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 dst_release(sk->sk_rx_dst);
161 sk_refcnt_debug_dec(sk);
163 EXPORT_SYMBOL(inet_sock_destruct);
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
172 * Automatically bind an unbound socket.
175 static int inet_autobind(struct sock *sk)
177 struct inet_sock *inet;
178 /* We may need to bind the socket. */
181 if (!inet->inet_num) {
182 if (sk->sk_prot->get_port(sk, 0)) {
186 inet->inet_sport = htons(inet->inet_num);
193 * Move a socket into listening state.
195 int inet_listen(struct socket *sock, int backlog)
197 struct sock *sk = sock->sk;
198 unsigned char old_state;
204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
207 old_state = sk->sk_state;
208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
211 /* Really, if the socket is already in listen state
212 * we can only allow the backlog to be adjusted.
214 if (old_state != TCP_LISTEN) {
215 err = inet_csk_listen_start(sk, backlog);
219 sk->sk_max_ack_backlog = backlog;
226 EXPORT_SYMBOL(inet_listen);
228 u32 inet_ehash_secret __read_mostly;
229 EXPORT_SYMBOL(inet_ehash_secret);
232 * inet_ehash_secret must be set exactly once
234 void build_ehash_secret(void)
239 get_random_bytes(&rnd, sizeof(rnd));
242 cmpxchg(&inet_ehash_secret, 0, rnd);
244 EXPORT_SYMBOL(build_ehash_secret);
246 static inline int inet_netns_ok(struct net *net, __u8 protocol)
248 const struct net_protocol *ipprot;
250 if (net_eq(net, &init_net))
253 ipprot = rcu_dereference(inet_protos[protocol]);
254 if (ipprot == NULL) {
258 return ipprot->netns_ok;
262 * Create an inet socket.
265 static int inet_create(struct net *net, struct socket *sock, int protocol,
269 struct inet_protosw *answer;
270 struct inet_sock *inet;
271 struct proto *answer_prot;
272 unsigned char answer_flags;
273 char answer_no_check;
274 int try_loading_module = 0;
277 if (unlikely(!inet_ehash_secret))
278 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
279 build_ehash_secret();
281 sock->state = SS_UNCONNECTED;
283 /* Look for the requested type/protocol pair. */
285 err = -ESOCKTNOSUPPORT;
287 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
290 /* Check the non-wild match. */
291 if (protocol == answer->protocol) {
292 if (protocol != IPPROTO_IP)
295 /* Check for the two wild cases. */
296 if (IPPROTO_IP == protocol) {
297 protocol = answer->protocol;
300 if (IPPROTO_IP == answer->protocol)
303 err = -EPROTONOSUPPORT;
307 if (try_loading_module < 2) {
310 * Be more specific, e.g. net-pf-2-proto-132-type-1
311 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
313 if (++try_loading_module == 1)
314 request_module("net-pf-%d-proto-%d-type-%d",
315 PF_INET, protocol, sock->type);
317 * Fall back to generic, e.g. net-pf-2-proto-132
318 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
321 request_module("net-pf-%d-proto-%d",
323 goto lookup_protocol;
329 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
333 if (!inet_netns_ok(net, protocol))
336 sock->ops = answer->ops;
337 answer_prot = answer->prot;
338 answer_no_check = answer->no_check;
339 answer_flags = answer->flags;
342 WARN_ON(answer_prot->slab == NULL);
345 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
350 sk->sk_no_check = answer_no_check;
351 if (INET_PROTOSW_REUSE & answer_flags)
352 sk->sk_reuse = SK_CAN_REUSE;
355 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
359 if (SOCK_RAW == sock->type) {
360 inet->inet_num = protocol;
361 if (IPPROTO_RAW == protocol)
365 if (ipv4_config.no_pmtu_disc)
366 inet->pmtudisc = IP_PMTUDISC_DONT;
368 inet->pmtudisc = IP_PMTUDISC_WANT;
372 sock_init_data(sock, sk);
374 sk->sk_destruct = inet_sock_destruct;
375 sk->sk_protocol = protocol;
376 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
383 inet->mc_list = NULL;
386 sk_refcnt_debug_inc(sk);
388 if (inet->inet_num) {
389 /* It assumes that any protocol which allows
390 * the user to assign a number at socket
391 * creation time automatically
394 inet->inet_sport = htons(inet->inet_num);
395 /* Add to protocol hash chains. */
396 sk->sk_prot->hash(sk);
399 if (sk->sk_prot->init) {
400 err = sk->sk_prot->init(sk);
402 sk_common_release(sk);
413 * The peer socket should always be NULL (or else). When we call this
414 * function we are destroying the object and from then on nobody
415 * should refer to it.
417 int inet_release(struct socket *sock)
419 struct sock *sk = sock->sk;
424 sock_rps_reset_flow(sk);
426 /* Applications forget to leave groups before exiting */
427 ip_mc_drop_socket(sk);
429 /* If linger is set, we don't return until the close
430 * is complete. Otherwise we return immediately. The
431 * actually closing is done the same either way.
433 * If the close is due to the process exiting, we never
437 if (sock_flag(sk, SOCK_LINGER) &&
438 !(current->flags & PF_EXITING))
439 timeout = sk->sk_lingertime;
441 sk->sk_prot->close(sk, timeout);
445 EXPORT_SYMBOL(inet_release);
447 /* It is off by default, see below. */
448 int sysctl_ip_nonlocal_bind __read_mostly;
449 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
451 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
453 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
454 struct sock *sk = sock->sk;
455 struct inet_sock *inet = inet_sk(sk);
460 /* If the socket has its own bind function then use it. (RAW) */
461 if (sk->sk_prot->bind) {
462 err = sk->sk_prot->bind(sk, uaddr, addr_len);
466 if (addr_len < sizeof(struct sockaddr_in))
469 if (addr->sin_family != AF_INET) {
470 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
471 * only if s_addr is INADDR_ANY.
474 if (addr->sin_family != AF_UNSPEC ||
475 addr->sin_addr.s_addr != htonl(INADDR_ANY))
479 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
481 /* Not specified by any standard per-se, however it breaks too
482 * many applications when removed. It is unfortunate since
483 * allowing applications to make a non-local bind solves
484 * several problems with systems using dynamic addressing.
485 * (ie. your servers still start up even if your ISDN link
486 * is temporarily down)
488 err = -EADDRNOTAVAIL;
489 if (!sysctl_ip_nonlocal_bind &&
490 !(inet->freebind || inet->transparent) &&
491 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
492 chk_addr_ret != RTN_LOCAL &&
493 chk_addr_ret != RTN_MULTICAST &&
494 chk_addr_ret != RTN_BROADCAST)
497 snum = ntohs(addr->sin_port);
499 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
502 /* We keep a pair of addresses. rcv_saddr is the one
503 * used by hash lookups, and saddr is used for transmit.
505 * In the BSD API these are the same except where it
506 * would be illegal to use them (multicast/broadcast) in
507 * which case the sending device address is used.
511 /* Check these errors (active socket, double bind). */
513 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
514 goto out_release_sock;
516 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
517 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
518 inet->inet_saddr = 0; /* Use device */
520 /* Make sure we are allowed to bind here. */
521 if (sk->sk_prot->get_port(sk, snum)) {
522 inet->inet_saddr = inet->inet_rcv_saddr = 0;
524 goto out_release_sock;
527 if (inet->inet_rcv_saddr)
528 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
530 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
531 inet->inet_sport = htons(inet->inet_num);
532 inet->inet_daddr = 0;
533 inet->inet_dport = 0;
541 EXPORT_SYMBOL(inet_bind);
543 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
544 int addr_len, int flags)
546 struct sock *sk = sock->sk;
548 if (addr_len < sizeof(uaddr->sa_family))
550 if (uaddr->sa_family == AF_UNSPEC)
551 return sk->sk_prot->disconnect(sk, flags);
553 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
555 return sk->sk_prot->connect(sk, uaddr, addr_len);
557 EXPORT_SYMBOL(inet_dgram_connect);
559 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
563 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
564 sk->sk_write_pending += writebias;
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);
580 sk->sk_write_pending -= writebias;
585 * Connect to a remote host. There is regrettably still a little
586 * TCP 'magic' in here.
588 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
589 int addr_len, int flags)
591 struct sock *sk = sock->sk;
595 if (addr_len < sizeof(uaddr->sa_family))
598 if (uaddr->sa_family == AF_UNSPEC) {
599 err = sk->sk_prot->disconnect(sk, flags);
600 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
604 switch (sock->state) {
613 /* Fall out of switch with err, set for this state */
617 if (sk->sk_state != TCP_CLOSE)
620 err = sk->sk_prot->connect(sk, uaddr, addr_len);
624 sock->state = SS_CONNECTING;
626 /* Just entered SS_CONNECTING state; the only
627 * difference is that return value in non-blocking
628 * case is EINPROGRESS, rather than EALREADY.
634 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
636 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
637 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
638 tcp_sk(sk)->fastopen_req &&
639 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
641 /* Error code is set above */
642 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
645 err = sock_intr_errno(timeo);
646 if (signal_pending(current))
650 /* Connection was closed by RST, timeout, ICMP error
651 * or another process disconnected us.
653 if (sk->sk_state == TCP_CLOSE)
656 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
657 * and error was received after socket entered established state.
658 * Hence, it is handled normally after connect() return successfully.
661 sock->state = SS_CONNECTED;
667 err = sock_error(sk) ? : -ECONNABORTED;
668 sock->state = SS_UNCONNECTED;
669 if (sk->sk_prot->disconnect(sk, flags))
670 sock->state = SS_DISCONNECTING;
673 EXPORT_SYMBOL(__inet_stream_connect);
675 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
676 int addr_len, int flags)
681 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
682 release_sock(sock->sk);
685 EXPORT_SYMBOL(inet_stream_connect);
688 * Accept a pending connection. The TCP layer now gives BSD semantics.
691 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
693 struct sock *sk1 = sock->sk;
695 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
702 sock_rps_record_flow(sk2);
703 WARN_ON(!((1 << sk2->sk_state) &
704 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
706 sock_graft(sk2, newsock);
708 newsock->state = SS_CONNECTED;
714 EXPORT_SYMBOL(inet_accept);
718 * This does both peername and sockname.
720 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
721 int *uaddr_len, int peer)
723 struct sock *sk = sock->sk;
724 struct inet_sock *inet = inet_sk(sk);
725 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
727 sin->sin_family = AF_INET;
729 if (!inet->inet_dport ||
730 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
733 sin->sin_port = inet->inet_dport;
734 sin->sin_addr.s_addr = inet->inet_daddr;
736 __be32 addr = inet->inet_rcv_saddr;
738 addr = inet->inet_saddr;
739 sin->sin_port = inet->inet_sport;
740 sin->sin_addr.s_addr = addr;
742 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
743 *uaddr_len = sizeof(*sin);
746 EXPORT_SYMBOL(inet_getname);
748 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
751 struct sock *sk = sock->sk;
753 sock_rps_record_flow(sk);
755 /* We may need to bind the socket. */
756 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
760 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
762 EXPORT_SYMBOL(inet_sendmsg);
764 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
765 size_t size, int flags)
767 struct sock *sk = sock->sk;
769 sock_rps_record_flow(sk);
771 /* We may need to bind the socket. */
772 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
776 if (sk->sk_prot->sendpage)
777 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
778 return sock_no_sendpage(sock, page, offset, size, flags);
780 EXPORT_SYMBOL(inet_sendpage);
782 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
783 size_t size, int flags)
785 struct sock *sk = sock->sk;
789 sock_rps_record_flow(sk);
791 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
792 flags & ~MSG_DONTWAIT, &addr_len);
794 msg->msg_namelen = addr_len;
797 EXPORT_SYMBOL(inet_recvmsg);
799 int inet_shutdown(struct socket *sock, int how)
801 struct sock *sk = sock->sk;
804 /* This should really check to make sure
805 * the socket is a TCP socket. (WHY AC...)
807 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
810 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
814 if (sock->state == SS_CONNECTING) {
815 if ((1 << sk->sk_state) &
816 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
817 sock->state = SS_DISCONNECTING;
819 sock->state = SS_CONNECTED;
822 switch (sk->sk_state) {
825 /* Hack to wake up other listeners, who can poll for
826 POLLHUP, even on eg. unconnected UDP sockets -- RR */
828 sk->sk_shutdown |= how;
829 if (sk->sk_prot->shutdown)
830 sk->sk_prot->shutdown(sk, how);
833 /* Remaining two branches are temporary solution for missing
834 * close() in multithreaded environment. It is _not_ a good idea,
835 * but we have no choice until close() is repaired at VFS level.
838 if (!(how & RCV_SHUTDOWN))
842 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
843 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
847 /* Wake up anyone sleeping in poll. */
848 sk->sk_state_change(sk);
852 EXPORT_SYMBOL(inet_shutdown);
855 * ioctl() calls you can issue on an INET socket. Most of these are
856 * device configuration and stuff and very rarely used. Some ioctls
857 * pass on to the socket itself.
859 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
860 * loads the devconfigure module does its configuring and unloads it.
861 * There's a good 20K of config code hanging around the kernel.
864 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
866 struct sock *sk = sock->sk;
868 struct net *net = sock_net(sk);
872 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
875 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
880 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
885 err = arp_ioctl(net, cmd, (void __user *)arg);
898 err = devinet_ioctl(net, cmd, (void __user *)arg);
901 if (sk->sk_prot->ioctl)
902 err = sk->sk_prot->ioctl(sk, cmd, arg);
909 EXPORT_SYMBOL(inet_ioctl);
912 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
914 struct sock *sk = sock->sk;
915 int err = -ENOIOCTLCMD;
917 if (sk->sk_prot->compat_ioctl)
918 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
924 const struct proto_ops inet_stream_ops = {
926 .owner = THIS_MODULE,
927 .release = inet_release,
929 .connect = inet_stream_connect,
930 .socketpair = sock_no_socketpair,
931 .accept = inet_accept,
932 .getname = inet_getname,
935 .listen = inet_listen,
936 .shutdown = inet_shutdown,
937 .setsockopt = sock_common_setsockopt,
938 .getsockopt = sock_common_getsockopt,
939 .sendmsg = inet_sendmsg,
940 .recvmsg = inet_recvmsg,
941 .mmap = sock_no_mmap,
942 .sendpage = inet_sendpage,
943 .splice_read = tcp_splice_read,
945 .compat_setsockopt = compat_sock_common_setsockopt,
946 .compat_getsockopt = compat_sock_common_getsockopt,
947 .compat_ioctl = inet_compat_ioctl,
950 EXPORT_SYMBOL(inet_stream_ops);
952 const struct proto_ops inet_dgram_ops = {
954 .owner = THIS_MODULE,
955 .release = inet_release,
957 .connect = inet_dgram_connect,
958 .socketpair = sock_no_socketpair,
959 .accept = sock_no_accept,
960 .getname = inet_getname,
963 .listen = sock_no_listen,
964 .shutdown = inet_shutdown,
965 .setsockopt = sock_common_setsockopt,
966 .getsockopt = sock_common_getsockopt,
967 .sendmsg = inet_sendmsg,
968 .recvmsg = inet_recvmsg,
969 .mmap = sock_no_mmap,
970 .sendpage = inet_sendpage,
972 .compat_setsockopt = compat_sock_common_setsockopt,
973 .compat_getsockopt = compat_sock_common_getsockopt,
974 .compat_ioctl = inet_compat_ioctl,
977 EXPORT_SYMBOL(inet_dgram_ops);
980 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
983 static const struct proto_ops inet_sockraw_ops = {
985 .owner = THIS_MODULE,
986 .release = inet_release,
988 .connect = inet_dgram_connect,
989 .socketpair = sock_no_socketpair,
990 .accept = sock_no_accept,
991 .getname = inet_getname,
992 .poll = datagram_poll,
994 .listen = sock_no_listen,
995 .shutdown = inet_shutdown,
996 .setsockopt = sock_common_setsockopt,
997 .getsockopt = sock_common_getsockopt,
998 .sendmsg = inet_sendmsg,
999 .recvmsg = inet_recvmsg,
1000 .mmap = sock_no_mmap,
1001 .sendpage = inet_sendpage,
1002 #ifdef CONFIG_COMPAT
1003 .compat_setsockopt = compat_sock_common_setsockopt,
1004 .compat_getsockopt = compat_sock_common_getsockopt,
1005 .compat_ioctl = inet_compat_ioctl,
1009 static const struct net_proto_family inet_family_ops = {
1011 .create = inet_create,
1012 .owner = THIS_MODULE,
1015 /* Upon startup we insert all the elements in inetsw_array[] into
1016 * the linked list inetsw.
1018 static struct inet_protosw inetsw_array[] =
1021 .type = SOCK_STREAM,
1022 .protocol = IPPROTO_TCP,
1024 .ops = &inet_stream_ops,
1026 .flags = INET_PROTOSW_PERMANENT |
1032 .protocol = IPPROTO_UDP,
1034 .ops = &inet_dgram_ops,
1035 .no_check = UDP_CSUM_DEFAULT,
1036 .flags = INET_PROTOSW_PERMANENT,
1041 .protocol = IPPROTO_ICMP,
1043 .ops = &inet_dgram_ops,
1044 .no_check = UDP_CSUM_DEFAULT,
1045 .flags = INET_PROTOSW_REUSE,
1050 .protocol = IPPROTO_IP, /* wild card */
1052 .ops = &inet_sockraw_ops,
1053 .no_check = UDP_CSUM_DEFAULT,
1054 .flags = INET_PROTOSW_REUSE,
1058 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1060 void inet_register_protosw(struct inet_protosw *p)
1062 struct list_head *lh;
1063 struct inet_protosw *answer;
1064 int protocol = p->protocol;
1065 struct list_head *last_perm;
1067 spin_lock_bh(&inetsw_lock);
1069 if (p->type >= SOCK_MAX)
1072 /* If we are trying to override a permanent protocol, bail. */
1074 last_perm = &inetsw[p->type];
1075 list_for_each(lh, &inetsw[p->type]) {
1076 answer = list_entry(lh, struct inet_protosw, list);
1078 /* Check only the non-wild match. */
1079 if (INET_PROTOSW_PERMANENT & answer->flags) {
1080 if (protocol == answer->protocol)
1090 /* Add the new entry after the last permanent entry if any, so that
1091 * the new entry does not override a permanent entry when matched with
1092 * a wild-card protocol. But it is allowed to override any existing
1093 * non-permanent entry. This means that when we remove this entry, the
1094 * system automatically returns to the old behavior.
1096 list_add_rcu(&p->list, last_perm);
1098 spin_unlock_bh(&inetsw_lock);
1103 pr_err("Attempt to override permanent protocol %d\n", protocol);
1107 pr_err("Ignoring attempt to register invalid socket type %d\n",
1111 EXPORT_SYMBOL(inet_register_protosw);
1113 void inet_unregister_protosw(struct inet_protosw *p)
1115 if (INET_PROTOSW_PERMANENT & p->flags) {
1116 pr_err("Attempt to unregister permanent protocol %d\n",
1119 spin_lock_bh(&inetsw_lock);
1120 list_del_rcu(&p->list);
1121 spin_unlock_bh(&inetsw_lock);
1126 EXPORT_SYMBOL(inet_unregister_protosw);
1129 * Shall we try to damage output packets if routing dev changes?
1132 int sysctl_ip_dynaddr __read_mostly;
1134 static int inet_sk_reselect_saddr(struct sock *sk)
1136 struct inet_sock *inet = inet_sk(sk);
1137 __be32 old_saddr = inet->inet_saddr;
1138 __be32 daddr = inet->inet_daddr;
1142 struct ip_options_rcu *inet_opt;
1144 inet_opt = rcu_dereference_protected(inet->inet_opt,
1145 sock_owned_by_user(sk));
1146 if (inet_opt && inet_opt->opt.srr)
1147 daddr = inet_opt->opt.faddr;
1149 /* Query new route. */
1150 fl4 = &inet->cork.fl.u.ip4;
1151 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1152 sk->sk_bound_dev_if, sk->sk_protocol,
1153 inet->inet_sport, inet->inet_dport, sk, false);
1157 sk_setup_caps(sk, &rt->dst);
1159 new_saddr = fl4->saddr;
1161 if (new_saddr == old_saddr)
1164 if (sysctl_ip_dynaddr > 1) {
1165 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1166 __func__, &old_saddr, &new_saddr);
1169 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1172 * XXX The only one ugly spot where we need to
1173 * XXX really change the sockets identity after
1174 * XXX it has entered the hashes. -DaveM
1176 * Besides that, it does not check for connection
1177 * uniqueness. Wait for troubles.
1179 __sk_prot_rehash(sk);
1183 int inet_sk_rebuild_header(struct sock *sk)
1185 struct inet_sock *inet = inet_sk(sk);
1186 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1188 struct ip_options_rcu *inet_opt;
1192 /* Route is OK, nothing to do. */
1198 inet_opt = rcu_dereference(inet->inet_opt);
1199 daddr = inet->inet_daddr;
1200 if (inet_opt && inet_opt->opt.srr)
1201 daddr = inet_opt->opt.faddr;
1203 fl4 = &inet->cork.fl.u.ip4;
1204 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1205 inet->inet_dport, inet->inet_sport,
1206 sk->sk_protocol, RT_CONN_FLAGS(sk),
1207 sk->sk_bound_dev_if);
1210 sk_setup_caps(sk, &rt->dst);
1214 /* Routing failed... */
1215 sk->sk_route_caps = 0;
1217 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1218 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1220 if (!sysctl_ip_dynaddr ||
1221 sk->sk_state != TCP_SYN_SENT ||
1222 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1223 (err = inet_sk_reselect_saddr(sk)) != 0)
1224 sk->sk_err_soft = -err;
1229 EXPORT_SYMBOL(inet_sk_rebuild_header);
1231 static int inet_gso_send_check(struct sk_buff *skb)
1233 const struct net_protocol *ops;
1234 const struct iphdr *iph;
1239 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1244 if (ihl < sizeof(*iph))
1247 if (unlikely(!pskb_may_pull(skb, ihl)))
1250 __skb_pull(skb, ihl);
1251 skb_reset_transport_header(skb);
1253 proto = iph->protocol;
1254 err = -EPROTONOSUPPORT;
1257 ops = rcu_dereference(inet_protos[proto]);
1258 if (likely(ops && ops->gso_send_check))
1259 err = ops->gso_send_check(skb);
1266 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1267 netdev_features_t features)
1269 struct sk_buff *segs = ERR_PTR(-EINVAL);
1270 const struct net_protocol *ops;
1275 unsigned int offset = 0;
1277 if (!(features & NETIF_F_V4_CSUM))
1278 features &= ~NETIF_F_SG;
1280 if (unlikely(skb_shinfo(skb)->gso_type &
1288 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1293 if (ihl < sizeof(*iph))
1296 if (unlikely(!pskb_may_pull(skb, ihl)))
1299 __skb_pull(skb, ihl);
1300 skb_reset_transport_header(skb);
1302 id = ntohs(iph->id);
1303 proto = iph->protocol;
1304 segs = ERR_PTR(-EPROTONOSUPPORT);
1307 ops = rcu_dereference(inet_protos[proto]);
1308 if (likely(ops && ops->gso_segment))
1309 segs = ops->gso_segment(skb, features);
1312 if (!segs || IS_ERR(segs))
1318 if (proto == IPPROTO_UDP) {
1319 iph->id = htons(id);
1320 iph->frag_off = htons(offset >> 3);
1321 if (skb->next != NULL)
1322 iph->frag_off |= htons(IP_MF);
1323 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1325 iph->id = htons(id++);
1326 iph->tot_len = htons(skb->len - skb->mac_len);
1328 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1329 } while ((skb = skb->next));
1335 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1336 struct sk_buff *skb)
1338 const struct net_protocol *ops;
1339 struct sk_buff **pp = NULL;
1341 const struct iphdr *iph;
1348 off = skb_gro_offset(skb);
1349 hlen = off + sizeof(*iph);
1350 iph = skb_gro_header_fast(skb, off);
1351 if (skb_gro_header_hard(skb, hlen)) {
1352 iph = skb_gro_header_slow(skb, hlen, off);
1357 proto = iph->protocol;
1360 ops = rcu_dereference(inet_protos[proto]);
1361 if (!ops || !ops->gro_receive)
1364 if (*(u8 *)iph != 0x45)
1367 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1370 id = ntohl(*(__be32 *)&iph->id);
1371 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1374 for (p = *head; p; p = p->next) {
1377 if (!NAPI_GRO_CB(p)->same_flow)
1382 if ((iph->protocol ^ iph2->protocol) |
1383 (iph->tos ^ iph2->tos) |
1384 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1385 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1386 NAPI_GRO_CB(p)->same_flow = 0;
1390 /* All fields must match except length and checksum. */
1391 NAPI_GRO_CB(p)->flush |=
1392 (iph->ttl ^ iph2->ttl) |
1393 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1395 NAPI_GRO_CB(p)->flush |= flush;
1398 NAPI_GRO_CB(skb)->flush |= flush;
1399 skb_gro_pull(skb, sizeof(*iph));
1400 skb_set_transport_header(skb, skb_gro_offset(skb));
1402 pp = ops->gro_receive(head, skb);
1408 NAPI_GRO_CB(skb)->flush |= flush;
1413 static int inet_gro_complete(struct sk_buff *skb)
1415 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1416 struct iphdr *iph = ip_hdr(skb);
1417 const struct net_protocol *ops;
1418 int proto = iph->protocol;
1421 csum_replace2(&iph->check, iph->tot_len, newlen);
1422 iph->tot_len = newlen;
1425 ops = rcu_dereference(inet_protos[proto]);
1426 if (WARN_ON(!ops || !ops->gro_complete))
1429 err = ops->gro_complete(skb);
1437 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1438 unsigned short type, unsigned char protocol,
1441 struct socket *sock;
1442 int rc = sock_create_kern(family, type, protocol, &sock);
1446 (*sk)->sk_allocation = GFP_ATOMIC;
1448 * Unhash it so that IP input processing does not even see it,
1449 * we do not wish this socket to see incoming packets.
1451 (*sk)->sk_prot->unhash(*sk);
1453 sk_change_net(*sk, net);
1457 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1459 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1461 unsigned long res = 0;
1464 for_each_possible_cpu(i) {
1465 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1466 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1470 EXPORT_SYMBOL_GPL(snmp_fold_field);
1472 #if BITS_PER_LONG==32
1474 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1479 for_each_possible_cpu(cpu) {
1481 struct u64_stats_sync *syncp;
1485 bhptr = per_cpu_ptr(mib[0], cpu);
1486 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1488 start = u64_stats_fetch_begin_bh(syncp);
1489 v = *(((u64 *) bhptr) + offt);
1490 } while (u64_stats_fetch_retry_bh(syncp, start));
1496 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1499 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1501 BUG_ON(ptr == NULL);
1502 ptr[0] = __alloc_percpu(mibsize, align);
1505 #if SNMP_ARRAY_SZ == 2
1506 ptr[1] = __alloc_percpu(mibsize, align);
1508 free_percpu(ptr[0]);
1515 EXPORT_SYMBOL_GPL(snmp_mib_init);
1517 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1521 BUG_ON(ptr == NULL);
1522 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1523 free_percpu(ptr[i]);
1527 EXPORT_SYMBOL_GPL(snmp_mib_free);
1529 #ifdef CONFIG_IP_MULTICAST
1530 static const struct net_protocol igmp_protocol = {
1531 .handler = igmp_rcv,
1536 static const struct net_protocol tcp_protocol = {
1537 .early_demux = tcp_v4_early_demux,
1538 .handler = tcp_v4_rcv,
1539 .err_handler = tcp_v4_err,
1540 .gso_send_check = tcp_v4_gso_send_check,
1541 .gso_segment = tcp_tso_segment,
1542 .gro_receive = tcp4_gro_receive,
1543 .gro_complete = tcp4_gro_complete,
1548 static const struct net_protocol udp_protocol = {
1550 .err_handler = udp_err,
1551 .gso_send_check = udp4_ufo_send_check,
1552 .gso_segment = udp4_ufo_fragment,
1557 static const struct net_protocol icmp_protocol = {
1558 .handler = icmp_rcv,
1559 .err_handler = ping_err,
1564 static __net_init int ipv4_mib_init_net(struct net *net)
1566 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1567 sizeof(struct tcp_mib),
1568 __alignof__(struct tcp_mib)) < 0)
1570 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1571 sizeof(struct ipstats_mib),
1572 __alignof__(struct ipstats_mib)) < 0)
1574 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1575 sizeof(struct linux_mib),
1576 __alignof__(struct linux_mib)) < 0)
1578 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1579 sizeof(struct udp_mib),
1580 __alignof__(struct udp_mib)) < 0)
1582 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1583 sizeof(struct udp_mib),
1584 __alignof__(struct udp_mib)) < 0)
1585 goto err_udplite_mib;
1586 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1587 sizeof(struct icmp_mib),
1588 __alignof__(struct icmp_mib)) < 0)
1590 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1592 if (!net->mib.icmpmsg_statistics)
1593 goto err_icmpmsg_mib;
1599 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1601 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1603 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1605 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1607 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1609 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1614 static __net_exit void ipv4_mib_exit_net(struct net *net)
1616 kfree(net->mib.icmpmsg_statistics);
1617 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1618 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1619 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1620 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1621 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1622 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1625 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1626 .init = ipv4_mib_init_net,
1627 .exit = ipv4_mib_exit_net,
1630 static int __init init_ipv4_mibs(void)
1632 return register_pernet_subsys(&ipv4_mib_ops);
1635 static int ipv4_proc_init(void);
1638 * IP protocol layer initialiser
1641 static struct packet_type ip_packet_type __read_mostly = {
1642 .type = cpu_to_be16(ETH_P_IP),
1644 .gso_send_check = inet_gso_send_check,
1645 .gso_segment = inet_gso_segment,
1646 .gro_receive = inet_gro_receive,
1647 .gro_complete = inet_gro_complete,
1650 static int __init inet_init(void)
1652 struct sk_buff *dummy_skb;
1653 struct inet_protosw *q;
1654 struct list_head *r;
1657 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1659 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1660 if (!sysctl_local_reserved_ports)
1663 rc = proto_register(&tcp_prot, 1);
1665 goto out_free_reserved_ports;
1667 rc = proto_register(&udp_prot, 1);
1669 goto out_unregister_tcp_proto;
1671 rc = proto_register(&raw_prot, 1);
1673 goto out_unregister_udp_proto;
1675 rc = proto_register(&ping_prot, 1);
1677 goto out_unregister_raw_proto;
1680 * Tell SOCKET that we are alive...
1683 (void)sock_register(&inet_family_ops);
1685 #ifdef CONFIG_SYSCTL
1686 ip_static_sysctl_init();
1689 tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1692 * Add all the base protocols.
1695 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1696 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1697 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1698 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1699 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1700 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1701 #ifdef CONFIG_IP_MULTICAST
1702 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1703 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1706 /* Register the socket-side information for inet_create. */
1707 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1710 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1711 inet_register_protosw(q);
1714 * Set the ARP module up
1720 * Set the IP module up
1727 /* Setup TCP slab cache for open requests. */
1730 /* Setup UDP memory threshold */
1733 /* Add UDP-Lite (RFC 3828) */
1734 udplite4_register();
1739 * Set the ICMP layer up
1742 if (icmp_init() < 0)
1743 panic("Failed to create the ICMP control socket.\n");
1746 * Initialise the multicast router
1748 #if defined(CONFIG_IP_MROUTE)
1750 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1753 * Initialise per-cpu ipv4 mibs
1756 if (init_ipv4_mibs())
1757 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1763 dev_add_pack(&ip_packet_type);
1768 out_unregister_raw_proto:
1769 proto_unregister(&raw_prot);
1770 out_unregister_udp_proto:
1771 proto_unregister(&udp_prot);
1772 out_unregister_tcp_proto:
1773 proto_unregister(&tcp_prot);
1774 out_free_reserved_ports:
1775 kfree(sysctl_local_reserved_ports);
1779 fs_initcall(inet_init);
1781 /* ------------------------------------------------------------------------ */
1783 #ifdef CONFIG_PROC_FS
1784 static int __init ipv4_proc_init(void)
1788 if (raw_proc_init())
1790 if (tcp4_proc_init())
1792 if (udp4_proc_init())
1794 if (ping_proc_init())
1796 if (ip_misc_proc_init())
1813 #else /* CONFIG_PROC_FS */
1814 static int __init ipv4_proc_init(void)
1818 #endif /* CONFIG_PROC_FS */
1820 MODULE_ALIAS_NETPROTO(PF_INET);