 struct inet_sock *inet = inet_sk(sk);
1090 __be32 old_saddr = inet->inet_saddr;
1092 __be32 daddr = inet->inet_daddr;
1094 if (inet->opt && inet->opt->srr)
1095 daddr = inet->opt->faddr;
1097 /* Query new route. */
1098 err = ip_route_connect(&rt, daddr, 0,
1100 sk->sk_bound_dev_if,
1102 inet->inet_sport, inet->inet_dport, sk, 0);
1106 sk_setup_caps(sk, &rt->dst);
1108 new_saddr = rt->rt_src;
1110 if (new_saddr == old_saddr)
1113 if (sysctl_ip_dynaddr > 1) {
1114 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1115 __func__, &old_saddr, &new_saddr);
1118 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1121 * XXX The only one ugly spot where we need to
1122 * XXX really change the sockets identity after
1123 * XXX it has entered the hashes. -DaveM
1125 * Besides that, it does not check for connection
1126 * uniqueness. Wait for troubles.
1128 __sk_prot_rehash(sk);
1132 int inet_sk_rebuild_header(struct sock *sk)
1134 struct inet_sock *inet = inet_sk(sk);
1135 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1139 /* Route is OK, nothing to do. */
1144 daddr = inet->inet_daddr;
1145 if (inet->opt && inet->opt->srr)
1146 daddr = inet->opt->faddr;
1149 .oif = sk->sk_bound_dev_if,
1150 .mark = sk->sk_mark,
1152 .fl4_src = inet->inet_saddr,
1153 .fl4_tos = RT_CONN_FLAGS(sk),
1154 .proto = sk->sk_protocol,
1155 .flags = inet_sk_flowi_flags(sk),
1156 .fl_ip_sport = inet->inet_sport,
1157 .fl_ip_dport = inet->inet_dport,
1160 security_sk_classify_flow(sk, &fl);
1161 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1164 sk_setup_caps(sk, &rt->dst);
1166 /* Routing failed... */
1167 sk->sk_route_caps = 0;
1169 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1170 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1172 if (!sysctl_ip_dynaddr ||
1173 sk->sk_state != TCP_SYN_SENT ||
1174 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1175 (err = inet_sk_reselect_saddr(sk)) != 0)
1176 sk->sk_err_soft = -err;
1181 EXPORT_SYMBOL(inet_sk_rebuild_header);
1183 static int inet_gso_send_check(struct sk_buff *skb)
1186 const struct net_protocol *ops;
1191 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1196 if (ihl < sizeof(*iph))
1199 if (unlikely(!pskb_may_pull(skb, ihl)))
1202 __skb_pull(skb, ihl);
1203 skb_reset_transport_header(skb);
1205 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1206 err = -EPROTONOSUPPORT;
1209 ops = rcu_dereference(inet_protos[proto]);
1210 if (likely(ops && ops->gso_send_check))
1211 err = ops->gso_send_check(skb);
1218 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1220 struct sk_buff *segs = ERR_PTR(-EINVAL);
1222 const struct net_protocol *ops;
1226 unsigned int offset = 0;
1228 if (!(features & NETIF_F_V4_CSUM))
1229 features &= ~NETIF_F_SG;
1231 if (unlikely(skb_shinfo(skb)->gso_type &
1239 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1244 if (ihl < sizeof(*iph))
1247 if (unlikely(!pskb_may_pull(skb, ihl)))
1250 __skb_pull(skb, ihl);
1251 skb_reset_transport_header(skb);
1253 id = ntohs(iph->id);
1254 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1255 segs = ERR_PTR(-EPROTONOSUPPORT);
1258 ops = rcu_dereference(inet_protos[proto]);
1259 if (likely(ops && ops->gso_segment))
1260 segs = ops->gso_segment(skb, features);
1263 if (!segs || IS_ERR(segs))
1269 if (proto == IPPROTO_UDP) {
1270 iph->id = htons(id);
1271 iph->frag_off = htons(offset >> 3);
1272 if (skb->next != NULL)
1273 iph->frag_off |= htons(IP_MF);
1274 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1276 iph->id = htons(id++);
1277 iph->tot_len = htons(skb->len - skb->mac_len);
1279 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1280 } while ((skb = skb->next));
1286 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1287 struct sk_buff *skb)
1289 const struct net_protocol *ops;
1290 struct sk_buff **pp = NULL;
1299 off = skb_gro_offset(skb);
1300 hlen = off + sizeof(*iph);
1301 iph = skb_gro_header_fast(skb, off);
1302 if (skb_gro_header_hard(skb, hlen)) {
1303 iph = skb_gro_header_slow(skb, hlen, off);
1308 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1311 ops = rcu_dereference(inet_protos[proto]);
1312 if (!ops || !ops->gro_receive)
1315 if (*(u8 *)iph != 0x45)
1318 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1321 id = ntohl(*(__be32 *)&iph->id);
1322 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1325 for (p = *head; p; p = p->next) {
1328 if (!NAPI_GRO_CB(p)->same_flow)
1333 if ((iph->protocol ^ iph2->protocol) |
1334 (iph->tos ^ iph2->tos) |
1335 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1336 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1337 NAPI_GRO_CB(p)->same_flow = 0;
1341 /* All fields must match except length and checksum. */
1342 NAPI_GRO_CB(p)->flush |=
1343 (iph->ttl ^ iph2->ttl) |
1344 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1346 NAPI_GRO_CB(p)->flush |= flush;
1349 NAPI_GRO_CB(skb)->flush |= flush;
1350 skb_gro_pull(skb, sizeof(*iph));
1351 skb_set_transport_header(skb, skb_gro_offset(skb));
1353 pp = ops->gro_receive(head, skb);
1359 NAPI_GRO_CB(skb)->flush |= flush;
1364 static int inet_gro_complete(struct sk_buff *skb)
1366 const struct net_protocol *ops;
1367 struct iphdr *iph = ip_hdr(skb);
1368 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1370 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1372 csum_replace2(&iph->check, iph->tot_len, newlen);
1373 iph->tot_len = newlen;
1376 ops = rcu_dereference(inet_protos[proto]);
1377 if (WARN_ON(!ops || !ops->gro_complete))
1380 err = ops->gro_complete(skb);
1388 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1389 unsigned short type, unsigned char protocol,
1392 struct socket *sock;
1393 int rc = sock_create_kern(family, type, protocol, &sock);
1397 (*sk)->sk_allocation = GFP_ATOMIC;
1399 * Unhash it so that IP input processing does not even see it,
1400 * we do not wish this socket to see incoming packets.
1402 (*sk)->sk_prot->unhash(*sk);
1404 sk_change_net(*sk, net);
1408 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1410 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1412 unsigned long res = 0;
1415 for_each_possible_cpu(i) {
1416 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1417 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1421 EXPORT_SYMBOL_GPL(snmp_fold_field);
1423 #if BITS_PER_LONG==32
1425 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1430 for_each_possible_cpu(cpu) {
1431 void *bhptr, *userptr;
1432 struct u64_stats_sync *syncp;
1436 /* first mib used by softirq context, we must use _bh() accessors */
1437 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1438 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1440 start = u64_stats_fetch_begin_bh(syncp);
1441 v_bh = *(((u64 *) bhptr) + offt);
1442 } while (u64_stats_fetch_retry_bh(syncp, start));
1444 /* second mib used in USER context */
1445 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1446 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1448 start = u64_stats_fetch_begin(syncp);
1449 v_user = *(((u64 *) userptr) + offt);
1450 } while (u64_stats_fetch_retry(syncp, start));
1452 res += v_bh + v_user;
1456 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1459 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1461 BUG_ON(ptr == NULL);
1462 ptr[0] = __alloc_percpu(mibsize, align);
1465 ptr[1] = __alloc_percpu(mibsize, align);
1470 free_percpu(ptr[0]);
1475 EXPORT_SYMBOL_GPL(snmp_mib_init);
1477 void snmp_mib_free(void __percpu *ptr[2])
1479 BUG_ON(ptr == NULL);
1480 free_percpu(ptr[0]);
1481 free_percpu(ptr[1]);
1482 ptr[0] = ptr[1] = NULL;
1484 EXPORT_SYMBOL_GPL(snmp_mib_free);
1486 #ifdef CONFIG_IP_MULTICAST
1487 static const struct net_protocol igmp_protocol = {
1488 .handler = igmp_rcv,
1493 static const struct net_protocol tcp_protocol = {
1494 .handler = tcp_v4_rcv,
1495 .err_handler = tcp_v4_err,
1496 .gso_send_check = tcp_v4_gso_send_check,
1497 .gso_segment = tcp_tso_segment,
1498 .gro_receive = tcp4_gro_receive,
1499 .gro_complete = tcp4_gro_complete,
1504 static const struct net_protocol udp_protocol = {
1506 .err_handler = udp_err,
1507 .gso_send_check = udp4_ufo_send_check,
1508 .gso_segment = udp4_ufo_fragment,
1513 static const struct net_protocol icmp_protocol = {
1514 .handler = icmp_rcv,
1519 static __net_init int ipv4_mib_init_net(struct net *net)
1521 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1522 sizeof(struct tcp_mib),
1523 __alignof__(struct tcp_mib)) < 0)
1525 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1526 sizeof(struct ipstats_mib),
1527 __alignof__(struct ipstats_mib)) < 0)
1529 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1530 sizeof(struct linux_mib),
1531 __alignof__(struct linux_mib)) < 0)
1533 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1534 sizeof(struct udp_mib),
1535 __alignof__(struct udp_mib)) < 0)
1537 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1538 sizeof(struct udp_mib),
1539 __alignof__(struct udp_mib)) < 0)
1540 goto err_udplite_mib;
1541 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1542 sizeof(struct icmp_mib),
1543 __alignof__(struct icmp_mib)) < 0)
1545 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1546 sizeof(struct icmpmsg_mib),
1547 __alignof__(struct icmpmsg_mib)) < 0)
1548 goto err_icmpmsg_mib;
1554 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1556 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1558 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1560 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1562 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1564 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1569 static __net_exit void ipv4_mib_exit_net(struct net *net)
1571 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1572 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1573 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1574 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1575 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1576 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1577 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1580 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1581 .init = ipv4_mib_init_net,
1582 .exit = ipv4_mib_exit_net,
1585 static int __init init_ipv4_mibs(void)
1587 return register_pernet_subsys(&ipv4_mib_ops);
1590 static int ipv4_proc_init(void);
1593 * IP protocol layer initialiser
1596 static struct packet_type ip_packet_type __read_mostly = {
1597 .type = cpu_to_be16(ETH_P_IP),
1599 .gso_send_check = inet_gso_send_check,
1600 .gso_segment = inet_gso_segment,
1601 .gro_receive = inet_gro_receive,
1602 .gro_complete = inet_gro_complete,
1605 static int __init inet_init(void)
1607 struct sk_buff *dummy_skb;
1608 struct inet_protosw *q;
1609 struct list_head *r;
1612 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1614 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1615 if (!sysctl_local_reserved_ports)
1618 rc = proto_register(&tcp_prot, 1);
1620 goto out_free_reserved_ports;
1622 rc = proto_register(&udp_prot, 1);
1624 goto out_unregister_tcp_proto;
1626 rc = proto_register(&raw_prot, 1);
1628 goto out_unregister_udp_proto;
1631 * Tell SOCKET that we are alive...
1634 (void)sock_register(&inet_family_ops);
1636 #ifdef CONFIG_SYSCTL
1637 ip_static_sysctl_init();
1641 * Add all the base protocols.
1644 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1645 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1646 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1647 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1648 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1649 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1650 #ifdef CONFIG_IP_MULTICAST
1651 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1652 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1655 /* Register the socket-side information for inet_create. */
1656 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1659 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1660 inet_register_protosw(q);
1663 * Set the ARP module up
1669 * Set the IP module up
1676 /* Setup TCP slab cache for open requests. */
1679 /* Setup UDP memory threshold */
1682 /* Add UDP-Lite (RFC 3828) */
1683 udplite4_register();
1686 * Set the ICMP layer up
1689 if (icmp_init() < 0)
1690 panic("Failed to create the ICMP control socket.\n");
1693 * Initialise the multicast router
1695 #if defined(CONFIG_IP_MROUTE)
1697 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1700 * Initialise per-cpu ipv4 mibs
1703 if (init_ipv4_mibs())
1704 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1710 dev_add_pack(&ip_packet_type);
1715 out_unregister_udp_proto:
1716 proto_unregister(&udp_prot);
1717 out_unregister_tcp_proto:
1718 proto_unregister(&tcp_prot);
1719 out_free_reserved_ports:
1720 kfree(sysctl_local_reserved_ports);
1724 fs_initcall(inet_init);
1726 /* ------------------------------------------------------------------------ */
1728 #ifdef CONFIG_PROC_FS
1729 static int __init ipv4_proc_init(void)
1733 if (raw_proc_init())
1735 if (tcp4_proc_init())
1737 if (udp4_proc_init())
1739 if (ip_misc_proc_init())
1754 #else /* CONFIG_PROC_FS */
1755 static int __init ipv4_proc_init(void)
1759 #endif /* CONFIG_PROC_FS */
1761 MODULE_ALIAS_NETPROTO(PF_INET);