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>
98 #include <net/protocol.h>
100 #include <net/route.h>
101 #include <net/ip_fib.h>
102 #include <net/inet_connection_sock.h>
105 #include <net/udplite.h>
106 #include <linux/skbuff.h>
107 #include <net/sock.h>
109 #include <net/icmp.h>
110 #include <net/ipip.h>
111 #include <net/inet_common.h>
112 #include <net/xfrm.h>
113 #include <net/net_namespace.h>
114 #ifdef CONFIG_IP_MROUTE
115 #include <linux/mroute.h>
118 DEFINE_SNMP_STAT(struct linux_mib
, net_statistics
) __read_mostly
;
120 extern void ip_mc_drop_socket(struct sock
*sk
);
122 /* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
125 static struct list_head inetsw
[SOCK_MAX
];
126 static DEFINE_SPINLOCK(inetsw_lock
);
128 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 printk("Attempt to release TCP socket in state %d %p\n",
148 if (!sock_flag(sk
, SOCK_DEAD
)) {
149 printk("Attempt to release alive inet socket %p\n", sk
);
153 BUG_TRAP(!atomic_read(&sk
->sk_rmem_alloc
));
154 BUG_TRAP(!atomic_read(&sk
->sk_wmem_alloc
));
155 BUG_TRAP(!sk
->sk_wmem_queued
);
156 BUG_TRAP(!sk
->sk_forward_alloc
);
159 dst_release(sk
->sk_dst_cache
);
160 sk_refcnt_debug_dec(sk
);
164 * The routines beyond this point handle the behaviour of an AF_INET
165 * socket object. Mostly it punts to the subprotocols of IP to do
170 * Automatically bind an unbound socket.
173 static int inet_autobind(struct sock
*sk
)
175 struct inet_sock
*inet
;
176 /* We may need to bind the socket. */
180 if (sk
->sk_prot
->get_port(sk
, 0)) {
184 inet
->sport
= htons(inet
->num
);
191 * Move a socket into listening state.
193 int inet_listen(struct socket
*sock
, int backlog
)
195 struct sock
*sk
= sock
->sk
;
196 unsigned char old_state
;
202 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_STREAM
)
205 old_state
= sk
->sk_state
;
206 if (!((1 << old_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
209 /* Really, if the socket is already in listen state
210 * we can only allow the backlog to be adjusted.
212 if (old_state
!= TCP_LISTEN
) {
213 err
= inet_csk_listen_start(sk
, backlog
);
217 sk
->sk_max_ack_backlog
= backlog
;
225 u32 inet_ehash_secret __read_mostly
;
226 EXPORT_SYMBOL(inet_ehash_secret
);
229 * inet_ehash_secret must be set exactly once
230 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
232 void build_ehash_secret(void)
236 get_random_bytes(&rnd
, sizeof(rnd
));
238 spin_lock_bh(&inetsw_lock
);
239 if (!inet_ehash_secret
)
240 inet_ehash_secret
= rnd
;
241 spin_unlock_bh(&inetsw_lock
);
243 EXPORT_SYMBOL(build_ehash_secret
);
245 static inline int inet_netns_ok(struct net
*net
, int protocol
)
248 struct net_protocol
*ipprot
;
250 if (net
== &init_net
)
253 hash
= protocol
& (MAX_INET_PROTOS
- 1);
254 ipprot
= rcu_dereference(inet_protos
[hash
]);
259 return ipprot
->netns_ok
;
263 * Create an inet socket.
266 static int inet_create(struct net
*net
, struct socket
*sock
, int protocol
)
270 struct inet_protosw
*answer
;
271 struct inet_sock
*inet
;
272 struct proto
*answer_prot
;
273 unsigned char answer_flags
;
274 char answer_no_check
;
275 int try_loading_module
= 0;
278 if (sock
->type
!= SOCK_RAW
&&
279 sock
->type
!= SOCK_DGRAM
&&
281 build_ehash_secret();
283 sock
->state
= SS_UNCONNECTED
;
285 /* Look for the requested type/protocol pair. */
288 err
= -ESOCKTNOSUPPORT
;
290 list_for_each_rcu(p
, &inetsw
[sock
->type
]) {
291 answer
= list_entry(p
, struct inet_protosw
, list
);
293 /* Check the non-wild match. */
294 if (protocol
== answer
->protocol
) {
295 if (protocol
!= IPPROTO_IP
)
298 /* Check for the two wild cases. */
299 if (IPPROTO_IP
== protocol
) {
300 protocol
= answer
->protocol
;
303 if (IPPROTO_IP
== answer
->protocol
)
306 err
= -EPROTONOSUPPORT
;
310 if (unlikely(answer
== NULL
)) {
311 if (try_loading_module
< 2) {
314 * Be more specific, e.g. net-pf-2-proto-132-type-1
315 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
317 if (++try_loading_module
== 1)
318 request_module("net-pf-%d-proto-%d-type-%d",
319 PF_INET
, protocol
, sock
->type
);
321 * Fall back to generic, e.g. net-pf-2-proto-132
322 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
325 request_module("net-pf-%d-proto-%d",
327 goto lookup_protocol
;
333 if (answer
->capability
> 0 && !capable(answer
->capability
))
337 if (!inet_netns_ok(net
, protocol
))
340 sock
->ops
= answer
->ops
;
341 answer_prot
= answer
->prot
;
342 answer_no_check
= answer
->no_check
;
343 answer_flags
= answer
->flags
;
346 BUG_TRAP(answer_prot
->slab
!= NULL
);
349 sk
= sk_alloc(net
, PF_INET
, GFP_KERNEL
, answer_prot
);
354 sk
->sk_no_check
= answer_no_check
;
355 if (INET_PROTOSW_REUSE
& answer_flags
)
359 inet
->is_icsk
= (INET_PROTOSW_ICSK
& answer_flags
) != 0;
361 if (SOCK_RAW
== sock
->type
) {
362 inet
->num
= protocol
;
363 if (IPPROTO_RAW
== protocol
)
367 if (ipv4_config
.no_pmtu_disc
)
368 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
370 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
374 sock_init_data(sock
, sk
);
376 sk
->sk_destruct
= inet_sock_destruct
;
377 sk
->sk_family
= PF_INET
;
378 sk
->sk_protocol
= protocol
;
379 sk
->sk_backlog_rcv
= sk
->sk_prot
->backlog_rcv
;
385 inet
->mc_list
= NULL
;
387 sk_refcnt_debug_inc(sk
);
390 /* It assumes that any protocol which allows
391 * the user to assign a number at socket
392 * creation time automatically
395 inet
->sport
= htons(inet
->num
);
396 /* Add to protocol hash chains. */
397 sk
->sk_prot
->hash(sk
);
400 if (sk
->sk_prot
->init
) {
401 err
= sk
->sk_prot
->init(sk
);
403 sk_common_release(sk
);
414 * The peer socket should always be NULL (or else). When we call this
415 * function we are destroying the object and from then on nobody
416 * should refer to it.
418 int inet_release(struct socket
*sock
)
420 struct sock
*sk
= sock
->sk
;
425 /* Applications forget to leave groups before exiting */
426 ip_mc_drop_socket(sk
);
428 /* If linger is set, we don't return until the close
429 * is complete. Otherwise we return immediately. The
430 * actually closing is done the same either way.
432 * If the close is due to the process exiting, we never
436 if (sock_flag(sk
, SOCK_LINGER
) &&
437 !(current
->flags
& PF_EXITING
))
438 timeout
= sk
->sk_lingertime
;
440 sk
->sk_prot
->close(sk
, timeout
);
445 /* It is off by default, see below. */
446 int sysctl_ip_nonlocal_bind __read_mostly
;
448 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
450 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
451 struct sock
*sk
= sock
->sk
;
452 struct inet_sock
*inet
= inet_sk(sk
);
457 /* If the socket has its own bind function then use it. (RAW) */
458 if (sk
->sk_prot
->bind
) {
459 err
= sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
463 if (addr_len
< sizeof(struct sockaddr_in
))
466 chk_addr_ret
= inet_addr_type(sock_net(sk
), addr
->sin_addr
.s_addr
);
468 /* Not specified by any standard per-se, however it breaks too
469 * many applications when removed. It is unfortunate since
470 * allowing applications to make a non-local bind solves
471 * several problems with systems using dynamic addressing.
472 * (ie. your servers still start up even if your ISDN link
473 * is temporarily down)
475 err
= -EADDRNOTAVAIL
;
476 if (!sysctl_ip_nonlocal_bind
&&
478 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
) &&
479 chk_addr_ret
!= RTN_LOCAL
&&
480 chk_addr_ret
!= RTN_MULTICAST
&&
481 chk_addr_ret
!= RTN_BROADCAST
)
484 snum
= ntohs(addr
->sin_port
);
486 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
489 /* We keep a pair of addresses. rcv_saddr is the one
490 * used by hash lookups, and saddr is used for transmit.
492 * In the BSD API these are the same except where it
493 * would be illegal to use them (multicast/broadcast) in
494 * which case the sending device address is used.
498 /* Check these errors (active socket, double bind). */
500 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->num
)
501 goto out_release_sock
;
503 inet
->rcv_saddr
= inet
->saddr
= addr
->sin_addr
.s_addr
;
504 if (chk_addr_ret
== RTN_MULTICAST
|| chk_addr_ret
== RTN_BROADCAST
)
505 inet
->saddr
= 0; /* Use device */
507 /* Make sure we are allowed to bind here. */
508 if (sk
->sk_prot
->get_port(sk
, snum
)) {
509 inet
->saddr
= inet
->rcv_saddr
= 0;
511 goto out_release_sock
;
515 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
517 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
518 inet
->sport
= htons(inet
->num
);
529 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
* uaddr
,
530 int addr_len
, int flags
)
532 struct sock
*sk
= sock
->sk
;
534 if (uaddr
->sa_family
== AF_UNSPEC
)
535 return sk
->sk_prot
->disconnect(sk
, flags
);
537 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
539 return sk
->sk_prot
->connect(sk
, (struct sockaddr
*)uaddr
, addr_len
);
542 static long inet_wait_for_connect(struct sock
*sk
, long timeo
)
546 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
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
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
561 finish_wait(sk
->sk_sleep
, &wait
);
566 * Connect to a remote host. There is regrettably still a little
567 * TCP 'magic' in here.
569 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
570 int addr_len
, int flags
)
572 struct sock
*sk
= sock
->sk
;
578 if (uaddr
->sa_family
== AF_UNSPEC
) {
579 err
= sk
->sk_prot
->disconnect(sk
, flags
);
580 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
584 switch (sock
->state
) {
593 /* Fall out of switch with err, set for this state */
597 if (sk
->sk_state
!= TCP_CLOSE
)
600 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
604 sock
->state
= SS_CONNECTING
;
606 /* Just entered SS_CONNECTING state; the only
607 * difference is that return value in non-blocking
608 * case is EINPROGRESS, rather than EALREADY.
614 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
616 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
617 /* Error code is set above */
618 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
))
621 err
= sock_intr_errno(timeo
);
622 if (signal_pending(current
))
626 /* Connection was closed by RST, timeout, ICMP error
627 * or another process disconnected us.
629 if (sk
->sk_state
== TCP_CLOSE
)
632 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
633 * and error was received after socket entered established state.
634 * Hence, it is handled normally after connect() return successfully.
637 sock
->state
= SS_CONNECTED
;
644 err
= sock_error(sk
) ? : -ECONNABORTED
;
645 sock
->state
= SS_UNCONNECTED
;
646 if (sk
->sk_prot
->disconnect(sk
, flags
))
647 sock
->state
= SS_DISCONNECTING
;
652 * Accept a pending connection. The TCP layer now gives BSD semantics.
655 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
657 struct sock
*sk1
= sock
->sk
;
659 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
);
666 BUG_TRAP((1 << sk2
->sk_state
) &
667 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_CLOSE
));
669 sock_graft(sk2
, newsock
);
671 newsock
->state
= SS_CONNECTED
;
680 * This does both peername and sockname.
682 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
683 int *uaddr_len
, int peer
)
685 struct sock
*sk
= sock
->sk
;
686 struct inet_sock
*inet
= inet_sk(sk
);
687 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
689 sin
->sin_family
= AF_INET
;
692 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
695 sin
->sin_port
= inet
->dport
;
696 sin
->sin_addr
.s_addr
= inet
->daddr
;
698 __be32 addr
= inet
->rcv_saddr
;
701 sin
->sin_port
= inet
->sport
;
702 sin
->sin_addr
.s_addr
= addr
;
704 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
705 *uaddr_len
= sizeof(*sin
);
709 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
712 struct sock
*sk
= sock
->sk
;
714 /* We may need to bind the socket. */
715 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
718 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
722 static ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
, size_t size
, int flags
)
724 struct sock
*sk
= sock
->sk
;
726 /* We may need to bind the socket. */
727 if (!inet_sk(sk
)->num
&& inet_autobind(sk
))
730 if (sk
->sk_prot
->sendpage
)
731 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
732 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
736 int inet_shutdown(struct socket
*sock
, int how
)
738 struct sock
*sk
= sock
->sk
;
741 /* This should really check to make sure
742 * the socket is a TCP socket. (WHY AC...)
744 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
747 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
751 if (sock
->state
== SS_CONNECTING
) {
752 if ((1 << sk
->sk_state
) &
753 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
754 sock
->state
= SS_DISCONNECTING
;
756 sock
->state
= SS_CONNECTED
;
759 switch (sk
->sk_state
) {
762 /* Hack to wake up other listeners, who can poll for
763 POLLHUP, even on eg. unconnected UDP sockets -- RR */
765 sk
->sk_shutdown
|= how
;
766 if (sk
->sk_prot
->shutdown
)
767 sk
->sk_prot
->shutdown(sk
, how
);
770 /* Remaining two branches are temporary solution for missing
771 * close() in multithreaded environment. It is _not_ a good idea,
772 * but we have no choice until close() is repaired at VFS level.
775 if (!(how
& RCV_SHUTDOWN
))
779 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
780 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
784 /* Wake up anyone sleeping in poll. */
785 sk
->sk_state_change(sk
);
791 * ioctl() calls you can issue on an INET socket. Most of these are
792 * device configuration and stuff and very rarely used. Some ioctls
793 * pass on to the socket itself.
795 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
796 * loads the devconfigure module does its configuring and unloads it.
797 * There's a good 20K of config code hanging around the kernel.
800 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
802 struct sock
*sk
= sock
->sk
;
804 struct net
*net
= sock_net(sk
);
808 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
811 err
= sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
816 err
= ip_rt_ioctl(net
, cmd
, (void __user
*)arg
);
821 err
= arp_ioctl(net
, cmd
, (void __user
*)arg
);
834 err
= devinet_ioctl(net
, cmd
, (void __user
*)arg
);
837 if (sk
->sk_prot
->ioctl
)
838 err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
);
846 const struct proto_ops inet_stream_ops
= {
848 .owner
= THIS_MODULE
,
849 .release
= inet_release
,
851 .connect
= inet_stream_connect
,
852 .socketpair
= sock_no_socketpair
,
853 .accept
= inet_accept
,
854 .getname
= inet_getname
,
857 .listen
= inet_listen
,
858 .shutdown
= inet_shutdown
,
859 .setsockopt
= sock_common_setsockopt
,
860 .getsockopt
= sock_common_getsockopt
,
861 .sendmsg
= tcp_sendmsg
,
862 .recvmsg
= sock_common_recvmsg
,
863 .mmap
= sock_no_mmap
,
864 .sendpage
= tcp_sendpage
,
865 .splice_read
= tcp_splice_read
,
867 .compat_setsockopt
= compat_sock_common_setsockopt
,
868 .compat_getsockopt
= compat_sock_common_getsockopt
,
872 const struct proto_ops inet_dgram_ops
= {
874 .owner
= THIS_MODULE
,
875 .release
= inet_release
,
877 .connect
= inet_dgram_connect
,
878 .socketpair
= sock_no_socketpair
,
879 .accept
= sock_no_accept
,
880 .getname
= inet_getname
,
883 .listen
= sock_no_listen
,
884 .shutdown
= inet_shutdown
,
885 .setsockopt
= sock_common_setsockopt
,
886 .getsockopt
= sock_common_getsockopt
,
887 .sendmsg
= inet_sendmsg
,
888 .recvmsg
= sock_common_recvmsg
,
889 .mmap
= sock_no_mmap
,
890 .sendpage
= inet_sendpage
,
892 .compat_setsockopt
= compat_sock_common_setsockopt
,
893 .compat_getsockopt
= compat_sock_common_getsockopt
,
898 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
901 static const struct proto_ops inet_sockraw_ops
= {
903 .owner
= THIS_MODULE
,
904 .release
= inet_release
,
906 .connect
= inet_dgram_connect
,
907 .socketpair
= sock_no_socketpair
,
908 .accept
= sock_no_accept
,
909 .getname
= inet_getname
,
910 .poll
= datagram_poll
,
912 .listen
= sock_no_listen
,
913 .shutdown
= inet_shutdown
,
914 .setsockopt
= sock_common_setsockopt
,
915 .getsockopt
= sock_common_getsockopt
,
916 .sendmsg
= inet_sendmsg
,
917 .recvmsg
= sock_common_recvmsg
,
918 .mmap
= sock_no_mmap
,
919 .sendpage
= inet_sendpage
,
921 .compat_setsockopt
= compat_sock_common_setsockopt
,
922 .compat_getsockopt
= compat_sock_common_getsockopt
,
926 static struct net_proto_family inet_family_ops
= {
928 .create
= inet_create
,
929 .owner
= THIS_MODULE
,
932 /* Upon startup we insert all the elements in inetsw_array[] into
933 * the linked list inetsw.
935 static struct inet_protosw inetsw_array
[] =
939 .protocol
= IPPROTO_TCP
,
941 .ops
= &inet_stream_ops
,
944 .flags
= INET_PROTOSW_PERMANENT
|
950 .protocol
= IPPROTO_UDP
,
952 .ops
= &inet_dgram_ops
,
954 .no_check
= UDP_CSUM_DEFAULT
,
955 .flags
= INET_PROTOSW_PERMANENT
,
961 .protocol
= IPPROTO_IP
, /* wild card */
963 .ops
= &inet_sockraw_ops
,
964 .capability
= CAP_NET_RAW
,
965 .no_check
= UDP_CSUM_DEFAULT
,
966 .flags
= INET_PROTOSW_REUSE
,
970 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
972 void inet_register_protosw(struct inet_protosw
*p
)
974 struct list_head
*lh
;
975 struct inet_protosw
*answer
;
976 int protocol
= p
->protocol
;
977 struct list_head
*last_perm
;
979 spin_lock_bh(&inetsw_lock
);
981 if (p
->type
>= SOCK_MAX
)
984 /* If we are trying to override a permanent protocol, bail. */
986 last_perm
= &inetsw
[p
->type
];
987 list_for_each(lh
, &inetsw
[p
->type
]) {
988 answer
= list_entry(lh
, struct inet_protosw
, list
);
990 /* Check only the non-wild match. */
991 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
992 if (protocol
== answer
->protocol
)
1002 /* Add the new entry after the last permanent entry if any, so that
1003 * the new entry does not override a permanent entry when matched with
1004 * a wild-card protocol. But it is allowed to override any existing
1005 * non-permanent entry. This means that when we remove this entry, the
1006 * system automatically returns to the old behavior.
1008 list_add_rcu(&p
->list
, last_perm
);
1010 spin_unlock_bh(&inetsw_lock
);
1017 printk(KERN_ERR
"Attempt to override permanent protocol %d.\n",
1023 "Ignoring attempt to register invalid socket type %d.\n",
1028 void inet_unregister_protosw(struct inet_protosw
*p
)
1030 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
1032 "Attempt to unregister permanent protocol %d.\n",
1035 spin_lock_bh(&inetsw_lock
);
1036 list_del_rcu(&p
->list
);
1037 spin_unlock_bh(&inetsw_lock
);
1044 * Shall we try to damage output packets if routing dev changes?
1047 int sysctl_ip_dynaddr __read_mostly
;
1049 static int inet_sk_reselect_saddr(struct sock
*sk
)
1051 struct inet_sock
*inet
= inet_sk(sk
);
1054 __be32 old_saddr
= inet
->saddr
;
1056 __be32 daddr
= inet
->daddr
;
1058 if (inet
->opt
&& inet
->opt
->srr
)
1059 daddr
= inet
->opt
->faddr
;
1061 /* Query new route. */
1062 err
= ip_route_connect(&rt
, daddr
, 0,
1064 sk
->sk_bound_dev_if
,
1066 inet
->sport
, inet
->dport
, sk
, 0);
1070 sk_setup_caps(sk
, &rt
->u
.dst
);
1072 new_saddr
= rt
->rt_src
;
1074 if (new_saddr
== old_saddr
)
1077 if (sysctl_ip_dynaddr
> 1) {
1078 printk(KERN_INFO
"%s(): shifting inet->"
1079 "saddr from " NIPQUAD_FMT
" to " NIPQUAD_FMT
"\n",
1082 NIPQUAD(new_saddr
));
1085 inet
->saddr
= inet
->rcv_saddr
= new_saddr
;
1088 * XXX The only one ugly spot where we need to
1089 * XXX really change the sockets identity after
1090 * XXX it has entered the hashes. -DaveM
1092 * Besides that, it does not check for connection
1093 * uniqueness. Wait for troubles.
1095 __sk_prot_rehash(sk
);
1099 int inet_sk_rebuild_header(struct sock
*sk
)
1101 struct inet_sock
*inet
= inet_sk(sk
);
1102 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1106 /* Route is OK, nothing to do. */
1111 daddr
= inet
->daddr
;
1112 if (inet
->opt
&& inet
->opt
->srr
)
1113 daddr
= inet
->opt
->faddr
;
1116 .oif
= sk
->sk_bound_dev_if
,
1120 .saddr
= inet
->saddr
,
1121 .tos
= RT_CONN_FLAGS(sk
),
1124 .proto
= sk
->sk_protocol
,
1127 .sport
= inet
->sport
,
1128 .dport
= inet
->dport
,
1133 security_sk_classify_flow(sk
, &fl
);
1134 err
= ip_route_output_flow(sock_net(sk
), &rt
, &fl
, sk
, 0);
1137 sk_setup_caps(sk
, &rt
->u
.dst
);
1139 /* Routing failed... */
1140 sk
->sk_route_caps
= 0;
1142 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1143 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1145 if (!sysctl_ip_dynaddr
||
1146 sk
->sk_state
!= TCP_SYN_SENT
||
1147 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1148 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1149 sk
->sk_err_soft
= -err
;
1155 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1157 static int inet_gso_send_check(struct sk_buff
*skb
)
1160 struct net_protocol
*ops
;
1165 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1170 if (ihl
< sizeof(*iph
))
1173 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1176 __skb_pull(skb
, ihl
);
1177 skb_reset_transport_header(skb
);
1179 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1180 err
= -EPROTONOSUPPORT
;
1183 ops
= rcu_dereference(inet_protos
[proto
]);
1184 if (likely(ops
&& ops
->gso_send_check
))
1185 err
= ops
->gso_send_check(skb
);
1192 static struct sk_buff
*inet_gso_segment(struct sk_buff
*skb
, int features
)
1194 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1196 struct net_protocol
*ops
;
1201 if (!(features
& NETIF_F_V4_CSUM
))
1202 features
&= ~NETIF_F_SG
;
1204 if (unlikely(skb_shinfo(skb
)->gso_type
&
1212 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1217 if (ihl
< sizeof(*iph
))
1220 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1223 __skb_pull(skb
, ihl
);
1224 skb_reset_transport_header(skb
);
1226 id
= ntohs(iph
->id
);
1227 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1228 segs
= ERR_PTR(-EPROTONOSUPPORT
);
1231 ops
= rcu_dereference(inet_protos
[proto
]);
1232 if (likely(ops
&& ops
->gso_segment
))
1233 segs
= ops
->gso_segment(skb
, features
);
1236 if (!segs
|| IS_ERR(segs
))
1242 iph
->id
= htons(id
++);
1243 iph
->tot_len
= htons(skb
->len
- skb
->mac_len
);
1245 iph
->check
= ip_fast_csum(skb_network_header(skb
), iph
->ihl
);
1246 } while ((skb
= skb
->next
));
1252 int inet_ctl_sock_create(struct sock
**sk
, unsigned short family
,
1253 unsigned short type
, unsigned char protocol
,
1256 struct socket
*sock
;
1257 int rc
= sock_create_kern(family
, type
, protocol
, &sock
);
1261 (*sk
)->sk_allocation
= GFP_ATOMIC
;
1263 * Unhash it so that IP input processing does not even see it,
1264 * we do not wish this socket to see incoming packets.
1266 (*sk
)->sk_prot
->unhash(*sk
);
1268 sk_change_net(*sk
, net
);
1273 EXPORT_SYMBOL_GPL(inet_ctl_sock_create
);
1275 unsigned long snmp_fold_field(void *mib
[], int offt
)
1277 unsigned long res
= 0;
1280 for_each_possible_cpu(i
) {
1281 res
+= *(((unsigned long *) per_cpu_ptr(mib
[0], i
)) + offt
);
1282 res
+= *(((unsigned long *) per_cpu_ptr(mib
[1], i
)) + offt
);
1286 EXPORT_SYMBOL_GPL(snmp_fold_field
);
1288 int snmp_mib_init(void *ptr
[2], size_t mibsize
)
1290 BUG_ON(ptr
== NULL
);
1291 ptr
[0] = __alloc_percpu(mibsize
);
1294 ptr
[1] = __alloc_percpu(mibsize
);
1299 free_percpu(ptr
[0]);
1304 EXPORT_SYMBOL_GPL(snmp_mib_init
);
1306 void snmp_mib_free(void *ptr
[2])
1308 BUG_ON(ptr
== NULL
);
1309 free_percpu(ptr
[0]);
1310 free_percpu(ptr
[1]);
1311 ptr
[0] = ptr
[1] = NULL
;
1313 EXPORT_SYMBOL_GPL(snmp_mib_free
);
1315 #ifdef CONFIG_IP_MULTICAST
1316 static struct net_protocol igmp_protocol
= {
1317 .handler
= igmp_rcv
,
1321 static struct net_protocol tcp_protocol
= {
1322 .handler
= tcp_v4_rcv
,
1323 .err_handler
= tcp_v4_err
,
1324 .gso_send_check
= tcp_v4_gso_send_check
,
1325 .gso_segment
= tcp_tso_segment
,
1330 static struct net_protocol udp_protocol
= {
1332 .err_handler
= udp_err
,
1337 static struct net_protocol icmp_protocol
= {
1338 .handler
= icmp_rcv
,
1343 static __net_init
int ipv4_mib_init_net(struct net
*net
)
1345 if (snmp_mib_init((void **)net
->mib
.tcp_statistics
,
1346 sizeof(struct tcp_mib
)) < 0)
1356 static __net_exit
void ipv4_mib_exit_net(struct net
*net
)
1358 snmp_mib_free((void **)net
->mib
.tcp_statistics
);
1361 static __net_initdata
struct pernet_operations ipv4_mib_ops
= {
1362 .init
= ipv4_mib_init_net
,
1363 .exit
= ipv4_mib_exit_net
,
1366 static int __init
init_ipv4_mibs(void)
1368 if (snmp_mib_init((void **)net_statistics
,
1369 sizeof(struct linux_mib
)) < 0)
1371 if (snmp_mib_init((void **)ip_statistics
,
1372 sizeof(struct ipstats_mib
)) < 0)
1374 if (snmp_mib_init((void **)icmp_statistics
,
1375 sizeof(struct icmp_mib
)) < 0)
1377 if (snmp_mib_init((void **)icmpmsg_statistics
,
1378 sizeof(struct icmpmsg_mib
)) < 0)
1379 goto err_icmpmsg_mib
;
1380 if (snmp_mib_init((void **)udp_statistics
,
1381 sizeof(struct udp_mib
)) < 0)
1383 if (snmp_mib_init((void **)udplite_statistics
,
1384 sizeof(struct udp_mib
)) < 0)
1385 goto err_udplite_mib
;
1387 if (register_pernet_subsys(&ipv4_mib_ops
))
1393 snmp_mib_free((void **)udplite_statistics
);
1395 snmp_mib_free((void **)udp_statistics
);
1397 snmp_mib_free((void **)icmpmsg_statistics
);
1399 snmp_mib_free((void **)icmp_statistics
);
1401 snmp_mib_free((void **)ip_statistics
);
1403 snmp_mib_free((void **)net_statistics
);
1408 static int ipv4_proc_init(void);
1411 * IP protocol layer initialiser
1414 static struct packet_type ip_packet_type
= {
1415 .type
= __constant_htons(ETH_P_IP
),
1417 .gso_send_check
= inet_gso_send_check
,
1418 .gso_segment
= inet_gso_segment
,
1421 static int __init
inet_init(void)
1423 struct sk_buff
*dummy_skb
;
1424 struct inet_protosw
*q
;
1425 struct list_head
*r
;
1428 BUILD_BUG_ON(sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
));
1430 rc
= proto_register(&tcp_prot
, 1);
1434 rc
= proto_register(&udp_prot
, 1);
1436 goto out_unregister_tcp_proto
;
1438 rc
= proto_register(&raw_prot
, 1);
1440 goto out_unregister_udp_proto
;
1443 * Tell SOCKET that we are alive...
1446 (void)sock_register(&inet_family_ops
);
1449 * Add all the base protocols.
1452 if (inet_add_protocol(&icmp_protocol
, IPPROTO_ICMP
) < 0)
1453 printk(KERN_CRIT
"inet_init: Cannot add ICMP protocol\n");
1454 if (inet_add_protocol(&udp_protocol
, IPPROTO_UDP
) < 0)
1455 printk(KERN_CRIT
"inet_init: Cannot add UDP protocol\n");
1456 if (inet_add_protocol(&tcp_protocol
, IPPROTO_TCP
) < 0)
1457 printk(KERN_CRIT
"inet_init: Cannot add TCP protocol\n");
1458 #ifdef CONFIG_IP_MULTICAST
1459 if (inet_add_protocol(&igmp_protocol
, IPPROTO_IGMP
) < 0)
1460 printk(KERN_CRIT
"inet_init: Cannot add IGMP protocol\n");
1463 /* Register the socket-side information for inet_create. */
1464 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1467 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1468 inet_register_protosw(q
);
1471 * Set the ARP module up
1477 * Set the IP module up
1484 /* Setup TCP slab cache for open requests. */
1487 /* Setup UDP memory threshold */
1490 /* Add UDP-Lite (RFC 3828) */
1491 udplite4_register();
1494 * Set the ICMP layer up
1497 if (icmp_init() < 0)
1498 panic("Failed to create the ICMP control socket.\n");
1501 * Initialise the multicast router
1503 #if defined(CONFIG_IP_MROUTE)
1505 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mroute\n");
1508 * Initialise per-cpu ipv4 mibs
1511 if (init_ipv4_mibs())
1512 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mibs\n");
1518 dev_add_pack(&ip_packet_type
);
1523 out_unregister_udp_proto
:
1524 proto_unregister(&udp_prot
);
1525 out_unregister_tcp_proto
:
1526 proto_unregister(&tcp_prot
);
1530 fs_initcall(inet_init
);
1532 /* ------------------------------------------------------------------------ */
1534 #ifdef CONFIG_PROC_FS
1535 static int __init
ipv4_proc_init(void)
1539 if (raw_proc_init())
1541 if (tcp4_proc_init())
1543 if (udp4_proc_init())
1545 if (ip_misc_proc_init())
1560 #else /* CONFIG_PROC_FS */
1561 static int __init
ipv4_proc_init(void)
1565 #endif /* CONFIG_PROC_FS */
1567 MODULE_ALIAS_NETPROTO(PF_INET
);
1569 EXPORT_SYMBOL(inet_accept
);
1570 EXPORT_SYMBOL(inet_bind
);
1571 EXPORT_SYMBOL(inet_dgram_connect
);
1572 EXPORT_SYMBOL(inet_dgram_ops
);
1573 EXPORT_SYMBOL(inet_getname
);
1574 EXPORT_SYMBOL(inet_ioctl
);
1575 EXPORT_SYMBOL(inet_listen
);
1576 EXPORT_SYMBOL(inet_register_protosw
);
1577 EXPORT_SYMBOL(inet_release
);
1578 EXPORT_SYMBOL(inet_sendmsg
);
1579 EXPORT_SYMBOL(inet_shutdown
);
1580 EXPORT_SYMBOL(inet_sock_destruct
);
1581 EXPORT_SYMBOL(inet_stream_connect
);
1582 EXPORT_SYMBOL(inet_stream_ops
);
1583 EXPORT_SYMBOL(inet_unregister_protosw
);
1584 EXPORT_SYMBOL(net_statistics
);
1585 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);