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>
121 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
122 #include <linux/android_aid.h>
124 static inline int current_has_network(void)
126 return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
129 static inline int current_has_network(void)
135 /* The inetsw table contains everything that inet_create needs to
136 * build a new socket.
138 static struct list_head inetsw[SOCK_MAX];
139 static DEFINE_SPINLOCK(inetsw_lock);
141 struct ipv4_config ipv4_config;
142 EXPORT_SYMBOL(ipv4_config);
144 /* New destruction routine */
146 void inet_sock_destruct(struct sock *sk)
148 struct inet_sock *inet = inet_sk(sk);
150 __skb_queue_purge(&sk->sk_receive_queue);
151 __skb_queue_purge(&sk->sk_error_queue);
155 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
156 pr_err("Attempt to release TCP socket in state %d %p\n",
160 if (!sock_flag(sk, SOCK_DEAD)) {
161 pr_err("Attempt to release alive inet socket %p\n", sk);
165 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
166 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
167 WARN_ON(sk->sk_wmem_queued);
168 WARN_ON(sk->sk_forward_alloc);
170 kfree(rcu_dereference_protected(inet->inet_opt, 1));
171 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
172 sk_refcnt_debug_dec(sk);
174 EXPORT_SYMBOL(inet_sock_destruct);
177 * The routines beyond this point handle the behaviour of an AF_INET
178 * socket object. Mostly it punts to the subprotocols of IP to do
183 * Automatically bind an unbound socket.
186 static int inet_autobind(struct sock *sk)
188 struct inet_sock *inet;
189 /* We may need to bind the socket. */
192 if (!inet->inet_num) {
193 if (sk->sk_prot->get_port(sk, 0)) {
197 inet->inet_sport = htons(inet->inet_num);
204 * Move a socket into listening state.
206 int inet_listen(struct socket *sock, int backlog)
208 struct sock *sk = sock->sk;
209 unsigned char old_state;
215 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
218 old_state = sk->sk_state;
219 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
222 /* Really, if the socket is already in listen state
223 * we can only allow the backlog to be adjusted.
225 if (old_state != TCP_LISTEN) {
226 err = inet_csk_listen_start(sk, backlog);
230 sk->sk_max_ack_backlog = backlog;
237 EXPORT_SYMBOL(inet_listen);
239 u32 inet_ehash_secret __read_mostly;
240 EXPORT_SYMBOL(inet_ehash_secret);
243 * inet_ehash_secret must be set exactly once
245 void build_ehash_secret(void)
250 get_random_bytes(&rnd, sizeof(rnd));
253 cmpxchg(&inet_ehash_secret, 0, rnd);
255 EXPORT_SYMBOL(build_ehash_secret);
257 static inline int inet_netns_ok(struct net *net, int protocol)
260 const struct net_protocol *ipprot;
262 if (net_eq(net, &init_net))
265 hash = protocol & (MAX_INET_PROTOS - 1);
266 ipprot = rcu_dereference(inet_protos[hash]);
271 return ipprot->netns_ok;
276 * Create an inet socket.
279 static int inet_create(struct net *net, struct socket *sock, int protocol,
283 struct inet_protosw *answer;
284 struct inet_sock *inet;
285 struct proto *answer_prot;
286 unsigned char answer_flags;
287 char answer_no_check;
288 int try_loading_module = 0;
291 if (!current_has_network())
294 if (unlikely(!inet_ehash_secret))
295 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
296 build_ehash_secret();
298 sock->state = SS_UNCONNECTED;
300 /* Look for the requested type/protocol pair. */
302 err = -ESOCKTNOSUPPORT;
304 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
307 /* Check the non-wild match. */
308 if (protocol == answer->protocol) {
309 if (protocol != IPPROTO_IP)
312 /* Check for the two wild cases. */
313 if (IPPROTO_IP == protocol) {
314 protocol = answer->protocol;
317 if (IPPROTO_IP == answer->protocol)
320 err = -EPROTONOSUPPORT;
324 if (try_loading_module < 2) {
327 * Be more specific, e.g. net-pf-2-proto-132-type-1
328 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
330 if (++try_loading_module == 1)
331 request_module("net-pf-%d-proto-%d-type-%d",
332 PF_INET, protocol, sock->type);
334 * Fall back to generic, e.g. net-pf-2-proto-132
335 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
338 request_module("net-pf-%d-proto-%d",
340 goto lookup_protocol;
346 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
350 if (!inet_netns_ok(net, protocol))
353 sock->ops = answer->ops;
354 answer_prot = answer->prot;
355 answer_no_check = answer->no_check;
356 answer_flags = answer->flags;
359 WARN_ON(answer_prot->slab == NULL);
362 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
367 sk->sk_no_check = answer_no_check;
368 if (INET_PROTOSW_REUSE & answer_flags)
372 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
376 if (SOCK_RAW == sock->type) {
377 inet->inet_num = protocol;
378 if (IPPROTO_RAW == protocol)
382 if (ipv4_config.no_pmtu_disc)
383 inet->pmtudisc = IP_PMTUDISC_DONT;
385 inet->pmtudisc = IP_PMTUDISC_WANT;
389 sock_init_data(sock, sk);
391 sk->sk_destruct = inet_sock_destruct;
392 sk->sk_protocol = protocol;
393 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
400 inet->mc_list = NULL;
402 sk_refcnt_debug_inc(sk);
404 if (inet->inet_num) {
405 /* It assumes that any protocol which allows
406 * the user to assign a number at socket
407 * creation time automatically
410 inet->inet_sport = htons(inet->inet_num);
411 /* Add to protocol hash chains. */
412 sk->sk_prot->hash(sk);
415 if (sk->sk_prot->init) {
416 err = sk->sk_prot->init(sk);
418 sk_common_release(sk);
429 * The peer socket should always be NULL (or else). When we call this
430 * function we are destroying the object and from then on nobody
431 * should refer to it.
433 int inet_release(struct socket *sock)
435 struct sock *sk = sock->sk;
440 sock_rps_reset_flow(sk);
442 /* Applications forget to leave groups before exiting */
443 ip_mc_drop_socket(sk);
445 /* If linger is set, we don't return until the close
446 * is complete. Otherwise we return immediately. The
447 * actually closing is done the same either way.
449 * If the close is due to the process exiting, we never
453 if (sock_flag(sk, SOCK_LINGER) &&
454 !(current->flags & PF_EXITING))
455 timeout = sk->sk_lingertime;
457 sk->sk_prot->close(sk, timeout);
461 EXPORT_SYMBOL(inet_release);
463 /* It is off by default, see below. */
464 int sysctl_ip_nonlocal_bind __read_mostly;
465 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
467 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
469 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
470 struct sock *sk = sock->sk;
471 struct inet_sock *inet = inet_sk(sk);
476 /* If the socket has its own bind function then use it. (RAW) */
477 if (sk->sk_prot->bind) {
478 err = sk->sk_prot->bind(sk, uaddr, addr_len);
482 if (addr_len < sizeof(struct sockaddr_in))
485 if (addr->sin_family != AF_INET) {
490 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
492 /* Not specified by any standard per-se, however it breaks too
493 * many applications when removed. It is unfortunate since
494 * allowing applications to make a non-local bind solves
495 * several problems with systems using dynamic addressing.
496 * (ie. your servers still start up even if your ISDN link
497 * is temporarily down)
499 err = -EADDRNOTAVAIL;
500 if (!sysctl_ip_nonlocal_bind &&
501 !(inet->freebind || inet->transparent) &&
502 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
503 chk_addr_ret != RTN_LOCAL &&
504 chk_addr_ret != RTN_MULTICAST &&
505 chk_addr_ret != RTN_BROADCAST)
508 snum = ntohs(addr->sin_port);
510 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
513 /* We keep a pair of addresses. rcv_saddr is the one
514 * used by hash lookups, and saddr is used for transmit.
516 * In the BSD API these are the same except where it
517 * would be illegal to use them (multicast/broadcast) in
518 * which case the sending device address is used.
522 /* Check these errors (active socket, double bind). */
524 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
525 goto out_release_sock;
527 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
528 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
529 inet->inet_saddr = 0; /* Use device */
531 /* Make sure we are allowed to bind here. */
532 if (sk->sk_prot->get_port(sk, snum)) {
533 inet->inet_saddr = inet->inet_rcv_saddr = 0;
535 goto out_release_sock;
538 if (inet->inet_rcv_saddr)
539 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
541 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
542 inet->inet_sport = htons(inet->inet_num);
543 inet->inet_daddr = 0;
544 inet->inet_dport = 0;
552 EXPORT_SYMBOL(inet_bind);
554 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
555 int addr_len, int flags)
557 struct sock *sk = sock->sk;
559 if (addr_len < sizeof(uaddr->sa_family))
561 if (uaddr->sa_family == AF_UNSPEC)
562 return sk->sk_prot->disconnect(sk, flags);
564 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
566 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
568 EXPORT_SYMBOL(inet_dgram_connect);
570 static long inet_wait_for_connect(struct sock *sk, long timeo)
574 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
576 /* Basic assumption: if someone sets sk->sk_err, he _must_
577 * change state of the socket from TCP_SYN_*.
578 * Connect() does not allow to get error notifications
579 * without closing the socket.
581 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
583 timeo = schedule_timeout(timeo);
585 if (signal_pending(current) || !timeo)
587 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
589 finish_wait(sk_sleep(sk), &wait);
594 * Connect to a remote host. There is regrettably still a little
595 * TCP 'magic' in here.
597 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
598 int addr_len, int flags)
600 struct sock *sk = sock->sk;
604 if (addr_len < sizeof(uaddr->sa_family))
609 if (uaddr->sa_family == AF_UNSPEC) {
610 err = sk->sk_prot->disconnect(sk, flags);
611 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
615 switch (sock->state) {
624 /* Fall out of switch with err, set for this state */
628 if (sk->sk_state != TCP_CLOSE)
631 err = sk->sk_prot->connect(sk, uaddr, addr_len);
635 sock->state = SS_CONNECTING;
637 /* Just entered SS_CONNECTING state; the only
638 * difference is that return value in non-blocking
639 * case is EINPROGRESS, rather than EALREADY.
645 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
647 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
648 /* Error code is set above */
649 if (!timeo || !inet_wait_for_connect(sk, timeo))
652 err = sock_intr_errno(timeo);
653 if (signal_pending(current))
657 /* Connection was closed by RST, timeout, ICMP error
658 * or another process disconnected us.
660 if (sk->sk_state == TCP_CLOSE)
663 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
664 * and error was received after socket entered established state.
665 * Hence, it is handled normally after connect() return successfully.
668 sock->state = SS_CONNECTED;
675 err = sock_error(sk) ? : -ECONNABORTED;
676 sock->state = SS_UNCONNECTED;
677 if (sk->sk_prot->disconnect(sk, flags))
678 sock->state = SS_DISCONNECTING;
681 EXPORT_SYMBOL(inet_stream_connect);
684 * Accept a pending connection. The TCP layer now gives BSD semantics.
687 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
689 struct sock *sk1 = sock->sk;
691 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
698 sock_rps_record_flow(sk2);
699 WARN_ON(!((1 << sk2->sk_state) &
700 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
702 sock_graft(sk2, newsock);
704 newsock->state = SS_CONNECTED;
710 EXPORT_SYMBOL(inet_accept);
714 * This does both peername and sockname.
716 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
717 int *uaddr_len, int peer)
719 struct sock *sk = sock->sk;
720 struct inet_sock *inet = inet_sk(sk);
721 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
723 sin->sin_family = AF_INET;
725 if (!inet->inet_dport ||
726 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
729 sin->sin_port = inet->inet_dport;
730 sin->sin_addr.s_addr = inet->inet_daddr;
732 __be32 addr = inet->inet_rcv_saddr;
734 addr = inet->inet_saddr;
735 sin->sin_port = inet->inet_sport;
736 sin->sin_addr.s_addr = addr;
738 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
739 *uaddr_len = sizeof(*sin);
742 EXPORT_SYMBOL(inet_getname);
744 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
747 struct sock *sk = sock->sk;
749 sock_rps_record_flow(sk);
751 /* We may need to bind the socket. */
752 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
756 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
758 EXPORT_SYMBOL(inet_sendmsg);
760 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
761 size_t size, int flags)
763 struct sock *sk = sock->sk;
765 sock_rps_record_flow(sk);
767 /* We may need to bind the socket. */
768 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
772 if (sk->sk_prot->sendpage)
773 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
774 return sock_no_sendpage(sock, page, offset, size, flags);
776 EXPORT_SYMBOL(inet_sendpage);
778 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
779 size_t size, int flags)
781 struct sock *sk = sock->sk;
785 sock_rps_record_flow(sk);
787 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
788 flags & ~MSG_DONTWAIT, &addr_len);
790 msg->msg_namelen = addr_len;
793 EXPORT_SYMBOL(inet_recvmsg);
795 int inet_shutdown(struct socket *sock, int how)
797 struct sock *sk = sock->sk;
800 /* This should really check to make sure
801 * the socket is a TCP socket. (WHY AC...)
803 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
806 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
810 if (sock->state == SS_CONNECTING) {
811 if ((1 << sk->sk_state) &
812 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
813 sock->state = SS_DISCONNECTING;
815 sock->state = SS_CONNECTED;
818 switch (sk->sk_state) {
821 /* Hack to wake up other listeners, who can poll for
822 POLLHUP, even on eg. unconnected UDP sockets -- RR */
824 sk->sk_shutdown |= how;
825 if (sk->sk_prot->shutdown)
826 sk->sk_prot->shutdown(sk, how);
829 /* Remaining two branches are temporary solution for missing
830 * close() in multithreaded environment. It is _not_ a good idea,
831 * but we have no choice until close() is repaired at VFS level.
834 if (!(how & RCV_SHUTDOWN))
838 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
839 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
843 /* Wake up anyone sleeping in poll. */
844 sk->sk_state_change(sk);
848 EXPORT_SYMBOL(inet_shutdown);
851 * ioctl() calls you can issue on an INET socket. Most of these are
852 * device configuration and stuff and very rarely used. Some ioctls
853 * pass on to the socket itself.
855 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
856 * loads the devconfigure module does its configuring and unloads it.
857 * There's a good 20K of config code hanging around the kernel.
860 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
862 struct sock *sk = sock->sk;
864 struct net *net = sock_net(sk);
868 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
871 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
876 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
881 err = arp_ioctl(net, cmd, (void __user *)arg);
895 err = devinet_ioctl(net, cmd, (void __user *)arg);
898 if (sk->sk_prot->ioctl)
899 err = sk->sk_prot->ioctl(sk, cmd, arg);
906 EXPORT_SYMBOL(inet_ioctl);
909 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
911 struct sock *sk = sock->sk;
912 int err = -ENOIOCTLCMD;
914 if (sk->sk_prot->compat_ioctl)
915 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
921 const struct proto_ops inet_stream_ops = {
923 .owner = THIS_MODULE,
924 .release = inet_release,
926 .connect = inet_stream_connect,
927 .socketpair = sock_no_socketpair,
928 .accept = inet_accept,
929 .getname = inet_getname,
932 .listen = inet_listen,
933 .shutdown = inet_shutdown,
934 .setsockopt = sock_common_setsockopt,
935 .getsockopt = sock_common_getsockopt,
936 .sendmsg = inet_sendmsg,
937 .recvmsg = inet_recvmsg,
938 .mmap = sock_no_mmap,
939 .sendpage = inet_sendpage,
940 .splice_read = tcp_splice_read,
942 .compat_setsockopt = compat_sock_common_setsockopt,
943 .compat_getsockopt = compat_sock_common_getsockopt,
944 .compat_ioctl = inet_compat_ioctl,
947 EXPORT_SYMBOL(inet_stream_ops);
949 const struct proto_ops inet_dgram_ops = {
951 .owner = THIS_MODULE,
952 .release = inet_release,
954 .connect = inet_dgram_connect,
955 .socketpair = sock_no_socketpair,
956 .accept = sock_no_accept,
957 .getname = inet_getname,
960 .listen = sock_no_listen,
961 .shutdown = inet_shutdown,
962 .setsockopt = sock_common_setsockopt,
963 .getsockopt = sock_common_getsockopt,
964 .sendmsg = inet_sendmsg,
965 .recvmsg = inet_recvmsg,
966 .mmap = sock_no_mmap,
967 .sendpage = inet_sendpage,
969 .compat_setsockopt = compat_sock_common_setsockopt,
970 .compat_getsockopt = compat_sock_common_getsockopt,
971 .compat_ioctl = inet_compat_ioctl,
974 EXPORT_SYMBOL(inet_dgram_ops);
977 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
980 static const struct proto_ops inet_sockraw_ops = {
982 .owner = THIS_MODULE,
983 .release = inet_release,
985 .connect = inet_dgram_connect,
986 .socketpair = sock_no_socketpair,
987 .accept = sock_no_accept,
988 .getname = inet_getname,
989 .poll = datagram_poll,
991 .listen = sock_no_listen,
992 .shutdown = inet_shutdown,
993 .setsockopt = sock_common_setsockopt,
994 .getsockopt = sock_common_getsockopt,
995 .sendmsg = inet_sendmsg,
996 .recvmsg = inet_recvmsg,
997 .mmap = sock_no_mmap,
998 .sendpage = inet_sendpage,
1000 .compat_setsockopt = compat_sock_common_setsockopt,
1001 .compat_getsockopt = compat_sock_common_getsockopt,
1002 .compat_ioctl = inet_compat_ioctl,
1006 static const struct net_proto_family inet_family_ops = {
1008 .create = inet_create,
1009 .owner = THIS_MODULE,
1012 /* Upon startup we insert all the elements in inetsw_array[] into
1013 * the linked list inetsw.
1015 static struct inet_protosw inetsw_array[] =
1018 .type = SOCK_STREAM,
1019 .protocol = IPPROTO_TCP,
1021 .ops = &inet_stream_ops,
1023 .flags = INET_PROTOSW_PERMANENT |
1029 .protocol = IPPROTO_UDP,
1031 .ops = &inet_dgram_ops,
1032 .no_check = UDP_CSUM_DEFAULT,
1033 .flags = INET_PROTOSW_PERMANENT,
1038 .protocol = IPPROTO_ICMP,
1040 .ops = &inet_dgram_ops,
1041 .no_check = UDP_CSUM_DEFAULT,
1042 .flags = INET_PROTOSW_REUSE,
1047 .protocol = IPPROTO_IP, /* wild card */
1049 .ops = &inet_sockraw_ops,
1050 .no_check = UDP_CSUM_DEFAULT,
1051 .flags = INET_PROTOSW_REUSE,
1055 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1057 void inet_register_protosw(struct inet_protosw *p)
1059 struct list_head *lh;
1060 struct inet_protosw *answer;
1061 int protocol = p->protocol;
1062 struct list_head *last_perm;
1064 spin_lock_bh(&inetsw_lock);
1066 if (p->type >= SOCK_MAX)
1069 /* If we are trying to override a permanent protocol, bail. */
1071 last_perm = &inetsw[p->type];
1072 list_for_each(lh, &inetsw[p->type]) {
1073 answer = list_entry(lh, struct inet_protosw, list);
1075 /* Check only the non-wild match. */
1076 if (INET_PROTOSW_PERMANENT & answer->flags) {
1077 if (protocol == answer->protocol)
1087 /* Add the new entry after the last permanent entry if any, so that
1088 * the new entry does not override a permanent entry when matched with
1089 * a wild-card protocol. But it is allowed to override any existing
1090 * non-permanent entry. This means that when we remove this entry, the
1091 * system automatically returns to the old behavior.
1093 list_add_rcu(&p->list, last_perm);
1095 spin_unlock_bh(&inetsw_lock);
1100 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1106 "Ignoring attempt to register invalid socket type %d.\n",
1110 EXPORT_SYMBOL(inet_register_protosw);
1112 void inet_unregister_protosw(struct inet_protosw *p)
1114 if (INET_PROTOSW_PERMANENT & p->flags) {
1116 "Attempt to unregister permanent protocol %d.\n",
1119 spin_lock_bh(&inetsw_lock);
1120 list_del_rcu(&p->list);
1121 spin_unlock_bh(&inetsw_lock);
1126 EXPORT_SYMBOL(inet_unregister_protosw);
1129 * Shall we try to damage output packets if routing dev changes?
1132 int sysctl_ip_dynaddr __read_mostly;
1134 static int inet_sk_reselect_saddr(struct sock *sk)
1136 struct inet_sock *inet = inet_sk(sk);
1137 __be32 old_saddr = inet->inet_saddr;
1138 __be32 daddr = inet->inet_daddr;
1142 struct ip_options_rcu *inet_opt;
1144 inet_opt = rcu_dereference_protected(inet->inet_opt,
1145 sock_owned_by_user(sk));
1146 if (inet_opt && inet_opt->opt.srr)
1147 daddr = inet_opt->opt.faddr;
1149 /* Query new route. */
1150 fl4 = &inet->cork.fl.u.ip4;
1151 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1152 sk->sk_bound_dev_if, sk->sk_protocol,
1153 inet->inet_sport, inet->inet_dport, sk, false);
1157 sk_setup_caps(sk, &rt->dst);
1159 new_saddr = fl4->saddr;
1161 if (new_saddr == old_saddr)
1164 if (sysctl_ip_dynaddr > 1) {
1165 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1166 __func__, &old_saddr, &new_saddr);
1169 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1172 * XXX The only one ugly spot where we need to
1173 * XXX really change the sockets identity after
1174 * XXX it has entered the hashes. -DaveM
1176 * Besides that, it does not check for connection
1177 * uniqueness. Wait for troubles.
1179 __sk_prot_rehash(sk);
1183 int inet_sk_rebuild_header(struct sock *sk)
1185 struct inet_sock *inet = inet_sk(sk);
1186 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1188 struct ip_options_rcu *inet_opt;
1192 /* Route is OK, nothing to do. */
1198 inet_opt = rcu_dereference(inet->inet_opt);
1199 daddr = inet->inet_daddr;
1200 if (inet_opt && inet_opt->opt.srr)
1201 daddr = inet_opt->opt.faddr;
1203 fl4 = &inet->cork.fl.u.ip4;
1204 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1205 inet->inet_dport, inet->inet_sport,
1206 sk->sk_protocol, RT_CONN_FLAGS(sk),
1207 sk->sk_bound_dev_if);
1210 sk_setup_caps(sk, &rt->dst);
1214 /* Routing failed... */
1215 sk->sk_route_caps = 0;
1217 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1218 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1220 if (!sysctl_ip_dynaddr ||
1221 sk->sk_state != TCP_SYN_SENT ||
1222 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1223 (err = inet_sk_reselect_saddr(sk)) != 0)
1224 sk->sk_err_soft = -err;
1229 EXPORT_SYMBOL(inet_sk_rebuild_header);
1231 static int inet_gso_send_check(struct sk_buff *skb)
1233 const struct iphdr *iph;
1234 const struct net_protocol *ops;
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 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1254 err = -EPROTONOSUPPORT;
1257 ops = rcu_dereference(inet_protos[proto]);
1258 if (likely(ops && ops->gso_send_check))
1259 err = ops->gso_send_check(skb);
1266 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1268 struct sk_buff *segs = ERR_PTR(-EINVAL);
1270 const struct net_protocol *ops;
1274 unsigned int offset = 0;
1276 if (!(features & NETIF_F_V4_CSUM))
1277 features &= ~NETIF_F_SG;
1279 if (unlikely(skb_shinfo(skb)->gso_type &
1287 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1292 if (ihl < sizeof(*iph))
1295 if (unlikely(!pskb_may_pull(skb, ihl)))
1298 __skb_pull(skb, ihl);
1299 skb_reset_transport_header(skb);
1301 id = ntohs(iph->id);
1302 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1303 segs = ERR_PTR(-EPROTONOSUPPORT);
1306 ops = rcu_dereference(inet_protos[proto]);
1307 if (likely(ops && ops->gso_segment))
1308 segs = ops->gso_segment(skb, features);
1311 if (!segs || IS_ERR(segs))
1317 if (proto == IPPROTO_UDP) {
1318 iph->id = htons(id);
1319 iph->frag_off = htons(offset >> 3);
1320 if (skb->next != NULL)
1321 iph->frag_off |= htons(IP_MF);
1322 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1324 iph->id = htons(id++);
1325 iph->tot_len = htons(skb->len - skb->mac_len);
1327 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1328 } while ((skb = skb->next));
1334 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1335 struct sk_buff *skb)
1337 const struct net_protocol *ops;
1338 struct sk_buff **pp = NULL;
1340 const struct iphdr *iph;
1347 off = skb_gro_offset(skb);
1348 hlen = off + sizeof(*iph);
1349 iph = skb_gro_header_fast(skb, off);
1350 if (skb_gro_header_hard(skb, hlen)) {
1351 iph = skb_gro_header_slow(skb, hlen, off);
1356 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1359 ops = rcu_dereference(inet_protos[proto]);
1360 if (!ops || !ops->gro_receive)
1363 if (*(u8 *)iph != 0x45)
1366 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1369 id = ntohl(*(__be32 *)&iph->id);
1370 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1373 for (p = *head; p; p = p->next) {
1376 if (!NAPI_GRO_CB(p)->same_flow)
1381 if ((iph->protocol ^ iph2->protocol) |
1382 (iph->tos ^ iph2->tos) |
1383 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1384 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1385 NAPI_GRO_CB(p)->same_flow = 0;
1389 /* All fields must match except length and checksum. */
1390 NAPI_GRO_CB(p)->flush |=
1391 (iph->ttl ^ iph2->ttl) |
1392 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1394 NAPI_GRO_CB(p)->flush |= flush;
1397 NAPI_GRO_CB(skb)->flush |= flush;
1398 skb_gro_pull(skb, sizeof(*iph));
1399 skb_set_transport_header(skb, skb_gro_offset(skb));
1401 pp = ops->gro_receive(head, skb);
1407 NAPI_GRO_CB(skb)->flush |= flush;
1412 static int inet_gro_complete(struct sk_buff *skb)
1414 const struct net_protocol *ops;
1415 struct iphdr *iph = ip_hdr(skb);
1416 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1418 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1420 csum_replace2(&iph->check, iph->tot_len, newlen);
1421 iph->tot_len = newlen;
1424 ops = rcu_dereference(inet_protos[proto]);
1425 if (WARN_ON(!ops || !ops->gro_complete))
1428 err = ops->gro_complete(skb);
1436 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1437 unsigned short type, unsigned char protocol,
1440 struct socket *sock;
1441 int rc = sock_create_kern(family, type, protocol, &sock);
1445 (*sk)->sk_allocation = GFP_ATOMIC;
1447 * Unhash it so that IP input processing does not even see it,
1448 * we do not wish this socket to see incoming packets.
1450 (*sk)->sk_prot->unhash(*sk);
1452 sk_change_net(*sk, net);
1456 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1458 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1460 unsigned long res = 0;
1463 for_each_possible_cpu(i) {
1464 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1465 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1469 EXPORT_SYMBOL_GPL(snmp_fold_field);
1471 #if BITS_PER_LONG==32
1473 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1478 for_each_possible_cpu(cpu) {
1479 void *bhptr, *userptr;
1480 struct u64_stats_sync *syncp;
1484 /* first mib used by softirq context, we must use _bh() accessors */
1485 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1486 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1488 start = u64_stats_fetch_begin_bh(syncp);
1489 v_bh = *(((u64 *) bhptr) + offt);
1490 } while (u64_stats_fetch_retry_bh(syncp, start));
1492 /* second mib used in USER context */
1493 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1494 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1496 start = u64_stats_fetch_begin(syncp);
1497 v_user = *(((u64 *) userptr) + offt);
1498 } while (u64_stats_fetch_retry(syncp, start));
1500 res += v_bh + v_user;
1504 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1507 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1509 BUG_ON(ptr == NULL);
1510 ptr[0] = __alloc_percpu(mibsize, align);
1513 ptr[1] = __alloc_percpu(mibsize, align);
1518 free_percpu(ptr[0]);
1523 EXPORT_SYMBOL_GPL(snmp_mib_init);
1525 void snmp_mib_free(void __percpu *ptr[2])
1527 BUG_ON(ptr == NULL);
1528 free_percpu(ptr[0]);
1529 free_percpu(ptr[1]);
1530 ptr[0] = ptr[1] = NULL;
1532 EXPORT_SYMBOL_GPL(snmp_mib_free);
1534 #ifdef CONFIG_IP_MULTICAST
1535 static const struct net_protocol igmp_protocol = {
1536 .handler = igmp_rcv,
1541 static const struct net_protocol tcp_protocol = {
1542 .handler = tcp_v4_rcv,
1543 .err_handler = tcp_v4_err,
1544 .gso_send_check = tcp_v4_gso_send_check,
1545 .gso_segment = tcp_tso_segment,
1546 .gro_receive = tcp4_gro_receive,
1547 .gro_complete = tcp4_gro_complete,
1552 static const struct net_protocol udp_protocol = {
1554 .err_handler = udp_err,
1555 .gso_send_check = udp4_ufo_send_check,
1556 .gso_segment = udp4_ufo_fragment,
1561 static const struct net_protocol icmp_protocol = {
1562 .handler = icmp_rcv,
1563 .err_handler = ping_err,
1568 static __net_init int ipv4_mib_init_net(struct net *net)
1570 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1571 sizeof(struct tcp_mib),
1572 __alignof__(struct tcp_mib)) < 0)
1574 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1575 sizeof(struct ipstats_mib),
1576 __alignof__(struct ipstats_mib)) < 0)
1578 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1579 sizeof(struct linux_mib),
1580 __alignof__(struct linux_mib)) < 0)
1582 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1583 sizeof(struct udp_mib),
1584 __alignof__(struct udp_mib)) < 0)
1586 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1587 sizeof(struct udp_mib),
1588 __alignof__(struct udp_mib)) < 0)
1589 goto err_udplite_mib;
1590 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1591 sizeof(struct icmp_mib),
1592 __alignof__(struct icmp_mib)) < 0)
1594 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1595 sizeof(struct icmpmsg_mib),
1596 __alignof__(struct icmpmsg_mib)) < 0)
1597 goto err_icmpmsg_mib;
1603 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1605 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1607 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1609 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1611 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1613 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1618 static __net_exit void ipv4_mib_exit_net(struct net *net)
1620 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1621 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1622 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1623 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1624 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1625 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1626 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1629 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1630 .init = ipv4_mib_init_net,
1631 .exit = ipv4_mib_exit_net,
1634 static int __init init_ipv4_mibs(void)
1636 return register_pernet_subsys(&ipv4_mib_ops);
1639 static int ipv4_proc_init(void);
1642 * IP protocol layer initialiser
1645 static struct packet_type ip_packet_type __read_mostly = {
1646 .type = cpu_to_be16(ETH_P_IP),
1648 .gso_send_check = inet_gso_send_check,
1649 .gso_segment = inet_gso_segment,
1650 .gro_receive = inet_gro_receive,
1651 .gro_complete = inet_gro_complete,
1654 static int __init inet_init(void)
1656 struct sk_buff *dummy_skb;
1657 struct inet_protosw *q;
1658 struct list_head *r;
1661 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1663 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1664 if (!sysctl_local_reserved_ports)
1667 rc = proto_register(&tcp_prot, 1);
1669 goto out_free_reserved_ports;
1671 rc = proto_register(&udp_prot, 1);
1673 goto out_unregister_tcp_proto;
1675 rc = proto_register(&raw_prot, 1);
1677 goto out_unregister_udp_proto;
1679 rc = proto_register(&ping_prot, 1);
1681 goto out_unregister_raw_proto;
1684 * Tell SOCKET that we are alive...
1687 (void)sock_register(&inet_family_ops);
1689 #ifdef CONFIG_SYSCTL
1690 ip_static_sysctl_init();
1694 * Add all the base protocols.
1697 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1698 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1699 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1700 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1701 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1702 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1703 #ifdef CONFIG_IP_MULTICAST
1704 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1705 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1708 /* Register the socket-side information for inet_create. */
1709 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1712 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1713 inet_register_protosw(q);
1716 * Set the ARP module up
1722 * Set the IP module up
1729 /* Setup TCP slab cache for open requests. */
1732 /* Setup UDP memory threshold */
1735 /* Add UDP-Lite (RFC 3828) */
1736 udplite4_register();
1741 * Set the ICMP layer up
1744 if (icmp_init() < 0)
1745 panic("Failed to create the ICMP control socket.\n");
1748 * Initialise the multicast router
1750 #if defined(CONFIG_IP_MROUTE)
1752 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1755 * Initialise per-cpu ipv4 mibs
1758 if (init_ipv4_mibs())
1759 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1765 dev_add_pack(&ip_packet_type);
1770 out_unregister_raw_proto:
1771 proto_unregister(&raw_prot);
1772 out_unregister_udp_proto:
1773 proto_unregister(&udp_prot);
1774 out_unregister_tcp_proto:
1775 proto_unregister(&tcp_prot);
1776 out_free_reserved_ports:
1777 kfree(sysctl_local_reserved_ports);
1781 fs_initcall(inet_init);
1783 /* ------------------------------------------------------------------------ */
1785 #ifdef CONFIG_PROC_FS
1786 static int __init ipv4_proc_init(void)
1790 if (raw_proc_init())
1792 if (tcp4_proc_init())
1794 if (udp4_proc_init())
1796 if (ping_proc_init())
1798 if (ip_misc_proc_init())
1815 #else /* CONFIG_PROC_FS */
1816 static int __init ipv4_proc_init(void)
1820 #endif /* CONFIG_PROC_FS */
1822 MODULE_ALIAS_NETPROTO(PF_INET);