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 <linux/skbuff.h>
109 #include <net/sock.h>
111 #include <net/icmp.h>
112 #include <net/ipip.h>
113 #include <net/inet_common.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #ifdef CONFIG_IP_MROUTE
117 #include <linux/mroute.h>
120 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
121 #include <linux/android_aid.h>
124 /* The inetsw table contains everything that inet_create needs to
125 * build a new socket.
127 static struct list_head inetsw[SOCK_MAX];
128 static DEFINE_SPINLOCK(inetsw_lock);
130 struct ipv4_config ipv4_config;
131 EXPORT_SYMBOL(ipv4_config);
133 /* New destruction routine */
135 void inet_sock_destruct(struct sock *sk)
137 struct inet_sock *inet = inet_sk(sk);
139 __skb_queue_purge(&sk->sk_receive_queue);
140 __skb_queue_purge(&sk->sk_error_queue);
144 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
145 pr_err("Attempt to release TCP socket in state %d %p\n",
149 if (!sock_flag(sk, SOCK_DEAD)) {
150 pr_err("Attempt to release alive inet socket %p\n", sk);
154 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
155 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
156 WARN_ON(sk->sk_wmem_queued);
157 WARN_ON(sk->sk_forward_alloc);
160 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
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
233 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
235 void build_ehash_secret(void)
239 get_random_bytes(&rnd, sizeof(rnd));
241 spin_lock_bh(&inetsw_lock);
242 if (!inet_ehash_secret)
243 inet_ehash_secret = rnd;
244 spin_unlock_bh(&inetsw_lock);
246 EXPORT_SYMBOL(build_ehash_secret);
248 static inline int inet_netns_ok(struct net *net, int protocol)
251 const struct net_protocol *ipprot;
253 if (net_eq(net, &init_net))
256 hash = protocol & (MAX_INET_PROTOS - 1);
257 ipprot = rcu_dereference(inet_protos[hash]);
262 return ipprot->netns_ok;
265 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
266 static inline int current_has_network(void)
268 return (!current_euid() || in_egroup_p(AID_INET) ||
269 in_egroup_p(AID_NET_RAW));
271 static inline int current_has_cap(int cap)
273 if (cap == CAP_NET_RAW && in_egroup_p(AID_NET_RAW))
278 static inline int current_has_network(void)
282 static inline int current_has_cap(int cap)
289 * Create an inet socket.
292 static int inet_create(struct net *net, struct socket *sock, int protocol,
296 struct inet_protosw *answer;
297 struct inet_sock *inet;
298 struct proto *answer_prot;
299 unsigned char answer_flags;
300 char answer_no_check;
301 int try_loading_module = 0;
304 if (!current_has_network())
307 if (unlikely(!inet_ehash_secret))
308 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
309 build_ehash_secret();
311 sock->state = SS_UNCONNECTED;
313 /* Look for the requested type/protocol pair. */
315 err = -ESOCKTNOSUPPORT;
317 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
320 /* Check the non-wild match. */
321 if (protocol == answer->protocol) {
322 if (protocol != IPPROTO_IP)
325 /* Check for the two wild cases. */
326 if (IPPROTO_IP == protocol) {
327 protocol = answer->protocol;
330 if (IPPROTO_IP == answer->protocol)
333 err = -EPROTONOSUPPORT;
337 if (try_loading_module < 2) {
340 * Be more specific, e.g. net-pf-2-proto-132-type-1
341 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
343 if (++try_loading_module == 1)
344 request_module("net-pf-%d-proto-%d-type-%d",
345 PF_INET, protocol, sock->type);
347 * Fall back to generic, e.g. net-pf-2-proto-132
348 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
351 request_module("net-pf-%d-proto-%d",
353 goto lookup_protocol;
359 if (sock->type == SOCK_RAW && !kern && !current_has_cap(CAP_NET_RAW))
363 if (!inet_netns_ok(net, protocol))
366 sock->ops = answer->ops;
367 answer_prot = answer->prot;
368 answer_no_check = answer->no_check;
369 answer_flags = answer->flags;
372 WARN_ON(answer_prot->slab == NULL);
375 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
380 sk->sk_no_check = answer_no_check;
381 if (INET_PROTOSW_REUSE & answer_flags)
385 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
389 if (SOCK_RAW == sock->type) {
390 inet->inet_num = protocol;
391 if (IPPROTO_RAW == protocol)
395 if (ipv4_config.no_pmtu_disc)
396 inet->pmtudisc = IP_PMTUDISC_DONT;
398 inet->pmtudisc = IP_PMTUDISC_WANT;
402 sock_init_data(sock, sk);
404 sk->sk_destruct = inet_sock_destruct;
405 sk->sk_protocol = protocol;
406 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
413 inet->mc_list = NULL;
415 sk_refcnt_debug_inc(sk);
417 if (inet->inet_num) {
418 /* It assumes that any protocol which allows
419 * the user to assign a number at socket
420 * creation time automatically
423 inet->inet_sport = htons(inet->inet_num);
424 /* Add to protocol hash chains. */
425 sk->sk_prot->hash(sk);
428 if (sk->sk_prot->init) {
429 err = sk->sk_prot->init(sk);
431 sk_common_release(sk);
442 * The peer socket should always be NULL (or else). When we call this
443 * function we are destroying the object and from then on nobody
444 * should refer to it.
446 int inet_release(struct socket *sock)
448 struct sock *sk = sock->sk;
453 sock_rps_reset_flow(sk);
455 /* Applications forget to leave groups before exiting */
456 ip_mc_drop_socket(sk);
458 /* If linger is set, we don't return until the close
459 * is complete. Otherwise we return immediately. The
460 * actually closing is done the same either way.
462 * If the close is due to the process exiting, we never
466 if (sock_flag(sk, SOCK_LINGER) &&
467 !(current->flags & PF_EXITING))
468 timeout = sk->sk_lingertime;
470 sk->sk_prot->close(sk, timeout);
474 EXPORT_SYMBOL(inet_release);
476 /* It is off by default, see below. */
477 int sysctl_ip_nonlocal_bind __read_mostly;
478 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
480 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
482 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
483 struct sock *sk = sock->sk;
484 struct inet_sock *inet = inet_sk(sk);
489 /* If the socket has its own bind function then use it. (RAW) */
490 if (sk->sk_prot->bind) {
491 err = sk->sk_prot->bind(sk, uaddr, addr_len);
495 if (addr_len < sizeof(struct sockaddr_in))
498 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
500 /* Not specified by any standard per-se, however it breaks too
501 * many applications when removed. It is unfortunate since
502 * allowing applications to make a non-local bind solves
503 * several problems with systems using dynamic addressing.
504 * (ie. your servers still start up even if your ISDN link
505 * is temporarily down)
507 err = -EADDRNOTAVAIL;
508 if (!sysctl_ip_nonlocal_bind &&
509 !(inet->freebind || inet->transparent) &&
510 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
511 chk_addr_ret != RTN_LOCAL &&
512 chk_addr_ret != RTN_MULTICAST &&
513 chk_addr_ret != RTN_BROADCAST)
516 snum = ntohs(addr->sin_port);
518 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
521 /* We keep a pair of addresses. rcv_saddr is the one
522 * used by hash lookups, and saddr is used for transmit.
524 * In the BSD API these are the same except where it
525 * would be illegal to use them (multicast/broadcast) in
526 * which case the sending device address is used.
530 /* Check these errors (active socket, double bind). */
532 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
533 goto out_release_sock;
535 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
536 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
537 inet->inet_saddr = 0; /* Use device */
539 /* Make sure we are allowed to bind here. */
540 if (sk->sk_prot->get_port(sk, snum)) {
541 inet->inet_saddr = inet->inet_rcv_saddr = 0;
543 goto out_release_sock;
546 if (inet->inet_rcv_saddr)
547 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
549 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
550 inet->inet_sport = htons(inet->inet_num);
551 inet->inet_daddr = 0;
552 inet->inet_dport = 0;
560 EXPORT_SYMBOL(inet_bind);
562 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
563 int addr_len, int flags)
565 struct sock *sk = sock->sk;
567 if (addr_len < sizeof(uaddr->sa_family))
569 if (uaddr->sa_family == AF_UNSPEC)
570 return sk->sk_prot->disconnect(sk, flags);
572 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
574 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
576 EXPORT_SYMBOL(inet_dgram_connect);
578 static long inet_wait_for_connect(struct sock *sk, long timeo)
582 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
584 /* Basic assumption: if someone sets sk->sk_err, he _must_
585 * change state of the socket from TCP_SYN_*.
586 * Connect() does not allow to get error notifications
587 * without closing the socket.
589 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
591 timeo = schedule_timeout(timeo);
593 if (signal_pending(current) || !timeo)
595 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
597 finish_wait(sk_sleep(sk), &wait);
602 * Connect to a remote host. There is regrettably still a little
603 * TCP 'magic' in here.
605 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
606 int addr_len, int flags)
608 struct sock *sk = sock->sk;
612 if (addr_len < sizeof(uaddr->sa_family))
617 if (uaddr->sa_family == AF_UNSPEC) {
618 err = sk->sk_prot->disconnect(sk, flags);
619 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
623 switch (sock->state) {
632 /* Fall out of switch with err, set for this state */
636 if (sk->sk_state != TCP_CLOSE)
639 err = sk->sk_prot->connect(sk, uaddr, addr_len);
643 sock->state = SS_CONNECTING;
645 /* Just entered SS_CONNECTING state; the only
646 * difference is that return value in non-blocking
647 * case is EINPROGRESS, rather than EALREADY.
653 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
655 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
656 /* Error code is set above */
657 if (!timeo || !inet_wait_for_connect(sk, timeo))
660 err = sock_intr_errno(timeo);
661 if (signal_pending(current))
665 /* Connection was closed by RST, timeout, ICMP error
666 * or another process disconnected us.
668 if (sk->sk_state == TCP_CLOSE)
671 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
672 * and error was received after socket entered established state.
673 * Hence, it is handled normally after connect() return successfully.
676 sock->state = SS_CONNECTED;
683 err = sock_error(sk) ? : -ECONNABORTED;
684 sock->state = SS_UNCONNECTED;
685 if (sk->sk_prot->disconnect(sk, flags))
686 sock->state = SS_DISCONNECTING;
689 EXPORT_SYMBOL(inet_stream_connect);
692 * Accept a pending connection. The TCP layer now gives BSD semantics.
695 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
697 struct sock *sk1 = sock->sk;
699 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
706 WARN_ON(!((1 << sk2->sk_state) &
707 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
709 sock_graft(sk2, newsock);
711 newsock->state = SS_CONNECTED;
717 EXPORT_SYMBOL(inet_accept);
721 * This does both peername and sockname.
723 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
724 int *uaddr_len, int peer)
726 struct sock *sk = sock->sk;
727 struct inet_sock *inet = inet_sk(sk);
728 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
730 sin->sin_family = AF_INET;
732 if (!inet->inet_dport ||
733 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
736 sin->sin_port = inet->inet_dport;
737 sin->sin_addr.s_addr = inet->inet_daddr;
739 __be32 addr = inet->inet_rcv_saddr;
741 addr = inet->inet_saddr;
742 sin->sin_port = inet->inet_sport;
743 sin->sin_addr.s_addr = addr;
745 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
746 *uaddr_len = sizeof(*sin);
749 EXPORT_SYMBOL(inet_getname);
751 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
754 struct sock *sk = sock->sk;
756 sock_rps_record_flow(sk);
758 /* We may need to bind the socket. */
759 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
763 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
765 EXPORT_SYMBOL(inet_sendmsg);
767 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
768 size_t size, int flags)
770 struct sock *sk = sock->sk;
772 sock_rps_record_flow(sk);
774 /* We may need to bind the socket. */
775 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
779 if (sk->sk_prot->sendpage)
780 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
781 return sock_no_sendpage(sock, page, offset, size, flags);
783 EXPORT_SYMBOL(inet_sendpage);
785 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
786 size_t size, int flags)
788 struct sock *sk = sock->sk;
792 sock_rps_record_flow(sk);
794 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
795 flags & ~MSG_DONTWAIT, &addr_len);
797 msg->msg_namelen = addr_len;
800 EXPORT_SYMBOL(inet_recvmsg);
802 int inet_shutdown(struct socket *sock, int how)
804 struct sock *sk = sock->sk;
807 /* This should really check to make sure
808 * the socket is a TCP socket. (WHY AC...)
810 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
813 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
817 if (sock->state == SS_CONNECTING) {
818 if ((1 << sk->sk_state) &
819 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
820 sock->state = SS_DISCONNECTING;
822 sock->state = SS_CONNECTED;
825 switch (sk->sk_state) {
828 /* Hack to wake up other listeners, who can poll for
829 POLLHUP, even on eg. unconnected UDP sockets -- RR */
831 sk->sk_shutdown |= how;
832 if (sk->sk_prot->shutdown)
833 sk->sk_prot->shutdown(sk, how);
836 /* Remaining two branches are temporary solution for missing
837 * close() in multithreaded environment. It is _not_ a good idea,
838 * but we have no choice until close() is repaired at VFS level.
841 if (!(how & RCV_SHUTDOWN))
845 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
846 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
850 /* Wake up anyone sleeping in poll. */
851 sk->sk_state_change(sk);
855 EXPORT_SYMBOL(inet_shutdown);
858 * ioctl() calls you can issue on an INET socket. Most of these are
859 * device configuration and stuff and very rarely used. Some ioctls
860 * pass on to the socket itself.
862 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
863 * loads the devconfigure module does its configuring and unloads it.
864 * There's a good 20K of config code hanging around the kernel.
867 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
869 struct sock *sk = sock->sk;
871 struct net *net = sock_net(sk);
875 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
878 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
883 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
888 err = arp_ioctl(net, cmd, (void __user *)arg);
901 err = devinet_ioctl(net, cmd, (void __user *)arg);
904 if (sk->sk_prot->ioctl)
905 err = sk->sk_prot->ioctl(sk, cmd, arg);
912 EXPORT_SYMBOL(inet_ioctl);
914 const struct proto_ops inet_stream_ops = {
916 .owner = THIS_MODULE,
917 .release = inet_release,
919 .connect = inet_stream_connect,
920 .socketpair = sock_no_socketpair,
921 .accept = inet_accept,
922 .getname = inet_getname,
925 .listen = inet_listen,
926 .shutdown = inet_shutdown,
927 .setsockopt = sock_common_setsockopt,
928 .getsockopt = sock_common_getsockopt,
929 .sendmsg = inet_sendmsg,
930 .recvmsg = inet_recvmsg,
931 .mmap = sock_no_mmap,
932 .sendpage = inet_sendpage,
933 .splice_read = tcp_splice_read,
935 .compat_setsockopt = compat_sock_common_setsockopt,
936 .compat_getsockopt = compat_sock_common_getsockopt,
939 EXPORT_SYMBOL(inet_stream_ops);
941 const struct proto_ops inet_dgram_ops = {
943 .owner = THIS_MODULE,
944 .release = inet_release,
946 .connect = inet_dgram_connect,
947 .socketpair = sock_no_socketpair,
948 .accept = sock_no_accept,
949 .getname = inet_getname,
952 .listen = sock_no_listen,
953 .shutdown = inet_shutdown,
954 .setsockopt = sock_common_setsockopt,
955 .getsockopt = sock_common_getsockopt,
956 .sendmsg = inet_sendmsg,
957 .recvmsg = inet_recvmsg,
958 .mmap = sock_no_mmap,
959 .sendpage = inet_sendpage,
961 .compat_setsockopt = compat_sock_common_setsockopt,
962 .compat_getsockopt = compat_sock_common_getsockopt,
965 EXPORT_SYMBOL(inet_dgram_ops);
968 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
971 static const struct proto_ops inet_sockraw_ops = {
973 .owner = THIS_MODULE,
974 .release = inet_release,
976 .connect = inet_dgram_connect,
977 .socketpair = sock_no_socketpair,
978 .accept = sock_no_accept,
979 .getname = inet_getname,
980 .poll = datagram_poll,
982 .listen = sock_no_listen,
983 .shutdown = inet_shutdown,
984 .setsockopt = sock_common_setsockopt,
985 .getsockopt = sock_common_getsockopt,
986 .sendmsg = inet_sendmsg,
987 .recvmsg = inet_recvmsg,
988 .mmap = sock_no_mmap,
989 .sendpage = inet_sendpage,
991 .compat_setsockopt = compat_sock_common_setsockopt,
992 .compat_getsockopt = compat_sock_common_getsockopt,
996 static const struct net_proto_family inet_family_ops = {
998 .create = inet_create,
999 .owner = THIS_MODULE,
1002 /* Upon startup we insert all the elements in inetsw_array[] into
1003 * the linked list inetsw.
1005 static struct inet_protosw inetsw_array[] =
1008 .type = SOCK_STREAM,
1009 .protocol = IPPROTO_TCP,
1011 .ops = &inet_stream_ops,
1013 .flags = INET_PROTOSW_PERMANENT |
1019 .protocol = IPPROTO_UDP,
1021 .ops = &inet_dgram_ops,
1022 .no_check = UDP_CSUM_DEFAULT,
1023 .flags = INET_PROTOSW_PERMANENT,
1029 .protocol = IPPROTO_IP, /* wild card */
1031 .ops = &inet_sockraw_ops,
1032 .no_check = UDP_CSUM_DEFAULT,
1033 .flags = INET_PROTOSW_REUSE,
1037 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1039 void inet_register_protosw(struct inet_protosw *p)
1041 struct list_head *lh;
1042 struct inet_protosw *answer;
1043 int protocol = p->protocol;
1044 struct list_head *last_perm;
1046 spin_lock_bh(&inetsw_lock);
1048 if (p->type >= SOCK_MAX)
1051 /* If we are trying to override a permanent protocol, bail. */
1053 last_perm = &inetsw[p->type];
1054 list_for_each(lh, &inetsw[p->type]) {
1055 answer = list_entry(lh, struct inet_protosw, list);
1057 /* Check only the non-wild match. */
1058 if (INET_PROTOSW_PERMANENT & answer->flags) {
1059 if (protocol == answer->protocol)
1069 /* Add the new entry after the last permanent entry if any, so that
1070 * the new entry does not override a permanent entry when matched with
1071 * a wild-card protocol. But it is allowed to override any existing
1072 * non-permanent entry. This means that when we remove this entry, the
1073 * system automatically returns to the old behavior.
1075 list_add_rcu(&p->list, last_perm);
1077 spin_unlock_bh(&inetsw_lock);
1082 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1088 "Ignoring attempt to register invalid socket type %d.\n",
1092 EXPORT_SYMBOL(inet_register_protosw);
1094 void inet_unregister_protosw(struct inet_protosw *p)
1096 if (INET_PROTOSW_PERMANENT & p->flags) {
1098 "Attempt to unregister permanent protocol %d.\n",
1101 spin_lock_bh(&inetsw_lock);
1102 list_del_rcu(&p->list);
1103 spin_unlock_bh(&inetsw_lock);
1108 EXPORT_SYMBOL(inet_unregister_protosw);
1111 * Shall we try to damage output packets if routing dev changes?
1114 int sysctl_ip_dynaddr __read_mostly;
1116 static int inet_sk_reselect_saddr(struct sock *sk)
1118 struct inet_sock *inet = inet_sk(sk);
1121 __be32 old_saddr = inet->inet_saddr;
1123 __be32 daddr = inet->inet_daddr;
1125 if (inet->opt && inet->opt->srr)
1126 daddr = inet->opt->faddr;
1128 /* Query new route. */
1129 err = ip_route_connect(&rt, daddr, 0,
1131 sk->sk_bound_dev_if,
1133 inet->inet_sport, inet->inet_dport, sk, 0);
1137 sk_setup_caps(sk, &rt->dst);
1139 new_saddr = rt->rt_src;
1141 if (new_saddr == old_saddr)
1144 if (sysctl_ip_dynaddr > 1) {
1145 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1146 __func__, &old_saddr, &new_saddr);
1149 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1152 * XXX The only one ugly spot where we need to
1153 * XXX really change the sockets identity after
1154 * XXX it has entered the hashes. -DaveM
1156 * Besides that, it does not check for connection
1157 * uniqueness. Wait for troubles.
1159 __sk_prot_rehash(sk);
1163 int inet_sk_rebuild_header(struct sock *sk)
1165 struct inet_sock *inet = inet_sk(sk);
1166 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1170 /* Route is OK, nothing to do. */
1175 daddr = inet->inet_daddr;
1176 if (inet->opt && inet->opt->srr)
1177 daddr = inet->opt->faddr;
1180 .oif = sk->sk_bound_dev_if,
1181 .mark = sk->sk_mark,
1185 .saddr = inet->inet_saddr,
1186 .tos = RT_CONN_FLAGS(sk),
1189 .proto = sk->sk_protocol,
1190 .flags = inet_sk_flowi_flags(sk),
1193 .sport = inet->inet_sport,
1194 .dport = inet->inet_dport,
1199 security_sk_classify_flow(sk, &fl);
1200 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1203 sk_setup_caps(sk, &rt->dst);
1205 /* Routing failed... */
1206 sk->sk_route_caps = 0;
1208 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1209 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1211 if (!sysctl_ip_dynaddr ||
1212 sk->sk_state != TCP_SYN_SENT ||
1213 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1214 (err = inet_sk_reselect_saddr(sk)) != 0)
1215 sk->sk_err_soft = -err;
1220 EXPORT_SYMBOL(inet_sk_rebuild_header);
1222 static int inet_gso_send_check(struct sk_buff *skb)
1225 const struct net_protocol *ops;
1230 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1235 if (ihl < sizeof(*iph))
1238 if (unlikely(!pskb_may_pull(skb, ihl)))
1241 __skb_pull(skb, ihl);
1242 skb_reset_transport_header(skb);
1244 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1245 err = -EPROTONOSUPPORT;
1248 ops = rcu_dereference(inet_protos[proto]);
1249 if (likely(ops && ops->gso_send_check))
1250 err = ops->gso_send_check(skb);
1257 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1259 struct sk_buff *segs = ERR_PTR(-EINVAL);
1261 const struct net_protocol *ops;
1265 unsigned int offset = 0;
1267 if (!(features & NETIF_F_V4_CSUM))
1268 features &= ~NETIF_F_SG;
1270 if (unlikely(skb_shinfo(skb)->gso_type &
1278 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1283 if (ihl < sizeof(*iph))
1286 if (unlikely(!pskb_may_pull(skb, ihl)))
1289 __skb_pull(skb, ihl);
1290 skb_reset_transport_header(skb);
1292 id = ntohs(iph->id);
1293 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1294 segs = ERR_PTR(-EPROTONOSUPPORT);
1297 ops = rcu_dereference(inet_protos[proto]);
1298 if (likely(ops && ops->gso_segment))
1299 segs = ops->gso_segment(skb, features);
1302 if (!segs || IS_ERR(segs))
1308 if (proto == IPPROTO_UDP) {
1309 iph->id = htons(id);
1310 iph->frag_off = htons(offset >> 3);
1311 if (skb->next != NULL)
1312 iph->frag_off |= htons(IP_MF);
1313 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1315 iph->id = htons(id++);
1316 iph->tot_len = htons(skb->len - skb->mac_len);
1318 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1319 } while ((skb = skb->next));
1325 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1326 struct sk_buff *skb)
1328 const struct net_protocol *ops;
1329 struct sk_buff **pp = NULL;
1338 off = skb_gro_offset(skb);
1339 hlen = off + sizeof(*iph);
1340 iph = skb_gro_header_fast(skb, off);
1341 if (skb_gro_header_hard(skb, hlen)) {
1342 iph = skb_gro_header_slow(skb, hlen, off);
1347 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1350 ops = rcu_dereference(inet_protos[proto]);
1351 if (!ops || !ops->gro_receive)
1354 if (*(u8 *)iph != 0x45)
1357 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1360 id = ntohl(*(__be32 *)&iph->id);
1361 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1364 for (p = *head; p; p = p->next) {
1367 if (!NAPI_GRO_CB(p)->same_flow)
1372 if ((iph->protocol ^ iph2->protocol) |
1373 (iph->tos ^ iph2->tos) |
1374 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1375 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1376 NAPI_GRO_CB(p)->same_flow = 0;
1380 /* All fields must match except length and checksum. */
1381 NAPI_GRO_CB(p)->flush |=
1382 (iph->ttl ^ iph2->ttl) |
1383 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1385 NAPI_GRO_CB(p)->flush |= flush;
1388 NAPI_GRO_CB(skb)->flush |= flush;
1389 skb_gro_pull(skb, sizeof(*iph));
1390 skb_set_transport_header(skb, skb_gro_offset(skb));
1392 pp = ops->gro_receive(head, skb);
1398 NAPI_GRO_CB(skb)->flush |= flush;
1403 static int inet_gro_complete(struct sk_buff *skb)
1405 const struct net_protocol *ops;
1406 struct iphdr *iph = ip_hdr(skb);
1407 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1409 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1411 csum_replace2(&iph->check, iph->tot_len, newlen);
1412 iph->tot_len = newlen;
1415 ops = rcu_dereference(inet_protos[proto]);
1416 if (WARN_ON(!ops || !ops->gro_complete))
1419 err = ops->gro_complete(skb);
1427 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1428 unsigned short type, unsigned char protocol,
1431 struct socket *sock;
1432 int rc = sock_create_kern(family, type, protocol, &sock);
1436 (*sk)->sk_allocation = GFP_ATOMIC;
1438 * Unhash it so that IP input processing does not even see it,
1439 * we do not wish this socket to see incoming packets.
1441 (*sk)->sk_prot->unhash(*sk);
1443 sk_change_net(*sk, net);
1447 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1449 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1451 unsigned long res = 0;
1454 for_each_possible_cpu(i) {
1455 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1456 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1460 EXPORT_SYMBOL_GPL(snmp_fold_field);
1462 #if BITS_PER_LONG==32
1464 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1469 for_each_possible_cpu(cpu) {
1470 void *bhptr, *userptr;
1471 struct u64_stats_sync *syncp;
1475 /* first mib used by softirq context, we must use _bh() accessors */
1476 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1477 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1479 start = u64_stats_fetch_begin_bh(syncp);
1480 v_bh = *(((u64 *) bhptr) + offt);
1481 } while (u64_stats_fetch_retry_bh(syncp, start));
1483 /* second mib used in USER context */
1484 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1485 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1487 start = u64_stats_fetch_begin(syncp);
1488 v_user = *(((u64 *) userptr) + offt);
1489 } while (u64_stats_fetch_retry(syncp, start));
1491 res += v_bh + v_user;
1495 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1498 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1500 BUG_ON(ptr == NULL);
1501 ptr[0] = __alloc_percpu(mibsize, align);
1504 ptr[1] = __alloc_percpu(mibsize, align);
1509 free_percpu(ptr[0]);
1514 EXPORT_SYMBOL_GPL(snmp_mib_init);
1516 void snmp_mib_free(void __percpu *ptr[2])
1518 BUG_ON(ptr == NULL);
1519 free_percpu(ptr[0]);
1520 free_percpu(ptr[1]);
1521 ptr[0] = ptr[1] = NULL;
1523 EXPORT_SYMBOL_GPL(snmp_mib_free);
1525 #ifdef CONFIG_IP_MULTICAST
1526 static const struct net_protocol igmp_protocol = {
1527 .handler = igmp_rcv,
1532 static const struct net_protocol tcp_protocol = {
1533 .handler = tcp_v4_rcv,
1534 .err_handler = tcp_v4_err,
1535 .gso_send_check = tcp_v4_gso_send_check,
1536 .gso_segment = tcp_tso_segment,
1537 .gro_receive = tcp4_gro_receive,
1538 .gro_complete = tcp4_gro_complete,
1543 static const struct net_protocol udp_protocol = {
1545 .err_handler = udp_err,
1546 .gso_send_check = udp4_ufo_send_check,
1547 .gso_segment = udp4_ufo_fragment,
1552 static const struct net_protocol icmp_protocol = {
1553 .handler = icmp_rcv,
1558 static __net_init int ipv4_mib_init_net(struct net *net)
1560 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1561 sizeof(struct tcp_mib),
1562 __alignof__(struct tcp_mib)) < 0)
1564 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1565 sizeof(struct ipstats_mib),
1566 __alignof__(struct ipstats_mib)) < 0)
1568 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1569 sizeof(struct linux_mib),
1570 __alignof__(struct linux_mib)) < 0)
1572 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1573 sizeof(struct udp_mib),
1574 __alignof__(struct udp_mib)) < 0)
1576 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1577 sizeof(struct udp_mib),
1578 __alignof__(struct udp_mib)) < 0)
1579 goto err_udplite_mib;
1580 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1581 sizeof(struct icmp_mib),
1582 __alignof__(struct icmp_mib)) < 0)
1584 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1585 sizeof(struct icmpmsg_mib),
1586 __alignof__(struct icmpmsg_mib)) < 0)
1587 goto err_icmpmsg_mib;
1593 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1595 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1597 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1599 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1601 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1603 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1608 static __net_exit void ipv4_mib_exit_net(struct net *net)
1610 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1611 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1612 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1613 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1614 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1615 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1616 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1619 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1620 .init = ipv4_mib_init_net,
1621 .exit = ipv4_mib_exit_net,
1624 static int __init init_ipv4_mibs(void)
1626 return register_pernet_subsys(&ipv4_mib_ops);
1629 static int ipv4_proc_init(void);
1632 * IP protocol layer initialiser
1635 static struct packet_type ip_packet_type __read_mostly = {
1636 .type = cpu_to_be16(ETH_P_IP),
1638 .gso_send_check = inet_gso_send_check,
1639 .gso_segment = inet_gso_segment,
1640 .gro_receive = inet_gro_receive,
1641 .gro_complete = inet_gro_complete,
1644 static int __init inet_init(void)
1646 struct sk_buff *dummy_skb;
1647 struct inet_protosw *q;
1648 struct list_head *r;
1651 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1653 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1654 if (!sysctl_local_reserved_ports)
1657 rc = proto_register(&tcp_prot, 1);
1659 goto out_free_reserved_ports;
1661 rc = proto_register(&udp_prot, 1);
1663 goto out_unregister_tcp_proto;
1665 rc = proto_register(&raw_prot, 1);
1667 goto out_unregister_udp_proto;
1670 * Tell SOCKET that we are alive...
1673 (void)sock_register(&inet_family_ops);
1675 #ifdef CONFIG_SYSCTL
1676 ip_static_sysctl_init();
1680 * Add all the base protocols.
1683 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1684 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1685 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1686 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1687 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1688 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1689 #ifdef CONFIG_IP_MULTICAST
1690 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1691 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1694 /* Register the socket-side information for inet_create. */
1695 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1698 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1699 inet_register_protosw(q);
1702 * Set the ARP module up
1708 * Set the IP module up
1715 /* Setup TCP slab cache for open requests. */
1718 /* Setup UDP memory threshold */
1721 /* Add UDP-Lite (RFC 3828) */
1722 udplite4_register();
1725 * Set the ICMP layer up
1728 if (icmp_init() < 0)
1729 panic("Failed to create the ICMP control socket.\n");
1732 * Initialise the multicast router
1734 #if defined(CONFIG_IP_MROUTE)
1736 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1739 * Initialise per-cpu ipv4 mibs
1742 if (init_ipv4_mibs())
1743 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1749 dev_add_pack(&ip_packet_type);
1754 out_unregister_udp_proto:
1755 proto_unregister(&udp_prot);
1756 out_unregister_tcp_proto:
1757 proto_unregister(&tcp_prot);
1758 out_free_reserved_ports:
1759 kfree(sysctl_local_reserved_ports);
1763 fs_initcall(inet_init);
1765 /* ------------------------------------------------------------------------ */
1767 #ifdef CONFIG_PROC_FS
1768 static int __init ipv4_proc_init(void)
1772 if (raw_proc_init())
1774 if (tcp4_proc_init())
1776 if (udp4_proc_init())
1778 if (ip_misc_proc_init())
1793 #else /* CONFIG_PROC_FS */
1794 static int __init ipv4_proc_init(void)
1798 #endif /* CONFIG_PROC_FS */
1800 MODULE_ALIAS_NETPROTO(PF_INET);