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);
230 * inet_ehash_secret must be set exactly once
232 void build_ehash_secret(void)
237 get_random_bytes(&rnd, sizeof(rnd));
240 cmpxchg(&inet_ehash_secret, 0, rnd);
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;
359 if (SOCK_RAW == sock->type) {
360 inet->inet_num = protocol;
361 if (IPPROTO_RAW == protocol)
365 if (ipv4_config.no_pmtu_disc)
366 inet->pmtudisc = IP_PMTUDISC_DONT;
368 inet->pmtudisc = IP_PMTUDISC_WANT;
372 sock_init_data(sock, sk);
374 sk->sk_destruct = inet_sock_destruct;
375 sk->sk_protocol = protocol;
376 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
383 inet->mc_list = NULL;
385 sk_refcnt_debug_inc(sk);
387 if (inet->inet_num) {
388 /* It assumes that any protocol which allows
389 * the user to assign a number at socket
390 * creation time automatically
393 inet->inet_sport = htons(inet->inet_num);
394 /* Add to protocol hash chains. */
395 sk->sk_prot->hash(sk);
398 if (sk->sk_prot->init) {
399 err = sk->sk_prot->init(sk);
401 sk_common_release(sk);
412 * The peer socket should always be NULL (or else). When we call this
413 * function we are destroying the object and from then on nobody
414 * should refer to it.
416 int inet_release(struct socket *sock)
418 struct sock *sk = sock->sk;
423 sock_rps_reset_flow(sk);
425 /* Applications forget to leave groups before exiting */
426 ip_mc_drop_socket(sk);
428 /* If linger is set, we don't return until the close
429 * is complete. Otherwise we return immediately. The
430 * actually closing is done the same either way.
432 * If the close is due to the process exiting, we never
436 if (sock_flag(sk, SOCK_LINGER) &&
437 !(current->flags & PF_EXITING))
438 timeout = sk->sk_lingertime;
440 sk->sk_prot->close(sk, timeout);
444 EXPORT_SYMBOL(inet_release);
446 /* It is off by default, see below. */
447 int sysctl_ip_nonlocal_bind __read_mostly;
448 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
450 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
452 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
453 struct sock *sk = sock->sk;
454 struct inet_sock *inet = inet_sk(sk);
459 /* If the socket has its own bind function then use it. (RAW) */
460 if (sk->sk_prot->bind) {
461 err = sk->sk_prot->bind(sk, uaddr, addr_len);
465 if (addr_len < sizeof(struct sockaddr_in))
468 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
470 /* Not specified by any standard per-se, however it breaks too
471 * many applications when removed. It is unfortunate since
472 * allowing applications to make a non-local bind solves
473 * several problems with systems using dynamic addressing.
474 * (ie. your servers still start up even if your ISDN link
475 * is temporarily down)
477 err = -EADDRNOTAVAIL;
478 if (!sysctl_ip_nonlocal_bind &&
479 !(inet->freebind || inet->transparent) &&
480 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
481 chk_addr_ret != RTN_LOCAL &&
482 chk_addr_ret != RTN_MULTICAST &&
483 chk_addr_ret != RTN_BROADCAST)
486 snum = ntohs(addr->sin_port);
488 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
491 /* We keep a pair of addresses. rcv_saddr is the one
492 * used by hash lookups, and saddr is used for transmit.
494 * In the BSD API these are the same except where it
495 * would be illegal to use them (multicast/broadcast) in
496 * which case the sending device address is used.
500 /* Check these errors (active socket, double bind). */
502 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
503 goto out_release_sock;
505 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
506 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
507 inet->inet_saddr = 0; /* Use device */
509 /* Make sure we are allowed to bind here. */
510 if (sk->sk_prot->get_port(sk, snum)) {
511 inet->inet_saddr = inet->inet_rcv_saddr = 0;
513 goto out_release_sock;
516 if (inet->inet_rcv_saddr)
517 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
519 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
520 inet->inet_sport = htons(inet->inet_num);
521 inet->inet_daddr = 0;
522 inet->inet_dport = 0;
530 EXPORT_SYMBOL(inet_bind);
532 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
533 int addr_len, int flags)
535 struct sock *sk = sock->sk;
537 if (addr_len < sizeof(uaddr->sa_family))
539 if (uaddr->sa_family == AF_UNSPEC)
540 return sk->sk_prot->disconnect(sk, flags);
542 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
544 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
546 EXPORT_SYMBOL(inet_dgram_connect);
548 static long inet_wait_for_connect(struct sock *sk, long timeo)
552 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
554 /* Basic assumption: if someone sets sk->sk_err, he _must_
555 * change state of the socket from TCP_SYN_*.
556 * Connect() does not allow to get error notifications
557 * without closing the socket.
559 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
561 timeo = schedule_timeout(timeo);
563 if (signal_pending(current) || !timeo)
565 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
567 finish_wait(sk_sleep(sk), &wait);
572 * Connect to a remote host. There is regrettably still a little
573 * TCP 'magic' in here.
575 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
576 int addr_len, int flags)
578 struct sock *sk = sock->sk;
582 if (addr_len < sizeof(uaddr->sa_family))
587 if (uaddr->sa_family == AF_UNSPEC) {
588 err = sk->sk_prot->disconnect(sk, flags);
589 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
593 switch (sock->state) {
602 /* Fall out of switch with err, set for this state */
606 if (sk->sk_state != TCP_CLOSE)
609 err = sk->sk_prot->connect(sk, uaddr, addr_len);
613 sock->state = SS_CONNECTING;
615 /* Just entered SS_CONNECTING state; the only
616 * difference is that return value in non-blocking
617 * case is EINPROGRESS, rather than EALREADY.
623 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
625 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
626 /* Error code is set above */
627 if (!timeo || !inet_wait_for_connect(sk, timeo))
630 err = sock_intr_errno(timeo);
631 if (signal_pending(current))
635 /* Connection was closed by RST, timeout, ICMP error
636 * or another process disconnected us.
638 if (sk->sk_state == TCP_CLOSE)
641 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
642 * and error was received after socket entered established state.
643 * Hence, it is handled normally after connect() return successfully.
646 sock->state = SS_CONNECTED;
653 err = sock_error(sk) ? : -ECONNABORTED;
654 sock->state = SS_UNCONNECTED;
655 if (sk->sk_prot->disconnect(sk, flags))
656 sock->state = SS_DISCONNECTING;
659 EXPORT_SYMBOL(inet_stream_connect);
662 * Accept a pending connection. The TCP layer now gives BSD semantics.
665 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
667 struct sock *sk1 = sock->sk;
669 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
676 WARN_ON(!((1 << sk2->sk_state) &
677 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
679 sock_graft(sk2, newsock);
681 newsock->state = SS_CONNECTED;
687 EXPORT_SYMBOL(inet_accept);
691 * This does both peername and sockname.
693 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
694 int *uaddr_len, int peer)
696 struct sock *sk = sock->sk;
697 struct inet_sock *inet = inet_sk(sk);
698 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
700 sin->sin_family = AF_INET;
702 if (!inet->inet_dport ||
703 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
706 sin->sin_port = inet->inet_dport;
707 sin->sin_addr.s_addr = inet->inet_daddr;
709 __be32 addr = inet->inet_rcv_saddr;
711 addr = inet->inet_saddr;
712 sin->sin_port = inet->inet_sport;
713 sin->sin_addr.s_addr = addr;
715 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
716 *uaddr_len = sizeof(*sin);
719 EXPORT_SYMBOL(inet_getname);
721 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
724 struct sock *sk = sock->sk;
726 sock_rps_record_flow(sk);
728 /* We may need to bind the socket. */
729 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
733 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
735 EXPORT_SYMBOL(inet_sendmsg);
737 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
738 size_t size, int flags)
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 if (sk->sk_prot->sendpage)
750 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
751 return sock_no_sendpage(sock, page, offset, size, flags);
753 EXPORT_SYMBOL(inet_sendpage);
755 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
756 size_t size, int flags)
758 struct sock *sk = sock->sk;
762 sock_rps_record_flow(sk);
764 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
765 flags & ~MSG_DONTWAIT, &addr_len);
767 msg->msg_namelen = addr_len;
770 EXPORT_SYMBOL(inet_recvmsg);
772 int inet_shutdown(struct socket *sock, int how)
774 struct sock *sk = sock->sk;
777 /* This should really check to make sure
778 * the socket is a TCP socket. (WHY AC...)
780 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
783 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
787 if (sock->state == SS_CONNECTING) {
788 if ((1 << sk->sk_state) &
789 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
790 sock->state = SS_DISCONNECTING;
792 sock->state = SS_CONNECTED;
795 switch (sk->sk_state) {
798 /* Hack to wake up other listeners, who can poll for
799 POLLHUP, even on eg. unconnected UDP sockets -- RR */
801 sk->sk_shutdown |= how;
802 if (sk->sk_prot->shutdown)
803 sk->sk_prot->shutdown(sk, how);
806 /* Remaining two branches are temporary solution for missing
807 * close() in multithreaded environment. It is _not_ a good idea,
808 * but we have no choice until close() is repaired at VFS level.
811 if (!(how & RCV_SHUTDOWN))
815 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
816 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
820 /* Wake up anyone sleeping in poll. */
821 sk->sk_state_change(sk);
825 EXPORT_SYMBOL(inet_shutdown);
828 * ioctl() calls you can issue on an INET socket. Most of these are
829 * device configuration and stuff and very rarely used. Some ioctls
830 * pass on to the socket itself.
832 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
833 * loads the devconfigure module does its configuring and unloads it.
834 * There's a good 20K of config code hanging around the kernel.
837 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
839 struct sock *sk = sock->sk;
841 struct net *net = sock_net(sk);
845 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
848 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
853 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
858 err = arp_ioctl(net, cmd, (void __user *)arg);
871 err = devinet_ioctl(net, cmd, (void __user *)arg);
874 if (sk->sk_prot->ioctl)
875 err = sk->sk_prot->ioctl(sk, cmd, arg);
882 EXPORT_SYMBOL(inet_ioctl);
885 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
887 struct sock *sk = sock->sk;
888 int err = -ENOIOCTLCMD;
890 if (sk->sk_prot->compat_ioctl)
891 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
897 const struct proto_ops inet_stream_ops = {
899 .owner = THIS_MODULE,
900 .release = inet_release,
902 .connect = inet_stream_connect,
903 .socketpair = sock_no_socketpair,
904 .accept = inet_accept,
905 .getname = inet_getname,
908 .listen = inet_listen,
909 .shutdown = inet_shutdown,
910 .setsockopt = sock_common_setsockopt,
911 .getsockopt = sock_common_getsockopt,
912 .sendmsg = inet_sendmsg,
913 .recvmsg = inet_recvmsg,
914 .mmap = sock_no_mmap,
915 .sendpage = inet_sendpage,
916 .splice_read = tcp_splice_read,
918 .compat_setsockopt = compat_sock_common_setsockopt,
919 .compat_getsockopt = compat_sock_common_getsockopt,
920 .compat_ioctl = inet_compat_ioctl,
923 EXPORT_SYMBOL(inet_stream_ops);
925 const struct proto_ops inet_dgram_ops = {
927 .owner = THIS_MODULE,
928 .release = inet_release,
930 .connect = inet_dgram_connect,
931 .socketpair = sock_no_socketpair,
932 .accept = sock_no_accept,
933 .getname = inet_getname,
936 .listen = sock_no_listen,
937 .shutdown = inet_shutdown,
938 .setsockopt = sock_common_setsockopt,
939 .getsockopt = sock_common_getsockopt,
940 .sendmsg = inet_sendmsg,
941 .recvmsg = inet_recvmsg,
942 .mmap = sock_no_mmap,
943 .sendpage = inet_sendpage,
945 .compat_setsockopt = compat_sock_common_setsockopt,
946 .compat_getsockopt = compat_sock_common_getsockopt,
947 .compat_ioctl = inet_compat_ioctl,
950 EXPORT_SYMBOL(inet_dgram_ops);
953 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
956 static const struct proto_ops inet_sockraw_ops = {
958 .owner = THIS_MODULE,
959 .release = inet_release,
961 .connect = inet_dgram_connect,
962 .socketpair = sock_no_socketpair,
963 .accept = sock_no_accept,
964 .getname = inet_getname,
965 .poll = datagram_poll,
967 .listen = sock_no_listen,
968 .shutdown = inet_shutdown,
969 .setsockopt = sock_common_setsockopt,
970 .getsockopt = sock_common_getsockopt,
971 .sendmsg = inet_sendmsg,
972 .recvmsg = inet_recvmsg,
973 .mmap = sock_no_mmap,
974 .sendpage = inet_sendpage,
976 .compat_setsockopt = compat_sock_common_setsockopt,
977 .compat_getsockopt = compat_sock_common_getsockopt,
978 .compat_ioctl = inet_compat_ioctl,
982 static const struct net_proto_family inet_family_ops = {
984 .create = inet_create,
985 .owner = THIS_MODULE,
988 /* Upon startup we insert all the elements in inetsw_array[] into
989 * the linked list inetsw.
991 static struct inet_protosw inetsw_array[] =
995 .protocol = IPPROTO_TCP,
997 .ops = &inet_stream_ops,
999 .flags = INET_PROTOSW_PERMANENT |
1005 .protocol = IPPROTO_UDP,
1007 .ops = &inet_dgram_ops,
1008 .no_check = UDP_CSUM_DEFAULT,
1009 .flags = INET_PROTOSW_PERMANENT,
1014 .protocol = IPPROTO_ICMP,
1016 .ops = &inet_dgram_ops,
1017 .no_check = UDP_CSUM_DEFAULT,
1018 .flags = INET_PROTOSW_REUSE,
1023 .protocol = IPPROTO_IP, /* wild card */
1025 .ops = &inet_sockraw_ops,
1026 .no_check = UDP_CSUM_DEFAULT,
1027 .flags = INET_PROTOSW_REUSE,
1031 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1033 void inet_register_protosw(struct inet_protosw *p)
1035 struct list_head *lh;
1036 struct inet_protosw *answer;
1037 int protocol = p->protocol;
1038 struct list_head *last_perm;
1040 spin_lock_bh(&inetsw_lock);
1042 if (p->type >= SOCK_MAX)
1045 /* If we are trying to override a permanent protocol, bail. */
1047 last_perm = &inetsw[p->type];
1048 list_for_each(lh, &inetsw[p->type]) {
1049 answer = list_entry(lh, struct inet_protosw, list);
1051 /* Check only the non-wild match. */
1052 if (INET_PROTOSW_PERMANENT & answer->flags) {
1053 if (protocol == answer->protocol)
1063 /* Add the new entry after the last permanent entry if any, so that
1064 * the new entry does not override a permanent entry when matched with
1065 * a wild-card protocol. But it is allowed to override any existing
1066 * non-permanent entry. This means that when we remove this entry, the
1067 * system automatically returns to the old behavior.
1069 list_add_rcu(&p->list, last_perm);
1071 spin_unlock_bh(&inetsw_lock);
1076 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1082 "Ignoring attempt to register invalid socket type %d.\n",
1086 EXPORT_SYMBOL(inet_register_protosw);
1088 void inet_unregister_protosw(struct inet_protosw *p)
1090 if (INET_PROTOSW_PERMANENT & p->flags) {
1092 "Attempt to unregister permanent protocol %d.\n",
1095 spin_lock_bh(&inetsw_lock);
1096 list_del_rcu(&p->list);
1097 spin_unlock_bh(&inetsw_lock);
1102 EXPORT_SYMBOL(inet_unregister_protosw);
1105 * Shall we try to damage output packets if routing dev changes?
1108 int sysctl_ip_dynaddr __read_mostly;
1110 static int inet_sk_reselect_saddr(struct sock *sk)
1112 struct inet_sock *inet = inet_sk(sk);
1113 __be32 old_saddr = inet->inet_saddr;
1114 __be32 daddr = inet->inet_daddr;
1118 struct ip_options_rcu *inet_opt;
1120 inet_opt = rcu_dereference_protected(inet->inet_opt,
1121 sock_owned_by_user(sk));
1122 if (inet_opt && inet_opt->opt.srr)
1123 daddr = inet_opt->opt.faddr;
1125 /* Query new route. */
1126 fl4 = &inet->cork.fl.u.ip4;
1127 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1128 sk->sk_bound_dev_if, sk->sk_protocol,
1129 inet->inet_sport, inet->inet_dport, sk, false);
1133 sk_setup_caps(sk, &rt->dst);
1135 new_saddr = fl4->saddr;
1137 if (new_saddr == old_saddr)
1140 if (sysctl_ip_dynaddr > 1) {
1141 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1142 __func__, &old_saddr, &new_saddr);
1145 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1148 * XXX The only one ugly spot where we need to
1149 * XXX really change the sockets identity after
1150 * XXX it has entered the hashes. -DaveM
1152 * Besides that, it does not check for connection
1153 * uniqueness. Wait for troubles.
1155 __sk_prot_rehash(sk);
1159 int inet_sk_rebuild_header(struct sock *sk)
1161 struct inet_sock *inet = inet_sk(sk);
1162 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1164 struct ip_options_rcu *inet_opt;
1168 /* Route is OK, nothing to do. */
1174 inet_opt = rcu_dereference(inet->inet_opt);
1175 daddr = inet->inet_daddr;
1176 if (inet_opt && inet_opt->opt.srr)
1177 daddr = inet_opt->opt.faddr;
1179 fl4 = &inet->cork.fl.u.ip4;
1180 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1181 inet->inet_dport, inet->inet_sport,
1182 sk->sk_protocol, RT_CONN_FLAGS(sk),
1183 sk->sk_bound_dev_if);
1186 sk_setup_caps(sk, &rt->dst);
1190 /* Routing failed... */
1191 sk->sk_route_caps = 0;
1193 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1194 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1196 if (!sysctl_ip_dynaddr ||
1197 sk->sk_state != TCP_SYN_SENT ||
1198 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1199 (err = inet_sk_reselect_saddr(sk)) != 0)
1200 sk->sk_err_soft = -err;
1205 EXPORT_SYMBOL(inet_sk_rebuild_header);
1207 static int inet_gso_send_check(struct sk_buff *skb)
1209 const struct iphdr *iph;
1210 const struct net_protocol *ops;
1215 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1220 if (ihl < sizeof(*iph))
1223 if (unlikely(!pskb_may_pull(skb, ihl)))
1226 __skb_pull(skb, ihl);
1227 skb_reset_transport_header(skb);
1229 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1230 err = -EPROTONOSUPPORT;
1233 ops = rcu_dereference(inet_protos[proto]);
1234 if (likely(ops && ops->gso_send_check))
1235 err = ops->gso_send_check(skb);
1242 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1244 struct sk_buff *segs = ERR_PTR(-EINVAL);
1246 const struct net_protocol *ops;
1250 unsigned int offset = 0;
1252 if (!(features & NETIF_F_V4_CSUM))
1253 features &= ~NETIF_F_SG;
1255 if (unlikely(skb_shinfo(skb)->gso_type &
1263 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1268 if (ihl < sizeof(*iph))
1271 if (unlikely(!pskb_may_pull(skb, ihl)))
1274 __skb_pull(skb, ihl);
1275 skb_reset_transport_header(skb);
1277 id = ntohs(iph->id);
1278 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1279 segs = ERR_PTR(-EPROTONOSUPPORT);
1282 ops = rcu_dereference(inet_protos[proto]);
1283 if (likely(ops && ops->gso_segment))
1284 segs = ops->gso_segment(skb, features);
1287 if (!segs || IS_ERR(segs))
1293 if (proto == IPPROTO_UDP) {
1294 iph->id = htons(id);
1295 iph->frag_off = htons(offset >> 3);
1296 if (skb->next != NULL)
1297 iph->frag_off |= htons(IP_MF);
1298 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1300 iph->id = htons(id++);
1301 iph->tot_len = htons(skb->len - skb->mac_len);
1303 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1304 } while ((skb = skb->next));
1310 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1311 struct sk_buff *skb)
1313 const struct net_protocol *ops;
1314 struct sk_buff **pp = NULL;
1316 const struct iphdr *iph;
1323 off = skb_gro_offset(skb);
1324 hlen = off + sizeof(*iph);
1325 iph = skb_gro_header_fast(skb, off);
1326 if (skb_gro_header_hard(skb, hlen)) {
1327 iph = skb_gro_header_slow(skb, hlen, off);
1332 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1335 ops = rcu_dereference(inet_protos[proto]);
1336 if (!ops || !ops->gro_receive)
1339 if (*(u8 *)iph != 0x45)
1342 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1345 id = ntohl(*(__be32 *)&iph->id);
1346 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1349 for (p = *head; p; p = p->next) {
1352 if (!NAPI_GRO_CB(p)->same_flow)
1357 if ((iph->protocol ^ iph2->protocol) |
1358 (iph->tos ^ iph2->tos) |
1359 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1360 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1361 NAPI_GRO_CB(p)->same_flow = 0;
1365 /* All fields must match except length and checksum. */
1366 NAPI_GRO_CB(p)->flush |=
1367 (iph->ttl ^ iph2->ttl) |
1368 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1370 NAPI_GRO_CB(p)->flush |= flush;
1373 NAPI_GRO_CB(skb)->flush |= flush;
1374 skb_gro_pull(skb, sizeof(*iph));
1375 skb_set_transport_header(skb, skb_gro_offset(skb));
1377 pp = ops->gro_receive(head, skb);
1383 NAPI_GRO_CB(skb)->flush |= flush;
1388 static int inet_gro_complete(struct sk_buff *skb)
1390 const struct net_protocol *ops;
1391 struct iphdr *iph = ip_hdr(skb);
1392 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1394 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1396 csum_replace2(&iph->check, iph->tot_len, newlen);
1397 iph->tot_len = newlen;
1400 ops = rcu_dereference(inet_protos[proto]);
1401 if (WARN_ON(!ops || !ops->gro_complete))
1404 err = ops->gro_complete(skb);
1412 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1413 unsigned short type, unsigned char protocol,
1416 struct socket *sock;
1417 int rc = sock_create_kern(family, type, protocol, &sock);
1421 (*sk)->sk_allocation = GFP_ATOMIC;
1423 * Unhash it so that IP input processing does not even see it,
1424 * we do not wish this socket to see incoming packets.
1426 (*sk)->sk_prot->unhash(*sk);
1428 sk_change_net(*sk, net);
1432 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1434 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1436 unsigned long res = 0;
1439 for_each_possible_cpu(i) {
1440 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1441 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1445 EXPORT_SYMBOL_GPL(snmp_fold_field);
1447 #if BITS_PER_LONG==32
1449 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1454 for_each_possible_cpu(cpu) {
1455 void *bhptr, *userptr;
1456 struct u64_stats_sync *syncp;
1460 /* first mib used by softirq context, we must use _bh() accessors */
1461 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1462 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1464 start = u64_stats_fetch_begin_bh(syncp);
1465 v_bh = *(((u64 *) bhptr) + offt);
1466 } while (u64_stats_fetch_retry_bh(syncp, start));
1468 /* second mib used in USER context */
1469 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1470 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1472 start = u64_stats_fetch_begin(syncp);
1473 v_user = *(((u64 *) userptr) + offt);
1474 } while (u64_stats_fetch_retry(syncp, start));
1476 res += v_bh + v_user;
1480 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1483 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1485 BUG_ON(ptr == NULL);
1486 ptr[0] = __alloc_percpu(mibsize, align);
1489 ptr[1] = __alloc_percpu(mibsize, align);
1494 free_percpu(ptr[0]);
1499 EXPORT_SYMBOL_GPL(snmp_mib_init);
1501 void snmp_mib_free(void __percpu *ptr[2])
1503 BUG_ON(ptr == NULL);
1504 free_percpu(ptr[0]);
1505 free_percpu(ptr[1]);
1506 ptr[0] = ptr[1] = NULL;
1508 EXPORT_SYMBOL_GPL(snmp_mib_free);
1510 #ifdef CONFIG_IP_MULTICAST
1511 static const struct net_protocol igmp_protocol = {
1512 .handler = igmp_rcv,
1517 static const struct net_protocol tcp_protocol = {
1518 .handler = tcp_v4_rcv,
1519 .err_handler = tcp_v4_err,
1520 .gso_send_check = tcp_v4_gso_send_check,
1521 .gso_segment = tcp_tso_segment,
1522 .gro_receive = tcp4_gro_receive,
1523 .gro_complete = tcp4_gro_complete,
1528 static const struct net_protocol udp_protocol = {
1530 .err_handler = udp_err,
1531 .gso_send_check = udp4_ufo_send_check,
1532 .gso_segment = udp4_ufo_fragment,
1537 static const struct net_protocol icmp_protocol = {
1538 .handler = icmp_rcv,
1539 .err_handler = ping_err,
1544 static __net_init int ipv4_mib_init_net(struct net *net)
1546 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1547 sizeof(struct tcp_mib),
1548 __alignof__(struct tcp_mib)) < 0)
1550 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1551 sizeof(struct ipstats_mib),
1552 __alignof__(struct ipstats_mib)) < 0)
1554 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1555 sizeof(struct linux_mib),
1556 __alignof__(struct linux_mib)) < 0)
1558 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1559 sizeof(struct udp_mib),
1560 __alignof__(struct udp_mib)) < 0)
1562 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1563 sizeof(struct udp_mib),
1564 __alignof__(struct udp_mib)) < 0)
1565 goto err_udplite_mib;
1566 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1567 sizeof(struct icmp_mib),
1568 __alignof__(struct icmp_mib)) < 0)
1570 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1571 sizeof(struct icmpmsg_mib),
1572 __alignof__(struct icmpmsg_mib)) < 0)
1573 goto err_icmpmsg_mib;
1579 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1581 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1583 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1585 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1587 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1589 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1594 static __net_exit void ipv4_mib_exit_net(struct net *net)
1596 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1597 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1598 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1599 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1600 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1601 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1602 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1605 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1606 .init = ipv4_mib_init_net,
1607 .exit = ipv4_mib_exit_net,
1610 static int __init init_ipv4_mibs(void)
1612 return register_pernet_subsys(&ipv4_mib_ops);
1615 static int ipv4_proc_init(void);
1618 * IP protocol layer initialiser
1621 static struct packet_type ip_packet_type __read_mostly = {
1622 .type = cpu_to_be16(ETH_P_IP),
1624 .gso_send_check = inet_gso_send_check,
1625 .gso_segment = inet_gso_segment,
1626 .gro_receive = inet_gro_receive,
1627 .gro_complete = inet_gro_complete,
1630 static int __init inet_init(void)
1632 struct sk_buff *dummy_skb;
1633 struct inet_protosw *q;
1634 struct list_head *r;
1637 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1639 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1640 if (!sysctl_local_reserved_ports)
1643 rc = proto_register(&tcp_prot, 1);
1645 goto out_free_reserved_ports;
1647 rc = proto_register(&udp_prot, 1);
1649 goto out_unregister_tcp_proto;
1651 rc = proto_register(&raw_prot, 1);
1653 goto out_unregister_udp_proto;
1655 rc = proto_register(&ping_prot, 1);
1657 goto out_unregister_raw_proto;
1660 * Tell SOCKET that we are alive...
1663 (void)sock_register(&inet_family_ops);
1665 #ifdef CONFIG_SYSCTL
1666 ip_static_sysctl_init();
1670 * Add all the base protocols.
1673 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1674 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1675 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1676 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1677 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1678 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1679 #ifdef CONFIG_IP_MULTICAST
1680 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1681 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1684 /* Register the socket-side information for inet_create. */
1685 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1688 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1689 inet_register_protosw(q);
1692 * Set the ARP module up
1698 * Set the IP module up
1705 /* Setup TCP slab cache for open requests. */
1708 /* Setup UDP memory threshold */
1711 /* Add UDP-Lite (RFC 3828) */
1712 udplite4_register();
1717 * Set the ICMP layer up
1720 if (icmp_init() < 0)
1721 panic("Failed to create the ICMP control socket.\n");
1724 * Initialise the multicast router
1726 #if defined(CONFIG_IP_MROUTE)
1728 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1731 * Initialise per-cpu ipv4 mibs
1734 if (init_ipv4_mibs())
1735 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1741 dev_add_pack(&ip_packet_type);
1746 out_unregister_raw_proto:
1747 proto_unregister(&raw_prot);
1748 out_unregister_udp_proto:
1749 proto_unregister(&udp_prot);
1750 out_unregister_tcp_proto:
1751 proto_unregister(&tcp_prot);
1752 out_free_reserved_ports:
1753 kfree(sysctl_local_reserved_ports);
1757 fs_initcall(inet_init);
1759 /* ------------------------------------------------------------------------ */
1761 #ifdef CONFIG_PROC_FS
1762 static int __init ipv4_proc_init(void)
1766 if (raw_proc_init())
1768 if (tcp4_proc_init())
1770 if (udp4_proc_init())
1772 if (ping_proc_init())
1774 if (ip_misc_proc_init())
1791 #else /* CONFIG_PROC_FS */
1792 static int __init ipv4_proc_init(void)
1796 #endif /* CONFIG_PROC_FS */
1798 MODULE_ALIAS_NETPROTO(PF_INET);