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 #define pr_fmt(fmt) "IPv4: " fmt
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
93 #include <asm/uaccess.h>
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
101 #include <net/protocol.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
113 #include <net/icmp.h>
114 #include <net/ipip.h>
115 #include <net/inet_common.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
123 /* The inetsw table contains everything that inet_create needs to
124 * build a new socket.
126 static struct list_head inetsw
[SOCK_MAX
];
127 static DEFINE_SPINLOCK(inetsw_lock
);
129 struct ipv4_config ipv4_config
;
130 EXPORT_SYMBOL(ipv4_config
);
132 /* New destruction routine */
134 void inet_sock_destruct(struct sock
*sk
)
136 struct inet_sock
*inet
= inet_sk(sk
);
138 __skb_queue_purge(&sk
->sk_receive_queue
);
139 __skb_queue_purge(&sk
->sk_error_queue
);
143 if (sk
->sk_type
== SOCK_STREAM
&& sk
->sk_state
!= TCP_CLOSE
) {
144 pr_err("Attempt to release TCP socket in state %d %p\n",
148 if (!sock_flag(sk
, SOCK_DEAD
)) {
149 pr_err("Attempt to release alive inet socket %p\n", sk
);
153 WARN_ON(atomic_read(&sk
->sk_rmem_alloc
));
154 WARN_ON(atomic_read(&sk
->sk_wmem_alloc
));
155 WARN_ON(sk
->sk_wmem_queued
);
156 WARN_ON(sk
->sk_forward_alloc
);
158 kfree(rcu_dereference_protected(inet
->inet_opt
, 1));
159 dst_release(rcu_dereference_check(sk
->sk_dst_cache
, 1));
160 dst_release(sk
->sk_rx_dst
);
161 sk_refcnt_debug_dec(sk
);
163 EXPORT_SYMBOL(inet_sock_destruct
);
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
172 * Automatically bind an unbound socket.
175 static int inet_autobind(struct sock
*sk
)
177 struct inet_sock
*inet
;
178 /* We may need to bind the socket. */
181 if (!inet
->inet_num
) {
182 if (sk
->sk_prot
->get_port(sk
, 0)) {
186 inet
->inet_sport
= htons(inet
->inet_num
);
193 * Move a socket into listening state.
195 int inet_listen(struct socket
*sock
, int backlog
)
197 struct sock
*sk
= sock
->sk
;
198 unsigned char old_state
;
204 if (sock
->state
!= SS_UNCONNECTED
|| sock
->type
!= SOCK_STREAM
)
207 old_state
= sk
->sk_state
;
208 if (!((1 << old_state
) & (TCPF_CLOSE
| TCPF_LISTEN
)))
211 /* Really, if the socket is already in listen state
212 * we can only allow the backlog to be adjusted.
214 if (old_state
!= TCP_LISTEN
) {
215 /* Check special setups for testing purpose to enable TFO w/o
216 * requiring TCP_FASTOPEN sockopt.
217 * Note that only TCP sockets (SOCK_STREAM) will reach here.
218 * Also fastopenq may already been allocated because this
219 * socket was in TCP_LISTEN state previously but was
220 * shutdown() (rather than close()).
222 if ((sysctl_tcp_fastopen
& TFO_SERVER_ENABLE
) != 0 &&
223 inet_csk(sk
)->icsk_accept_queue
.fastopenq
== NULL
) {
224 if ((sysctl_tcp_fastopen
& TFO_SERVER_WO_SOCKOPT1
) != 0)
225 err
= fastopen_init_queue(sk
, backlog
);
226 else if ((sysctl_tcp_fastopen
&
227 TFO_SERVER_WO_SOCKOPT2
) != 0)
228 err
= fastopen_init_queue(sk
,
229 ((uint
)sysctl_tcp_fastopen
) >> 16);
235 err
= inet_csk_listen_start(sk
, backlog
);
239 sk
->sk_max_ack_backlog
= backlog
;
246 EXPORT_SYMBOL(inet_listen
);
248 u32 inet_ehash_secret __read_mostly
;
249 EXPORT_SYMBOL(inet_ehash_secret
);
252 * inet_ehash_secret must be set exactly once
254 void build_ehash_secret(void)
259 get_random_bytes(&rnd
, sizeof(rnd
));
262 cmpxchg(&inet_ehash_secret
, 0, rnd
);
264 EXPORT_SYMBOL(build_ehash_secret
);
266 static inline int inet_netns_ok(struct net
*net
, __u8 protocol
)
268 const struct net_protocol
*ipprot
;
270 if (net_eq(net
, &init_net
))
273 ipprot
= rcu_dereference(inet_protos
[protocol
]);
274 if (ipprot
== NULL
) {
278 return ipprot
->netns_ok
;
282 * Create an inet socket.
285 static int inet_create(struct net
*net
, struct socket
*sock
, int protocol
,
289 struct inet_protosw
*answer
;
290 struct inet_sock
*inet
;
291 struct proto
*answer_prot
;
292 unsigned char answer_flags
;
293 char answer_no_check
;
294 int try_loading_module
= 0;
297 if (unlikely(!inet_ehash_secret
))
298 if (sock
->type
!= SOCK_RAW
&& sock
->type
!= SOCK_DGRAM
)
299 build_ehash_secret();
301 sock
->state
= SS_UNCONNECTED
;
303 /* Look for the requested type/protocol pair. */
305 err
= -ESOCKTNOSUPPORT
;
307 list_for_each_entry_rcu(answer
, &inetsw
[sock
->type
], list
) {
310 /* Check the non-wild match. */
311 if (protocol
== answer
->protocol
) {
312 if (protocol
!= IPPROTO_IP
)
315 /* Check for the two wild cases. */
316 if (IPPROTO_IP
== protocol
) {
317 protocol
= answer
->protocol
;
320 if (IPPROTO_IP
== answer
->protocol
)
323 err
= -EPROTONOSUPPORT
;
327 if (try_loading_module
< 2) {
330 * Be more specific, e.g. net-pf-2-proto-132-type-1
331 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
333 if (++try_loading_module
== 1)
334 request_module("net-pf-%d-proto-%d-type-%d",
335 PF_INET
, protocol
, sock
->type
);
337 * Fall back to generic, e.g. net-pf-2-proto-132
338 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
341 request_module("net-pf-%d-proto-%d",
343 goto lookup_protocol
;
349 if (sock
->type
== SOCK_RAW
&& !kern
&& !capable(CAP_NET_RAW
))
353 if (!inet_netns_ok(net
, protocol
))
356 sock
->ops
= answer
->ops
;
357 answer_prot
= answer
->prot
;
358 answer_no_check
= answer
->no_check
;
359 answer_flags
= answer
->flags
;
362 WARN_ON(answer_prot
->slab
== NULL
);
365 sk
= sk_alloc(net
, PF_INET
, GFP_KERNEL
, answer_prot
);
370 sk
->sk_no_check
= answer_no_check
;
371 if (INET_PROTOSW_REUSE
& answer_flags
)
372 sk
->sk_reuse
= SK_CAN_REUSE
;
375 inet
->is_icsk
= (INET_PROTOSW_ICSK
& answer_flags
) != 0;
379 if (SOCK_RAW
== sock
->type
) {
380 inet
->inet_num
= protocol
;
381 if (IPPROTO_RAW
== protocol
)
385 if (ipv4_config
.no_pmtu_disc
)
386 inet
->pmtudisc
= IP_PMTUDISC_DONT
;
388 inet
->pmtudisc
= IP_PMTUDISC_WANT
;
392 sock_init_data(sock
, sk
);
394 sk
->sk_destruct
= inet_sock_destruct
;
395 sk
->sk_protocol
= protocol
;
396 sk
->sk_backlog_rcv
= sk
->sk_prot
->backlog_rcv
;
403 inet
->mc_list
= NULL
;
406 sk_refcnt_debug_inc(sk
);
408 if (inet
->inet_num
) {
409 /* It assumes that any protocol which allows
410 * the user to assign a number at socket
411 * creation time automatically
414 inet
->inet_sport
= htons(inet
->inet_num
);
415 /* Add to protocol hash chains. */
416 sk
->sk_prot
->hash(sk
);
419 if (sk
->sk_prot
->init
) {
420 err
= sk
->sk_prot
->init(sk
);
422 sk_common_release(sk
);
433 * The peer socket should always be NULL (or else). When we call this
434 * function we are destroying the object and from then on nobody
435 * should refer to it.
437 int inet_release(struct socket
*sock
)
439 struct sock
*sk
= sock
->sk
;
444 sock_rps_reset_flow(sk
);
446 /* Applications forget to leave groups before exiting */
447 ip_mc_drop_socket(sk
);
449 /* If linger is set, we don't return until the close
450 * is complete. Otherwise we return immediately. The
451 * actually closing is done the same either way.
453 * If the close is due to the process exiting, we never
457 if (sock_flag(sk
, SOCK_LINGER
) &&
458 !(current
->flags
& PF_EXITING
))
459 timeout
= sk
->sk_lingertime
;
461 sk
->sk_prot
->close(sk
, timeout
);
465 EXPORT_SYMBOL(inet_release
);
467 /* It is off by default, see below. */
468 int sysctl_ip_nonlocal_bind __read_mostly
;
469 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind
);
471 int inet_bind(struct socket
*sock
, struct sockaddr
*uaddr
, int addr_len
)
473 struct sockaddr_in
*addr
= (struct sockaddr_in
*)uaddr
;
474 struct sock
*sk
= sock
->sk
;
475 struct inet_sock
*inet
= inet_sk(sk
);
480 /* If the socket has its own bind function then use it. (RAW) */
481 if (sk
->sk_prot
->bind
) {
482 err
= sk
->sk_prot
->bind(sk
, uaddr
, addr_len
);
486 if (addr_len
< sizeof(struct sockaddr_in
))
489 if (addr
->sin_family
!= AF_INET
) {
490 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
491 * only if s_addr is INADDR_ANY.
494 if (addr
->sin_family
!= AF_UNSPEC
||
495 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
))
499 chk_addr_ret
= inet_addr_type(sock_net(sk
), addr
->sin_addr
.s_addr
);
501 /* Not specified by any standard per-se, however it breaks too
502 * many applications when removed. It is unfortunate since
503 * allowing applications to make a non-local bind solves
504 * several problems with systems using dynamic addressing.
505 * (ie. your servers still start up even if your ISDN link
506 * is temporarily down)
508 err
= -EADDRNOTAVAIL
;
509 if (!sysctl_ip_nonlocal_bind
&&
510 !(inet
->freebind
|| inet
->transparent
) &&
511 addr
->sin_addr
.s_addr
!= htonl(INADDR_ANY
) &&
512 chk_addr_ret
!= RTN_LOCAL
&&
513 chk_addr_ret
!= RTN_MULTICAST
&&
514 chk_addr_ret
!= RTN_BROADCAST
)
517 snum
= ntohs(addr
->sin_port
);
519 if (snum
&& snum
< PROT_SOCK
&& !capable(CAP_NET_BIND_SERVICE
))
522 /* We keep a pair of addresses. rcv_saddr is the one
523 * used by hash lookups, and saddr is used for transmit.
525 * In the BSD API these are the same except where it
526 * would be illegal to use them (multicast/broadcast) in
527 * which case the sending device address is used.
531 /* Check these errors (active socket, double bind). */
533 if (sk
->sk_state
!= TCP_CLOSE
|| inet
->inet_num
)
534 goto out_release_sock
;
536 inet
->inet_rcv_saddr
= inet
->inet_saddr
= addr
->sin_addr
.s_addr
;
537 if (chk_addr_ret
== RTN_MULTICAST
|| chk_addr_ret
== RTN_BROADCAST
)
538 inet
->inet_saddr
= 0; /* Use device */
540 /* Make sure we are allowed to bind here. */
541 if (sk
->sk_prot
->get_port(sk
, snum
)) {
542 inet
->inet_saddr
= inet
->inet_rcv_saddr
= 0;
544 goto out_release_sock
;
547 if (inet
->inet_rcv_saddr
)
548 sk
->sk_userlocks
|= SOCK_BINDADDR_LOCK
;
550 sk
->sk_userlocks
|= SOCK_BINDPORT_LOCK
;
551 inet
->inet_sport
= htons(inet
->inet_num
);
552 inet
->inet_daddr
= 0;
553 inet
->inet_dport
= 0;
561 EXPORT_SYMBOL(inet_bind
);
563 int inet_dgram_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
564 int addr_len
, int flags
)
566 struct sock
*sk
= sock
->sk
;
568 if (addr_len
< sizeof(uaddr
->sa_family
))
570 if (uaddr
->sa_family
== AF_UNSPEC
)
571 return sk
->sk_prot
->disconnect(sk
, flags
);
573 if (!inet_sk(sk
)->inet_num
&& inet_autobind(sk
))
575 return sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
577 EXPORT_SYMBOL(inet_dgram_connect
);
579 static long inet_wait_for_connect(struct sock
*sk
, long timeo
, int writebias
)
583 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
584 sk
->sk_write_pending
+= writebias
;
586 /* Basic assumption: if someone sets sk->sk_err, he _must_
587 * change state of the socket from TCP_SYN_*.
588 * Connect() does not allow to get error notifications
589 * without closing the socket.
591 while ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
593 timeo
= schedule_timeout(timeo
);
595 if (signal_pending(current
) || !timeo
)
597 prepare_to_wait(sk_sleep(sk
), &wait
, TASK_INTERRUPTIBLE
);
599 finish_wait(sk_sleep(sk
), &wait
);
600 sk
->sk_write_pending
-= writebias
;
605 * Connect to a remote host. There is regrettably still a little
606 * TCP 'magic' in here.
608 int __inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
609 int addr_len
, int flags
)
611 struct sock
*sk
= sock
->sk
;
615 if (addr_len
< sizeof(uaddr
->sa_family
))
618 if (uaddr
->sa_family
== AF_UNSPEC
) {
619 err
= sk
->sk_prot
->disconnect(sk
, flags
);
620 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
624 switch (sock
->state
) {
633 /* Fall out of switch with err, set for this state */
637 if (sk
->sk_state
!= TCP_CLOSE
)
640 err
= sk
->sk_prot
->connect(sk
, uaddr
, addr_len
);
644 sock
->state
= SS_CONNECTING
;
646 /* Just entered SS_CONNECTING state; the only
647 * difference is that return value in non-blocking
648 * case is EINPROGRESS, rather than EALREADY.
654 timeo
= sock_sndtimeo(sk
, flags
& O_NONBLOCK
);
656 if ((1 << sk
->sk_state
) & (TCPF_SYN_SENT
| TCPF_SYN_RECV
)) {
657 int writebias
= (sk
->sk_protocol
== IPPROTO_TCP
) &&
658 tcp_sk(sk
)->fastopen_req
&&
659 tcp_sk(sk
)->fastopen_req
->data
? 1 : 0;
661 /* Error code is set above */
662 if (!timeo
|| !inet_wait_for_connect(sk
, timeo
, writebias
))
665 err
= sock_intr_errno(timeo
);
666 if (signal_pending(current
))
670 /* Connection was closed by RST, timeout, ICMP error
671 * or another process disconnected us.
673 if (sk
->sk_state
== TCP_CLOSE
)
676 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
677 * and error was received after socket entered established state.
678 * Hence, it is handled normally after connect() return successfully.
681 sock
->state
= SS_CONNECTED
;
687 err
= sock_error(sk
) ? : -ECONNABORTED
;
688 sock
->state
= SS_UNCONNECTED
;
689 if (sk
->sk_prot
->disconnect(sk
, flags
))
690 sock
->state
= SS_DISCONNECTING
;
693 EXPORT_SYMBOL(__inet_stream_connect
);
695 int inet_stream_connect(struct socket
*sock
, struct sockaddr
*uaddr
,
696 int addr_len
, int flags
)
701 err
= __inet_stream_connect(sock
, uaddr
, addr_len
, flags
);
702 release_sock(sock
->sk
);
705 EXPORT_SYMBOL(inet_stream_connect
);
708 * Accept a pending connection. The TCP layer now gives BSD semantics.
711 int inet_accept(struct socket
*sock
, struct socket
*newsock
, int flags
)
713 struct sock
*sk1
= sock
->sk
;
715 struct sock
*sk2
= sk1
->sk_prot
->accept(sk1
, flags
, &err
);
722 sock_rps_record_flow(sk2
);
723 WARN_ON(!((1 << sk2
->sk_state
) &
724 (TCPF_ESTABLISHED
| TCPF_SYN_RECV
|
725 TCPF_CLOSE_WAIT
| TCPF_CLOSE
)));
727 sock_graft(sk2
, newsock
);
729 newsock
->state
= SS_CONNECTED
;
735 EXPORT_SYMBOL(inet_accept
);
739 * This does both peername and sockname.
741 int inet_getname(struct socket
*sock
, struct sockaddr
*uaddr
,
742 int *uaddr_len
, int peer
)
744 struct sock
*sk
= sock
->sk
;
745 struct inet_sock
*inet
= inet_sk(sk
);
746 DECLARE_SOCKADDR(struct sockaddr_in
*, sin
, uaddr
);
748 sin
->sin_family
= AF_INET
;
750 if (!inet
->inet_dport
||
751 (((1 << sk
->sk_state
) & (TCPF_CLOSE
| TCPF_SYN_SENT
)) &&
754 sin
->sin_port
= inet
->inet_dport
;
755 sin
->sin_addr
.s_addr
= inet
->inet_daddr
;
757 __be32 addr
= inet
->inet_rcv_saddr
;
759 addr
= inet
->inet_saddr
;
760 sin
->sin_port
= inet
->inet_sport
;
761 sin
->sin_addr
.s_addr
= addr
;
763 memset(sin
->sin_zero
, 0, sizeof(sin
->sin_zero
));
764 *uaddr_len
= sizeof(*sin
);
767 EXPORT_SYMBOL(inet_getname
);
769 int inet_sendmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
772 struct sock
*sk
= sock
->sk
;
774 sock_rps_record_flow(sk
);
776 /* We may need to bind the socket. */
777 if (!inet_sk(sk
)->inet_num
&& !sk
->sk_prot
->no_autobind
&&
781 return sk
->sk_prot
->sendmsg(iocb
, sk
, msg
, size
);
783 EXPORT_SYMBOL(inet_sendmsg
);
785 ssize_t
inet_sendpage(struct socket
*sock
, struct page
*page
, int offset
,
786 size_t size
, int flags
)
788 struct sock
*sk
= sock
->sk
;
790 sock_rps_record_flow(sk
);
792 /* We may need to bind the socket. */
793 if (!inet_sk(sk
)->inet_num
&& !sk
->sk_prot
->no_autobind
&&
797 if (sk
->sk_prot
->sendpage
)
798 return sk
->sk_prot
->sendpage(sk
, page
, offset
, size
, flags
);
799 return sock_no_sendpage(sock
, page
, offset
, size
, flags
);
801 EXPORT_SYMBOL(inet_sendpage
);
803 int inet_recvmsg(struct kiocb
*iocb
, struct socket
*sock
, struct msghdr
*msg
,
804 size_t size
, int flags
)
806 struct sock
*sk
= sock
->sk
;
810 sock_rps_record_flow(sk
);
812 err
= sk
->sk_prot
->recvmsg(iocb
, sk
, msg
, size
, flags
& MSG_DONTWAIT
,
813 flags
& ~MSG_DONTWAIT
, &addr_len
);
815 msg
->msg_namelen
= addr_len
;
818 EXPORT_SYMBOL(inet_recvmsg
);
820 int inet_shutdown(struct socket
*sock
, int how
)
822 struct sock
*sk
= sock
->sk
;
825 /* This should really check to make sure
826 * the socket is a TCP socket. (WHY AC...)
828 how
++; /* maps 0->1 has the advantage of making bit 1 rcvs and
831 if ((how
& ~SHUTDOWN_MASK
) || !how
) /* MAXINT->0 */
835 if (sock
->state
== SS_CONNECTING
) {
836 if ((1 << sk
->sk_state
) &
837 (TCPF_SYN_SENT
| TCPF_SYN_RECV
| TCPF_CLOSE
))
838 sock
->state
= SS_DISCONNECTING
;
840 sock
->state
= SS_CONNECTED
;
843 switch (sk
->sk_state
) {
846 /* Hack to wake up other listeners, who can poll for
847 POLLHUP, even on eg. unconnected UDP sockets -- RR */
849 sk
->sk_shutdown
|= how
;
850 if (sk
->sk_prot
->shutdown
)
851 sk
->sk_prot
->shutdown(sk
, how
);
854 /* Remaining two branches are temporary solution for missing
855 * close() in multithreaded environment. It is _not_ a good idea,
856 * but we have no choice until close() is repaired at VFS level.
859 if (!(how
& RCV_SHUTDOWN
))
863 err
= sk
->sk_prot
->disconnect(sk
, O_NONBLOCK
);
864 sock
->state
= err
? SS_DISCONNECTING
: SS_UNCONNECTED
;
868 /* Wake up anyone sleeping in poll. */
869 sk
->sk_state_change(sk
);
873 EXPORT_SYMBOL(inet_shutdown
);
876 * ioctl() calls you can issue on an INET socket. Most of these are
877 * device configuration and stuff and very rarely used. Some ioctls
878 * pass on to the socket itself.
880 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
881 * loads the devconfigure module does its configuring and unloads it.
882 * There's a good 20K of config code hanging around the kernel.
885 int inet_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
887 struct sock
*sk
= sock
->sk
;
889 struct net
*net
= sock_net(sk
);
893 err
= sock_get_timestamp(sk
, (struct timeval __user
*)arg
);
896 err
= sock_get_timestampns(sk
, (struct timespec __user
*)arg
);
901 err
= ip_rt_ioctl(net
, cmd
, (void __user
*)arg
);
906 err
= arp_ioctl(net
, cmd
, (void __user
*)arg
);
919 err
= devinet_ioctl(net
, cmd
, (void __user
*)arg
);
922 if (sk
->sk_prot
->ioctl
)
923 err
= sk
->sk_prot
->ioctl(sk
, cmd
, arg
);
930 EXPORT_SYMBOL(inet_ioctl
);
933 static int inet_compat_ioctl(struct socket
*sock
, unsigned int cmd
, unsigned long arg
)
935 struct sock
*sk
= sock
->sk
;
936 int err
= -ENOIOCTLCMD
;
938 if (sk
->sk_prot
->compat_ioctl
)
939 err
= sk
->sk_prot
->compat_ioctl(sk
, cmd
, arg
);
945 const struct proto_ops inet_stream_ops
= {
947 .owner
= THIS_MODULE
,
948 .release
= inet_release
,
950 .connect
= inet_stream_connect
,
951 .socketpair
= sock_no_socketpair
,
952 .accept
= inet_accept
,
953 .getname
= inet_getname
,
956 .listen
= inet_listen
,
957 .shutdown
= inet_shutdown
,
958 .setsockopt
= sock_common_setsockopt
,
959 .getsockopt
= sock_common_getsockopt
,
960 .sendmsg
= inet_sendmsg
,
961 .recvmsg
= inet_recvmsg
,
962 .mmap
= sock_no_mmap
,
963 .sendpage
= inet_sendpage
,
964 .splice_read
= tcp_splice_read
,
966 .compat_setsockopt
= compat_sock_common_setsockopt
,
967 .compat_getsockopt
= compat_sock_common_getsockopt
,
968 .compat_ioctl
= inet_compat_ioctl
,
971 EXPORT_SYMBOL(inet_stream_ops
);
973 const struct proto_ops inet_dgram_ops
= {
975 .owner
= THIS_MODULE
,
976 .release
= inet_release
,
978 .connect
= inet_dgram_connect
,
979 .socketpair
= sock_no_socketpair
,
980 .accept
= sock_no_accept
,
981 .getname
= inet_getname
,
984 .listen
= sock_no_listen
,
985 .shutdown
= inet_shutdown
,
986 .setsockopt
= sock_common_setsockopt
,
987 .getsockopt
= sock_common_getsockopt
,
988 .sendmsg
= inet_sendmsg
,
989 .recvmsg
= inet_recvmsg
,
990 .mmap
= sock_no_mmap
,
991 .sendpage
= inet_sendpage
,
993 .compat_setsockopt
= compat_sock_common_setsockopt
,
994 .compat_getsockopt
= compat_sock_common_getsockopt
,
995 .compat_ioctl
= inet_compat_ioctl
,
998 EXPORT_SYMBOL(inet_dgram_ops
);
1001 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1004 static const struct proto_ops inet_sockraw_ops
= {
1006 .owner
= THIS_MODULE
,
1007 .release
= inet_release
,
1009 .connect
= inet_dgram_connect
,
1010 .socketpair
= sock_no_socketpair
,
1011 .accept
= sock_no_accept
,
1012 .getname
= inet_getname
,
1013 .poll
= datagram_poll
,
1014 .ioctl
= inet_ioctl
,
1015 .listen
= sock_no_listen
,
1016 .shutdown
= inet_shutdown
,
1017 .setsockopt
= sock_common_setsockopt
,
1018 .getsockopt
= sock_common_getsockopt
,
1019 .sendmsg
= inet_sendmsg
,
1020 .recvmsg
= inet_recvmsg
,
1021 .mmap
= sock_no_mmap
,
1022 .sendpage
= inet_sendpage
,
1023 #ifdef CONFIG_COMPAT
1024 .compat_setsockopt
= compat_sock_common_setsockopt
,
1025 .compat_getsockopt
= compat_sock_common_getsockopt
,
1026 .compat_ioctl
= inet_compat_ioctl
,
1030 static const struct net_proto_family inet_family_ops
= {
1032 .create
= inet_create
,
1033 .owner
= THIS_MODULE
,
1036 /* Upon startup we insert all the elements in inetsw_array[] into
1037 * the linked list inetsw.
1039 static struct inet_protosw inetsw_array
[] =
1042 .type
= SOCK_STREAM
,
1043 .protocol
= IPPROTO_TCP
,
1045 .ops
= &inet_stream_ops
,
1047 .flags
= INET_PROTOSW_PERMANENT
|
1053 .protocol
= IPPROTO_UDP
,
1055 .ops
= &inet_dgram_ops
,
1056 .no_check
= UDP_CSUM_DEFAULT
,
1057 .flags
= INET_PROTOSW_PERMANENT
,
1062 .protocol
= IPPROTO_ICMP
,
1064 .ops
= &inet_dgram_ops
,
1065 .no_check
= UDP_CSUM_DEFAULT
,
1066 .flags
= INET_PROTOSW_REUSE
,
1071 .protocol
= IPPROTO_IP
, /* wild card */
1073 .ops
= &inet_sockraw_ops
,
1074 .no_check
= UDP_CSUM_DEFAULT
,
1075 .flags
= INET_PROTOSW_REUSE
,
1079 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1081 void inet_register_protosw(struct inet_protosw
*p
)
1083 struct list_head
*lh
;
1084 struct inet_protosw
*answer
;
1085 int protocol
= p
->protocol
;
1086 struct list_head
*last_perm
;
1088 spin_lock_bh(&inetsw_lock
);
1090 if (p
->type
>= SOCK_MAX
)
1093 /* If we are trying to override a permanent protocol, bail. */
1095 last_perm
= &inetsw
[p
->type
];
1096 list_for_each(lh
, &inetsw
[p
->type
]) {
1097 answer
= list_entry(lh
, struct inet_protosw
, list
);
1099 /* Check only the non-wild match. */
1100 if (INET_PROTOSW_PERMANENT
& answer
->flags
) {
1101 if (protocol
== answer
->protocol
)
1111 /* Add the new entry after the last permanent entry if any, so that
1112 * the new entry does not override a permanent entry when matched with
1113 * a wild-card protocol. But it is allowed to override any existing
1114 * non-permanent entry. This means that when we remove this entry, the
1115 * system automatically returns to the old behavior.
1117 list_add_rcu(&p
->list
, last_perm
);
1119 spin_unlock_bh(&inetsw_lock
);
1124 pr_err("Attempt to override permanent protocol %d\n", protocol
);
1128 pr_err("Ignoring attempt to register invalid socket type %d\n",
1132 EXPORT_SYMBOL(inet_register_protosw
);
1134 void inet_unregister_protosw(struct inet_protosw
*p
)
1136 if (INET_PROTOSW_PERMANENT
& p
->flags
) {
1137 pr_err("Attempt to unregister permanent protocol %d\n",
1140 spin_lock_bh(&inetsw_lock
);
1141 list_del_rcu(&p
->list
);
1142 spin_unlock_bh(&inetsw_lock
);
1147 EXPORT_SYMBOL(inet_unregister_protosw
);
1150 * Shall we try to damage output packets if routing dev changes?
1153 int sysctl_ip_dynaddr __read_mostly
;
1155 static int inet_sk_reselect_saddr(struct sock
*sk
)
1157 struct inet_sock
*inet
= inet_sk(sk
);
1158 __be32 old_saddr
= inet
->inet_saddr
;
1159 __be32 daddr
= inet
->inet_daddr
;
1163 struct ip_options_rcu
*inet_opt
;
1165 inet_opt
= rcu_dereference_protected(inet
->inet_opt
,
1166 sock_owned_by_user(sk
));
1167 if (inet_opt
&& inet_opt
->opt
.srr
)
1168 daddr
= inet_opt
->opt
.faddr
;
1170 /* Query new route. */
1171 fl4
= &inet
->cork
.fl
.u
.ip4
;
1172 rt
= ip_route_connect(fl4
, daddr
, 0, RT_CONN_FLAGS(sk
),
1173 sk
->sk_bound_dev_if
, sk
->sk_protocol
,
1174 inet
->inet_sport
, inet
->inet_dport
, sk
, false);
1178 sk_setup_caps(sk
, &rt
->dst
);
1180 new_saddr
= fl4
->saddr
;
1182 if (new_saddr
== old_saddr
)
1185 if (sysctl_ip_dynaddr
> 1) {
1186 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1187 __func__
, &old_saddr
, &new_saddr
);
1190 inet
->inet_saddr
= inet
->inet_rcv_saddr
= new_saddr
;
1193 * XXX The only one ugly spot where we need to
1194 * XXX really change the sockets identity after
1195 * XXX it has entered the hashes. -DaveM
1197 * Besides that, it does not check for connection
1198 * uniqueness. Wait for troubles.
1200 __sk_prot_rehash(sk
);
1204 int inet_sk_rebuild_header(struct sock
*sk
)
1206 struct inet_sock
*inet
= inet_sk(sk
);
1207 struct rtable
*rt
= (struct rtable
*)__sk_dst_check(sk
, 0);
1209 struct ip_options_rcu
*inet_opt
;
1213 /* Route is OK, nothing to do. */
1219 inet_opt
= rcu_dereference(inet
->inet_opt
);
1220 daddr
= inet
->inet_daddr
;
1221 if (inet_opt
&& inet_opt
->opt
.srr
)
1222 daddr
= inet_opt
->opt
.faddr
;
1224 fl4
= &inet
->cork
.fl
.u
.ip4
;
1225 rt
= ip_route_output_ports(sock_net(sk
), fl4
, sk
, daddr
, inet
->inet_saddr
,
1226 inet
->inet_dport
, inet
->inet_sport
,
1227 sk
->sk_protocol
, RT_CONN_FLAGS(sk
),
1228 sk
->sk_bound_dev_if
);
1231 sk_setup_caps(sk
, &rt
->dst
);
1235 /* Routing failed... */
1236 sk
->sk_route_caps
= 0;
1238 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1239 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1241 if (!sysctl_ip_dynaddr
||
1242 sk
->sk_state
!= TCP_SYN_SENT
||
1243 (sk
->sk_userlocks
& SOCK_BINDADDR_LOCK
) ||
1244 (err
= inet_sk_reselect_saddr(sk
)) != 0)
1245 sk
->sk_err_soft
= -err
;
1250 EXPORT_SYMBOL(inet_sk_rebuild_header
);
1252 static int inet_gso_send_check(struct sk_buff
*skb
)
1254 const struct net_protocol
*ops
;
1255 const struct iphdr
*iph
;
1260 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1265 if (ihl
< sizeof(*iph
))
1268 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1271 __skb_pull(skb
, ihl
);
1272 skb_reset_transport_header(skb
);
1274 proto
= iph
->protocol
;
1275 err
= -EPROTONOSUPPORT
;
1278 ops
= rcu_dereference(inet_protos
[proto
]);
1279 if (likely(ops
&& ops
->gso_send_check
))
1280 err
= ops
->gso_send_check(skb
);
1287 static struct sk_buff
*inet_gso_segment(struct sk_buff
*skb
,
1288 netdev_features_t features
)
1290 struct sk_buff
*segs
= ERR_PTR(-EINVAL
);
1291 const struct net_protocol
*ops
;
1296 unsigned int offset
= 0;
1298 if (!(features
& NETIF_F_V4_CSUM
))
1299 features
&= ~NETIF_F_SG
;
1301 if (unlikely(skb_shinfo(skb
)->gso_type
&
1309 if (unlikely(!pskb_may_pull(skb
, sizeof(*iph
))))
1314 if (ihl
< sizeof(*iph
))
1317 if (unlikely(!pskb_may_pull(skb
, ihl
)))
1320 __skb_pull(skb
, ihl
);
1321 skb_reset_transport_header(skb
);
1323 id
= ntohs(iph
->id
);
1324 proto
= iph
->protocol
;
1325 segs
= ERR_PTR(-EPROTONOSUPPORT
);
1328 ops
= rcu_dereference(inet_protos
[proto
]);
1329 if (likely(ops
&& ops
->gso_segment
))
1330 segs
= ops
->gso_segment(skb
, features
);
1333 if (!segs
|| IS_ERR(segs
))
1339 if (proto
== IPPROTO_UDP
) {
1340 iph
->id
= htons(id
);
1341 iph
->frag_off
= htons(offset
>> 3);
1342 if (skb
->next
!= NULL
)
1343 iph
->frag_off
|= htons(IP_MF
);
1344 offset
+= (skb
->len
- skb
->mac_len
- iph
->ihl
* 4);
1346 iph
->id
= htons(id
++);
1347 iph
->tot_len
= htons(skb
->len
- skb
->mac_len
);
1349 iph
->check
= ip_fast_csum(skb_network_header(skb
), iph
->ihl
);
1350 } while ((skb
= skb
->next
));
1356 static struct sk_buff
**inet_gro_receive(struct sk_buff
**head
,
1357 struct sk_buff
*skb
)
1359 const struct net_protocol
*ops
;
1360 struct sk_buff
**pp
= NULL
;
1362 const struct iphdr
*iph
;
1369 off
= skb_gro_offset(skb
);
1370 hlen
= off
+ sizeof(*iph
);
1371 iph
= skb_gro_header_fast(skb
, off
);
1372 if (skb_gro_header_hard(skb
, hlen
)) {
1373 iph
= skb_gro_header_slow(skb
, hlen
, off
);
1378 proto
= iph
->protocol
;
1381 ops
= rcu_dereference(inet_protos
[proto
]);
1382 if (!ops
|| !ops
->gro_receive
)
1385 if (*(u8
*)iph
!= 0x45)
1388 if (unlikely(ip_fast_csum((u8
*)iph
, 5)))
1391 id
= ntohl(*(__be32
*)&iph
->id
);
1392 flush
= (u16
)((ntohl(*(__be32
*)iph
) ^ skb_gro_len(skb
)) | (id
^ IP_DF
));
1395 for (p
= *head
; p
; p
= p
->next
) {
1398 if (!NAPI_GRO_CB(p
)->same_flow
)
1403 if ((iph
->protocol
^ iph2
->protocol
) |
1404 ((__force u32
)iph
->saddr
^ (__force u32
)iph2
->saddr
) |
1405 ((__force u32
)iph
->daddr
^ (__force u32
)iph2
->daddr
)) {
1406 NAPI_GRO_CB(p
)->same_flow
= 0;
1410 /* All fields must match except length and checksum. */
1411 NAPI_GRO_CB(p
)->flush
|=
1412 (iph
->ttl
^ iph2
->ttl
) |
1413 (iph
->tos
^ iph2
->tos
) |
1414 ((u16
)(ntohs(iph2
->id
) + NAPI_GRO_CB(p
)->count
) ^ id
);
1416 NAPI_GRO_CB(p
)->flush
|= flush
;
1419 NAPI_GRO_CB(skb
)->flush
|= flush
;
1420 skb_gro_pull(skb
, sizeof(*iph
));
1421 skb_set_transport_header(skb
, skb_gro_offset(skb
));
1423 pp
= ops
->gro_receive(head
, skb
);
1429 NAPI_GRO_CB(skb
)->flush
|= flush
;
1434 static int inet_gro_complete(struct sk_buff
*skb
)
1436 __be16 newlen
= htons(skb
->len
- skb_network_offset(skb
));
1437 struct iphdr
*iph
= ip_hdr(skb
);
1438 const struct net_protocol
*ops
;
1439 int proto
= iph
->protocol
;
1442 csum_replace2(&iph
->check
, iph
->tot_len
, newlen
);
1443 iph
->tot_len
= newlen
;
1446 ops
= rcu_dereference(inet_protos
[proto
]);
1447 if (WARN_ON(!ops
|| !ops
->gro_complete
))
1450 err
= ops
->gro_complete(skb
);
1458 int inet_ctl_sock_create(struct sock
**sk
, unsigned short family
,
1459 unsigned short type
, unsigned char protocol
,
1462 struct socket
*sock
;
1463 int rc
= sock_create_kern(family
, type
, protocol
, &sock
);
1467 (*sk
)->sk_allocation
= GFP_ATOMIC
;
1469 * Unhash it so that IP input processing does not even see it,
1470 * we do not wish this socket to see incoming packets.
1472 (*sk
)->sk_prot
->unhash(*sk
);
1474 sk_change_net(*sk
, net
);
1478 EXPORT_SYMBOL_GPL(inet_ctl_sock_create
);
1480 unsigned long snmp_fold_field(void __percpu
*mib
[], int offt
)
1482 unsigned long res
= 0;
1485 for_each_possible_cpu(i
) {
1486 for (j
= 0; j
< SNMP_ARRAY_SZ
; j
++)
1487 res
+= *(((unsigned long *) per_cpu_ptr(mib
[j
], i
)) + offt
);
1491 EXPORT_SYMBOL_GPL(snmp_fold_field
);
1493 #if BITS_PER_LONG==32
1495 u64
snmp_fold_field64(void __percpu
*mib
[], int offt
, size_t syncp_offset
)
1500 for_each_possible_cpu(cpu
) {
1502 struct u64_stats_sync
*syncp
;
1506 bhptr
= per_cpu_ptr(mib
[0], cpu
);
1507 syncp
= (struct u64_stats_sync
*)(bhptr
+ syncp_offset
);
1509 start
= u64_stats_fetch_begin_bh(syncp
);
1510 v
= *(((u64
*) bhptr
) + offt
);
1511 } while (u64_stats_fetch_retry_bh(syncp
, start
));
1517 EXPORT_SYMBOL_GPL(snmp_fold_field64
);
1520 int snmp_mib_init(void __percpu
*ptr
[2], size_t mibsize
, size_t align
)
1522 BUG_ON(ptr
== NULL
);
1523 ptr
[0] = __alloc_percpu(mibsize
, align
);
1526 #if SNMP_ARRAY_SZ == 2
1527 ptr
[1] = __alloc_percpu(mibsize
, align
);
1529 free_percpu(ptr
[0]);
1536 EXPORT_SYMBOL_GPL(snmp_mib_init
);
1538 void snmp_mib_free(void __percpu
*ptr
[SNMP_ARRAY_SZ
])
1542 BUG_ON(ptr
== NULL
);
1543 for (i
= 0; i
< SNMP_ARRAY_SZ
; i
++) {
1544 free_percpu(ptr
[i
]);
1548 EXPORT_SYMBOL_GPL(snmp_mib_free
);
1550 #ifdef CONFIG_IP_MULTICAST
1551 static const struct net_protocol igmp_protocol
= {
1552 .handler
= igmp_rcv
,
1557 static const struct net_protocol tcp_protocol
= {
1558 .early_demux
= tcp_v4_early_demux
,
1559 .handler
= tcp_v4_rcv
,
1560 .err_handler
= tcp_v4_err
,
1561 .gso_send_check
= tcp_v4_gso_send_check
,
1562 .gso_segment
= tcp_tso_segment
,
1563 .gro_receive
= tcp4_gro_receive
,
1564 .gro_complete
= tcp4_gro_complete
,
1569 static const struct net_protocol udp_protocol
= {
1571 .err_handler
= udp_err
,
1572 .gso_send_check
= udp4_ufo_send_check
,
1573 .gso_segment
= udp4_ufo_fragment
,
1578 static const struct net_protocol icmp_protocol
= {
1579 .handler
= icmp_rcv
,
1580 .err_handler
= ping_err
,
1585 static __net_init
int ipv4_mib_init_net(struct net
*net
)
1587 if (snmp_mib_init((void __percpu
**)net
->mib
.tcp_statistics
,
1588 sizeof(struct tcp_mib
),
1589 __alignof__(struct tcp_mib
)) < 0)
1591 if (snmp_mib_init((void __percpu
**)net
->mib
.ip_statistics
,
1592 sizeof(struct ipstats_mib
),
1593 __alignof__(struct ipstats_mib
)) < 0)
1595 if (snmp_mib_init((void __percpu
**)net
->mib
.net_statistics
,
1596 sizeof(struct linux_mib
),
1597 __alignof__(struct linux_mib
)) < 0)
1599 if (snmp_mib_init((void __percpu
**)net
->mib
.udp_statistics
,
1600 sizeof(struct udp_mib
),
1601 __alignof__(struct udp_mib
)) < 0)
1603 if (snmp_mib_init((void __percpu
**)net
->mib
.udplite_statistics
,
1604 sizeof(struct udp_mib
),
1605 __alignof__(struct udp_mib
)) < 0)
1606 goto err_udplite_mib
;
1607 if (snmp_mib_init((void __percpu
**)net
->mib
.icmp_statistics
,
1608 sizeof(struct icmp_mib
),
1609 __alignof__(struct icmp_mib
)) < 0)
1611 net
->mib
.icmpmsg_statistics
= kzalloc(sizeof(struct icmpmsg_mib
),
1613 if (!net
->mib
.icmpmsg_statistics
)
1614 goto err_icmpmsg_mib
;
1620 snmp_mib_free((void __percpu
**)net
->mib
.icmp_statistics
);
1622 snmp_mib_free((void __percpu
**)net
->mib
.udplite_statistics
);
1624 snmp_mib_free((void __percpu
**)net
->mib
.udp_statistics
);
1626 snmp_mib_free((void __percpu
**)net
->mib
.net_statistics
);
1628 snmp_mib_free((void __percpu
**)net
->mib
.ip_statistics
);
1630 snmp_mib_free((void __percpu
**)net
->mib
.tcp_statistics
);
1635 static __net_exit
void ipv4_mib_exit_net(struct net
*net
)
1637 kfree(net
->mib
.icmpmsg_statistics
);
1638 snmp_mib_free((void __percpu
**)net
->mib
.icmp_statistics
);
1639 snmp_mib_free((void __percpu
**)net
->mib
.udplite_statistics
);
1640 snmp_mib_free((void __percpu
**)net
->mib
.udp_statistics
);
1641 snmp_mib_free((void __percpu
**)net
->mib
.net_statistics
);
1642 snmp_mib_free((void __percpu
**)net
->mib
.ip_statistics
);
1643 snmp_mib_free((void __percpu
**)net
->mib
.tcp_statistics
);
1646 static __net_initdata
struct pernet_operations ipv4_mib_ops
= {
1647 .init
= ipv4_mib_init_net
,
1648 .exit
= ipv4_mib_exit_net
,
1651 static int __init
init_ipv4_mibs(void)
1653 return register_pernet_subsys(&ipv4_mib_ops
);
1656 static int ipv4_proc_init(void);
1659 * IP protocol layer initialiser
1662 static struct packet_type ip_packet_type __read_mostly
= {
1663 .type
= cpu_to_be16(ETH_P_IP
),
1665 .gso_send_check
= inet_gso_send_check
,
1666 .gso_segment
= inet_gso_segment
,
1667 .gro_receive
= inet_gro_receive
,
1668 .gro_complete
= inet_gro_complete
,
1671 static int __init
inet_init(void)
1673 struct sk_buff
*dummy_skb
;
1674 struct inet_protosw
*q
;
1675 struct list_head
*r
;
1678 BUILD_BUG_ON(sizeof(struct inet_skb_parm
) > sizeof(dummy_skb
->cb
));
1680 sysctl_local_reserved_ports
= kzalloc(65536 / 8, GFP_KERNEL
);
1681 if (!sysctl_local_reserved_ports
)
1684 rc
= proto_register(&tcp_prot
, 1);
1686 goto out_free_reserved_ports
;
1688 rc
= proto_register(&udp_prot
, 1);
1690 goto out_unregister_tcp_proto
;
1692 rc
= proto_register(&raw_prot
, 1);
1694 goto out_unregister_udp_proto
;
1696 rc
= proto_register(&ping_prot
, 1);
1698 goto out_unregister_raw_proto
;
1701 * Tell SOCKET that we are alive...
1704 (void)sock_register(&inet_family_ops
);
1706 #ifdef CONFIG_SYSCTL
1707 ip_static_sysctl_init();
1710 tcp_prot
.sysctl_mem
= init_net
.ipv4
.sysctl_tcp_mem
;
1713 * Add all the base protocols.
1716 if (inet_add_protocol(&icmp_protocol
, IPPROTO_ICMP
) < 0)
1717 pr_crit("%s: Cannot add ICMP protocol\n", __func__
);
1718 if (inet_add_protocol(&udp_protocol
, IPPROTO_UDP
) < 0)
1719 pr_crit("%s: Cannot add UDP protocol\n", __func__
);
1720 if (inet_add_protocol(&tcp_protocol
, IPPROTO_TCP
) < 0)
1721 pr_crit("%s: Cannot add TCP protocol\n", __func__
);
1722 #ifdef CONFIG_IP_MULTICAST
1723 if (inet_add_protocol(&igmp_protocol
, IPPROTO_IGMP
) < 0)
1724 pr_crit("%s: Cannot add IGMP protocol\n", __func__
);
1727 /* Register the socket-side information for inet_create. */
1728 for (r
= &inetsw
[0]; r
< &inetsw
[SOCK_MAX
]; ++r
)
1731 for (q
= inetsw_array
; q
< &inetsw_array
[INETSW_ARRAY_LEN
]; ++q
)
1732 inet_register_protosw(q
);
1735 * Set the ARP module up
1741 * Set the IP module up
1748 /* Setup TCP slab cache for open requests. */
1751 /* Setup UDP memory threshold */
1754 /* Add UDP-Lite (RFC 3828) */
1755 udplite4_register();
1760 * Set the ICMP layer up
1763 if (icmp_init() < 0)
1764 panic("Failed to create the ICMP control socket.\n");
1767 * Initialise the multicast router
1769 #if defined(CONFIG_IP_MROUTE)
1771 pr_crit("%s: Cannot init ipv4 mroute\n", __func__
);
1774 * Initialise per-cpu ipv4 mibs
1777 if (init_ipv4_mibs())
1778 pr_crit("%s: Cannot init ipv4 mibs\n", __func__
);
1784 dev_add_pack(&ip_packet_type
);
1789 out_unregister_raw_proto
:
1790 proto_unregister(&raw_prot
);
1791 out_unregister_udp_proto
:
1792 proto_unregister(&udp_prot
);
1793 out_unregister_tcp_proto
:
1794 proto_unregister(&tcp_prot
);
1795 out_free_reserved_ports
:
1796 kfree(sysctl_local_reserved_ports
);
1800 fs_initcall(inet_init
);
1802 /* ------------------------------------------------------------------------ */
1804 #ifdef CONFIG_PROC_FS
1805 static int __init
ipv4_proc_init(void)
1809 if (raw_proc_init())
1811 if (tcp4_proc_init())
1813 if (udp4_proc_init())
1815 if (ping_proc_init())
1817 if (ip_misc_proc_init())
1834 #else /* CONFIG_PROC_FS */
1835 static int __init
ipv4_proc_init(void)
1839 #endif /* CONFIG_PROC_FS */
1841 MODULE_ALIAS_NETPROTO(PF_INET
);