2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PF_INET protocol family socket handler.
13 * Changes (see also sock.c)
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
68 #define pr_fmt(fmt) "IPv4: " fmt
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
93 #include <asm/uaccess.h>
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
101 #include <net/protocol.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
113 #include <net/icmp.h>
114 #include <net/inet_common.h>
115 #include <net/ip_tunnels.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #include <net/secure_seq.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
122 #include <net/l3mdev.h>
125 /* The inetsw table contains everything that inet_create needs to
126 * build a new socket.
128 static struct list_head inetsw[SOCK_MAX];
129 static DEFINE_SPINLOCK(inetsw_lock);
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 dst_release(sk->sk_rx_dst);
160 sk_refcnt_debug_dec(sk);
162 EXPORT_SYMBOL(inet_sock_destruct);
165 * The routines beyond this point handle the behaviour of an AF_INET
166 * socket object. Mostly it punts to the subprotocols of IP to do
171 * Automatically bind an unbound socket.
174 static int inet_autobind(struct sock *sk)
176 struct inet_sock *inet;
177 /* We may need to bind the socket. */
180 if (!inet->inet_num) {
181 if (sk->sk_prot->get_port(sk, 0)) {
185 inet->inet_sport = htons(inet->inet_num);
192 * Move a socket into listening state.
194 int inet_listen(struct socket *sock, int backlog)
196 struct sock *sk = sock->sk;
197 unsigned char old_state;
203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
206 old_state = sk->sk_state;
207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
213 if (old_state != TCP_LISTEN) {
214 /* Check special setups for testing purpose to enable TFO w/o
215 * requiring TCP_FASTOPEN sockopt.
216 * Note that only TCP sockets (SOCK_STREAM) will reach here.
217 * Also fastopenq may already been allocated because this
218 * socket was in TCP_LISTEN state previously but was
219 * shutdown() (rather than close()).
221 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
222 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
223 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
224 fastopen_queue_tune(sk, backlog);
225 else if ((sysctl_tcp_fastopen &
226 TFO_SERVER_WO_SOCKOPT2) != 0)
227 fastopen_queue_tune(sk,
228 ((uint)sysctl_tcp_fastopen) >> 16);
230 tcp_fastopen_init_key_once(true);
232 err = inet_csk_listen_start(sk, backlog);
236 sk->sk_max_ack_backlog = backlog;
243 EXPORT_SYMBOL(inet_listen);
246 * Create an inet socket.
249 static int inet_create(struct net *net, struct socket *sock, int protocol,
253 struct inet_protosw *answer;
254 struct inet_sock *inet;
255 struct proto *answer_prot;
256 unsigned char answer_flags;
257 int try_loading_module = 0;
260 if (protocol < 0 || protocol >= IPPROTO_MAX)
263 sock->state = SS_UNCONNECTED;
265 /* Look for the requested type/protocol pair. */
267 err = -ESOCKTNOSUPPORT;
269 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
272 /* Check the non-wild match. */
273 if (protocol == answer->protocol) {
274 if (protocol != IPPROTO_IP)
277 /* Check for the two wild cases. */
278 if (IPPROTO_IP == protocol) {
279 protocol = answer->protocol;
282 if (IPPROTO_IP == answer->protocol)
285 err = -EPROTONOSUPPORT;
289 if (try_loading_module < 2) {
292 * Be more specific, e.g. net-pf-2-proto-132-type-1
293 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
295 if (++try_loading_module == 1)
296 request_module("net-pf-%d-proto-%d-type-%d",
297 PF_INET, protocol, sock->type);
299 * Fall back to generic, e.g. net-pf-2-proto-132
300 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
303 request_module("net-pf-%d-proto-%d",
305 goto lookup_protocol;
311 if (sock->type == SOCK_RAW && !kern &&
312 !ns_capable(net->user_ns, CAP_NET_RAW))
315 sock->ops = answer->ops;
316 answer_prot = answer->prot;
317 answer_flags = answer->flags;
320 WARN_ON(!answer_prot->slab);
323 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
328 if (INET_PROTOSW_REUSE & answer_flags)
329 sk->sk_reuse = SK_CAN_REUSE;
332 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
336 if (SOCK_RAW == sock->type) {
337 inet->inet_num = protocol;
338 if (IPPROTO_RAW == protocol)
342 if (net->ipv4.sysctl_ip_no_pmtu_disc)
343 inet->pmtudisc = IP_PMTUDISC_DONT;
345 inet->pmtudisc = IP_PMTUDISC_WANT;
349 sock_init_data(sock, sk);
351 sk->sk_destruct = inet_sock_destruct;
352 sk->sk_protocol = protocol;
353 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
360 inet->mc_list = NULL;
363 sk_refcnt_debug_inc(sk);
365 if (inet->inet_num) {
366 /* It assumes that any protocol which allows
367 * the user to assign a number at socket
368 * creation time automatically
371 inet->inet_sport = htons(inet->inet_num);
372 /* Add to protocol hash chains. */
373 err = sk->sk_prot->hash(sk);
375 sk_common_release(sk);
380 if (sk->sk_prot->init) {
381 err = sk->sk_prot->init(sk);
383 sk_common_release(sk);
394 * The peer socket should always be NULL (or else). When we call this
395 * function we are destroying the object and from then on nobody
396 * should refer to it.
398 int inet_release(struct socket *sock)
400 struct sock *sk = sock->sk;
405 /* Applications forget to leave groups before exiting */
406 ip_mc_drop_socket(sk);
408 /* If linger is set, we don't return until the close
409 * is complete. Otherwise we return immediately. The
410 * actually closing is done the same either way.
412 * If the close is due to the process exiting, we never
416 if (sock_flag(sk, SOCK_LINGER) &&
417 !(current->flags & PF_EXITING))
418 timeout = sk->sk_lingertime;
420 sk->sk_prot->close(sk, timeout);
424 EXPORT_SYMBOL(inet_release);
426 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
428 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
429 struct sock *sk = sock->sk;
430 struct inet_sock *inet = inet_sk(sk);
431 struct net *net = sock_net(sk);
434 u32 tb_id = RT_TABLE_LOCAL;
437 /* If the socket has its own bind function then use it. (RAW) */
438 if (sk->sk_prot->bind) {
439 err = sk->sk_prot->bind(sk, uaddr, addr_len);
443 if (addr_len < sizeof(struct sockaddr_in))
446 if (addr->sin_family != AF_INET) {
447 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
448 * only if s_addr is INADDR_ANY.
451 if (addr->sin_family != AF_UNSPEC ||
452 addr->sin_addr.s_addr != htonl(INADDR_ANY))
456 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
457 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
459 /* Not specified by any standard per-se, however it breaks too
460 * many applications when removed. It is unfortunate since
461 * allowing applications to make a non-local bind solves
462 * several problems with systems using dynamic addressing.
463 * (ie. your servers still start up even if your ISDN link
464 * is temporarily down)
466 err = -EADDRNOTAVAIL;
467 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
468 !(inet->freebind || inet->transparent) &&
469 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
470 chk_addr_ret != RTN_LOCAL &&
471 chk_addr_ret != RTN_MULTICAST &&
472 chk_addr_ret != RTN_BROADCAST)
475 snum = ntohs(addr->sin_port);
477 if (snum && snum < PROT_SOCK &&
478 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
481 /* We keep a pair of addresses. rcv_saddr is the one
482 * used by hash lookups, and saddr is used for transmit.
484 * In the BSD API these are the same except where it
485 * would be illegal to use them (multicast/broadcast) in
486 * which case the sending device address is used.
490 /* Check these errors (active socket, double bind). */
492 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
493 goto out_release_sock;
495 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
496 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
497 inet->inet_saddr = 0; /* Use device */
499 /* Make sure we are allowed to bind here. */
500 if ((snum || !inet->bind_address_no_port) &&
501 sk->sk_prot->get_port(sk, snum)) {
502 inet->inet_saddr = inet->inet_rcv_saddr = 0;
504 goto out_release_sock;
507 if (inet->inet_rcv_saddr)
508 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
510 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
511 inet->inet_sport = htons(inet->inet_num);
512 inet->inet_daddr = 0;
513 inet->inet_dport = 0;
521 EXPORT_SYMBOL(inet_bind);
523 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
524 int addr_len, int flags)
526 struct sock *sk = sock->sk;
528 if (addr_len < sizeof(uaddr->sa_family))
530 if (uaddr->sa_family == AF_UNSPEC)
531 return sk->sk_prot->disconnect(sk, flags);
533 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
535 return sk->sk_prot->connect(sk, uaddr, addr_len);
537 EXPORT_SYMBOL(inet_dgram_connect);
539 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
543 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
544 sk->sk_write_pending += writebias;
546 /* Basic assumption: if someone sets sk->sk_err, he _must_
547 * change state of the socket from TCP_SYN_*.
548 * Connect() does not allow to get error notifications
549 * without closing the socket.
551 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
553 timeo = schedule_timeout(timeo);
555 if (signal_pending(current) || !timeo)
557 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
559 finish_wait(sk_sleep(sk), &wait);
560 sk->sk_write_pending -= writebias;
565 * Connect to a remote host. There is regrettably still a little
566 * TCP 'magic' in here.
568 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
569 int addr_len, int flags)
571 struct sock *sk = sock->sk;
575 if (addr_len < sizeof(uaddr->sa_family))
578 if (uaddr->sa_family == AF_UNSPEC) {
579 err = sk->sk_prot->disconnect(sk, flags);
580 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
584 switch (sock->state) {
593 /* Fall out of switch with err, set for this state */
597 if (sk->sk_state != TCP_CLOSE)
600 err = sk->sk_prot->connect(sk, uaddr, addr_len);
604 sock->state = SS_CONNECTING;
606 /* Just entered SS_CONNECTING state; the only
607 * difference is that return value in non-blocking
608 * case is EINPROGRESS, rather than EALREADY.
614 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
616 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
617 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
618 tcp_sk(sk)->fastopen_req &&
619 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
621 /* Error code is set above */
622 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
625 err = sock_intr_errno(timeo);
626 if (signal_pending(current))
630 /* Connection was closed by RST, timeout, ICMP error
631 * or another process disconnected us.
633 if (sk->sk_state == TCP_CLOSE)
636 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
637 * and error was received after socket entered established state.
638 * Hence, it is handled normally after connect() return successfully.
641 sock->state = SS_CONNECTED;
647 err = sock_error(sk) ? : -ECONNABORTED;
648 sock->state = SS_UNCONNECTED;
649 if (sk->sk_prot->disconnect(sk, flags))
650 sock->state = SS_DISCONNECTING;
653 EXPORT_SYMBOL(__inet_stream_connect);
655 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
656 int addr_len, int flags)
661 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
662 release_sock(sock->sk);
665 EXPORT_SYMBOL(inet_stream_connect);
668 * Accept a pending connection. The TCP layer now gives BSD semantics.
671 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
673 struct sock *sk1 = sock->sk;
675 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
682 sock_rps_record_flow(sk2);
683 WARN_ON(!((1 << sk2->sk_state) &
684 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
685 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
687 sock_graft(sk2, newsock);
689 newsock->state = SS_CONNECTED;
695 EXPORT_SYMBOL(inet_accept);
699 * This does both peername and sockname.
701 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
702 int *uaddr_len, int peer)
704 struct sock *sk = sock->sk;
705 struct inet_sock *inet = inet_sk(sk);
706 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
708 sin->sin_family = AF_INET;
710 if (!inet->inet_dport ||
711 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
714 sin->sin_port = inet->inet_dport;
715 sin->sin_addr.s_addr = inet->inet_daddr;
717 __be32 addr = inet->inet_rcv_saddr;
719 addr = inet->inet_saddr;
720 sin->sin_port = inet->inet_sport;
721 sin->sin_addr.s_addr = addr;
723 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
724 *uaddr_len = sizeof(*sin);
727 EXPORT_SYMBOL(inet_getname);
729 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
731 struct sock *sk = sock->sk;
733 sock_rps_record_flow(sk);
735 /* We may need to bind the socket. */
736 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
740 return sk->sk_prot->sendmsg(sk, msg, size);
742 EXPORT_SYMBOL(inet_sendmsg);
744 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
745 size_t size, int flags)
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 if (sk->sk_prot->sendpage)
757 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
758 return sock_no_sendpage(sock, page, offset, size, flags);
760 EXPORT_SYMBOL(inet_sendpage);
762 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
765 struct sock *sk = sock->sk;
769 sock_rps_record_flow(sk);
771 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
772 flags & ~MSG_DONTWAIT, &addr_len);
774 msg->msg_namelen = addr_len;
777 EXPORT_SYMBOL(inet_recvmsg);
779 int inet_shutdown(struct socket *sock, int how)
781 struct sock *sk = sock->sk;
784 /* This should really check to make sure
785 * the socket is a TCP socket. (WHY AC...)
787 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
790 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
794 if (sock->state == SS_CONNECTING) {
795 if ((1 << sk->sk_state) &
796 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
797 sock->state = SS_DISCONNECTING;
799 sock->state = SS_CONNECTED;
802 switch (sk->sk_state) {
805 /* Hack to wake up other listeners, who can poll for
806 POLLHUP, even on eg. unconnected UDP sockets -- RR */
808 sk->sk_shutdown |= how;
809 if (sk->sk_prot->shutdown)
810 sk->sk_prot->shutdown(sk, how);
813 /* Remaining two branches are temporary solution for missing
814 * close() in multithreaded environment. It is _not_ a good idea,
815 * but we have no choice until close() is repaired at VFS level.
818 if (!(how & RCV_SHUTDOWN))
822 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
823 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
827 /* Wake up anyone sleeping in poll. */
828 sk->sk_state_change(sk);
832 EXPORT_SYMBOL(inet_shutdown);
835 * ioctl() calls you can issue on an INET socket. Most of these are
836 * device configuration and stuff and very rarely used. Some ioctls
837 * pass on to the socket itself.
839 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
840 * loads the devconfigure module does its configuring and unloads it.
841 * There's a good 20K of config code hanging around the kernel.
844 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
846 struct sock *sk = sock->sk;
848 struct net *net = sock_net(sk);
852 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
855 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
860 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
865 err = arp_ioctl(net, cmd, (void __user *)arg);
878 err = devinet_ioctl(net, cmd, (void __user *)arg);
881 if (sk->sk_prot->ioctl)
882 err = sk->sk_prot->ioctl(sk, cmd, arg);
889 EXPORT_SYMBOL(inet_ioctl);
892 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
894 struct sock *sk = sock->sk;
895 int err = -ENOIOCTLCMD;
897 if (sk->sk_prot->compat_ioctl)
898 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
904 const struct proto_ops inet_stream_ops = {
906 .owner = THIS_MODULE,
907 .release = inet_release,
909 .connect = inet_stream_connect,
910 .socketpair = sock_no_socketpair,
911 .accept = inet_accept,
912 .getname = inet_getname,
915 .listen = inet_listen,
916 .shutdown = inet_shutdown,
917 .setsockopt = sock_common_setsockopt,
918 .getsockopt = sock_common_getsockopt,
919 .sendmsg = inet_sendmsg,
920 .recvmsg = inet_recvmsg,
921 .mmap = sock_no_mmap,
922 .sendpage = inet_sendpage,
923 .splice_read = tcp_splice_read,
925 .compat_setsockopt = compat_sock_common_setsockopt,
926 .compat_getsockopt = compat_sock_common_getsockopt,
927 .compat_ioctl = inet_compat_ioctl,
930 EXPORT_SYMBOL(inet_stream_ops);
932 const struct proto_ops inet_dgram_ops = {
934 .owner = THIS_MODULE,
935 .release = inet_release,
937 .connect = inet_dgram_connect,
938 .socketpair = sock_no_socketpair,
939 .accept = sock_no_accept,
940 .getname = inet_getname,
943 .listen = sock_no_listen,
944 .shutdown = inet_shutdown,
945 .setsockopt = sock_common_setsockopt,
946 .getsockopt = sock_common_getsockopt,
947 .sendmsg = inet_sendmsg,
948 .recvmsg = inet_recvmsg,
949 .mmap = sock_no_mmap,
950 .sendpage = inet_sendpage,
952 .compat_setsockopt = compat_sock_common_setsockopt,
953 .compat_getsockopt = compat_sock_common_getsockopt,
954 .compat_ioctl = inet_compat_ioctl,
957 EXPORT_SYMBOL(inet_dgram_ops);
960 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
963 static const struct proto_ops inet_sockraw_ops = {
965 .owner = THIS_MODULE,
966 .release = inet_release,
968 .connect = inet_dgram_connect,
969 .socketpair = sock_no_socketpair,
970 .accept = sock_no_accept,
971 .getname = inet_getname,
972 .poll = datagram_poll,
974 .listen = sock_no_listen,
975 .shutdown = inet_shutdown,
976 .setsockopt = sock_common_setsockopt,
977 .getsockopt = sock_common_getsockopt,
978 .sendmsg = inet_sendmsg,
979 .recvmsg = inet_recvmsg,
980 .mmap = sock_no_mmap,
981 .sendpage = inet_sendpage,
983 .compat_setsockopt = compat_sock_common_setsockopt,
984 .compat_getsockopt = compat_sock_common_getsockopt,
985 .compat_ioctl = inet_compat_ioctl,
989 static const struct net_proto_family inet_family_ops = {
991 .create = inet_create,
992 .owner = THIS_MODULE,
995 /* Upon startup we insert all the elements in inetsw_array[] into
996 * the linked list inetsw.
998 static struct inet_protosw inetsw_array[] =
1001 .type = SOCK_STREAM,
1002 .protocol = IPPROTO_TCP,
1004 .ops = &inet_stream_ops,
1005 .flags = INET_PROTOSW_PERMANENT |
1011 .protocol = IPPROTO_UDP,
1013 .ops = &inet_dgram_ops,
1014 .flags = INET_PROTOSW_PERMANENT,
1019 .protocol = IPPROTO_ICMP,
1021 .ops = &inet_dgram_ops,
1022 .flags = INET_PROTOSW_REUSE,
1027 .protocol = IPPROTO_IP, /* wild card */
1029 .ops = &inet_sockraw_ops,
1030 .flags = INET_PROTOSW_REUSE,
1034 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1036 void inet_register_protosw(struct inet_protosw *p)
1038 struct list_head *lh;
1039 struct inet_protosw *answer;
1040 int protocol = p->protocol;
1041 struct list_head *last_perm;
1043 spin_lock_bh(&inetsw_lock);
1045 if (p->type >= SOCK_MAX)
1048 /* If we are trying to override a permanent protocol, bail. */
1049 last_perm = &inetsw[p->type];
1050 list_for_each(lh, &inetsw[p->type]) {
1051 answer = list_entry(lh, struct inet_protosw, list);
1052 /* Check only the non-wild match. */
1053 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1055 if (protocol == answer->protocol)
1060 /* Add the new entry after the last permanent entry if any, so that
1061 * the new entry does not override a permanent entry when matched with
1062 * a wild-card protocol. But it is allowed to override any existing
1063 * non-permanent entry. This means that when we remove this entry, the
1064 * system automatically returns to the old behavior.
1066 list_add_rcu(&p->list, last_perm);
1068 spin_unlock_bh(&inetsw_lock);
1073 pr_err("Attempt to override permanent protocol %d\n", protocol);
1077 pr_err("Ignoring attempt to register invalid socket type %d\n",
1081 EXPORT_SYMBOL(inet_register_protosw);
1083 void inet_unregister_protosw(struct inet_protosw *p)
1085 if (INET_PROTOSW_PERMANENT & p->flags) {
1086 pr_err("Attempt to unregister permanent protocol %d\n",
1089 spin_lock_bh(&inetsw_lock);
1090 list_del_rcu(&p->list);
1091 spin_unlock_bh(&inetsw_lock);
1096 EXPORT_SYMBOL(inet_unregister_protosw);
1098 static int inet_sk_reselect_saddr(struct sock *sk)
1100 struct inet_sock *inet = inet_sk(sk);
1101 __be32 old_saddr = inet->inet_saddr;
1102 __be32 daddr = inet->inet_daddr;
1106 struct ip_options_rcu *inet_opt;
1108 inet_opt = rcu_dereference_protected(inet->inet_opt,
1109 sock_owned_by_user(sk));
1110 if (inet_opt && inet_opt->opt.srr)
1111 daddr = inet_opt->opt.faddr;
1113 /* Query new route. */
1114 fl4 = &inet->cork.fl.u.ip4;
1115 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1116 sk->sk_bound_dev_if, sk->sk_protocol,
1117 inet->inet_sport, inet->inet_dport, sk);
1121 sk_setup_caps(sk, &rt->dst);
1123 new_saddr = fl4->saddr;
1125 if (new_saddr == old_saddr)
1128 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1129 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1130 __func__, &old_saddr, &new_saddr);
1133 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1136 * XXX The only one ugly spot where we need to
1137 * XXX really change the sockets identity after
1138 * XXX it has entered the hashes. -DaveM
1140 * Besides that, it does not check for connection
1141 * uniqueness. Wait for troubles.
1143 return __sk_prot_rehash(sk);
1146 int inet_sk_rebuild_header(struct sock *sk)
1148 struct inet_sock *inet = inet_sk(sk);
1149 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1151 struct ip_options_rcu *inet_opt;
1155 /* Route is OK, nothing to do. */
1161 inet_opt = rcu_dereference(inet->inet_opt);
1162 daddr = inet->inet_daddr;
1163 if (inet_opt && inet_opt->opt.srr)
1164 daddr = inet_opt->opt.faddr;
1166 fl4 = &inet->cork.fl.u.ip4;
1167 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1168 inet->inet_dport, inet->inet_sport,
1169 sk->sk_protocol, RT_CONN_FLAGS(sk),
1170 sk->sk_bound_dev_if);
1173 sk_setup_caps(sk, &rt->dst);
1177 /* Routing failed... */
1178 sk->sk_route_caps = 0;
1180 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1181 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1183 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1184 sk->sk_state != TCP_SYN_SENT ||
1185 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1186 (err = inet_sk_reselect_saddr(sk)) != 0)
1187 sk->sk_err_soft = -err;
1192 EXPORT_SYMBOL(inet_sk_rebuild_header);
1194 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1195 netdev_features_t features)
1197 struct sk_buff *segs = ERR_PTR(-EINVAL);
1198 const struct net_offload *ops;
1199 unsigned int offset = 0;
1200 bool udpfrag, encap;
1207 if (unlikely(skb_shinfo(skb)->gso_type &
1217 SKB_GSO_UDP_TUNNEL |
1218 SKB_GSO_UDP_TUNNEL_CSUM |
1219 SKB_GSO_TUNNEL_REMCSUM |
1223 skb_reset_network_header(skb);
1224 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1225 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1230 if (ihl < sizeof(*iph))
1233 id = ntohs(iph->id);
1234 proto = iph->protocol;
1236 /* Warning: after this point, iph might be no longer valid */
1237 if (unlikely(!pskb_may_pull(skb, ihl)))
1239 __skb_pull(skb, ihl);
1241 encap = SKB_GSO_CB(skb)->encap_level > 0;
1243 features &= skb->dev->hw_enc_features;
1244 SKB_GSO_CB(skb)->encap_level += ihl;
1246 skb_reset_transport_header(skb);
1248 segs = ERR_PTR(-EPROTONOSUPPORT);
1250 if (skb->encapsulation &&
1251 skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1252 udpfrag = proto == IPPROTO_UDP && encap;
1254 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1256 ops = rcu_dereference(inet_offloads[proto]);
1257 if (likely(ops && ops->callbacks.gso_segment))
1258 segs = ops->callbacks.gso_segment(skb, features);
1260 if (IS_ERR_OR_NULL(segs))
1265 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1267 iph->id = htons(id);
1268 iph->frag_off = htons(offset >> 3);
1270 iph->frag_off |= htons(IP_MF);
1271 offset += skb->len - nhoff - ihl;
1273 iph->id = htons(id++);
1275 iph->tot_len = htons(skb->len - nhoff);
1278 skb_reset_inner_headers(skb);
1279 skb->network_header = (u8 *)iph - skb->head;
1280 } while ((skb = skb->next));
1286 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1287 struct sk_buff *skb)
1289 const struct net_offload *ops;
1290 struct sk_buff **pp = NULL;
1292 const struct iphdr *iph;
1299 off = skb_gro_offset(skb);
1300 hlen = off + sizeof(*iph);
1301 iph = skb_gro_header_fast(skb, off);
1302 if (skb_gro_header_hard(skb, hlen)) {
1303 iph = skb_gro_header_slow(skb, hlen, off);
1308 proto = iph->protocol;
1311 ops = rcu_dereference(inet_offloads[proto]);
1312 if (!ops || !ops->callbacks.gro_receive)
1315 if (*(u8 *)iph != 0x45)
1318 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1321 id = ntohl(*(__be32 *)&iph->id);
1322 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1325 for (p = *head; p; p = p->next) {
1328 if (!NAPI_GRO_CB(p)->same_flow)
1331 iph2 = (struct iphdr *)(p->data + off);
1332 /* The above works because, with the exception of the top
1333 * (inner most) layer, we only aggregate pkts with the same
1334 * hdr length so all the hdrs we'll need to verify will start
1335 * at the same offset.
1337 if ((iph->protocol ^ iph2->protocol) |
1338 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1339 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1340 NAPI_GRO_CB(p)->same_flow = 0;
1344 /* All fields must match except length and checksum. */
1345 NAPI_GRO_CB(p)->flush |=
1346 (iph->ttl ^ iph2->ttl) |
1347 (iph->tos ^ iph2->tos) |
1348 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1350 /* Save the IP ID check to be included later when we get to
1351 * the transport layer so only the inner most IP ID is checked.
1352 * This is because some GSO/TSO implementations do not
1353 * correctly increment the IP ID for the outer hdrs.
1355 NAPI_GRO_CB(p)->flush_id =
1356 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1357 NAPI_GRO_CB(p)->flush |= flush;
1360 NAPI_GRO_CB(skb)->flush |= flush;
1361 skb_set_network_header(skb, off);
1362 /* The above will be needed by the transport layer if there is one
1363 * immediately following this IP hdr.
1366 /* Note : No need to call skb_gro_postpull_rcsum() here,
1367 * as we already checked checksum over ipv4 header was 0
1369 skb_gro_pull(skb, sizeof(*iph));
1370 skb_set_transport_header(skb, skb_gro_offset(skb));
1372 pp = ops->callbacks.gro_receive(head, skb);
1378 NAPI_GRO_CB(skb)->flush |= flush;
1383 #define SECONDS_PER_DAY 86400
1385 /* inet_current_timestamp - Return IP network timestamp
1387 * Return milliseconds since midnight in network byte order.
1389 __be32 inet_current_timestamp(void)
1393 struct timespec64 ts;
1395 ktime_get_real_ts64(&ts);
1397 /* Get secs since midnight. */
1398 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1399 /* Convert to msecs. */
1400 msecs = secs * MSEC_PER_SEC;
1401 /* Convert nsec to msec. */
1402 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1404 /* Convert to network byte order. */
1405 return htons(msecs);
1407 EXPORT_SYMBOL(inet_current_timestamp);
1409 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1411 if (sk->sk_family == AF_INET)
1412 return ip_recv_error(sk, msg, len, addr_len);
1413 #if IS_ENABLED(CONFIG_IPV6)
1414 if (sk->sk_family == AF_INET6)
1415 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1420 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1422 __be16 newlen = htons(skb->len - nhoff);
1423 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1424 const struct net_offload *ops;
1425 int proto = iph->protocol;
1428 if (skb->encapsulation)
1429 skb_set_inner_network_header(skb, nhoff);
1431 csum_replace2(&iph->check, iph->tot_len, newlen);
1432 iph->tot_len = newlen;
1435 ops = rcu_dereference(inet_offloads[proto]);
1436 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1439 /* Only need to add sizeof(*iph) to get to the next hdr below
1440 * because any hdr with option will have been flushed in
1441 * inet_gro_receive().
1443 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1451 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1452 unsigned short type, unsigned char protocol,
1455 struct socket *sock;
1456 int rc = sock_create_kern(net, family, type, protocol, &sock);
1460 (*sk)->sk_allocation = GFP_ATOMIC;
1462 * Unhash it so that IP input processing does not even see it,
1463 * we do not wish this socket to see incoming packets.
1465 (*sk)->sk_prot->unhash(*sk);
1469 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1471 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1473 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1475 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1477 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1479 unsigned long res = 0;
1482 for_each_possible_cpu(i)
1483 res += snmp_get_cpu_field(mib, i, offt);
1486 EXPORT_SYMBOL_GPL(snmp_fold_field);
1488 #if BITS_PER_LONG==32
1490 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1491 size_t syncp_offset)
1494 struct u64_stats_sync *syncp;
1498 bhptr = per_cpu_ptr(mib, cpu);
1499 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1501 start = u64_stats_fetch_begin_irq(syncp);
1502 v = *(((u64 *)bhptr) + offt);
1503 } while (u64_stats_fetch_retry_irq(syncp, start));
1507 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1509 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1514 for_each_possible_cpu(cpu) {
1515 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1519 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1522 #ifdef CONFIG_IP_MULTICAST
1523 static const struct net_protocol igmp_protocol = {
1524 .handler = igmp_rcv,
1529 static const struct net_protocol tcp_protocol = {
1530 .early_demux = tcp_v4_early_demux,
1531 .handler = tcp_v4_rcv,
1532 .err_handler = tcp_v4_err,
1535 .icmp_strict_tag_validation = 1,
1538 static const struct net_protocol udp_protocol = {
1539 .early_demux = udp_v4_early_demux,
1541 .err_handler = udp_err,
1546 static const struct net_protocol icmp_protocol = {
1547 .handler = icmp_rcv,
1548 .err_handler = icmp_err,
1553 static __net_init int ipv4_mib_init_net(struct net *net)
1557 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1558 if (!net->mib.tcp_statistics)
1560 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1561 if (!net->mib.ip_statistics)
1564 for_each_possible_cpu(i) {
1565 struct ipstats_mib *af_inet_stats;
1566 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1567 u64_stats_init(&af_inet_stats->syncp);
1570 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1571 if (!net->mib.net_statistics)
1573 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1574 if (!net->mib.udp_statistics)
1576 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1577 if (!net->mib.udplite_statistics)
1578 goto err_udplite_mib;
1579 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1580 if (!net->mib.icmp_statistics)
1582 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1584 if (!net->mib.icmpmsg_statistics)
1585 goto err_icmpmsg_mib;
1591 free_percpu(net->mib.icmp_statistics);
1593 free_percpu(net->mib.udplite_statistics);
1595 free_percpu(net->mib.udp_statistics);
1597 free_percpu(net->mib.net_statistics);
1599 free_percpu(net->mib.ip_statistics);
1601 free_percpu(net->mib.tcp_statistics);
1606 static __net_exit void ipv4_mib_exit_net(struct net *net)
1608 kfree(net->mib.icmpmsg_statistics);
1609 free_percpu(net->mib.icmp_statistics);
1610 free_percpu(net->mib.udplite_statistics);
1611 free_percpu(net->mib.udp_statistics);
1612 free_percpu(net->mib.net_statistics);
1613 free_percpu(net->mib.ip_statistics);
1614 free_percpu(net->mib.tcp_statistics);
1617 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1618 .init = ipv4_mib_init_net,
1619 .exit = ipv4_mib_exit_net,
1622 static int __init init_ipv4_mibs(void)
1624 return register_pernet_subsys(&ipv4_mib_ops);
1627 static __net_init int inet_init_net(struct net *net)
1630 * Set defaults for local port range
1632 seqlock_init(&net->ipv4.ip_local_ports.lock);
1633 net->ipv4.ip_local_ports.range[0] = 32768;
1634 net->ipv4.ip_local_ports.range[1] = 60999;
1636 seqlock_init(&net->ipv4.ping_group_range.lock);
1638 * Sane defaults - nobody may create ping sockets.
1639 * Boot scripts should set this to distro-specific group.
1641 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1642 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1646 static __net_exit void inet_exit_net(struct net *net)
1650 static __net_initdata struct pernet_operations af_inet_ops = {
1651 .init = inet_init_net,
1652 .exit = inet_exit_net,
1655 static int __init init_inet_pernet_ops(void)
1657 return register_pernet_subsys(&af_inet_ops);
1660 static int ipv4_proc_init(void);
1663 * IP protocol layer initialiser
1666 static struct packet_offload ip_packet_offload __read_mostly = {
1667 .type = cpu_to_be16(ETH_P_IP),
1669 .gso_segment = inet_gso_segment,
1670 .gro_receive = inet_gro_receive,
1671 .gro_complete = inet_gro_complete,
1675 static const struct net_offload ipip_offload = {
1677 .gso_segment = inet_gso_segment,
1678 .gro_receive = inet_gro_receive,
1679 .gro_complete = inet_gro_complete,
1683 static int __init ipv4_offload_init(void)
1688 if (udpv4_offload_init() < 0)
1689 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1690 if (tcpv4_offload_init() < 0)
1691 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1693 dev_add_offload(&ip_packet_offload);
1694 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1698 fs_initcall(ipv4_offload_init);
1700 static struct packet_type ip_packet_type __read_mostly = {
1701 .type = cpu_to_be16(ETH_P_IP),
1705 static int __init inet_init(void)
1707 struct inet_protosw *q;
1708 struct list_head *r;
1711 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1713 rc = proto_register(&tcp_prot, 1);
1717 rc = proto_register(&udp_prot, 1);
1719 goto out_unregister_tcp_proto;
1721 rc = proto_register(&raw_prot, 1);
1723 goto out_unregister_udp_proto;
1725 rc = proto_register(&ping_prot, 1);
1727 goto out_unregister_raw_proto;
1730 * Tell SOCKET that we are alive...
1733 (void)sock_register(&inet_family_ops);
1735 #ifdef CONFIG_SYSCTL
1736 ip_static_sysctl_init();
1740 * Add all the base protocols.
1743 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1744 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1745 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1746 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1747 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1748 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1749 #ifdef CONFIG_IP_MULTICAST
1750 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1751 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1754 /* Register the socket-side information for inet_create. */
1755 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1758 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1759 inet_register_protosw(q);
1762 * Set the ARP module up
1768 * Set the IP module up
1775 /* Setup TCP slab cache for open requests. */
1778 /* Setup UDP memory threshold */
1781 /* Add UDP-Lite (RFC 3828) */
1782 udplite4_register();
1787 * Set the ICMP layer up
1790 if (icmp_init() < 0)
1791 panic("Failed to create the ICMP control socket.\n");
1794 * Initialise the multicast router
1796 #if defined(CONFIG_IP_MROUTE)
1798 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1801 if (init_inet_pernet_ops())
1802 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1804 * Initialise per-cpu ipv4 mibs
1807 if (init_ipv4_mibs())
1808 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1814 dev_add_pack(&ip_packet_type);
1816 ip_tunnel_core_init();
1821 out_unregister_raw_proto:
1822 proto_unregister(&raw_prot);
1823 out_unregister_udp_proto:
1824 proto_unregister(&udp_prot);
1825 out_unregister_tcp_proto:
1826 proto_unregister(&tcp_prot);
1830 fs_initcall(inet_init);
1832 /* ------------------------------------------------------------------------ */
1834 #ifdef CONFIG_PROC_FS
1835 static int __init ipv4_proc_init(void)
1839 if (raw_proc_init())
1841 if (tcp4_proc_init())
1843 if (udp4_proc_init())
1845 if (ping_proc_init())
1847 if (ip_misc_proc_init())
1864 #else /* CONFIG_PROC_FS */
1865 static int __init ipv4_proc_init(void)
1869 #endif /* CONFIG_PROC_FS */
1871 MODULE_ALIAS_NETPROTO(PF_INET);