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 #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>
124 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
125 #include <linux/android_aid.h>
127 static inline int current_has_network(void)
129 return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
132 static inline int current_has_network(void)
138 /* The inetsw table contains everything that inet_create needs to
139 * build a new socket.
141 static struct list_head inetsw[SOCK_MAX];
142 static DEFINE_SPINLOCK(inetsw_lock);
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 dst_release(sk->sk_rx_dst);
173 sk_refcnt_debug_dec(sk);
175 EXPORT_SYMBOL(inet_sock_destruct);
178 * The routines beyond this point handle the behaviour of an AF_INET
179 * socket object. Mostly it punts to the subprotocols of IP to do
184 * Automatically bind an unbound socket.
187 static int inet_autobind(struct sock *sk)
189 struct inet_sock *inet;
190 /* We may need to bind the socket. */
193 if (!inet->inet_num) {
194 if (sk->sk_prot->get_port(sk, 0)) {
198 inet->inet_sport = htons(inet->inet_num);
205 * Move a socket into listening state.
207 int inet_listen(struct socket *sock, int backlog)
209 struct sock *sk = sock->sk;
210 unsigned char old_state;
216 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
219 old_state = sk->sk_state;
220 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
223 /* Really, if the socket is already in listen state
224 * we can only allow the backlog to be adjusted.
226 if (old_state != TCP_LISTEN) {
227 /* Check special setups for testing purpose to enable TFO w/o
228 * requiring TCP_FASTOPEN sockopt.
229 * Note that only TCP sockets (SOCK_STREAM) will reach here.
230 * Also fastopenq may already been allocated because this
231 * socket was in TCP_LISTEN state previously but was
232 * shutdown() (rather than close()).
234 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
235 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
236 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
237 fastopen_queue_tune(sk, backlog);
238 else if ((sysctl_tcp_fastopen &
239 TFO_SERVER_WO_SOCKOPT2) != 0)
240 fastopen_queue_tune(sk,
241 ((uint)sysctl_tcp_fastopen) >> 16);
243 tcp_fastopen_init_key_once(true);
245 err = inet_csk_listen_start(sk, backlog);
249 sk->sk_max_ack_backlog = backlog;
256 EXPORT_SYMBOL(inet_listen);
259 * Create an inet socket.
262 static int inet_create(struct net *net, struct socket *sock, int protocol,
266 struct inet_protosw *answer;
267 struct inet_sock *inet;
268 struct proto *answer_prot;
269 unsigned char answer_flags;
270 int try_loading_module = 0;
273 if (protocol < 0 || protocol >= IPPROTO_MAX)
276 if (!current_has_network())
279 sock->state = SS_UNCONNECTED;
281 /* Look for the requested type/protocol pair. */
283 err = -ESOCKTNOSUPPORT;
285 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
288 /* Check the non-wild match. */
289 if (protocol == answer->protocol) {
290 if (protocol != IPPROTO_IP)
293 /* Check for the two wild cases. */
294 if (IPPROTO_IP == protocol) {
295 protocol = answer->protocol;
298 if (IPPROTO_IP == answer->protocol)
301 err = -EPROTONOSUPPORT;
305 if (try_loading_module < 2) {
308 * Be more specific, e.g. net-pf-2-proto-132-type-1
309 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
311 if (++try_loading_module == 1)
312 request_module("net-pf-%d-proto-%d-type-%d",
313 PF_INET, protocol, sock->type);
315 * Fall back to generic, e.g. net-pf-2-proto-132
316 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
319 request_module("net-pf-%d-proto-%d",
321 goto lookup_protocol;
327 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
330 sock->ops = answer->ops;
331 answer_prot = answer->prot;
332 answer_flags = answer->flags;
335 WARN_ON(!answer_prot->slab);
338 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
343 if (INET_PROTOSW_REUSE & answer_flags)
344 sk->sk_reuse = SK_CAN_REUSE;
347 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
351 if (SOCK_RAW == sock->type) {
352 inet->inet_num = protocol;
353 if (IPPROTO_RAW == protocol)
357 if (net->ipv4.sysctl_ip_no_pmtu_disc)
358 inet->pmtudisc = IP_PMTUDISC_DONT;
360 inet->pmtudisc = IP_PMTUDISC_WANT;
364 sock_init_data(sock, sk);
366 sk->sk_destruct = inet_sock_destruct;
367 sk->sk_protocol = protocol;
368 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
375 inet->mc_list = NULL;
378 sk_refcnt_debug_inc(sk);
380 if (inet->inet_num) {
381 /* It assumes that any protocol which allows
382 * the user to assign a number at socket
383 * creation time automatically
386 inet->inet_sport = htons(inet->inet_num);
387 /* Add to protocol hash chains. */
388 sk->sk_prot->hash(sk);
391 if (sk->sk_prot->init) {
392 err = sk->sk_prot->init(sk);
394 sk_common_release(sk);
405 * The peer socket should always be NULL (or else). When we call this
406 * function we are destroying the object and from then on nobody
407 * should refer to it.
409 int inet_release(struct socket *sock)
411 struct sock *sk = sock->sk;
416 /* Applications forget to leave groups before exiting */
417 ip_mc_drop_socket(sk);
419 /* If linger is set, we don't return until the close
420 * is complete. Otherwise we return immediately. The
421 * actually closing is done the same either way.
423 * If the close is due to the process exiting, we never
427 if (sock_flag(sk, SOCK_LINGER) &&
428 !(current->flags & PF_EXITING))
429 timeout = sk->sk_lingertime;
431 sk->sk_prot->close(sk, timeout);
435 EXPORT_SYMBOL(inet_release);
437 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
439 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
440 struct sock *sk = sock->sk;
441 struct inet_sock *inet = inet_sk(sk);
442 struct net *net = sock_net(sk);
445 u32 tb_id = RT_TABLE_LOCAL;
448 /* If the socket has its own bind function then use it. (RAW) */
449 if (sk->sk_prot->bind) {
450 err = sk->sk_prot->bind(sk, uaddr, addr_len);
454 if (addr_len < sizeof(struct sockaddr_in))
457 if (addr->sin_family != AF_INET) {
458 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
459 * only if s_addr is INADDR_ANY.
462 if (addr->sin_family != AF_UNSPEC ||
463 addr->sin_addr.s_addr != htonl(INADDR_ANY))
467 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
468 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
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 (!net->ipv4.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 &&
489 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
492 /* We keep a pair of addresses. rcv_saddr is the one
493 * used by hash lookups, and saddr is used for transmit.
495 * In the BSD API these are the same except where it
496 * would be illegal to use them (multicast/broadcast) in
497 * which case the sending device address is used.
501 /* Check these errors (active socket, double bind). */
503 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
504 goto out_release_sock;
506 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
507 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
508 inet->inet_saddr = 0; /* Use device */
510 /* Make sure we are allowed to bind here. */
511 if ((snum || !inet->bind_address_no_port) &&
512 sk->sk_prot->get_port(sk, snum)) {
513 inet->inet_saddr = inet->inet_rcv_saddr = 0;
515 goto out_release_sock;
518 if (inet->inet_rcv_saddr)
519 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
521 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
522 inet->inet_sport = htons(inet->inet_num);
523 inet->inet_daddr = 0;
524 inet->inet_dport = 0;
532 EXPORT_SYMBOL(inet_bind);
534 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
535 int addr_len, int flags)
537 struct sock *sk = sock->sk;
539 if (addr_len < sizeof(uaddr->sa_family))
541 if (uaddr->sa_family == AF_UNSPEC)
542 return sk->sk_prot->disconnect(sk, flags);
544 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
546 return sk->sk_prot->connect(sk, uaddr, addr_len);
548 EXPORT_SYMBOL(inet_dgram_connect);
550 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
554 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
555 sk->sk_write_pending += writebias;
557 /* Basic assumption: if someone sets sk->sk_err, he _must_
558 * change state of the socket from TCP_SYN_*.
559 * Connect() does not allow to get error notifications
560 * without closing the socket.
562 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
564 timeo = schedule_timeout(timeo);
566 if (signal_pending(current) || !timeo)
568 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
570 finish_wait(sk_sleep(sk), &wait);
571 sk->sk_write_pending -= writebias;
576 * Connect to a remote host. There is regrettably still a little
577 * TCP 'magic' in here.
579 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
580 int addr_len, int flags)
582 struct sock *sk = sock->sk;
586 if (addr_len < sizeof(uaddr->sa_family))
589 if (uaddr->sa_family == AF_UNSPEC) {
590 err = sk->sk_prot->disconnect(sk, flags);
591 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
595 switch (sock->state) {
604 /* Fall out of switch with err, set for this state */
608 if (sk->sk_state != TCP_CLOSE)
611 err = sk->sk_prot->connect(sk, uaddr, addr_len);
615 sock->state = SS_CONNECTING;
617 /* Just entered SS_CONNECTING state; the only
618 * difference is that return value in non-blocking
619 * case is EINPROGRESS, rather than EALREADY.
625 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
627 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
628 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
629 tcp_sk(sk)->fastopen_req &&
630 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
632 /* Error code is set above */
633 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
636 err = sock_intr_errno(timeo);
637 if (signal_pending(current))
641 /* Connection was closed by RST, timeout, ICMP error
642 * or another process disconnected us.
644 if (sk->sk_state == TCP_CLOSE)
647 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
648 * and error was received after socket entered established state.
649 * Hence, it is handled normally after connect() return successfully.
652 sock->state = SS_CONNECTED;
658 err = sock_error(sk) ? : -ECONNABORTED;
659 sock->state = SS_UNCONNECTED;
660 if (sk->sk_prot->disconnect(sk, flags))
661 sock->state = SS_DISCONNECTING;
664 EXPORT_SYMBOL(__inet_stream_connect);
666 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
667 int addr_len, int flags)
672 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
673 release_sock(sock->sk);
676 EXPORT_SYMBOL(inet_stream_connect);
679 * Accept a pending connection. The TCP layer now gives BSD semantics.
682 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
684 struct sock *sk1 = sock->sk;
686 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
693 sock_rps_record_flow(sk2);
694 WARN_ON(!((1 << sk2->sk_state) &
695 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
696 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
698 sock_graft(sk2, newsock);
700 newsock->state = SS_CONNECTED;
706 EXPORT_SYMBOL(inet_accept);
710 * This does both peername and sockname.
712 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
713 int *uaddr_len, int peer)
715 struct sock *sk = sock->sk;
716 struct inet_sock *inet = inet_sk(sk);
717 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
719 sin->sin_family = AF_INET;
721 if (!inet->inet_dport ||
722 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
725 sin->sin_port = inet->inet_dport;
726 sin->sin_addr.s_addr = inet->inet_daddr;
728 __be32 addr = inet->inet_rcv_saddr;
730 addr = inet->inet_saddr;
731 sin->sin_port = inet->inet_sport;
732 sin->sin_addr.s_addr = addr;
734 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
735 *uaddr_len = sizeof(*sin);
738 EXPORT_SYMBOL(inet_getname);
740 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
742 struct sock *sk = sock->sk;
744 sock_rps_record_flow(sk);
746 /* We may need to bind the socket. */
747 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
751 return sk->sk_prot->sendmsg(sk, msg, size);
753 EXPORT_SYMBOL(inet_sendmsg);
755 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
756 size_t size, int flags)
758 struct sock *sk = sock->sk;
760 sock_rps_record_flow(sk);
762 /* We may need to bind the socket. */
763 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
767 if (sk->sk_prot->sendpage)
768 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
769 return sock_no_sendpage(sock, page, offset, size, flags);
771 EXPORT_SYMBOL(inet_sendpage);
773 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
776 struct sock *sk = sock->sk;
780 sock_rps_record_flow(sk);
782 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
783 flags & ~MSG_DONTWAIT, &addr_len);
785 msg->msg_namelen = addr_len;
788 EXPORT_SYMBOL(inet_recvmsg);
790 int inet_shutdown(struct socket *sock, int how)
792 struct sock *sk = sock->sk;
795 /* This should really check to make sure
796 * the socket is a TCP socket. (WHY AC...)
798 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
801 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
805 if (sock->state == SS_CONNECTING) {
806 if ((1 << sk->sk_state) &
807 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
808 sock->state = SS_DISCONNECTING;
810 sock->state = SS_CONNECTED;
813 switch (sk->sk_state) {
816 /* Hack to wake up other listeners, who can poll for
817 POLLHUP, even on eg. unconnected UDP sockets -- RR */
819 sk->sk_shutdown |= how;
820 if (sk->sk_prot->shutdown)
821 sk->sk_prot->shutdown(sk, how);
824 /* Remaining two branches are temporary solution for missing
825 * close() in multithreaded environment. It is _not_ a good idea,
826 * but we have no choice until close() is repaired at VFS level.
829 if (!(how & RCV_SHUTDOWN))
833 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
834 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
838 /* Wake up anyone sleeping in poll. */
839 sk->sk_state_change(sk);
843 EXPORT_SYMBOL(inet_shutdown);
846 * ioctl() calls you can issue on an INET socket. Most of these are
847 * device configuration and stuff and very rarely used. Some ioctls
848 * pass on to the socket itself.
850 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
851 * loads the devconfigure module does its configuring and unloads it.
852 * There's a good 20K of config code hanging around the kernel.
855 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
857 struct sock *sk = sock->sk;
859 struct net *net = sock_net(sk);
863 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
866 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
871 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
876 err = arp_ioctl(net, cmd, (void __user *)arg);
889 err = devinet_ioctl(net, cmd, (void __user *)arg);
892 if (sk->sk_prot->ioctl)
893 err = sk->sk_prot->ioctl(sk, cmd, arg);
900 EXPORT_SYMBOL(inet_ioctl);
903 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
905 struct sock *sk = sock->sk;
906 int err = -ENOIOCTLCMD;
908 if (sk->sk_prot->compat_ioctl)
909 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
915 const struct proto_ops inet_stream_ops = {
917 .owner = THIS_MODULE,
918 .release = inet_release,
920 .connect = inet_stream_connect,
921 .socketpair = sock_no_socketpair,
922 .accept = inet_accept,
923 .getname = inet_getname,
926 .listen = inet_listen,
927 .shutdown = inet_shutdown,
928 .setsockopt = sock_common_setsockopt,
929 .getsockopt = sock_common_getsockopt,
930 .sendmsg = inet_sendmsg,
931 .recvmsg = inet_recvmsg,
932 .mmap = sock_no_mmap,
933 .sendpage = inet_sendpage,
934 .splice_read = tcp_splice_read,
936 .compat_setsockopt = compat_sock_common_setsockopt,
937 .compat_getsockopt = compat_sock_common_getsockopt,
938 .compat_ioctl = inet_compat_ioctl,
941 EXPORT_SYMBOL(inet_stream_ops);
943 const struct proto_ops inet_dgram_ops = {
945 .owner = THIS_MODULE,
946 .release = inet_release,
948 .connect = inet_dgram_connect,
949 .socketpair = sock_no_socketpair,
950 .accept = sock_no_accept,
951 .getname = inet_getname,
954 .listen = sock_no_listen,
955 .shutdown = inet_shutdown,
956 .setsockopt = sock_common_setsockopt,
957 .getsockopt = sock_common_getsockopt,
958 .sendmsg = inet_sendmsg,
959 .recvmsg = inet_recvmsg,
960 .mmap = sock_no_mmap,
961 .sendpage = inet_sendpage,
963 .compat_setsockopt = compat_sock_common_setsockopt,
964 .compat_getsockopt = compat_sock_common_getsockopt,
965 .compat_ioctl = inet_compat_ioctl,
968 EXPORT_SYMBOL(inet_dgram_ops);
971 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
974 static const struct proto_ops inet_sockraw_ops = {
976 .owner = THIS_MODULE,
977 .release = inet_release,
979 .connect = inet_dgram_connect,
980 .socketpair = sock_no_socketpair,
981 .accept = sock_no_accept,
982 .getname = inet_getname,
983 .poll = datagram_poll,
985 .listen = sock_no_listen,
986 .shutdown = inet_shutdown,
987 .setsockopt = sock_common_setsockopt,
988 .getsockopt = sock_common_getsockopt,
989 .sendmsg = inet_sendmsg,
990 .recvmsg = inet_recvmsg,
991 .mmap = sock_no_mmap,
992 .sendpage = inet_sendpage,
994 .compat_setsockopt = compat_sock_common_setsockopt,
995 .compat_getsockopt = compat_sock_common_getsockopt,
996 .compat_ioctl = inet_compat_ioctl,
1000 static const struct net_proto_family inet_family_ops = {
1002 .create = inet_create,
1003 .owner = THIS_MODULE,
1006 /* Upon startup we insert all the elements in inetsw_array[] into
1007 * the linked list inetsw.
1009 static struct inet_protosw inetsw_array[] =
1012 .type = SOCK_STREAM,
1013 .protocol = IPPROTO_TCP,
1015 .ops = &inet_stream_ops,
1016 .flags = INET_PROTOSW_PERMANENT |
1022 .protocol = IPPROTO_UDP,
1024 .ops = &inet_dgram_ops,
1025 .flags = INET_PROTOSW_PERMANENT,
1030 .protocol = IPPROTO_ICMP,
1032 .ops = &inet_dgram_ops,
1033 .flags = INET_PROTOSW_REUSE,
1038 .protocol = IPPROTO_IP, /* wild card */
1040 .ops = &inet_sockraw_ops,
1041 .flags = INET_PROTOSW_REUSE,
1045 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1047 void inet_register_protosw(struct inet_protosw *p)
1049 struct list_head *lh;
1050 struct inet_protosw *answer;
1051 int protocol = p->protocol;
1052 struct list_head *last_perm;
1054 spin_lock_bh(&inetsw_lock);
1056 if (p->type >= SOCK_MAX)
1059 /* If we are trying to override a permanent protocol, bail. */
1060 last_perm = &inetsw[p->type];
1061 list_for_each(lh, &inetsw[p->type]) {
1062 answer = list_entry(lh, struct inet_protosw, list);
1063 /* Check only the non-wild match. */
1064 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1066 if (protocol == answer->protocol)
1071 /* Add the new entry after the last permanent entry if any, so that
1072 * the new entry does not override a permanent entry when matched with
1073 * a wild-card protocol. But it is allowed to override any existing
1074 * non-permanent entry. This means that when we remove this entry, the
1075 * system automatically returns to the old behavior.
1077 list_add_rcu(&p->list, last_perm);
1079 spin_unlock_bh(&inetsw_lock);
1084 pr_err("Attempt to override permanent protocol %d\n", protocol);
1088 pr_err("Ignoring attempt to register invalid socket type %d\n",
1092 EXPORT_SYMBOL(inet_register_protosw);
1094 void inet_unregister_protosw(struct inet_protosw *p)
1096 if (INET_PROTOSW_PERMANENT & p->flags) {
1097 pr_err("Attempt to unregister permanent protocol %d\n",
1100 spin_lock_bh(&inetsw_lock);
1101 list_del_rcu(&p->list);
1102 spin_unlock_bh(&inetsw_lock);
1107 EXPORT_SYMBOL(inet_unregister_protosw);
1110 * Shall we try to damage output packets if routing dev changes?
1113 int sysctl_ip_dynaddr __read_mostly;
1115 static int inet_sk_reselect_saddr(struct sock *sk)
1117 struct inet_sock *inet = inet_sk(sk);
1118 __be32 old_saddr = inet->inet_saddr;
1119 __be32 daddr = inet->inet_daddr;
1123 struct ip_options_rcu *inet_opt;
1125 inet_opt = rcu_dereference_protected(inet->inet_opt,
1126 sock_owned_by_user(sk));
1127 if (inet_opt && inet_opt->opt.srr)
1128 daddr = inet_opt->opt.faddr;
1130 /* Query new route. */
1131 fl4 = &inet->cork.fl.u.ip4;
1132 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1133 sk->sk_bound_dev_if, sk->sk_protocol,
1134 inet->inet_sport, inet->inet_dport, sk);
1138 sk_setup_caps(sk, &rt->dst);
1140 new_saddr = fl4->saddr;
1142 if (new_saddr == old_saddr)
1145 if (sysctl_ip_dynaddr > 1) {
1146 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1147 __func__, &old_saddr, &new_saddr);
1150 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1153 * XXX The only one ugly spot where we need to
1154 * XXX really change the sockets identity after
1155 * XXX it has entered the hashes. -DaveM
1157 * Besides that, it does not check for connection
1158 * uniqueness. Wait for troubles.
1160 __sk_prot_rehash(sk);
1164 int inet_sk_rebuild_header(struct sock *sk)
1166 struct inet_sock *inet = inet_sk(sk);
1167 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1169 struct ip_options_rcu *inet_opt;
1173 /* Route is OK, nothing to do. */
1179 inet_opt = rcu_dereference(inet->inet_opt);
1180 daddr = inet->inet_daddr;
1181 if (inet_opt && inet_opt->opt.srr)
1182 daddr = inet_opt->opt.faddr;
1184 fl4 = &inet->cork.fl.u.ip4;
1185 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1186 inet->inet_dport, inet->inet_sport,
1187 sk->sk_protocol, RT_CONN_FLAGS(sk),
1188 sk->sk_bound_dev_if);
1191 sk_setup_caps(sk, &rt->dst);
1195 /* Routing failed... */
1196 sk->sk_route_caps = 0;
1198 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1199 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1201 if (!sysctl_ip_dynaddr ||
1202 sk->sk_state != TCP_SYN_SENT ||
1203 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1204 (err = inet_sk_reselect_saddr(sk)) != 0)
1205 sk->sk_err_soft = -err;
1210 EXPORT_SYMBOL(inet_sk_rebuild_header);
1212 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1213 netdev_features_t features)
1215 struct sk_buff *segs = ERR_PTR(-EINVAL);
1216 const struct net_offload *ops;
1217 unsigned int offset = 0;
1218 bool udpfrag, encap;
1225 if (unlikely(skb_shinfo(skb)->gso_type &
1235 SKB_GSO_UDP_TUNNEL |
1236 SKB_GSO_UDP_TUNNEL_CSUM |
1237 SKB_GSO_TUNNEL_REMCSUM |
1241 skb_reset_network_header(skb);
1242 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1243 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1248 if (ihl < sizeof(*iph))
1251 id = ntohs(iph->id);
1252 proto = iph->protocol;
1254 /* Warning: after this point, iph might be no longer valid */
1255 if (unlikely(!pskb_may_pull(skb, ihl)))
1257 __skb_pull(skb, ihl);
1259 encap = SKB_GSO_CB(skb)->encap_level > 0;
1261 features &= skb->dev->hw_enc_features;
1262 SKB_GSO_CB(skb)->encap_level += ihl;
1264 skb_reset_transport_header(skb);
1266 segs = ERR_PTR(-EPROTONOSUPPORT);
1268 if (skb->encapsulation &&
1269 skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1270 udpfrag = proto == IPPROTO_UDP && encap;
1272 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1274 ops = rcu_dereference(inet_offloads[proto]);
1275 if (likely(ops && ops->callbacks.gso_segment))
1276 segs = ops->callbacks.gso_segment(skb, features);
1278 if (IS_ERR_OR_NULL(segs))
1283 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1285 iph->id = htons(id);
1286 iph->frag_off = htons(offset >> 3);
1288 iph->frag_off |= htons(IP_MF);
1289 offset += skb->len - nhoff - ihl;
1291 iph->id = htons(id++);
1293 iph->tot_len = htons(skb->len - nhoff);
1296 skb_reset_inner_headers(skb);
1297 skb->network_header = (u8 *)iph - skb->head;
1298 } while ((skb = skb->next));
1304 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1305 struct sk_buff *skb)
1307 const struct net_offload *ops;
1308 struct sk_buff **pp = NULL;
1310 const struct iphdr *iph;
1317 off = skb_gro_offset(skb);
1318 hlen = off + sizeof(*iph);
1319 iph = skb_gro_header_fast(skb, off);
1320 if (skb_gro_header_hard(skb, hlen)) {
1321 iph = skb_gro_header_slow(skb, hlen, off);
1326 proto = iph->protocol;
1329 ops = rcu_dereference(inet_offloads[proto]);
1330 if (!ops || !ops->callbacks.gro_receive)
1333 if (*(u8 *)iph != 0x45)
1336 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1339 id = ntohl(*(__be32 *)&iph->id);
1340 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1343 for (p = *head; p; p = p->next) {
1346 if (!NAPI_GRO_CB(p)->same_flow)
1349 iph2 = (struct iphdr *)(p->data + off);
1350 /* The above works because, with the exception of the top
1351 * (inner most) layer, we only aggregate pkts with the same
1352 * hdr length so all the hdrs we'll need to verify will start
1353 * at the same offset.
1355 if ((iph->protocol ^ iph2->protocol) |
1356 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1357 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1358 NAPI_GRO_CB(p)->same_flow = 0;
1362 /* All fields must match except length and checksum. */
1363 NAPI_GRO_CB(p)->flush |=
1364 (iph->ttl ^ iph2->ttl) |
1365 (iph->tos ^ iph2->tos) |
1366 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1368 /* Save the IP ID check to be included later when we get to
1369 * the transport layer so only the inner most IP ID is checked.
1370 * This is because some GSO/TSO implementations do not
1371 * correctly increment the IP ID for the outer hdrs.
1373 NAPI_GRO_CB(p)->flush_id =
1374 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1375 NAPI_GRO_CB(p)->flush |= flush;
1378 NAPI_GRO_CB(skb)->flush |= flush;
1379 skb_set_network_header(skb, off);
1380 /* The above will be needed by the transport layer if there is one
1381 * immediately following this IP hdr.
1384 /* Note : No need to call skb_gro_postpull_rcsum() here,
1385 * as we already checked checksum over ipv4 header was 0
1387 skb_gro_pull(skb, sizeof(*iph));
1388 skb_set_transport_header(skb, skb_gro_offset(skb));
1390 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1396 NAPI_GRO_CB(skb)->flush |= flush;
1401 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1402 struct sk_buff *skb)
1404 if (NAPI_GRO_CB(skb)->encap_mark) {
1405 NAPI_GRO_CB(skb)->flush = 1;
1409 NAPI_GRO_CB(skb)->encap_mark = 1;
1411 return inet_gro_receive(head, skb);
1414 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1416 if (sk->sk_family == AF_INET)
1417 return ip_recv_error(sk, msg, len, addr_len);
1418 #if IS_ENABLED(CONFIG_IPV6)
1419 if (sk->sk_family == AF_INET6)
1420 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1425 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1427 __be16 newlen = htons(skb->len - nhoff);
1428 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1429 const struct net_offload *ops;
1430 int proto = iph->protocol;
1433 if (skb->encapsulation)
1434 skb_set_inner_network_header(skb, nhoff);
1436 csum_replace2(&iph->check, iph->tot_len, newlen);
1437 iph->tot_len = newlen;
1440 ops = rcu_dereference(inet_offloads[proto]);
1441 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1444 /* Only need to add sizeof(*iph) to get to the next hdr below
1445 * because any hdr with option will have been flushed in
1446 * inet_gro_receive().
1448 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1456 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1458 skb->encapsulation = 1;
1459 skb_shinfo(skb)->gso_type |= SKB_GSO_IPIP;
1460 return inet_gro_complete(skb, nhoff);
1463 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1464 unsigned short type, unsigned char protocol,
1467 struct socket *sock;
1468 int rc = sock_create_kern(net, family, type, protocol, &sock);
1472 (*sk)->sk_allocation = GFP_ATOMIC;
1474 * Unhash it so that IP input processing does not even see it,
1475 * we do not wish this socket to see incoming packets.
1477 (*sk)->sk_prot->unhash(*sk);
1481 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1483 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1485 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1487 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1489 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1491 unsigned long res = 0;
1494 for_each_possible_cpu(i)
1495 res += snmp_get_cpu_field(mib, i, offt);
1498 EXPORT_SYMBOL_GPL(snmp_fold_field);
1500 #if BITS_PER_LONG==32
1502 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1503 size_t syncp_offset)
1506 struct u64_stats_sync *syncp;
1510 bhptr = per_cpu_ptr(mib, cpu);
1511 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1513 start = u64_stats_fetch_begin_irq(syncp);
1514 v = *(((u64 *)bhptr) + offt);
1515 } while (u64_stats_fetch_retry_irq(syncp, start));
1519 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1521 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1526 for_each_possible_cpu(cpu) {
1527 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1531 EXPORT_SYMBOL_GPL(snmp_fold_field64);
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 .early_demux = tcp_v4_early_demux,
1543 .handler = tcp_v4_rcv,
1544 .err_handler = tcp_v4_err,
1547 .icmp_strict_tag_validation = 1,
1550 static const struct net_protocol udp_protocol = {
1551 .early_demux = udp_v4_early_demux,
1553 .err_handler = udp_err,
1558 static const struct net_protocol icmp_protocol = {
1559 .handler = icmp_rcv,
1560 .err_handler = icmp_err,
1565 static __net_init int ipv4_mib_init_net(struct net *net)
1569 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1570 if (!net->mib.tcp_statistics)
1572 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1573 if (!net->mib.ip_statistics)
1576 for_each_possible_cpu(i) {
1577 struct ipstats_mib *af_inet_stats;
1578 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1579 u64_stats_init(&af_inet_stats->syncp);
1582 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1583 if (!net->mib.net_statistics)
1585 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1586 if (!net->mib.udp_statistics)
1588 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1589 if (!net->mib.udplite_statistics)
1590 goto err_udplite_mib;
1591 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1592 if (!net->mib.icmp_statistics)
1594 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1596 if (!net->mib.icmpmsg_statistics)
1597 goto err_icmpmsg_mib;
1603 free_percpu(net->mib.icmp_statistics);
1605 free_percpu(net->mib.udplite_statistics);
1607 free_percpu(net->mib.udp_statistics);
1609 free_percpu(net->mib.net_statistics);
1611 free_percpu(net->mib.ip_statistics);
1613 free_percpu(net->mib.tcp_statistics);
1618 static __net_exit void ipv4_mib_exit_net(struct net *net)
1620 kfree(net->mib.icmpmsg_statistics);
1621 free_percpu(net->mib.icmp_statistics);
1622 free_percpu(net->mib.udplite_statistics);
1623 free_percpu(net->mib.udp_statistics);
1624 free_percpu(net->mib.net_statistics);
1625 free_percpu(net->mib.ip_statistics);
1626 free_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 __net_init int inet_init_net(struct net *net)
1642 * Set defaults for local port range
1644 seqlock_init(&net->ipv4.ip_local_ports.lock);
1645 net->ipv4.ip_local_ports.range[0] = 32768;
1646 net->ipv4.ip_local_ports.range[1] = 60999;
1648 seqlock_init(&net->ipv4.ping_group_range.lock);
1650 * Sane defaults - nobody may create ping sockets.
1651 * Boot scripts should set this to distro-specific group.
1653 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1654 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1658 static __net_exit void inet_exit_net(struct net *net)
1662 static __net_initdata struct pernet_operations af_inet_ops = {
1663 .init = inet_init_net,
1664 .exit = inet_exit_net,
1667 static int __init init_inet_pernet_ops(void)
1669 return register_pernet_subsys(&af_inet_ops);
1672 static int ipv4_proc_init(void);
1675 * IP protocol layer initialiser
1678 static struct packet_offload ip_packet_offload __read_mostly = {
1679 .type = cpu_to_be16(ETH_P_IP),
1681 .gso_segment = inet_gso_segment,
1682 .gro_receive = inet_gro_receive,
1683 .gro_complete = inet_gro_complete,
1687 static const struct net_offload ipip_offload = {
1689 .gso_segment = inet_gso_segment,
1690 .gro_receive = ipip_gro_receive,
1691 .gro_complete = ipip_gro_complete,
1695 static int __init ipv4_offload_init(void)
1700 if (udpv4_offload_init() < 0)
1701 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1702 if (tcpv4_offload_init() < 0)
1703 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1705 dev_add_offload(&ip_packet_offload);
1706 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1710 fs_initcall(ipv4_offload_init);
1712 static struct packet_type ip_packet_type __read_mostly = {
1713 .type = cpu_to_be16(ETH_P_IP),
1717 static int __init inet_init(void)
1719 struct inet_protosw *q;
1720 struct list_head *r;
1723 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1725 rc = proto_register(&tcp_prot, 1);
1729 rc = proto_register(&udp_prot, 1);
1731 goto out_unregister_tcp_proto;
1733 rc = proto_register(&raw_prot, 1);
1735 goto out_unregister_udp_proto;
1737 rc = proto_register(&ping_prot, 1);
1739 goto out_unregister_raw_proto;
1742 * Tell SOCKET that we are alive...
1745 (void)sock_register(&inet_family_ops);
1747 #ifdef CONFIG_SYSCTL
1748 ip_static_sysctl_init();
1752 * Add all the base protocols.
1755 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1756 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1757 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1758 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1759 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1760 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1761 #ifdef CONFIG_IP_MULTICAST
1762 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1763 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1766 /* Register the socket-side information for inet_create. */
1767 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1770 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1771 inet_register_protosw(q);
1774 * Set the ARP module up
1780 * Set the IP module up
1787 /* Setup TCP slab cache for open requests. */
1790 /* Setup UDP memory threshold */
1793 /* Add UDP-Lite (RFC 3828) */
1794 udplite4_register();
1799 * Set the ICMP layer up
1802 if (icmp_init() < 0)
1803 panic("Failed to create the ICMP control socket.\n");
1806 * Initialise the multicast router
1808 #if defined(CONFIG_IP_MROUTE)
1810 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1813 if (init_inet_pernet_ops())
1814 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1816 * Initialise per-cpu ipv4 mibs
1819 if (init_ipv4_mibs())
1820 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1826 dev_add_pack(&ip_packet_type);
1828 ip_tunnel_core_init();
1833 out_unregister_raw_proto:
1834 proto_unregister(&raw_prot);
1835 out_unregister_udp_proto:
1836 proto_unregister(&udp_prot);
1837 out_unregister_tcp_proto:
1838 proto_unregister(&tcp_prot);
1842 fs_initcall(inet_init);
1844 /* ------------------------------------------------------------------------ */
1846 #ifdef CONFIG_PROC_FS
1847 static int __init ipv4_proc_init(void)
1851 if (raw_proc_init())
1853 if (tcp4_proc_init())
1855 if (udp4_proc_init())
1857 if (ping_proc_init())
1859 if (ip_misc_proc_init())
1876 #else /* CONFIG_PROC_FS */
1877 static int __init ipv4_proc_init(void)
1881 #endif /* CONFIG_PROC_FS */
1883 MODULE_ALIAS_NETPROTO(PF_INET);