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>
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 <net/ping.h>
109 #include <linux/skbuff.h>
110 #include <net/sock.h>
112 #include <net/icmp.h>
113 #include <net/ipip.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
122 /* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
125 static struct list_head inetsw[SOCK_MAX];
126 static DEFINE_SPINLOCK(inetsw_lock);
128 struct ipv4_config ipv4_config;
129 EXPORT_SYMBOL(ipv4_config);
131 /* New destruction routine */
133 void inet_sock_destruct(struct sock *sk)
135 struct inet_sock *inet = inet_sk(sk);
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 pr_err("Attempt to release TCP socket in state %d %p\n",
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 pr_err("Attempt to release alive inet socket %p\n", sk);
152 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 WARN_ON(sk->sk_wmem_queued);
155 WARN_ON(sk->sk_forward_alloc);
157 kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 sk_refcnt_debug_dec(sk);
161 EXPORT_SYMBOL(inet_sock_destruct);
164 * The routines beyond this point handle the behaviour of an AF_INET
165 * socket object. Mostly it punts to the subprotocols of IP to do
170 * Automatically bind an unbound socket.
173 static int inet_autobind(struct sock *sk)
175 struct inet_sock *inet;
176 /* We may need to bind the socket. */
179 if (!inet->inet_num) {
180 if (sk->sk_prot->get_port(sk, 0)) {
184 inet->inet_sport = htons(inet->inet_num);
191 * Move a socket into listening state.
193 int inet_listen(struct socket *sock, int backlog)
195 struct sock *sk = sock->sk;
196 unsigned char old_state;
202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 old_state = sk->sk_state;
206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 /* Really, if the socket is already in listen state
210 * we can only allow the backlog to be adjusted.
212 if (old_state != TCP_LISTEN) {
213 err = inet_csk_listen_start(sk, backlog);
217 sk->sk_max_ack_backlog = backlog;
224 EXPORT_SYMBOL(inet_listen);
226 u32 inet_ehash_secret __read_mostly;
227 EXPORT_SYMBOL(inet_ehash_secret);
229 u32 ipv6_hash_secret __read_mostly;
230 EXPORT_SYMBOL(ipv6_hash_secret);
233 * inet_ehash_secret must be set exactly once, and to a non nul value
234 * ipv6_hash_secret must be set exactly once.
236 void build_ehash_secret(void)
241 get_random_bytes(&rnd, sizeof(rnd));
244 if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
245 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
247 EXPORT_SYMBOL(build_ehash_secret);
249 static inline int inet_netns_ok(struct net *net, int protocol)
252 const struct net_protocol *ipprot;
254 if (net_eq(net, &init_net))
257 hash = protocol & (MAX_INET_PROTOS - 1);
258 ipprot = rcu_dereference(inet_protos[hash]);
263 return ipprot->netns_ok;
267 * Create an inet socket.
270 static int inet_create(struct net *net, struct socket *sock, int protocol,
274 struct inet_protosw *answer;
275 struct inet_sock *inet;
276 struct proto *answer_prot;
277 unsigned char answer_flags;
278 char answer_no_check;
279 int try_loading_module = 0;
282 if (unlikely(!inet_ehash_secret))
283 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
284 build_ehash_secret();
286 sock->state = SS_UNCONNECTED;
288 /* Look for the requested type/protocol pair. */
290 err = -ESOCKTNOSUPPORT;
292 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
295 /* Check the non-wild match. */
296 if (protocol == answer->protocol) {
297 if (protocol != IPPROTO_IP)
300 /* Check for the two wild cases. */
301 if (IPPROTO_IP == protocol) {
302 protocol = answer->protocol;
305 if (IPPROTO_IP == answer->protocol)
308 err = -EPROTONOSUPPORT;
312 if (try_loading_module < 2) {
315 * Be more specific, e.g. net-pf-2-proto-132-type-1
316 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
318 if (++try_loading_module == 1)
319 request_module("net-pf-%d-proto-%d-type-%d",
320 PF_INET, protocol, sock->type);
322 * Fall back to generic, e.g. net-pf-2-proto-132
323 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
326 request_module("net-pf-%d-proto-%d",
328 goto lookup_protocol;
334 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
338 if (!inet_netns_ok(net, protocol))
341 sock->ops = answer->ops;
342 answer_prot = answer->prot;
343 answer_no_check = answer->no_check;
344 answer_flags = answer->flags;
347 WARN_ON(answer_prot->slab == NULL);
350 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
355 sk->sk_no_check = answer_no_check;
356 if (INET_PROTOSW_REUSE & answer_flags)
360 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
364 if (SOCK_RAW == sock->type) {
365 inet->inet_num = protocol;
366 if (IPPROTO_RAW == protocol)
370 if (ipv4_config.no_pmtu_disc)
371 inet->pmtudisc = IP_PMTUDISC_DONT;
373 inet->pmtudisc = IP_PMTUDISC_WANT;
377 sock_init_data(sock, sk);
379 sk->sk_destruct = inet_sock_destruct;
380 sk->sk_protocol = protocol;
381 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
388 inet->mc_list = NULL;
390 sk_refcnt_debug_inc(sk);
392 if (inet->inet_num) {
393 /* It assumes that any protocol which allows
394 * the user to assign a number at socket
395 * creation time automatically
398 inet->inet_sport = htons(inet->inet_num);
399 /* Add to protocol hash chains. */
400 sk->sk_prot->hash(sk);
403 if (sk->sk_prot->init) {
404 err = sk->sk_prot->init(sk);
406 sk_common_release(sk);
417 * The peer socket should always be NULL (or else). When we call this
418 * function we are destroying the object and from then on nobody
419 * should refer to it.
421 int inet_release(struct socket *sock)
423 struct sock *sk = sock->sk;
428 sock_rps_reset_flow(sk);
430 /* Applications forget to leave groups before exiting */
431 ip_mc_drop_socket(sk);
433 /* If linger is set, we don't return until the close
434 * is complete. Otherwise we return immediately. The
435 * actually closing is done the same either way.
437 * If the close is due to the process exiting, we never
441 if (sock_flag(sk, SOCK_LINGER) &&
442 !(current->flags & PF_EXITING))
443 timeout = sk->sk_lingertime;
445 sk->sk_prot->close(sk, timeout);
449 EXPORT_SYMBOL(inet_release);
451 /* It is off by default, see below. */
452 int sysctl_ip_nonlocal_bind __read_mostly;
453 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
455 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
457 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
458 struct sock *sk = sock->sk;
459 struct inet_sock *inet = inet_sk(sk);
464 /* If the socket has its own bind function then use it. (RAW) */
465 if (sk->sk_prot->bind) {
466 err = sk->sk_prot->bind(sk, uaddr, addr_len);
470 if (addr_len < sizeof(struct sockaddr_in))
473 if (addr->sin_family != AF_INET) {
474 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
475 * only if s_addr is INADDR_ANY.
478 if (addr->sin_family != AF_UNSPEC ||
479 addr->sin_addr.s_addr != htonl(INADDR_ANY))
483 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
485 /* Not specified by any standard per-se, however it breaks too
486 * many applications when removed. It is unfortunate since
487 * allowing applications to make a non-local bind solves
488 * several problems with systems using dynamic addressing.
489 * (ie. your servers still start up even if your ISDN link
490 * is temporarily down)
492 err = -EADDRNOTAVAIL;
493 if (!sysctl_ip_nonlocal_bind &&
494 !(inet->freebind || inet->transparent) &&
495 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
496 chk_addr_ret != RTN_LOCAL &&
497 chk_addr_ret != RTN_MULTICAST &&
498 chk_addr_ret != RTN_BROADCAST)
501 snum = ntohs(addr->sin_port);
503 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
506 /* We keep a pair of addresses. rcv_saddr is the one
507 * used by hash lookups, and saddr is used for transmit.
509 * In the BSD API these are the same except where it
510 * would be illegal to use them (multicast/broadcast) in
511 * which case the sending device address is used.
515 /* Check these errors (active socket, double bind). */
517 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
518 goto out_release_sock;
520 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
521 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
522 inet->inet_saddr = 0; /* Use device */
524 /* Make sure we are allowed to bind here. */
525 if (sk->sk_prot->get_port(sk, snum)) {
526 inet->inet_saddr = inet->inet_rcv_saddr = 0;
528 goto out_release_sock;
531 if (inet->inet_rcv_saddr)
532 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
534 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
535 inet->inet_sport = htons(inet->inet_num);
536 inet->inet_daddr = 0;
537 inet->inet_dport = 0;
545 EXPORT_SYMBOL(inet_bind);
547 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
548 int addr_len, int flags)
550 struct sock *sk = sock->sk;
552 if (addr_len < sizeof(uaddr->sa_family))
554 if (uaddr->sa_family == AF_UNSPEC)
555 return sk->sk_prot->disconnect(sk, flags);
557 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
559 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
561 EXPORT_SYMBOL(inet_dgram_connect);
563 static long inet_wait_for_connect(struct sock *sk, long timeo)
567 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
569 /* Basic assumption: if someone sets sk->sk_err, he _must_
570 * change state of the socket from TCP_SYN_*.
571 * Connect() does not allow to get error notifications
572 * without closing the socket.
574 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
576 timeo = schedule_timeout(timeo);
578 if (signal_pending(current) || !timeo)
580 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
582 finish_wait(sk_sleep(sk), &wait);
587 * Connect to a remote host. There is regrettably still a little
588 * TCP 'magic' in here.
590 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
591 int addr_len, int flags)
593 struct sock *sk = sock->sk;
597 if (addr_len < sizeof(uaddr->sa_family))
602 if (uaddr->sa_family == AF_UNSPEC) {
603 err = sk->sk_prot->disconnect(sk, flags);
604 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
608 switch (sock->state) {
617 /* Fall out of switch with err, set for this state */
621 if (sk->sk_state != TCP_CLOSE)
624 err = sk->sk_prot->connect(sk, uaddr, addr_len);
628 sock->state = SS_CONNECTING;
630 /* Just entered SS_CONNECTING state; the only
631 * difference is that return value in non-blocking
632 * case is EINPROGRESS, rather than EALREADY.
638 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
640 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
641 /* Error code is set above */
642 if (!timeo || !inet_wait_for_connect(sk, timeo))
645 err = sock_intr_errno(timeo);
646 if (signal_pending(current))
650 /* Connection was closed by RST, timeout, ICMP error
651 * or another process disconnected us.
653 if (sk->sk_state == TCP_CLOSE)
656 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
657 * and error was received after socket entered established state.
658 * Hence, it is handled normally after connect() return successfully.
661 sock->state = SS_CONNECTED;
668 err = sock_error(sk) ? : -ECONNABORTED;
669 sock->state = SS_UNCONNECTED;
670 if (sk->sk_prot->disconnect(sk, flags))
671 sock->state = SS_DISCONNECTING;
674 EXPORT_SYMBOL(inet_stream_connect);
677 * Accept a pending connection. The TCP layer now gives BSD semantics.
680 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
682 struct sock *sk1 = sock->sk;
684 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
691 sock_rps_record_flow(sk2);
692 WARN_ON(!((1 << sk2->sk_state) &
693 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
695 sock_graft(sk2, newsock);
697 newsock->state = SS_CONNECTED;
703 EXPORT_SYMBOL(inet_accept);
707 * This does both peername and sockname.
709 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
710 int *uaddr_len, int peer)
712 struct sock *sk = sock->sk;
713 struct inet_sock *inet = inet_sk(sk);
714 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
716 sin->sin_family = AF_INET;
718 if (!inet->inet_dport ||
719 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
722 sin->sin_port = inet->inet_dport;
723 sin->sin_addr.s_addr = inet->inet_daddr;
725 __be32 addr = inet->inet_rcv_saddr;
727 addr = inet->inet_saddr;
728 sin->sin_port = inet->inet_sport;
729 sin->sin_addr.s_addr = addr;
731 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
732 *uaddr_len = sizeof(*sin);
735 EXPORT_SYMBOL(inet_getname);
737 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
740 struct sock *sk = sock->sk;
742 sock_rps_record_flow(sk);
744 /* We may need to bind the socket. */
745 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
749 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
751 EXPORT_SYMBOL(inet_sendmsg);
753 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
754 size_t size, int flags)
756 struct sock *sk = sock->sk;
758 sock_rps_record_flow(sk);
760 /* We may need to bind the socket. */
761 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
765 if (sk->sk_prot->sendpage)
766 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
767 return sock_no_sendpage(sock, page, offset, size, flags);
769 EXPORT_SYMBOL(inet_sendpage);
771 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
772 size_t size, int flags)
774 struct sock *sk = sock->sk;
778 sock_rps_record_flow(sk);
780 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
781 flags & ~MSG_DONTWAIT, &addr_len);
783 msg->msg_namelen = addr_len;
786 EXPORT_SYMBOL(inet_recvmsg);
788 int inet_shutdown(struct socket *sock, int how)
790 struct sock *sk = sock->sk;
793 /* This should really check to make sure
794 * the socket is a TCP socket. (WHY AC...)
796 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
799 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
803 if (sock->state == SS_CONNECTING) {
804 if ((1 << sk->sk_state) &
805 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
806 sock->state = SS_DISCONNECTING;
808 sock->state = SS_CONNECTED;
811 switch (sk->sk_state) {
814 /* Hack to wake up other listeners, who can poll for
815 POLLHUP, even on eg. unconnected UDP sockets -- RR */
817 sk->sk_shutdown |= how;
818 if (sk->sk_prot->shutdown)
819 sk->sk_prot->shutdown(sk, how);
822 /* Remaining two branches are temporary solution for missing
823 * close() in multithreaded environment. It is _not_ a good idea,
824 * but we have no choice until close() is repaired at VFS level.
827 if (!(how & RCV_SHUTDOWN))
831 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
832 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
836 /* Wake up anyone sleeping in poll. */
837 sk->sk_state_change(sk);
841 EXPORT_SYMBOL(inet_shutdown);
844 * ioctl() calls you can issue on an INET socket. Most of these are
845 * device configuration and stuff and very rarely used. Some ioctls
846 * pass on to the socket itself.
848 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
849 * loads the devconfigure module does its configuring and unloads it.
850 * There's a good 20K of config code hanging around the kernel.
853 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
855 struct sock *sk = sock->sk;
857 struct net *net = sock_net(sk);
861 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
864 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
869 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
874 err = arp_ioctl(net, cmd, (void __user *)arg);
887 err = devinet_ioctl(net, cmd, (void __user *)arg);
890 if (sk->sk_prot->ioctl)
891 err = sk->sk_prot->ioctl(sk, cmd, arg);
898 EXPORT_SYMBOL(inet_ioctl);
901 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
903 struct sock *sk = sock->sk;
904 int err = -ENOIOCTLCMD;
906 if (sk->sk_prot->compat_ioctl)
907 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
913 const struct proto_ops inet_stream_ops = {
915 .owner = THIS_MODULE,
916 .release = inet_release,
918 .connect = inet_stream_connect,
919 .socketpair = sock_no_socketpair,
920 .accept = inet_accept,
921 .getname = inet_getname,
924 .listen = inet_listen,
925 .shutdown = inet_shutdown,
926 .setsockopt = sock_common_setsockopt,
927 .getsockopt = sock_common_getsockopt,
928 .sendmsg = inet_sendmsg,
929 .recvmsg = inet_recvmsg,
930 .mmap = sock_no_mmap,
931 .sendpage = inet_sendpage,
932 .splice_read = tcp_splice_read,
934 .compat_setsockopt = compat_sock_common_setsockopt,
935 .compat_getsockopt = compat_sock_common_getsockopt,
936 .compat_ioctl = inet_compat_ioctl,
939 EXPORT_SYMBOL(inet_stream_ops);
941 const struct proto_ops inet_dgram_ops = {
943 .owner = THIS_MODULE,
944 .release = inet_release,
946 .connect = inet_dgram_connect,
947 .socketpair = sock_no_socketpair,
948 .accept = sock_no_accept,
949 .getname = inet_getname,
952 .listen = sock_no_listen,
953 .shutdown = inet_shutdown,
954 .setsockopt = sock_common_setsockopt,
955 .getsockopt = sock_common_getsockopt,
956 .sendmsg = inet_sendmsg,
957 .recvmsg = inet_recvmsg,
958 .mmap = sock_no_mmap,
959 .sendpage = inet_sendpage,
961 .compat_setsockopt = compat_sock_common_setsockopt,
962 .compat_getsockopt = compat_sock_common_getsockopt,
963 .compat_ioctl = inet_compat_ioctl,
966 EXPORT_SYMBOL(inet_dgram_ops);
969 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
972 static const struct proto_ops inet_sockraw_ops = {
974 .owner = THIS_MODULE,
975 .release = inet_release,
977 .connect = inet_dgram_connect,
978 .socketpair = sock_no_socketpair,
979 .accept = sock_no_accept,
980 .getname = inet_getname,
981 .poll = datagram_poll,
983 .listen = sock_no_listen,
984 .shutdown = inet_shutdown,
985 .setsockopt = sock_common_setsockopt,
986 .getsockopt = sock_common_getsockopt,
987 .sendmsg = inet_sendmsg,
988 .recvmsg = inet_recvmsg,
989 .mmap = sock_no_mmap,
990 .sendpage = inet_sendpage,
992 .compat_setsockopt = compat_sock_common_setsockopt,
993 .compat_getsockopt = compat_sock_common_getsockopt,
994 .compat_ioctl = inet_compat_ioctl,
998 static const struct net_proto_family inet_family_ops = {
1000 .create = inet_create,
1001 .owner = THIS_MODULE,
1004 /* Upon startup we insert all the elements in inetsw_array[] into
1005 * the linked list inetsw.
1007 static struct inet_protosw inetsw_array[] =
1010 .type = SOCK_STREAM,
1011 .protocol = IPPROTO_TCP,
1013 .ops = &inet_stream_ops,
1015 .flags = INET_PROTOSW_PERMANENT |
1021 .protocol = IPPROTO_UDP,
1023 .ops = &inet_dgram_ops,
1024 .no_check = UDP_CSUM_DEFAULT,
1025 .flags = INET_PROTOSW_PERMANENT,
1030 .protocol = IPPROTO_ICMP,
1032 .ops = &inet_dgram_ops,
1033 .no_check = UDP_CSUM_DEFAULT,
1034 .flags = INET_PROTOSW_REUSE,
1039 .protocol = IPPROTO_IP, /* wild card */
1041 .ops = &inet_sockraw_ops,
1042 .no_check = UDP_CSUM_DEFAULT,
1043 .flags = INET_PROTOSW_REUSE,
1047 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1049 void inet_register_protosw(struct inet_protosw *p)
1051 struct list_head *lh;
1052 struct inet_protosw *answer;
1053 int protocol = p->protocol;
1054 struct list_head *last_perm;
1056 spin_lock_bh(&inetsw_lock);
1058 if (p->type >= SOCK_MAX)
1061 /* If we are trying to override a permanent protocol, bail. */
1063 last_perm = &inetsw[p->type];
1064 list_for_each(lh, &inetsw[p->type]) {
1065 answer = list_entry(lh, struct inet_protosw, list);
1067 /* Check only the non-wild match. */
1068 if (INET_PROTOSW_PERMANENT & answer->flags) {
1069 if (protocol == answer->protocol)
1079 /* Add the new entry after the last permanent entry if any, so that
1080 * the new entry does not override a permanent entry when matched with
1081 * a wild-card protocol. But it is allowed to override any existing
1082 * non-permanent entry. This means that when we remove this entry, the
1083 * system automatically returns to the old behavior.
1085 list_add_rcu(&p->list, last_perm);
1087 spin_unlock_bh(&inetsw_lock);
1092 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1098 "Ignoring attempt to register invalid socket type %d.\n",
1102 EXPORT_SYMBOL(inet_register_protosw);
1104 void inet_unregister_protosw(struct inet_protosw *p)
1106 if (INET_PROTOSW_PERMANENT & p->flags) {
1108 "Attempt to unregister permanent protocol %d.\n",
1111 spin_lock_bh(&inetsw_lock);
1112 list_del_rcu(&p->list);
1113 spin_unlock_bh(&inetsw_lock);
1118 EXPORT_SYMBOL(inet_unregister_protosw);
1121 * Shall we try to damage output packets if routing dev changes?
1124 int sysctl_ip_dynaddr __read_mostly;
1126 static int inet_sk_reselect_saddr(struct sock *sk)
1128 struct inet_sock *inet = inet_sk(sk);
1129 __be32 old_saddr = inet->inet_saddr;
1130 __be32 daddr = inet->inet_daddr;
1134 struct ip_options_rcu *inet_opt;
1136 inet_opt = rcu_dereference_protected(inet->inet_opt,
1137 sock_owned_by_user(sk));
1138 if (inet_opt && inet_opt->opt.srr)
1139 daddr = inet_opt->opt.faddr;
1141 /* Query new route. */
1142 fl4 = &inet->cork.fl.u.ip4;
1143 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1144 sk->sk_bound_dev_if, sk->sk_protocol,
1145 inet->inet_sport, inet->inet_dport, sk, false);
1149 sk_setup_caps(sk, &rt->dst);
1151 new_saddr = fl4->saddr;
1153 if (new_saddr == old_saddr)
1156 if (sysctl_ip_dynaddr > 1) {
1157 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1158 __func__, &old_saddr, &new_saddr);
1161 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1164 * XXX The only one ugly spot where we need to
1165 * XXX really change the sockets identity after
1166 * XXX it has entered the hashes. -DaveM
1168 * Besides that, it does not check for connection
1169 * uniqueness. Wait for troubles.
1171 __sk_prot_rehash(sk);
1175 int inet_sk_rebuild_header(struct sock *sk)
1177 struct inet_sock *inet = inet_sk(sk);
1178 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1180 struct ip_options_rcu *inet_opt;
1184 /* Route is OK, nothing to do. */
1190 inet_opt = rcu_dereference(inet->inet_opt);
1191 daddr = inet->inet_daddr;
1192 if (inet_opt && inet_opt->opt.srr)
1193 daddr = inet_opt->opt.faddr;
1195 fl4 = &inet->cork.fl.u.ip4;
1196 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1197 inet->inet_dport, inet->inet_sport,
1198 sk->sk_protocol, RT_CONN_FLAGS(sk),
1199 sk->sk_bound_dev_if);
1202 sk_setup_caps(sk, &rt->dst);
1206 /* Routing failed... */
1207 sk->sk_route_caps = 0;
1209 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1210 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1212 if (!sysctl_ip_dynaddr ||
1213 sk->sk_state != TCP_SYN_SENT ||
1214 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1215 (err = inet_sk_reselect_saddr(sk)) != 0)
1216 sk->sk_err_soft = -err;
1221 EXPORT_SYMBOL(inet_sk_rebuild_header);
1223 static int inet_gso_send_check(struct sk_buff *skb)
1225 const struct iphdr *iph;
1226 const struct net_protocol *ops;
1231 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1236 if (ihl < sizeof(*iph))
1239 if (unlikely(!pskb_may_pull(skb, ihl)))
1242 __skb_pull(skb, ihl);
1243 skb_reset_transport_header(skb);
1245 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1246 err = -EPROTONOSUPPORT;
1249 ops = rcu_dereference(inet_protos[proto]);
1250 if (likely(ops && ops->gso_send_check))
1251 err = ops->gso_send_check(skb);
1258 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1260 struct sk_buff *segs = ERR_PTR(-EINVAL);
1262 const struct net_protocol *ops;
1266 unsigned int offset = 0;
1268 if (!(features & NETIF_F_V4_CSUM))
1269 features &= ~NETIF_F_SG;
1271 if (unlikely(skb_shinfo(skb)->gso_type &
1279 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1284 if (ihl < sizeof(*iph))
1287 if (unlikely(!pskb_may_pull(skb, ihl)))
1290 __skb_pull(skb, ihl);
1291 skb_reset_transport_header(skb);
1293 id = ntohs(iph->id);
1294 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1295 segs = ERR_PTR(-EPROTONOSUPPORT);
1298 ops = rcu_dereference(inet_protos[proto]);
1299 if (likely(ops && ops->gso_segment))
1300 segs = ops->gso_segment(skb, features);
1303 if (!segs || IS_ERR(segs))
1309 if (proto == IPPROTO_UDP) {
1310 iph->id = htons(id);
1311 iph->frag_off = htons(offset >> 3);
1312 if (skb->next != NULL)
1313 iph->frag_off |= htons(IP_MF);
1314 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1316 iph->id = htons(id++);
1317 iph->tot_len = htons(skb->len - skb->mac_len);
1319 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1320 } while ((skb = skb->next));
1326 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1327 struct sk_buff *skb)
1329 const struct net_protocol *ops;
1330 struct sk_buff **pp = NULL;
1332 const struct iphdr *iph;
1339 off = skb_gro_offset(skb);
1340 hlen = off + sizeof(*iph);
1341 iph = skb_gro_header_fast(skb, off);
1342 if (skb_gro_header_hard(skb, hlen)) {
1343 iph = skb_gro_header_slow(skb, hlen, off);
1348 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1351 ops = rcu_dereference(inet_protos[proto]);
1352 if (!ops || !ops->gro_receive)
1355 if (*(u8 *)iph != 0x45)
1358 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1361 id = ntohl(*(__be32 *)&iph->id);
1362 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1365 for (p = *head; p; p = p->next) {
1368 if (!NAPI_GRO_CB(p)->same_flow)
1373 if ((iph->protocol ^ iph2->protocol) |
1374 (iph->tos ^ iph2->tos) |
1375 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1376 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1377 NAPI_GRO_CB(p)->same_flow = 0;
1381 /* All fields must match except length and checksum. */
1382 NAPI_GRO_CB(p)->flush |=
1383 (iph->ttl ^ iph2->ttl) |
1384 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1386 NAPI_GRO_CB(p)->flush |= flush;
1389 NAPI_GRO_CB(skb)->flush |= flush;
1390 skb_gro_pull(skb, sizeof(*iph));
1391 skb_set_transport_header(skb, skb_gro_offset(skb));
1393 pp = ops->gro_receive(head, skb);
1399 NAPI_GRO_CB(skb)->flush |= flush;
1404 static int inet_gro_complete(struct sk_buff *skb)
1406 const struct net_protocol *ops;
1407 struct iphdr *iph = ip_hdr(skb);
1408 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1410 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1412 csum_replace2(&iph->check, iph->tot_len, newlen);
1413 iph->tot_len = newlen;
1416 ops = rcu_dereference(inet_protos[proto]);
1417 if (WARN_ON(!ops || !ops->gro_complete))
1420 err = ops->gro_complete(skb);
1428 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1429 unsigned short type, unsigned char protocol,
1432 struct socket *sock;
1433 int rc = sock_create_kern(family, type, protocol, &sock);
1437 (*sk)->sk_allocation = GFP_ATOMIC;
1439 * Unhash it so that IP input processing does not even see it,
1440 * we do not wish this socket to see incoming packets.
1442 (*sk)->sk_prot->unhash(*sk);
1444 sk_change_net(*sk, net);
1448 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1450 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1452 unsigned long res = 0;
1455 for_each_possible_cpu(i) {
1456 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1457 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1461 EXPORT_SYMBOL_GPL(snmp_fold_field);
1463 #if BITS_PER_LONG==32
1465 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1470 for_each_possible_cpu(cpu) {
1472 struct u64_stats_sync *syncp;
1476 bhptr = per_cpu_ptr(mib[0], cpu);
1477 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1479 start = u64_stats_fetch_begin_bh(syncp);
1480 v = *(((u64 *) bhptr) + offt);
1481 } while (u64_stats_fetch_retry_bh(syncp, start));
1487 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1490 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1492 BUG_ON(ptr == NULL);
1493 ptr[0] = __alloc_percpu(mibsize, align);
1496 #if SNMP_ARRAY_SZ == 2
1497 ptr[1] = __alloc_percpu(mibsize, align);
1499 free_percpu(ptr[0]);
1506 EXPORT_SYMBOL_GPL(snmp_mib_init);
1508 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1512 BUG_ON(ptr == NULL);
1513 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1514 free_percpu(ptr[i]);
1518 EXPORT_SYMBOL_GPL(snmp_mib_free);
1520 #ifdef CONFIG_IP_MULTICAST
1521 static const struct net_protocol igmp_protocol = {
1522 .handler = igmp_rcv,
1527 static const struct net_protocol tcp_protocol = {
1528 .handler = tcp_v4_rcv,
1529 .err_handler = tcp_v4_err,
1530 .gso_send_check = tcp_v4_gso_send_check,
1531 .gso_segment = tcp_tso_segment,
1532 .gro_receive = tcp4_gro_receive,
1533 .gro_complete = tcp4_gro_complete,
1538 static const struct net_protocol udp_protocol = {
1540 .err_handler = udp_err,
1541 .gso_send_check = udp4_ufo_send_check,
1542 .gso_segment = udp4_ufo_fragment,
1547 static const struct net_protocol icmp_protocol = {
1548 .handler = icmp_rcv,
1549 .err_handler = ping_err,
1554 static __net_init int ipv4_mib_init_net(struct net *net)
1556 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1557 sizeof(struct tcp_mib),
1558 __alignof__(struct tcp_mib)) < 0)
1560 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1561 sizeof(struct ipstats_mib),
1562 __alignof__(struct ipstats_mib)) < 0)
1564 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1565 sizeof(struct linux_mib),
1566 __alignof__(struct linux_mib)) < 0)
1568 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1569 sizeof(struct udp_mib),
1570 __alignof__(struct udp_mib)) < 0)
1572 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1573 sizeof(struct udp_mib),
1574 __alignof__(struct udp_mib)) < 0)
1575 goto err_udplite_mib;
1576 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1577 sizeof(struct icmp_mib),
1578 __alignof__(struct icmp_mib)) < 0)
1580 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1581 sizeof(struct icmpmsg_mib),
1582 __alignof__(struct icmpmsg_mib)) < 0)
1583 goto err_icmpmsg_mib;
1589 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1591 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1593 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1595 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1597 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1599 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1604 static __net_exit void ipv4_mib_exit_net(struct net *net)
1606 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1607 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1608 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1609 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1610 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1611 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1612 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1615 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1616 .init = ipv4_mib_init_net,
1617 .exit = ipv4_mib_exit_net,
1620 static int __init init_ipv4_mibs(void)
1622 return register_pernet_subsys(&ipv4_mib_ops);
1625 static int ipv4_proc_init(void);
1628 * IP protocol layer initialiser
1631 static struct packet_type ip_packet_type __read_mostly = {
1632 .type = cpu_to_be16(ETH_P_IP),
1634 .gso_send_check = inet_gso_send_check,
1635 .gso_segment = inet_gso_segment,
1636 .gro_receive = inet_gro_receive,
1637 .gro_complete = inet_gro_complete,
1640 static int __init inet_init(void)
1642 struct sk_buff *dummy_skb;
1643 struct inet_protosw *q;
1644 struct list_head *r;
1647 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1649 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1650 if (!sysctl_local_reserved_ports)
1653 rc = proto_register(&tcp_prot, 1);
1655 goto out_free_reserved_ports;
1657 rc = proto_register(&udp_prot, 1);
1659 goto out_unregister_tcp_proto;
1661 rc = proto_register(&raw_prot, 1);
1663 goto out_unregister_udp_proto;
1665 rc = proto_register(&ping_prot, 1);
1667 goto out_unregister_raw_proto;
1670 * Tell SOCKET that we are alive...
1673 (void)sock_register(&inet_family_ops);
1675 #ifdef CONFIG_SYSCTL
1676 ip_static_sysctl_init();
1680 * Add all the base protocols.
1683 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1684 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1685 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1686 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1687 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1688 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1689 #ifdef CONFIG_IP_MULTICAST
1690 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1691 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1694 /* Register the socket-side information for inet_create. */
1695 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1698 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1699 inet_register_protosw(q);
1702 * Set the ARP module up
1708 * Set the IP module up
1715 /* Setup TCP slab cache for open requests. */
1718 /* Setup UDP memory threshold */
1721 /* Add UDP-Lite (RFC 3828) */
1722 udplite4_register();
1727 * Set the ICMP layer up
1730 if (icmp_init() < 0)
1731 panic("Failed to create the ICMP control socket.\n");
1734 * Initialise the multicast router
1736 #if defined(CONFIG_IP_MROUTE)
1738 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1741 * Initialise per-cpu ipv4 mibs
1744 if (init_ipv4_mibs())
1745 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1751 dev_add_pack(&ip_packet_type);
1756 out_unregister_raw_proto:
1757 proto_unregister(&raw_prot);
1758 out_unregister_udp_proto:
1759 proto_unregister(&udp_prot);
1760 out_unregister_tcp_proto:
1761 proto_unregister(&tcp_prot);
1762 out_free_reserved_ports:
1763 kfree(sysctl_local_reserved_ports);
1767 fs_initcall(inet_init);
1769 /* ------------------------------------------------------------------------ */
1771 #ifdef CONFIG_PROC_FS
1772 static int __init ipv4_proc_init(void)
1776 if (raw_proc_init())
1778 if (tcp4_proc_init())
1780 if (udp4_proc_init())
1782 if (ping_proc_init())
1784 if (ip_misc_proc_init())
1801 #else /* CONFIG_PROC_FS */
1802 static int __init ipv4_proc_init(void)
1806 #endif /* CONFIG_PROC_FS */
1808 MODULE_ALIAS_NETPROTO(PF_INET);