2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PF_INET protocol family socket handler.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
13 * Changes (see also sock.c)
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
68 #define pr_fmt(fmt) "IPv4: " fmt
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
93 #include <asm/uaccess.h>
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
101 #include <net/protocol.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
113 #include <net/icmp.h>
114 #include <net/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 struct ipv4_config ipv4_config;
130 EXPORT_SYMBOL(ipv4_config);
132 /* New destruction routine */
134 void inet_sock_destruct(struct sock *sk)
136 struct inet_sock *inet = inet_sk(sk);
138 __skb_queue_purge(&sk->sk_receive_queue);
139 __skb_queue_purge(&sk->sk_error_queue);
143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 pr_err("Attempt to release TCP socket in state %d %p\n",
148 if (!sock_flag(sk, SOCK_DEAD)) {
149 pr_err("Attempt to release alive inet socket %p\n", sk);
153 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
155 WARN_ON(sk->sk_wmem_queued);
156 WARN_ON(sk->sk_forward_alloc);
158 kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
160 dst_release(sk->sk_rx_dst);
161 sk_refcnt_debug_dec(sk);
163 EXPORT_SYMBOL(inet_sock_destruct);
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
172 * Automatically bind an unbound socket.
175 static int inet_autobind(struct sock *sk)
177 struct inet_sock *inet;
178 /* We may need to bind the socket. */
181 if (!inet->inet_num) {
182 if (sk->sk_prot->get_port(sk, 0)) {
186 inet->inet_sport = htons(inet->inet_num);
193 * Move a socket into listening state.
195 int inet_listen(struct socket *sock, int backlog)
197 struct sock *sk = sock->sk;
198 unsigned char old_state;
204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
207 old_state = sk->sk_state;
208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
211 /* Really, if the socket is already in listen state
212 * we can only allow the backlog to be adjusted.
214 if (old_state != TCP_LISTEN) {
215 /* Check special setups for testing purpose to enable TFO w/o
216 * requiring TCP_FASTOPEN sockopt.
217 * Note that only TCP sockets (SOCK_STREAM) will reach here.
218 * Also fastopenq may already been allocated because this
219 * socket was in TCP_LISTEN state previously but was
220 * shutdown() (rather than close()).
222 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
223 inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
224 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
225 err = fastopen_init_queue(sk, backlog);
226 else if ((sysctl_tcp_fastopen &
227 TFO_SERVER_WO_SOCKOPT2) != 0)
228 err = fastopen_init_queue(sk,
229 ((uint)sysctl_tcp_fastopen) >> 16);
235 err = inet_csk_listen_start(sk, backlog);
239 sk->sk_max_ack_backlog = backlog;
246 EXPORT_SYMBOL(inet_listen);
249 * Create an inet socket.
252 static int inet_create(struct net *net, struct socket *sock, int protocol,
256 struct inet_protosw *answer;
257 struct inet_sock *inet;
258 struct proto *answer_prot;
259 unsigned char answer_flags;
260 char answer_no_check;
261 int try_loading_module = 0;
264 sock->state = SS_UNCONNECTED;
266 /* Look for the requested type/protocol pair. */
268 err = -ESOCKTNOSUPPORT;
270 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
273 /* Check the non-wild match. */
274 if (protocol == answer->protocol) {
275 if (protocol != IPPROTO_IP)
278 /* Check for the two wild cases. */
279 if (IPPROTO_IP == protocol) {
280 protocol = answer->protocol;
283 if (IPPROTO_IP == answer->protocol)
286 err = -EPROTONOSUPPORT;
290 if (try_loading_module < 2) {
293 * Be more specific, e.g. net-pf-2-proto-132-type-1
294 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
296 if (++try_loading_module == 1)
297 request_module("net-pf-%d-proto-%d-type-%d",
298 PF_INET, protocol, sock->type);
300 * Fall back to generic, e.g. net-pf-2-proto-132
301 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
304 request_module("net-pf-%d-proto-%d",
306 goto lookup_protocol;
312 if (sock->type == SOCK_RAW && !kern &&
313 !ns_capable(net->user_ns, CAP_NET_RAW))
316 sock->ops = answer->ops;
317 answer_prot = answer->prot;
318 answer_no_check = answer->no_check;
319 answer_flags = answer->flags;
322 WARN_ON(answer_prot->slab == NULL);
325 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
330 sk->sk_no_check = answer_no_check;
331 if (INET_PROTOSW_REUSE & answer_flags)
332 sk->sk_reuse = SK_CAN_REUSE;
335 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
339 if (SOCK_RAW == sock->type) {
340 inet->inet_num = protocol;
341 if (IPPROTO_RAW == protocol)
345 if (ipv4_config.no_pmtu_disc)
346 inet->pmtudisc = IP_PMTUDISC_DONT;
348 inet->pmtudisc = IP_PMTUDISC_WANT;
352 sock_init_data(sock, sk);
354 sk->sk_destruct = inet_sock_destruct;
355 sk->sk_protocol = protocol;
356 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
363 inet->mc_list = NULL;
366 sk_refcnt_debug_inc(sk);
368 if (inet->inet_num) {
369 /* It assumes that any protocol which allows
370 * the user to assign a number at socket
371 * creation time automatically
374 inet->inet_sport = htons(inet->inet_num);
375 /* Add to protocol hash chains. */
376 sk->sk_prot->hash(sk);
379 if (sk->sk_prot->init) {
380 err = sk->sk_prot->init(sk);
382 sk_common_release(sk);
393 * The peer socket should always be NULL (or else). When we call this
394 * function we are destroying the object and from then on nobody
395 * should refer to it.
397 int inet_release(struct socket *sock)
399 struct sock *sk = sock->sk;
404 sock_rps_reset_flow(sk);
406 /* Applications forget to leave groups before exiting */
407 ip_mc_drop_socket(sk);
409 /* If linger is set, we don't return until the close
410 * is complete. Otherwise we return immediately. The
411 * actually closing is done the same either way.
413 * If the close is due to the process exiting, we never
417 if (sock_flag(sk, SOCK_LINGER) &&
418 !(current->flags & PF_EXITING))
419 timeout = sk->sk_lingertime;
421 sk->sk_prot->close(sk, timeout);
425 EXPORT_SYMBOL(inet_release);
427 /* It is off by default, see below. */
428 int sysctl_ip_nonlocal_bind __read_mostly;
429 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
431 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
433 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
434 struct sock *sk = sock->sk;
435 struct inet_sock *inet = inet_sk(sk);
436 struct net *net = sock_net(sk);
441 /* If the socket has its own bind function then use it. (RAW) */
442 if (sk->sk_prot->bind) {
443 err = sk->sk_prot->bind(sk, uaddr, addr_len);
447 if (addr_len < sizeof(struct sockaddr_in))
450 if (addr->sin_family != AF_INET) {
451 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
452 * only if s_addr is INADDR_ANY.
455 if (addr->sin_family != AF_UNSPEC ||
456 addr->sin_addr.s_addr != htonl(INADDR_ANY))
460 chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
462 /* Not specified by any standard per-se, however it breaks too
463 * many applications when removed. It is unfortunate since
464 * allowing applications to make a non-local bind solves
465 * several problems with systems using dynamic addressing.
466 * (ie. your servers still start up even if your ISDN link
467 * is temporarily down)
469 err = -EADDRNOTAVAIL;
470 if (!sysctl_ip_nonlocal_bind &&
471 !(inet->freebind || inet->transparent) &&
472 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
473 chk_addr_ret != RTN_LOCAL &&
474 chk_addr_ret != RTN_MULTICAST &&
475 chk_addr_ret != RTN_BROADCAST)
478 snum = ntohs(addr->sin_port);
480 if (snum && snum < PROT_SOCK &&
481 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
484 /* We keep a pair of addresses. rcv_saddr is the one
485 * used by hash lookups, and saddr is used for transmit.
487 * In the BSD API these are the same except where it
488 * would be illegal to use them (multicast/broadcast) in
489 * which case the sending device address is used.
493 /* Check these errors (active socket, double bind). */
495 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
496 goto out_release_sock;
498 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
499 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
500 inet->inet_saddr = 0; /* Use device */
502 /* Make sure we are allowed to bind here. */
503 if (sk->sk_prot->get_port(sk, snum)) {
504 inet->inet_saddr = inet->inet_rcv_saddr = 0;
506 goto out_release_sock;
509 if (inet->inet_rcv_saddr)
510 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
512 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
513 inet->inet_sport = htons(inet->inet_num);
514 inet->inet_daddr = 0;
515 inet->inet_dport = 0;
523 EXPORT_SYMBOL(inet_bind);
525 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
526 int addr_len, int flags)
528 struct sock *sk = sock->sk;
530 if (addr_len < sizeof(uaddr->sa_family))
532 if (uaddr->sa_family == AF_UNSPEC)
533 return sk->sk_prot->disconnect(sk, flags);
535 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
537 return sk->sk_prot->connect(sk, uaddr, addr_len);
539 EXPORT_SYMBOL(inet_dgram_connect);
541 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
545 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
546 sk->sk_write_pending += writebias;
548 /* Basic assumption: if someone sets sk->sk_err, he _must_
549 * change state of the socket from TCP_SYN_*.
550 * Connect() does not allow to get error notifications
551 * without closing the socket.
553 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
555 timeo = schedule_timeout(timeo);
557 if (signal_pending(current) || !timeo)
559 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
561 finish_wait(sk_sleep(sk), &wait);
562 sk->sk_write_pending -= writebias;
567 * Connect to a remote host. There is regrettably still a little
568 * TCP 'magic' in here.
570 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
571 int addr_len, int flags)
573 struct sock *sk = sock->sk;
577 if (addr_len < sizeof(uaddr->sa_family))
580 if (uaddr->sa_family == AF_UNSPEC) {
581 err = sk->sk_prot->disconnect(sk, flags);
582 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
586 switch (sock->state) {
595 /* Fall out of switch with err, set for this state */
599 if (sk->sk_state != TCP_CLOSE)
602 err = sk->sk_prot->connect(sk, uaddr, addr_len);
606 sock->state = SS_CONNECTING;
608 /* Just entered SS_CONNECTING state; the only
609 * difference is that return value in non-blocking
610 * case is EINPROGRESS, rather than EALREADY.
616 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
618 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
619 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
620 tcp_sk(sk)->fastopen_req &&
621 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
623 /* Error code is set above */
624 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
627 err = sock_intr_errno(timeo);
628 if (signal_pending(current))
632 /* Connection was closed by RST, timeout, ICMP error
633 * or another process disconnected us.
635 if (sk->sk_state == TCP_CLOSE)
638 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
639 * and error was received after socket entered established state.
640 * Hence, it is handled normally after connect() return successfully.
643 sock->state = SS_CONNECTED;
649 err = sock_error(sk) ? : -ECONNABORTED;
650 sock->state = SS_UNCONNECTED;
651 if (sk->sk_prot->disconnect(sk, flags))
652 sock->state = SS_DISCONNECTING;
655 EXPORT_SYMBOL(__inet_stream_connect);
657 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
658 int addr_len, int flags)
663 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
664 release_sock(sock->sk);
667 EXPORT_SYMBOL(inet_stream_connect);
670 * Accept a pending connection. The TCP layer now gives BSD semantics.
673 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
675 struct sock *sk1 = sock->sk;
677 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
684 sock_rps_record_flow(sk2);
685 WARN_ON(!((1 << sk2->sk_state) &
686 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
687 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
689 sock_graft(sk2, newsock);
691 newsock->state = SS_CONNECTED;
697 EXPORT_SYMBOL(inet_accept);
701 * This does both peername and sockname.
703 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
704 int *uaddr_len, int peer)
706 struct sock *sk = sock->sk;
707 struct inet_sock *inet = inet_sk(sk);
708 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
710 sin->sin_family = AF_INET;
712 if (!inet->inet_dport ||
713 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
716 sin->sin_port = inet->inet_dport;
717 sin->sin_addr.s_addr = inet->inet_daddr;
719 __be32 addr = inet->inet_rcv_saddr;
721 addr = inet->inet_saddr;
722 sin->sin_port = inet->inet_sport;
723 sin->sin_addr.s_addr = addr;
725 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
726 *uaddr_len = sizeof(*sin);
729 EXPORT_SYMBOL(inet_getname);
731 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
734 struct sock *sk = sock->sk;
736 sock_rps_record_flow(sk);
738 /* We may need to bind the socket. */
739 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
743 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
745 EXPORT_SYMBOL(inet_sendmsg);
747 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
748 size_t size, int flags)
750 struct sock *sk = sock->sk;
752 sock_rps_record_flow(sk);
754 /* We may need to bind the socket. */
755 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
759 if (sk->sk_prot->sendpage)
760 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
761 return sock_no_sendpage(sock, page, offset, size, flags);
763 EXPORT_SYMBOL(inet_sendpage);
765 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
766 size_t size, int flags)
768 struct sock *sk = sock->sk;
772 sock_rps_record_flow(sk);
774 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
775 flags & ~MSG_DONTWAIT, &addr_len);
777 msg->msg_namelen = addr_len;
780 EXPORT_SYMBOL(inet_recvmsg);
782 int inet_shutdown(struct socket *sock, int how)
784 struct sock *sk = sock->sk;
787 /* This should really check to make sure
788 * the socket is a TCP socket. (WHY AC...)
790 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
793 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
797 if (sock->state == SS_CONNECTING) {
798 if ((1 << sk->sk_state) &
799 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
800 sock->state = SS_DISCONNECTING;
802 sock->state = SS_CONNECTED;
805 switch (sk->sk_state) {
808 /* Hack to wake up other listeners, who can poll for
809 POLLHUP, even on eg. unconnected UDP sockets -- RR */
811 sk->sk_shutdown |= how;
812 if (sk->sk_prot->shutdown)
813 sk->sk_prot->shutdown(sk, how);
816 /* Remaining two branches are temporary solution for missing
817 * close() in multithreaded environment. It is _not_ a good idea,
818 * but we have no choice until close() is repaired at VFS level.
821 if (!(how & RCV_SHUTDOWN))
825 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
826 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
830 /* Wake up anyone sleeping in poll. */
831 sk->sk_state_change(sk);
835 EXPORT_SYMBOL(inet_shutdown);
838 * ioctl() calls you can issue on an INET socket. Most of these are
839 * device configuration and stuff and very rarely used. Some ioctls
840 * pass on to the socket itself.
842 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
843 * loads the devconfigure module does its configuring and unloads it.
844 * There's a good 20K of config code hanging around the kernel.
847 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
849 struct sock *sk = sock->sk;
851 struct net *net = sock_net(sk);
855 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
858 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
863 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
868 err = arp_ioctl(net, cmd, (void __user *)arg);
881 err = devinet_ioctl(net, cmd, (void __user *)arg);
884 if (sk->sk_prot->ioctl)
885 err = sk->sk_prot->ioctl(sk, cmd, arg);
892 EXPORT_SYMBOL(inet_ioctl);
895 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
897 struct sock *sk = sock->sk;
898 int err = -ENOIOCTLCMD;
900 if (sk->sk_prot->compat_ioctl)
901 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
907 const struct proto_ops inet_stream_ops = {
909 .owner = THIS_MODULE,
910 .release = inet_release,
912 .connect = inet_stream_connect,
913 .socketpair = sock_no_socketpair,
914 .accept = inet_accept,
915 .getname = inet_getname,
918 .listen = inet_listen,
919 .shutdown = inet_shutdown,
920 .setsockopt = sock_common_setsockopt,
921 .getsockopt = sock_common_getsockopt,
922 .sendmsg = inet_sendmsg,
923 .recvmsg = inet_recvmsg,
924 .mmap = sock_no_mmap,
925 .sendpage = inet_sendpage,
926 .splice_read = tcp_splice_read,
928 .compat_setsockopt = compat_sock_common_setsockopt,
929 .compat_getsockopt = compat_sock_common_getsockopt,
930 .compat_ioctl = inet_compat_ioctl,
933 EXPORT_SYMBOL(inet_stream_ops);
935 const struct proto_ops inet_dgram_ops = {
937 .owner = THIS_MODULE,
938 .release = inet_release,
940 .connect = inet_dgram_connect,
941 .socketpair = sock_no_socketpair,
942 .accept = sock_no_accept,
943 .getname = inet_getname,
946 .listen = sock_no_listen,
947 .shutdown = inet_shutdown,
948 .setsockopt = sock_common_setsockopt,
949 .getsockopt = sock_common_getsockopt,
950 .sendmsg = inet_sendmsg,
951 .recvmsg = inet_recvmsg,
952 .mmap = sock_no_mmap,
953 .sendpage = inet_sendpage,
955 .compat_setsockopt = compat_sock_common_setsockopt,
956 .compat_getsockopt = compat_sock_common_getsockopt,
957 .compat_ioctl = inet_compat_ioctl,
960 EXPORT_SYMBOL(inet_dgram_ops);
963 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
966 static const struct proto_ops inet_sockraw_ops = {
968 .owner = THIS_MODULE,
969 .release = inet_release,
971 .connect = inet_dgram_connect,
972 .socketpair = sock_no_socketpair,
973 .accept = sock_no_accept,
974 .getname = inet_getname,
975 .poll = datagram_poll,
977 .listen = sock_no_listen,
978 .shutdown = inet_shutdown,
979 .setsockopt = sock_common_setsockopt,
980 .getsockopt = sock_common_getsockopt,
981 .sendmsg = inet_sendmsg,
982 .recvmsg = inet_recvmsg,
983 .mmap = sock_no_mmap,
984 .sendpage = inet_sendpage,
986 .compat_setsockopt = compat_sock_common_setsockopt,
987 .compat_getsockopt = compat_sock_common_getsockopt,
988 .compat_ioctl = inet_compat_ioctl,
992 static const struct net_proto_family inet_family_ops = {
994 .create = inet_create,
995 .owner = THIS_MODULE,
998 /* Upon startup we insert all the elements in inetsw_array[] into
999 * the linked list inetsw.
1001 static struct inet_protosw inetsw_array[] =
1004 .type = SOCK_STREAM,
1005 .protocol = IPPROTO_TCP,
1007 .ops = &inet_stream_ops,
1009 .flags = INET_PROTOSW_PERMANENT |
1015 .protocol = IPPROTO_UDP,
1017 .ops = &inet_dgram_ops,
1018 .no_check = UDP_CSUM_DEFAULT,
1019 .flags = INET_PROTOSW_PERMANENT,
1024 .protocol = IPPROTO_ICMP,
1026 .ops = &inet_dgram_ops,
1027 .no_check = UDP_CSUM_DEFAULT,
1028 .flags = INET_PROTOSW_REUSE,
1033 .protocol = IPPROTO_IP, /* wild card */
1035 .ops = &inet_sockraw_ops,
1036 .no_check = UDP_CSUM_DEFAULT,
1037 .flags = INET_PROTOSW_REUSE,
1041 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1043 void inet_register_protosw(struct inet_protosw *p)
1045 struct list_head *lh;
1046 struct inet_protosw *answer;
1047 int protocol = p->protocol;
1048 struct list_head *last_perm;
1050 spin_lock_bh(&inetsw_lock);
1052 if (p->type >= SOCK_MAX)
1055 /* If we are trying to override a permanent protocol, bail. */
1057 last_perm = &inetsw[p->type];
1058 list_for_each(lh, &inetsw[p->type]) {
1059 answer = list_entry(lh, struct inet_protosw, list);
1061 /* Check only the non-wild match. */
1062 if (INET_PROTOSW_PERMANENT & answer->flags) {
1063 if (protocol == answer->protocol)
1073 /* Add the new entry after the last permanent entry if any, so that
1074 * the new entry does not override a permanent entry when matched with
1075 * a wild-card protocol. But it is allowed to override any existing
1076 * non-permanent entry. This means that when we remove this entry, the
1077 * system automatically returns to the old behavior.
1079 list_add_rcu(&p->list, last_perm);
1081 spin_unlock_bh(&inetsw_lock);
1086 pr_err("Attempt to override permanent protocol %d\n", protocol);
1090 pr_err("Ignoring attempt to register invalid socket type %d\n",
1094 EXPORT_SYMBOL(inet_register_protosw);
1096 void inet_unregister_protosw(struct inet_protosw *p)
1098 if (INET_PROTOSW_PERMANENT & p->flags) {
1099 pr_err("Attempt to unregister permanent protocol %d\n",
1102 spin_lock_bh(&inetsw_lock);
1103 list_del_rcu(&p->list);
1104 spin_unlock_bh(&inetsw_lock);
1109 EXPORT_SYMBOL(inet_unregister_protosw);
1112 * Shall we try to damage output packets if routing dev changes?
1115 int sysctl_ip_dynaddr __read_mostly;
1117 static int inet_sk_reselect_saddr(struct sock *sk)
1119 struct inet_sock *inet = inet_sk(sk);
1120 __be32 old_saddr = inet->inet_saddr;
1121 __be32 daddr = inet->inet_daddr;
1125 struct ip_options_rcu *inet_opt;
1127 inet_opt = rcu_dereference_protected(inet->inet_opt,
1128 sock_owned_by_user(sk));
1129 if (inet_opt && inet_opt->opt.srr)
1130 daddr = inet_opt->opt.faddr;
1132 /* Query new route. */
1133 fl4 = &inet->cork.fl.u.ip4;
1134 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1135 sk->sk_bound_dev_if, sk->sk_protocol,
1136 inet->inet_sport, inet->inet_dport, sk, false);
1140 sk_setup_caps(sk, &rt->dst);
1142 new_saddr = fl4->saddr;
1144 if (new_saddr == old_saddr)
1147 if (sysctl_ip_dynaddr > 1) {
1148 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1149 __func__, &old_saddr, &new_saddr);
1152 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1155 * XXX The only one ugly spot where we need to
1156 * XXX really change the sockets identity after
1157 * XXX it has entered the hashes. -DaveM
1159 * Besides that, it does not check for connection
1160 * uniqueness. Wait for troubles.
1162 __sk_prot_rehash(sk);
1166 int inet_sk_rebuild_header(struct sock *sk)
1168 struct inet_sock *inet = inet_sk(sk);
1169 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1171 struct ip_options_rcu *inet_opt;
1175 /* Route is OK, nothing to do. */
1181 inet_opt = rcu_dereference(inet->inet_opt);
1182 daddr = inet->inet_daddr;
1183 if (inet_opt && inet_opt->opt.srr)
1184 daddr = inet_opt->opt.faddr;
1186 fl4 = &inet->cork.fl.u.ip4;
1187 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1188 inet->inet_dport, inet->inet_sport,
1189 sk->sk_protocol, RT_CONN_FLAGS(sk),
1190 sk->sk_bound_dev_if);
1193 sk_setup_caps(sk, &rt->dst);
1197 /* Routing failed... */
1198 sk->sk_route_caps = 0;
1200 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1201 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1203 if (!sysctl_ip_dynaddr ||
1204 sk->sk_state != TCP_SYN_SENT ||
1205 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1206 (err = inet_sk_reselect_saddr(sk)) != 0)
1207 sk->sk_err_soft = -err;
1212 EXPORT_SYMBOL(inet_sk_rebuild_header);
1214 static int inet_gso_send_check(struct sk_buff *skb)
1216 const struct net_offload *ops;
1217 const struct iphdr *iph;
1222 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1227 if (ihl < sizeof(*iph))
1230 proto = iph->protocol;
1232 /* Warning: after this point, iph might be no longer valid */
1233 if (unlikely(!pskb_may_pull(skb, ihl)))
1235 __skb_pull(skb, ihl);
1237 skb_reset_transport_header(skb);
1238 err = -EPROTONOSUPPORT;
1240 ops = rcu_dereference(inet_offloads[proto]);
1241 if (likely(ops && ops->callbacks.gso_send_check))
1242 err = ops->callbacks.gso_send_check(skb);
1248 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1249 netdev_features_t features)
1251 struct sk_buff *segs = ERR_PTR(-EINVAL);
1252 const struct net_offload *ops;
1253 unsigned int offset = 0;
1261 if (unlikely(skb_shinfo(skb)->gso_type &
1269 SKB_GSO_UDP_TUNNEL |
1274 skb_reset_network_header(skb);
1275 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1276 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1281 if (ihl < sizeof(*iph))
1284 id = ntohs(iph->id);
1285 proto = iph->protocol;
1287 /* Warning: after this point, iph might be no longer valid */
1288 if (unlikely(!pskb_may_pull(skb, ihl)))
1290 __skb_pull(skb, ihl);
1292 tunnel = SKB_GSO_CB(skb)->encap_level > 0;
1294 features = skb->dev->hw_enc_features & netif_skb_features(skb);
1295 SKB_GSO_CB(skb)->encap_level += ihl;
1297 skb_reset_transport_header(skb);
1299 segs = ERR_PTR(-EPROTONOSUPPORT);
1301 ops = rcu_dereference(inet_offloads[proto]);
1302 if (likely(ops && ops->callbacks.gso_segment))
1303 segs = ops->callbacks.gso_segment(skb, features);
1305 if (IS_ERR_OR_NULL(segs))
1310 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1311 if (!tunnel && proto == IPPROTO_UDP) {
1312 iph->id = htons(id);
1313 iph->frag_off = htons(offset >> 3);
1314 if (skb->next != NULL)
1315 iph->frag_off |= htons(IP_MF);
1316 offset += skb->len - nhoff - ihl;
1318 iph->id = htons(id++);
1320 iph->tot_len = htons(skb->len - nhoff);
1323 skb_reset_inner_headers(skb);
1324 skb->encapsulation = 1;
1326 skb->network_header = (u8 *)iph - skb->head;
1327 } while ((skb = skb->next));
1333 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1334 struct sk_buff *skb)
1336 const struct net_offload *ops;
1337 struct sk_buff **pp = NULL;
1339 const struct iphdr *iph;
1346 off = skb_gro_offset(skb);
1347 hlen = off + sizeof(*iph);
1348 iph = skb_gro_header_fast(skb, off);
1349 if (skb_gro_header_hard(skb, hlen)) {
1350 iph = skb_gro_header_slow(skb, hlen, off);
1355 proto = iph->protocol;
1358 ops = rcu_dereference(inet_offloads[proto]);
1359 if (!ops || !ops->callbacks.gro_receive)
1362 if (*(u8 *)iph != 0x45)
1365 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1368 id = ntohl(*(__be32 *)&iph->id);
1369 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1372 for (p = *head; p; p = p->next) {
1375 if (!NAPI_GRO_CB(p)->same_flow)
1380 if ((iph->protocol ^ iph2->protocol) |
1381 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1382 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1383 NAPI_GRO_CB(p)->same_flow = 0;
1387 /* All fields must match except length and checksum. */
1388 NAPI_GRO_CB(p)->flush |=
1389 (iph->ttl ^ iph2->ttl) |
1390 (iph->tos ^ iph2->tos) |
1391 (__force int)((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)) |
1392 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1394 NAPI_GRO_CB(p)->flush |= flush;
1397 NAPI_GRO_CB(skb)->flush |= flush;
1398 skb_gro_pull(skb, sizeof(*iph));
1399 skb_set_transport_header(skb, skb_gro_offset(skb));
1401 pp = ops->callbacks.gro_receive(head, skb);
1407 NAPI_GRO_CB(skb)->flush |= flush;
1412 static int inet_gro_complete(struct sk_buff *skb)
1414 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1415 struct iphdr *iph = ip_hdr(skb);
1416 const struct net_offload *ops;
1417 int proto = iph->protocol;
1420 csum_replace2(&iph->check, iph->tot_len, newlen);
1421 iph->tot_len = newlen;
1424 ops = rcu_dereference(inet_offloads[proto]);
1425 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1428 err = ops->callbacks.gro_complete(skb);
1436 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1437 unsigned short type, unsigned char protocol,
1440 struct socket *sock;
1441 int rc = sock_create_kern(family, type, protocol, &sock);
1445 (*sk)->sk_allocation = GFP_ATOMIC;
1447 * Unhash it so that IP input processing does not even see it,
1448 * we do not wish this socket to see incoming packets.
1450 (*sk)->sk_prot->unhash(*sk);
1452 sk_change_net(*sk, net);
1456 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1458 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1460 unsigned long res = 0;
1463 for_each_possible_cpu(i) {
1464 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1465 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1469 EXPORT_SYMBOL_GPL(snmp_fold_field);
1471 #if BITS_PER_LONG==32
1473 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1478 for_each_possible_cpu(cpu) {
1480 struct u64_stats_sync *syncp;
1484 bhptr = per_cpu_ptr(mib[0], cpu);
1485 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1487 start = u64_stats_fetch_begin_bh(syncp);
1488 v = *(((u64 *) bhptr) + offt);
1489 } while (u64_stats_fetch_retry_bh(syncp, start));
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 #if SNMP_ARRAY_SZ == 2
1505 ptr[1] = __alloc_percpu(mibsize, align);
1507 free_percpu(ptr[0]);
1514 EXPORT_SYMBOL_GPL(snmp_mib_init);
1516 #ifdef CONFIG_IP_MULTICAST
1517 static const struct net_protocol igmp_protocol = {
1518 .handler = igmp_rcv,
1523 static const struct net_protocol tcp_protocol = {
1524 .early_demux = tcp_v4_early_demux,
1525 .handler = tcp_v4_rcv,
1526 .err_handler = tcp_v4_err,
1531 static const struct net_protocol udp_protocol = {
1532 .early_demux = udp_v4_early_demux,
1534 .err_handler = udp_err,
1539 static const struct net_protocol icmp_protocol = {
1540 .handler = icmp_rcv,
1541 .err_handler = icmp_err,
1546 static __net_init int ipv4_mib_init_net(struct net *net)
1548 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1549 sizeof(struct tcp_mib),
1550 __alignof__(struct tcp_mib)) < 0)
1552 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1553 sizeof(struct ipstats_mib),
1554 __alignof__(struct ipstats_mib)) < 0)
1556 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1557 sizeof(struct linux_mib),
1558 __alignof__(struct linux_mib)) < 0)
1560 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1561 sizeof(struct udp_mib),
1562 __alignof__(struct udp_mib)) < 0)
1564 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1565 sizeof(struct udp_mib),
1566 __alignof__(struct udp_mib)) < 0)
1567 goto err_udplite_mib;
1568 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1569 sizeof(struct icmp_mib),
1570 __alignof__(struct icmp_mib)) < 0)
1572 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1574 if (!net->mib.icmpmsg_statistics)
1575 goto err_icmpmsg_mib;
1581 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1583 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1585 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1587 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1589 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1591 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1596 static __net_exit void ipv4_mib_exit_net(struct net *net)
1598 kfree(net->mib.icmpmsg_statistics);
1599 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1600 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1601 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1602 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1603 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1604 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1607 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1608 .init = ipv4_mib_init_net,
1609 .exit = ipv4_mib_exit_net,
1612 static int __init init_ipv4_mibs(void)
1614 return register_pernet_subsys(&ipv4_mib_ops);
1617 static int ipv4_proc_init(void);
1620 * IP protocol layer initialiser
1623 static struct packet_offload ip_packet_offload __read_mostly = {
1624 .type = cpu_to_be16(ETH_P_IP),
1626 .gso_send_check = inet_gso_send_check,
1627 .gso_segment = inet_gso_segment,
1628 .gro_receive = inet_gro_receive,
1629 .gro_complete = inet_gro_complete,
1633 static const struct net_offload ipip_offload = {
1635 .gso_send_check = inet_gso_send_check,
1636 .gso_segment = inet_gso_segment,
1640 static int __init ipv4_offload_init(void)
1645 if (udpv4_offload_init() < 0)
1646 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1647 if (tcpv4_offload_init() < 0)
1648 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1650 dev_add_offload(&ip_packet_offload);
1651 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1655 fs_initcall(ipv4_offload_init);
1657 static struct packet_type ip_packet_type __read_mostly = {
1658 .type = cpu_to_be16(ETH_P_IP),
1662 static int __init inet_init(void)
1664 struct inet_protosw *q;
1665 struct list_head *r;
1668 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1670 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1671 if (!sysctl_local_reserved_ports)
1674 rc = proto_register(&tcp_prot, 1);
1676 goto out_free_reserved_ports;
1678 rc = proto_register(&udp_prot, 1);
1680 goto out_unregister_tcp_proto;
1682 rc = proto_register(&raw_prot, 1);
1684 goto out_unregister_udp_proto;
1686 rc = proto_register(&ping_prot, 1);
1688 goto out_unregister_raw_proto;
1691 * Tell SOCKET that we are alive...
1694 (void)sock_register(&inet_family_ops);
1696 #ifdef CONFIG_SYSCTL
1697 ip_static_sysctl_init();
1701 * Add all the base protocols.
1704 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1705 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1706 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1707 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1708 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1709 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1710 #ifdef CONFIG_IP_MULTICAST
1711 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1712 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1715 /* Register the socket-side information for inet_create. */
1716 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1719 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1720 inet_register_protosw(q);
1723 * Set the ARP module up
1729 * Set the IP module up
1736 /* Setup TCP slab cache for open requests. */
1739 /* Setup UDP memory threshold */
1742 /* Add UDP-Lite (RFC 3828) */
1743 udplite4_register();
1748 * Set the ICMP layer up
1751 if (icmp_init() < 0)
1752 panic("Failed to create the ICMP control socket.\n");
1755 * Initialise the multicast router
1757 #if defined(CONFIG_IP_MROUTE)
1759 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1762 * Initialise per-cpu ipv4 mibs
1765 if (init_ipv4_mibs())
1766 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1772 dev_add_pack(&ip_packet_type);
1777 out_unregister_raw_proto:
1778 proto_unregister(&raw_prot);
1779 out_unregister_udp_proto:
1780 proto_unregister(&udp_prot);
1781 out_unregister_tcp_proto:
1782 proto_unregister(&tcp_prot);
1783 out_free_reserved_ports:
1784 kfree(sysctl_local_reserved_ports);
1788 fs_initcall(inet_init);
1790 /* ------------------------------------------------------------------------ */
1792 #ifdef CONFIG_PROC_FS
1793 static int __init ipv4_proc_init(void)
1797 if (raw_proc_init())
1799 if (tcp4_proc_init())
1801 if (udp4_proc_init())
1803 if (ping_proc_init())
1805 if (ip_misc_proc_init())
1822 #else /* CONFIG_PROC_FS */
1823 static int __init ipv4_proc_init(void)
1827 #endif /* CONFIG_PROC_FS */
1829 MODULE_ALIAS_NETPROTO(PF_INET);