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.
13 * Changes (see also sock.c)
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
68 #define pr_fmt(fmt) "IPv4: " fmt
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
93 #include <asm/uaccess.h>
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
101 #include <net/protocol.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
113 #include <net/icmp.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #include <net/secure_seq.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
129 /* New destruction routine */
131 void inet_sock_destruct(struct sock *sk)
133 struct inet_sock *inet = inet_sk(sk);
135 __skb_queue_purge(&sk->sk_receive_queue);
136 __skb_queue_purge(&sk->sk_error_queue);
140 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
141 pr_err("Attempt to release TCP socket in state %d %p\n",
145 if (!sock_flag(sk, SOCK_DEAD)) {
146 pr_err("Attempt to release alive inet socket %p\n", sk);
150 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
151 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
152 WARN_ON(sk->sk_wmem_queued);
153 WARN_ON(sk->sk_forward_alloc);
155 kfree(rcu_dereference_protected(inet->inet_opt, 1));
156 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
157 dst_release(sk->sk_rx_dst);
158 sk_refcnt_debug_dec(sk);
160 EXPORT_SYMBOL(inet_sock_destruct);
163 * The routines beyond this point handle the behaviour of an AF_INET
164 * socket object. Mostly it punts to the subprotocols of IP to do
169 * Automatically bind an unbound socket.
172 static int inet_autobind(struct sock *sk)
174 struct inet_sock *inet;
175 /* We may need to bind the socket. */
178 if (!inet->inet_num) {
179 if (sk->sk_prot->get_port(sk, 0)) {
183 inet->inet_sport = htons(inet->inet_num);
190 * Move a socket into listening state.
192 int inet_listen(struct socket *sock, int backlog)
194 struct sock *sk = sock->sk;
195 unsigned char old_state;
201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 old_state = sk->sk_state;
205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 /* Really, if the socket is already in listen state
209 * we can only allow the backlog to be adjusted.
211 if (old_state != TCP_LISTEN) {
212 /* Check special setups for testing purpose to enable TFO w/o
213 * requiring TCP_FASTOPEN sockopt.
214 * Note that only TCP sockets (SOCK_STREAM) will reach here.
215 * Also fastopenq may already been allocated because this
216 * socket was in TCP_LISTEN state previously but was
217 * shutdown() (rather than close()).
219 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
220 !inet_csk(sk)->icsk_accept_queue.fastopenq) {
221 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
222 err = fastopen_init_queue(sk, backlog);
223 else if ((sysctl_tcp_fastopen &
224 TFO_SERVER_WO_SOCKOPT2) != 0)
225 err = fastopen_init_queue(sk,
226 ((uint)sysctl_tcp_fastopen) >> 16);
232 err = inet_csk_listen_start(sk, backlog);
236 sk->sk_max_ack_backlog = backlog;
243 EXPORT_SYMBOL(inet_listen);
246 * Create an inet socket.
249 static int inet_create(struct net *net, struct socket *sock, int protocol,
253 struct inet_protosw *answer;
254 struct inet_sock *inet;
255 struct proto *answer_prot;
256 unsigned char answer_flags;
257 int try_loading_module = 0;
260 sock->state = SS_UNCONNECTED;
262 /* Look for the requested type/protocol pair. */
264 err = -ESOCKTNOSUPPORT;
266 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
269 /* Check the non-wild match. */
270 if (protocol == answer->protocol) {
271 if (protocol != IPPROTO_IP)
274 /* Check for the two wild cases. */
275 if (IPPROTO_IP == protocol) {
276 protocol = answer->protocol;
279 if (IPPROTO_IP == answer->protocol)
282 err = -EPROTONOSUPPORT;
286 if (try_loading_module < 2) {
289 * Be more specific, e.g. net-pf-2-proto-132-type-1
290 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
292 if (++try_loading_module == 1)
293 request_module("net-pf-%d-proto-%d-type-%d",
294 PF_INET, protocol, sock->type);
296 * Fall back to generic, e.g. net-pf-2-proto-132
297 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
300 request_module("net-pf-%d-proto-%d",
302 goto lookup_protocol;
308 if (sock->type == SOCK_RAW && !kern &&
309 !ns_capable(net->user_ns, CAP_NET_RAW))
312 sock->ops = answer->ops;
313 answer_prot = answer->prot;
314 answer_flags = answer->flags;
317 WARN_ON(!answer_prot->slab);
320 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
325 if (INET_PROTOSW_REUSE & answer_flags)
326 sk->sk_reuse = SK_CAN_REUSE;
329 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
333 if (SOCK_RAW == sock->type) {
334 inet->inet_num = protocol;
335 if (IPPROTO_RAW == protocol)
339 if (net->ipv4.sysctl_ip_no_pmtu_disc)
340 inet->pmtudisc = IP_PMTUDISC_DONT;
342 inet->pmtudisc = IP_PMTUDISC_WANT;
346 sock_init_data(sock, sk);
348 sk->sk_destruct = inet_sock_destruct;
349 sk->sk_protocol = protocol;
350 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
357 inet->mc_list = NULL;
360 sk_refcnt_debug_inc(sk);
362 if (inet->inet_num) {
363 /* It assumes that any protocol which allows
364 * the user to assign a number at socket
365 * creation time automatically
368 inet->inet_sport = htons(inet->inet_num);
369 /* Add to protocol hash chains. */
370 sk->sk_prot->hash(sk);
373 if (sk->sk_prot->init) {
374 err = sk->sk_prot->init(sk);
376 sk_common_release(sk);
387 * The peer socket should always be NULL (or else). When we call this
388 * function we are destroying the object and from then on nobody
389 * should refer to it.
391 int inet_release(struct socket *sock)
393 struct sock *sk = sock->sk;
398 /* Applications forget to leave groups before exiting */
399 ip_mc_drop_socket(sk);
401 /* If linger is set, we don't return until the close
402 * is complete. Otherwise we return immediately. The
403 * actually closing is done the same either way.
405 * If the close is due to the process exiting, we never
409 if (sock_flag(sk, SOCK_LINGER) &&
410 !(current->flags & PF_EXITING))
411 timeout = sk->sk_lingertime;
413 sk->sk_prot->close(sk, timeout);
417 EXPORT_SYMBOL(inet_release);
419 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
421 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
422 struct sock *sk = sock->sk;
423 struct inet_sock *inet = inet_sk(sk);
424 struct net *net = sock_net(sk);
429 /* If the socket has its own bind function then use it. (RAW) */
430 if (sk->sk_prot->bind) {
431 err = sk->sk_prot->bind(sk, uaddr, addr_len);
435 if (addr_len < sizeof(struct sockaddr_in))
438 if (addr->sin_family != AF_INET) {
439 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
440 * only if s_addr is INADDR_ANY.
443 if (addr->sin_family != AF_UNSPEC ||
444 addr->sin_addr.s_addr != htonl(INADDR_ANY))
448 chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
450 /* Not specified by any standard per-se, however it breaks too
451 * many applications when removed. It is unfortunate since
452 * allowing applications to make a non-local bind solves
453 * several problems with systems using dynamic addressing.
454 * (ie. your servers still start up even if your ISDN link
455 * is temporarily down)
457 err = -EADDRNOTAVAIL;
458 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
459 !(inet->freebind || inet->transparent) &&
460 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
461 chk_addr_ret != RTN_LOCAL &&
462 chk_addr_ret != RTN_MULTICAST &&
463 chk_addr_ret != RTN_BROADCAST)
466 snum = ntohs(addr->sin_port);
468 if (snum && snum < PROT_SOCK &&
469 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
472 /* We keep a pair of addresses. rcv_saddr is the one
473 * used by hash lookups, and saddr is used for transmit.
475 * In the BSD API these are the same except where it
476 * would be illegal to use them (multicast/broadcast) in
477 * which case the sending device address is used.
481 /* Check these errors (active socket, double bind). */
483 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
484 goto out_release_sock;
486 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
487 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
488 inet->inet_saddr = 0; /* Use device */
490 /* Make sure we are allowed to bind here. */
491 if (sk->sk_prot->get_port(sk, snum)) {
492 inet->inet_saddr = inet->inet_rcv_saddr = 0;
494 goto out_release_sock;
497 if (inet->inet_rcv_saddr)
498 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
500 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
501 inet->inet_sport = htons(inet->inet_num);
502 inet->inet_daddr = 0;
503 inet->inet_dport = 0;
511 EXPORT_SYMBOL(inet_bind);
513 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
514 int addr_len, int flags)
516 struct sock *sk = sock->sk;
518 if (addr_len < sizeof(uaddr->sa_family))
520 if (uaddr->sa_family == AF_UNSPEC)
521 return sk->sk_prot->disconnect(sk, flags);
523 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
525 return sk->sk_prot->connect(sk, uaddr, addr_len);
527 EXPORT_SYMBOL(inet_dgram_connect);
529 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
533 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
534 sk->sk_write_pending += writebias;
536 /* Basic assumption: if someone sets sk->sk_err, he _must_
537 * change state of the socket from TCP_SYN_*.
538 * Connect() does not allow to get error notifications
539 * without closing the socket.
541 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
543 timeo = schedule_timeout(timeo);
545 if (signal_pending(current) || !timeo)
547 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
549 finish_wait(sk_sleep(sk), &wait);
550 sk->sk_write_pending -= writebias;
555 * Connect to a remote host. There is regrettably still a little
556 * TCP 'magic' in here.
558 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
559 int addr_len, int flags)
561 struct sock *sk = sock->sk;
565 if (addr_len < sizeof(uaddr->sa_family))
568 if (uaddr->sa_family == AF_UNSPEC) {
569 err = sk->sk_prot->disconnect(sk, flags);
570 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
574 switch (sock->state) {
583 /* Fall out of switch with err, set for this state */
587 if (sk->sk_state != TCP_CLOSE)
590 err = sk->sk_prot->connect(sk, uaddr, addr_len);
594 sock->state = SS_CONNECTING;
596 /* Just entered SS_CONNECTING state; the only
597 * difference is that return value in non-blocking
598 * case is EINPROGRESS, rather than EALREADY.
604 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
606 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
607 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
608 tcp_sk(sk)->fastopen_req &&
609 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
611 /* Error code is set above */
612 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
615 err = sock_intr_errno(timeo);
616 if (signal_pending(current))
620 /* Connection was closed by RST, timeout, ICMP error
621 * or another process disconnected us.
623 if (sk->sk_state == TCP_CLOSE)
626 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
627 * and error was received after socket entered established state.
628 * Hence, it is handled normally after connect() return successfully.
631 sock->state = SS_CONNECTED;
637 err = sock_error(sk) ? : -ECONNABORTED;
638 sock->state = SS_UNCONNECTED;
639 if (sk->sk_prot->disconnect(sk, flags))
640 sock->state = SS_DISCONNECTING;
643 EXPORT_SYMBOL(__inet_stream_connect);
645 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
646 int addr_len, int flags)
651 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
652 release_sock(sock->sk);
655 EXPORT_SYMBOL(inet_stream_connect);
658 * Accept a pending connection. The TCP layer now gives BSD semantics.
661 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
663 struct sock *sk1 = sock->sk;
665 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
672 sock_rps_record_flow(sk2);
673 WARN_ON(!((1 << sk2->sk_state) &
674 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
675 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
677 sock_graft(sk2, newsock);
679 newsock->state = SS_CONNECTED;
685 EXPORT_SYMBOL(inet_accept);
689 * This does both peername and sockname.
691 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
692 int *uaddr_len, int peer)
694 struct sock *sk = sock->sk;
695 struct inet_sock *inet = inet_sk(sk);
696 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
698 sin->sin_family = AF_INET;
700 if (!inet->inet_dport ||
701 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
704 sin->sin_port = inet->inet_dport;
705 sin->sin_addr.s_addr = inet->inet_daddr;
707 __be32 addr = inet->inet_rcv_saddr;
709 addr = inet->inet_saddr;
710 sin->sin_port = inet->inet_sport;
711 sin->sin_addr.s_addr = addr;
713 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
714 *uaddr_len = sizeof(*sin);
717 EXPORT_SYMBOL(inet_getname);
719 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
721 struct sock *sk = sock->sk;
723 sock_rps_record_flow(sk);
725 /* We may need to bind the socket. */
726 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
730 return sk->sk_prot->sendmsg(sk, msg, size);
732 EXPORT_SYMBOL(inet_sendmsg);
734 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
735 size_t size, int flags)
737 struct sock *sk = sock->sk;
739 sock_rps_record_flow(sk);
741 /* We may need to bind the socket. */
742 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
746 if (sk->sk_prot->sendpage)
747 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
748 return sock_no_sendpage(sock, page, offset, size, flags);
750 EXPORT_SYMBOL(inet_sendpage);
752 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
755 struct sock *sk = sock->sk;
759 sock_rps_record_flow(sk);
761 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
762 flags & ~MSG_DONTWAIT, &addr_len);
764 msg->msg_namelen = addr_len;
767 EXPORT_SYMBOL(inet_recvmsg);
769 int inet_shutdown(struct socket *sock, int how)
771 struct sock *sk = sock->sk;
774 /* This should really check to make sure
775 * the socket is a TCP socket. (WHY AC...)
777 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
780 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
784 if (sock->state == SS_CONNECTING) {
785 if ((1 << sk->sk_state) &
786 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
787 sock->state = SS_DISCONNECTING;
789 sock->state = SS_CONNECTED;
792 switch (sk->sk_state) {
795 /* Hack to wake up other listeners, who can poll for
796 POLLHUP, even on eg. unconnected UDP sockets -- RR */
798 sk->sk_shutdown |= how;
799 if (sk->sk_prot->shutdown)
800 sk->sk_prot->shutdown(sk, how);
803 /* Remaining two branches are temporary solution for missing
804 * close() in multithreaded environment. It is _not_ a good idea,
805 * but we have no choice until close() is repaired at VFS level.
808 if (!(how & RCV_SHUTDOWN))
812 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
813 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
817 /* Wake up anyone sleeping in poll. */
818 sk->sk_state_change(sk);
822 EXPORT_SYMBOL(inet_shutdown);
825 * ioctl() calls you can issue on an INET socket. Most of these are
826 * device configuration and stuff and very rarely used. Some ioctls
827 * pass on to the socket itself.
829 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
830 * loads the devconfigure module does its configuring and unloads it.
831 * There's a good 20K of config code hanging around the kernel.
834 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
836 struct sock *sk = sock->sk;
838 struct net *net = sock_net(sk);
842 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
845 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
850 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
855 err = arp_ioctl(net, cmd, (void __user *)arg);
868 err = devinet_ioctl(net, cmd, (void __user *)arg);
871 if (sk->sk_prot->ioctl)
872 err = sk->sk_prot->ioctl(sk, cmd, arg);
879 EXPORT_SYMBOL(inet_ioctl);
882 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
884 struct sock *sk = sock->sk;
885 int err = -ENOIOCTLCMD;
887 if (sk->sk_prot->compat_ioctl)
888 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
894 const struct proto_ops inet_stream_ops = {
896 .owner = THIS_MODULE,
897 .release = inet_release,
899 .connect = inet_stream_connect,
900 .socketpair = sock_no_socketpair,
901 .accept = inet_accept,
902 .getname = inet_getname,
905 .listen = inet_listen,
906 .shutdown = inet_shutdown,
907 .setsockopt = sock_common_setsockopt,
908 .getsockopt = sock_common_getsockopt,
909 .sendmsg = inet_sendmsg,
910 .recvmsg = inet_recvmsg,
911 .mmap = sock_no_mmap,
912 .sendpage = inet_sendpage,
913 .splice_read = tcp_splice_read,
915 .compat_setsockopt = compat_sock_common_setsockopt,
916 .compat_getsockopt = compat_sock_common_getsockopt,
917 .compat_ioctl = inet_compat_ioctl,
920 EXPORT_SYMBOL(inet_stream_ops);
922 const struct proto_ops inet_dgram_ops = {
924 .owner = THIS_MODULE,
925 .release = inet_release,
927 .connect = inet_dgram_connect,
928 .socketpair = sock_no_socketpair,
929 .accept = sock_no_accept,
930 .getname = inet_getname,
933 .listen = sock_no_listen,
934 .shutdown = inet_shutdown,
935 .setsockopt = sock_common_setsockopt,
936 .getsockopt = sock_common_getsockopt,
937 .sendmsg = inet_sendmsg,
938 .recvmsg = inet_recvmsg,
939 .mmap = sock_no_mmap,
940 .sendpage = inet_sendpage,
942 .compat_setsockopt = compat_sock_common_setsockopt,
943 .compat_getsockopt = compat_sock_common_getsockopt,
944 .compat_ioctl = inet_compat_ioctl,
947 EXPORT_SYMBOL(inet_dgram_ops);
950 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
953 static const struct proto_ops inet_sockraw_ops = {
955 .owner = THIS_MODULE,
956 .release = inet_release,
958 .connect = inet_dgram_connect,
959 .socketpair = sock_no_socketpair,
960 .accept = sock_no_accept,
961 .getname = inet_getname,
962 .poll = datagram_poll,
964 .listen = sock_no_listen,
965 .shutdown = inet_shutdown,
966 .setsockopt = sock_common_setsockopt,
967 .getsockopt = sock_common_getsockopt,
968 .sendmsg = inet_sendmsg,
969 .recvmsg = inet_recvmsg,
970 .mmap = sock_no_mmap,
971 .sendpage = inet_sendpage,
973 .compat_setsockopt = compat_sock_common_setsockopt,
974 .compat_getsockopt = compat_sock_common_getsockopt,
975 .compat_ioctl = inet_compat_ioctl,
979 static const struct net_proto_family inet_family_ops = {
981 .create = inet_create,
982 .owner = THIS_MODULE,
985 /* Upon startup we insert all the elements in inetsw_array[] into
986 * the linked list inetsw.
988 static struct inet_protosw inetsw_array[] =
992 .protocol = IPPROTO_TCP,
994 .ops = &inet_stream_ops,
995 .flags = INET_PROTOSW_PERMANENT |
1001 .protocol = IPPROTO_UDP,
1003 .ops = &inet_dgram_ops,
1004 .flags = INET_PROTOSW_PERMANENT,
1009 .protocol = IPPROTO_ICMP,
1011 .ops = &inet_dgram_ops,
1012 .flags = INET_PROTOSW_REUSE,
1017 .protocol = IPPROTO_IP, /* wild card */
1019 .ops = &inet_sockraw_ops,
1020 .flags = INET_PROTOSW_REUSE,
1024 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1026 void inet_register_protosw(struct inet_protosw *p)
1028 struct list_head *lh;
1029 struct inet_protosw *answer;
1030 int protocol = p->protocol;
1031 struct list_head *last_perm;
1033 spin_lock_bh(&inetsw_lock);
1035 if (p->type >= SOCK_MAX)
1038 /* If we are trying to override a permanent protocol, bail. */
1040 last_perm = &inetsw[p->type];
1041 list_for_each(lh, &inetsw[p->type]) {
1042 answer = list_entry(lh, struct inet_protosw, list);
1044 /* Check only the non-wild match. */
1045 if (INET_PROTOSW_PERMANENT & answer->flags) {
1046 if (protocol == answer->protocol)
1056 /* Add the new entry after the last permanent entry if any, so that
1057 * the new entry does not override a permanent entry when matched with
1058 * a wild-card protocol. But it is allowed to override any existing
1059 * non-permanent entry. This means that when we remove this entry, the
1060 * system automatically returns to the old behavior.
1062 list_add_rcu(&p->list, last_perm);
1064 spin_unlock_bh(&inetsw_lock);
1069 pr_err("Attempt to override permanent protocol %d\n", protocol);
1073 pr_err("Ignoring attempt to register invalid socket type %d\n",
1077 EXPORT_SYMBOL(inet_register_protosw);
1079 void inet_unregister_protosw(struct inet_protosw *p)
1081 if (INET_PROTOSW_PERMANENT & p->flags) {
1082 pr_err("Attempt to unregister permanent protocol %d\n",
1085 spin_lock_bh(&inetsw_lock);
1086 list_del_rcu(&p->list);
1087 spin_unlock_bh(&inetsw_lock);
1092 EXPORT_SYMBOL(inet_unregister_protosw);
1095 * Shall we try to damage output packets if routing dev changes?
1098 int sysctl_ip_dynaddr __read_mostly;
1100 static int inet_sk_reselect_saddr(struct sock *sk)
1102 struct inet_sock *inet = inet_sk(sk);
1103 __be32 old_saddr = inet->inet_saddr;
1104 __be32 daddr = inet->inet_daddr;
1108 struct ip_options_rcu *inet_opt;
1110 inet_opt = rcu_dereference_protected(inet->inet_opt,
1111 sock_owned_by_user(sk));
1112 if (inet_opt && inet_opt->opt.srr)
1113 daddr = inet_opt->opt.faddr;
1115 /* Query new route. */
1116 fl4 = &inet->cork.fl.u.ip4;
1117 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1118 sk->sk_bound_dev_if, sk->sk_protocol,
1119 inet->inet_sport, inet->inet_dport, sk);
1123 sk_setup_caps(sk, &rt->dst);
1125 new_saddr = fl4->saddr;
1127 if (new_saddr == old_saddr)
1130 if (sysctl_ip_dynaddr > 1) {
1131 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1132 __func__, &old_saddr, &new_saddr);
1135 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1138 * XXX The only one ugly spot where we need to
1139 * XXX really change the sockets identity after
1140 * XXX it has entered the hashes. -DaveM
1142 * Besides that, it does not check for connection
1143 * uniqueness. Wait for troubles.
1145 __sk_prot_rehash(sk);
1149 int inet_sk_rebuild_header(struct sock *sk)
1151 struct inet_sock *inet = inet_sk(sk);
1152 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1154 struct ip_options_rcu *inet_opt;
1158 /* Route is OK, nothing to do. */
1164 inet_opt = rcu_dereference(inet->inet_opt);
1165 daddr = inet->inet_daddr;
1166 if (inet_opt && inet_opt->opt.srr)
1167 daddr = inet_opt->opt.faddr;
1169 fl4 = &inet->cork.fl.u.ip4;
1170 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1171 inet->inet_dport, inet->inet_sport,
1172 sk->sk_protocol, RT_CONN_FLAGS(sk),
1173 sk->sk_bound_dev_if);
1176 sk_setup_caps(sk, &rt->dst);
1180 /* Routing failed... */
1181 sk->sk_route_caps = 0;
1183 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1184 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1186 if (!sysctl_ip_dynaddr ||
1187 sk->sk_state != TCP_SYN_SENT ||
1188 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1189 (err = inet_sk_reselect_saddr(sk)) != 0)
1190 sk->sk_err_soft = -err;
1195 EXPORT_SYMBOL(inet_sk_rebuild_header);
1197 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1198 netdev_features_t features)
1200 struct sk_buff *segs = ERR_PTR(-EINVAL);
1201 const struct net_offload *ops;
1202 unsigned int offset = 0;
1203 bool udpfrag, encap;
1210 if (unlikely(skb_shinfo(skb)->gso_type &
1220 SKB_GSO_UDP_TUNNEL |
1221 SKB_GSO_UDP_TUNNEL_CSUM |
1222 SKB_GSO_TUNNEL_REMCSUM |
1226 skb_reset_network_header(skb);
1227 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1228 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1233 if (ihl < sizeof(*iph))
1236 id = ntohs(iph->id);
1237 proto = iph->protocol;
1239 /* Warning: after this point, iph might be no longer valid */
1240 if (unlikely(!pskb_may_pull(skb, ihl)))
1242 __skb_pull(skb, ihl);
1244 encap = SKB_GSO_CB(skb)->encap_level > 0;
1246 features &= skb->dev->hw_enc_features;
1247 SKB_GSO_CB(skb)->encap_level += ihl;
1249 skb_reset_transport_header(skb);
1251 segs = ERR_PTR(-EPROTONOSUPPORT);
1253 if (skb->encapsulation &&
1254 skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
1255 udpfrag = proto == IPPROTO_UDP && encap;
1257 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
1259 ops = rcu_dereference(inet_offloads[proto]);
1260 if (likely(ops && ops->callbacks.gso_segment))
1261 segs = ops->callbacks.gso_segment(skb, features);
1263 if (IS_ERR_OR_NULL(segs))
1268 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1270 iph->id = htons(id);
1271 iph->frag_off = htons(offset >> 3);
1273 iph->frag_off |= htons(IP_MF);
1274 offset += skb->len - nhoff - ihl;
1276 iph->id = htons(id++);
1278 iph->tot_len = htons(skb->len - nhoff);
1281 skb_reset_inner_headers(skb);
1282 skb->network_header = (u8 *)iph - skb->head;
1283 } while ((skb = skb->next));
1289 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1290 struct sk_buff *skb)
1292 const struct net_offload *ops;
1293 struct sk_buff **pp = NULL;
1295 const struct iphdr *iph;
1302 off = skb_gro_offset(skb);
1303 hlen = off + sizeof(*iph);
1304 iph = skb_gro_header_fast(skb, off);
1305 if (skb_gro_header_hard(skb, hlen)) {
1306 iph = skb_gro_header_slow(skb, hlen, off);
1311 proto = iph->protocol;
1314 ops = rcu_dereference(inet_offloads[proto]);
1315 if (!ops || !ops->callbacks.gro_receive)
1318 if (*(u8 *)iph != 0x45)
1321 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1324 id = ntohl(*(__be32 *)&iph->id);
1325 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1328 for (p = *head; p; p = p->next) {
1331 if (!NAPI_GRO_CB(p)->same_flow)
1334 iph2 = (struct iphdr *)(p->data + off);
1335 /* The above works because, with the exception of the top
1336 * (inner most) layer, we only aggregate pkts with the same
1337 * hdr length so all the hdrs we'll need to verify will start
1338 * at the same offset.
1340 if ((iph->protocol ^ iph2->protocol) |
1341 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1342 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1343 NAPI_GRO_CB(p)->same_flow = 0;
1347 /* All fields must match except length and checksum. */
1348 NAPI_GRO_CB(p)->flush |=
1349 (iph->ttl ^ iph2->ttl) |
1350 (iph->tos ^ iph2->tos) |
1351 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1353 /* Save the IP ID check to be included later when we get to
1354 * the transport layer so only the inner most IP ID is checked.
1355 * This is because some GSO/TSO implementations do not
1356 * correctly increment the IP ID for the outer hdrs.
1358 NAPI_GRO_CB(p)->flush_id =
1359 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1360 NAPI_GRO_CB(p)->flush |= flush;
1363 NAPI_GRO_CB(skb)->flush |= flush;
1364 skb_set_network_header(skb, off);
1365 /* The above will be needed by the transport layer if there is one
1366 * immediately following this IP hdr.
1369 /* Note : No need to call skb_gro_postpull_rcsum() here,
1370 * as we already checked checksum over ipv4 header was 0
1372 skb_gro_pull(skb, sizeof(*iph));
1373 skb_set_transport_header(skb, skb_gro_offset(skb));
1375 pp = ops->callbacks.gro_receive(head, skb);
1381 NAPI_GRO_CB(skb)->flush |= flush;
1386 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1388 if (sk->sk_family == AF_INET)
1389 return ip_recv_error(sk, msg, len, addr_len);
1390 #if IS_ENABLED(CONFIG_IPV6)
1391 if (sk->sk_family == AF_INET6)
1392 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1397 static int inet_gro_complete(struct sk_buff *skb, int nhoff)
1399 __be16 newlen = htons(skb->len - nhoff);
1400 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1401 const struct net_offload *ops;
1402 int proto = iph->protocol;
1405 if (skb->encapsulation)
1406 skb_set_inner_network_header(skb, nhoff);
1408 csum_replace2(&iph->check, iph->tot_len, newlen);
1409 iph->tot_len = newlen;
1412 ops = rcu_dereference(inet_offloads[proto]);
1413 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1416 /* Only need to add sizeof(*iph) to get to the next hdr below
1417 * because any hdr with option will have been flushed in
1418 * inet_gro_receive().
1420 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1428 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1429 unsigned short type, unsigned char protocol,
1432 struct socket *sock;
1433 int rc = sock_create_kern(family, type, protocol, &sock);
1437 (*sk)->sk_allocation = GFP_ATOMIC;
1439 * Unhash it so that IP input processing does not even see it,
1440 * we do not wish this socket to see incoming packets.
1442 (*sk)->sk_prot->unhash(*sk);
1444 sk_change_net(*sk, net);
1448 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1450 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1452 unsigned long res = 0;
1455 for_each_possible_cpu(i)
1456 res += *(((unsigned long *) per_cpu_ptr(mib, i)) + offt);
1459 EXPORT_SYMBOL_GPL(snmp_fold_field);
1461 #if BITS_PER_LONG==32
1463 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1468 for_each_possible_cpu(cpu) {
1470 struct u64_stats_sync *syncp;
1474 bhptr = per_cpu_ptr(mib, cpu);
1475 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1477 start = u64_stats_fetch_begin_irq(syncp);
1478 v = *(((u64 *) bhptr) + offt);
1479 } while (u64_stats_fetch_retry_irq(syncp, start));
1485 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1488 #ifdef CONFIG_IP_MULTICAST
1489 static const struct net_protocol igmp_protocol = {
1490 .handler = igmp_rcv,
1495 static const struct net_protocol tcp_protocol = {
1496 .early_demux = tcp_v4_early_demux,
1497 .handler = tcp_v4_rcv,
1498 .err_handler = tcp_v4_err,
1501 .icmp_strict_tag_validation = 1,
1504 static const struct net_protocol udp_protocol = {
1505 .early_demux = udp_v4_early_demux,
1507 .err_handler = udp_err,
1512 static const struct net_protocol icmp_protocol = {
1513 .handler = icmp_rcv,
1514 .err_handler = icmp_err,
1519 static __net_init int ipv4_mib_init_net(struct net *net)
1523 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1524 if (!net->mib.tcp_statistics)
1526 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1527 if (!net->mib.ip_statistics)
1530 for_each_possible_cpu(i) {
1531 struct ipstats_mib *af_inet_stats;
1532 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1533 u64_stats_init(&af_inet_stats->syncp);
1536 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1537 if (!net->mib.net_statistics)
1539 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1540 if (!net->mib.udp_statistics)
1542 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1543 if (!net->mib.udplite_statistics)
1544 goto err_udplite_mib;
1545 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1546 if (!net->mib.icmp_statistics)
1548 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1550 if (!net->mib.icmpmsg_statistics)
1551 goto err_icmpmsg_mib;
1557 free_percpu(net->mib.icmp_statistics);
1559 free_percpu(net->mib.udplite_statistics);
1561 free_percpu(net->mib.udp_statistics);
1563 free_percpu(net->mib.net_statistics);
1565 free_percpu(net->mib.ip_statistics);
1567 free_percpu(net->mib.tcp_statistics);
1572 static __net_exit void ipv4_mib_exit_net(struct net *net)
1574 kfree(net->mib.icmpmsg_statistics);
1575 free_percpu(net->mib.icmp_statistics);
1576 free_percpu(net->mib.udplite_statistics);
1577 free_percpu(net->mib.udp_statistics);
1578 free_percpu(net->mib.net_statistics);
1579 free_percpu(net->mib.ip_statistics);
1580 free_percpu(net->mib.tcp_statistics);
1583 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1584 .init = ipv4_mib_init_net,
1585 .exit = ipv4_mib_exit_net,
1588 static int __init init_ipv4_mibs(void)
1590 return register_pernet_subsys(&ipv4_mib_ops);
1593 static __net_init int inet_init_net(struct net *net)
1596 * Set defaults for local port range
1598 seqlock_init(&net->ipv4.ip_local_ports.lock);
1599 net->ipv4.ip_local_ports.range[0] = 32768;
1600 net->ipv4.ip_local_ports.range[1] = 61000;
1602 seqlock_init(&net->ipv4.ping_group_range.lock);
1604 * Sane defaults - nobody may create ping sockets.
1605 * Boot scripts should set this to distro-specific group.
1607 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1608 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1612 static __net_exit void inet_exit_net(struct net *net)
1616 static __net_initdata struct pernet_operations af_inet_ops = {
1617 .init = inet_init_net,
1618 .exit = inet_exit_net,
1621 static int __init init_inet_pernet_ops(void)
1623 return register_pernet_subsys(&af_inet_ops);
1626 static int ipv4_proc_init(void);
1629 * IP protocol layer initialiser
1632 static struct packet_offload ip_packet_offload __read_mostly = {
1633 .type = cpu_to_be16(ETH_P_IP),
1635 .gso_segment = inet_gso_segment,
1636 .gro_receive = inet_gro_receive,
1637 .gro_complete = inet_gro_complete,
1641 static const struct net_offload ipip_offload = {
1643 .gso_segment = inet_gso_segment,
1644 .gro_receive = inet_gro_receive,
1645 .gro_complete = inet_gro_complete,
1649 static int __init ipv4_offload_init(void)
1654 if (udpv4_offload_init() < 0)
1655 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1656 if (tcpv4_offload_init() < 0)
1657 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1659 dev_add_offload(&ip_packet_offload);
1660 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1664 fs_initcall(ipv4_offload_init);
1666 static struct packet_type ip_packet_type __read_mostly = {
1667 .type = cpu_to_be16(ETH_P_IP),
1671 static int __init inet_init(void)
1673 struct inet_protosw *q;
1674 struct list_head *r;
1677 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1679 rc = proto_register(&tcp_prot, 1);
1683 rc = proto_register(&udp_prot, 1);
1685 goto out_unregister_tcp_proto;
1687 rc = proto_register(&raw_prot, 1);
1689 goto out_unregister_udp_proto;
1691 rc = proto_register(&ping_prot, 1);
1693 goto out_unregister_raw_proto;
1696 * Tell SOCKET that we are alive...
1699 (void)sock_register(&inet_family_ops);
1701 #ifdef CONFIG_SYSCTL
1702 ip_static_sysctl_init();
1706 * Add all the base protocols.
1709 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1710 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1711 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1712 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1713 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1714 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1715 #ifdef CONFIG_IP_MULTICAST
1716 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1717 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1720 /* Register the socket-side information for inet_create. */
1721 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1724 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1725 inet_register_protosw(q);
1728 * Set the ARP module up
1734 * Set the IP module up
1741 /* Setup TCP slab cache for open requests. */
1744 /* Setup UDP memory threshold */
1747 /* Add UDP-Lite (RFC 3828) */
1748 udplite4_register();
1753 * Set the ICMP layer up
1756 if (icmp_init() < 0)
1757 panic("Failed to create the ICMP control socket.\n");
1760 * Initialise the multicast router
1762 #if defined(CONFIG_IP_MROUTE)
1764 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1767 if (init_inet_pernet_ops())
1768 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1770 * Initialise per-cpu ipv4 mibs
1773 if (init_ipv4_mibs())
1774 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1780 dev_add_pack(&ip_packet_type);
1785 out_unregister_raw_proto:
1786 proto_unregister(&raw_prot);
1787 out_unregister_udp_proto:
1788 proto_unregister(&udp_prot);
1789 out_unregister_tcp_proto:
1790 proto_unregister(&tcp_prot);
1794 fs_initcall(inet_init);
1796 /* ------------------------------------------------------------------------ */
1798 #ifdef CONFIG_PROC_FS
1799 static int __init ipv4_proc_init(void)
1803 if (raw_proc_init())
1805 if (tcp4_proc_init())
1807 if (udp4_proc_init())
1809 if (ping_proc_init())
1811 if (ip_misc_proc_init())
1828 #else /* CONFIG_PROC_FS */
1829 static int __init ipv4_proc_init(void)
1833 #endif /* CONFIG_PROC_FS */
1835 MODULE_ALIAS_NETPROTO(PF_INET);