2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PF_INET protocol family socket handler.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
13 * Changes (see also sock.c)
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
68 #include <linux/err.h>
69 #include <linux/errno.h>
70 #include <linux/types.h>
71 #include <linux/socket.h>
73 #include <linux/kernel.h>
74 #include <linux/module.h>
75 #include <linux/sched.h>
76 #include <linux/timer.h>
77 #include <linux/string.h>
78 #include <linux/sockios.h>
79 #include <linux/net.h>
80 #include <linux/capability.h>
81 #include <linux/fcntl.h>
83 #include <linux/interrupt.h>
84 #include <linux/stat.h>
85 #include <linux/init.h>
86 #include <linux/poll.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/slab.h>
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/inetdevice.h>
97 #include <linux/netdevice.h>
98 #include <net/checksum.h>
100 #include <net/protocol.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
107 #include <net/udplite.h>
108 #include <net/ping.h>
109 #include <linux/skbuff.h>
110 #include <net/sock.h>
112 #include <net/icmp.h>
113 #include <net/ipip.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
122 /* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
125 static struct list_head inetsw[SOCK_MAX];
126 static DEFINE_SPINLOCK(inetsw_lock);
128 struct ipv4_config ipv4_config;
129 EXPORT_SYMBOL(ipv4_config);
131 /* New destruction routine */
133 void inet_sock_destruct(struct sock *sk)
135 struct inet_sock *inet = inet_sk(sk);
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 pr_err("Attempt to release TCP socket in state %d %p\n",
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 pr_err("Attempt to release alive inet socket %p\n", sk);
152 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 WARN_ON(sk->sk_wmem_queued);
155 WARN_ON(sk->sk_forward_alloc);
157 kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 sk_refcnt_debug_dec(sk);
161 EXPORT_SYMBOL(inet_sock_destruct);
164 * The routines beyond this point handle the behaviour of an AF_INET
165 * socket object. Mostly it punts to the subprotocols of IP to do
170 * Automatically bind an unbound socket.
173 static int inet_autobind(struct sock *sk)
175 struct inet_sock *inet;
176 /* We may need to bind the socket. */
179 if (!inet->inet_num) {
180 if (sk->sk_prot->get_port(sk, 0)) {
184 inet->inet_sport = htons(inet->inet_num);
191 * Move a socket into listening state.
193 int inet_listen(struct socket *sock, int backlog)
195 struct sock *sk = sock->sk;
196 unsigned char old_state;
202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 old_state = sk->sk_state;
206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 /* Really, if the socket is already in listen state
210 * we can only allow the backlog to be adjusted.
212 if (old_state != TCP_LISTEN) {
213 err = inet_csk_listen_start(sk, backlog);
217 sk->sk_max_ack_backlog = backlog;
224 EXPORT_SYMBOL(inet_listen);
226 u32 inet_ehash_secret __read_mostly;
227 EXPORT_SYMBOL(inet_ehash_secret);
229 u32 ipv6_hash_secret __read_mostly;
230 EXPORT_SYMBOL(ipv6_hash_secret);
233 * inet_ehash_secret must be set exactly once, and to a non nul value
234 * ipv6_hash_secret must be set exactly once.
236 void build_ehash_secret(void)
241 get_random_bytes(&rnd, sizeof(rnd));
244 if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
245 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
247 EXPORT_SYMBOL(build_ehash_secret);
249 static inline int inet_netns_ok(struct net *net, int protocol)
252 const struct net_protocol *ipprot;
254 if (net_eq(net, &init_net))
257 hash = protocol & (MAX_INET_PROTOS - 1);
258 ipprot = rcu_dereference(inet_protos[hash]);
263 return ipprot->netns_ok;
267 * Create an inet socket.
270 static int inet_create(struct net *net, struct socket *sock, int protocol,
274 struct inet_protosw *answer;
275 struct inet_sock *inet;
276 struct proto *answer_prot;
277 unsigned char answer_flags;
278 char answer_no_check;
279 int try_loading_module = 0;
282 if (protocol < 0 || protocol >= IPPROTO_MAX)
285 if (unlikely(!inet_ehash_secret))
286 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
287 build_ehash_secret();
289 sock->state = SS_UNCONNECTED;
291 /* Look for the requested type/protocol pair. */
293 err = -ESOCKTNOSUPPORT;
295 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
298 /* Check the non-wild match. */
299 if (protocol == answer->protocol) {
300 if (protocol != IPPROTO_IP)
303 /* Check for the two wild cases. */
304 if (IPPROTO_IP == protocol) {
305 protocol = answer->protocol;
308 if (IPPROTO_IP == answer->protocol)
311 err = -EPROTONOSUPPORT;
315 if (try_loading_module < 2) {
318 * Be more specific, e.g. net-pf-2-proto-132-type-1
319 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
321 if (++try_loading_module == 1)
322 request_module("net-pf-%d-proto-%d-type-%d",
323 PF_INET, protocol, sock->type);
325 * Fall back to generic, e.g. net-pf-2-proto-132
326 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
329 request_module("net-pf-%d-proto-%d",
331 goto lookup_protocol;
337 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
341 if (!inet_netns_ok(net, protocol))
344 sock->ops = answer->ops;
345 answer_prot = answer->prot;
346 answer_no_check = answer->no_check;
347 answer_flags = answer->flags;
350 WARN_ON(answer_prot->slab == NULL);
353 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
358 sk->sk_no_check = answer_no_check;
359 if (INET_PROTOSW_REUSE & answer_flags)
363 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
367 if (SOCK_RAW == sock->type) {
368 inet->inet_num = protocol;
369 if (IPPROTO_RAW == protocol)
373 if (ipv4_config.no_pmtu_disc)
374 inet->pmtudisc = IP_PMTUDISC_DONT;
376 inet->pmtudisc = IP_PMTUDISC_WANT;
380 sock_init_data(sock, sk);
382 sk->sk_destruct = inet_sock_destruct;
383 sk->sk_protocol = protocol;
384 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
391 inet->mc_list = NULL;
393 sk_refcnt_debug_inc(sk);
395 if (inet->inet_num) {
396 /* It assumes that any protocol which allows
397 * the user to assign a number at socket
398 * creation time automatically
401 inet->inet_sport = htons(inet->inet_num);
402 /* Add to protocol hash chains. */
403 sk->sk_prot->hash(sk);
406 if (sk->sk_prot->init) {
407 err = sk->sk_prot->init(sk);
409 sk_common_release(sk);
420 * The peer socket should always be NULL (or else). When we call this
421 * function we are destroying the object and from then on nobody
422 * should refer to it.
424 int inet_release(struct socket *sock)
426 struct sock *sk = sock->sk;
431 sock_rps_reset_flow(sk);
433 /* Applications forget to leave groups before exiting */
434 ip_mc_drop_socket(sk);
436 /* If linger is set, we don't return until the close
437 * is complete. Otherwise we return immediately. The
438 * actually closing is done the same either way.
440 * If the close is due to the process exiting, we never
444 if (sock_flag(sk, SOCK_LINGER) &&
445 !(current->flags & PF_EXITING))
446 timeout = sk->sk_lingertime;
448 sk->sk_prot->close(sk, timeout);
452 EXPORT_SYMBOL(inet_release);
454 /* It is off by default, see below. */
455 int sysctl_ip_nonlocal_bind __read_mostly;
456 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
458 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
460 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
461 struct sock *sk = sock->sk;
462 struct inet_sock *inet = inet_sk(sk);
467 /* If the socket has its own bind function then use it. (RAW) */
468 if (sk->sk_prot->bind) {
469 err = sk->sk_prot->bind(sk, uaddr, addr_len);
473 if (addr_len < sizeof(struct sockaddr_in))
476 if (addr->sin_family != AF_INET) {
477 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
478 * only if s_addr is INADDR_ANY.
481 if (addr->sin_family != AF_UNSPEC ||
482 addr->sin_addr.s_addr != htonl(INADDR_ANY))
486 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
488 /* Not specified by any standard per-se, however it breaks too
489 * many applications when removed. It is unfortunate since
490 * allowing applications to make a non-local bind solves
491 * several problems with systems using dynamic addressing.
492 * (ie. your servers still start up even if your ISDN link
493 * is temporarily down)
495 err = -EADDRNOTAVAIL;
496 if (!sysctl_ip_nonlocal_bind &&
497 !(inet->freebind || inet->transparent) &&
498 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
499 chk_addr_ret != RTN_LOCAL &&
500 chk_addr_ret != RTN_MULTICAST &&
501 chk_addr_ret != RTN_BROADCAST)
504 snum = ntohs(addr->sin_port);
506 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
509 /* We keep a pair of addresses. rcv_saddr is the one
510 * used by hash lookups, and saddr is used for transmit.
512 * In the BSD API these are the same except where it
513 * would be illegal to use them (multicast/broadcast) in
514 * which case the sending device address is used.
518 /* Check these errors (active socket, double bind). */
520 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
521 goto out_release_sock;
523 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
524 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
525 inet->inet_saddr = 0; /* Use device */
527 /* Make sure we are allowed to bind here. */
528 if (sk->sk_prot->get_port(sk, snum)) {
529 inet->inet_saddr = inet->inet_rcv_saddr = 0;
531 goto out_release_sock;
534 if (inet->inet_rcv_saddr)
535 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
537 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
538 inet->inet_sport = htons(inet->inet_num);
539 inet->inet_daddr = 0;
540 inet->inet_dport = 0;
548 EXPORT_SYMBOL(inet_bind);
550 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
551 int addr_len, int flags)
553 struct sock *sk = sock->sk;
555 if (addr_len < sizeof(uaddr->sa_family))
557 if (uaddr->sa_family == AF_UNSPEC)
558 return sk->sk_prot->disconnect(sk, flags);
560 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
562 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
564 EXPORT_SYMBOL(inet_dgram_connect);
566 static long inet_wait_for_connect(struct sock *sk, long timeo)
570 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
572 /* Basic assumption: if someone sets sk->sk_err, he _must_
573 * change state of the socket from TCP_SYN_*.
574 * Connect() does not allow to get error notifications
575 * without closing the socket.
577 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
579 timeo = schedule_timeout(timeo);
581 if (signal_pending(current) || !timeo)
583 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
585 finish_wait(sk_sleep(sk), &wait);
590 * Connect to a remote host. There is regrettably still a little
591 * TCP 'magic' in here.
593 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
594 int addr_len, int flags)
596 struct sock *sk = sock->sk;
600 if (addr_len < sizeof(uaddr->sa_family))
605 if (uaddr->sa_family == AF_UNSPEC) {
606 err = sk->sk_prot->disconnect(sk, flags);
607 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
611 switch (sock->state) {
620 /* Fall out of switch with err, set for this state */
624 if (sk->sk_state != TCP_CLOSE)
627 err = sk->sk_prot->connect(sk, uaddr, addr_len);
631 sock->state = SS_CONNECTING;
633 /* Just entered SS_CONNECTING state; the only
634 * difference is that return value in non-blocking
635 * case is EINPROGRESS, rather than EALREADY.
641 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
643 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
644 /* Error code is set above */
645 if (!timeo || !inet_wait_for_connect(sk, timeo))
648 err = sock_intr_errno(timeo);
649 if (signal_pending(current))
653 /* Connection was closed by RST, timeout, ICMP error
654 * or another process disconnected us.
656 if (sk->sk_state == TCP_CLOSE)
659 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
660 * and error was received after socket entered established state.
661 * Hence, it is handled normally after connect() return successfully.
664 sock->state = SS_CONNECTED;
671 err = sock_error(sk) ? : -ECONNABORTED;
672 sock->state = SS_UNCONNECTED;
673 if (sk->sk_prot->disconnect(sk, flags))
674 sock->state = SS_DISCONNECTING;
677 EXPORT_SYMBOL(inet_stream_connect);
680 * Accept a pending connection. The TCP layer now gives BSD semantics.
683 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
685 struct sock *sk1 = sock->sk;
687 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
694 sock_rps_record_flow(sk2);
695 WARN_ON(!((1 << sk2->sk_state) &
696 (TCPF_ESTABLISHED | 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 kiocb *iocb, struct socket *sock, struct msghdr *msg,
743 struct sock *sk = sock->sk;
745 sock_rps_record_flow(sk);
747 /* We may need to bind the socket. */
748 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
752 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
754 EXPORT_SYMBOL(inet_sendmsg);
756 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
757 size_t size, int flags)
759 struct sock *sk = sock->sk;
761 sock_rps_record_flow(sk);
763 /* We may need to bind the socket. */
764 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
768 if (sk->sk_prot->sendpage)
769 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
770 return sock_no_sendpage(sock, page, offset, size, flags);
772 EXPORT_SYMBOL(inet_sendpage);
774 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
775 size_t size, int flags)
777 struct sock *sk = sock->sk;
781 sock_rps_record_flow(sk);
783 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
784 flags & ~MSG_DONTWAIT, &addr_len);
786 msg->msg_namelen = addr_len;
789 EXPORT_SYMBOL(inet_recvmsg);
791 int inet_shutdown(struct socket *sock, int how)
793 struct sock *sk = sock->sk;
796 /* This should really check to make sure
797 * the socket is a TCP socket. (WHY AC...)
799 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
802 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
806 if (sock->state == SS_CONNECTING) {
807 if ((1 << sk->sk_state) &
808 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
809 sock->state = SS_DISCONNECTING;
811 sock->state = SS_CONNECTED;
814 switch (sk->sk_state) {
817 /* Hack to wake up other listeners, who can poll for
818 POLLHUP, even on eg. unconnected UDP sockets -- RR */
820 sk->sk_shutdown |= how;
821 if (sk->sk_prot->shutdown)
822 sk->sk_prot->shutdown(sk, how);
825 /* Remaining two branches are temporary solution for missing
826 * close() in multithreaded environment. It is _not_ a good idea,
827 * but we have no choice until close() is repaired at VFS level.
830 if (!(how & RCV_SHUTDOWN))
834 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
835 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
839 /* Wake up anyone sleeping in poll. */
840 sk->sk_state_change(sk);
844 EXPORT_SYMBOL(inet_shutdown);
847 * ioctl() calls you can issue on an INET socket. Most of these are
848 * device configuration and stuff and very rarely used. Some ioctls
849 * pass on to the socket itself.
851 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
852 * loads the devconfigure module does its configuring and unloads it.
853 * There's a good 20K of config code hanging around the kernel.
856 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
858 struct sock *sk = sock->sk;
860 struct net *net = sock_net(sk);
864 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
867 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
872 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
877 err = arp_ioctl(net, cmd, (void __user *)arg);
890 err = devinet_ioctl(net, cmd, (void __user *)arg);
893 if (sk->sk_prot->ioctl)
894 err = sk->sk_prot->ioctl(sk, cmd, arg);
901 EXPORT_SYMBOL(inet_ioctl);
904 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
906 struct sock *sk = sock->sk;
907 int err = -ENOIOCTLCMD;
909 if (sk->sk_prot->compat_ioctl)
910 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
916 const struct proto_ops inet_stream_ops = {
918 .owner = THIS_MODULE,
919 .release = inet_release,
921 .connect = inet_stream_connect,
922 .socketpair = sock_no_socketpair,
923 .accept = inet_accept,
924 .getname = inet_getname,
927 .listen = inet_listen,
928 .shutdown = inet_shutdown,
929 .setsockopt = sock_common_setsockopt,
930 .getsockopt = sock_common_getsockopt,
931 .sendmsg = inet_sendmsg,
932 .recvmsg = inet_recvmsg,
933 .mmap = sock_no_mmap,
934 .sendpage = inet_sendpage,
935 .splice_read = tcp_splice_read,
937 .compat_setsockopt = compat_sock_common_setsockopt,
938 .compat_getsockopt = compat_sock_common_getsockopt,
939 .compat_ioctl = inet_compat_ioctl,
942 EXPORT_SYMBOL(inet_stream_ops);
944 const struct proto_ops inet_dgram_ops = {
946 .owner = THIS_MODULE,
947 .release = inet_release,
949 .connect = inet_dgram_connect,
950 .socketpair = sock_no_socketpair,
951 .accept = sock_no_accept,
952 .getname = inet_getname,
955 .listen = sock_no_listen,
956 .shutdown = inet_shutdown,
957 .setsockopt = sock_common_setsockopt,
958 .getsockopt = sock_common_getsockopt,
959 .sendmsg = inet_sendmsg,
960 .recvmsg = inet_recvmsg,
961 .mmap = sock_no_mmap,
962 .sendpage = inet_sendpage,
964 .compat_setsockopt = compat_sock_common_setsockopt,
965 .compat_getsockopt = compat_sock_common_getsockopt,
966 .compat_ioctl = inet_compat_ioctl,
969 EXPORT_SYMBOL(inet_dgram_ops);
972 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
975 static const struct proto_ops inet_sockraw_ops = {
977 .owner = THIS_MODULE,
978 .release = inet_release,
980 .connect = inet_dgram_connect,
981 .socketpair = sock_no_socketpair,
982 .accept = sock_no_accept,
983 .getname = inet_getname,
984 .poll = datagram_poll,
986 .listen = sock_no_listen,
987 .shutdown = inet_shutdown,
988 .setsockopt = sock_common_setsockopt,
989 .getsockopt = sock_common_getsockopt,
990 .sendmsg = inet_sendmsg,
991 .recvmsg = inet_recvmsg,
992 .mmap = sock_no_mmap,
993 .sendpage = inet_sendpage,
995 .compat_setsockopt = compat_sock_common_setsockopt,
996 .compat_getsockopt = compat_sock_common_getsockopt,
997 .compat_ioctl = inet_compat_ioctl,
1001 static const struct net_proto_family inet_family_ops = {
1003 .create = inet_create,
1004 .owner = THIS_MODULE,
1007 /* Upon startup we insert all the elements in inetsw_array[] into
1008 * the linked list inetsw.
1010 static struct inet_protosw inetsw_array[] =
1013 .type = SOCK_STREAM,
1014 .protocol = IPPROTO_TCP,
1016 .ops = &inet_stream_ops,
1018 .flags = INET_PROTOSW_PERMANENT |
1024 .protocol = IPPROTO_UDP,
1026 .ops = &inet_dgram_ops,
1027 .no_check = UDP_CSUM_DEFAULT,
1028 .flags = INET_PROTOSW_PERMANENT,
1033 .protocol = IPPROTO_ICMP,
1035 .ops = &inet_sockraw_ops,
1036 .no_check = UDP_CSUM_DEFAULT,
1037 .flags = INET_PROTOSW_REUSE,
1042 .protocol = IPPROTO_IP, /* wild card */
1044 .ops = &inet_sockraw_ops,
1045 .no_check = UDP_CSUM_DEFAULT,
1046 .flags = INET_PROTOSW_REUSE,
1050 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1052 void inet_register_protosw(struct inet_protosw *p)
1054 struct list_head *lh;
1055 struct inet_protosw *answer;
1056 int protocol = p->protocol;
1057 struct list_head *last_perm;
1059 spin_lock_bh(&inetsw_lock);
1061 if (p->type >= SOCK_MAX)
1064 /* If we are trying to override a permanent protocol, bail. */
1066 last_perm = &inetsw[p->type];
1067 list_for_each(lh, &inetsw[p->type]) {
1068 answer = list_entry(lh, struct inet_protosw, list);
1070 /* Check only the non-wild match. */
1071 if (INET_PROTOSW_PERMANENT & answer->flags) {
1072 if (protocol == answer->protocol)
1082 /* Add the new entry after the last permanent entry if any, so that
1083 * the new entry does not override a permanent entry when matched with
1084 * a wild-card protocol. But it is allowed to override any existing
1085 * non-permanent entry. This means that when we remove this entry, the
1086 * system automatically returns to the old behavior.
1088 list_add_rcu(&p->list, last_perm);
1090 spin_unlock_bh(&inetsw_lock);
1095 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1101 "Ignoring attempt to register invalid socket type %d.\n",
1105 EXPORT_SYMBOL(inet_register_protosw);
1107 void inet_unregister_protosw(struct inet_protosw *p)
1109 if (INET_PROTOSW_PERMANENT & p->flags) {
1111 "Attempt to unregister permanent protocol %d.\n",
1114 spin_lock_bh(&inetsw_lock);
1115 list_del_rcu(&p->list);
1116 spin_unlock_bh(&inetsw_lock);
1121 EXPORT_SYMBOL(inet_unregister_protosw);
1124 * Shall we try to damage output packets if routing dev changes?
1127 int sysctl_ip_dynaddr __read_mostly;
1129 static int inet_sk_reselect_saddr(struct sock *sk)
1131 struct inet_sock *inet = inet_sk(sk);
1132 __be32 old_saddr = inet->inet_saddr;
1133 __be32 daddr = inet->inet_daddr;
1137 struct ip_options_rcu *inet_opt;
1139 inet_opt = rcu_dereference_protected(inet->inet_opt,
1140 sock_owned_by_user(sk));
1141 if (inet_opt && inet_opt->opt.srr)
1142 daddr = inet_opt->opt.faddr;
1144 /* Query new route. */
1145 fl4 = &inet->cork.fl.u.ip4;
1146 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1147 sk->sk_bound_dev_if, sk->sk_protocol,
1148 inet->inet_sport, inet->inet_dport, sk, false);
1152 sk_setup_caps(sk, &rt->dst);
1154 new_saddr = fl4->saddr;
1156 if (new_saddr == old_saddr)
1159 if (sysctl_ip_dynaddr > 1) {
1160 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1161 __func__, &old_saddr, &new_saddr);
1164 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1167 * XXX The only one ugly spot where we need to
1168 * XXX really change the sockets identity after
1169 * XXX it has entered the hashes. -DaveM
1171 * Besides that, it does not check for connection
1172 * uniqueness. Wait for troubles.
1174 __sk_prot_rehash(sk);
1178 int inet_sk_rebuild_header(struct sock *sk)
1180 struct inet_sock *inet = inet_sk(sk);
1181 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1183 struct ip_options_rcu *inet_opt;
1187 /* Route is OK, nothing to do. */
1193 inet_opt = rcu_dereference(inet->inet_opt);
1194 daddr = inet->inet_daddr;
1195 if (inet_opt && inet_opt->opt.srr)
1196 daddr = inet_opt->opt.faddr;
1198 fl4 = &inet->cork.fl.u.ip4;
1199 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1200 inet->inet_dport, inet->inet_sport,
1201 sk->sk_protocol, RT_CONN_FLAGS(sk),
1202 sk->sk_bound_dev_if);
1205 sk_setup_caps(sk, &rt->dst);
1209 /* Routing failed... */
1210 sk->sk_route_caps = 0;
1212 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1213 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1215 if (!sysctl_ip_dynaddr ||
1216 sk->sk_state != TCP_SYN_SENT ||
1217 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1218 (err = inet_sk_reselect_saddr(sk)) != 0)
1219 sk->sk_err_soft = -err;
1224 EXPORT_SYMBOL(inet_sk_rebuild_header);
1226 static int inet_gso_send_check(struct sk_buff *skb)
1228 const struct iphdr *iph;
1229 const struct net_protocol *ops;
1234 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1239 if (ihl < sizeof(*iph))
1242 if (unlikely(!pskb_may_pull(skb, ihl)))
1245 __skb_pull(skb, ihl);
1246 skb_reset_transport_header(skb);
1248 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1249 err = -EPROTONOSUPPORT;
1252 ops = rcu_dereference(inet_protos[proto]);
1253 if (likely(ops && ops->gso_send_check))
1254 err = ops->gso_send_check(skb);
1261 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1263 struct sk_buff *segs = ERR_PTR(-EINVAL);
1265 const struct net_protocol *ops;
1269 unsigned int offset = 0;
1271 if (!(features & NETIF_F_V4_CSUM))
1272 features &= ~NETIF_F_SG;
1274 if (unlikely(skb_shinfo(skb)->gso_type &
1282 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1287 if (ihl < sizeof(*iph))
1290 if (unlikely(!pskb_may_pull(skb, ihl)))
1293 __skb_pull(skb, ihl);
1294 skb_reset_transport_header(skb);
1296 id = ntohs(iph->id);
1297 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1298 segs = ERR_PTR(-EPROTONOSUPPORT);
1301 ops = rcu_dereference(inet_protos[proto]);
1302 if (likely(ops && ops->gso_segment))
1303 segs = ops->gso_segment(skb, features);
1306 if (!segs || IS_ERR(segs))
1312 if (proto == IPPROTO_UDP) {
1313 iph->id = htons(id);
1314 iph->frag_off = htons(offset >> 3);
1315 if (skb->next != NULL)
1316 iph->frag_off |= htons(IP_MF);
1317 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1319 iph->id = htons(id++);
1320 iph->tot_len = htons(skb->len - skb->mac_len);
1322 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1323 } while ((skb = skb->next));
1329 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1330 struct sk_buff *skb)
1332 const struct net_protocol *ops;
1333 struct sk_buff **pp = NULL;
1335 const struct iphdr *iph;
1342 off = skb_gro_offset(skb);
1343 hlen = off + sizeof(*iph);
1344 iph = skb_gro_header_fast(skb, off);
1345 if (skb_gro_header_hard(skb, hlen)) {
1346 iph = skb_gro_header_slow(skb, hlen, off);
1351 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1354 ops = rcu_dereference(inet_protos[proto]);
1355 if (!ops || !ops->gro_receive)
1358 if (*(u8 *)iph != 0x45)
1361 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1364 id = ntohl(*(__be32 *)&iph->id);
1365 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1368 for (p = *head; p; p = p->next) {
1371 if (!NAPI_GRO_CB(p)->same_flow)
1376 if ((iph->protocol ^ iph2->protocol) |
1377 (iph->tos ^ iph2->tos) |
1378 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1379 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1380 NAPI_GRO_CB(p)->same_flow = 0;
1384 /* All fields must match except length and checksum. */
1385 NAPI_GRO_CB(p)->flush |=
1386 (iph->ttl ^ iph2->ttl) |
1387 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1389 NAPI_GRO_CB(p)->flush |= flush;
1392 NAPI_GRO_CB(skb)->flush |= flush;
1393 skb_gro_pull(skb, sizeof(*iph));
1394 skb_set_transport_header(skb, skb_gro_offset(skb));
1396 pp = ops->gro_receive(head, skb);
1402 NAPI_GRO_CB(skb)->flush |= flush;
1407 static int inet_gro_complete(struct sk_buff *skb)
1409 const struct net_protocol *ops;
1410 struct iphdr *iph = ip_hdr(skb);
1411 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1413 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1415 csum_replace2(&iph->check, iph->tot_len, newlen);
1416 iph->tot_len = newlen;
1419 ops = rcu_dereference(inet_protos[proto]);
1420 if (WARN_ON(!ops || !ops->gro_complete))
1423 err = ops->gro_complete(skb);
1431 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1432 unsigned short type, unsigned char protocol,
1435 struct socket *sock;
1436 int rc = sock_create_kern(family, type, protocol, &sock);
1440 (*sk)->sk_allocation = GFP_ATOMIC;
1442 * Unhash it so that IP input processing does not even see it,
1443 * we do not wish this socket to see incoming packets.
1445 (*sk)->sk_prot->unhash(*sk);
1447 sk_change_net(*sk, net);
1451 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1453 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1455 unsigned long res = 0;
1458 for_each_possible_cpu(i) {
1459 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1460 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1464 EXPORT_SYMBOL_GPL(snmp_fold_field);
1466 #if BITS_PER_LONG==32
1468 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1473 for_each_possible_cpu(cpu) {
1475 struct u64_stats_sync *syncp;
1479 bhptr = per_cpu_ptr(mib[0], cpu);
1480 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1482 start = u64_stats_fetch_begin_bh(syncp);
1483 v = *(((u64 *) bhptr) + offt);
1484 } while (u64_stats_fetch_retry_bh(syncp, start));
1490 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1493 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1495 BUG_ON(ptr == NULL);
1496 ptr[0] = __alloc_percpu(mibsize, align);
1499 #if SNMP_ARRAY_SZ == 2
1500 ptr[1] = __alloc_percpu(mibsize, align);
1502 free_percpu(ptr[0]);
1509 EXPORT_SYMBOL_GPL(snmp_mib_init);
1511 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1515 BUG_ON(ptr == NULL);
1516 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1517 free_percpu(ptr[i]);
1521 EXPORT_SYMBOL_GPL(snmp_mib_free);
1523 #ifdef CONFIG_IP_MULTICAST
1524 static const struct net_protocol igmp_protocol = {
1525 .handler = igmp_rcv,
1530 static const struct net_protocol tcp_protocol = {
1531 .handler = tcp_v4_rcv,
1532 .err_handler = tcp_v4_err,
1533 .gso_send_check = tcp_v4_gso_send_check,
1534 .gso_segment = tcp_tso_segment,
1535 .gro_receive = tcp4_gro_receive,
1536 .gro_complete = tcp4_gro_complete,
1541 static const struct net_protocol udp_protocol = {
1543 .err_handler = udp_err,
1544 .gso_send_check = udp4_ufo_send_check,
1545 .gso_segment = udp4_ufo_fragment,
1550 static const struct net_protocol icmp_protocol = {
1551 .handler = icmp_rcv,
1552 .err_handler = ping_err,
1557 static __net_init int ipv4_mib_init_net(struct net *net)
1559 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1560 sizeof(struct tcp_mib),
1561 __alignof__(struct tcp_mib)) < 0)
1563 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1564 sizeof(struct ipstats_mib),
1565 __alignof__(struct ipstats_mib)) < 0)
1567 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1568 sizeof(struct linux_mib),
1569 __alignof__(struct linux_mib)) < 0)
1571 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1572 sizeof(struct udp_mib),
1573 __alignof__(struct udp_mib)) < 0)
1575 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1576 sizeof(struct udp_mib),
1577 __alignof__(struct udp_mib)) < 0)
1578 goto err_udplite_mib;
1579 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1580 sizeof(struct icmp_mib),
1581 __alignof__(struct icmp_mib)) < 0)
1583 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1584 sizeof(struct icmpmsg_mib),
1585 __alignof__(struct icmpmsg_mib)) < 0)
1586 goto err_icmpmsg_mib;
1592 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1594 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1596 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1598 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1600 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1602 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1607 static __net_exit void ipv4_mib_exit_net(struct net *net)
1609 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1610 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1611 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1612 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1613 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1614 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1615 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1618 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1619 .init = ipv4_mib_init_net,
1620 .exit = ipv4_mib_exit_net,
1623 static int __init init_ipv4_mibs(void)
1625 return register_pernet_subsys(&ipv4_mib_ops);
1628 static int ipv4_proc_init(void);
1631 * IP protocol layer initialiser
1634 static struct packet_type ip_packet_type __read_mostly = {
1635 .type = cpu_to_be16(ETH_P_IP),
1637 .gso_send_check = inet_gso_send_check,
1638 .gso_segment = inet_gso_segment,
1639 .gro_receive = inet_gro_receive,
1640 .gro_complete = inet_gro_complete,
1643 static int __init inet_init(void)
1645 struct sk_buff *dummy_skb;
1646 struct inet_protosw *q;
1647 struct list_head *r;
1650 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1652 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1653 if (!sysctl_local_reserved_ports)
1656 rc = proto_register(&tcp_prot, 1);
1658 goto out_free_reserved_ports;
1660 rc = proto_register(&udp_prot, 1);
1662 goto out_unregister_tcp_proto;
1664 rc = proto_register(&raw_prot, 1);
1666 goto out_unregister_udp_proto;
1668 rc = proto_register(&ping_prot, 1);
1670 goto out_unregister_raw_proto;
1673 * Tell SOCKET that we are alive...
1676 (void)sock_register(&inet_family_ops);
1678 #ifdef CONFIG_SYSCTL
1679 ip_static_sysctl_init();
1683 * Add all the base protocols.
1686 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1687 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1688 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1689 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1690 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1691 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1692 #ifdef CONFIG_IP_MULTICAST
1693 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1694 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1697 /* Register the socket-side information for inet_create. */
1698 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1701 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1702 inet_register_protosw(q);
1705 * Set the ARP module up
1711 * Set the IP module up
1718 /* Setup TCP slab cache for open requests. */
1721 /* Setup UDP memory threshold */
1724 /* Add UDP-Lite (RFC 3828) */
1725 udplite4_register();
1730 * Set the ICMP layer up
1733 if (icmp_init() < 0)
1734 panic("Failed to create the ICMP control socket.\n");
1737 * Initialise the multicast router
1739 #if defined(CONFIG_IP_MROUTE)
1741 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1744 * Initialise per-cpu ipv4 mibs
1747 if (init_ipv4_mibs())
1748 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1754 dev_add_pack(&ip_packet_type);
1759 out_unregister_raw_proto:
1760 proto_unregister(&raw_prot);
1761 out_unregister_udp_proto:
1762 proto_unregister(&udp_prot);
1763 out_unregister_tcp_proto:
1764 proto_unregister(&tcp_prot);
1765 out_free_reserved_ports:
1766 kfree(sysctl_local_reserved_ports);
1770 fs_initcall(inet_init);
1772 /* ------------------------------------------------------------------------ */
1774 #ifdef CONFIG_PROC_FS
1775 static int __init ipv4_proc_init(void)
1779 if (raw_proc_init())
1781 if (tcp4_proc_init())
1783 if (udp4_proc_init())
1785 if (ping_proc_init())
1787 if (ip_misc_proc_init())
1804 #else /* CONFIG_PROC_FS */
1805 static int __init ipv4_proc_init(void)
1809 #endif /* CONFIG_PROC_FS */
1811 MODULE_ALIAS_NETPROTO(PF_INET);