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(sk
->sk_dst_cache
);
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 /* Applications forget to leave groups before exiting */
423 ip_mc_drop_socket(sk
);
425 /* If linger is set, we don't return until the close
426 * is complete. Otherwise we return immediately. The
427 * actually closing is done the same either way.
429 * If the close is due to the process exiting, we never
433 if (sock_flag(sk
, SOCK_LINGER
) &&
434 !(current
->flags
& PF_EXITING
))
435 timeout
= sk
->sk_lingertime
;
437 sk
->sk_prot
->close(sk
, timeout
);
441 EXPORT_SYMBOL(inet_release
);
443 /* It is off by default, see below. */
444 int sysctl_ip_nonlocal_bind __read_mostly
;
445 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);
447 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
449 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
450 struct sock
*sk
= sock
->sk
;
451 struct inet_sock
*inet
= inet_sk(sk
);
456 /* If the socket has its own bind function then use it. (RAW) */
457 if (sk
->sk_prot
->bind
) {
458 err
= sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
462 if (addr_len
< sizeof(struct sockaddr_in
))
465 chk_addr_ret
= inet_addr_type(sock_net(sk
), addr
->sin_addr
.s_addr
);
467 /* Not specified by any standard per-se, however it breaks too
468 * many applications when removed. It is unfortunate since
469 * allowing applications to make a non-local bind solves
470 * several problems with systems using dynamic addressing.
471 * (ie. your servers still start up even if your ISDN link
472 * is temporarily down)
474 err
= -EADDRNOTAVAIL
;
475 if (!sysctl_ip_nonlocal_bind
&&
476 !(inet
->freebind
|| inet
->transparent
) &&
477 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
) &&
478 chk_addr_ret
!= RTN_LOCAL
&&
479 chk_addr_ret
!= RTN_MULTICAST
&&
480 chk_addr_ret
!= RTN_BROADCAST
)
483 snum
= ntohs(addr
->sin_port
);
485 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
488 /* We keep a pair of addresses. rcv_saddr is the one
489 * used by hash lookups, and saddr is used for transmit.
491 * In the BSD API these are the same except where it
492 * would be illegal to use them (multicast/broadcast) in
493 * which case the sending device address is used.
497 /* Check these errors (active socket, double bind). */
499 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->inet_num
)
500 goto out_release_sock
;
502 inet
->inet_rcv_saddr
= inet
->inet_saddr
= addr
->sin_addr
.s_addr
;
503 if (chk_addr_ret
== RTN_MULTICAST
|| chk_addr_ret
== RTN_BROADCAST
)
504 inet
->inet_saddr
= 0; /* Use device */
506 /* Make sure we are allowed to bind here. */
507 if (sk
->sk_prot
->get_port(sk
, snum
)) {
508 inet
->inet_saddr
= inet
->inet_rcv_saddr
= 0;
510 goto out_release_sock
;
513 if (inet
->inet_rcv_saddr
)
514 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
516 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
517 inet
->inet_sport
= htons(inet
->inet_num
);
518 inet
->inet_daddr
= 0;
519 inet
->inet_dport
= 0;
527 EXPORT_SYMBOL(inet_bind
);
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 (addr_len
< sizeof(uaddr
->sa_family
))
536 if (uaddr
->sa_family
== AF_UNSPEC
)
537 return sk
->sk_prot
->disconnect(sk
, flags
);
539 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
541 return sk
->sk_prot
->connect(sk
, (struct sockaddr
*)uaddr
, addr_len
);
543 EXPORT_SYMBOL(inet_dgram_connect
);
545 static long inet_wait_for_connect(struct sock
*sk
, long timeo
)
549 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
551 /* Basic assumption: if someone sets sk->sk_err, he _must_
552 * change state of the socket from TCP_SYN_*.
553 * Connect() does not allow to get error notifications
554 * without closing the socket.
556 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
558 timeo
= schedule_timeout(timeo
);
560 if (signal_pending(current
) || !timeo
)
562 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
564 finish_wait(sk
->sk_sleep
, &wait
);
569 * Connect to a remote host. There is regrettably still a little
570 * TCP 'magic' in here.
572 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
573 int addr_len
, int flags
)
575 struct sock
*sk
= sock
->sk
;
579 if (addr_len
< sizeof(uaddr
->sa_family
))
584 if (uaddr
->sa_family
== AF_UNSPEC
) {
585 err
= sk
->sk_prot
->disconnect(sk
, flags
);
586 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
590 switch (sock
->state
) {
599 /* Fall out of switch with err, set for this state */
603 if (sk
->sk_state
!= TCP_CLOSE
)
606 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
610 sock
->state
= SS_CONNECTING
;
612 /* Just entered SS_CONNECTING state; the only
613 * difference is that return value in non-blocking
614 * case is EINPROGRESS, rather than EALREADY.
620 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
622 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
623 /* Error code is set above */
624 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
))
627 err
= sock_intr_errno(timeo
);
628 if (signal_pending(current
))
632 /* Connection was closed by RST, timeout, ICMP error
633 * or another process disconnected us.
635 if (sk
->sk_state
== TCP_CLOSE
)
638 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
639 * and error was received after socket entered established state.
640 * Hence, it is handled normally after connect() return successfully.
643 sock
->state
= SS_CONNECTED
;
650 err
= sock_error(sk
) ? : -ECONNABORTED
;
651 sock
->state
= SS_UNCONNECTED
;
652 if (sk
->sk_prot
->disconnect(sk
, flags
))
653 sock
->state
= SS_DISCONNECTING
;
656 EXPORT_SYMBOL(inet_stream_connect
);
659 * Accept a pending connection. The TCP layer now gives BSD semantics.
662 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
664 struct sock
*sk1
= sock
->sk
;
666 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
);
673 WARN_ON(!((1 << sk2
->sk_state
) &
674 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_CLOSE
)));
676 sock_graft(sk2
, newsock
);
678 newsock
->state
= SS_CONNECTED
;
684 EXPORT_SYMBOL(inet_accept
);
688 * This does both peername and sockname.
690 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
691 int *uaddr_len
, int peer
)
693 struct sock
*sk
= sock
->sk
;
694 struct inet_sock
*inet
= inet_sk(sk
);
695 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, uaddr
);
697 sin
->sin_family
= AF_INET
;
699 if (!inet
->inet_dport
||
700 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
703 sin
->sin_port
= inet
->inet_dport
;
704 sin
->sin_addr
.s_addr
= inet
->inet_daddr
;
706 __be32 addr
= inet
->inet_rcv_saddr
;
708 addr
= inet
->inet_saddr
;
709 sin
->sin_port
= inet
->inet_sport
;
710 sin
->sin_addr
.s_addr
= addr
;
712 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
713 *uaddr_len
= sizeof(*sin
);
716 EXPORT_SYMBOL(inet_getname
);
718 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
721 struct sock
*sk
= sock
->sk
;
723 /* We may need to bind the socket. */
724 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
727 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
729 EXPORT_SYMBOL(inet_sendmsg
);
732 static ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
733 size_t size
, int flags
)
735 struct sock
*sk
= sock
->sk
;
737 /* We may need to bind the socket. */
738 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
741 if (sk
->sk_prot
->sendpage
)
742 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
743 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
747 int inet_shutdown(struct socket
*sock
, int how
)
749 struct sock
*sk
= sock
->sk
;
752 /* This should really check to make sure
753 * the socket is a TCP socket. (WHY AC...)
755 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
758 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
762 if (sock
->state
== SS_CONNECTING
) {
763 if ((1 << sk
->sk_state
) &
764 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
765 sock
->state
= SS_DISCONNECTING
;
767 sock
->state
= SS_CONNECTED
;
770 switch (sk
->sk_state
) {
773 /* Hack to wake up other listeners, who can poll for
774 POLLHUP, even on eg. unconnected UDP sockets -- RR */
776 sk
->sk_shutdown
|= how
;
777 if (sk
->sk_prot
->shutdown
)
778 sk
->sk_prot
->shutdown(sk
, how
);
781 /* Remaining two branches are temporary solution for missing
782 * close() in multithreaded environment. It is _not_ a good idea,
783 * but we have no choice until close() is repaired at VFS level.
786 if (!(how
& RCV_SHUTDOWN
))
790 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
791 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
795 /* Wake up anyone sleeping in poll. */
796 sk
->sk_state_change(sk
);
800 EXPORT_SYMBOL(inet_shutdown
);
803 * ioctl() calls you can issue on an INET socket. Most of these are
804 * device configuration and stuff and very rarely used. Some ioctls
805 * pass on to the socket itself.
807 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
808 * loads the devconfigure module does its configuring and unloads it.
809 * There's a good 20K of config code hanging around the kernel.
812 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
814 struct sock
*sk
= sock
->sk
;
816 struct net
*net
= sock_net(sk
);
820 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
823 err
= sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
828 err
= ip_rt_ioctl(net
, cmd
, (void __user
*)arg
);
833 err
= arp_ioctl(net
, cmd
, (void __user
*)arg
);
846 err
= devinet_ioctl(net
, cmd
, (void __user
*)arg
);
849 if (sk
->sk_prot
->ioctl
)
850 err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
);
857 EXPORT_SYMBOL(inet_ioctl
);
859 const struct proto_ops inet_stream_ops
= {
861 .owner
= THIS_MODULE
,
862 .release
= inet_release
,
864 .connect
= inet_stream_connect
,
865 .socketpair
= sock_no_socketpair
,
866 .accept
= inet_accept
,
867 .getname
= inet_getname
,
870 .listen
= inet_listen
,
871 .shutdown
= inet_shutdown
,
872 .setsockopt
= sock_common_setsockopt
,
873 .getsockopt
= sock_common_getsockopt
,
874 .sendmsg
= tcp_sendmsg
,
875 .recvmsg
= sock_common_recvmsg
,
876 .mmap
= sock_no_mmap
,
877 .sendpage
= tcp_sendpage
,
878 .splice_read
= tcp_splice_read
,
880 .compat_setsockopt
= compat_sock_common_setsockopt
,
881 .compat_getsockopt
= compat_sock_common_getsockopt
,
884 EXPORT_SYMBOL(inet_stream_ops
);
886 const struct proto_ops inet_dgram_ops
= {
888 .owner
= THIS_MODULE
,
889 .release
= inet_release
,
891 .connect
= inet_dgram_connect
,
892 .socketpair
= sock_no_socketpair
,
893 .accept
= sock_no_accept
,
894 .getname
= inet_getname
,
897 .listen
= sock_no_listen
,
898 .shutdown
= inet_shutdown
,
899 .setsockopt
= sock_common_setsockopt
,
900 .getsockopt
= sock_common_getsockopt
,
901 .sendmsg
= inet_sendmsg
,
902 .recvmsg
= sock_common_recvmsg
,
903 .mmap
= sock_no_mmap
,
904 .sendpage
= inet_sendpage
,
906 .compat_setsockopt
= compat_sock_common_setsockopt
,
907 .compat_getsockopt
= compat_sock_common_getsockopt
,
910 EXPORT_SYMBOL(inet_dgram_ops
);
913 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
916 static const struct proto_ops inet_sockraw_ops
= {
918 .owner
= THIS_MODULE
,
919 .release
= inet_release
,
921 .connect
= inet_dgram_connect
,
922 .socketpair
= sock_no_socketpair
,
923 .accept
= sock_no_accept
,
924 .getname
= inet_getname
,
925 .poll
= datagram_poll
,
927 .listen
= sock_no_listen
,
928 .shutdown
= inet_shutdown
,
929 .setsockopt
= sock_common_setsockopt
,
930 .getsockopt
= sock_common_getsockopt
,
931 .sendmsg
= inet_sendmsg
,
932 .recvmsg
= sock_common_recvmsg
,
933 .mmap
= sock_no_mmap
,
934 .sendpage
= inet_sendpage
,
936 .compat_setsockopt
= compat_sock_common_setsockopt
,
937 .compat_getsockopt
= compat_sock_common_getsockopt
,
941 static const struct net_proto_family inet_family_ops
= {
943 .create
= inet_create
,
944 .owner
= THIS_MODULE
,
947 /* Upon startup we insert all the elements in inetsw_array[] into
948 * the linked list inetsw.
950 static struct inet_protosw inetsw_array
[] =
954 .protocol
= IPPROTO_TCP
,
956 .ops
= &inet_stream_ops
,
958 .flags
= INET_PROTOSW_PERMANENT
|
964 .protocol
= IPPROTO_UDP
,
966 .ops
= &inet_dgram_ops
,
967 .no_check
= UDP_CSUM_DEFAULT
,
968 .flags
= INET_PROTOSW_PERMANENT
,
974 .protocol
= IPPROTO_IP
, /* wild card */
976 .ops
= &inet_sockraw_ops
,
977 .no_check
= UDP_CSUM_DEFAULT
,
978 .flags
= INET_PROTOSW_REUSE
,
982 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
984 void inet_register_protosw(struct inet_protosw
*p
)
986 struct list_head
*lh
;
987 struct inet_protosw
*answer
;
988 int protocol
= p
->protocol
;
989 struct list_head
*last_perm
;
991 spin_lock_bh(&inetsw_lock
);
993 if (p
->type
>= SOCK_MAX
)
996 /* If we are trying to override a permanent protocol, bail. */
998 last_perm
= &inetsw
[p
->type
];
999 list_for_each(lh
, &inetsw
[p
->type
]) {
1000 answer
= list_entry(lh
, struct inet_protosw
, list
);
1002 /* Check only the non-wild match. */
1003 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
1004 if (protocol
== answer
->protocol
)
1014 /* Add the new entry after the last permanent entry if any, so that
1015 * the new entry does not override a permanent entry when matched with
1016 * a wild-card protocol. But it is allowed to override any existing
1017 * non-permanent entry. This means that when we remove this entry, the
1018 * system automatically returns to the old behavior.
1020 list_add_rcu(&p
->list
, last_perm
);
1022 spin_unlock_bh(&inetsw_lock
);
1027 printk(KERN_ERR
"Attempt to override permanent protocol %d.\n",
1033 "Ignoring attempt to register invalid socket type %d.\n",
1037 EXPORT_SYMBOL(inet_register_protosw
);
1039 void inet_unregister_protosw(struct inet_protosw
*p
)
1041 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
1043 "Attempt to unregister permanent protocol %d.\n",
1046 spin_lock_bh(&inetsw_lock
);
1047 list_del_rcu(&p
->list
);
1048 spin_unlock_bh(&inetsw_lock
);
1053 EXPORT_SYMBOL(inet_unregister_protosw
);
1056 * Shall we try to damage output packets if routing dev changes?
1059 int sysctl_ip_dynaddr __read_mostly
;
1061 static int inet_sk_reselect_saddr(struct sock
*sk
)
1063 struct inet_sock
*inet
= inet_sk(sk
);
1066 __be32 old_saddr
= inet
->inet_saddr
;
1068 __be32 daddr
= inet
->inet_daddr
;
1070 if (inet
->opt
&& inet
->opt
->srr
)
1071 daddr
= inet
->opt
->faddr
;
1073 /* Query new route. */
1074 err
= ip_route_connect(&rt
, daddr
, 0,
1076 sk
->sk_bound_dev_if
,
1078 inet
->inet_sport
, inet
->inet_dport
, sk
, 0);
1082 sk_setup_caps(sk
, &rt
->u
.dst
);
1084 new_saddr
= rt
->rt_src
;
1086 if (new_saddr
== old_saddr
)
1089 if (sysctl_ip_dynaddr
> 1) {
1090 printk(KERN_INFO
"%s(): shifting inet->saddr from %pI4 to %pI4\n",
1091 __func__
, &old_saddr
, &new_saddr
);
1094 inet
->inet_saddr
= inet
->inet_rcv_saddr
= new_saddr
;
1097 * XXX The only one ugly spot where we need to
1098 * XXX really change the sockets identity after
1099 * XXX it has entered the hashes. -DaveM
1101 * Besides that, it does not check for connection
1102 * uniqueness. Wait for troubles.
1104 __sk_prot_rehash(sk
);
1108 int inet_sk_rebuild_header(struct sock
*sk
)
1110 struct inet_sock
*inet
= inet_sk(sk
);
1111 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1115 /* Route is OK, nothing to do. */
1120 daddr
= inet
->inet_daddr
;
1121 if (inet
->opt
&& inet
->opt
->srr
)
1122 daddr
= inet
->opt
->faddr
;
1125 .oif
= sk
->sk_bound_dev_if
,
1126 .mark
= sk
->sk_mark
,
1130 .saddr
= inet
->inet_saddr
,
1131 .tos
= RT_CONN_FLAGS(sk
),
1134 .proto
= sk
->sk_protocol
,
1135 .flags
= inet_sk_flowi_flags(sk
),
1138 .sport
= inet
->inet_sport
,
1139 .dport
= inet
->inet_dport
,
1144 security_sk_classify_flow(sk
, &fl
);
1145 err
= ip_route_output_flow(sock_net(sk
), &rt
, &fl
, sk
, 0);
1148 sk_setup_caps(sk
, &rt
->u
.dst
);
1150 /* Routing failed... */
1151 sk
->sk_route_caps
= 0;
1153 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1154 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1156 if (!sysctl_ip_dynaddr
||
1157 sk
->sk_state
!= TCP_SYN_SENT
||
1158 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1159 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1160 sk
->sk_err_soft
= -err
;
1165 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1167 static int inet_gso_send_check(struct sk_buff
*skb
)
1170 const struct net_protocol
*ops
;
1175 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1180 if (ihl
< sizeof(*iph
))
1183 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1186 __skb_pull(skb
, ihl
);
1187 skb_reset_transport_header(skb
);
1189 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1190 err
= -EPROTONOSUPPORT
;
1193 ops
= rcu_dereference(inet_protos
[proto
]);
1194 if (likely(ops
&& ops
->gso_send_check
))
1195 err
= ops
->gso_send_check(skb
);
1202 static struct sk_buff
*inet_gso_segment(struct sk_buff
*skb
, int features
)
1204 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1206 const struct net_protocol
*ops
;
1210 unsigned int offset
= 0;
1212 if (!(features
& NETIF_F_V4_CSUM
))
1213 features
&= ~NETIF_F_SG
;
1215 if (unlikely(skb_shinfo(skb
)->gso_type
&
1223 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1228 if (ihl
< sizeof(*iph
))
1231 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1234 __skb_pull(skb
, ihl
);
1235 skb_reset_transport_header(skb
);
1237 id
= ntohs(iph
->id
);
1238 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1239 segs
= ERR_PTR(-EPROTONOSUPPORT
);
1242 ops
= rcu_dereference(inet_protos
[proto
]);
1243 if (likely(ops
&& ops
->gso_segment
))
1244 segs
= ops
->gso_segment(skb
, features
);
1247 if (!segs
|| IS_ERR(segs
))
1253 if (proto
== IPPROTO_UDP
) {
1254 iph
->id
= htons(id
);
1255 iph
->frag_off
= htons(offset
>> 3);
1256 if (skb
->next
!= NULL
)
1257 iph
->frag_off
|= htons(IP_MF
);
1258 offset
+= (skb
->len
- skb
->mac_len
- iph
->ihl
* 4);
1260 iph
->id
= htons(id
++);
1261 iph
->tot_len
= htons(skb
->len
- skb
->mac_len
);
1263 iph
->check
= ip_fast_csum(skb_network_header(skb
), iph
->ihl
);
1264 } while ((skb
= skb
->next
));
1270 static struct sk_buff
**inet_gro_receive(struct sk_buff
**head
,
1271 struct sk_buff
*skb
)
1273 const struct net_protocol
*ops
;
1274 struct sk_buff
**pp
= NULL
;
1283 off
= skb_gro_offset(skb
);
1284 hlen
= off
+ sizeof(*iph
);
1285 iph
= skb_gro_header_fast(skb
, off
);
1286 if (skb_gro_header_hard(skb
, hlen
)) {
1287 iph
= skb_gro_header_slow(skb
, hlen
, off
);
1292 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1295 ops
= rcu_dereference(inet_protos
[proto
]);
1296 if (!ops
|| !ops
->gro_receive
)
1299 if (*(u8
*)iph
!= 0x45)
1302 if (unlikely(ip_fast_csum((u8
*)iph
, iph
->ihl
)))
1305 id
= ntohl(*(u32
*)&iph
->id
);
1306 flush
= (u16
)((ntohl(*(u32
*)iph
) ^ skb_gro_len(skb
)) | (id
^ IP_DF
));
1309 for (p
= *head
; p
; p
= p
->next
) {
1312 if (!NAPI_GRO_CB(p
)->same_flow
)
1317 if ((iph
->protocol
^ iph2
->protocol
) |
1318 (iph
->tos
^ iph2
->tos
) |
1319 (iph
->saddr
^ iph2
->saddr
) |
1320 (iph
->daddr
^ iph2
->daddr
)) {
1321 NAPI_GRO_CB(p
)->same_flow
= 0;
1325 /* All fields must match except length and checksum. */
1326 NAPI_GRO_CB(p
)->flush
|=
1327 (iph
->ttl
^ iph2
->ttl
) |
1328 ((u16
)(ntohs(iph2
->id
) + NAPI_GRO_CB(p
)->count
) ^ id
);
1330 NAPI_GRO_CB(p
)->flush
|= flush
;
1333 NAPI_GRO_CB(skb
)->flush
|= flush
;
1334 skb_gro_pull(skb
, sizeof(*iph
));
1335 skb_set_transport_header(skb
, skb_gro_offset(skb
));
1337 pp
= ops
->gro_receive(head
, skb
);
1343 NAPI_GRO_CB(skb
)->flush
|= flush
;
1348 static int inet_gro_complete(struct sk_buff
*skb
)
1350 const struct net_protocol
*ops
;
1351 struct iphdr
*iph
= ip_hdr(skb
);
1352 int proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1354 __be16 newlen
= htons(skb
->len
- skb_network_offset(skb
));
1356 csum_replace2(&iph
->check
, iph
->tot_len
, newlen
);
1357 iph
->tot_len
= newlen
;
1360 ops
= rcu_dereference(inet_protos
[proto
]);
1361 if (WARN_ON(!ops
|| !ops
->gro_complete
))
1364 err
= ops
->gro_complete(skb
);
1372 int inet_ctl_sock_create(struct sock
**sk
, unsigned short family
,
1373 unsigned short type
, unsigned char protocol
,
1376 struct socket
*sock
;
1377 int rc
= sock_create_kern(family
, type
, protocol
, &sock
);
1381 (*sk
)->sk_allocation
= GFP_ATOMIC
;
1383 * Unhash it so that IP input processing does not even see it,
1384 * we do not wish this socket to see incoming packets.
1386 (*sk
)->sk_prot
->unhash(*sk
);
1388 sk_change_net(*sk
, net
);
1392 EXPORT_SYMBOL_GPL(inet_ctl_sock_create
);
1394 unsigned long snmp_fold_field(void __percpu
*mib
[], int offt
)
1396 unsigned long res
= 0;
1399 for_each_possible_cpu(i
) {
1400 res
+= *(((unsigned long *) per_cpu_ptr(mib
[0], i
)) + offt
);
1401 res
+= *(((unsigned long *) per_cpu_ptr(mib
[1], i
)) + offt
);
1405 EXPORT_SYMBOL_GPL(snmp_fold_field
);
1407 int snmp_mib_init(void __percpu
*ptr
[2], size_t mibsize
)
1409 BUG_ON(ptr
== NULL
);
1410 ptr
[0] = __alloc_percpu(mibsize
, __alignof__(unsigned long long));
1413 ptr
[1] = __alloc_percpu(mibsize
, __alignof__(unsigned long long));
1418 free_percpu(ptr
[0]);
1423 EXPORT_SYMBOL_GPL(snmp_mib_init
);
1425 void snmp_mib_free(void __percpu
*ptr
[2])
1427 BUG_ON(ptr
== NULL
);
1428 free_percpu(ptr
[0]);
1429 free_percpu(ptr
[1]);
1430 ptr
[0] = ptr
[1] = NULL
;
1432 EXPORT_SYMBOL_GPL(snmp_mib_free
);
1434 #ifdef CONFIG_IP_MULTICAST
1435 static const struct net_protocol igmp_protocol
= {
1436 .handler
= igmp_rcv
,
1441 static const struct net_protocol tcp_protocol
= {
1442 .handler
= tcp_v4_rcv
,
1443 .err_handler
= tcp_v4_err
,
1444 .gso_send_check
= tcp_v4_gso_send_check
,
1445 .gso_segment
= tcp_tso_segment
,
1446 .gro_receive
= tcp4_gro_receive
,
1447 .gro_complete
= tcp4_gro_complete
,
1452 static const struct net_protocol udp_protocol
= {
1454 .err_handler
= udp_err
,
1455 .gso_send_check
= udp4_ufo_send_check
,
1456 .gso_segment
= udp4_ufo_fragment
,
1461 static const struct net_protocol icmp_protocol
= {
1462 .handler
= icmp_rcv
,
1467 static __net_init
int ipv4_mib_init_net(struct net
*net
)
1469 if (snmp_mib_init((void __percpu
**)net
->mib
.tcp_statistics
,
1470 sizeof(struct tcp_mib
)) < 0)
1472 if (snmp_mib_init((void __percpu
**)net
->mib
.ip_statistics
,
1473 sizeof(struct ipstats_mib
)) < 0)
1475 if (snmp_mib_init((void __percpu
**)net
->mib
.net_statistics
,
1476 sizeof(struct linux_mib
)) < 0)
1478 if (snmp_mib_init((void __percpu
**)net
->mib
.udp_statistics
,
1479 sizeof(struct udp_mib
)) < 0)
1481 if (snmp_mib_init((void __percpu
**)net
->mib
.udplite_statistics
,
1482 sizeof(struct udp_mib
)) < 0)
1483 goto err_udplite_mib
;
1484 if (snmp_mib_init((void __percpu
**)net
->mib
.icmp_statistics
,
1485 sizeof(struct icmp_mib
)) < 0)
1487 if (snmp_mib_init((void __percpu
**)net
->mib
.icmpmsg_statistics
,
1488 sizeof(struct icmpmsg_mib
)) < 0)
1489 goto err_icmpmsg_mib
;
1495 snmp_mib_free((void __percpu
**)net
->mib
.icmp_statistics
);
1497 snmp_mib_free((void __percpu
**)net
->mib
.udplite_statistics
);
1499 snmp_mib_free((void __percpu
**)net
->mib
.udp_statistics
);
1501 snmp_mib_free((void __percpu
**)net
->mib
.net_statistics
);
1503 snmp_mib_free((void __percpu
**)net
->mib
.ip_statistics
);
1505 snmp_mib_free((void __percpu
**)net
->mib
.tcp_statistics
);
1510 static __net_exit
void ipv4_mib_exit_net(struct net
*net
)
1512 snmp_mib_free((void __percpu
**)net
->mib
.icmpmsg_statistics
);
1513 snmp_mib_free((void __percpu
**)net
->mib
.icmp_statistics
);
1514 snmp_mib_free((void __percpu
**)net
->mib
.udplite_statistics
);
1515 snmp_mib_free((void __percpu
**)net
->mib
.udp_statistics
);
1516 snmp_mib_free((void __percpu
**)net
->mib
.net_statistics
);
1517 snmp_mib_free((void __percpu
**)net
->mib
.ip_statistics
);
1518 snmp_mib_free((void __percpu
**)net
->mib
.tcp_statistics
);
1521 static __net_initdata
struct pernet_operations ipv4_mib_ops
= {
1522 .init
= ipv4_mib_init_net
,
1523 .exit
= ipv4_mib_exit_net
,
1526 static int __init
init_ipv4_mibs(void)
1528 return register_pernet_subsys(&ipv4_mib_ops
);
1531 static int ipv4_proc_init(void);
1534 * IP protocol layer initialiser
1537 static struct packet_type ip_packet_type __read_mostly
= {
1538 .type
= cpu_to_be16(ETH_P_IP
),
1540 .gso_send_check
= inet_gso_send_check
,
1541 .gso_segment
= inet_gso_segment
,
1542 .gro_receive
= inet_gro_receive
,
1543 .gro_complete
= inet_gro_complete
,
1546 static int __init
inet_init(void)
1548 struct sk_buff
*dummy_skb
;
1549 struct inet_protosw
*q
;
1550 struct list_head
*r
;
1553 BUILD_BUG_ON(sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
));
1555 rc
= proto_register(&tcp_prot
, 1);
1559 rc
= proto_register(&udp_prot
, 1);
1561 goto out_unregister_tcp_proto
;
1563 rc
= proto_register(&raw_prot
, 1);
1565 goto out_unregister_udp_proto
;
1568 * Tell SOCKET that we are alive...
1571 (void)sock_register(&inet_family_ops
);
1573 #ifdef CONFIG_SYSCTL
1574 ip_static_sysctl_init();
1578 * Add all the base protocols.
1581 if (inet_add_protocol(&icmp_protocol
, IPPROTO_ICMP
) < 0)
1582 printk(KERN_CRIT
"inet_init: Cannot add ICMP protocol\n");
1583 if (inet_add_protocol(&udp_protocol
, IPPROTO_UDP
) < 0)
1584 printk(KERN_CRIT
"inet_init: Cannot add UDP protocol\n");
1585 if (inet_add_protocol(&tcp_protocol
, IPPROTO_TCP
) < 0)
1586 printk(KERN_CRIT
"inet_init: Cannot add TCP protocol\n");
1587 #ifdef CONFIG_IP_MULTICAST
1588 if (inet_add_protocol(&igmp_protocol
, IPPROTO_IGMP
) < 0)
1589 printk(KERN_CRIT
"inet_init: Cannot add IGMP protocol\n");
1592 /* Register the socket-side information for inet_create. */
1593 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1596 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1597 inet_register_protosw(q
);
1600 * Set the ARP module up
1606 * Set the IP module up
1613 /* Setup TCP slab cache for open requests. */
1616 /* Setup UDP memory threshold */
1619 /* Add UDP-Lite (RFC 3828) */
1620 udplite4_register();
1623 * Set the ICMP layer up
1626 if (icmp_init() < 0)
1627 panic("Failed to create the ICMP control socket.\n");
1630 * Initialise the multicast router
1632 #if defined(CONFIG_IP_MROUTE)
1634 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mroute\n");
1637 * Initialise per-cpu ipv4 mibs
1640 if (init_ipv4_mibs())
1641 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mibs\n");
1647 dev_add_pack(&ip_packet_type
);
1652 out_unregister_udp_proto
:
1653 proto_unregister(&udp_prot
);
1654 out_unregister_tcp_proto
:
1655 proto_unregister(&tcp_prot
);
1659 fs_initcall(inet_init
);
1661 /* ------------------------------------------------------------------------ */
1663 #ifdef CONFIG_PROC_FS
1664 static int __init
ipv4_proc_init(void)
1668 if (raw_proc_init())
1670 if (tcp4_proc_init())
1672 if (udp4_proc_init())
1674 if (ip_misc_proc_init())
1689 #else /* CONFIG_PROC_FS */
1690 static int __init
ipv4_proc_init(void)
1694 #endif /* CONFIG_PROC_FS */
1696 MODULE_ALIAS_NETPROTO(PF_INET
);