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>
90 #include <asm/uaccess.h>
91 #include <asm/system.h>
93 #include <linux/inet.h>
94 #include <linux/igmp.h>
95 #include <linux/inetdevice.h>
96 #include <linux/netdevice.h>
97 #include <net/checksum.h>
99 #include <net/protocol.h>
101 #include <net/route.h>
102 #include <net/ip_fib.h>
103 #include <net/inet_connection_sock.h>
106 #include <net/udplite.h>
107 #include <linux/skbuff.h>
108 #include <net/sock.h>
110 #include <net/icmp.h>
111 #include <net/ipip.h>
112 #include <net/inet_common.h>
113 #include <net/xfrm.h>
114 #include <net/net_namespace.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
119 extern void ip_mc_drop_socket(struct sock *sk);
121 /* The inetsw table contains everything that inet_create needs to
122 * build a new socket.
124 static struct list_head inetsw[SOCK_MAX];
125 static DEFINE_SPINLOCK(inetsw_lock);
127 struct ipv4_config ipv4_config;
129 EXPORT_SYMBOL(ipv4_config);
131 /* New destruction routine */
133 void inet_sock_destruct(struct sock *sk)
135 struct inet_sock *inet = inet_sk(sk);
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 printk("Attempt to release TCP socket in state %d %p\n",
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 printk("Attempt to release alive inet socket %p\n", sk);
152 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 WARN_ON(sk->sk_wmem_queued);
155 WARN_ON(sk->sk_forward_alloc);
158 dst_release(sk->sk_dst_cache);
159 sk_refcnt_debug_dec(sk);
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. */
179 if (sk->sk_prot->get_port(sk, 0)) {
183 inet->sport = htons(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 err = inet_csk_listen_start(sk, backlog);
216 sk->sk_max_ack_backlog = backlog;
224 u32 inet_ehash_secret __read_mostly;
225 EXPORT_SYMBOL(inet_ehash_secret);
228 * inet_ehash_secret must be set exactly once
229 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
231 void build_ehash_secret(void)
235 get_random_bytes(&rnd, sizeof(rnd));
237 spin_lock_bh(&inetsw_lock);
238 if (!inet_ehash_secret)
239 inet_ehash_secret = rnd;
240 spin_unlock_bh(&inetsw_lock);
242 EXPORT_SYMBOL(build_ehash_secret);
244 static inline int inet_netns_ok(struct net *net, int protocol)
247 struct net_protocol *ipprot;
249 if (net_eq(net, &init_net))
252 hash = protocol & (MAX_INET_PROTOS - 1);
253 ipprot = rcu_dereference(inet_protos[hash]);
258 return ipprot->netns_ok;
262 * Create an inet socket.
265 static int inet_create(struct net *net, struct socket *sock, int protocol)
268 struct inet_protosw *answer;
269 struct inet_sock *inet;
270 struct proto *answer_prot;
271 unsigned char answer_flags;
272 char answer_no_check;
273 int try_loading_module = 0;
276 if (unlikely(!inet_ehash_secret))
277 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
278 build_ehash_secret();
280 sock->state = SS_UNCONNECTED;
282 /* Look for the requested type/protocol pair. */
284 err = -ESOCKTNOSUPPORT;
286 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
289 /* Check the non-wild match. */
290 if (protocol == answer->protocol) {
291 if (protocol != IPPROTO_IP)
294 /* Check for the two wild cases. */
295 if (IPPROTO_IP == protocol) {
296 protocol = answer->protocol;
299 if (IPPROTO_IP == answer->protocol)
302 err = -EPROTONOSUPPORT;
306 if (try_loading_module < 2) {
309 * Be more specific, e.g. net-pf-2-proto-132-type-1
310 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
312 if (++try_loading_module == 1)
313 request_module("net-pf-%d-proto-%d-type-%d",
314 PF_INET, protocol, sock->type);
316 * Fall back to generic, e.g. net-pf-2-proto-132
317 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
320 request_module("net-pf-%d-proto-%d",
322 goto lookup_protocol;
328 if (answer->capability > 0 && !capable(answer->capability))
332 if (!inet_netns_ok(net, protocol))
335 sock->ops = answer->ops;
336 answer_prot = answer->prot;
337 answer_no_check = answer->no_check;
338 answer_flags = answer->flags;
341 WARN_ON(answer_prot->slab == NULL);
344 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
349 sk->sk_no_check = answer_no_check;
350 if (INET_PROTOSW_REUSE & answer_flags)
354 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
356 if (SOCK_RAW == sock->type) {
357 inet->num = protocol;
358 if (IPPROTO_RAW == protocol)
362 if (ipv4_config.no_pmtu_disc)
363 inet->pmtudisc = IP_PMTUDISC_DONT;
365 inet->pmtudisc = IP_PMTUDISC_WANT;
369 sock_init_data(sock, sk);
371 sk->sk_destruct = inet_sock_destruct;
372 sk->sk_family = PF_INET;
373 sk->sk_protocol = protocol;
374 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
380 inet->mc_list = NULL;
382 sk_refcnt_debug_inc(sk);
385 /* It assumes that any protocol which allows
386 * the user to assign a number at socket
387 * creation time automatically
390 inet->sport = htons(inet->num);
391 /* Add to protocol hash chains. */
392 sk->sk_prot->hash(sk);
395 if (sk->sk_prot->init) {
396 err = sk->sk_prot->init(sk);
398 sk_common_release(sk);
409 * The peer socket should always be NULL (or else). When we call this
410 * function we are destroying the object and from then on nobody
411 * should refer to it.
413 int inet_release(struct socket *sock)
415 struct sock *sk = sock->sk;
420 /* Applications forget to leave groups before exiting */
421 ip_mc_drop_socket(sk);
423 /* If linger is set, we don't return until the close
424 * is complete. Otherwise we return immediately. The
425 * actually closing is done the same either way.
427 * If the close is due to the process exiting, we never
431 if (sock_flag(sk, SOCK_LINGER) &&
432 !(current->flags & PF_EXITING))
433 timeout = sk->sk_lingertime;
435 sk->sk_prot->close(sk, timeout);
440 /* It is off by default, see below. */
441 int sysctl_ip_nonlocal_bind __read_mostly;
443 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
445 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
446 struct sock *sk = sock->sk;
447 struct inet_sock *inet = inet_sk(sk);
452 /* If the socket has its own bind function then use it. (RAW) */
453 if (sk->sk_prot->bind) {
454 err = sk->sk_prot->bind(sk, uaddr, addr_len);
458 if (addr_len < sizeof(struct sockaddr_in))
461 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
463 /* Not specified by any standard per-se, however it breaks too
464 * many applications when removed. It is unfortunate since
465 * allowing applications to make a non-local bind solves
466 * several problems with systems using dynamic addressing.
467 * (ie. your servers still start up even if your ISDN link
468 * is temporarily down)
470 err = -EADDRNOTAVAIL;
471 if (!sysctl_ip_nonlocal_bind &&
472 !(inet->freebind || inet->transparent) &&
473 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
474 chk_addr_ret != RTN_LOCAL &&
475 chk_addr_ret != RTN_MULTICAST &&
476 chk_addr_ret != RTN_BROADCAST)
479 snum = ntohs(addr->sin_port);
481 if (snum && snum < PROT_SOCK && !capable(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->num)
496 goto out_release_sock;
498 inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
499 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
500 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->saddr = inet->rcv_saddr = 0;
506 goto out_release_sock;
510 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
512 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
513 inet->sport = htons(inet->num);
524 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
525 int addr_len, int flags)
527 struct sock *sk = sock->sk;
529 if (uaddr->sa_family == AF_UNSPEC)
530 return sk->sk_prot->disconnect(sk, flags);
532 if (!inet_sk(sk)->num && inet_autobind(sk))
534 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
537 static long inet_wait_for_connect(struct sock *sk, long timeo)
541 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
543 /* Basic assumption: if someone sets sk->sk_err, he _must_
544 * change state of the socket from TCP_SYN_*.
545 * Connect() does not allow to get error notifications
546 * without closing the socket.
548 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
550 timeo = schedule_timeout(timeo);
552 if (signal_pending(current) || !timeo)
554 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
556 finish_wait(sk->sk_sleep, &wait);
561 * Connect to a remote host. There is regrettably still a little
562 * TCP 'magic' in here.
564 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
565 int addr_len, int flags)
567 struct sock *sk = sock->sk;
573 if (uaddr->sa_family == AF_UNSPEC) {
574 err = sk->sk_prot->disconnect(sk, flags);
575 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
579 switch (sock->state) {
588 /* Fall out of switch with err, set for this state */
592 if (sk->sk_state != TCP_CLOSE)
595 err = sk->sk_prot->connect(sk, uaddr, addr_len);
599 sock->state = SS_CONNECTING;
601 /* Just entered SS_CONNECTING state; the only
602 * difference is that return value in non-blocking
603 * case is EINPROGRESS, rather than EALREADY.
609 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
611 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
612 /* Error code is set above */
613 if (!timeo || !inet_wait_for_connect(sk, timeo))
616 err = sock_intr_errno(timeo);
617 if (signal_pending(current))
621 /* Connection was closed by RST, timeout, ICMP error
622 * or another process disconnected us.
624 if (sk->sk_state == TCP_CLOSE)
627 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
628 * and error was received after socket entered established state.
629 * Hence, it is handled normally after connect() return successfully.
632 sock->state = SS_CONNECTED;
639 err = sock_error(sk) ? : -ECONNABORTED;
640 sock->state = SS_UNCONNECTED;
641 if (sk->sk_prot->disconnect(sk, flags))
642 sock->state = SS_DISCONNECTING;
647 * Accept a pending connection. The TCP layer now gives BSD semantics.
650 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
652 struct sock *sk1 = sock->sk;
654 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
661 WARN_ON(!((1 << sk2->sk_state) &
662 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
664 sock_graft(sk2, newsock);
666 newsock->state = SS_CONNECTED;
675 * This does both peername and sockname.
677 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
678 int *uaddr_len, int peer)
680 struct sock *sk = sock->sk;
681 struct inet_sock *inet = inet_sk(sk);
682 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
684 sin->sin_family = AF_INET;
687 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
690 sin->sin_port = inet->dport;
691 sin->sin_addr.s_addr = inet->daddr;
693 __be32 addr = inet->rcv_saddr;
696 sin->sin_port = inet->sport;
697 sin->sin_addr.s_addr = addr;
699 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
700 *uaddr_len = sizeof(*sin);
704 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
707 struct sock *sk = sock->sk;
709 /* We may need to bind the socket. */
710 if (!inet_sk(sk)->num && inet_autobind(sk))
713 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
717 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
719 struct sock *sk = sock->sk;
721 /* We may need to bind the socket. */
722 if (!inet_sk(sk)->num && inet_autobind(sk))
725 if (sk->sk_prot->sendpage)
726 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
727 return sock_no_sendpage(sock, page, offset, size, flags);
731 int inet_shutdown(struct socket *sock, int how)
733 struct sock *sk = sock->sk;
736 /* This should really check to make sure
737 * the socket is a TCP socket. (WHY AC...)
739 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
742 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
746 if (sock->state == SS_CONNECTING) {
747 if ((1 << sk->sk_state) &
748 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
749 sock->state = SS_DISCONNECTING;
751 sock->state = SS_CONNECTED;
754 switch (sk->sk_state) {
757 /* Hack to wake up other listeners, who can poll for
758 POLLHUP, even on eg. unconnected UDP sockets -- RR */
760 sk->sk_shutdown |= how;
761 if (sk->sk_prot->shutdown)
762 sk->sk_prot->shutdown(sk, how);
765 /* Remaining two branches are temporary solution for missing
766 * close() in multithreaded environment. It is _not_ a good idea,
767 * but we have no choice until close() is repaired at VFS level.
770 if (!(how & RCV_SHUTDOWN))
774 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
775 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
779 /* Wake up anyone sleeping in poll. */
780 sk->sk_state_change(sk);
786 * ioctl() calls you can issue on an INET socket. Most of these are
787 * device configuration and stuff and very rarely used. Some ioctls
788 * pass on to the socket itself.
790 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
791 * loads the devconfigure module does its configuring and unloads it.
792 * There's a good 20K of config code hanging around the kernel.
795 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
797 struct sock *sk = sock->sk;
799 struct net *net = sock_net(sk);
803 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
806 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
811 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
816 err = arp_ioctl(net, cmd, (void __user *)arg);
829 err = devinet_ioctl(net, cmd, (void __user *)arg);
832 if (sk->sk_prot->ioctl)
833 err = sk->sk_prot->ioctl(sk, cmd, arg);
841 const struct proto_ops inet_stream_ops = {
843 .owner = THIS_MODULE,
844 .release = inet_release,
846 .connect = inet_stream_connect,
847 .socketpair = sock_no_socketpair,
848 .accept = inet_accept,
849 .getname = inet_getname,
852 .listen = inet_listen,
853 .shutdown = inet_shutdown,
854 .setsockopt = sock_common_setsockopt,
855 .getsockopt = sock_common_getsockopt,
856 .sendmsg = tcp_sendmsg,
857 .recvmsg = sock_common_recvmsg,
858 .mmap = sock_no_mmap,
859 .sendpage = tcp_sendpage,
860 .splice_read = tcp_splice_read,
862 .compat_setsockopt = compat_sock_common_setsockopt,
863 .compat_getsockopt = compat_sock_common_getsockopt,
867 const struct proto_ops inet_dgram_ops = {
869 .owner = THIS_MODULE,
870 .release = inet_release,
872 .connect = inet_dgram_connect,
873 .socketpair = sock_no_socketpair,
874 .accept = sock_no_accept,
875 .getname = inet_getname,
878 .listen = sock_no_listen,
879 .shutdown = inet_shutdown,
880 .setsockopt = sock_common_setsockopt,
881 .getsockopt = sock_common_getsockopt,
882 .sendmsg = inet_sendmsg,
883 .recvmsg = sock_common_recvmsg,
884 .mmap = sock_no_mmap,
885 .sendpage = inet_sendpage,
887 .compat_setsockopt = compat_sock_common_setsockopt,
888 .compat_getsockopt = compat_sock_common_getsockopt,
893 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
896 static const struct proto_ops inet_sockraw_ops = {
898 .owner = THIS_MODULE,
899 .release = inet_release,
901 .connect = inet_dgram_connect,
902 .socketpair = sock_no_socketpair,
903 .accept = sock_no_accept,
904 .getname = inet_getname,
905 .poll = datagram_poll,
907 .listen = sock_no_listen,
908 .shutdown = inet_shutdown,
909 .setsockopt = sock_common_setsockopt,
910 .getsockopt = sock_common_getsockopt,
911 .sendmsg = inet_sendmsg,
912 .recvmsg = sock_common_recvmsg,
913 .mmap = sock_no_mmap,
914 .sendpage = inet_sendpage,
916 .compat_setsockopt = compat_sock_common_setsockopt,
917 .compat_getsockopt = compat_sock_common_getsockopt,
921 static struct net_proto_family inet_family_ops = {
923 .create = inet_create,
924 .owner = THIS_MODULE,
927 /* Upon startup we insert all the elements in inetsw_array[] into
928 * the linked list inetsw.
930 static struct inet_protosw inetsw_array[] =
934 .protocol = IPPROTO_TCP,
936 .ops = &inet_stream_ops,
939 .flags = INET_PROTOSW_PERMANENT |
945 .protocol = IPPROTO_UDP,
947 .ops = &inet_dgram_ops,
949 .no_check = UDP_CSUM_DEFAULT,
950 .flags = INET_PROTOSW_PERMANENT,
956 .protocol = IPPROTO_IP, /* wild card */
958 .ops = &inet_sockraw_ops,
959 .capability = CAP_NET_RAW,
960 .no_check = UDP_CSUM_DEFAULT,
961 .flags = INET_PROTOSW_REUSE,
965 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
967 void inet_register_protosw(struct inet_protosw *p)
969 struct list_head *lh;
970 struct inet_protosw *answer;
971 int protocol = p->protocol;
972 struct list_head *last_perm;
974 spin_lock_bh(&inetsw_lock);
976 if (p->type >= SOCK_MAX)
979 /* If we are trying to override a permanent protocol, bail. */
981 last_perm = &inetsw[p->type];
982 list_for_each(lh, &inetsw[p->type]) {
983 answer = list_entry(lh, struct inet_protosw, list);
985 /* Check only the non-wild match. */
986 if (INET_PROTOSW_PERMANENT & answer->flags) {
987 if (protocol == answer->protocol)
997 /* Add the new entry after the last permanent entry if any, so that
998 * the new entry does not override a permanent entry when matched with
999 * a wild-card protocol. But it is allowed to override any existing
1000 * non-permanent entry. This means that when we remove this entry, the
1001 * system automatically returns to the old behavior.
1003 list_add_rcu(&p->list, last_perm);
1005 spin_unlock_bh(&inetsw_lock);
1012 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1018 "Ignoring attempt to register invalid socket type %d.\n",
1023 void inet_unregister_protosw(struct inet_protosw *p)
1025 if (INET_PROTOSW_PERMANENT & p->flags) {
1027 "Attempt to unregister permanent protocol %d.\n",
1030 spin_lock_bh(&inetsw_lock);
1031 list_del_rcu(&p->list);
1032 spin_unlock_bh(&inetsw_lock);
1039 * Shall we try to damage output packets if routing dev changes?
1042 int sysctl_ip_dynaddr __read_mostly;
1044 static int inet_sk_reselect_saddr(struct sock *sk)
1046 struct inet_sock *inet = inet_sk(sk);
1049 __be32 old_saddr = inet->saddr;
1051 __be32 daddr = inet->daddr;
1053 if (inet->opt && inet->opt->srr)
1054 daddr = inet->opt->faddr;
1056 /* Query new route. */
1057 err = ip_route_connect(&rt, daddr, 0,
1059 sk->sk_bound_dev_if,
1061 inet->sport, inet->dport, sk, 0);
1065 sk_setup_caps(sk, &rt->u.dst);
1067 new_saddr = rt->rt_src;
1069 if (new_saddr == old_saddr)
1072 if (sysctl_ip_dynaddr > 1) {
1073 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1074 __func__, &old_saddr, &new_saddr);
1077 inet->saddr = inet->rcv_saddr = new_saddr;
1080 * XXX The only one ugly spot where we need to
1081 * XXX really change the sockets identity after
1082 * XXX it has entered the hashes. -DaveM
1084 * Besides that, it does not check for connection
1085 * uniqueness. Wait for troubles.
1087 __sk_prot_rehash(sk);
1091 int inet_sk_rebuild_header(struct sock *sk)
1093 struct inet_sock *inet = inet_sk(sk);
1094 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1098 /* Route is OK, nothing to do. */
1103 daddr = inet->daddr;
1104 if (inet->opt && inet->opt->srr)
1105 daddr = inet->opt->faddr;
1108 .oif = sk->sk_bound_dev_if,
1112 .saddr = inet->saddr,
1113 .tos = RT_CONN_FLAGS(sk),
1116 .proto = sk->sk_protocol,
1117 .flags = inet_sk_flowi_flags(sk),
1120 .sport = inet->sport,
1121 .dport = inet->dport,
1126 security_sk_classify_flow(sk, &fl);
1127 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1130 sk_setup_caps(sk, &rt->u.dst);
1132 /* Routing failed... */
1133 sk->sk_route_caps = 0;
1135 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1136 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1138 if (!sysctl_ip_dynaddr ||
1139 sk->sk_state != TCP_SYN_SENT ||
1140 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1141 (err = inet_sk_reselect_saddr(sk)) != 0)
1142 sk->sk_err_soft = -err;
1148 EXPORT_SYMBOL(inet_sk_rebuild_header);
1150 static int inet_gso_send_check(struct sk_buff *skb)
1153 struct net_protocol *ops;
1158 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1163 if (ihl < sizeof(*iph))
1166 if (unlikely(!pskb_may_pull(skb, ihl)))
1169 __skb_pull(skb, ihl);
1170 skb_reset_transport_header(skb);
1172 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1173 err = -EPROTONOSUPPORT;
1176 ops = rcu_dereference(inet_protos[proto]);
1177 if (likely(ops && ops->gso_send_check))
1178 err = ops->gso_send_check(skb);
1185 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1187 struct sk_buff *segs = ERR_PTR(-EINVAL);
1189 struct net_protocol *ops;
1194 if (!(features & NETIF_F_V4_CSUM))
1195 features &= ~NETIF_F_SG;
1197 if (unlikely(skb_shinfo(skb)->gso_type &
1205 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1210 if (ihl < sizeof(*iph))
1213 if (unlikely(!pskb_may_pull(skb, ihl)))
1216 __skb_pull(skb, ihl);
1217 skb_reset_transport_header(skb);
1219 id = ntohs(iph->id);
1220 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1221 segs = ERR_PTR(-EPROTONOSUPPORT);
1224 ops = rcu_dereference(inet_protos[proto]);
1225 if (likely(ops && ops->gso_segment))
1226 segs = ops->gso_segment(skb, features);
1229 if (!segs || IS_ERR(segs))
1235 iph->id = htons(id++);
1236 iph->tot_len = htons(skb->len - skb->mac_len);
1238 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1239 } while ((skb = skb->next));
1245 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1246 struct sk_buff *skb)
1248 struct net_protocol *ops;
1249 struct sk_buff **pp = NULL;
1256 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1260 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1263 ops = rcu_dereference(inet_protos[proto]);
1264 if (!ops || !ops->gro_receive)
1267 if (iph->version != 4 || iph->ihl != 5)
1270 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1273 flush = ntohs(iph->tot_len) != skb->len ||
1274 iph->frag_off != htons(IP_DF);
1275 id = ntohs(iph->id);
1277 for (p = *head; p; p = p->next) {
1280 if (!NAPI_GRO_CB(p)->same_flow)
1285 if (iph->protocol != iph2->protocol ||
1286 iph->tos != iph2->tos ||
1287 memcmp(&iph->saddr, &iph2->saddr, 8)) {
1288 NAPI_GRO_CB(p)->same_flow = 0;
1292 /* All fields must match except length and checksum. */
1293 NAPI_GRO_CB(p)->flush |=
1294 memcmp(&iph->frag_off, &iph2->frag_off, 4) ||
1295 (u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) != id;
1297 NAPI_GRO_CB(p)->flush |= flush;
1300 NAPI_GRO_CB(skb)->flush |= flush;
1301 __skb_pull(skb, sizeof(*iph));
1302 skb_reset_transport_header(skb);
1304 pp = ops->gro_receive(head, skb);
1310 NAPI_GRO_CB(skb)->flush |= flush;
1315 static int inet_gro_complete(struct sk_buff *skb)
1317 struct net_protocol *ops;
1318 struct iphdr *iph = ip_hdr(skb);
1319 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1321 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1323 csum_replace2(&iph->check, iph->tot_len, newlen);
1324 iph->tot_len = newlen;
1327 ops = rcu_dereference(inet_protos[proto]);
1328 if (WARN_ON(!ops || !ops->gro_complete))
1331 err = ops->gro_complete(skb);
1339 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1340 unsigned short type, unsigned char protocol,
1343 struct socket *sock;
1344 int rc = sock_create_kern(family, type, protocol, &sock);
1348 (*sk)->sk_allocation = GFP_ATOMIC;
1350 * Unhash it so that IP input processing does not even see it,
1351 * we do not wish this socket to see incoming packets.
1353 (*sk)->sk_prot->unhash(*sk);
1355 sk_change_net(*sk, net);
1360 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1362 unsigned long snmp_fold_field(void *mib[], int offt)
1364 unsigned long res = 0;
1367 for_each_possible_cpu(i) {
1368 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1369 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1373 EXPORT_SYMBOL_GPL(snmp_fold_field);
1375 int snmp_mib_init(void *ptr[2], size_t mibsize)
1377 BUG_ON(ptr == NULL);
1378 ptr[0] = __alloc_percpu(mibsize);
1381 ptr[1] = __alloc_percpu(mibsize);
1386 free_percpu(ptr[0]);
1391 EXPORT_SYMBOL_GPL(snmp_mib_init);
1393 void snmp_mib_free(void *ptr[2])
1395 BUG_ON(ptr == NULL);
1396 free_percpu(ptr[0]);
1397 free_percpu(ptr[1]);
1398 ptr[0] = ptr[1] = NULL;
1400 EXPORT_SYMBOL_GPL(snmp_mib_free);
1402 #ifdef CONFIG_IP_MULTICAST
1403 static struct net_protocol igmp_protocol = {
1404 .handler = igmp_rcv,
1409 static struct net_protocol tcp_protocol = {
1410 .handler = tcp_v4_rcv,
1411 .err_handler = tcp_v4_err,
1412 .gso_send_check = tcp_v4_gso_send_check,
1413 .gso_segment = tcp_tso_segment,
1414 .gro_receive = tcp4_gro_receive,
1415 .gro_complete = tcp4_gro_complete,
1420 static struct net_protocol udp_protocol = {
1422 .err_handler = udp_err,
1427 static struct net_protocol icmp_protocol = {
1428 .handler = icmp_rcv,
1433 static __net_init int ipv4_mib_init_net(struct net *net)
1435 if (snmp_mib_init((void **)net->mib.tcp_statistics,
1436 sizeof(struct tcp_mib)) < 0)
1438 if (snmp_mib_init((void **)net->mib.ip_statistics,
1439 sizeof(struct ipstats_mib)) < 0)
1441 if (snmp_mib_init((void **)net->mib.net_statistics,
1442 sizeof(struct linux_mib)) < 0)
1444 if (snmp_mib_init((void **)net->mib.udp_statistics,
1445 sizeof(struct udp_mib)) < 0)
1447 if (snmp_mib_init((void **)net->mib.udplite_statistics,
1448 sizeof(struct udp_mib)) < 0)
1449 goto err_udplite_mib;
1450 if (snmp_mib_init((void **)net->mib.icmp_statistics,
1451 sizeof(struct icmp_mib)) < 0)
1453 if (snmp_mib_init((void **)net->mib.icmpmsg_statistics,
1454 sizeof(struct icmpmsg_mib)) < 0)
1455 goto err_icmpmsg_mib;
1461 snmp_mib_free((void **)net->mib.icmp_statistics);
1463 snmp_mib_free((void **)net->mib.udplite_statistics);
1465 snmp_mib_free((void **)net->mib.udp_statistics);
1467 snmp_mib_free((void **)net->mib.net_statistics);
1469 snmp_mib_free((void **)net->mib.ip_statistics);
1471 snmp_mib_free((void **)net->mib.tcp_statistics);
1476 static __net_exit void ipv4_mib_exit_net(struct net *net)
1478 snmp_mib_free((void **)net->mib.icmpmsg_statistics);
1479 snmp_mib_free((void **)net->mib.icmp_statistics);
1480 snmp_mib_free((void **)net->mib.udplite_statistics);
1481 snmp_mib_free((void **)net->mib.udp_statistics);
1482 snmp_mib_free((void **)net->mib.net_statistics);
1483 snmp_mib_free((void **)net->mib.ip_statistics);
1484 snmp_mib_free((void **)net->mib.tcp_statistics);
1487 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1488 .init = ipv4_mib_init_net,
1489 .exit = ipv4_mib_exit_net,
1492 static int __init init_ipv4_mibs(void)
1494 return register_pernet_subsys(&ipv4_mib_ops);
1497 static int ipv4_proc_init(void);
1500 * IP protocol layer initialiser
1503 static struct packet_type ip_packet_type = {
1504 .type = __constant_htons(ETH_P_IP),
1506 .gso_send_check = inet_gso_send_check,
1507 .gso_segment = inet_gso_segment,
1508 .gro_receive = inet_gro_receive,
1509 .gro_complete = inet_gro_complete,
1512 static int __init inet_init(void)
1514 struct sk_buff *dummy_skb;
1515 struct inet_protosw *q;
1516 struct list_head *r;
1519 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1521 rc = proto_register(&tcp_prot, 1);
1525 rc = proto_register(&udp_prot, 1);
1527 goto out_unregister_tcp_proto;
1529 rc = proto_register(&raw_prot, 1);
1531 goto out_unregister_udp_proto;
1534 * Tell SOCKET that we are alive...
1537 (void)sock_register(&inet_family_ops);
1539 #ifdef CONFIG_SYSCTL
1540 ip_static_sysctl_init();
1544 * Add all the base protocols.
1547 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1548 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1549 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1550 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1551 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1552 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1553 #ifdef CONFIG_IP_MULTICAST
1554 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1555 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1558 /* Register the socket-side information for inet_create. */
1559 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1562 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1563 inet_register_protosw(q);
1566 * Set the ARP module up
1572 * Set the IP module up
1579 /* Setup TCP slab cache for open requests. */
1582 /* Setup UDP memory threshold */
1585 /* Add UDP-Lite (RFC 3828) */
1586 udplite4_register();
1589 * Set the ICMP layer up
1592 if (icmp_init() < 0)
1593 panic("Failed to create the ICMP control socket.\n");
1596 * Initialise the multicast router
1598 #if defined(CONFIG_IP_MROUTE)
1600 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1603 * Initialise per-cpu ipv4 mibs
1606 if (init_ipv4_mibs())
1607 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1613 dev_add_pack(&ip_packet_type);
1618 out_unregister_udp_proto:
1619 proto_unregister(&udp_prot);
1620 out_unregister_tcp_proto:
1621 proto_unregister(&tcp_prot);
1625 fs_initcall(inet_init);
1627 /* ------------------------------------------------------------------------ */
1629 #ifdef CONFIG_PROC_FS
1630 static int __init ipv4_proc_init(void)
1634 if (raw_proc_init())
1636 if (tcp4_proc_init())
1638 if (udp4_proc_init())
1640 if (ip_misc_proc_init())
1655 #else /* CONFIG_PROC_FS */
1656 static int __init ipv4_proc_init(void)
1660 #endif /* CONFIG_PROC_FS */
1662 MODULE_ALIAS_NETPROTO(PF_INET);
1664 EXPORT_SYMBOL(inet_accept);
1665 EXPORT_SYMBOL(inet_bind);
1666 EXPORT_SYMBOL(inet_dgram_connect);
1667 EXPORT_SYMBOL(inet_dgram_ops);
1668 EXPORT_SYMBOL(inet_getname);
1669 EXPORT_SYMBOL(inet_ioctl);
1670 EXPORT_SYMBOL(inet_listen);
1671 EXPORT_SYMBOL(inet_register_protosw);
1672 EXPORT_SYMBOL(inet_release);
1673 EXPORT_SYMBOL(inet_sendmsg);
1674 EXPORT_SYMBOL(inet_shutdown);
1675 EXPORT_SYMBOL(inet_sock_destruct);
1676 EXPORT_SYMBOL(inet_stream_connect);
1677 EXPORT_SYMBOL(inet_stream_ops);
1678 EXPORT_SYMBOL(inet_unregister_protosw);
1679 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);