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>
121 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
122 #include <linux/android_aid.h>
124 static inline int current_has_network(void)
126 return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
129 static inline int current_has_network(void)
135 /* The inetsw table contains everything that inet_create needs to
136 * build a new socket.
138 static struct list_head inetsw[SOCK_MAX];
139 static DEFINE_SPINLOCK(inetsw_lock);
141 struct ipv4_config ipv4_config;
142 EXPORT_SYMBOL(ipv4_config);
144 /* New destruction routine */
146 void inet_sock_destruct(struct sock *sk)
148 struct inet_sock *inet = inet_sk(sk);
150 __skb_queue_purge(&sk->sk_receive_queue);
151 __skb_queue_purge(&sk->sk_error_queue);
155 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
156 pr_err("Attempt to release TCP socket in state %d %p\n",
160 if (!sock_flag(sk, SOCK_DEAD)) {
161 pr_err("Attempt to release alive inet socket %p\n", sk);
165 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
166 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
167 WARN_ON(sk->sk_wmem_queued);
168 WARN_ON(sk->sk_forward_alloc);
170 kfree(rcu_dereference_protected(inet->inet_opt, 1));
171 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
172 sk_refcnt_debug_dec(sk);
174 EXPORT_SYMBOL(inet_sock_destruct);
177 * The routines beyond this point handle the behaviour of an AF_INET
178 * socket object. Mostly it punts to the subprotocols of IP to do
183 * Automatically bind an unbound socket.
186 static int inet_autobind(struct sock *sk)
188 struct inet_sock *inet;
189 /* We may need to bind the socket. */
192 if (!inet->inet_num) {
193 if (sk->sk_prot->get_port(sk, 0)) {
197 inet->inet_sport = htons(inet->inet_num);
204 * Move a socket into listening state.
206 int inet_listen(struct socket *sock, int backlog)
208 struct sock *sk = sock->sk;
209 unsigned char old_state;
215 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
218 old_state = sk->sk_state;
219 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
222 /* Really, if the socket is already in listen state
223 * we can only allow the backlog to be adjusted.
225 if (old_state != TCP_LISTEN) {
226 err = inet_csk_listen_start(sk, backlog);
230 sk->sk_max_ack_backlog = backlog;
237 EXPORT_SYMBOL(inet_listen);
239 u32 inet_ehash_secret __read_mostly;
240 EXPORT_SYMBOL(inet_ehash_secret);
242 u32 ipv6_hash_secret __read_mostly;
243 EXPORT_SYMBOL(ipv6_hash_secret);
246 * inet_ehash_secret must be set exactly once, and to a non nul value
247 * ipv6_hash_secret must be set exactly once.
249 void build_ehash_secret(void)
254 get_random_bytes(&rnd, sizeof(rnd));
257 if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
258 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
260 EXPORT_SYMBOL(build_ehash_secret);
262 static inline int inet_netns_ok(struct net *net, int protocol)
265 const struct net_protocol *ipprot;
267 if (net_eq(net, &init_net))
270 hash = protocol & (MAX_INET_PROTOS - 1);
271 ipprot = rcu_dereference(inet_protos[hash]);
276 return ipprot->netns_ok;
281 * Create an inet socket.
284 static int inet_create(struct net *net, struct socket *sock, int protocol,
288 struct inet_protosw *answer;
289 struct inet_sock *inet;
290 struct proto *answer_prot;
291 unsigned char answer_flags;
292 char answer_no_check;
293 int try_loading_module = 0;
296 if (!current_has_network())
299 if (unlikely(!inet_ehash_secret))
300 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
301 build_ehash_secret();
303 sock->state = SS_UNCONNECTED;
305 /* Look for the requested type/protocol pair. */
307 err = -ESOCKTNOSUPPORT;
309 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
312 /* Check the non-wild match. */
313 if (protocol == answer->protocol) {
314 if (protocol != IPPROTO_IP)
317 /* Check for the two wild cases. */
318 if (IPPROTO_IP == protocol) {
319 protocol = answer->protocol;
322 if (IPPROTO_IP == answer->protocol)
325 err = -EPROTONOSUPPORT;
329 if (try_loading_module < 2) {
332 * Be more specific, e.g. net-pf-2-proto-132-type-1
333 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
335 if (++try_loading_module == 1)
336 request_module("net-pf-%d-proto-%d-type-%d",
337 PF_INET, protocol, sock->type);
339 * Fall back to generic, e.g. net-pf-2-proto-132
340 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
343 request_module("net-pf-%d-proto-%d",
345 goto lookup_protocol;
351 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
355 if (!inet_netns_ok(net, protocol))
358 sock->ops = answer->ops;
359 answer_prot = answer->prot;
360 answer_no_check = answer->no_check;
361 answer_flags = answer->flags;
364 WARN_ON(answer_prot->slab == NULL);
367 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
372 sk->sk_no_check = answer_no_check;
373 if (INET_PROTOSW_REUSE & answer_flags)
377 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
381 if (SOCK_RAW == sock->type) {
382 inet->inet_num = protocol;
383 if (IPPROTO_RAW == protocol)
387 if (ipv4_config.no_pmtu_disc)
388 inet->pmtudisc = IP_PMTUDISC_DONT;
390 inet->pmtudisc = IP_PMTUDISC_WANT;
394 sock_init_data(sock, sk);
396 sk->sk_destruct = inet_sock_destruct;
397 sk->sk_protocol = protocol;
398 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
405 inet->mc_list = NULL;
407 sk_refcnt_debug_inc(sk);
409 if (inet->inet_num) {
410 /* It assumes that any protocol which allows
411 * the user to assign a number at socket
412 * creation time automatically
415 inet->inet_sport = htons(inet->inet_num);
416 /* Add to protocol hash chains. */
417 sk->sk_prot->hash(sk);
420 if (sk->sk_prot->init) {
421 err = sk->sk_prot->init(sk);
423 sk_common_release(sk);
434 * The peer socket should always be NULL (or else). When we call this
435 * function we are destroying the object and from then on nobody
436 * should refer to it.
438 int inet_release(struct socket *sock)
440 struct sock *sk = sock->sk;
445 sock_rps_reset_flow(sk);
447 /* Applications forget to leave groups before exiting */
448 ip_mc_drop_socket(sk);
450 /* If linger is set, we don't return until the close
451 * is complete. Otherwise we return immediately. The
452 * actually closing is done the same either way.
454 * If the close is due to the process exiting, we never
458 if (sock_flag(sk, SOCK_LINGER) &&
459 !(current->flags & PF_EXITING))
460 timeout = sk->sk_lingertime;
462 sk->sk_prot->close(sk, timeout);
466 EXPORT_SYMBOL(inet_release);
468 /* It is off by default, see below. */
469 int sysctl_ip_nonlocal_bind __read_mostly;
470 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
472 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
474 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
475 struct sock *sk = sock->sk;
476 struct inet_sock *inet = inet_sk(sk);
481 /* If the socket has its own bind function then use it. (RAW) */
482 if (sk->sk_prot->bind) {
483 err = sk->sk_prot->bind(sk, uaddr, addr_len);
487 if (addr_len < sizeof(struct sockaddr_in))
490 if (addr->sin_family != AF_INET) {
495 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
497 /* Not specified by any standard per-se, however it breaks too
498 * many applications when removed. It is unfortunate since
499 * allowing applications to make a non-local bind solves
500 * several problems with systems using dynamic addressing.
501 * (ie. your servers still start up even if your ISDN link
502 * is temporarily down)
504 err = -EADDRNOTAVAIL;
505 if (!sysctl_ip_nonlocal_bind &&
506 !(inet->freebind || inet->transparent) &&
507 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
508 chk_addr_ret != RTN_LOCAL &&
509 chk_addr_ret != RTN_MULTICAST &&
510 chk_addr_ret != RTN_BROADCAST)
513 snum = ntohs(addr->sin_port);
515 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
518 /* We keep a pair of addresses. rcv_saddr is the one
519 * used by hash lookups, and saddr is used for transmit.
521 * In the BSD API these are the same except where it
522 * would be illegal to use them (multicast/broadcast) in
523 * which case the sending device address is used.
527 /* Check these errors (active socket, double bind). */
529 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
530 goto out_release_sock;
532 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
533 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
534 inet->inet_saddr = 0; /* Use device */
536 /* Make sure we are allowed to bind here. */
537 if (sk->sk_prot->get_port(sk, snum)) {
538 inet->inet_saddr = inet->inet_rcv_saddr = 0;
540 goto out_release_sock;
543 if (inet->inet_rcv_saddr)
544 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
546 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
547 inet->inet_sport = htons(inet->inet_num);
548 inet->inet_daddr = 0;
549 inet->inet_dport = 0;
557 EXPORT_SYMBOL(inet_bind);
559 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
560 int addr_len, int flags)
562 struct sock *sk = sock->sk;
564 if (addr_len < sizeof(uaddr->sa_family))
566 if (uaddr->sa_family == AF_UNSPEC)
567 return sk->sk_prot->disconnect(sk, flags);
569 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
571 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
573 EXPORT_SYMBOL(inet_dgram_connect);
575 static long inet_wait_for_connect(struct sock *sk, long timeo)
579 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
581 /* Basic assumption: if someone sets sk->sk_err, he _must_
582 * change state of the socket from TCP_SYN_*.
583 * Connect() does not allow to get error notifications
584 * without closing the socket.
586 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
588 timeo = schedule_timeout(timeo);
590 if (signal_pending(current) || !timeo)
592 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
594 finish_wait(sk_sleep(sk), &wait);
599 * Connect to a remote host. There is regrettably still a little
600 * TCP 'magic' in here.
602 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
603 int addr_len, int flags)
605 struct sock *sk = sock->sk;
609 if (addr_len < sizeof(uaddr->sa_family))
614 if (uaddr->sa_family == AF_UNSPEC) {
615 err = sk->sk_prot->disconnect(sk, flags);
616 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
620 switch (sock->state) {
629 /* Fall out of switch with err, set for this state */
633 if (sk->sk_state != TCP_CLOSE)
636 err = sk->sk_prot->connect(sk, uaddr, addr_len);
640 sock->state = SS_CONNECTING;
642 /* Just entered SS_CONNECTING state; the only
643 * difference is that return value in non-blocking
644 * case is EINPROGRESS, rather than EALREADY.
650 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
652 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
653 /* Error code is set above */
654 if (!timeo || !inet_wait_for_connect(sk, timeo))
657 err = sock_intr_errno(timeo);
658 if (signal_pending(current))
662 /* Connection was closed by RST, timeout, ICMP error
663 * or another process disconnected us.
665 if (sk->sk_state == TCP_CLOSE)
668 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
669 * and error was received after socket entered established state.
670 * Hence, it is handled normally after connect() return successfully.
673 sock->state = SS_CONNECTED;
680 err = sock_error(sk) ? : -ECONNABORTED;
681 sock->state = SS_UNCONNECTED;
682 if (sk->sk_prot->disconnect(sk, flags))
683 sock->state = SS_DISCONNECTING;
686 EXPORT_SYMBOL(inet_stream_connect);
689 * Accept a pending connection. The TCP layer now gives BSD semantics.
692 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
694 struct sock *sk1 = sock->sk;
696 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
703 sock_rps_record_flow(sk2);
704 WARN_ON(!((1 << sk2->sk_state) &
705 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
707 sock_graft(sk2, newsock);
709 newsock->state = SS_CONNECTED;
715 EXPORT_SYMBOL(inet_accept);
719 * This does both peername and sockname.
721 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
722 int *uaddr_len, int peer)
724 struct sock *sk = sock->sk;
725 struct inet_sock *inet = inet_sk(sk);
726 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
728 sin->sin_family = AF_INET;
730 if (!inet->inet_dport ||
731 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
734 sin->sin_port = inet->inet_dport;
735 sin->sin_addr.s_addr = inet->inet_daddr;
737 __be32 addr = inet->inet_rcv_saddr;
739 addr = inet->inet_saddr;
740 sin->sin_port = inet->inet_sport;
741 sin->sin_addr.s_addr = addr;
743 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
744 *uaddr_len = sizeof(*sin);
747 EXPORT_SYMBOL(inet_getname);
749 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
752 struct sock *sk = sock->sk;
754 sock_rps_record_flow(sk);
756 /* We may need to bind the socket. */
757 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
761 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
763 EXPORT_SYMBOL(inet_sendmsg);
765 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
766 size_t size, int flags)
768 struct sock *sk = sock->sk;
770 sock_rps_record_flow(sk);
772 /* We may need to bind the socket. */
773 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
777 if (sk->sk_prot->sendpage)
778 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
779 return sock_no_sendpage(sock, page, offset, size, flags);
781 EXPORT_SYMBOL(inet_sendpage);
783 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
784 size_t size, int flags)
786 struct sock *sk = sock->sk;
790 sock_rps_record_flow(sk);
792 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
793 flags & ~MSG_DONTWAIT, &addr_len);
795 msg->msg_namelen = addr_len;
798 EXPORT_SYMBOL(inet_recvmsg);
800 int inet_shutdown(struct socket *sock, int how)
802 struct sock *sk = sock->sk;
805 /* This should really check to make sure
806 * the socket is a TCP socket. (WHY AC...)
808 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
811 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
815 if (sock->state == SS_CONNECTING) {
816 if ((1 << sk->sk_state) &
817 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
818 sock->state = SS_DISCONNECTING;
820 sock->state = SS_CONNECTED;
823 switch (sk->sk_state) {
826 /* Hack to wake up other listeners, who can poll for
827 POLLHUP, even on eg. unconnected UDP sockets -- RR */
829 sk->sk_shutdown |= how;
830 if (sk->sk_prot->shutdown)
831 sk->sk_prot->shutdown(sk, how);
834 /* Remaining two branches are temporary solution for missing
835 * close() in multithreaded environment. It is _not_ a good idea,
836 * but we have no choice until close() is repaired at VFS level.
839 if (!(how & RCV_SHUTDOWN))
843 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
844 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
848 /* Wake up anyone sleeping in poll. */
849 sk->sk_state_change(sk);
853 EXPORT_SYMBOL(inet_shutdown);
856 * ioctl() calls you can issue on an INET socket. Most of these are
857 * device configuration and stuff and very rarely used. Some ioctls
858 * pass on to the socket itself.
860 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
861 * loads the devconfigure module does its configuring and unloads it.
862 * There's a good 20K of config code hanging around the kernel.
865 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
867 struct sock *sk = sock->sk;
869 struct net *net = sock_net(sk);
873 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
876 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
881 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
886 err = arp_ioctl(net, cmd, (void __user *)arg);
900 err = devinet_ioctl(net, cmd, (void __user *)arg);
903 if (sk->sk_prot->ioctl)
904 err = sk->sk_prot->ioctl(sk, cmd, arg);
911 EXPORT_SYMBOL(inet_ioctl);
914 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
916 struct sock *sk = sock->sk;
917 int err = -ENOIOCTLCMD;
919 if (sk->sk_prot->compat_ioctl)
920 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
926 const struct proto_ops inet_stream_ops = {
928 .owner = THIS_MODULE,
929 .release = inet_release,
931 .connect = inet_stream_connect,
932 .socketpair = sock_no_socketpair,
933 .accept = inet_accept,
934 .getname = inet_getname,
937 .listen = inet_listen,
938 .shutdown = inet_shutdown,
939 .setsockopt = sock_common_setsockopt,
940 .getsockopt = sock_common_getsockopt,
941 .sendmsg = inet_sendmsg,
942 .recvmsg = inet_recvmsg,
943 .mmap = sock_no_mmap,
944 .sendpage = inet_sendpage,
945 .splice_read = tcp_splice_read,
947 .compat_setsockopt = compat_sock_common_setsockopt,
948 .compat_getsockopt = compat_sock_common_getsockopt,
949 .compat_ioctl = inet_compat_ioctl,
952 EXPORT_SYMBOL(inet_stream_ops);
954 const struct proto_ops inet_dgram_ops = {
956 .owner = THIS_MODULE,
957 .release = inet_release,
959 .connect = inet_dgram_connect,
960 .socketpair = sock_no_socketpair,
961 .accept = sock_no_accept,
962 .getname = inet_getname,
965 .listen = sock_no_listen,
966 .shutdown = inet_shutdown,
967 .setsockopt = sock_common_setsockopt,
968 .getsockopt = sock_common_getsockopt,
969 .sendmsg = inet_sendmsg,
970 .recvmsg = inet_recvmsg,
971 .mmap = sock_no_mmap,
972 .sendpage = inet_sendpage,
974 .compat_setsockopt = compat_sock_common_setsockopt,
975 .compat_getsockopt = compat_sock_common_getsockopt,
976 .compat_ioctl = inet_compat_ioctl,
979 EXPORT_SYMBOL(inet_dgram_ops);
982 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
985 static const struct proto_ops inet_sockraw_ops = {
987 .owner = THIS_MODULE,
988 .release = inet_release,
990 .connect = inet_dgram_connect,
991 .socketpair = sock_no_socketpair,
992 .accept = sock_no_accept,
993 .getname = inet_getname,
994 .poll = datagram_poll,
996 .listen = sock_no_listen,
997 .shutdown = inet_shutdown,
998 .setsockopt = sock_common_setsockopt,
999 .getsockopt = sock_common_getsockopt,
1000 .sendmsg = inet_sendmsg,
1001 .recvmsg = inet_recvmsg,
1002 .mmap = sock_no_mmap,
1003 .sendpage = inet_sendpage,
1004 #ifdef CONFIG_COMPAT
1005 .compat_setsockopt = compat_sock_common_setsockopt,
1006 .compat_getsockopt = compat_sock_common_getsockopt,
1007 .compat_ioctl = inet_compat_ioctl,
1011 static const struct net_proto_family inet_family_ops = {
1013 .create = inet_create,
1014 .owner = THIS_MODULE,
1017 /* Upon startup we insert all the elements in inetsw_array[] into
1018 * the linked list inetsw.
1020 static struct inet_protosw inetsw_array[] =
1023 .type = SOCK_STREAM,
1024 .protocol = IPPROTO_TCP,
1026 .ops = &inet_stream_ops,
1028 .flags = INET_PROTOSW_PERMANENT |
1034 .protocol = IPPROTO_UDP,
1036 .ops = &inet_dgram_ops,
1037 .no_check = UDP_CSUM_DEFAULT,
1038 .flags = INET_PROTOSW_PERMANENT,
1043 .protocol = IPPROTO_ICMP,
1045 .ops = &inet_dgram_ops,
1046 .no_check = UDP_CSUM_DEFAULT,
1047 .flags = INET_PROTOSW_REUSE,
1052 .protocol = IPPROTO_IP, /* wild card */
1054 .ops = &inet_sockraw_ops,
1055 .no_check = UDP_CSUM_DEFAULT,
1056 .flags = INET_PROTOSW_REUSE,
1060 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1062 void inet_register_protosw(struct inet_protosw *p)
1064 struct list_head *lh;
1065 struct inet_protosw *answer;
1066 int protocol = p->protocol;
1067 struct list_head *last_perm;
1069 spin_lock_bh(&inetsw_lock);
1071 if (p->type >= SOCK_MAX)
1074 /* If we are trying to override a permanent protocol, bail. */
1076 last_perm = &inetsw[p->type];
1077 list_for_each(lh, &inetsw[p->type]) {
1078 answer = list_entry(lh, struct inet_protosw, list);
1080 /* Check only the non-wild match. */
1081 if (INET_PROTOSW_PERMANENT & answer->flags) {
1082 if (protocol == answer->protocol)
1092 /* Add the new entry after the last permanent entry if any, so that
1093 * the new entry does not override a permanent entry when matched with
1094 * a wild-card protocol. But it is allowed to override any existing
1095 * non-permanent entry. This means that when we remove this entry, the
1096 * system automatically returns to the old behavior.
1098 list_add_rcu(&p->list, last_perm);
1100 spin_unlock_bh(&inetsw_lock);
1105 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1111 "Ignoring attempt to register invalid socket type %d.\n",
1115 EXPORT_SYMBOL(inet_register_protosw);
1117 void inet_unregister_protosw(struct inet_protosw *p)
1119 if (INET_PROTOSW_PERMANENT & p->flags) {
1121 "Attempt to unregister permanent protocol %d.\n",
1124 spin_lock_bh(&inetsw_lock);
1125 list_del_rcu(&p->list);
1126 spin_unlock_bh(&inetsw_lock);
1131 EXPORT_SYMBOL(inet_unregister_protosw);
1134 * Shall we try to damage output packets if routing dev changes?
1137 int sysctl_ip_dynaddr __read_mostly;
1139 static int inet_sk_reselect_saddr(struct sock *sk)
1141 struct inet_sock *inet = inet_sk(sk);
1142 __be32 old_saddr = inet->inet_saddr;
1143 __be32 daddr = inet->inet_daddr;
1147 struct ip_options_rcu *inet_opt;
1149 inet_opt = rcu_dereference_protected(inet->inet_opt,
1150 sock_owned_by_user(sk));
1151 if (inet_opt && inet_opt->opt.srr)
1152 daddr = inet_opt->opt.faddr;
1154 /* Query new route. */
1155 fl4 = &inet->cork.fl.u.ip4;
1156 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1157 sk->sk_bound_dev_if, sk->sk_protocol,
1158 inet->inet_sport, inet->inet_dport, sk, false);
1162 sk_setup_caps(sk, &rt->dst);
1164 new_saddr = fl4->saddr;
1166 if (new_saddr == old_saddr)
1169 if (sysctl_ip_dynaddr > 1) {
1170 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1171 __func__, &old_saddr, &new_saddr);
1174 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1177 * XXX The only one ugly spot where we need to
1178 * XXX really change the sockets identity after
1179 * XXX it has entered the hashes. -DaveM
1181 * Besides that, it does not check for connection
1182 * uniqueness. Wait for troubles.
1184 __sk_prot_rehash(sk);
1188 int inet_sk_rebuild_header(struct sock *sk)
1190 struct inet_sock *inet = inet_sk(sk);
1191 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1193 struct ip_options_rcu *inet_opt;
1197 /* Route is OK, nothing to do. */
1203 inet_opt = rcu_dereference(inet->inet_opt);
1204 daddr = inet->inet_daddr;
1205 if (inet_opt && inet_opt->opt.srr)
1206 daddr = inet_opt->opt.faddr;
1208 fl4 = &inet->cork.fl.u.ip4;
1209 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1210 inet->inet_dport, inet->inet_sport,
1211 sk->sk_protocol, RT_CONN_FLAGS(sk),
1212 sk->sk_bound_dev_if);
1215 sk_setup_caps(sk, &rt->dst);
1219 /* Routing failed... */
1220 sk->sk_route_caps = 0;
1222 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1223 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1225 if (!sysctl_ip_dynaddr ||
1226 sk->sk_state != TCP_SYN_SENT ||
1227 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1228 (err = inet_sk_reselect_saddr(sk)) != 0)
1229 sk->sk_err_soft = -err;
1234 EXPORT_SYMBOL(inet_sk_rebuild_header);
1236 static int inet_gso_send_check(struct sk_buff *skb)
1238 const struct iphdr *iph;
1239 const struct net_protocol *ops;
1244 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1249 if (ihl < sizeof(*iph))
1252 if (unlikely(!pskb_may_pull(skb, ihl)))
1255 __skb_pull(skb, ihl);
1256 skb_reset_transport_header(skb);
1258 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1259 err = -EPROTONOSUPPORT;
1262 ops = rcu_dereference(inet_protos[proto]);
1263 if (likely(ops && ops->gso_send_check))
1264 err = ops->gso_send_check(skb);
1271 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1273 struct sk_buff *segs = ERR_PTR(-EINVAL);
1275 const struct net_protocol *ops;
1279 unsigned int offset = 0;
1281 if (!(features & NETIF_F_V4_CSUM))
1282 features &= ~NETIF_F_SG;
1284 if (unlikely(skb_shinfo(skb)->gso_type &
1292 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1297 if (ihl < sizeof(*iph))
1300 if (unlikely(!pskb_may_pull(skb, ihl)))
1303 __skb_pull(skb, ihl);
1304 skb_reset_transport_header(skb);
1306 id = ntohs(iph->id);
1307 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1308 segs = ERR_PTR(-EPROTONOSUPPORT);
1311 ops = rcu_dereference(inet_protos[proto]);
1312 if (likely(ops && ops->gso_segment))
1313 segs = ops->gso_segment(skb, features);
1316 if (!segs || IS_ERR(segs))
1322 if (proto == IPPROTO_UDP) {
1323 iph->id = htons(id);
1324 iph->frag_off = htons(offset >> 3);
1325 if (skb->next != NULL)
1326 iph->frag_off |= htons(IP_MF);
1327 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1329 iph->id = htons(id++);
1330 iph->tot_len = htons(skb->len - skb->mac_len);
1332 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1333 } while ((skb = skb->next));
1339 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1340 struct sk_buff *skb)
1342 const struct net_protocol *ops;
1343 struct sk_buff **pp = NULL;
1345 const struct iphdr *iph;
1352 off = skb_gro_offset(skb);
1353 hlen = off + sizeof(*iph);
1354 iph = skb_gro_header_fast(skb, off);
1355 if (skb_gro_header_hard(skb, hlen)) {
1356 iph = skb_gro_header_slow(skb, hlen, off);
1361 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1364 ops = rcu_dereference(inet_protos[proto]);
1365 if (!ops || !ops->gro_receive)
1368 if (*(u8 *)iph != 0x45)
1371 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1374 id = ntohl(*(__be32 *)&iph->id);
1375 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1378 for (p = *head; p; p = p->next) {
1381 if (!NAPI_GRO_CB(p)->same_flow)
1386 if ((iph->protocol ^ iph2->protocol) |
1387 (iph->tos ^ iph2->tos) |
1388 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1389 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1390 NAPI_GRO_CB(p)->same_flow = 0;
1394 /* All fields must match except length and checksum. */
1395 NAPI_GRO_CB(p)->flush |=
1396 (iph->ttl ^ iph2->ttl) |
1397 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1399 NAPI_GRO_CB(p)->flush |= flush;
1402 NAPI_GRO_CB(skb)->flush |= flush;
1403 skb_gro_pull(skb, sizeof(*iph));
1404 skb_set_transport_header(skb, skb_gro_offset(skb));
1406 pp = ops->gro_receive(head, skb);
1412 NAPI_GRO_CB(skb)->flush |= flush;
1417 static int inet_gro_complete(struct sk_buff *skb)
1419 const struct net_protocol *ops;
1420 struct iphdr *iph = ip_hdr(skb);
1421 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1423 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1425 csum_replace2(&iph->check, iph->tot_len, newlen);
1426 iph->tot_len = newlen;
1429 ops = rcu_dereference(inet_protos[proto]);
1430 if (WARN_ON(!ops || !ops->gro_complete))
1433 err = ops->gro_complete(skb);
1441 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1442 unsigned short type, unsigned char protocol,
1445 struct socket *sock;
1446 int rc = sock_create_kern(family, type, protocol, &sock);
1450 (*sk)->sk_allocation = GFP_ATOMIC;
1452 * Unhash it so that IP input processing does not even see it,
1453 * we do not wish this socket to see incoming packets.
1455 (*sk)->sk_prot->unhash(*sk);
1457 sk_change_net(*sk, net);
1461 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1463 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1465 unsigned long res = 0;
1468 for_each_possible_cpu(i) {
1469 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1470 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1474 EXPORT_SYMBOL_GPL(snmp_fold_field);
1476 #if BITS_PER_LONG==32
1478 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1483 for_each_possible_cpu(cpu) {
1484 void *bhptr, *userptr;
1485 struct u64_stats_sync *syncp;
1489 /* first mib used by softirq context, we must use _bh() accessors */
1490 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1491 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1493 start = u64_stats_fetch_begin_bh(syncp);
1494 v_bh = *(((u64 *) bhptr) + offt);
1495 } while (u64_stats_fetch_retry_bh(syncp, start));
1497 /* second mib used in USER context */
1498 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1499 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1501 start = u64_stats_fetch_begin(syncp);
1502 v_user = *(((u64 *) userptr) + offt);
1503 } while (u64_stats_fetch_retry(syncp, start));
1505 res += v_bh + v_user;
1509 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1512 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1514 BUG_ON(ptr == NULL);
1515 ptr[0] = __alloc_percpu(mibsize, align);
1518 ptr[1] = __alloc_percpu(mibsize, align);
1523 free_percpu(ptr[0]);
1528 EXPORT_SYMBOL_GPL(snmp_mib_init);
1530 void snmp_mib_free(void __percpu *ptr[2])
1532 BUG_ON(ptr == NULL);
1533 free_percpu(ptr[0]);
1534 free_percpu(ptr[1]);
1535 ptr[0] = ptr[1] = NULL;
1537 EXPORT_SYMBOL_GPL(snmp_mib_free);
1539 #ifdef CONFIG_IP_MULTICAST
1540 static const struct net_protocol igmp_protocol = {
1541 .handler = igmp_rcv,
1546 static const struct net_protocol tcp_protocol = {
1547 .handler = tcp_v4_rcv,
1548 .err_handler = tcp_v4_err,
1549 .gso_send_check = tcp_v4_gso_send_check,
1550 .gso_segment = tcp_tso_segment,
1551 .gro_receive = tcp4_gro_receive,
1552 .gro_complete = tcp4_gro_complete,
1557 static const struct net_protocol udp_protocol = {
1559 .err_handler = udp_err,
1560 .gso_send_check = udp4_ufo_send_check,
1561 .gso_segment = udp4_ufo_fragment,
1566 static const struct net_protocol icmp_protocol = {
1567 .handler = icmp_rcv,
1568 .err_handler = ping_err,
1573 static __net_init int ipv4_mib_init_net(struct net *net)
1575 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1576 sizeof(struct tcp_mib),
1577 __alignof__(struct tcp_mib)) < 0)
1579 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1580 sizeof(struct ipstats_mib),
1581 __alignof__(struct ipstats_mib)) < 0)
1583 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1584 sizeof(struct linux_mib),
1585 __alignof__(struct linux_mib)) < 0)
1587 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1588 sizeof(struct udp_mib),
1589 __alignof__(struct udp_mib)) < 0)
1591 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1592 sizeof(struct udp_mib),
1593 __alignof__(struct udp_mib)) < 0)
1594 goto err_udplite_mib;
1595 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1596 sizeof(struct icmp_mib),
1597 __alignof__(struct icmp_mib)) < 0)
1599 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1600 sizeof(struct icmpmsg_mib),
1601 __alignof__(struct icmpmsg_mib)) < 0)
1602 goto err_icmpmsg_mib;
1608 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1610 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1612 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1614 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1616 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1618 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1623 static __net_exit void ipv4_mib_exit_net(struct net *net)
1625 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1626 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1627 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1628 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1629 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1630 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1631 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1634 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1635 .init = ipv4_mib_init_net,
1636 .exit = ipv4_mib_exit_net,
1639 static int __init init_ipv4_mibs(void)
1641 return register_pernet_subsys(&ipv4_mib_ops);
1644 static int ipv4_proc_init(void);
1647 * IP protocol layer initialiser
1650 static struct packet_type ip_packet_type __read_mostly = {
1651 .type = cpu_to_be16(ETH_P_IP),
1653 .gso_send_check = inet_gso_send_check,
1654 .gso_segment = inet_gso_segment,
1655 .gro_receive = inet_gro_receive,
1656 .gro_complete = inet_gro_complete,
1659 static int __init inet_init(void)
1661 struct sk_buff *dummy_skb;
1662 struct inet_protosw *q;
1663 struct list_head *r;
1666 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1668 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1669 if (!sysctl_local_reserved_ports)
1672 rc = proto_register(&tcp_prot, 1);
1674 goto out_free_reserved_ports;
1676 rc = proto_register(&udp_prot, 1);
1678 goto out_unregister_tcp_proto;
1680 rc = proto_register(&raw_prot, 1);
1682 goto out_unregister_udp_proto;
1684 rc = proto_register(&ping_prot, 1);
1686 goto out_unregister_raw_proto;
1689 * Tell SOCKET that we are alive...
1692 (void)sock_register(&inet_family_ops);
1694 #ifdef CONFIG_SYSCTL
1695 ip_static_sysctl_init();
1699 * Add all the base protocols.
1702 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1703 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1704 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1705 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1706 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1707 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1708 #ifdef CONFIG_IP_MULTICAST
1709 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1710 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1713 /* Register the socket-side information for inet_create. */
1714 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1717 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1718 inet_register_protosw(q);
1721 * Set the ARP module up
1727 * Set the IP module up
1734 /* Setup TCP slab cache for open requests. */
1737 /* Setup UDP memory threshold */
1740 /* Add UDP-Lite (RFC 3828) */
1741 udplite4_register();
1746 * Set the ICMP layer up
1749 if (icmp_init() < 0)
1750 panic("Failed to create the ICMP control socket.\n");
1753 * Initialise the multicast router
1755 #if defined(CONFIG_IP_MROUTE)
1757 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1760 * Initialise per-cpu ipv4 mibs
1763 if (init_ipv4_mibs())
1764 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1770 dev_add_pack(&ip_packet_type);
1775 out_unregister_raw_proto:
1776 proto_unregister(&raw_prot);
1777 out_unregister_udp_proto:
1778 proto_unregister(&udp_prot);
1779 out_unregister_tcp_proto:
1780 proto_unregister(&tcp_prot);
1781 out_free_reserved_ports:
1782 kfree(sysctl_local_reserved_ports);
1786 fs_initcall(inet_init);
1788 /* ------------------------------------------------------------------------ */
1790 #ifdef CONFIG_PROC_FS
1791 static int __init ipv4_proc_init(void)
1795 if (raw_proc_init())
1797 if (tcp4_proc_init())
1799 if (udp4_proc_init())
1801 if (ping_proc_init())
1803 if (ip_misc_proc_init())
1820 #else /* CONFIG_PROC_FS */
1821 static int __init ipv4_proc_init(void)
1825 #endif /* CONFIG_PROC_FS */
1827 MODULE_ALIAS_NETPROTO(PF_INET);