Merge branch 'linus' into core/iommu
[linux-2.6] / net / ipv4 / af_inet.c
1 /*
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.
5  *
6  *              PF_INET protocol family socket handler.
7  *
8  * Authors:     Ross Biro
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>
12  *
13  * Changes (see also sock.c)
14  *
15  *              piggy,
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
23  *                                      structures
24  *                                      when accept() ed
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
30  *              Alan Cox,
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
40  *                                      compactness.
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
46  *                                      dumbly.
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.
61  *
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.
66  */
67
68 #include <linux/err.h>
69 #include <linux/errno.h>
70 #include <linux/types.h>
71 #include <linux/socket.h>
72 #include <linux/in.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>
82 #include <linux/mm.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
90 #include <asm/uaccess.h>
91 #include <asm/system.h>
92
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>
98 #include <net/ip.h>
99 #include <net/protocol.h>
100 #include <net/arp.h>
101 #include <net/route.h>
102 #include <net/ip_fib.h>
103 #include <net/inet_connection_sock.h>
104 #include <net/tcp.h>
105 #include <net/udp.h>
106 #include <net/udplite.h>
107 #include <linux/skbuff.h>
108 #include <net/sock.h>
109 #include <net/raw.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>
117 #endif
118
119 extern void ip_mc_drop_socket(struct sock *sk);
120
121 /* The inetsw table contains everything that inet_create needs to
122  * build a new socket.
123  */
124 static struct list_head inetsw[SOCK_MAX];
125 static DEFINE_SPINLOCK(inetsw_lock);
126
127 struct ipv4_config ipv4_config;
128
129 EXPORT_SYMBOL(ipv4_config);
130
131 /* New destruction routine */
132
133 void inet_sock_destruct(struct sock *sk)
134 {
135         struct inet_sock *inet = inet_sk(sk);
136
137         __skb_queue_purge(&sk->sk_receive_queue);
138         __skb_queue_purge(&sk->sk_error_queue);
139
140         sk_mem_reclaim(sk);
141
142         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143                 printk("Attempt to release TCP socket in state %d %p\n",
144                        sk->sk_state, sk);
145                 return;
146         }
147         if (!sock_flag(sk, SOCK_DEAD)) {
148                 printk("Attempt to release alive inet socket %p\n", sk);
149                 return;
150         }
151
152         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154         WARN_ON(sk->sk_wmem_queued);
155         WARN_ON(sk->sk_forward_alloc);
156
157         kfree(inet->opt);
158         dst_release(sk->sk_dst_cache);
159         sk_refcnt_debug_dec(sk);
160 }
161
162 /*
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
165  *      the work.
166  */
167
168 /*
169  *      Automatically bind an unbound socket.
170  */
171
172 static int inet_autobind(struct sock *sk)
173 {
174         struct inet_sock *inet;
175         /* We may need to bind the socket. */
176         lock_sock(sk);
177         inet = inet_sk(sk);
178         if (!inet->num) {
179                 if (sk->sk_prot->get_port(sk, 0)) {
180                         release_sock(sk);
181                         return -EAGAIN;
182                 }
183                 inet->sport = htons(inet->num);
184         }
185         release_sock(sk);
186         return 0;
187 }
188
189 /*
190  *      Move a socket into listening state.
191  */
192 int inet_listen(struct socket *sock, int backlog)
193 {
194         struct sock *sk = sock->sk;
195         unsigned char old_state;
196         int err;
197
198         lock_sock(sk);
199
200         err = -EINVAL;
201         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
202                 goto out;
203
204         old_state = sk->sk_state;
205         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
206                 goto out;
207
208         /* Really, if the socket is already in listen state
209          * we can only allow the backlog to be adjusted.
210          */
211         if (old_state != TCP_LISTEN) {
212                 err = inet_csk_listen_start(sk, backlog);
213                 if (err)
214                         goto out;
215         }
216         sk->sk_max_ack_backlog = backlog;
217         err = 0;
218
219 out:
220         release_sock(sk);
221         return err;
222 }
223
224 u32 inet_ehash_secret __read_mostly;
225 EXPORT_SYMBOL(inet_ehash_secret);
226
227 /*
228  * inet_ehash_secret must be set exactly once
229  * Instead of using a dedicated spinlock, we (ab)use inetsw_lock
230  */
231 void build_ehash_secret(void)
232 {
233         u32 rnd;
234         do {
235                 get_random_bytes(&rnd, sizeof(rnd));
236         } while (rnd == 0);
237         spin_lock_bh(&inetsw_lock);
238         if (!inet_ehash_secret)
239                 inet_ehash_secret = rnd;
240         spin_unlock_bh(&inetsw_lock);
241 }
242 EXPORT_SYMBOL(build_ehash_secret);
243
244 static inline int inet_netns_ok(struct net *net, int protocol)
245 {
246         int hash;
247         struct net_protocol *ipprot;
248
249         if (net_eq(net, &init_net))
250                 return 1;
251
252         hash = protocol & (MAX_INET_PROTOS - 1);
253         ipprot = rcu_dereference(inet_protos[hash]);
254
255         if (ipprot == NULL)
256                 /* raw IP is OK */
257                 return 1;
258         return ipprot->netns_ok;
259 }
260
261 /*
262  *      Create an inet socket.
263  */
264
265 static int inet_create(struct net *net, struct socket *sock, int protocol)
266 {
267         struct sock *sk;
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;
274         int err;
275
276         if (unlikely(!inet_ehash_secret))
277                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
278                         build_ehash_secret();
279
280         sock->state = SS_UNCONNECTED;
281
282         /* Look for the requested type/protocol pair. */
283 lookup_protocol:
284         err = -ESOCKTNOSUPPORT;
285         rcu_read_lock();
286         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
287
288                 err = 0;
289                 /* Check the non-wild match. */
290                 if (protocol == answer->protocol) {
291                         if (protocol != IPPROTO_IP)
292                                 break;
293                 } else {
294                         /* Check for the two wild cases. */
295                         if (IPPROTO_IP == protocol) {
296                                 protocol = answer->protocol;
297                                 break;
298                         }
299                         if (IPPROTO_IP == answer->protocol)
300                                 break;
301                 }
302                 err = -EPROTONOSUPPORT;
303         }
304
305         if (unlikely(err)) {
306                 if (try_loading_module < 2) {
307                         rcu_read_unlock();
308                         /*
309                          * Be more specific, e.g. net-pf-2-proto-132-type-1
310                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
311                          */
312                         if (++try_loading_module == 1)
313                                 request_module("net-pf-%d-proto-%d-type-%d",
314                                                PF_INET, protocol, sock->type);
315                         /*
316                          * Fall back to generic, e.g. net-pf-2-proto-132
317                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
318                          */
319                         else
320                                 request_module("net-pf-%d-proto-%d",
321                                                PF_INET, protocol);
322                         goto lookup_protocol;
323                 } else
324                         goto out_rcu_unlock;
325         }
326
327         err = -EPERM;
328         if (answer->capability > 0 && !capable(answer->capability))
329                 goto out_rcu_unlock;
330
331         err = -EAFNOSUPPORT;
332         if (!inet_netns_ok(net, protocol))
333                 goto out_rcu_unlock;
334
335         sock->ops = answer->ops;
336         answer_prot = answer->prot;
337         answer_no_check = answer->no_check;
338         answer_flags = answer->flags;
339         rcu_read_unlock();
340
341         WARN_ON(answer_prot->slab == NULL);
342
343         err = -ENOBUFS;
344         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
345         if (sk == NULL)
346                 goto out;
347
348         err = 0;
349         sk->sk_no_check = answer_no_check;
350         if (INET_PROTOSW_REUSE & answer_flags)
351                 sk->sk_reuse = 1;
352
353         inet = inet_sk(sk);
354         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
355
356         if (SOCK_RAW == sock->type) {
357                 inet->num = protocol;
358                 if (IPPROTO_RAW == protocol)
359                         inet->hdrincl = 1;
360         }
361
362         if (ipv4_config.no_pmtu_disc)
363                 inet->pmtudisc = IP_PMTUDISC_DONT;
364         else
365                 inet->pmtudisc = IP_PMTUDISC_WANT;
366
367         inet->id = 0;
368
369         sock_init_data(sock, sk);
370
371         sk->sk_destruct    = inet_sock_destruct;
372         sk->sk_protocol    = protocol;
373         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
374
375         inet->uc_ttl    = -1;
376         inet->mc_loop   = 1;
377         inet->mc_ttl    = 1;
378         inet->mc_index  = 0;
379         inet->mc_list   = NULL;
380
381         sk_refcnt_debug_inc(sk);
382
383         if (inet->num) {
384                 /* It assumes that any protocol which allows
385                  * the user to assign a number at socket
386                  * creation time automatically
387                  * shares.
388                  */
389                 inet->sport = htons(inet->num);
390                 /* Add to protocol hash chains. */
391                 sk->sk_prot->hash(sk);
392         }
393
394         if (sk->sk_prot->init) {
395                 err = sk->sk_prot->init(sk);
396                 if (err)
397                         sk_common_release(sk);
398         }
399 out:
400         return err;
401 out_rcu_unlock:
402         rcu_read_unlock();
403         goto out;
404 }
405
406
407 /*
408  *      The peer socket should always be NULL (or else). When we call this
409  *      function we are destroying the object and from then on nobody
410  *      should refer to it.
411  */
412 int inet_release(struct socket *sock)
413 {
414         struct sock *sk = sock->sk;
415
416         if (sk) {
417                 long timeout;
418
419                 /* Applications forget to leave groups before exiting */
420                 ip_mc_drop_socket(sk);
421
422                 /* If linger is set, we don't return until the close
423                  * is complete.  Otherwise we return immediately. The
424                  * actually closing is done the same either way.
425                  *
426                  * If the close is due to the process exiting, we never
427                  * linger..
428                  */
429                 timeout = 0;
430                 if (sock_flag(sk, SOCK_LINGER) &&
431                     !(current->flags & PF_EXITING))
432                         timeout = sk->sk_lingertime;
433                 sock->sk = NULL;
434                 sk->sk_prot->close(sk, timeout);
435         }
436         return 0;
437 }
438
439 /* It is off by default, see below. */
440 int sysctl_ip_nonlocal_bind __read_mostly;
441
442 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
443 {
444         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
445         struct sock *sk = sock->sk;
446         struct inet_sock *inet = inet_sk(sk);
447         unsigned short snum;
448         int chk_addr_ret;
449         int err;
450
451         /* If the socket has its own bind function then use it. (RAW) */
452         if (sk->sk_prot->bind) {
453                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
454                 goto out;
455         }
456         err = -EINVAL;
457         if (addr_len < sizeof(struct sockaddr_in))
458                 goto out;
459
460         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
461
462         /* Not specified by any standard per-se, however it breaks too
463          * many applications when removed.  It is unfortunate since
464          * allowing applications to make a non-local bind solves
465          * several problems with systems using dynamic addressing.
466          * (ie. your servers still start up even if your ISDN link
467          *  is temporarily down)
468          */
469         err = -EADDRNOTAVAIL;
470         if (!sysctl_ip_nonlocal_bind &&
471             !(inet->freebind || inet->transparent) &&
472             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
473             chk_addr_ret != RTN_LOCAL &&
474             chk_addr_ret != RTN_MULTICAST &&
475             chk_addr_ret != RTN_BROADCAST)
476                 goto out;
477
478         snum = ntohs(addr->sin_port);
479         err = -EACCES;
480         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
481                 goto out;
482
483         /*      We keep a pair of addresses. rcv_saddr is the one
484          *      used by hash lookups, and saddr is used for transmit.
485          *
486          *      In the BSD API these are the same except where it
487          *      would be illegal to use them (multicast/broadcast) in
488          *      which case the sending device address is used.
489          */
490         lock_sock(sk);
491
492         /* Check these errors (active socket, double bind). */
493         err = -EINVAL;
494         if (sk->sk_state != TCP_CLOSE || inet->num)
495                 goto out_release_sock;
496
497         inet->rcv_saddr = inet->saddr = addr->sin_addr.s_addr;
498         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
499                 inet->saddr = 0;  /* Use device */
500
501         /* Make sure we are allowed to bind here. */
502         if (sk->sk_prot->get_port(sk, snum)) {
503                 inet->saddr = inet->rcv_saddr = 0;
504                 err = -EADDRINUSE;
505                 goto out_release_sock;
506         }
507
508         if (inet->rcv_saddr)
509                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
510         if (snum)
511                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
512         inet->sport = htons(inet->num);
513         inet->daddr = 0;
514         inet->dport = 0;
515         sk_dst_reset(sk);
516         err = 0;
517 out_release_sock:
518         release_sock(sk);
519 out:
520         return err;
521 }
522
523 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
524                        int addr_len, int flags)
525 {
526         struct sock *sk = sock->sk;
527
528         if (uaddr->sa_family == AF_UNSPEC)
529                 return sk->sk_prot->disconnect(sk, flags);
530
531         if (!inet_sk(sk)->num && inet_autobind(sk))
532                 return -EAGAIN;
533         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
534 }
535
536 static long inet_wait_for_connect(struct sock *sk, long timeo)
537 {
538         DEFINE_WAIT(wait);
539
540         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
541
542         /* Basic assumption: if someone sets sk->sk_err, he _must_
543          * change state of the socket from TCP_SYN_*.
544          * Connect() does not allow to get error notifications
545          * without closing the socket.
546          */
547         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
548                 release_sock(sk);
549                 timeo = schedule_timeout(timeo);
550                 lock_sock(sk);
551                 if (signal_pending(current) || !timeo)
552                         break;
553                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
554         }
555         finish_wait(sk->sk_sleep, &wait);
556         return timeo;
557 }
558
559 /*
560  *      Connect to a remote host. There is regrettably still a little
561  *      TCP 'magic' in here.
562  */
563 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
564                         int addr_len, int flags)
565 {
566         struct sock *sk = sock->sk;
567         int err;
568         long timeo;
569
570         lock_sock(sk);
571
572         if (uaddr->sa_family == AF_UNSPEC) {
573                 err = sk->sk_prot->disconnect(sk, flags);
574                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
575                 goto out;
576         }
577
578         switch (sock->state) {
579         default:
580                 err = -EINVAL;
581                 goto out;
582         case SS_CONNECTED:
583                 err = -EISCONN;
584                 goto out;
585         case SS_CONNECTING:
586                 err = -EALREADY;
587                 /* Fall out of switch with err, set for this state */
588                 break;
589         case SS_UNCONNECTED:
590                 err = -EISCONN;
591                 if (sk->sk_state != TCP_CLOSE)
592                         goto out;
593
594                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
595                 if (err < 0)
596                         goto out;
597
598                 sock->state = SS_CONNECTING;
599
600                 /* Just entered SS_CONNECTING state; the only
601                  * difference is that return value in non-blocking
602                  * case is EINPROGRESS, rather than EALREADY.
603                  */
604                 err = -EINPROGRESS;
605                 break;
606         }
607
608         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
609
610         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
611                 /* Error code is set above */
612                 if (!timeo || !inet_wait_for_connect(sk, timeo))
613                         goto out;
614
615                 err = sock_intr_errno(timeo);
616                 if (signal_pending(current))
617                         goto out;
618         }
619
620         /* Connection was closed by RST, timeout, ICMP error
621          * or another process disconnected us.
622          */
623         if (sk->sk_state == TCP_CLOSE)
624                 goto sock_error;
625
626         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
627          * and error was received after socket entered established state.
628          * Hence, it is handled normally after connect() return successfully.
629          */
630
631         sock->state = SS_CONNECTED;
632         err = 0;
633 out:
634         release_sock(sk);
635         return err;
636
637 sock_error:
638         err = sock_error(sk) ? : -ECONNABORTED;
639         sock->state = SS_UNCONNECTED;
640         if (sk->sk_prot->disconnect(sk, flags))
641                 sock->state = SS_DISCONNECTING;
642         goto out;
643 }
644
645 /*
646  *      Accept a pending connection. The TCP layer now gives BSD semantics.
647  */
648
649 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
650 {
651         struct sock *sk1 = sock->sk;
652         int err = -EINVAL;
653         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
654
655         if (!sk2)
656                 goto do_err;
657
658         lock_sock(sk2);
659
660         WARN_ON(!((1 << sk2->sk_state) &
661                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
662
663         sock_graft(sk2, newsock);
664
665         newsock->state = SS_CONNECTED;
666         err = 0;
667         release_sock(sk2);
668 do_err:
669         return err;
670 }
671
672
673 /*
674  *      This does both peername and sockname.
675  */
676 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
677                         int *uaddr_len, int peer)
678 {
679         struct sock *sk         = sock->sk;
680         struct inet_sock *inet  = inet_sk(sk);
681         struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
682
683         sin->sin_family = AF_INET;
684         if (peer) {
685                 if (!inet->dport ||
686                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
687                      peer == 1))
688                         return -ENOTCONN;
689                 sin->sin_port = inet->dport;
690                 sin->sin_addr.s_addr = inet->daddr;
691         } else {
692                 __be32 addr = inet->rcv_saddr;
693                 if (!addr)
694                         addr = inet->saddr;
695                 sin->sin_port = inet->sport;
696                 sin->sin_addr.s_addr = addr;
697         }
698         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
699         *uaddr_len = sizeof(*sin);
700         return 0;
701 }
702
703 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
704                  size_t size)
705 {
706         struct sock *sk = sock->sk;
707
708         /* We may need to bind the socket. */
709         if (!inet_sk(sk)->num && inet_autobind(sk))
710                 return -EAGAIN;
711
712         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
713 }
714
715
716 static ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
717 {
718         struct sock *sk = sock->sk;
719
720         /* We may need to bind the socket. */
721         if (!inet_sk(sk)->num && inet_autobind(sk))
722                 return -EAGAIN;
723
724         if (sk->sk_prot->sendpage)
725                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
726         return sock_no_sendpage(sock, page, offset, size, flags);
727 }
728
729
730 int inet_shutdown(struct socket *sock, int how)
731 {
732         struct sock *sk = sock->sk;
733         int err = 0;
734
735         /* This should really check to make sure
736          * the socket is a TCP socket. (WHY AC...)
737          */
738         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
739                        1->2 bit 2 snds.
740                        2->3 */
741         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
742                 return -EINVAL;
743
744         lock_sock(sk);
745         if (sock->state == SS_CONNECTING) {
746                 if ((1 << sk->sk_state) &
747                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
748                         sock->state = SS_DISCONNECTING;
749                 else
750                         sock->state = SS_CONNECTED;
751         }
752
753         switch (sk->sk_state) {
754         case TCP_CLOSE:
755                 err = -ENOTCONN;
756                 /* Hack to wake up other listeners, who can poll for
757                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
758         default:
759                 sk->sk_shutdown |= how;
760                 if (sk->sk_prot->shutdown)
761                         sk->sk_prot->shutdown(sk, how);
762                 break;
763
764         /* Remaining two branches are temporary solution for missing
765          * close() in multithreaded environment. It is _not_ a good idea,
766          * but we have no choice until close() is repaired at VFS level.
767          */
768         case TCP_LISTEN:
769                 if (!(how & RCV_SHUTDOWN))
770                         break;
771                 /* Fall through */
772         case TCP_SYN_SENT:
773                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
774                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
775                 break;
776         }
777
778         /* Wake up anyone sleeping in poll. */
779         sk->sk_state_change(sk);
780         release_sock(sk);
781         return err;
782 }
783
784 /*
785  *      ioctl() calls you can issue on an INET socket. Most of these are
786  *      device configuration and stuff and very rarely used. Some ioctls
787  *      pass on to the socket itself.
788  *
789  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
790  *      loads the devconfigure module does its configuring and unloads it.
791  *      There's a good 20K of config code hanging around the kernel.
792  */
793
794 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
795 {
796         struct sock *sk = sock->sk;
797         int err = 0;
798         struct net *net = sock_net(sk);
799
800         switch (cmd) {
801                 case SIOCGSTAMP:
802                         err = sock_get_timestamp(sk, (struct timeval __user *)arg);
803                         break;
804                 case SIOCGSTAMPNS:
805                         err = sock_get_timestampns(sk, (struct timespec __user *)arg);
806                         break;
807                 case SIOCADDRT:
808                 case SIOCDELRT:
809                 case SIOCRTMSG:
810                         err = ip_rt_ioctl(net, cmd, (void __user *)arg);
811                         break;
812                 case SIOCDARP:
813                 case SIOCGARP:
814                 case SIOCSARP:
815                         err = arp_ioctl(net, cmd, (void __user *)arg);
816                         break;
817                 case SIOCGIFADDR:
818                 case SIOCSIFADDR:
819                 case SIOCGIFBRDADDR:
820                 case SIOCSIFBRDADDR:
821                 case SIOCGIFNETMASK:
822                 case SIOCSIFNETMASK:
823                 case SIOCGIFDSTADDR:
824                 case SIOCSIFDSTADDR:
825                 case SIOCSIFPFLAGS:
826                 case SIOCGIFPFLAGS:
827                 case SIOCSIFFLAGS:
828                         err = devinet_ioctl(net, cmd, (void __user *)arg);
829                         break;
830                 default:
831                         if (sk->sk_prot->ioctl)
832                                 err = sk->sk_prot->ioctl(sk, cmd, arg);
833                         else
834                                 err = -ENOIOCTLCMD;
835                         break;
836         }
837         return err;
838 }
839
840 const struct proto_ops inet_stream_ops = {
841         .family            = PF_INET,
842         .owner             = THIS_MODULE,
843         .release           = inet_release,
844         .bind              = inet_bind,
845         .connect           = inet_stream_connect,
846         .socketpair        = sock_no_socketpair,
847         .accept            = inet_accept,
848         .getname           = inet_getname,
849         .poll              = tcp_poll,
850         .ioctl             = inet_ioctl,
851         .listen            = inet_listen,
852         .shutdown          = inet_shutdown,
853         .setsockopt        = sock_common_setsockopt,
854         .getsockopt        = sock_common_getsockopt,
855         .sendmsg           = tcp_sendmsg,
856         .recvmsg           = sock_common_recvmsg,
857         .mmap              = sock_no_mmap,
858         .sendpage          = tcp_sendpage,
859         .splice_read       = tcp_splice_read,
860 #ifdef CONFIG_COMPAT
861         .compat_setsockopt = compat_sock_common_setsockopt,
862         .compat_getsockopt = compat_sock_common_getsockopt,
863 #endif
864 };
865
866 const struct proto_ops inet_dgram_ops = {
867         .family            = PF_INET,
868         .owner             = THIS_MODULE,
869         .release           = inet_release,
870         .bind              = inet_bind,
871         .connect           = inet_dgram_connect,
872         .socketpair        = sock_no_socketpair,
873         .accept            = sock_no_accept,
874         .getname           = inet_getname,
875         .poll              = udp_poll,
876         .ioctl             = inet_ioctl,
877         .listen            = sock_no_listen,
878         .shutdown          = inet_shutdown,
879         .setsockopt        = sock_common_setsockopt,
880         .getsockopt        = sock_common_getsockopt,
881         .sendmsg           = inet_sendmsg,
882         .recvmsg           = sock_common_recvmsg,
883         .mmap              = sock_no_mmap,
884         .sendpage          = inet_sendpage,
885 #ifdef CONFIG_COMPAT
886         .compat_setsockopt = compat_sock_common_setsockopt,
887         .compat_getsockopt = compat_sock_common_getsockopt,
888 #endif
889 };
890
891 /*
892  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
893  * udp_poll
894  */
895 static const struct proto_ops inet_sockraw_ops = {
896         .family            = PF_INET,
897         .owner             = THIS_MODULE,
898         .release           = inet_release,
899         .bind              = inet_bind,
900         .connect           = inet_dgram_connect,
901         .socketpair        = sock_no_socketpair,
902         .accept            = sock_no_accept,
903         .getname           = inet_getname,
904         .poll              = datagram_poll,
905         .ioctl             = inet_ioctl,
906         .listen            = sock_no_listen,
907         .shutdown          = inet_shutdown,
908         .setsockopt        = sock_common_setsockopt,
909         .getsockopt        = sock_common_getsockopt,
910         .sendmsg           = inet_sendmsg,
911         .recvmsg           = sock_common_recvmsg,
912         .mmap              = sock_no_mmap,
913         .sendpage          = inet_sendpage,
914 #ifdef CONFIG_COMPAT
915         .compat_setsockopt = compat_sock_common_setsockopt,
916         .compat_getsockopt = compat_sock_common_getsockopt,
917 #endif
918 };
919
920 static struct net_proto_family inet_family_ops = {
921         .family = PF_INET,
922         .create = inet_create,
923         .owner  = THIS_MODULE,
924 };
925
926 /* Upon startup we insert all the elements in inetsw_array[] into
927  * the linked list inetsw.
928  */
929 static struct inet_protosw inetsw_array[] =
930 {
931         {
932                 .type =       SOCK_STREAM,
933                 .protocol =   IPPROTO_TCP,
934                 .prot =       &tcp_prot,
935                 .ops =        &inet_stream_ops,
936                 .capability = -1,
937                 .no_check =   0,
938                 .flags =      INET_PROTOSW_PERMANENT |
939                               INET_PROTOSW_ICSK,
940         },
941
942         {
943                 .type =       SOCK_DGRAM,
944                 .protocol =   IPPROTO_UDP,
945                 .prot =       &udp_prot,
946                 .ops =        &inet_dgram_ops,
947                 .capability = -1,
948                 .no_check =   UDP_CSUM_DEFAULT,
949                 .flags =      INET_PROTOSW_PERMANENT,
950        },
951
952
953        {
954                .type =       SOCK_RAW,
955                .protocol =   IPPROTO_IP,        /* wild card */
956                .prot =       &raw_prot,
957                .ops =        &inet_sockraw_ops,
958                .capability = CAP_NET_RAW,
959                .no_check =   UDP_CSUM_DEFAULT,
960                .flags =      INET_PROTOSW_REUSE,
961        }
962 };
963
964 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
965
966 void inet_register_protosw(struct inet_protosw *p)
967 {
968         struct list_head *lh;
969         struct inet_protosw *answer;
970         int protocol = p->protocol;
971         struct list_head *last_perm;
972
973         spin_lock_bh(&inetsw_lock);
974
975         if (p->type >= SOCK_MAX)
976                 goto out_illegal;
977
978         /* If we are trying to override a permanent protocol, bail. */
979         answer = NULL;
980         last_perm = &inetsw[p->type];
981         list_for_each(lh, &inetsw[p->type]) {
982                 answer = list_entry(lh, struct inet_protosw, list);
983
984                 /* Check only the non-wild match. */
985                 if (INET_PROTOSW_PERMANENT & answer->flags) {
986                         if (protocol == answer->protocol)
987                                 break;
988                         last_perm = lh;
989                 }
990
991                 answer = NULL;
992         }
993         if (answer)
994                 goto out_permanent;
995
996         /* Add the new entry after the last permanent entry if any, so that
997          * the new entry does not override a permanent entry when matched with
998          * a wild-card protocol. But it is allowed to override any existing
999          * non-permanent entry.  This means that when we remove this entry, the
1000          * system automatically returns to the old behavior.
1001          */
1002         list_add_rcu(&p->list, last_perm);
1003 out:
1004         spin_unlock_bh(&inetsw_lock);
1005
1006         synchronize_net();
1007
1008         return;
1009
1010 out_permanent:
1011         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1012                protocol);
1013         goto out;
1014
1015 out_illegal:
1016         printk(KERN_ERR
1017                "Ignoring attempt to register invalid socket type %d.\n",
1018                p->type);
1019         goto out;
1020 }
1021
1022 void inet_unregister_protosw(struct inet_protosw *p)
1023 {
1024         if (INET_PROTOSW_PERMANENT & p->flags) {
1025                 printk(KERN_ERR
1026                        "Attempt to unregister permanent protocol %d.\n",
1027                        p->protocol);
1028         } else {
1029                 spin_lock_bh(&inetsw_lock);
1030                 list_del_rcu(&p->list);
1031                 spin_unlock_bh(&inetsw_lock);
1032
1033                 synchronize_net();
1034         }
1035 }
1036
1037 /*
1038  *      Shall we try to damage output packets if routing dev changes?
1039  */
1040
1041 int sysctl_ip_dynaddr __read_mostly;
1042
1043 static int inet_sk_reselect_saddr(struct sock *sk)
1044 {
1045         struct inet_sock *inet = inet_sk(sk);
1046         int err;
1047         struct rtable *rt;
1048         __be32 old_saddr = inet->saddr;
1049         __be32 new_saddr;
1050         __be32 daddr = inet->daddr;
1051
1052         if (inet->opt && inet->opt->srr)
1053                 daddr = inet->opt->faddr;
1054
1055         /* Query new route. */
1056         err = ip_route_connect(&rt, daddr, 0,
1057                                RT_CONN_FLAGS(sk),
1058                                sk->sk_bound_dev_if,
1059                                sk->sk_protocol,
1060                                inet->sport, inet->dport, sk, 0);
1061         if (err)
1062                 return err;
1063
1064         sk_setup_caps(sk, &rt->u.dst);
1065
1066         new_saddr = rt->rt_src;
1067
1068         if (new_saddr == old_saddr)
1069                 return 0;
1070
1071         if (sysctl_ip_dynaddr > 1) {
1072                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1073                        __func__, &old_saddr, &new_saddr);
1074         }
1075
1076         inet->saddr = inet->rcv_saddr = new_saddr;
1077
1078         /*
1079          * XXX The only one ugly spot where we need to
1080          * XXX really change the sockets identity after
1081          * XXX it has entered the hashes. -DaveM
1082          *
1083          * Besides that, it does not check for connection
1084          * uniqueness. Wait for troubles.
1085          */
1086         __sk_prot_rehash(sk);
1087         return 0;
1088 }
1089
1090 int inet_sk_rebuild_header(struct sock *sk)
1091 {
1092         struct inet_sock *inet = inet_sk(sk);
1093         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1094         __be32 daddr;
1095         int err;
1096
1097         /* Route is OK, nothing to do. */
1098         if (rt)
1099                 return 0;
1100
1101         /* Reroute. */
1102         daddr = inet->daddr;
1103         if (inet->opt && inet->opt->srr)
1104                 daddr = inet->opt->faddr;
1105 {
1106         struct flowi fl = {
1107                 .oif = sk->sk_bound_dev_if,
1108                 .nl_u = {
1109                         .ip4_u = {
1110                                 .daddr  = daddr,
1111                                 .saddr  = inet->saddr,
1112                                 .tos    = RT_CONN_FLAGS(sk),
1113                         },
1114                 },
1115                 .proto = sk->sk_protocol,
1116                 .flags = inet_sk_flowi_flags(sk),
1117                 .uli_u = {
1118                         .ports = {
1119                                 .sport = inet->sport,
1120                                 .dport = inet->dport,
1121                         },
1122                 },
1123         };
1124
1125         security_sk_classify_flow(sk, &fl);
1126         err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 0);
1127 }
1128         if (!err)
1129                 sk_setup_caps(sk, &rt->u.dst);
1130         else {
1131                 /* Routing failed... */
1132                 sk->sk_route_caps = 0;
1133                 /*
1134                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1135                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1136                  */
1137                 if (!sysctl_ip_dynaddr ||
1138                     sk->sk_state != TCP_SYN_SENT ||
1139                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1140                     (err = inet_sk_reselect_saddr(sk)) != 0)
1141                         sk->sk_err_soft = -err;
1142         }
1143
1144         return err;
1145 }
1146
1147 EXPORT_SYMBOL(inet_sk_rebuild_header);
1148
1149 static int inet_gso_send_check(struct sk_buff *skb)
1150 {
1151         struct iphdr *iph;
1152         struct net_protocol *ops;
1153         int proto;
1154         int ihl;
1155         int err = -EINVAL;
1156
1157         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1158                 goto out;
1159
1160         iph = ip_hdr(skb);
1161         ihl = iph->ihl * 4;
1162         if (ihl < sizeof(*iph))
1163                 goto out;
1164
1165         if (unlikely(!pskb_may_pull(skb, ihl)))
1166                 goto out;
1167
1168         __skb_pull(skb, ihl);
1169         skb_reset_transport_header(skb);
1170         iph = ip_hdr(skb);
1171         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1172         err = -EPROTONOSUPPORT;
1173
1174         rcu_read_lock();
1175         ops = rcu_dereference(inet_protos[proto]);
1176         if (likely(ops && ops->gso_send_check))
1177                 err = ops->gso_send_check(skb);
1178         rcu_read_unlock();
1179
1180 out:
1181         return err;
1182 }
1183
1184 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, int features)
1185 {
1186         struct sk_buff *segs = ERR_PTR(-EINVAL);
1187         struct iphdr *iph;
1188         struct net_protocol *ops;
1189         int proto;
1190         int ihl;
1191         int id;
1192
1193         if (!(features & NETIF_F_V4_CSUM))
1194                 features &= ~NETIF_F_SG;
1195
1196         if (unlikely(skb_shinfo(skb)->gso_type &
1197                      ~(SKB_GSO_TCPV4 |
1198                        SKB_GSO_UDP |
1199                        SKB_GSO_DODGY |
1200                        SKB_GSO_TCP_ECN |
1201                        0)))
1202                 goto out;
1203
1204         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1205                 goto out;
1206
1207         iph = ip_hdr(skb);
1208         ihl = iph->ihl * 4;
1209         if (ihl < sizeof(*iph))
1210                 goto out;
1211
1212         if (unlikely(!pskb_may_pull(skb, ihl)))
1213                 goto out;
1214
1215         __skb_pull(skb, ihl);
1216         skb_reset_transport_header(skb);
1217         iph = ip_hdr(skb);
1218         id = ntohs(iph->id);
1219         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1220         segs = ERR_PTR(-EPROTONOSUPPORT);
1221
1222         rcu_read_lock();
1223         ops = rcu_dereference(inet_protos[proto]);
1224         if (likely(ops && ops->gso_segment))
1225                 segs = ops->gso_segment(skb, features);
1226         rcu_read_unlock();
1227
1228         if (!segs || IS_ERR(segs))
1229                 goto out;
1230
1231         skb = segs;
1232         do {
1233                 iph = ip_hdr(skb);
1234                 iph->id = htons(id++);
1235                 iph->tot_len = htons(skb->len - skb->mac_len);
1236                 iph->check = 0;
1237                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1238         } while ((skb = skb->next));
1239
1240 out:
1241         return segs;
1242 }
1243
1244 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1245                                          struct sk_buff *skb)
1246 {
1247         struct net_protocol *ops;
1248         struct sk_buff **pp = NULL;
1249         struct sk_buff *p;
1250         struct iphdr *iph;
1251         int flush = 1;
1252         int proto;
1253         int id;
1254
1255         iph = skb_gro_header(skb, sizeof(*iph));
1256         if (unlikely(!iph))
1257                 goto out;
1258
1259         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1260
1261         rcu_read_lock();
1262         ops = rcu_dereference(inet_protos[proto]);
1263         if (!ops || !ops->gro_receive)
1264                 goto out_unlock;
1265
1266         if (*(u8 *)iph != 0x45)
1267                 goto out_unlock;
1268
1269         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1270                 goto out_unlock;
1271
1272         flush = ntohs(iph->tot_len) != skb_gro_len(skb) ||
1273                 iph->frag_off != htons(IP_DF);
1274         id = ntohs(iph->id);
1275
1276         for (p = *head; p; p = p->next) {
1277                 struct iphdr *iph2;
1278
1279                 if (!NAPI_GRO_CB(p)->same_flow)
1280                         continue;
1281
1282                 iph2 = ip_hdr(p);
1283
1284                 if ((iph->protocol ^ iph2->protocol) |
1285                     (iph->tos ^ iph2->tos) |
1286                     (iph->saddr ^ iph2->saddr) |
1287                     (iph->daddr ^ iph2->daddr)) {
1288                         NAPI_GRO_CB(p)->same_flow = 0;
1289                         continue;
1290                 }
1291
1292                 /* All fields must match except length and checksum. */
1293                 NAPI_GRO_CB(p)->flush |=
1294                         (iph->ttl ^ iph2->ttl) |
1295                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1296
1297                 NAPI_GRO_CB(p)->flush |= flush;
1298         }
1299
1300         NAPI_GRO_CB(skb)->flush |= flush;
1301         skb_gro_pull(skb, sizeof(*iph));
1302         skb_set_transport_header(skb, skb_gro_offset(skb));
1303
1304         pp = ops->gro_receive(head, skb);
1305
1306 out_unlock:
1307         rcu_read_unlock();
1308
1309 out:
1310         NAPI_GRO_CB(skb)->flush |= flush;
1311
1312         return pp;
1313 }
1314
1315 static int inet_gro_complete(struct sk_buff *skb)
1316 {
1317         struct net_protocol *ops;
1318         struct iphdr *iph = ip_hdr(skb);
1319         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1320         int err = -ENOSYS;
1321         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1322
1323         csum_replace2(&iph->check, iph->tot_len, newlen);
1324         iph->tot_len = newlen;
1325
1326         rcu_read_lock();
1327         ops = rcu_dereference(inet_protos[proto]);
1328         if (WARN_ON(!ops || !ops->gro_complete))
1329                 goto out_unlock;
1330
1331         err = ops->gro_complete(skb);
1332
1333 out_unlock:
1334         rcu_read_unlock();
1335
1336         return err;
1337 }
1338
1339 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1340                          unsigned short type, unsigned char protocol,
1341                          struct net *net)
1342 {
1343         struct socket *sock;
1344         int rc = sock_create_kern(family, type, protocol, &sock);
1345
1346         if (rc == 0) {
1347                 *sk = sock->sk;
1348                 (*sk)->sk_allocation = GFP_ATOMIC;
1349                 /*
1350                  * Unhash it so that IP input processing does not even see it,
1351                  * we do not wish this socket to see incoming packets.
1352                  */
1353                 (*sk)->sk_prot->unhash(*sk);
1354
1355                 sk_change_net(*sk, net);
1356         }
1357         return rc;
1358 }
1359
1360 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1361
1362 unsigned long snmp_fold_field(void *mib[], int offt)
1363 {
1364         unsigned long res = 0;
1365         int i;
1366
1367         for_each_possible_cpu(i) {
1368                 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1369                 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1370         }
1371         return res;
1372 }
1373 EXPORT_SYMBOL_GPL(snmp_fold_field);
1374
1375 int snmp_mib_init(void *ptr[2], size_t mibsize)
1376 {
1377         BUG_ON(ptr == NULL);
1378         ptr[0] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1379         if (!ptr[0])
1380                 goto err0;
1381         ptr[1] = __alloc_percpu(mibsize, __alignof__(unsigned long long));
1382         if (!ptr[1])
1383                 goto err1;
1384         return 0;
1385 err1:
1386         free_percpu(ptr[0]);
1387         ptr[0] = NULL;
1388 err0:
1389         return -ENOMEM;
1390 }
1391 EXPORT_SYMBOL_GPL(snmp_mib_init);
1392
1393 void snmp_mib_free(void *ptr[2])
1394 {
1395         BUG_ON(ptr == NULL);
1396         free_percpu(ptr[0]);
1397         free_percpu(ptr[1]);
1398         ptr[0] = ptr[1] = NULL;
1399 }
1400 EXPORT_SYMBOL_GPL(snmp_mib_free);
1401
1402 #ifdef CONFIG_IP_MULTICAST
1403 static struct net_protocol igmp_protocol = {
1404         .handler =      igmp_rcv,
1405         .netns_ok =     1,
1406 };
1407 #endif
1408
1409 static struct net_protocol tcp_protocol = {
1410         .handler =      tcp_v4_rcv,
1411         .err_handler =  tcp_v4_err,
1412         .gso_send_check = tcp_v4_gso_send_check,
1413         .gso_segment =  tcp_tso_segment,
1414         .gro_receive =  tcp4_gro_receive,
1415         .gro_complete = tcp4_gro_complete,
1416         .no_policy =    1,
1417         .netns_ok =     1,
1418 };
1419
1420 static struct net_protocol udp_protocol = {
1421         .handler =      udp_rcv,
1422         .err_handler =  udp_err,
1423         .no_policy =    1,
1424         .netns_ok =     1,
1425 };
1426
1427 static struct net_protocol icmp_protocol = {
1428         .handler =      icmp_rcv,
1429         .no_policy =    1,
1430         .netns_ok =     1,
1431 };
1432
1433 static __net_init int ipv4_mib_init_net(struct net *net)
1434 {
1435         if (snmp_mib_init((void **)net->mib.tcp_statistics,
1436                           sizeof(struct tcp_mib)) < 0)
1437                 goto err_tcp_mib;
1438         if (snmp_mib_init((void **)net->mib.ip_statistics,
1439                           sizeof(struct ipstats_mib)) < 0)
1440                 goto err_ip_mib;
1441         if (snmp_mib_init((void **)net->mib.net_statistics,
1442                           sizeof(struct linux_mib)) < 0)
1443                 goto err_net_mib;
1444         if (snmp_mib_init((void **)net->mib.udp_statistics,
1445                           sizeof(struct udp_mib)) < 0)
1446                 goto err_udp_mib;
1447         if (snmp_mib_init((void **)net->mib.udplite_statistics,
1448                           sizeof(struct udp_mib)) < 0)
1449                 goto err_udplite_mib;
1450         if (snmp_mib_init((void **)net->mib.icmp_statistics,
1451                           sizeof(struct icmp_mib)) < 0)
1452                 goto err_icmp_mib;
1453         if (snmp_mib_init((void **)net->mib.icmpmsg_statistics,
1454                           sizeof(struct icmpmsg_mib)) < 0)
1455                 goto err_icmpmsg_mib;
1456
1457         tcp_mib_init(net);
1458         return 0;
1459
1460 err_icmpmsg_mib:
1461         snmp_mib_free((void **)net->mib.icmp_statistics);
1462 err_icmp_mib:
1463         snmp_mib_free((void **)net->mib.udplite_statistics);
1464 err_udplite_mib:
1465         snmp_mib_free((void **)net->mib.udp_statistics);
1466 err_udp_mib:
1467         snmp_mib_free((void **)net->mib.net_statistics);
1468 err_net_mib:
1469         snmp_mib_free((void **)net->mib.ip_statistics);
1470 err_ip_mib:
1471         snmp_mib_free((void **)net->mib.tcp_statistics);
1472 err_tcp_mib:
1473         return -ENOMEM;
1474 }
1475
1476 static __net_exit void ipv4_mib_exit_net(struct net *net)
1477 {
1478         snmp_mib_free((void **)net->mib.icmpmsg_statistics);
1479         snmp_mib_free((void **)net->mib.icmp_statistics);
1480         snmp_mib_free((void **)net->mib.udplite_statistics);
1481         snmp_mib_free((void **)net->mib.udp_statistics);
1482         snmp_mib_free((void **)net->mib.net_statistics);
1483         snmp_mib_free((void **)net->mib.ip_statistics);
1484         snmp_mib_free((void **)net->mib.tcp_statistics);
1485 }
1486
1487 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1488         .init = ipv4_mib_init_net,
1489         .exit = ipv4_mib_exit_net,
1490 };
1491
1492 static int __init init_ipv4_mibs(void)
1493 {
1494         return register_pernet_subsys(&ipv4_mib_ops);
1495 }
1496
1497 static int ipv4_proc_init(void);
1498
1499 /*
1500  *      IP protocol layer initialiser
1501  */
1502
1503 static struct packet_type ip_packet_type __read_mostly = {
1504         .type = cpu_to_be16(ETH_P_IP),
1505         .func = ip_rcv,
1506         .gso_send_check = inet_gso_send_check,
1507         .gso_segment = inet_gso_segment,
1508         .gro_receive = inet_gro_receive,
1509         .gro_complete = inet_gro_complete,
1510 };
1511
1512 static int __init inet_init(void)
1513 {
1514         struct sk_buff *dummy_skb;
1515         struct inet_protosw *q;
1516         struct list_head *r;
1517         int rc = -EINVAL;
1518
1519         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1520
1521         rc = proto_register(&tcp_prot, 1);
1522         if (rc)
1523                 goto out;
1524
1525         rc = proto_register(&udp_prot, 1);
1526         if (rc)
1527                 goto out_unregister_tcp_proto;
1528
1529         rc = proto_register(&raw_prot, 1);
1530         if (rc)
1531                 goto out_unregister_udp_proto;
1532
1533         /*
1534          *      Tell SOCKET that we are alive...
1535          */
1536
1537         (void)sock_register(&inet_family_ops);
1538
1539 #ifdef CONFIG_SYSCTL
1540         ip_static_sysctl_init();
1541 #endif
1542
1543         /*
1544          *      Add all the base protocols.
1545          */
1546
1547         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1548                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1549         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1550                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1551         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1552                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1553 #ifdef CONFIG_IP_MULTICAST
1554         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1555                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1556 #endif
1557
1558         /* Register the socket-side information for inet_create. */
1559         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1560                 INIT_LIST_HEAD(r);
1561
1562         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1563                 inet_register_protosw(q);
1564
1565         /*
1566          *      Set the ARP module up
1567          */
1568
1569         arp_init();
1570
1571         /*
1572          *      Set the IP module up
1573          */
1574
1575         ip_init();
1576
1577         tcp_v4_init();
1578
1579         /* Setup TCP slab cache for open requests. */
1580         tcp_init();
1581
1582         /* Setup UDP memory threshold */
1583         udp_init();
1584
1585         /* Add UDP-Lite (RFC 3828) */
1586         udplite4_register();
1587
1588         /*
1589          *      Set the ICMP layer up
1590          */
1591
1592         if (icmp_init() < 0)
1593                 panic("Failed to create the ICMP control socket.\n");
1594
1595         /*
1596          *      Initialise the multicast router
1597          */
1598 #if defined(CONFIG_IP_MROUTE)
1599         if (ip_mr_init())
1600                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1601 #endif
1602         /*
1603          *      Initialise per-cpu ipv4 mibs
1604          */
1605
1606         if (init_ipv4_mibs())
1607                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1608
1609         ipv4_proc_init();
1610
1611         ipfrag_init();
1612
1613         dev_add_pack(&ip_packet_type);
1614
1615         rc = 0;
1616 out:
1617         return rc;
1618 out_unregister_udp_proto:
1619         proto_unregister(&udp_prot);
1620 out_unregister_tcp_proto:
1621         proto_unregister(&tcp_prot);
1622         goto out;
1623 }
1624
1625 fs_initcall(inet_init);
1626
1627 /* ------------------------------------------------------------------------ */
1628
1629 #ifdef CONFIG_PROC_FS
1630 static int __init ipv4_proc_init(void)
1631 {
1632         int rc = 0;
1633
1634         if (raw_proc_init())
1635                 goto out_raw;
1636         if (tcp4_proc_init())
1637                 goto out_tcp;
1638         if (udp4_proc_init())
1639                 goto out_udp;
1640         if (ip_misc_proc_init())
1641                 goto out_misc;
1642 out:
1643         return rc;
1644 out_misc:
1645         udp4_proc_exit();
1646 out_udp:
1647         tcp4_proc_exit();
1648 out_tcp:
1649         raw_proc_exit();
1650 out_raw:
1651         rc = -ENOMEM;
1652         goto out;
1653 }
1654
1655 #else /* CONFIG_PROC_FS */
1656 static int __init ipv4_proc_init(void)
1657 {
1658         return 0;
1659 }
1660 #endif /* CONFIG_PROC_FS */
1661
1662 MODULE_ALIAS_NETPROTO(PF_INET);
1663
1664 EXPORT_SYMBOL(inet_accept);
1665 EXPORT_SYMBOL(inet_bind);
1666 EXPORT_SYMBOL(inet_dgram_connect);
1667 EXPORT_SYMBOL(inet_dgram_ops);
1668 EXPORT_SYMBOL(inet_getname);
1669 EXPORT_SYMBOL(inet_ioctl);
1670 EXPORT_SYMBOL(inet_listen);
1671 EXPORT_SYMBOL(inet_register_protosw);
1672 EXPORT_SYMBOL(inet_release);
1673 EXPORT_SYMBOL(inet_sendmsg);
1674 EXPORT_SYMBOL(inet_shutdown);
1675 EXPORT_SYMBOL(inet_sock_destruct);
1676 EXPORT_SYMBOL(inet_stream_connect);
1677 EXPORT_SYMBOL(inet_stream_ops);
1678 EXPORT_SYMBOL(inet_unregister_protosw);
1679 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);