2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * PF_INET protocol family socket handler.
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
13 * Changes (see also sock.c)
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
68 #include <linux/err.h>
69 #include <linux/errno.h>
70 #include <linux/types.h>
71 #include <linux/socket.h>
73 #include <linux/kernel.h>
74 #include <linux/module.h>
75 #include <linux/sched.h>
76 #include <linux/timer.h>
77 #include <linux/string.h>
78 #include <linux/sockios.h>
79 #include <linux/net.h>
80 #include <linux/capability.h>
81 #include <linux/fcntl.h>
83 #include <linux/interrupt.h>
84 #include <linux/stat.h>
85 #include <linux/init.h>
86 #include <linux/poll.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
90 #include <asm/uaccess.h>
91 #include <asm/system.h>
93 #include <linux/inet.h>
94 #include <linux/igmp.h>
95 #include <linux/inetdevice.h>
96 #include <linux/netdevice.h>
97 #include <net/checksum.h>
99 #include <net/protocol.h>
101 #include <net/route.h>
102 #include <net/ip_fib.h>
103 #include <net/inet_connection_sock.h>
106 #include <net/udplite.h>
107 #include <linux/skbuff.h>
108 #include <net/sock.h>
110 #include <net/icmp.h>
111 #include <net/ipip.h>
112 #include <net/inet_common.h>
113 #include <net/xfrm.h>
114 #include <net/net_namespace.h>
115 #ifdef CONFIG_IP_MROUTE
116 #include <linux/mroute.h>
120 /* The inetsw table contains everything that inet_create needs to
121 * build a new socket.
123 static struct list_head inetsw
[SOCK_MAX
];
124 static DEFINE_SPINLOCK(inetsw_lock
);
126 struct ipv4_config ipv4_config
;
127 EXPORT_SYMBOL(ipv4_config
);
129 /* New destruction routine */
131 void inet_sock_destruct(struct sock
*sk
)
133 struct inet_sock
*inet
= inet_sk(sk
);
135 __skb_queue_purge(&sk
->sk_receive_queue
);
136 __skb_queue_purge(&sk
->sk_error_queue
);
140 if (sk
->sk_type
== SOCK_STREAM
&& sk
->sk_state
!= TCP_CLOSE
) {
141 pr_err("Attempt to release TCP socket in state %d %p\n",
145 if (!sock_flag(sk
, SOCK_DEAD
)) {
146 pr_err("Attempt to release alive inet socket %p\n", sk
);
150 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
151 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
152 WARN_ON(sk
->sk_wmem_queued
);
153 WARN_ON(sk
->sk_forward_alloc
);
156 dst_release(sk
->sk_dst_cache
);
157 sk_refcnt_debug_dec(sk
);
159 EXPORT_SYMBOL(inet_sock_destruct
);
162 * The routines beyond this point handle the behaviour of an AF_INET
163 * socket object. Mostly it punts to the subprotocols of IP to do
168 * Automatically bind an unbound socket.
171 static int inet_autobind(struct sock
*sk
)
173 struct inet_sock
*inet
;
174 /* We may need to bind the socket. */
177 if (!inet
->inet_num
) {
178 if (sk
->sk_prot
->get_port(sk
, 0)) {
182 inet
->inet_sport
= htons(inet
->inet_num
);
189 * Move a socket into listening state.
191 int inet_listen(struct socket
*sock
, int backlog
)
193 struct sock
*sk
= sock
->sk
;
194 unsigned char old_state
;
200 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_STREAM
)
203 old_state
= sk
->sk_state
;
204 if (!((1 << old_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
207 /* Really, if the socket is already in listen state
208 * we can only allow the backlog to be adjusted.
210 if (old_state
!= TCP_LISTEN
) {
211 err
= inet_csk_listen_start(sk
, backlog
);
215 sk
->sk_max_ack_backlog
= backlog
;
222 EXPORT_SYMBOL(inet_listen
);
224 u32 inet_ehash_secret __read_mostly
;
225 EXPORT_SYMBOL(inet_ehash_secret
);
228 * inet_ehash_secret must be set exactly once
229 * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
231 void build_ehash_secret(void)
235 get_random_bytes(&rnd
, sizeof(rnd
));
237 spin_lock_bh(&inetsw_lock
);
238 if (!inet_ehash_secret
)
239 inet_ehash_secret
= rnd
;
240 spin_unlock_bh(&inetsw_lock
);
242 EXPORT_SYMBOL(build_ehash_secret
);
244 static inline int inet_netns_ok(struct net
*net
, int protocol
)
247 const struct net_protocol
*ipprot
;
249 if (net_eq(net
, &init_net
))
252 hash
= protocol
& (MAX_INET_PROTOS
- 1);
253 ipprot
= rcu_dereference(inet_protos
[hash
]);
258 return ipprot
->netns_ok
;
262 * Create an inet socket.
265 static int inet_create(struct net
*net
, struct socket
*sock
, int protocol
)
268 struct inet_protosw
*answer
;
269 struct inet_sock
*inet
;
270 struct proto
*answer_prot
;
271 unsigned char answer_flags
;
272 char answer_no_check
;
273 int try_loading_module
= 0;
276 if (unlikely(!inet_ehash_secret
))
277 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
278 build_ehash_secret();
280 sock
->state
= SS_UNCONNECTED
;
282 /* Look for the requested type/protocol pair. */
284 err
= -ESOCKTNOSUPPORT
;
286 list_for_each_entry_rcu(answer
, &inetsw
[sock
->type
], list
) {
289 /* Check the non-wild match. */
290 if (protocol
== answer
->protocol
) {
291 if (protocol
!= IPPROTO_IP
)
294 /* Check for the two wild cases. */
295 if (IPPROTO_IP
== protocol
) {
296 protocol
= answer
->protocol
;
299 if (IPPROTO_IP
== answer
->protocol
)
302 err
= -EPROTONOSUPPORT
;
306 if (try_loading_module
< 2) {
309 * Be more specific, e.g. net-pf-2-proto-132-type-1
310 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
312 if (++try_loading_module
== 1)
313 request_module("net-pf-%d-proto-%d-type-%d",
314 PF_INET
, protocol
, sock
->type
);
316 * Fall back to generic, e.g. net-pf-2-proto-132
317 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
320 request_module("net-pf-%d-proto-%d",
322 goto lookup_protocol
;
328 if (answer
->capability
> 0 && !capable(answer
->capability
))
332 if (!inet_netns_ok(net
, protocol
))
335 sock
->ops
= answer
->ops
;
336 answer_prot
= answer
->prot
;
337 answer_no_check
= answer
->no_check
;
338 answer_flags
= answer
->flags
;
341 WARN_ON(answer_prot
->slab
== NULL
);
344 sk
= sk_alloc(net
, PF_INET
, GFP_KERNEL
, answer_prot
);
349 sk
->sk_no_check
= answer_no_check
;
350 if (INET_PROTOSW_REUSE
& answer_flags
)
354 inet
->is_icsk
= (INET_PROTOSW_ICSK
& answer_flags
) != 0;
356 if (SOCK_RAW
== sock
->type
) {
357 inet
->inet_num
= protocol
;
358 if (IPPROTO_RAW
== protocol
)
362 if (ipv4_config
.no_pmtu_disc
)
363 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
365 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
369 sock_init_data(sock
, sk
);
371 sk
->sk_destruct
= inet_sock_destruct
;
372 sk
->sk_protocol
= protocol
;
373 sk
->sk_backlog_rcv
= sk
->sk_prot
->backlog_rcv
;
380 inet
->mc_list
= NULL
;
382 sk_refcnt_debug_inc(sk
);
384 if (inet
->inet_num
) {
385 /* It assumes that any protocol which allows
386 * the user to assign a number at socket
387 * creation time automatically
390 inet
->inet_sport
= htons(inet
->inet_num
);
391 /* Add to protocol hash chains. */
392 sk
->sk_prot
->hash(sk
);
395 if (sk
->sk_prot
->init
) {
396 err
= sk
->sk_prot
->init(sk
);
398 sk_common_release(sk
);
409 * The peer socket should always be NULL (or else). When we call this
410 * function we are destroying the object and from then on nobody
411 * should refer to it.
413 int inet_release(struct socket
*sock
)
415 struct sock
*sk
= sock
->sk
;
420 /* Applications forget to leave groups before exiting */
421 ip_mc_drop_socket(sk
);
423 /* If linger is set, we don't return until the close
424 * is complete. Otherwise we return immediately. The
425 * actually closing is done the same either way.
427 * If the close is due to the process exiting, we never
431 if (sock_flag(sk
, SOCK_LINGER
) &&
432 !(current
->flags
& PF_EXITING
))
433 timeout
= sk
->sk_lingertime
;
435 sk
->sk_prot
->close(sk
, timeout
);
439 EXPORT_SYMBOL(inet_release
);
441 /* It is off by default, see below. */
442 int sysctl_ip_nonlocal_bind __read_mostly
;
443 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);
445 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
447 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
448 struct sock
*sk
= sock
->sk
;
449 struct inet_sock
*inet
= inet_sk(sk
);
454 /* If the socket has its own bind function then use it. (RAW) */
455 if (sk
->sk_prot
->bind
) {
456 err
= sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
460 if (addr_len
< sizeof(struct sockaddr_in
))
463 chk_addr_ret
= inet_addr_type(sock_net(sk
), addr
->sin_addr
.s_addr
);
465 /* Not specified by any standard per-se, however it breaks too
466 * many applications when removed. It is unfortunate since
467 * allowing applications to make a non-local bind solves
468 * several problems with systems using dynamic addressing.
469 * (ie. your servers still start up even if your ISDN link
470 * is temporarily down)
472 err
= -EADDRNOTAVAIL
;
473 if (!sysctl_ip_nonlocal_bind
&&
474 !(inet
->freebind
|| inet
->transparent
) &&
475 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
) &&
476 chk_addr_ret
!= RTN_LOCAL
&&
477 chk_addr_ret
!= RTN_MULTICAST
&&
478 chk_addr_ret
!= RTN_BROADCAST
)
481 snum
= ntohs(addr
->sin_port
);
483 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
486 /* We keep a pair of addresses. rcv_saddr is the one
487 * used by hash lookups, and saddr is used for transmit.
489 * In the BSD API these are the same except where it
490 * would be illegal to use them (multicast/broadcast) in
491 * which case the sending device address is used.
495 /* Check these errors (active socket, double bind). */
497 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->inet_num
)
498 goto out_release_sock
;
500 inet
->inet_rcv_saddr
= inet
->inet_saddr
= addr
->sin_addr
.s_addr
;
501 if (chk_addr_ret
== RTN_MULTICAST
|| chk_addr_ret
== RTN_BROADCAST
)
502 inet
->inet_saddr
= 0; /* Use device */
504 /* Make sure we are allowed to bind here. */
505 if (sk
->sk_prot
->get_port(sk
, snum
)) {
506 inet
->inet_saddr
= inet
->inet_rcv_saddr
= 0;
508 goto out_release_sock
;
511 if (inet
->inet_rcv_saddr
)
512 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
514 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
515 inet
->inet_sport
= htons(inet
->inet_num
);
516 inet
->inet_daddr
= 0;
517 inet
->inet_dport
= 0;
525 EXPORT_SYMBOL(inet_bind
);
527 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
* uaddr
,
528 int addr_len
, int flags
)
530 struct sock
*sk
= sock
->sk
;
532 if (uaddr
->sa_family
== AF_UNSPEC
)
533 return sk
->sk_prot
->disconnect(sk
, flags
);
535 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
537 return sk
->sk_prot
->connect(sk
, (struct sockaddr
*)uaddr
, addr_len
);
539 EXPORT_SYMBOL(inet_dgram_connect
);
541 static long inet_wait_for_connect(struct sock
*sk
, long timeo
)
545 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
547 /* Basic assumption: if someone sets sk->sk_err, he _must_
548 * change state of the socket from TCP_SYN_*.
549 * Connect() does not allow to get error notifications
550 * without closing the socket.
552 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
554 timeo
= schedule_timeout(timeo
);
556 if (signal_pending(current
) || !timeo
)
558 prepare_to_wait(sk
->sk_sleep
, &wait
, TASK_INTERRUPTIBLE
);
560 finish_wait(sk
->sk_sleep
, &wait
);
565 * Connect to a remote host. There is regrettably still a little
566 * TCP 'magic' in here.
568 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
569 int addr_len
, int flags
)
571 struct sock
*sk
= sock
->sk
;
577 if (uaddr
->sa_family
== AF_UNSPEC
) {
578 err
= sk
->sk_prot
->disconnect(sk
, flags
);
579 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
583 switch (sock
->state
) {
592 /* Fall out of switch with err, set for this state */
596 if (sk
->sk_state
!= TCP_CLOSE
)
599 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
603 sock
->state
= SS_CONNECTING
;
605 /* Just entered SS_CONNECTING state; the only
606 * difference is that return value in non-blocking
607 * case is EINPROGRESS, rather than EALREADY.
613 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
615 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
616 /* Error code is set above */
617 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
))
620 err
= sock_intr_errno(timeo
);
621 if (signal_pending(current
))
625 /* Connection was closed by RST, timeout, ICMP error
626 * or another process disconnected us.
628 if (sk
->sk_state
== TCP_CLOSE
)
631 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
632 * and error was received after socket entered established state.
633 * Hence, it is handled normally after connect() return successfully.
636 sock
->state
= SS_CONNECTED
;
643 err
= sock_error(sk
) ? : -ECONNABORTED
;
644 sock
->state
= SS_UNCONNECTED
;
645 if (sk
->sk_prot
->disconnect(sk
, flags
))
646 sock
->state
= SS_DISCONNECTING
;
649 EXPORT_SYMBOL(inet_stream_connect
);
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 WARN_ON(!((1 << sk2
->sk_state
) &
667 (TCPF_ESTABLISHED
| TCPF_CLOSE_WAIT
| TCPF_CLOSE
)));
669 sock_graft(sk2
, newsock
);
671 newsock
->state
= SS_CONNECTED
;
677 EXPORT_SYMBOL(inet_accept
);
681 * This does both peername and sockname.
683 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
684 int *uaddr_len
, int peer
)
686 struct sock
*sk
= sock
->sk
;
687 struct inet_sock
*inet
= inet_sk(sk
);
688 struct sockaddr_in
*sin
= (struct sockaddr_in
*)uaddr
;
690 sin
->sin_family
= AF_INET
;
692 if (!inet
->inet_dport
||
693 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
696 sin
->sin_port
= inet
->inet_dport
;
697 sin
->sin_addr
.s_addr
= inet
->inet_daddr
;
699 __be32 addr
= inet
->inet_rcv_saddr
;
701 addr
= inet
->inet_saddr
;
702 sin
->sin_port
= inet
->inet_sport
;
703 sin
->sin_addr
.s_addr
= addr
;
705 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
706 *uaddr_len
= sizeof(*sin
);
709 EXPORT_SYMBOL(inet_getname
);
711 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
714 struct sock
*sk
= sock
->sk
;
716 /* We may need to bind the socket. */
717 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
720 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
722 EXPORT_SYMBOL(inet_sendmsg
);
725 static ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
726 size_t size
, int flags
)
728 struct sock
*sk
= sock
->sk
;
730 /* We may need to bind the socket. */
731 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
734 if (sk
->sk_prot
->sendpage
)
735 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
736 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
740 int inet_shutdown(struct socket
*sock
, int how
)
742 struct sock
*sk
= sock
->sk
;
745 /* This should really check to make sure
746 * the socket is a TCP socket. (WHY AC...)
748 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
751 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
755 if (sock
->state
== SS_CONNECTING
) {
756 if ((1 << sk
->sk_state
) &
757 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
758 sock
->state
= SS_DISCONNECTING
;
760 sock
->state
= SS_CONNECTED
;
763 switch (sk
->sk_state
) {
766 /* Hack to wake up other listeners, who can poll for
767 POLLHUP, even on eg. unconnected UDP sockets -- RR */
769 sk
->sk_shutdown
|= how
;
770 if (sk
->sk_prot
->shutdown
)
771 sk
->sk_prot
->shutdown(sk
, how
);
774 /* Remaining two branches are temporary solution for missing
775 * close() in multithreaded environment. It is _not_ a good idea,
776 * but we have no choice until close() is repaired at VFS level.
779 if (!(how
& RCV_SHUTDOWN
))
783 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
784 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
788 /* Wake up anyone sleeping in poll. */
789 sk
->sk_state_change(sk
);
793 EXPORT_SYMBOL(inet_shutdown
);
796 * ioctl() calls you can issue on an INET socket. Most of these are
797 * device configuration and stuff and very rarely used. Some ioctls
798 * pass on to the socket itself.
800 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
801 * loads the devconfigure module does its configuring and unloads it.
802 * There's a good 20K of config code hanging around the kernel.
805 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
807 struct sock
*sk
= sock
->sk
;
809 struct net
*net
= sock_net(sk
);
813 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
816 err
= sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
821 err
= ip_rt_ioctl(net
, cmd
, (void __user
*)arg
);
826 err
= arp_ioctl(net
, cmd
, (void __user
*)arg
);
839 err
= devinet_ioctl(net
, cmd
, (void __user
*)arg
);
842 if (sk
->sk_prot
->ioctl
)
843 err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
);
850 EXPORT_SYMBOL(inet_ioctl
);
852 const struct proto_ops inet_stream_ops
= {
854 .owner
= THIS_MODULE
,
855 .release
= inet_release
,
857 .connect
= inet_stream_connect
,
858 .socketpair
= sock_no_socketpair
,
859 .accept
= inet_accept
,
860 .getname
= inet_getname
,
863 .listen
= inet_listen
,
864 .shutdown
= inet_shutdown
,
865 .setsockopt
= sock_common_setsockopt
,
866 .getsockopt
= sock_common_getsockopt
,
867 .sendmsg
= tcp_sendmsg
,
868 .recvmsg
= sock_common_recvmsg
,
869 .mmap
= sock_no_mmap
,
870 .sendpage
= tcp_sendpage
,
871 .splice_read
= tcp_splice_read
,
873 .compat_setsockopt
= compat_sock_common_setsockopt
,
874 .compat_getsockopt
= compat_sock_common_getsockopt
,
877 EXPORT_SYMBOL(inet_stream_ops
);
879 const struct proto_ops inet_dgram_ops
= {
881 .owner
= THIS_MODULE
,
882 .release
= inet_release
,
884 .connect
= inet_dgram_connect
,
885 .socketpair
= sock_no_socketpair
,
886 .accept
= sock_no_accept
,
887 .getname
= inet_getname
,
890 .listen
= sock_no_listen
,
891 .shutdown
= inet_shutdown
,
892 .setsockopt
= sock_common_setsockopt
,
893 .getsockopt
= sock_common_getsockopt
,
894 .sendmsg
= inet_sendmsg
,
895 .recvmsg
= sock_common_recvmsg
,
896 .mmap
= sock_no_mmap
,
897 .sendpage
= inet_sendpage
,
899 .compat_setsockopt
= compat_sock_common_setsockopt
,
900 .compat_getsockopt
= compat_sock_common_getsockopt
,
903 EXPORT_SYMBOL(inet_dgram_ops
);
906 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
909 static const struct proto_ops inet_sockraw_ops
= {
911 .owner
= THIS_MODULE
,
912 .release
= inet_release
,
914 .connect
= inet_dgram_connect
,
915 .socketpair
= sock_no_socketpair
,
916 .accept
= sock_no_accept
,
917 .getname
= inet_getname
,
918 .poll
= datagram_poll
,
920 .listen
= sock_no_listen
,
921 .shutdown
= inet_shutdown
,
922 .setsockopt
= sock_common_setsockopt
,
923 .getsockopt
= sock_common_getsockopt
,
924 .sendmsg
= inet_sendmsg
,
925 .recvmsg
= sock_common_recvmsg
,
926 .mmap
= sock_no_mmap
,
927 .sendpage
= inet_sendpage
,
929 .compat_setsockopt
= compat_sock_common_setsockopt
,
930 .compat_getsockopt
= compat_sock_common_getsockopt
,
934 static const struct net_proto_family inet_family_ops
= {
936 .create
= inet_create
,
937 .owner
= THIS_MODULE
,
940 /* Upon startup we insert all the elements in inetsw_array[] into
941 * the linked list inetsw.
943 static struct inet_protosw inetsw_array
[] =
947 .protocol
= IPPROTO_TCP
,
949 .ops
= &inet_stream_ops
,
952 .flags
= INET_PROTOSW_PERMANENT
|
958 .protocol
= IPPROTO_UDP
,
960 .ops
= &inet_dgram_ops
,
962 .no_check
= UDP_CSUM_DEFAULT
,
963 .flags
= INET_PROTOSW_PERMANENT
,
969 .protocol
= IPPROTO_IP
, /* wild card */
971 .ops
= &inet_sockraw_ops
,
972 .capability
= CAP_NET_RAW
,
973 .no_check
= UDP_CSUM_DEFAULT
,
974 .flags
= INET_PROTOSW_REUSE
,
978 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
980 void inet_register_protosw(struct inet_protosw
*p
)
982 struct list_head
*lh
;
983 struct inet_protosw
*answer
;
984 int protocol
= p
->protocol
;
985 struct list_head
*last_perm
;
987 spin_lock_bh(&inetsw_lock
);
989 if (p
->type
>= SOCK_MAX
)
992 /* If we are trying to override a permanent protocol, bail. */
994 last_perm
= &inetsw
[p
->type
];
995 list_for_each(lh
, &inetsw
[p
->type
]) {
996 answer
= list_entry(lh
, struct inet_protosw
, list
);
998 /* Check only the non-wild match. */
999 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
1000 if (protocol
== answer
->protocol
)
1010 /* Add the new entry after the last permanent entry if any, so that
1011 * the new entry does not override a permanent entry when matched with
1012 * a wild-card protocol. But it is allowed to override any existing
1013 * non-permanent entry. This means that when we remove this entry, the
1014 * system automatically returns to the old behavior.
1016 list_add_rcu(&p
->list
, last_perm
);
1018 spin_unlock_bh(&inetsw_lock
);
1023 printk(KERN_ERR
"Attempt to override permanent protocol %d.\n",
1029 "Ignoring attempt to register invalid socket type %d.\n",
1033 EXPORT_SYMBOL(inet_register_protosw
);
1035 void inet_unregister_protosw(struct inet_protosw
*p
)
1037 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
1039 "Attempt to unregister permanent protocol %d.\n",
1042 spin_lock_bh(&inetsw_lock
);
1043 list_del_rcu(&p
->list
);
1044 spin_unlock_bh(&inetsw_lock
);
1049 EXPORT_SYMBOL(inet_unregister_protosw
);
1052 * Shall we try to damage output packets if routing dev changes?
1055 int sysctl_ip_dynaddr __read_mostly
;
1057 static int inet_sk_reselect_saddr(struct sock
*sk
)
1059 struct inet_sock
*inet
= inet_sk(sk
);
1062 __be32 old_saddr
= inet
->inet_saddr
;
1064 __be32 daddr
= inet
->inet_daddr
;
1066 if (inet
->opt
&& inet
->opt
->srr
)
1067 daddr
= inet
->opt
->faddr
;
1069 /* Query new route. */
1070 err
= ip_route_connect(&rt
, daddr
, 0,
1072 sk
->sk_bound_dev_if
,
1074 inet
->inet_sport
, inet
->inet_dport
, sk
, 0);
1078 sk_setup_caps(sk
, &rt
->u
.dst
);
1080 new_saddr
= rt
->rt_src
;
1082 if (new_saddr
== old_saddr
)
1085 if (sysctl_ip_dynaddr
> 1) {
1086 printk(KERN_INFO
"%s(): shifting inet->saddr from %pI4 to %pI4\n",
1087 __func__
, &old_saddr
, &new_saddr
);
1090 inet
->inet_saddr
= inet
->inet_rcv_saddr
= new_saddr
;
1093 * XXX The only one ugly spot where we need to
1094 * XXX really change the sockets identity after
1095 * XXX it has entered the hashes. -DaveM
1097 * Besides that, it does not check for connection
1098 * uniqueness. Wait for troubles.
1100 __sk_prot_rehash(sk
);
1104 int inet_sk_rebuild_header(struct sock
*sk
)
1106 struct inet_sock
*inet
= inet_sk(sk
);
1107 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1111 /* Route is OK, nothing to do. */
1116 daddr
= inet
->inet_daddr
;
1117 if (inet
->opt
&& inet
->opt
->srr
)
1118 daddr
= inet
->opt
->faddr
;
1121 .oif
= sk
->sk_bound_dev_if
,
1122 .mark
= sk
->sk_mark
,
1126 .saddr
= inet
->inet_saddr
,
1127 .tos
= RT_CONN_FLAGS(sk
),
1130 .proto
= sk
->sk_protocol
,
1131 .flags
= inet_sk_flowi_flags(sk
),
1134 .sport
= inet
->inet_sport
,
1135 .dport
= inet
->inet_dport
,
1140 security_sk_classify_flow(sk
, &fl
);
1141 err
= ip_route_output_flow(sock_net(sk
), &rt
, &fl
, sk
, 0);
1144 sk_setup_caps(sk
, &rt
->u
.dst
);
1146 /* Routing failed... */
1147 sk
->sk_route_caps
= 0;
1149 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1150 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1152 if (!sysctl_ip_dynaddr
||
1153 sk
->sk_state
!= TCP_SYN_SENT
||
1154 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1155 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1156 sk
->sk_err_soft
= -err
;
1161 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1163 static int inet_gso_send_check(struct sk_buff
*skb
)
1166 const struct net_protocol
*ops
;
1171 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1176 if (ihl
< sizeof(*iph
))
1179 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1182 __skb_pull(skb
, ihl
);
1183 skb_reset_transport_header(skb
);
1185 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1186 err
= -EPROTONOSUPPORT
;
1189 ops
= rcu_dereference(inet_protos
[proto
]);
1190 if (likely(ops
&& ops
->gso_send_check
))
1191 err
= ops
->gso_send_check(skb
);
1198 static struct sk_buff
*inet_gso_segment(struct sk_buff
*skb
, int features
)
1200 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1202 const struct net_protocol
*ops
;
1206 unsigned int offset
= 0;
1208 if (!(features
& NETIF_F_V4_CSUM
))
1209 features
&= ~NETIF_F_SG
;
1211 if (unlikely(skb_shinfo(skb
)->gso_type
&
1219 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1224 if (ihl
< sizeof(*iph
))
1227 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1230 __skb_pull(skb
, ihl
);
1231 skb_reset_transport_header(skb
);
1233 id
= ntohs(iph
->id
);
1234 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1235 segs
= ERR_PTR(-EPROTONOSUPPORT
);
1238 ops
= rcu_dereference(inet_protos
[proto
]);
1239 if (likely(ops
&& ops
->gso_segment
))
1240 segs
= ops
->gso_segment(skb
, features
);
1243 if (!segs
|| IS_ERR(segs
))
1249 if (proto
== IPPROTO_UDP
) {
1250 iph
->id
= htons(id
);
1251 iph
->frag_off
= htons(offset
>> 3);
1252 if (skb
->next
!= NULL
)
1253 iph
->frag_off
|= htons(IP_MF
);
1254 offset
+= (skb
->len
- skb
->mac_len
- iph
->ihl
* 4);
1256 iph
->id
= htons(id
++);
1257 iph
->tot_len
= htons(skb
->len
- skb
->mac_len
);
1259 iph
->check
= ip_fast_csum(skb_network_header(skb
), iph
->ihl
);
1260 } while ((skb
= skb
->next
));
1266 static struct sk_buff
**inet_gro_receive(struct sk_buff
**head
,
1267 struct sk_buff
*skb
)
1269 const struct net_protocol
*ops
;
1270 struct sk_buff
**pp
= NULL
;
1279 off
= skb_gro_offset(skb
);
1280 hlen
= off
+ sizeof(*iph
);
1281 iph
= skb_gro_header_fast(skb
, off
);
1282 if (skb_gro_header_hard(skb
, hlen
)) {
1283 iph
= skb_gro_header_slow(skb
, hlen
, off
);
1288 proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1291 ops
= rcu_dereference(inet_protos
[proto
]);
1292 if (!ops
|| !ops
->gro_receive
)
1295 if (*(u8
*)iph
!= 0x45)
1298 if (unlikely(ip_fast_csum((u8
*)iph
, iph
->ihl
)))
1301 id
= ntohl(*(u32
*)&iph
->id
);
1302 flush
= (u16
)((ntohl(*(u32
*)iph
) ^ skb_gro_len(skb
)) | (id
^ IP_DF
));
1305 for (p
= *head
; p
; p
= p
->next
) {
1308 if (!NAPI_GRO_CB(p
)->same_flow
)
1313 if ((iph
->protocol
^ iph2
->protocol
) |
1314 (iph
->tos
^ iph2
->tos
) |
1315 (iph
->saddr
^ iph2
->saddr
) |
1316 (iph
->daddr
^ iph2
->daddr
)) {
1317 NAPI_GRO_CB(p
)->same_flow
= 0;
1321 /* All fields must match except length and checksum. */
1322 NAPI_GRO_CB(p
)->flush
|=
1323 (iph
->ttl
^ iph2
->ttl
) |
1324 ((u16
)(ntohs(iph2
->id
) + NAPI_GRO_CB(p
)->count
) ^ id
);
1326 NAPI_GRO_CB(p
)->flush
|= flush
;
1329 NAPI_GRO_CB(skb
)->flush
|= flush
;
1330 skb_gro_pull(skb
, sizeof(*iph
));
1331 skb_set_transport_header(skb
, skb_gro_offset(skb
));
1333 pp
= ops
->gro_receive(head
, skb
);
1339 NAPI_GRO_CB(skb
)->flush
|= flush
;
1344 static int inet_gro_complete(struct sk_buff
*skb
)
1346 const struct net_protocol
*ops
;
1347 struct iphdr
*iph
= ip_hdr(skb
);
1348 int proto
= iph
->protocol
& (MAX_INET_PROTOS
- 1);
1350 __be16 newlen
= htons(skb
->len
- skb_network_offset(skb
));
1352 csum_replace2(&iph
->check
, iph
->tot_len
, newlen
);
1353 iph
->tot_len
= newlen
;
1356 ops
= rcu_dereference(inet_protos
[proto
]);
1357 if (WARN_ON(!ops
|| !ops
->gro_complete
))
1360 err
= ops
->gro_complete(skb
);
1368 int inet_ctl_sock_create(struct sock
**sk
, unsigned short family
,
1369 unsigned short type
, unsigned char protocol
,
1372 struct socket
*sock
;
1373 int rc
= sock_create_kern(family
, type
, protocol
, &sock
);
1377 (*sk
)->sk_allocation
= GFP_ATOMIC
;
1379 * Unhash it so that IP input processing does not even see it,
1380 * we do not wish this socket to see incoming packets.
1382 (*sk
)->sk_prot
->unhash(*sk
);
1384 sk_change_net(*sk
, net
);
1388 EXPORT_SYMBOL_GPL(inet_ctl_sock_create
);
1390 unsigned long snmp_fold_field(void *mib
[], int offt
)
1392 unsigned long res
= 0;
1395 for_each_possible_cpu(i
) {
1396 res
+= *(((unsigned long *) per_cpu_ptr(mib
[0], i
)) + offt
);
1397 res
+= *(((unsigned long *) per_cpu_ptr(mib
[1], i
)) + offt
);
1401 EXPORT_SYMBOL_GPL(snmp_fold_field
);
1403 int snmp_mib_init(void *ptr
[2], size_t mibsize
)
1405 BUG_ON(ptr
== NULL
);
1406 ptr
[0] = __alloc_percpu(mibsize
, __alignof__(unsigned long long));
1409 ptr
[1] = __alloc_percpu(mibsize
, __alignof__(unsigned long long));
1414 free_percpu(ptr
[0]);
1419 EXPORT_SYMBOL_GPL(snmp_mib_init
);
1421 void snmp_mib_free(void *ptr
[2])
1423 BUG_ON(ptr
== NULL
);
1424 free_percpu(ptr
[0]);
1425 free_percpu(ptr
[1]);
1426 ptr
[0] = ptr
[1] = NULL
;
1428 EXPORT_SYMBOL_GPL(snmp_mib_free
);
1430 #ifdef CONFIG_IP_MULTICAST
1431 static const struct net_protocol igmp_protocol
= {
1432 .handler
= igmp_rcv
,
1437 static const struct net_protocol tcp_protocol
= {
1438 .handler
= tcp_v4_rcv
,
1439 .err_handler
= tcp_v4_err
,
1440 .gso_send_check
= tcp_v4_gso_send_check
,
1441 .gso_segment
= tcp_tso_segment
,
1442 .gro_receive
= tcp4_gro_receive
,
1443 .gro_complete
= tcp4_gro_complete
,
1448 static const struct net_protocol udp_protocol
= {
1450 .err_handler
= udp_err
,
1451 .gso_send_check
= udp4_ufo_send_check
,
1452 .gso_segment
= udp4_ufo_fragment
,
1457 static const struct net_protocol icmp_protocol
= {
1458 .handler
= icmp_rcv
,
1463 static __net_init
int ipv4_mib_init_net(struct net
*net
)
1465 if (snmp_mib_init((void **)net
->mib
.tcp_statistics
,
1466 sizeof(struct tcp_mib
)) < 0)
1468 if (snmp_mib_init((void **)net
->mib
.ip_statistics
,
1469 sizeof(struct ipstats_mib
)) < 0)
1471 if (snmp_mib_init((void **)net
->mib
.net_statistics
,
1472 sizeof(struct linux_mib
)) < 0)
1474 if (snmp_mib_init((void **)net
->mib
.udp_statistics
,
1475 sizeof(struct udp_mib
)) < 0)
1477 if (snmp_mib_init((void **)net
->mib
.udplite_statistics
,
1478 sizeof(struct udp_mib
)) < 0)
1479 goto err_udplite_mib
;
1480 if (snmp_mib_init((void **)net
->mib
.icmp_statistics
,
1481 sizeof(struct icmp_mib
)) < 0)
1483 if (snmp_mib_init((void **)net
->mib
.icmpmsg_statistics
,
1484 sizeof(struct icmpmsg_mib
)) < 0)
1485 goto err_icmpmsg_mib
;
1491 snmp_mib_free((void **)net
->mib
.icmp_statistics
);
1493 snmp_mib_free((void **)net
->mib
.udplite_statistics
);
1495 snmp_mib_free((void **)net
->mib
.udp_statistics
);
1497 snmp_mib_free((void **)net
->mib
.net_statistics
);
1499 snmp_mib_free((void **)net
->mib
.ip_statistics
);
1501 snmp_mib_free((void **)net
->mib
.tcp_statistics
);
1506 static __net_exit
void ipv4_mib_exit_net(struct net
*net
)
1508 snmp_mib_free((void **)net
->mib
.icmpmsg_statistics
);
1509 snmp_mib_free((void **)net
->mib
.icmp_statistics
);
1510 snmp_mib_free((void **)net
->mib
.udplite_statistics
);
1511 snmp_mib_free((void **)net
->mib
.udp_statistics
);
1512 snmp_mib_free((void **)net
->mib
.net_statistics
);
1513 snmp_mib_free((void **)net
->mib
.ip_statistics
);
1514 snmp_mib_free((void **)net
->mib
.tcp_statistics
);
1517 static __net_initdata
struct pernet_operations ipv4_mib_ops
= {
1518 .init
= ipv4_mib_init_net
,
1519 .exit
= ipv4_mib_exit_net
,
1522 static int __init
init_ipv4_mibs(void)
1524 return register_pernet_subsys(&ipv4_mib_ops
);
1527 static int ipv4_proc_init(void);
1530 * IP protocol layer initialiser
1533 static struct packet_type ip_packet_type __read_mostly
= {
1534 .type
= cpu_to_be16(ETH_P_IP
),
1536 .gso_send_check
= inet_gso_send_check
,
1537 .gso_segment
= inet_gso_segment
,
1538 .gro_receive
= inet_gro_receive
,
1539 .gro_complete
= inet_gro_complete
,
1542 static int __init
inet_init(void)
1544 struct sk_buff
*dummy_skb
;
1545 struct inet_protosw
*q
;
1546 struct list_head
*r
;
1549 BUILD_BUG_ON(sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
));
1551 rc
= proto_register(&tcp_prot
, 1);
1555 rc
= proto_register(&udp_prot
, 1);
1557 goto out_unregister_tcp_proto
;
1559 rc
= proto_register(&raw_prot
, 1);
1561 goto out_unregister_udp_proto
;
1564 * Tell SOCKET that we are alive...
1567 (void)sock_register(&inet_family_ops
);
1569 #ifdef CONFIG_SYSCTL
1570 ip_static_sysctl_init();
1574 * Add all the base protocols.
1577 if (inet_add_protocol(&icmp_protocol
, IPPROTO_ICMP
) < 0)
1578 printk(KERN_CRIT
"inet_init: Cannot add ICMP protocol\n");
1579 if (inet_add_protocol(&udp_protocol
, IPPROTO_UDP
) < 0)
1580 printk(KERN_CRIT
"inet_init: Cannot add UDP protocol\n");
1581 if (inet_add_protocol(&tcp_protocol
, IPPROTO_TCP
) < 0)
1582 printk(KERN_CRIT
"inet_init: Cannot add TCP protocol\n");
1583 #ifdef CONFIG_IP_MULTICAST
1584 if (inet_add_protocol(&igmp_protocol
, IPPROTO_IGMP
) < 0)
1585 printk(KERN_CRIT
"inet_init: Cannot add IGMP protocol\n");
1588 /* Register the socket-side information for inet_create. */
1589 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1592 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1593 inet_register_protosw(q
);
1596 * Set the ARP module up
1602 * Set the IP module up
1609 /* Setup TCP slab cache for open requests. */
1612 /* Setup UDP memory threshold */
1615 /* Add UDP-Lite (RFC 3828) */
1616 udplite4_register();
1619 * Set the ICMP layer up
1622 if (icmp_init() < 0)
1623 panic("Failed to create the ICMP control socket.\n");
1626 * Initialise the multicast router
1628 #if defined(CONFIG_IP_MROUTE)
1630 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mroute\n");
1633 * Initialise per-cpu ipv4 mibs
1636 if (init_ipv4_mibs())
1637 printk(KERN_CRIT
"inet_init: Cannot init ipv4 mibs\n");
1643 dev_add_pack(&ip_packet_type
);
1648 out_unregister_udp_proto
:
1649 proto_unregister(&udp_prot
);
1650 out_unregister_tcp_proto
:
1651 proto_unregister(&tcp_prot
);
1655 fs_initcall(inet_init
);
1657 /* ------------------------------------------------------------------------ */
1659 #ifdef CONFIG_PROC_FS
1660 static int __init
ipv4_proc_init(void)
1664 if (raw_proc_init())
1666 if (tcp4_proc_init())
1668 if (udp4_proc_init())
1670 if (ip_misc_proc_init())
1685 #else /* CONFIG_PROC_FS */
1686 static int __init
ipv4_proc_init(void)
1690 #endif /* CONFIG_PROC_FS */
1692 MODULE_ALIAS_NETPROTO(PF_INET
);