2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PF_INET protocol family socket handler.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
13 * Changes (see also sock.c)
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
68 #include <linux/err.h>
69 #include <linux/errno.h>
70 #include <linux/types.h>
71 #include <linux/socket.h>
73 #include <linux/kernel.h>
74 #include <linux/module.h>
75 #include <linux/sched.h>
76 #include <linux/timer.h>
77 #include <linux/string.h>
78 #include <linux/sockios.h>
79 #include <linux/net.h>
80 #include <linux/capability.h>
81 #include <linux/fcntl.h>
83 #include <linux/interrupt.h>
84 #include <linux/stat.h>
85 #include <linux/init.h>
86 #include <linux/poll.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/slab.h>
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/inetdevice.h>
97 #include <linux/netdevice.h>
98 #include <net/checksum.h>
100 #include <net/protocol.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
107 #include <net/udplite.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
111 #include <net/icmp.h>
112 #include <net/ipip.h>
113 #include <net/inet_common.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #ifdef CONFIG_IP_MROUTE
117 #include <linux/mroute.h>
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
;
128 EXPORT_SYMBOL(ipv4_config
);
130 /* New destruction routine */
132 void inet_sock_destruct(struct sock
*sk
)
134 struct inet_sock
*inet
= inet_sk(sk
);
136 __skb_queue_purge(&sk
->sk_receive_queue
);
137 __skb_queue_purge(&sk
->sk_error_queue
);
141 if (sk
->sk_type
== SOCK_STREAM
&& sk
->sk_state
!= TCP_CLOSE
) {
142 pr_err("Attempt to release TCP socket in state %d %p\n",
146 if (!sock_flag(sk
, SOCK_DEAD
)) {
147 pr_err("Attempt to release alive inet socket %p\n", sk
);
151 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
152 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
153 WARN_ON(sk
->sk_wmem_queued
);
154 WARN_ON(sk
->sk_forward_alloc
);
157 dst_release(rcu_dereference_check(sk
->sk_dst_cache
, 1));
158 sk_refcnt_debug_dec(sk
);
160 EXPORT_SYMBOL(inet_sock_destruct
);
163 * The routines beyond this point handle the behaviour of an AF_INET
164 * socket object. Mostly it punts to the subprotocols of IP to do
169 * Automatically bind an unbound socket.
172 static int inet_autobind(struct sock
*sk
)
174 struct inet_sock
*inet
;
175 /* We may need to bind the socket. */
178 if (!inet
->inet_num
) {
179 if (sk
->sk_prot
->get_port(sk
, 0)) {
183 inet
->inet_sport
= htons(inet
->inet_num
);
190 * Move a socket into listening state.
192 int inet_listen(struct socket
*sock
, int backlog
)
194 struct sock
*sk
= sock
->sk
;
195 unsigned char old_state
;
201 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_STREAM
)
204 old_state
= sk
->sk_state
;
205 if (!((1 << old_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
208 /* Really, if the socket is already in listen state
209 * we can only allow the backlog to be adjusted.
211 if (old_state
!= TCP_LISTEN
) {
212 err
= inet_csk_listen_start(sk
, backlog
);
216 sk
->sk_max_ack_backlog
= backlog
;
223 EXPORT_SYMBOL(inet_listen
);
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 const struct net_protocol
*ipprot
;
250 if (net_eq(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 (unlikely(!inet_ehash_secret
))
279 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
280 build_ehash_secret();
282 sock
->state
= SS_UNCONNECTED
;
284 /* Look for the requested type/protocol pair. */
286 err
= -ESOCKTNOSUPPORT
;
288 list_for_each_entry_rcu(answer
, &inetsw
[sock
->type
], list
) {
291 /* Check the non-wild match. */
292 if (protocol
== answer
->protocol
) {
293 if (protocol
!= IPPROTO_IP
)
296 /* Check for the two wild cases. */
297 if (IPPROTO_IP
== protocol
) {
298 protocol
= answer
->protocol
;
301 if (IPPROTO_IP
== answer
->protocol
)
304 err
= -EPROTONOSUPPORT
;
308 if (try_loading_module
< 2) {
311 * Be more specific, e.g. net-pf-2-proto-132-type-1
312 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
314 if (++try_loading_module
== 1)
315 request_module("net-pf-%d-proto-%d-type-%d",
316 PF_INET
, protocol
, sock
->type
);
318 * Fall back to generic, e.g. net-pf-2-proto-132
319 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
322 request_module("net-pf-%d-proto-%d",
324 goto lookup_protocol
;
330 if (sock
->type
== SOCK_RAW
&& !kern
&& !capable(CAP_NET_RAW
))
334 if (!inet_netns_ok(net
, protocol
))
337 sock
->ops
= answer
->ops
;
338 answer_prot
= answer
->prot
;
339 answer_no_check
= answer
->no_check
;
340 answer_flags
= answer
->flags
;
343 WARN_ON(answer_prot
->slab
== NULL
);
346 sk
= sk_alloc(net
, PF_INET
, GFP_KERNEL
, answer_prot
);
351 sk
->sk_no_check
= answer_no_check
;
352 if (INET_PROTOSW_REUSE
& answer_flags
)
356 inet
->is_icsk
= (INET_PROTOSW_ICSK
& answer_flags
) != 0;
358 if (SOCK_RAW
== sock
->type
) {
359 inet
->inet_num
= protocol
;
360 if (IPPROTO_RAW
== protocol
)
364 if (ipv4_config
.no_pmtu_disc
)
365 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
367 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
371 sock_init_data(sock
, sk
);
373 sk
->sk_destruct
= inet_sock_destruct
;
374 sk
->sk_protocol
= protocol
;
375 sk
->sk_backlog_rcv
= sk
->sk_prot
->backlog_rcv
;
382 inet
->mc_list
= NULL
;
384 sk_refcnt_debug_inc(sk
);
386 if (inet
->inet_num
) {
387 /* It assumes that any protocol which allows
388 * the user to assign a number at socket
389 * creation time automatically
392 inet
->inet_sport
= htons(inet
->inet_num
);
393 /* Add to protocol hash chains. */
394 sk
->sk_prot
->hash(sk
);
397 if (sk
->sk_prot
->init
) {
398 err
= sk
->sk_prot
->init(sk
);
400 sk_common_release(sk
);
411 * The peer socket should always be NULL (or else). When we call this
412 * function we are destroying the object and from then on nobody
413 * should refer to it.
415 int inet_release(struct socket
*sock
)
417 struct sock
*sk
= sock
->sk
;
422 inet_rps_reset_flow(sk
);
424 /* Applications forget to leave groups before exiting */
425 ip_mc_drop_socket(sk
);
427 /* If linger is set, we don't return until the close
428 * is complete. Otherwise we return immediately. The
429 * actually closing is done the same either way.
431 * If the close is due to the process exiting, we never
435 if (sock_flag(sk
, SOCK_LINGER
) &&
436 !(current
->flags
& PF_EXITING
))
437 timeout
= sk
->sk_lingertime
;
439 sk
->sk_prot
->close(sk
, timeout
);
443 EXPORT_SYMBOL(inet_release
);
445 /* It is off by default, see below. */
446 int sysctl_ip_nonlocal_bind __read_mostly
;
447 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);
449 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
451 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
452 struct sock
*sk
= sock
->sk
;
453 struct inet_sock
*inet
= inet_sk(sk
);
458 /* If the socket has its own bind function then use it. (RAW) */
459 if (sk
->sk_prot
->bind
) {
460 err
= sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
464 if (addr_len
< sizeof(struct sockaddr_in
))
467 chk_addr_ret
= inet_addr_type(sock_net(sk
), addr
->sin_addr
.s_addr
);
469 /* Not specified by any standard per-se, however it breaks too
470 * many applications when removed. It is unfortunate since
471 * allowing applications to make a non-local bind solves
472 * several problems with systems using dynamic addressing.
473 * (ie. your servers still start up even if your ISDN link
474 * is temporarily down)
476 err
= -EADDRNOTAVAIL
;
477 if (!sysctl_ip_nonlocal_bind
&&
478 !(inet
->freebind
|| inet
->transparent
) &&
479 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
) &&
480 chk_addr_ret
!= RTN_LOCAL
&&
481 chk_addr_ret
!= RTN_MULTICAST
&&
482 chk_addr_ret
!= RTN_BROADCAST
)
485 snum
= ntohs(addr
->sin_port
);
487 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
490 /* We keep a pair of addresses. rcv_saddr is the one
491 * used by hash lookups, and saddr is used for transmit.
493 * In the BSD API these are the same except where it
494 * would be illegal to use them (multicast/broadcast) in
495 * which case the sending device address is used.
499 /* Check these errors (active socket, double bind). */
501 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->inet_num
)
502 goto out_release_sock
;
504 inet
->inet_rcv_saddr
= inet
->inet_saddr
= addr
->sin_addr
.s_addr
;
505 if (chk_addr_ret
== RTN_MULTICAST
|| chk_addr_ret
== RTN_BROADCAST
)
506 inet
->inet_saddr
= 0; /* Use device */
508 /* Make sure we are allowed to bind here. */
509 if (sk
->sk_prot
->get_port(sk
, snum
)) {
510 inet
->inet_saddr
= inet
->inet_rcv_saddr
= 0;
512 goto out_release_sock
;
515 if (inet
->inet_rcv_saddr
)
516 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
518 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
519 inet
->inet_sport
= htons(inet
->inet_num
);
520 inet
->inet_daddr
= 0;
521 inet
->inet_dport
= 0;
529 EXPORT_SYMBOL(inet_bind
);
531 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
* uaddr
,
532 int addr_len
, int flags
)
534 struct sock
*sk
= sock
->sk
;
536 if (addr_len
< sizeof(uaddr
->sa_family
))
538 if (uaddr
->sa_family
== AF_UNSPEC
)
539 return sk
->sk_prot
->disconnect(sk
, flags
);
541 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
543 return sk
->sk_prot
->connect(sk
, (struct sockaddr
*)uaddr
, addr_len
);
545 EXPORT_SYMBOL(inet_dgram_connect
);
547 static long inet_wait_for_connect(struct sock
*sk
, long timeo
)
551 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
553 /* Basic assumption: if someone sets sk->sk_err, he _must_
554 * change state of the socket from TCP_SYN_*.
555 * Connect() does not allow to get error notifications
556 * without closing the socket.
558 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
560 timeo
= schedule_timeout(timeo
);
562 if (signal_pending(current
) || !timeo
)
564 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
566 finish_wait(sk
->sk_sleep
, &wait
);
571 * Connect to a remote host. There is regrettably still a little
572 * TCP 'magic' in here.
574 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
575 int addr_len
, int flags
)
577 struct sock
*sk
= sock
->sk
;
581 if (addr_len
< sizeof(uaddr
->sa_family
))
586 if (uaddr
->sa_family
== AF_UNSPEC
) {
587 err
= sk
->sk_prot
->disconnect(sk
, flags
);
588 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
592 switch (sock
->state
) {
601 /* Fall out of switch with err, set for this state */
605 if (sk
->sk_state
!= TCP_CLOSE
)
608 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
612 sock
->state
= SS_CONNECTING
;
614 /* Just entered SS_CONNECTING state; the only
615 * difference is that return value in non-blocking
616 * case is EINPROGRESS, rather than EALREADY.
622 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
624 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
625 /* Error code is set above */
626 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
))
629 err
= sock_intr_errno(timeo
);
630 if (signal_pending(current
))
634 /* Connection was closed by RST, timeout, ICMP error
635 * or another process disconnected us.
637 if (sk
->sk_state
== TCP_CLOSE
)
640 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
641 * and error was received after socket entered established state.
642 * Hence, it is handled normally after connect() return successfully.
645 sock
->state
= SS_CONNECTED
;
652 err
= sock_error(sk
) ? : -ECONNABORTED
;
653 sock
->state
= SS_UNCONNECTED
;
654 if (sk
->sk_prot
->disconnect(sk
, flags
))
655 sock
->state
= SS_DISCONNECTING
;
658 EXPORT_SYMBOL(inet_stream_connect
);
661 * Accept a pending connection. The TCP layer now gives BSD semantics.
664 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
666 struct sock
*sk1
= sock
->sk
;
668 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
);
675 WARN_ON(!((1 << sk2
->sk_state
) &
676 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_CLOSE
)));
678 sock_graft(sk2
, newsock
);
680 newsock
->state
= SS_CONNECTED
;
686 EXPORT_SYMBOL(inet_accept
);
690 * This does both peername and sockname.
692 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
693 int *uaddr_len
, int peer
)
695 struct sock
*sk
= sock
->sk
;
696 struct inet_sock
*inet
= inet_sk(sk
);
697 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, uaddr
);
699 sin
->sin_family
= AF_INET
;
701 if (!inet
->inet_dport
||
702 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
705 sin
->sin_port
= inet
->inet_dport
;
706 sin
->sin_addr
.s_addr
= inet
->inet_daddr
;
708 __be32 addr
= inet
->inet_rcv_saddr
;
710 addr
= inet
->inet_saddr
;
711 sin
->sin_port
= inet
->inet_sport
;
712 sin
->sin_addr
.s_addr
= addr
;
714 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
715 *uaddr_len
= sizeof(*sin
);
718 EXPORT_SYMBOL(inet_getname
);
720 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
723 struct sock
*sk
= sock
->sk
;
725 inet_rps_record_flow(sk
);
727 /* We may need to bind the socket. */
728 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
731 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
733 EXPORT_SYMBOL(inet_sendmsg
);
735 static ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
736 size_t size
, int flags
)
738 struct sock
*sk
= sock
->sk
;
740 inet_rps_record_flow(sk
);
742 /* We may need to bind the socket. */
743 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
746 if (sk
->sk_prot
->sendpage
)
747 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
748 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
751 int inet_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
752 size_t size
, int flags
)
754 struct sock
*sk
= sock
->sk
;
758 inet_rps_record_flow(sk
);
760 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
761 flags
& ~MSG_DONTWAIT
, &addr_len
);
763 msg
->msg_namelen
= addr_len
;
766 EXPORT_SYMBOL(inet_recvmsg
);
768 int inet_shutdown(struct socket
*sock
, int how
)
770 struct sock
*sk
= sock
->sk
;
773 /* This should really check to make sure
774 * the socket is a TCP socket. (WHY AC...)
776 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
779 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
783 if (sock
->state
== SS_CONNECTING
) {
784 if ((1 << sk
->sk_state
) &
785 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
786 sock
->state
= SS_DISCONNECTING
;
788 sock
->state
= SS_CONNECTED
;
791 switch (sk
->sk_state
) {
794 /* Hack to wake up other listeners, who can poll for
795 POLLHUP, even on eg. unconnected UDP sockets -- RR */
797 sk
->sk_shutdown
|= how
;
798 if (sk
->sk_prot
->shutdown
)
799 sk
->sk_prot
->shutdown(sk
, how
);
802 /* Remaining two branches are temporary solution for missing
803 * close() in multithreaded environment. It is _not_ a good idea,
804 * but we have no choice until close() is repaired at VFS level.
807 if (!(how
& RCV_SHUTDOWN
))
811 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
812 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
816 /* Wake up anyone sleeping in poll. */
817 sk
->sk_state_change(sk
);
821 EXPORT_SYMBOL(inet_shutdown
);
824 * ioctl() calls you can issue on an INET socket. Most of these are
825 * device configuration and stuff and very rarely used. Some ioctls
826 * pass on to the socket itself.
828 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
829 * loads the devconfigure module does its configuring and unloads it.
830 * There's a good 20K of config code hanging around the kernel.
833 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
835 struct sock
*sk
= sock
->sk
;
837 struct net
*net
= sock_net(sk
);
841 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
844 err
= sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
849 err
= ip_rt_ioctl(net
, cmd
, (void __user
*)arg
);
854 err
= arp_ioctl(net
, cmd
, (void __user
*)arg
);
867 err
= devinet_ioctl(net
, cmd
, (void __user
*)arg
);
870 if (sk
->sk_prot
->ioctl
)
871 err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
);
878 EXPORT_SYMBOL(inet_ioctl
);
880 const struct proto_ops inet_stream_ops
= {
882 .owner
= THIS_MODULE
,
883 .release
= inet_release
,
885 .connect
= inet_stream_connect
,
886 .socketpair
= sock_no_socketpair
,
887 .accept
= inet_accept
,
888 .getname
= inet_getname
,
891 .listen
= inet_listen
,
892 .shutdown
= inet_shutdown
,
893 .setsockopt
= sock_common_setsockopt
,
894 .getsockopt
= sock_common_getsockopt
,
895 .sendmsg
= tcp_sendmsg
,
896 .recvmsg
= inet_recvmsg
,
897 .mmap
= sock_no_mmap
,
898 .sendpage
= tcp_sendpage
,
899 .splice_read
= tcp_splice_read
,
901 .compat_setsockopt
= compat_sock_common_setsockopt
,
902 .compat_getsockopt
= compat_sock_common_getsockopt
,
905 EXPORT_SYMBOL(inet_stream_ops
);
907 const struct proto_ops inet_dgram_ops
= {
909 .owner
= THIS_MODULE
,
910 .release
= inet_release
,
912 .connect
= inet_dgram_connect
,
913 .socketpair
= sock_no_socketpair
,
914 .accept
= sock_no_accept
,
915 .getname
= inet_getname
,
918 .listen
= sock_no_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
,
927 .compat_setsockopt
= compat_sock_common_setsockopt
,
928 .compat_getsockopt
= compat_sock_common_getsockopt
,
931 EXPORT_SYMBOL(inet_dgram_ops
);
934 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
937 static const struct proto_ops inet_sockraw_ops
= {
939 .owner
= THIS_MODULE
,
940 .release
= inet_release
,
942 .connect
= inet_dgram_connect
,
943 .socketpair
= sock_no_socketpair
,
944 .accept
= sock_no_accept
,
945 .getname
= inet_getname
,
946 .poll
= datagram_poll
,
948 .listen
= sock_no_listen
,
949 .shutdown
= inet_shutdown
,
950 .setsockopt
= sock_common_setsockopt
,
951 .getsockopt
= sock_common_getsockopt
,
952 .sendmsg
= inet_sendmsg
,
953 .recvmsg
= inet_recvmsg
,
954 .mmap
= sock_no_mmap
,
955 .sendpage
= inet_sendpage
,
957 .compat_setsockopt
= compat_sock_common_setsockopt
,
958 .compat_getsockopt
= compat_sock_common_getsockopt
,
962 static const struct net_proto_family inet_family_ops
= {
964 .create
= inet_create
,
965 .owner
= THIS_MODULE
,
968 /* Upon startup we insert all the elements in inetsw_array[] into
969 * the linked list inetsw.
971 static struct inet_protosw inetsw_array
[] =
975 .protocol
= IPPROTO_TCP
,
977 .ops
= &inet_stream_ops
,
979 .flags
= INET_PROTOSW_PERMANENT
|
985 .protocol
= IPPROTO_UDP
,
987 .ops
= &inet_dgram_ops
,
988 .no_check
= UDP_CSUM_DEFAULT
,
989 .flags
= INET_PROTOSW_PERMANENT
,
995 .protocol
= IPPROTO_IP
, /* wild card */
997 .ops
= &inet_sockraw_ops
,
998 .no_check
= UDP_CSUM_DEFAULT
,
999 .flags
= INET_PROTOSW_REUSE
,
1003 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1005 void inet_register_protosw(struct inet_protosw
*p
)
1007 struct list_head
*lh
;
1008 struct inet_protosw
*answer
;
1009 int protocol
= p
->protocol
;
1010 struct list_head
*last_perm
;
1012 spin_lock_bh(&inetsw_lock
);
1014 if (p
->type
>= SOCK_MAX
)
1017 /* If we are trying to override a permanent protocol, bail. */
1019 last_perm
= &inetsw
[p
->type
];
1020 list_for_each(lh
, &inetsw
[p
->type
]) {
1021 answer
= list_entry(lh
, struct inet_protosw
, list
);
1023 /* Check only the non-wild match. */
1024 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
1025 if (protocol
== answer
->protocol
)
1035 /* Add the new entry after the last permanent entry if any, so that
1036 * the new entry does not override a permanent entry when matched with
1037 * a wild-card protocol. But it is allowed to override any existing
1038 * non-permanent entry. This means that when we remove this entry, the
1039 * system automatically returns to the old behavior.
1041 list_add_rcu(&p
->list
, last_perm
);
1043 spin_unlock_bh(&inetsw_lock
);
1048 printk(KERN_ERR
"Attempt to override permanent protocol %d.\n",
1054 "Ignoring attempt to register invalid socket type %d.\n",
1058 EXPORT_SYMBOL(inet_register_protosw
);
1060 void inet_unregister_protosw(struct inet_protosw
*p
)
1062 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
1064 "Attempt to unregister permanent protocol %d.\n",
1067 spin_lock_bh(&inetsw_lock
);
1068 list_del_rcu(&p
->list
);
1069 spin_unlock_bh(&inetsw_lock
);
1074 EXPORT_SYMBOL(inet_unregister_protosw
);
1077 * Shall we try to damage output packets if routing dev changes?
1080 int sysctl_ip_dynaddr __read_mostly
;
1082 static int inet_sk_reselect_saddr(struct sock
*sk
)
1084 struct inet_sock
*inet
= inet_sk(sk
);
1087 __be32 old_saddr
= inet
->inet_saddr
;
1089 __be32 daddr
= inet
->inet_daddr
;
1091 if (inet
->opt
&& inet
->opt
->srr
)
1092 daddr
= inet
->opt
->faddr
;
1094 /* Query new route. */
1095 err
= ip_route_connect(&rt
, daddr
, 0,
1097 sk
->sk_bound_dev_if
,
1099 inet
->inet_sport
, inet
->inet_dport
, sk
, 0);
1103 sk_setup_caps(sk
, &rt
->u
.dst
);
1105 new_saddr
= rt
->rt_src
;
1107 if (new_saddr
== old_saddr
)
1110 if (sysctl_ip_dynaddr
> 1) {
1111 printk(KERN_INFO
"%s(): shifting inet->saddr from %pI4 to %pI4\n",
1112 __func__
, &old_saddr
, &new_saddr
);
1115 inet
->inet_saddr
= inet
->inet_rcv_saddr
= new_saddr
;
1118 * XXX The only one ugly spot where we need to
1119 * XXX really change the sockets identity after
1120 * XXX it has entered the hashes. -DaveM
1122 * Besides that, it does not check for connection
1123 * uniqueness. Wait for troubles.
1125 __sk_prot_rehash(sk
);
1129 int inet_sk_rebuild_header(struct sock
*sk
)
1131 struct inet_sock
*inet
= inet_sk(sk
);
1132 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1136 /* Route is OK, nothing to do. */
1141 daddr
= inet
->inet_daddr
;
1142 if (inet
->opt
&& inet
->opt
->srr
)
1143 daddr
= inet
->opt
->faddr
;
1146 .oif
= sk
->sk_bound_dev_if
,
1147 .mark
= sk
->sk_mark
,
1151 .saddr
= inet
->inet_saddr
,
1152 .tos
= RT_CONN_FLAGS(sk
),
1155 .proto
= sk
->sk_protocol
,
1156 .flags
= inet_sk_flowi_flags(sk
),
1159 .sport
= inet
->inet_sport
,
1160 .dport
= inet
->inet_dport
,
1165 security_sk_classify_flow(sk
, &fl
);
1166 err
= ip_route_output_flow(sock_net(sk
), &rt
, &fl
, sk
, 0);
1169 sk_setup_caps(sk
, &rt
->u
.dst
);
1171 /* Routing failed... */
1172 sk
->sk_route_caps
= 0;
1174 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1175 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1177 if (!sysctl_ip_dynaddr
||
1178 sk
->sk_state
!= TCP_SYN_SENT
||
1179 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1180 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1181 sk
->sk_err_soft
= -err
;
1186 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1188 static int inet_gso_send_check(struct sk_buff
*skb
)
1191 const struct net_protocol
*ops
;
1196 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1201 if (ihl
< sizeof(*iph
))
1204 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1207 __skb_pull(skb
, ihl
);
1208 skb_reset_transport_header(skb
);
1210 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1211 err
= -EPROTONOSUPPORT
;
1214 ops
= rcu_dereference(inet_protos
[proto
]);
1215 if (likely(ops
&& ops
->gso_send_check
))
1216 err
= ops
->gso_send_check(skb
);
1223 static struct sk_buff
*inet_gso_segment(struct sk_buff
*skb
, int features
)
1225 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1227 const struct net_protocol
*ops
;
1231 unsigned int offset
= 0;
1233 if (!(features
& NETIF_F_V4_CSUM
))
1234 features
&= ~NETIF_F_SG
;
1236 if (unlikely(skb_shinfo(skb
)->gso_type
&
1244 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1249 if (ihl
< sizeof(*iph
))
1252 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1255 __skb_pull(skb
, ihl
);
1256 skb_reset_transport_header(skb
);
1258 id
= ntohs(iph
->id
);
1259 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1260 segs
= ERR_PTR(-EPROTONOSUPPORT
);
1263 ops
= rcu_dereference(inet_protos
[proto
]);
1264 if (likely(ops
&& ops
->gso_segment
))
1265 segs
= ops
->gso_segment(skb
, features
);
1268 if (!segs
|| IS_ERR(segs
))
1274 if (proto
== IPPROTO_UDP
) {
1275 iph
->id
= htons(id
);
1276 iph
->frag_off
= htons(offset
>> 3);
1277 if (skb
->next
!= NULL
)
1278 iph
->frag_off
|= htons(IP_MF
);
1279 offset
+= (skb
->len
- skb
->mac_len
- iph
->ihl
* 4);
1281 iph
->id
= htons(id
++);
1282 iph
->tot_len
= htons(skb
->len
- skb
->mac_len
);
1284 iph
->check
= ip_fast_csum(skb_network_header(skb
), iph
->ihl
);
1285 } while ((skb
= skb
->next
));
1291 static struct sk_buff
**inet_gro_receive(struct sk_buff
**head
,
1292 struct sk_buff
*skb
)
1294 const struct net_protocol
*ops
;
1295 struct sk_buff
**pp
= NULL
;
1304 off
= skb_gro_offset(skb
);
1305 hlen
= off
+ sizeof(*iph
);
1306 iph
= skb_gro_header_fast(skb
, off
);
1307 if (skb_gro_header_hard(skb
, hlen
)) {
1308 iph
= skb_gro_header_slow(skb
, hlen
, off
);
1313 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1316 ops
= rcu_dereference(inet_protos
[proto
]);
1317 if (!ops
|| !ops
->gro_receive
)
1320 if (*(u8
*)iph
!= 0x45)
1323 if (unlikely(ip_fast_csum((u8
*)iph
, iph
->ihl
)))
1326 id
= ntohl(*(u32
*)&iph
->id
);
1327 flush
= (u16
)((ntohl(*(u32
*)iph
) ^ skb_gro_len(skb
)) | (id
^ IP_DF
));
1330 for (p
= *head
; p
; p
= p
->next
) {
1333 if (!NAPI_GRO_CB(p
)->same_flow
)
1338 if ((iph
->protocol
^ iph2
->protocol
) |
1339 (iph
->tos
^ iph2
->tos
) |
1340 (iph
->saddr
^ iph2
->saddr
) |
1341 (iph
->daddr
^ iph2
->daddr
)) {
1342 NAPI_GRO_CB(p
)->same_flow
= 0;
1346 /* All fields must match except length and checksum. */
1347 NAPI_GRO_CB(p
)->flush
|=
1348 (iph
->ttl
^ iph2
->ttl
) |
1349 ((u16
)(ntohs(iph2
->id
) + NAPI_GRO_CB(p
)->count
) ^ id
);
1351 NAPI_GRO_CB(p
)->flush
|= flush
;
1354 NAPI_GRO_CB(skb
)->flush
|= flush
;
1355 skb_gro_pull(skb
, sizeof(*iph
));
1356 skb_set_transport_header(skb
, skb_gro_offset(skb
));
1358 pp
= ops
->gro_receive(head
, skb
);
1364 NAPI_GRO_CB(skb
)->flush
|= flush
;
1369 static int inet_gro_complete(struct sk_buff
*skb
)
1371 const struct net_protocol
*ops
;
1372 struct iphdr
*iph
= ip_hdr(skb
);
1373 int proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1375 __be16 newlen
= htons(skb
->len
- skb_network_offset(skb
));
1377 csum_replace2(&iph
->check
, iph
->tot_len
, newlen
);
1378 iph
->tot_len
= newlen
;
1381 ops
= rcu_dereference(inet_protos
[proto
]);
1382 if (WARN_ON(!ops
|| !ops
->gro_complete
))
1385 err
= ops
->gro_complete(skb
);
1393 int inet_ctl_sock_create(struct sock
**sk
, unsigned short family
,
1394 unsigned short type
, unsigned char protocol
,
1397 struct socket
*sock
;
1398 int rc
= sock_create_kern(family
, type
, protocol
, &sock
);
1402 (*sk
)->sk_allocation
= GFP_ATOMIC
;
1404 * Unhash it so that IP input processing does not even see it,
1405 * we do not wish this socket to see incoming packets.
1407 (*sk
)->sk_prot
->unhash(*sk
);
1409 sk_change_net(*sk
, net
);
1413 EXPORT_SYMBOL_GPL(inet_ctl_sock_create
);
1415 unsigned long snmp_fold_field(void __percpu
*mib
[], int offt
)
1417 unsigned long res
= 0;
1420 for_each_possible_cpu(i
) {
1421 res
+= *(((unsigned long *) per_cpu_ptr(mib
[0], i
)) + offt
);
1422 res
+= *(((unsigned long *) per_cpu_ptr(mib
[1], i
)) + offt
);
1426 EXPORT_SYMBOL_GPL(snmp_fold_field
);
1428 int snmp_mib_init(void __percpu
*ptr
[2], size_t mibsize
)
1430 BUG_ON(ptr
== NULL
);
1431 ptr
[0] = __alloc_percpu(mibsize
, __alignof__(unsigned long));
1434 ptr
[1] = __alloc_percpu(mibsize
, __alignof__(unsigned long));
1439 free_percpu(ptr
[0]);
1444 EXPORT_SYMBOL_GPL(snmp_mib_init
);
1446 void snmp_mib_free(void __percpu
*ptr
[2])
1448 BUG_ON(ptr
== NULL
);
1449 free_percpu(ptr
[0]);
1450 free_percpu(ptr
[1]);
1451 ptr
[0] = ptr
[1] = NULL
;
1453 EXPORT_SYMBOL_GPL(snmp_mib_free
);
1455 #ifdef CONFIG_IP_MULTICAST
1456 static const struct net_protocol igmp_protocol
= {
1457 .handler
= igmp_rcv
,
1462 static const struct net_protocol tcp_protocol
= {
1463 .handler
= tcp_v4_rcv
,
1464 .err_handler
= tcp_v4_err
,
1465 .gso_send_check
= tcp_v4_gso_send_check
,
1466 .gso_segment
= tcp_tso_segment
,
1467 .gro_receive
= tcp4_gro_receive
,
1468 .gro_complete
= tcp4_gro_complete
,
1473 static const struct net_protocol udp_protocol
= {
1475 .err_handler
= udp_err
,
1476 .gso_send_check
= udp4_ufo_send_check
,
1477 .gso_segment
= udp4_ufo_fragment
,
1482 static const struct net_protocol icmp_protocol
= {
1483 .handler
= icmp_rcv
,
1488 static __net_init
int ipv4_mib_init_net(struct net
*net
)
1490 if (snmp_mib_init((void __percpu
**)net
->mib
.tcp_statistics
,
1491 sizeof(struct tcp_mib
)) < 0)
1493 if (snmp_mib_init((void __percpu
**)net
->mib
.ip_statistics
,
1494 sizeof(struct ipstats_mib
)) < 0)
1496 if (snmp_mib_init((void __percpu
**)net
->mib
.net_statistics
,
1497 sizeof(struct linux_mib
)) < 0)
1499 if (snmp_mib_init((void __percpu
**)net
->mib
.udp_statistics
,
1500 sizeof(struct udp_mib
)) < 0)
1502 if (snmp_mib_init((void __percpu
**)net
->mib
.udplite_statistics
,
1503 sizeof(struct udp_mib
)) < 0)
1504 goto err_udplite_mib
;
1505 if (snmp_mib_init((void __percpu
**)net
->mib
.icmp_statistics
,
1506 sizeof(struct icmp_mib
)) < 0)
1508 if (snmp_mib_init((void __percpu
**)net
->mib
.icmpmsg_statistics
,
1509 sizeof(struct icmpmsg_mib
)) < 0)
1510 goto err_icmpmsg_mib
;
1516 snmp_mib_free((void __percpu
**)net
->mib
.icmp_statistics
);
1518 snmp_mib_free((void __percpu
**)net
->mib
.udplite_statistics
);
1520 snmp_mib_free((void __percpu
**)net
->mib
.udp_statistics
);
1522 snmp_mib_free((void __percpu
**)net
->mib
.net_statistics
);
1524 snmp_mib_free((void __percpu
**)net
->mib
.ip_statistics
);
1526 snmp_mib_free((void __percpu
**)net
->mib
.tcp_statistics
);
1531 static __net_exit
void ipv4_mib_exit_net(struct net
*net
)
1533 snmp_mib_free((void __percpu
**)net
->mib
.icmpmsg_statistics
);
1534 snmp_mib_free((void __percpu
**)net
->mib
.icmp_statistics
);
1535 snmp_mib_free((void __percpu
**)net
->mib
.udplite_statistics
);
1536 snmp_mib_free((void __percpu
**)net
->mib
.udp_statistics
);
1537 snmp_mib_free((void __percpu
**)net
->mib
.net_statistics
);
1538 snmp_mib_free((void __percpu
**)net
->mib
.ip_statistics
);
1539 snmp_mib_free((void __percpu
**)net
->mib
.tcp_statistics
);
1542 static __net_initdata
struct pernet_operations ipv4_mib_ops
= {
1543 .init
= ipv4_mib_init_net
,
1544 .exit
= ipv4_mib_exit_net
,
1547 static int __init
init_ipv4_mibs(void)
1549 return register_pernet_subsys(&ipv4_mib_ops
);
1552 static int ipv4_proc_init(void);
1555 * IP protocol layer initialiser
1558 static struct packet_type ip_packet_type __read_mostly
= {
1559 .type
= cpu_to_be16(ETH_P_IP
),
1561 .gso_send_check
= inet_gso_send_check
,
1562 .gso_segment
= inet_gso_segment
,
1563 .gro_receive
= inet_gro_receive
,
1564 .gro_complete
= inet_gro_complete
,
1567 static int __init
inet_init(void)
1569 struct sk_buff
*dummy_skb
;
1570 struct inet_protosw
*q
;
1571 struct list_head
*r
;
1574 BUILD_BUG_ON(sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
));
1576 rc
= proto_register(&tcp_prot
, 1);
1580 rc
= proto_register(&udp_prot
, 1);
1582 goto out_unregister_tcp_proto
;
1584 rc
= proto_register(&raw_prot
, 1);
1586 goto out_unregister_udp_proto
;
1589 * Tell SOCKET that we are alive...
1592 (void)sock_register(&inet_family_ops
);
1594 #ifdef CONFIG_SYSCTL
1595 ip_static_sysctl_init();
1599 * Add all the base protocols.
1602 if (inet_add_protocol(&icmp_protocol
, IPPROTO_ICMP
) < 0)
1603 printk(KERN_CRIT
"inet_init: Cannot add ICMP protocol\n");
1604 if (inet_add_protocol(&udp_protocol
, IPPROTO_UDP
) < 0)
1605 printk(KERN_CRIT
"inet_init: Cannot add UDP protocol\n");
1606 if (inet_add_protocol(&tcp_protocol
, IPPROTO_TCP
) < 0)
1607 printk(KERN_CRIT
"inet_init: Cannot add TCP protocol\n");
1608 #ifdef CONFIG_IP_MULTICAST
1609 if (inet_add_protocol(&igmp_protocol
, IPPROTO_IGMP
) < 0)
1610 printk(KERN_CRIT
"inet_init: Cannot add IGMP protocol\n");
1613 /* Register the socket-side information for inet_create. */
1614 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1617 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1618 inet_register_protosw(q
);
1621 * Set the ARP module up
1627 * Set the IP module up
1634 /* Setup TCP slab cache for open requests. */
1637 /* Setup UDP memory threshold */
1640 /* Add UDP-Lite (RFC 3828) */
1641 udplite4_register();
1644 * Set the ICMP layer up
1647 if (icmp_init() < 0)
1648 panic("Failed to create the ICMP control socket.\n");
1651 * Initialise the multicast router
1653 #if defined(CONFIG_IP_MROUTE)
1655 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mroute\n");
1658 * Initialise per-cpu ipv4 mibs
1661 if (init_ipv4_mibs())
1662 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mibs\n");
1668 dev_add_pack(&ip_packet_type
);
1673 out_unregister_udp_proto
:
1674 proto_unregister(&udp_prot
);
1675 out_unregister_tcp_proto
:
1676 proto_unregister(&tcp_prot
);
1680 fs_initcall(inet_init
);
1682 /* ------------------------------------------------------------------------ */
1684 #ifdef CONFIG_PROC_FS
1685 static int __init
ipv4_proc_init(void)
1689 if (raw_proc_init())
1691 if (tcp4_proc_init())
1693 if (udp4_proc_init())
1695 if (ip_misc_proc_init())
1710 #else /* CONFIG_PROC_FS */
1711 static int __init
ipv4_proc_init(void)
1715 #endif /* CONFIG_PROC_FS */
1717 MODULE_ALIAS_NETPROTO(PF_INET
);