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/ipip.h>
115 #include <net/inet_common.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
129 struct ipv4_config ipv4_config;
130 EXPORT_SYMBOL(ipv4_config);
132 /* New destruction routine */
134 void inet_sock_destruct(struct sock *sk)
136 struct inet_sock *inet = inet_sk(sk);
138 __skb_queue_purge(&sk->sk_receive_queue);
139 __skb_queue_purge(&sk->sk_error_queue);
143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 pr_err("Attempt to release TCP socket in state %d %p\n",
148 if (!sock_flag(sk, SOCK_DEAD)) {
149 pr_err("Attempt to release alive inet socket %p\n", sk);
153 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
155 WARN_ON(sk->sk_wmem_queued);
156 WARN_ON(sk->sk_forward_alloc);
158 kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
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 err = inet_csk_listen_start(sk, backlog);
218 sk->sk_max_ack_backlog = backlog;
225 EXPORT_SYMBOL(inet_listen);
227 u32 inet_ehash_secret __read_mostly;
228 EXPORT_SYMBOL(inet_ehash_secret);
231 * inet_ehash_secret must be set exactly once
233 void build_ehash_secret(void)
238 get_random_bytes(&rnd, sizeof(rnd));
241 cmpxchg(&inet_ehash_secret, 0, rnd);
243 EXPORT_SYMBOL(build_ehash_secret);
245 static inline int inet_netns_ok(struct net *net, int protocol)
248 const struct net_protocol *ipprot;
250 if (net_eq(net, &init_net))
253 hash = protocol & (MAX_INET_PROTOS - 1);
254 ipprot = rcu_dereference(inet_protos[hash]);
259 return ipprot->netns_ok;
263 * Create an inet socket.
266 static int inet_create(struct net *net, struct socket *sock, int protocol,
270 struct inet_protosw *answer;
271 struct inet_sock *inet;
272 struct proto *answer_prot;
273 unsigned char answer_flags;
274 char answer_no_check;
275 int try_loading_module = 0;
278 if (unlikely(!inet_ehash_secret))
279 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
280 build_ehash_secret();
282 sock->state = SS_UNCONNECTED;
284 /* Look for the requested type/protocol pair. */
286 err = -ESOCKTNOSUPPORT;
288 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
291 /* Check the non-wild match. */
292 if (protocol == answer->protocol) {
293 if (protocol != IPPROTO_IP)
296 /* Check for the two wild cases. */
297 if (IPPROTO_IP == protocol) {
298 protocol = answer->protocol;
301 if (IPPROTO_IP == answer->protocol)
304 err = -EPROTONOSUPPORT;
308 if (try_loading_module < 2) {
311 * Be more specific, e.g. net-pf-2-proto-132-type-1
312 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
314 if (++try_loading_module == 1)
315 request_module("net-pf-%d-proto-%d-type-%d",
316 PF_INET, protocol, sock->type);
318 * Fall back to generic, e.g. net-pf-2-proto-132
319 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
322 request_module("net-pf-%d-proto-%d",
324 goto lookup_protocol;
330 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
334 if (!inet_netns_ok(net, protocol))
337 sock->ops = answer->ops;
338 answer_prot = answer->prot;
339 answer_no_check = answer->no_check;
340 answer_flags = answer->flags;
343 WARN_ON(answer_prot->slab == NULL);
346 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
351 sk->sk_no_check = answer_no_check;
352 if (INET_PROTOSW_REUSE & answer_flags)
356 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
360 if (SOCK_RAW == sock->type) {
361 inet->inet_num = protocol;
362 if (IPPROTO_RAW == protocol)
366 if (ipv4_config.no_pmtu_disc)
367 inet->pmtudisc = IP_PMTUDISC_DONT;
369 inet->pmtudisc = IP_PMTUDISC_WANT;
373 sock_init_data(sock, sk);
375 sk->sk_destruct = inet_sock_destruct;
376 sk->sk_protocol = protocol;
377 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
384 inet->mc_list = NULL;
387 sk_refcnt_debug_inc(sk);
389 if (inet->inet_num) {
390 /* It assumes that any protocol which allows
391 * the user to assign a number at socket
392 * creation time automatically
395 inet->inet_sport = htons(inet->inet_num);
396 /* Add to protocol hash chains. */
397 sk->sk_prot->hash(sk);
400 if (sk->sk_prot->init) {
401 err = sk->sk_prot->init(sk);
403 sk_common_release(sk);
414 * The peer socket should always be NULL (or else). When we call this
415 * function we are destroying the object and from then on nobody
416 * should refer to it.
418 int inet_release(struct socket *sock)
420 struct sock *sk = sock->sk;
425 sock_rps_reset_flow(sk);
427 /* Applications forget to leave groups before exiting */
428 ip_mc_drop_socket(sk);
430 /* If linger is set, we don't return until the close
431 * is complete. Otherwise we return immediately. The
432 * actually closing is done the same either way.
434 * If the close is due to the process exiting, we never
438 if (sock_flag(sk, SOCK_LINGER) &&
439 !(current->flags & PF_EXITING))
440 timeout = sk->sk_lingertime;
442 sk->sk_prot->close(sk, timeout);
446 EXPORT_SYMBOL(inet_release);
448 /* It is off by default, see below. */
449 int sysctl_ip_nonlocal_bind __read_mostly;
450 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
452 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
454 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
455 struct sock *sk = sock->sk;
456 struct inet_sock *inet = inet_sk(sk);
461 /* If the socket has its own bind function then use it. (RAW) */
462 if (sk->sk_prot->bind) {
463 err = sk->sk_prot->bind(sk, uaddr, addr_len);
467 if (addr_len < sizeof(struct sockaddr_in))
470 if (addr->sin_family != AF_INET) {
471 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
472 * only if s_addr is INADDR_ANY.
475 if (addr->sin_family != AF_UNSPEC ||
476 addr->sin_addr.s_addr != htonl(INADDR_ANY))
480 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
482 /* Not specified by any standard per-se, however it breaks too
483 * many applications when removed. It is unfortunate since
484 * allowing applications to make a non-local bind solves
485 * several problems with systems using dynamic addressing.
486 * (ie. your servers still start up even if your ISDN link
487 * is temporarily down)
489 err = -EADDRNOTAVAIL;
490 if (!sysctl_ip_nonlocal_bind &&
491 !(inet->freebind || inet->transparent) &&
492 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
493 chk_addr_ret != RTN_LOCAL &&
494 chk_addr_ret != RTN_MULTICAST &&
495 chk_addr_ret != RTN_BROADCAST)
498 snum = ntohs(addr->sin_port);
500 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
503 /* We keep a pair of addresses. rcv_saddr is the one
504 * used by hash lookups, and saddr is used for transmit.
506 * In the BSD API these are the same except where it
507 * would be illegal to use them (multicast/broadcast) in
508 * which case the sending device address is used.
512 /* Check these errors (active socket, double bind). */
514 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
515 goto out_release_sock;
517 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
518 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
519 inet->inet_saddr = 0; /* Use device */
521 /* Make sure we are allowed to bind here. */
522 if (sk->sk_prot->get_port(sk, snum)) {
523 inet->inet_saddr = inet->inet_rcv_saddr = 0;
525 goto out_release_sock;
528 if (inet->inet_rcv_saddr)
529 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
531 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
532 inet->inet_sport = htons(inet->inet_num);
533 inet->inet_daddr = 0;
534 inet->inet_dport = 0;
542 EXPORT_SYMBOL(inet_bind);
544 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
545 int addr_len, int flags)
547 struct sock *sk = sock->sk;
549 if (addr_len < sizeof(uaddr->sa_family))
551 if (uaddr->sa_family == AF_UNSPEC)
552 return sk->sk_prot->disconnect(sk, flags);
554 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
556 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
558 EXPORT_SYMBOL(inet_dgram_connect);
560 static long inet_wait_for_connect(struct sock *sk, long timeo)
564 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
566 /* Basic assumption: if someone sets sk->sk_err, he _must_
567 * change state of the socket from TCP_SYN_*.
568 * Connect() does not allow to get error notifications
569 * without closing the socket.
571 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
573 timeo = schedule_timeout(timeo);
575 if (signal_pending(current) || !timeo)
577 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
579 finish_wait(sk_sleep(sk), &wait);
584 * Connect to a remote host. There is regrettably still a little
585 * TCP 'magic' in here.
587 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
588 int addr_len, int flags)
590 struct sock *sk = sock->sk;
594 if (addr_len < sizeof(uaddr->sa_family))
599 if (uaddr->sa_family == AF_UNSPEC) {
600 err = sk->sk_prot->disconnect(sk, flags);
601 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
605 switch (sock->state) {
614 /* Fall out of switch with err, set for this state */
618 if (sk->sk_state != TCP_CLOSE)
621 err = sk->sk_prot->connect(sk, uaddr, addr_len);
625 sock->state = SS_CONNECTING;
627 /* Just entered SS_CONNECTING state; the only
628 * difference is that return value in non-blocking
629 * case is EINPROGRESS, rather than EALREADY.
635 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
637 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
638 /* Error code is set above */
639 if (!timeo || !inet_wait_for_connect(sk, timeo))
642 err = sock_intr_errno(timeo);
643 if (signal_pending(current))
647 /* Connection was closed by RST, timeout, ICMP error
648 * or another process disconnected us.
650 if (sk->sk_state == TCP_CLOSE)
653 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
654 * and error was received after socket entered established state.
655 * Hence, it is handled normally after connect() return successfully.
658 sock->state = SS_CONNECTED;
665 err = sock_error(sk) ? : -ECONNABORTED;
666 sock->state = SS_UNCONNECTED;
667 if (sk->sk_prot->disconnect(sk, flags))
668 sock->state = SS_DISCONNECTING;
671 EXPORT_SYMBOL(inet_stream_connect);
674 * Accept a pending connection. The TCP layer now gives BSD semantics.
677 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
679 struct sock *sk1 = sock->sk;
681 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
688 sock_rps_record_flow(sk2);
689 WARN_ON(!((1 << sk2->sk_state) &
690 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
692 sock_graft(sk2, newsock);
694 newsock->state = SS_CONNECTED;
700 EXPORT_SYMBOL(inet_accept);
704 * This does both peername and sockname.
706 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
707 int *uaddr_len, int peer)
709 struct sock *sk = sock->sk;
710 struct inet_sock *inet = inet_sk(sk);
711 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
713 sin->sin_family = AF_INET;
715 if (!inet->inet_dport ||
716 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
719 sin->sin_port = inet->inet_dport;
720 sin->sin_addr.s_addr = inet->inet_daddr;
722 __be32 addr = inet->inet_rcv_saddr;
724 addr = inet->inet_saddr;
725 sin->sin_port = inet->inet_sport;
726 sin->sin_addr.s_addr = addr;
728 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
729 *uaddr_len = sizeof(*sin);
732 EXPORT_SYMBOL(inet_getname);
734 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
737 struct sock *sk = sock->sk;
739 sock_rps_record_flow(sk);
741 /* We may need to bind the socket. */
742 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
746 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
748 EXPORT_SYMBOL(inet_sendmsg);
750 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
751 size_t size, int flags)
753 struct sock *sk = sock->sk;
755 sock_rps_record_flow(sk);
757 /* We may need to bind the socket. */
758 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
762 if (sk->sk_prot->sendpage)
763 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
764 return sock_no_sendpage(sock, page, offset, size, flags);
766 EXPORT_SYMBOL(inet_sendpage);
768 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
769 size_t size, int flags)
771 struct sock *sk = sock->sk;
775 sock_rps_record_flow(sk);
777 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
778 flags & ~MSG_DONTWAIT, &addr_len);
780 msg->msg_namelen = addr_len;
783 EXPORT_SYMBOL(inet_recvmsg);
785 int inet_shutdown(struct socket *sock, int how)
787 struct sock *sk = sock->sk;
790 /* This should really check to make sure
791 * the socket is a TCP socket. (WHY AC...)
793 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
796 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
800 if (sock->state == SS_CONNECTING) {
801 if ((1 << sk->sk_state) &
802 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
803 sock->state = SS_DISCONNECTING;
805 sock->state = SS_CONNECTED;
808 switch (sk->sk_state) {
811 /* Hack to wake up other listeners, who can poll for
812 POLLHUP, even on eg. unconnected UDP sockets -- RR */
814 sk->sk_shutdown |= how;
815 if (sk->sk_prot->shutdown)
816 sk->sk_prot->shutdown(sk, how);
819 /* Remaining two branches are temporary solution for missing
820 * close() in multithreaded environment. It is _not_ a good idea,
821 * but we have no choice until close() is repaired at VFS level.
824 if (!(how & RCV_SHUTDOWN))
828 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
829 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
833 /* Wake up anyone sleeping in poll. */
834 sk->sk_state_change(sk);
838 EXPORT_SYMBOL(inet_shutdown);
841 * ioctl() calls you can issue on an INET socket. Most of these are
842 * device configuration and stuff and very rarely used. Some ioctls
843 * pass on to the socket itself.
845 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
846 * loads the devconfigure module does its configuring and unloads it.
847 * There's a good 20K of config code hanging around the kernel.
850 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
852 struct sock *sk = sock->sk;
854 struct net *net = sock_net(sk);
858 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
861 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
866 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
871 err = arp_ioctl(net, cmd, (void __user *)arg);
884 err = devinet_ioctl(net, cmd, (void __user *)arg);
887 if (sk->sk_prot->ioctl)
888 err = sk->sk_prot->ioctl(sk, cmd, arg);
895 EXPORT_SYMBOL(inet_ioctl);
898 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
900 struct sock *sk = sock->sk;
901 int err = -ENOIOCTLCMD;
903 if (sk->sk_prot->compat_ioctl)
904 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
910 const struct proto_ops inet_stream_ops = {
912 .owner = THIS_MODULE,
913 .release = inet_release,
915 .connect = inet_stream_connect,
916 .socketpair = sock_no_socketpair,
917 .accept = inet_accept,
918 .getname = inet_getname,
921 .listen = inet_listen,
922 .shutdown = inet_shutdown,
923 .setsockopt = sock_common_setsockopt,
924 .getsockopt = sock_common_getsockopt,
925 .sendmsg = inet_sendmsg,
926 .recvmsg = inet_recvmsg,
927 .mmap = sock_no_mmap,
928 .sendpage = inet_sendpage,
929 .splice_read = tcp_splice_read,
931 .compat_setsockopt = compat_sock_common_setsockopt,
932 .compat_getsockopt = compat_sock_common_getsockopt,
933 .compat_ioctl = inet_compat_ioctl,
936 EXPORT_SYMBOL(inet_stream_ops);
938 const struct proto_ops inet_dgram_ops = {
940 .owner = THIS_MODULE,
941 .release = inet_release,
943 .connect = inet_dgram_connect,
944 .socketpair = sock_no_socketpair,
945 .accept = sock_no_accept,
946 .getname = inet_getname,
949 .listen = sock_no_listen,
950 .shutdown = inet_shutdown,
951 .setsockopt = sock_common_setsockopt,
952 .getsockopt = sock_common_getsockopt,
953 .sendmsg = inet_sendmsg,
954 .recvmsg = inet_recvmsg,
955 .mmap = sock_no_mmap,
956 .sendpage = inet_sendpage,
958 .compat_setsockopt = compat_sock_common_setsockopt,
959 .compat_getsockopt = compat_sock_common_getsockopt,
960 .compat_ioctl = inet_compat_ioctl,
963 EXPORT_SYMBOL(inet_dgram_ops);
966 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
969 static const struct proto_ops inet_sockraw_ops = {
971 .owner = THIS_MODULE,
972 .release = inet_release,
974 .connect = inet_dgram_connect,
975 .socketpair = sock_no_socketpair,
976 .accept = sock_no_accept,
977 .getname = inet_getname,
978 .poll = datagram_poll,
980 .listen = sock_no_listen,
981 .shutdown = inet_shutdown,
982 .setsockopt = sock_common_setsockopt,
983 .getsockopt = sock_common_getsockopt,
984 .sendmsg = inet_sendmsg,
985 .recvmsg = inet_recvmsg,
986 .mmap = sock_no_mmap,
987 .sendpage = inet_sendpage,
989 .compat_setsockopt = compat_sock_common_setsockopt,
990 .compat_getsockopt = compat_sock_common_getsockopt,
991 .compat_ioctl = inet_compat_ioctl,
995 static const struct net_proto_family inet_family_ops = {
997 .create = inet_create,
998 .owner = THIS_MODULE,
1001 /* Upon startup we insert all the elements in inetsw_array[] into
1002 * the linked list inetsw.
1004 static struct inet_protosw inetsw_array[] =
1007 .type = SOCK_STREAM,
1008 .protocol = IPPROTO_TCP,
1010 .ops = &inet_stream_ops,
1012 .flags = INET_PROTOSW_PERMANENT |
1018 .protocol = IPPROTO_UDP,
1020 .ops = &inet_dgram_ops,
1021 .no_check = UDP_CSUM_DEFAULT,
1022 .flags = INET_PROTOSW_PERMANENT,
1027 .protocol = IPPROTO_ICMP,
1029 .ops = &inet_dgram_ops,
1030 .no_check = UDP_CSUM_DEFAULT,
1031 .flags = INET_PROTOSW_REUSE,
1036 .protocol = IPPROTO_IP, /* wild card */
1038 .ops = &inet_sockraw_ops,
1039 .no_check = UDP_CSUM_DEFAULT,
1040 .flags = INET_PROTOSW_REUSE,
1044 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1046 void inet_register_protosw(struct inet_protosw *p)
1048 struct list_head *lh;
1049 struct inet_protosw *answer;
1050 int protocol = p->protocol;
1051 struct list_head *last_perm;
1053 spin_lock_bh(&inetsw_lock);
1055 if (p->type >= SOCK_MAX)
1058 /* 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);
1064 /* Check only the non-wild match. */
1065 if (INET_PROTOSW_PERMANENT & answer->flags) {
1066 if (protocol == answer->protocol)
1076 /* Add the new entry after the last permanent entry if any, so that
1077 * the new entry does not override a permanent entry when matched with
1078 * a wild-card protocol. But it is allowed to override any existing
1079 * non-permanent entry. This means that when we remove this entry, the
1080 * system automatically returns to the old behavior.
1082 list_add_rcu(&p->list, last_perm);
1084 spin_unlock_bh(&inetsw_lock);
1089 pr_err("Attempt to override permanent protocol %d\n", protocol);
1093 pr_err("Ignoring attempt to register invalid socket type %d\n",
1097 EXPORT_SYMBOL(inet_register_protosw);
1099 void inet_unregister_protosw(struct inet_protosw *p)
1101 if (INET_PROTOSW_PERMANENT & p->flags) {
1102 pr_err("Attempt to unregister permanent protocol %d\n",
1105 spin_lock_bh(&inetsw_lock);
1106 list_del_rcu(&p->list);
1107 spin_unlock_bh(&inetsw_lock);
1112 EXPORT_SYMBOL(inet_unregister_protosw);
1115 * Shall we try to damage output packets if routing dev changes?
1118 int sysctl_ip_dynaddr __read_mostly;
1120 static int inet_sk_reselect_saddr(struct sock *sk)
1122 struct inet_sock *inet = inet_sk(sk);
1123 __be32 old_saddr = inet->inet_saddr;
1124 __be32 daddr = inet->inet_daddr;
1128 struct ip_options_rcu *inet_opt;
1130 inet_opt = rcu_dereference_protected(inet->inet_opt,
1131 sock_owned_by_user(sk));
1132 if (inet_opt && inet_opt->opt.srr)
1133 daddr = inet_opt->opt.faddr;
1135 /* Query new route. */
1136 fl4 = &inet->cork.fl.u.ip4;
1137 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1138 sk->sk_bound_dev_if, sk->sk_protocol,
1139 inet->inet_sport, inet->inet_dport, sk, false);
1143 sk_setup_caps(sk, &rt->dst);
1145 new_saddr = fl4->saddr;
1147 if (new_saddr == old_saddr)
1150 if (sysctl_ip_dynaddr > 1) {
1151 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1152 __func__, &old_saddr, &new_saddr);
1155 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1158 * XXX The only one ugly spot where we need to
1159 * XXX really change the sockets identity after
1160 * XXX it has entered the hashes. -DaveM
1162 * Besides that, it does not check for connection
1163 * uniqueness. Wait for troubles.
1165 __sk_prot_rehash(sk);
1169 int inet_sk_rebuild_header(struct sock *sk)
1171 struct inet_sock *inet = inet_sk(sk);
1172 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1174 struct ip_options_rcu *inet_opt;
1178 /* Route is OK, nothing to do. */
1184 inet_opt = rcu_dereference(inet->inet_opt);
1185 daddr = inet->inet_daddr;
1186 if (inet_opt && inet_opt->opt.srr)
1187 daddr = inet_opt->opt.faddr;
1189 fl4 = &inet->cork.fl.u.ip4;
1190 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1191 inet->inet_dport, inet->inet_sport,
1192 sk->sk_protocol, RT_CONN_FLAGS(sk),
1193 sk->sk_bound_dev_if);
1196 sk_setup_caps(sk, &rt->dst);
1200 /* Routing failed... */
1201 sk->sk_route_caps = 0;
1203 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1204 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1206 if (!sysctl_ip_dynaddr ||
1207 sk->sk_state != TCP_SYN_SENT ||
1208 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1209 (err = inet_sk_reselect_saddr(sk)) != 0)
1210 sk->sk_err_soft = -err;
1215 EXPORT_SYMBOL(inet_sk_rebuild_header);
1217 static int inet_gso_send_check(struct sk_buff *skb)
1219 const struct iphdr *iph;
1220 const struct net_protocol *ops;
1225 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1230 if (ihl < sizeof(*iph))
1233 if (unlikely(!pskb_may_pull(skb, ihl)))
1236 __skb_pull(skb, ihl);
1237 skb_reset_transport_header(skb);
1239 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1240 err = -EPROTONOSUPPORT;
1243 ops = rcu_dereference(inet_protos[proto]);
1244 if (likely(ops && ops->gso_send_check))
1245 err = ops->gso_send_check(skb);
1252 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1253 netdev_features_t features)
1255 struct sk_buff *segs = ERR_PTR(-EINVAL);
1257 const struct net_protocol *ops;
1261 unsigned int offset = 0;
1263 if (!(features & NETIF_F_V4_CSUM))
1264 features &= ~NETIF_F_SG;
1266 if (unlikely(skb_shinfo(skb)->gso_type &
1274 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1279 if (ihl < sizeof(*iph))
1282 if (unlikely(!pskb_may_pull(skb, ihl)))
1285 __skb_pull(skb, ihl);
1286 skb_reset_transport_header(skb);
1288 id = ntohs(iph->id);
1289 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1290 segs = ERR_PTR(-EPROTONOSUPPORT);
1293 ops = rcu_dereference(inet_protos[proto]);
1294 if (likely(ops && ops->gso_segment))
1295 segs = ops->gso_segment(skb, features);
1298 if (!segs || IS_ERR(segs))
1304 if (proto == IPPROTO_UDP) {
1305 iph->id = htons(id);
1306 iph->frag_off = htons(offset >> 3);
1307 if (skb->next != NULL)
1308 iph->frag_off |= htons(IP_MF);
1309 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1311 iph->id = htons(id++);
1312 iph->tot_len = htons(skb->len - skb->mac_len);
1314 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1315 } while ((skb = skb->next));
1321 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1322 struct sk_buff *skb)
1324 const struct net_protocol *ops;
1325 struct sk_buff **pp = NULL;
1327 const struct iphdr *iph;
1334 off = skb_gro_offset(skb);
1335 hlen = off + sizeof(*iph);
1336 iph = skb_gro_header_fast(skb, off);
1337 if (skb_gro_header_hard(skb, hlen)) {
1338 iph = skb_gro_header_slow(skb, hlen, off);
1343 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1346 ops = rcu_dereference(inet_protos[proto]);
1347 if (!ops || !ops->gro_receive)
1350 if (*(u8 *)iph != 0x45)
1353 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1356 id = ntohl(*(__be32 *)&iph->id);
1357 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1360 for (p = *head; p; p = p->next) {
1363 if (!NAPI_GRO_CB(p)->same_flow)
1368 if ((iph->protocol ^ iph2->protocol) |
1369 (iph->tos ^ iph2->tos) |
1370 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1371 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1372 NAPI_GRO_CB(p)->same_flow = 0;
1376 /* All fields must match except length and checksum. */
1377 NAPI_GRO_CB(p)->flush |=
1378 (iph->ttl ^ iph2->ttl) |
1379 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1381 NAPI_GRO_CB(p)->flush |= flush;
1384 NAPI_GRO_CB(skb)->flush |= flush;
1385 skb_gro_pull(skb, sizeof(*iph));
1386 skb_set_transport_header(skb, skb_gro_offset(skb));
1388 pp = ops->gro_receive(head, skb);
1394 NAPI_GRO_CB(skb)->flush |= flush;
1399 static int inet_gro_complete(struct sk_buff *skb)
1401 const struct net_protocol *ops;
1402 struct iphdr *iph = ip_hdr(skb);
1403 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1405 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1407 csum_replace2(&iph->check, iph->tot_len, newlen);
1408 iph->tot_len = newlen;
1411 ops = rcu_dereference(inet_protos[proto]);
1412 if (WARN_ON(!ops || !ops->gro_complete))
1415 err = ops->gro_complete(skb);
1423 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1424 unsigned short type, unsigned char protocol,
1427 struct socket *sock;
1428 int rc = sock_create_kern(family, type, protocol, &sock);
1432 (*sk)->sk_allocation = GFP_ATOMIC;
1434 * Unhash it so that IP input processing does not even see it,
1435 * we do not wish this socket to see incoming packets.
1437 (*sk)->sk_prot->unhash(*sk);
1439 sk_change_net(*sk, net);
1443 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1445 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1447 unsigned long res = 0;
1450 for_each_possible_cpu(i) {
1451 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1452 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1456 EXPORT_SYMBOL_GPL(snmp_fold_field);
1458 #if BITS_PER_LONG==32
1460 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1465 for_each_possible_cpu(cpu) {
1467 struct u64_stats_sync *syncp;
1471 bhptr = per_cpu_ptr(mib[0], cpu);
1472 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1474 start = u64_stats_fetch_begin_bh(syncp);
1475 v = *(((u64 *) bhptr) + offt);
1476 } while (u64_stats_fetch_retry_bh(syncp, start));
1482 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1485 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1487 BUG_ON(ptr == NULL);
1488 ptr[0] = __alloc_percpu(mibsize, align);
1491 #if SNMP_ARRAY_SZ == 2
1492 ptr[1] = __alloc_percpu(mibsize, align);
1494 free_percpu(ptr[0]);
1501 EXPORT_SYMBOL_GPL(snmp_mib_init);
1503 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1507 BUG_ON(ptr == NULL);
1508 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1509 free_percpu(ptr[i]);
1513 EXPORT_SYMBOL_GPL(snmp_mib_free);
1515 #ifdef CONFIG_IP_MULTICAST
1516 static const struct net_protocol igmp_protocol = {
1517 .handler = igmp_rcv,
1522 static const struct net_protocol tcp_protocol = {
1523 .handler = tcp_v4_rcv,
1524 .err_handler = tcp_v4_err,
1525 .gso_send_check = tcp_v4_gso_send_check,
1526 .gso_segment = tcp_tso_segment,
1527 .gro_receive = tcp4_gro_receive,
1528 .gro_complete = tcp4_gro_complete,
1533 static const struct net_protocol udp_protocol = {
1535 .err_handler = udp_err,
1536 .gso_send_check = udp4_ufo_send_check,
1537 .gso_segment = udp4_ufo_fragment,
1542 static const struct net_protocol icmp_protocol = {
1543 .handler = icmp_rcv,
1544 .err_handler = ping_err,
1549 static __net_init int ipv4_mib_init_net(struct net *net)
1551 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1552 sizeof(struct tcp_mib),
1553 __alignof__(struct tcp_mib)) < 0)
1555 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1556 sizeof(struct ipstats_mib),
1557 __alignof__(struct ipstats_mib)) < 0)
1559 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1560 sizeof(struct linux_mib),
1561 __alignof__(struct linux_mib)) < 0)
1563 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1564 sizeof(struct udp_mib),
1565 __alignof__(struct udp_mib)) < 0)
1567 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1568 sizeof(struct udp_mib),
1569 __alignof__(struct udp_mib)) < 0)
1570 goto err_udplite_mib;
1571 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1572 sizeof(struct icmp_mib),
1573 __alignof__(struct icmp_mib)) < 0)
1575 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1577 if (!net->mib.icmpmsg_statistics)
1578 goto err_icmpmsg_mib;
1584 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1586 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1588 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1590 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1592 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1594 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1599 static __net_exit void ipv4_mib_exit_net(struct net *net)
1601 kfree(net->mib.icmpmsg_statistics);
1602 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1603 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1604 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1605 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1606 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1607 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1610 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1611 .init = ipv4_mib_init_net,
1612 .exit = ipv4_mib_exit_net,
1615 static int __init init_ipv4_mibs(void)
1617 return register_pernet_subsys(&ipv4_mib_ops);
1620 static int ipv4_proc_init(void);
1623 * IP protocol layer initialiser
1626 static struct packet_type ip_packet_type __read_mostly = {
1627 .type = cpu_to_be16(ETH_P_IP),
1629 .gso_send_check = inet_gso_send_check,
1630 .gso_segment = inet_gso_segment,
1631 .gro_receive = inet_gro_receive,
1632 .gro_complete = inet_gro_complete,
1635 static int __init inet_init(void)
1637 struct sk_buff *dummy_skb;
1638 struct inet_protosw *q;
1639 struct list_head *r;
1642 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1644 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1645 if (!sysctl_local_reserved_ports)
1648 rc = proto_register(&tcp_prot, 1);
1650 goto out_free_reserved_ports;
1652 rc = proto_register(&udp_prot, 1);
1654 goto out_unregister_tcp_proto;
1656 rc = proto_register(&raw_prot, 1);
1658 goto out_unregister_udp_proto;
1660 rc = proto_register(&ping_prot, 1);
1662 goto out_unregister_raw_proto;
1665 * Tell SOCKET that we are alive...
1668 (void)sock_register(&inet_family_ops);
1670 #ifdef CONFIG_SYSCTL
1671 ip_static_sysctl_init();
1674 tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1677 * Add all the base protocols.
1680 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1681 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1682 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1683 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1684 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1685 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1686 #ifdef CONFIG_IP_MULTICAST
1687 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1688 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1691 /* Register the socket-side information for inet_create. */
1692 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1695 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1696 inet_register_protosw(q);
1699 * Set the ARP module up
1705 * Set the IP module up
1712 /* Setup TCP slab cache for open requests. */
1715 /* Setup UDP memory threshold */
1718 /* Add UDP-Lite (RFC 3828) */
1719 udplite4_register();
1724 * Set the ICMP layer up
1727 if (icmp_init() < 0)
1728 panic("Failed to create the ICMP control socket.\n");
1731 * Initialise the multicast router
1733 #if defined(CONFIG_IP_MROUTE)
1735 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1738 * Initialise per-cpu ipv4 mibs
1741 if (init_ipv4_mibs())
1742 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1748 dev_add_pack(&ip_packet_type);
1753 out_unregister_raw_proto:
1754 proto_unregister(&raw_prot);
1755 out_unregister_udp_proto:
1756 proto_unregister(&udp_prot);
1757 out_unregister_tcp_proto:
1758 proto_unregister(&tcp_prot);
1759 out_free_reserved_ports:
1760 kfree(sysctl_local_reserved_ports);
1764 fs_initcall(inet_init);
1766 /* ------------------------------------------------------------------------ */
1768 #ifdef CONFIG_PROC_FS
1769 static int __init ipv4_proc_init(void)
1773 if (raw_proc_init())
1775 if (tcp4_proc_init())
1777 if (udp4_proc_init())
1779 if (ping_proc_init())
1781 if (ip_misc_proc_init())
1798 #else /* CONFIG_PROC_FS */
1799 static int __init ipv4_proc_init(void)
1803 #endif /* CONFIG_PROC_FS */
1805 MODULE_ALIAS_NETPROTO(PF_INET);