1b745d412cf6eb8c0b2d424ab5e5c8cbbc439248
[linux-2.6.git] / 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 #include <linux/slab.h>
90
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
93
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/inetdevice.h>
97 #include <linux/netdevice.h>
98 #include <net/checksum.h>
99 #include <net/ip.h>
100 #include <net/protocol.h>
101 #include <net/arp.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
105 #include <net/tcp.h>
106 #include <net/udp.h>
107 #include <net/udplite.h>
108 #include <net/ping.h>
109 #include <linux/skbuff.h>
110 #include <net/sock.h>
111 #include <net/raw.h>
112 #include <net/icmp.h>
113 #include <net/ipip.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #ifdef CONFIG_IP_MROUTE
118 #include <linux/mroute.h>
119 #endif
120
121
122 /* The inetsw table contains everything that inet_create needs to
123  * build a new socket.
124  */
125 static struct list_head inetsw[SOCK_MAX];
126 static DEFINE_SPINLOCK(inetsw_lock);
127
128 struct ipv4_config ipv4_config;
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                 pr_err("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                 pr_err("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(rcu_dereference_protected(inet->inet_opt, 1));
158         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159         sk_refcnt_debug_dec(sk);
160 }
161 EXPORT_SYMBOL(inet_sock_destruct);
162
163 /*
164  *      The routines beyond this point handle the behaviour of an AF_INET
165  *      socket object. Mostly it punts to the subprotocols of IP to do
166  *      the work.
167  */
168
169 /*
170  *      Automatically bind an unbound socket.
171  */
172
173 static int inet_autobind(struct sock *sk)
174 {
175         struct inet_sock *inet;
176         /* We may need to bind the socket. */
177         lock_sock(sk);
178         inet = inet_sk(sk);
179         if (!inet->inet_num) {
180                 if (sk->sk_prot->get_port(sk, 0)) {
181                         release_sock(sk);
182                         return -EAGAIN;
183                 }
184                 inet->inet_sport = htons(inet->inet_num);
185         }
186         release_sock(sk);
187         return 0;
188 }
189
190 /*
191  *      Move a socket into listening state.
192  */
193 int inet_listen(struct socket *sock, int backlog)
194 {
195         struct sock *sk = sock->sk;
196         unsigned char old_state;
197         int err;
198
199         lock_sock(sk);
200
201         err = -EINVAL;
202         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
203                 goto out;
204
205         old_state = sk->sk_state;
206         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
207                 goto out;
208
209         /* Really, if the socket is already in listen state
210          * we can only allow the backlog to be adjusted.
211          */
212         if (old_state != TCP_LISTEN) {
213                 err = inet_csk_listen_start(sk, backlog);
214                 if (err)
215                         goto out;
216         }
217         sk->sk_max_ack_backlog = backlog;
218         err = 0;
219
220 out:
221         release_sock(sk);
222         return err;
223 }
224 EXPORT_SYMBOL(inet_listen);
225
226 u32 inet_ehash_secret __read_mostly;
227 EXPORT_SYMBOL(inet_ehash_secret);
228
229 /*
230  * inet_ehash_secret must be set exactly once
231  */
232 void build_ehash_secret(void)
233 {
234         u32 rnd;
235
236         do {
237                 get_random_bytes(&rnd, sizeof(rnd));
238         } while (rnd == 0);
239
240         cmpxchg(&inet_ehash_secret, 0, rnd);
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         const 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                        int kern)
267 {
268         struct sock *sk;
269         struct inet_protosw *answer;
270         struct inet_sock *inet;
271         struct proto *answer_prot;
272         unsigned char answer_flags;
273         char answer_no_check;
274         int try_loading_module = 0;
275         int err;
276
277         if (unlikely(!inet_ehash_secret))
278                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
279                         build_ehash_secret();
280
281         sock->state = SS_UNCONNECTED;
282
283         /* Look for the requested type/protocol pair. */
284 lookup_protocol:
285         err = -ESOCKTNOSUPPORT;
286         rcu_read_lock();
287         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
288
289                 err = 0;
290                 /* Check the non-wild match. */
291                 if (protocol == answer->protocol) {
292                         if (protocol != IPPROTO_IP)
293                                 break;
294                 } else {
295                         /* Check for the two wild cases. */
296                         if (IPPROTO_IP == protocol) {
297                                 protocol = answer->protocol;
298                                 break;
299                         }
300                         if (IPPROTO_IP == answer->protocol)
301                                 break;
302                 }
303                 err = -EPROTONOSUPPORT;
304         }
305
306         if (unlikely(err)) {
307                 if (try_loading_module < 2) {
308                         rcu_read_unlock();
309                         /*
310                          * Be more specific, e.g. net-pf-2-proto-132-type-1
311                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
312                          */
313                         if (++try_loading_module == 1)
314                                 request_module("net-pf-%d-proto-%d-type-%d",
315                                                PF_INET, protocol, sock->type);
316                         /*
317                          * Fall back to generic, e.g. net-pf-2-proto-132
318                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
319                          */
320                         else
321                                 request_module("net-pf-%d-proto-%d",
322                                                PF_INET, protocol);
323                         goto lookup_protocol;
324                 } else
325                         goto out_rcu_unlock;
326         }
327
328         err = -EPERM;
329         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
330                 goto out_rcu_unlock;
331
332         err = -EAFNOSUPPORT;
333         if (!inet_netns_ok(net, protocol))
334                 goto out_rcu_unlock;
335
336         sock->ops = answer->ops;
337         answer_prot = answer->prot;
338         answer_no_check = answer->no_check;
339         answer_flags = answer->flags;
340         rcu_read_unlock();
341
342         WARN_ON(answer_prot->slab == NULL);
343
344         err = -ENOBUFS;
345         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
346         if (sk == NULL)
347                 goto out;
348
349         err = 0;
350         sk->sk_no_check = answer_no_check;
351         if (INET_PROTOSW_REUSE & answer_flags)
352                 sk->sk_reuse = 1;
353
354         inet = inet_sk(sk);
355         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
356
357         inet->nodefrag = 0;
358
359         if (SOCK_RAW == sock->type) {
360                 inet->inet_num = protocol;
361                 if (IPPROTO_RAW == protocol)
362                         inet->hdrincl = 1;
363         }
364
365         if (ipv4_config.no_pmtu_disc)
366                 inet->pmtudisc = IP_PMTUDISC_DONT;
367         else
368                 inet->pmtudisc = IP_PMTUDISC_WANT;
369
370         inet->inet_id = 0;
371
372         sock_init_data(sock, sk);
373
374         sk->sk_destruct    = inet_sock_destruct;
375         sk->sk_protocol    = protocol;
376         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
377
378         inet->uc_ttl    = -1;
379         inet->mc_loop   = 1;
380         inet->mc_ttl    = 1;
381         inet->mc_all    = 1;
382         inet->mc_index  = 0;
383         inet->mc_list   = NULL;
384
385         sk_refcnt_debug_inc(sk);
386
387         if (inet->inet_num) {
388                 /* It assumes that any protocol which allows
389                  * the user to assign a number at socket
390                  * creation time automatically
391                  * shares.
392                  */
393                 inet->inet_sport = htons(inet->inet_num);
394                 /* Add to protocol hash chains. */
395                 sk->sk_prot->hash(sk);
396         }
397
398         if (sk->sk_prot->init) {
399                 err = sk->sk_prot->init(sk);
400                 if (err)
401                         sk_common_release(sk);
402         }
403 out:
404         return err;
405 out_rcu_unlock:
406         rcu_read_unlock();
407         goto out;
408 }
409
410
411 /*
412  *      The peer socket should always be NULL (or else). When we call this
413  *      function we are destroying the object and from then on nobody
414  *      should refer to it.
415  */
416 int inet_release(struct socket *sock)
417 {
418         struct sock *sk = sock->sk;
419
420         if (sk) {
421                 long timeout;
422
423                 sock_rps_reset_flow(sk);
424
425                 /* Applications forget to leave groups before exiting */
426                 ip_mc_drop_socket(sk);
427
428                 /* If linger is set, we don't return until the close
429                  * is complete.  Otherwise we return immediately. The
430                  * actually closing is done the same either way.
431                  *
432                  * If the close is due to the process exiting, we never
433                  * linger..
434                  */
435                 timeout = 0;
436                 if (sock_flag(sk, SOCK_LINGER) &&
437                     !(current->flags & PF_EXITING))
438                         timeout = sk->sk_lingertime;
439                 sock->sk = NULL;
440                 sk->sk_prot->close(sk, timeout);
441         }
442         return 0;
443 }
444 EXPORT_SYMBOL(inet_release);
445
446 /* It is off by default, see below. */
447 int sysctl_ip_nonlocal_bind __read_mostly;
448 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
449
450 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
451 {
452         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
453         struct sock *sk = sock->sk;
454         struct inet_sock *inet = inet_sk(sk);
455         unsigned short snum;
456         int chk_addr_ret;
457         int err;
458
459         /* If the socket has its own bind function then use it. (RAW) */
460         if (sk->sk_prot->bind) {
461                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
462                 goto out;
463         }
464         err = -EINVAL;
465         if (addr_len < sizeof(struct sockaddr_in))
466                 goto out;
467
468         if (addr->sin_family != AF_INET) {
469                 err = -EAFNOSUPPORT;
470                 goto out;
471         }
472
473         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
474
475         /* Not specified by any standard per-se, however it breaks too
476          * many applications when removed.  It is unfortunate since
477          * allowing applications to make a non-local bind solves
478          * several problems with systems using dynamic addressing.
479          * (ie. your servers still start up even if your ISDN link
480          *  is temporarily down)
481          */
482         err = -EADDRNOTAVAIL;
483         if (!sysctl_ip_nonlocal_bind &&
484             !(inet->freebind || inet->transparent) &&
485             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
486             chk_addr_ret != RTN_LOCAL &&
487             chk_addr_ret != RTN_MULTICAST &&
488             chk_addr_ret != RTN_BROADCAST)
489                 goto out;
490
491         snum = ntohs(addr->sin_port);
492         err = -EACCES;
493         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
494                 goto out;
495
496         /*      We keep a pair of addresses. rcv_saddr is the one
497          *      used by hash lookups, and saddr is used for transmit.
498          *
499          *      In the BSD API these are the same except where it
500          *      would be illegal to use them (multicast/broadcast) in
501          *      which case the sending device address is used.
502          */
503         lock_sock(sk);
504
505         /* Check these errors (active socket, double bind). */
506         err = -EINVAL;
507         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
508                 goto out_release_sock;
509
510         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
511         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
512                 inet->inet_saddr = 0;  /* Use device */
513
514         /* Make sure we are allowed to bind here. */
515         if (sk->sk_prot->get_port(sk, snum)) {
516                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
517                 err = -EADDRINUSE;
518                 goto out_release_sock;
519         }
520
521         if (inet->inet_rcv_saddr)
522                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
523         if (snum)
524                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
525         inet->inet_sport = htons(inet->inet_num);
526         inet->inet_daddr = 0;
527         inet->inet_dport = 0;
528         sk_dst_reset(sk);
529         err = 0;
530 out_release_sock:
531         release_sock(sk);
532 out:
533         return err;
534 }
535 EXPORT_SYMBOL(inet_bind);
536
537 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
538                        int addr_len, int flags)
539 {
540         struct sock *sk = sock->sk;
541
542         if (addr_len < sizeof(uaddr->sa_family))
543                 return -EINVAL;
544         if (uaddr->sa_family == AF_UNSPEC)
545                 return sk->sk_prot->disconnect(sk, flags);
546
547         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
548                 return -EAGAIN;
549         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
550 }
551 EXPORT_SYMBOL(inet_dgram_connect);
552
553 static long inet_wait_for_connect(struct sock *sk, long timeo)
554 {
555         DEFINE_WAIT(wait);
556
557         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
558
559         /* Basic assumption: if someone sets sk->sk_err, he _must_
560          * change state of the socket from TCP_SYN_*.
561          * Connect() does not allow to get error notifications
562          * without closing the socket.
563          */
564         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
565                 release_sock(sk);
566                 timeo = schedule_timeout(timeo);
567                 lock_sock(sk);
568                 if (signal_pending(current) || !timeo)
569                         break;
570                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
571         }
572         finish_wait(sk_sleep(sk), &wait);
573         return timeo;
574 }
575
576 /*
577  *      Connect to a remote host. There is regrettably still a little
578  *      TCP 'magic' in here.
579  */
580 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
581                         int addr_len, int flags)
582 {
583         struct sock *sk = sock->sk;
584         int err;
585         long timeo;
586
587         if (addr_len < sizeof(uaddr->sa_family))
588                 return -EINVAL;
589
590         lock_sock(sk);
591
592         if (uaddr->sa_family == AF_UNSPEC) {
593                 err = sk->sk_prot->disconnect(sk, flags);
594                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
595                 goto out;
596         }
597
598         switch (sock->state) {
599         default:
600                 err = -EINVAL;
601                 goto out;
602         case SS_CONNECTED:
603                 err = -EISCONN;
604                 goto out;
605         case SS_CONNECTING:
606                 err = -EALREADY;
607                 /* Fall out of switch with err, set for this state */
608                 break;
609         case SS_UNCONNECTED:
610                 err = -EISCONN;
611                 if (sk->sk_state != TCP_CLOSE)
612                         goto out;
613
614                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
615                 if (err < 0)
616                         goto out;
617
618                 sock->state = SS_CONNECTING;
619
620                 /* Just entered SS_CONNECTING state; the only
621                  * difference is that return value in non-blocking
622                  * case is EINPROGRESS, rather than EALREADY.
623                  */
624                 err = -EINPROGRESS;
625                 break;
626         }
627
628         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
629
630         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
631                 /* Error code is set above */
632                 if (!timeo || !inet_wait_for_connect(sk, timeo))
633                         goto out;
634
635                 err = sock_intr_errno(timeo);
636                 if (signal_pending(current))
637                         goto out;
638         }
639
640         /* Connection was closed by RST, timeout, ICMP error
641          * or another process disconnected us.
642          */
643         if (sk->sk_state == TCP_CLOSE)
644                 goto sock_error;
645
646         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
647          * and error was received after socket entered established state.
648          * Hence, it is handled normally after connect() return successfully.
649          */
650
651         sock->state = SS_CONNECTED;
652         err = 0;
653 out:
654         release_sock(sk);
655         return err;
656
657 sock_error:
658         err = sock_error(sk) ? : -ECONNABORTED;
659         sock->state = SS_UNCONNECTED;
660         if (sk->sk_prot->disconnect(sk, flags))
661                 sock->state = SS_DISCONNECTING;
662         goto out;
663 }
664 EXPORT_SYMBOL(inet_stream_connect);
665
666 /*
667  *      Accept a pending connection. The TCP layer now gives BSD semantics.
668  */
669
670 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
671 {
672         struct sock *sk1 = sock->sk;
673         int err = -EINVAL;
674         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
675
676         if (!sk2)
677                 goto do_err;
678
679         lock_sock(sk2);
680
681         sock_rps_record_flow(sk2);
682         WARN_ON(!((1 << sk2->sk_state) &
683                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
684
685         sock_graft(sk2, newsock);
686
687         newsock->state = SS_CONNECTED;
688         err = 0;
689         release_sock(sk2);
690 do_err:
691         return err;
692 }
693 EXPORT_SYMBOL(inet_accept);
694
695
696 /*
697  *      This does both peername and sockname.
698  */
699 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
700                         int *uaddr_len, int peer)
701 {
702         struct sock *sk         = sock->sk;
703         struct inet_sock *inet  = inet_sk(sk);
704         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
705
706         sin->sin_family = AF_INET;
707         if (peer) {
708                 if (!inet->inet_dport ||
709                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
710                      peer == 1))
711                         return -ENOTCONN;
712                 sin->sin_port = inet->inet_dport;
713                 sin->sin_addr.s_addr = inet->inet_daddr;
714         } else {
715                 __be32 addr = inet->inet_rcv_saddr;
716                 if (!addr)
717                         addr = inet->inet_saddr;
718                 sin->sin_port = inet->inet_sport;
719                 sin->sin_addr.s_addr = addr;
720         }
721         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
722         *uaddr_len = sizeof(*sin);
723         return 0;
724 }
725 EXPORT_SYMBOL(inet_getname);
726
727 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
728                  size_t size)
729 {
730         struct sock *sk = sock->sk;
731
732         sock_rps_record_flow(sk);
733
734         /* We may need to bind the socket. */
735         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
736             inet_autobind(sk))
737                 return -EAGAIN;
738
739         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
740 }
741 EXPORT_SYMBOL(inet_sendmsg);
742
743 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
744                       size_t size, int flags)
745 {
746         struct sock *sk = sock->sk;
747
748         sock_rps_record_flow(sk);
749
750         /* We may need to bind the socket. */
751         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
752             inet_autobind(sk))
753                 return -EAGAIN;
754
755         if (sk->sk_prot->sendpage)
756                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
757         return sock_no_sendpage(sock, page, offset, size, flags);
758 }
759 EXPORT_SYMBOL(inet_sendpage);
760
761 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
762                  size_t size, int flags)
763 {
764         struct sock *sk = sock->sk;
765         int addr_len = 0;
766         int err;
767
768         sock_rps_record_flow(sk);
769
770         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
771                                    flags & ~MSG_DONTWAIT, &addr_len);
772         if (err >= 0)
773                 msg->msg_namelen = addr_len;
774         return err;
775 }
776 EXPORT_SYMBOL(inet_recvmsg);
777
778 int inet_shutdown(struct socket *sock, int how)
779 {
780         struct sock *sk = sock->sk;
781         int err = 0;
782
783         /* This should really check to make sure
784          * the socket is a TCP socket. (WHY AC...)
785          */
786         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
787                        1->2 bit 2 snds.
788                        2->3 */
789         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
790                 return -EINVAL;
791
792         lock_sock(sk);
793         if (sock->state == SS_CONNECTING) {
794                 if ((1 << sk->sk_state) &
795                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
796                         sock->state = SS_DISCONNECTING;
797                 else
798                         sock->state = SS_CONNECTED;
799         }
800
801         switch (sk->sk_state) {
802         case TCP_CLOSE:
803                 err = -ENOTCONN;
804                 /* Hack to wake up other listeners, who can poll for
805                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
806         default:
807                 sk->sk_shutdown |= how;
808                 if (sk->sk_prot->shutdown)
809                         sk->sk_prot->shutdown(sk, how);
810                 break;
811
812         /* Remaining two branches are temporary solution for missing
813          * close() in multithreaded environment. It is _not_ a good idea,
814          * but we have no choice until close() is repaired at VFS level.
815          */
816         case TCP_LISTEN:
817                 if (!(how & RCV_SHUTDOWN))
818                         break;
819                 /* Fall through */
820         case TCP_SYN_SENT:
821                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
822                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
823                 break;
824         }
825
826         /* Wake up anyone sleeping in poll. */
827         sk->sk_state_change(sk);
828         release_sock(sk);
829         return err;
830 }
831 EXPORT_SYMBOL(inet_shutdown);
832
833 /*
834  *      ioctl() calls you can issue on an INET socket. Most of these are
835  *      device configuration and stuff and very rarely used. Some ioctls
836  *      pass on to the socket itself.
837  *
838  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
839  *      loads the devconfigure module does its configuring and unloads it.
840  *      There's a good 20K of config code hanging around the kernel.
841  */
842
843 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
844 {
845         struct sock *sk = sock->sk;
846         int err = 0;
847         struct net *net = sock_net(sk);
848
849         switch (cmd) {
850         case SIOCGSTAMP:
851                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
852                 break;
853         case SIOCGSTAMPNS:
854                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
855                 break;
856         case SIOCADDRT:
857         case SIOCDELRT:
858         case SIOCRTMSG:
859                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
860                 break;
861         case SIOCDARP:
862         case SIOCGARP:
863         case SIOCSARP:
864                 err = arp_ioctl(net, cmd, (void __user *)arg);
865                 break;
866         case SIOCGIFADDR:
867         case SIOCSIFADDR:
868         case SIOCGIFBRDADDR:
869         case SIOCSIFBRDADDR:
870         case SIOCGIFNETMASK:
871         case SIOCSIFNETMASK:
872         case SIOCGIFDSTADDR:
873         case SIOCSIFDSTADDR:
874         case SIOCSIFPFLAGS:
875         case SIOCGIFPFLAGS:
876         case SIOCSIFFLAGS:
877                 err = devinet_ioctl(net, cmd, (void __user *)arg);
878                 break;
879         default:
880                 if (sk->sk_prot->ioctl)
881                         err = sk->sk_prot->ioctl(sk, cmd, arg);
882                 else
883                         err = -ENOIOCTLCMD;
884                 break;
885         }
886         return err;
887 }
888 EXPORT_SYMBOL(inet_ioctl);
889
890 #ifdef CONFIG_COMPAT
891 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
892 {
893         struct sock *sk = sock->sk;
894         int err = -ENOIOCTLCMD;
895
896         if (sk->sk_prot->compat_ioctl)
897                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
898
899         return err;
900 }
901 #endif
902
903 const struct proto_ops inet_stream_ops = {
904         .family            = PF_INET,
905         .owner             = THIS_MODULE,
906         .release           = inet_release,
907         .bind              = inet_bind,
908         .connect           = inet_stream_connect,
909         .socketpair        = sock_no_socketpair,
910         .accept            = inet_accept,
911         .getname           = inet_getname,
912         .poll              = tcp_poll,
913         .ioctl             = inet_ioctl,
914         .listen            = inet_listen,
915         .shutdown          = inet_shutdown,
916         .setsockopt        = sock_common_setsockopt,
917         .getsockopt        = sock_common_getsockopt,
918         .sendmsg           = inet_sendmsg,
919         .recvmsg           = inet_recvmsg,
920         .mmap              = sock_no_mmap,
921         .sendpage          = inet_sendpage,
922         .splice_read       = tcp_splice_read,
923 #ifdef CONFIG_COMPAT
924         .compat_setsockopt = compat_sock_common_setsockopt,
925         .compat_getsockopt = compat_sock_common_getsockopt,
926         .compat_ioctl      = inet_compat_ioctl,
927 #endif
928 };
929 EXPORT_SYMBOL(inet_stream_ops);
930
931 const struct proto_ops inet_dgram_ops = {
932         .family            = PF_INET,
933         .owner             = THIS_MODULE,
934         .release           = inet_release,
935         .bind              = inet_bind,
936         .connect           = inet_dgram_connect,
937         .socketpair        = sock_no_socketpair,
938         .accept            = sock_no_accept,
939         .getname           = inet_getname,
940         .poll              = udp_poll,
941         .ioctl             = inet_ioctl,
942         .listen            = sock_no_listen,
943         .shutdown          = inet_shutdown,
944         .setsockopt        = sock_common_setsockopt,
945         .getsockopt        = sock_common_getsockopt,
946         .sendmsg           = inet_sendmsg,
947         .recvmsg           = inet_recvmsg,
948         .mmap              = sock_no_mmap,
949         .sendpage          = inet_sendpage,
950 #ifdef CONFIG_COMPAT
951         .compat_setsockopt = compat_sock_common_setsockopt,
952         .compat_getsockopt = compat_sock_common_getsockopt,
953         .compat_ioctl      = inet_compat_ioctl,
954 #endif
955 };
956 EXPORT_SYMBOL(inet_dgram_ops);
957
958 /*
959  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
960  * udp_poll
961  */
962 static const struct proto_ops inet_sockraw_ops = {
963         .family            = PF_INET,
964         .owner             = THIS_MODULE,
965         .release           = inet_release,
966         .bind              = inet_bind,
967         .connect           = inet_dgram_connect,
968         .socketpair        = sock_no_socketpair,
969         .accept            = sock_no_accept,
970         .getname           = inet_getname,
971         .poll              = datagram_poll,
972         .ioctl             = inet_ioctl,
973         .listen            = sock_no_listen,
974         .shutdown          = inet_shutdown,
975         .setsockopt        = sock_common_setsockopt,
976         .getsockopt        = sock_common_getsockopt,
977         .sendmsg           = inet_sendmsg,
978         .recvmsg           = inet_recvmsg,
979         .mmap              = sock_no_mmap,
980         .sendpage          = inet_sendpage,
981 #ifdef CONFIG_COMPAT
982         .compat_setsockopt = compat_sock_common_setsockopt,
983         .compat_getsockopt = compat_sock_common_getsockopt,
984         .compat_ioctl      = inet_compat_ioctl,
985 #endif
986 };
987
988 static const struct net_proto_family inet_family_ops = {
989         .family = PF_INET,
990         .create = inet_create,
991         .owner  = THIS_MODULE,
992 };
993
994 /* Upon startup we insert all the elements in inetsw_array[] into
995  * the linked list inetsw.
996  */
997 static struct inet_protosw inetsw_array[] =
998 {
999         {
1000                 .type =       SOCK_STREAM,
1001                 .protocol =   IPPROTO_TCP,
1002                 .prot =       &tcp_prot,
1003                 .ops =        &inet_stream_ops,
1004                 .no_check =   0,
1005                 .flags =      INET_PROTOSW_PERMANENT |
1006                               INET_PROTOSW_ICSK,
1007         },
1008
1009         {
1010                 .type =       SOCK_DGRAM,
1011                 .protocol =   IPPROTO_UDP,
1012                 .prot =       &udp_prot,
1013                 .ops =        &inet_dgram_ops,
1014                 .no_check =   UDP_CSUM_DEFAULT,
1015                 .flags =      INET_PROTOSW_PERMANENT,
1016        },
1017
1018        {
1019                 .type =       SOCK_DGRAM,
1020                 .protocol =   IPPROTO_ICMP,
1021                 .prot =       &ping_prot,
1022                 .ops =        &inet_dgram_ops,
1023                 .no_check =   UDP_CSUM_DEFAULT,
1024                 .flags =      INET_PROTOSW_REUSE,
1025        },
1026
1027        {
1028                .type =       SOCK_RAW,
1029                .protocol =   IPPROTO_IP,        /* wild card */
1030                .prot =       &raw_prot,
1031                .ops =        &inet_sockraw_ops,
1032                .no_check =   UDP_CSUM_DEFAULT,
1033                .flags =      INET_PROTOSW_REUSE,
1034        }
1035 };
1036
1037 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1038
1039 void inet_register_protosw(struct inet_protosw *p)
1040 {
1041         struct list_head *lh;
1042         struct inet_protosw *answer;
1043         int protocol = p->protocol;
1044         struct list_head *last_perm;
1045
1046         spin_lock_bh(&inetsw_lock);
1047
1048         if (p->type >= SOCK_MAX)
1049                 goto out_illegal;
1050
1051         /* If we are trying to override a permanent protocol, bail. */
1052         answer = NULL;
1053         last_perm = &inetsw[p->type];
1054         list_for_each(lh, &inetsw[p->type]) {
1055                 answer = list_entry(lh, struct inet_protosw, list);
1056
1057                 /* Check only the non-wild match. */
1058                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1059                         if (protocol == answer->protocol)
1060                                 break;
1061                         last_perm = lh;
1062                 }
1063
1064                 answer = NULL;
1065         }
1066         if (answer)
1067                 goto out_permanent;
1068
1069         /* Add the new entry after the last permanent entry if any, so that
1070          * the new entry does not override a permanent entry when matched with
1071          * a wild-card protocol. But it is allowed to override any existing
1072          * non-permanent entry.  This means that when we remove this entry, the
1073          * system automatically returns to the old behavior.
1074          */
1075         list_add_rcu(&p->list, last_perm);
1076 out:
1077         spin_unlock_bh(&inetsw_lock);
1078
1079         return;
1080
1081 out_permanent:
1082         printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1083                protocol);
1084         goto out;
1085
1086 out_illegal:
1087         printk(KERN_ERR
1088                "Ignoring attempt to register invalid socket type %d.\n",
1089                p->type);
1090         goto out;
1091 }
1092 EXPORT_SYMBOL(inet_register_protosw);
1093
1094 void inet_unregister_protosw(struct inet_protosw *p)
1095 {
1096         if (INET_PROTOSW_PERMANENT & p->flags) {
1097                 printk(KERN_ERR
1098                        "Attempt to unregister permanent protocol %d.\n",
1099                        p->protocol);
1100         } else {
1101                 spin_lock_bh(&inetsw_lock);
1102                 list_del_rcu(&p->list);
1103                 spin_unlock_bh(&inetsw_lock);
1104
1105                 synchronize_net();
1106         }
1107 }
1108 EXPORT_SYMBOL(inet_unregister_protosw);
1109
1110 /*
1111  *      Shall we try to damage output packets if routing dev changes?
1112  */
1113
1114 int sysctl_ip_dynaddr __read_mostly;
1115
1116 static int inet_sk_reselect_saddr(struct sock *sk)
1117 {
1118         struct inet_sock *inet = inet_sk(sk);
1119         __be32 old_saddr = inet->inet_saddr;
1120         __be32 daddr = inet->inet_daddr;
1121         struct flowi4 *fl4;
1122         struct rtable *rt;
1123         __be32 new_saddr;
1124         struct ip_options_rcu *inet_opt;
1125
1126         inet_opt = rcu_dereference_protected(inet->inet_opt,
1127                                              sock_owned_by_user(sk));
1128         if (inet_opt && inet_opt->opt.srr)
1129                 daddr = inet_opt->opt.faddr;
1130
1131         /* Query new route. */
1132         fl4 = &inet->cork.fl.u.ip4;
1133         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1134                               sk->sk_bound_dev_if, sk->sk_protocol,
1135                               inet->inet_sport, inet->inet_dport, sk, false);
1136         if (IS_ERR(rt))
1137                 return PTR_ERR(rt);
1138
1139         sk_setup_caps(sk, &rt->dst);
1140
1141         new_saddr = fl4->saddr;
1142
1143         if (new_saddr == old_saddr)
1144                 return 0;
1145
1146         if (sysctl_ip_dynaddr > 1) {
1147                 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1148                        __func__, &old_saddr, &new_saddr);
1149         }
1150
1151         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1152
1153         /*
1154          * XXX The only one ugly spot where we need to
1155          * XXX really change the sockets identity after
1156          * XXX it has entered the hashes. -DaveM
1157          *
1158          * Besides that, it does not check for connection
1159          * uniqueness. Wait for troubles.
1160          */
1161         __sk_prot_rehash(sk);
1162         return 0;
1163 }
1164
1165 int inet_sk_rebuild_header(struct sock *sk)
1166 {
1167         struct inet_sock *inet = inet_sk(sk);
1168         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1169         __be32 daddr;
1170         struct ip_options_rcu *inet_opt;
1171         struct flowi4 *fl4;
1172         int err;
1173
1174         /* Route is OK, nothing to do. */
1175         if (rt)
1176                 return 0;
1177
1178         /* Reroute. */
1179         rcu_read_lock();
1180         inet_opt = rcu_dereference(inet->inet_opt);
1181         daddr = inet->inet_daddr;
1182         if (inet_opt && inet_opt->opt.srr)
1183                 daddr = inet_opt->opt.faddr;
1184         rcu_read_unlock();
1185         fl4 = &inet->cork.fl.u.ip4;
1186         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1187                                    inet->inet_dport, inet->inet_sport,
1188                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1189                                    sk->sk_bound_dev_if);
1190         if (!IS_ERR(rt)) {
1191                 err = 0;
1192                 sk_setup_caps(sk, &rt->dst);
1193         } else {
1194                 err = PTR_ERR(rt);
1195
1196                 /* Routing failed... */
1197                 sk->sk_route_caps = 0;
1198                 /*
1199                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1200                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1201                  */
1202                 if (!sysctl_ip_dynaddr ||
1203                     sk->sk_state != TCP_SYN_SENT ||
1204                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1205                     (err = inet_sk_reselect_saddr(sk)) != 0)
1206                         sk->sk_err_soft = -err;
1207         }
1208
1209         return err;
1210 }
1211 EXPORT_SYMBOL(inet_sk_rebuild_header);
1212
1213 static int inet_gso_send_check(struct sk_buff *skb)
1214 {
1215         const struct iphdr *iph;
1216         const struct net_protocol *ops;
1217         int proto;
1218         int ihl;
1219         int err = -EINVAL;
1220
1221         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1222                 goto out;
1223
1224         iph = ip_hdr(skb);
1225         ihl = iph->ihl * 4;
1226         if (ihl < sizeof(*iph))
1227                 goto out;
1228
1229         if (unlikely(!pskb_may_pull(skb, ihl)))
1230                 goto out;
1231
1232         __skb_pull(skb, ihl);
1233         skb_reset_transport_header(skb);
1234         iph = ip_hdr(skb);
1235         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1236         err = -EPROTONOSUPPORT;
1237
1238         rcu_read_lock();
1239         ops = rcu_dereference(inet_protos[proto]);
1240         if (likely(ops && ops->gso_send_check))
1241                 err = ops->gso_send_check(skb);
1242         rcu_read_unlock();
1243
1244 out:
1245         return err;
1246 }
1247
1248 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1249 {
1250         struct sk_buff *segs = ERR_PTR(-EINVAL);
1251         struct iphdr *iph;
1252         const struct net_protocol *ops;
1253         int proto;
1254         int ihl;
1255         int id;
1256         unsigned int offset = 0;
1257
1258         if (!(features & NETIF_F_V4_CSUM))
1259                 features &= ~NETIF_F_SG;
1260
1261         if (unlikely(skb_shinfo(skb)->gso_type &
1262                      ~(SKB_GSO_TCPV4 |
1263                        SKB_GSO_UDP |
1264                        SKB_GSO_DODGY |
1265                        SKB_GSO_TCP_ECN |
1266                        0)))
1267                 goto out;
1268
1269         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1270                 goto out;
1271
1272         iph = ip_hdr(skb);
1273         ihl = iph->ihl * 4;
1274         if (ihl < sizeof(*iph))
1275                 goto out;
1276
1277         if (unlikely(!pskb_may_pull(skb, ihl)))
1278                 goto out;
1279
1280         __skb_pull(skb, ihl);
1281         skb_reset_transport_header(skb);
1282         iph = ip_hdr(skb);
1283         id = ntohs(iph->id);
1284         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1285         segs = ERR_PTR(-EPROTONOSUPPORT);
1286
1287         rcu_read_lock();
1288         ops = rcu_dereference(inet_protos[proto]);
1289         if (likely(ops && ops->gso_segment))
1290                 segs = ops->gso_segment(skb, features);
1291         rcu_read_unlock();
1292
1293         if (!segs || IS_ERR(segs))
1294                 goto out;
1295
1296         skb = segs;
1297         do {
1298                 iph = ip_hdr(skb);
1299                 if (proto == IPPROTO_UDP) {
1300                         iph->id = htons(id);
1301                         iph->frag_off = htons(offset >> 3);
1302                         if (skb->next != NULL)
1303                                 iph->frag_off |= htons(IP_MF);
1304                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1305                 } else
1306                         iph->id = htons(id++);
1307                 iph->tot_len = htons(skb->len - skb->mac_len);
1308                 iph->check = 0;
1309                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1310         } while ((skb = skb->next));
1311
1312 out:
1313         return segs;
1314 }
1315
1316 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1317                                          struct sk_buff *skb)
1318 {
1319         const struct net_protocol *ops;
1320         struct sk_buff **pp = NULL;
1321         struct sk_buff *p;
1322         const struct iphdr *iph;
1323         unsigned int hlen;
1324         unsigned int off;
1325         unsigned int id;
1326         int flush = 1;
1327         int proto;
1328
1329         off = skb_gro_offset(skb);
1330         hlen = off + sizeof(*iph);
1331         iph = skb_gro_header_fast(skb, off);
1332         if (skb_gro_header_hard(skb, hlen)) {
1333                 iph = skb_gro_header_slow(skb, hlen, off);
1334                 if (unlikely(!iph))
1335                         goto out;
1336         }
1337
1338         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1339
1340         rcu_read_lock();
1341         ops = rcu_dereference(inet_protos[proto]);
1342         if (!ops || !ops->gro_receive)
1343                 goto out_unlock;
1344
1345         if (*(u8 *)iph != 0x45)
1346                 goto out_unlock;
1347
1348         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1349                 goto out_unlock;
1350
1351         id = ntohl(*(__be32 *)&iph->id);
1352         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1353         id >>= 16;
1354
1355         for (p = *head; p; p = p->next) {
1356                 struct iphdr *iph2;
1357
1358                 if (!NAPI_GRO_CB(p)->same_flow)
1359                         continue;
1360
1361                 iph2 = ip_hdr(p);
1362
1363                 if ((iph->protocol ^ iph2->protocol) |
1364                     (iph->tos ^ iph2->tos) |
1365                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1366                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1367                         NAPI_GRO_CB(p)->same_flow = 0;
1368                         continue;
1369                 }
1370
1371                 /* All fields must match except length and checksum. */
1372                 NAPI_GRO_CB(p)->flush |=
1373                         (iph->ttl ^ iph2->ttl) |
1374                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1375
1376                 NAPI_GRO_CB(p)->flush |= flush;
1377         }
1378
1379         NAPI_GRO_CB(skb)->flush |= flush;
1380         skb_gro_pull(skb, sizeof(*iph));
1381         skb_set_transport_header(skb, skb_gro_offset(skb));
1382
1383         pp = ops->gro_receive(head, skb);
1384
1385 out_unlock:
1386         rcu_read_unlock();
1387
1388 out:
1389         NAPI_GRO_CB(skb)->flush |= flush;
1390
1391         return pp;
1392 }
1393
1394 static int inet_gro_complete(struct sk_buff *skb)
1395 {
1396         const struct net_protocol *ops;
1397         struct iphdr *iph = ip_hdr(skb);
1398         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1399         int err = -ENOSYS;
1400         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1401
1402         csum_replace2(&iph->check, iph->tot_len, newlen);
1403         iph->tot_len = newlen;
1404
1405         rcu_read_lock();
1406         ops = rcu_dereference(inet_protos[proto]);
1407         if (WARN_ON(!ops || !ops->gro_complete))
1408                 goto out_unlock;
1409
1410         err = ops->gro_complete(skb);
1411
1412 out_unlock:
1413         rcu_read_unlock();
1414
1415         return err;
1416 }
1417
1418 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1419                          unsigned short type, unsigned char protocol,
1420                          struct net *net)
1421 {
1422         struct socket *sock;
1423         int rc = sock_create_kern(family, type, protocol, &sock);
1424
1425         if (rc == 0) {
1426                 *sk = sock->sk;
1427                 (*sk)->sk_allocation = GFP_ATOMIC;
1428                 /*
1429                  * Unhash it so that IP input processing does not even see it,
1430                  * we do not wish this socket to see incoming packets.
1431                  */
1432                 (*sk)->sk_prot->unhash(*sk);
1433
1434                 sk_change_net(*sk, net);
1435         }
1436         return rc;
1437 }
1438 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1439
1440 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1441 {
1442         unsigned long res = 0;
1443         int i, j;
1444
1445         for_each_possible_cpu(i) {
1446                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1447                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1448         }
1449         return res;
1450 }
1451 EXPORT_SYMBOL_GPL(snmp_fold_field);
1452
1453 #if BITS_PER_LONG==32
1454
1455 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1456 {
1457         u64 res = 0;
1458         int cpu;
1459
1460         for_each_possible_cpu(cpu) {
1461                 void *bhptr;
1462                 struct u64_stats_sync *syncp;
1463                 u64 v;
1464                 unsigned int start;
1465
1466                 bhptr = per_cpu_ptr(mib[0], cpu);
1467                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1468                 do {
1469                         start = u64_stats_fetch_begin_bh(syncp);
1470                         v = *(((u64 *) bhptr) + offt);
1471                 } while (u64_stats_fetch_retry_bh(syncp, start));
1472
1473                 res += v;
1474         }
1475         return res;
1476 }
1477 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1478 #endif
1479
1480 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1481 {
1482         BUG_ON(ptr == NULL);
1483         ptr[0] = __alloc_percpu(mibsize, align);
1484         if (!ptr[0])
1485                 return -ENOMEM;
1486 #if SNMP_ARRAY_SZ == 2
1487         ptr[1] = __alloc_percpu(mibsize, align);
1488         if (!ptr[1]) {
1489                 free_percpu(ptr[0]);
1490                 ptr[0] = NULL;
1491                 return -ENOMEM;
1492         }
1493 #endif
1494         return 0;
1495 }
1496 EXPORT_SYMBOL_GPL(snmp_mib_init);
1497
1498 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1499 {
1500         int i;
1501
1502         BUG_ON(ptr == NULL);
1503         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1504                 free_percpu(ptr[i]);
1505                 ptr[i] = NULL;
1506         }
1507 }
1508 EXPORT_SYMBOL_GPL(snmp_mib_free);
1509
1510 #ifdef CONFIG_IP_MULTICAST
1511 static const struct net_protocol igmp_protocol = {
1512         .handler =      igmp_rcv,
1513         .netns_ok =     1,
1514 };
1515 #endif
1516
1517 static const struct net_protocol tcp_protocol = {
1518         .handler =      tcp_v4_rcv,
1519         .err_handler =  tcp_v4_err,
1520         .gso_send_check = tcp_v4_gso_send_check,
1521         .gso_segment =  tcp_tso_segment,
1522         .gro_receive =  tcp4_gro_receive,
1523         .gro_complete = tcp4_gro_complete,
1524         .no_policy =    1,
1525         .netns_ok =     1,
1526 };
1527
1528 static const struct net_protocol udp_protocol = {
1529         .handler =      udp_rcv,
1530         .err_handler =  udp_err,
1531         .gso_send_check = udp4_ufo_send_check,
1532         .gso_segment = udp4_ufo_fragment,
1533         .no_policy =    1,
1534         .netns_ok =     1,
1535 };
1536
1537 static const struct net_protocol icmp_protocol = {
1538         .handler =      icmp_rcv,
1539         .err_handler =  ping_err,
1540         .no_policy =    1,
1541         .netns_ok =     1,
1542 };
1543
1544 static __net_init int ipv4_mib_init_net(struct net *net)
1545 {
1546         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1547                           sizeof(struct tcp_mib),
1548                           __alignof__(struct tcp_mib)) < 0)
1549                 goto err_tcp_mib;
1550         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1551                           sizeof(struct ipstats_mib),
1552                           __alignof__(struct ipstats_mib)) < 0)
1553                 goto err_ip_mib;
1554         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1555                           sizeof(struct linux_mib),
1556                           __alignof__(struct linux_mib)) < 0)
1557                 goto err_net_mib;
1558         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1559                           sizeof(struct udp_mib),
1560                           __alignof__(struct udp_mib)) < 0)
1561                 goto err_udp_mib;
1562         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1563                           sizeof(struct udp_mib),
1564                           __alignof__(struct udp_mib)) < 0)
1565                 goto err_udplite_mib;
1566         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1567                           sizeof(struct icmp_mib),
1568                           __alignof__(struct icmp_mib)) < 0)
1569                 goto err_icmp_mib;
1570         if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1571                           sizeof(struct icmpmsg_mib),
1572                           __alignof__(struct icmpmsg_mib)) < 0)
1573                 goto err_icmpmsg_mib;
1574
1575         tcp_mib_init(net);
1576         return 0;
1577
1578 err_icmpmsg_mib:
1579         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1580 err_icmp_mib:
1581         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1582 err_udplite_mib:
1583         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1584 err_udp_mib:
1585         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1586 err_net_mib:
1587         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1588 err_ip_mib:
1589         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1590 err_tcp_mib:
1591         return -ENOMEM;
1592 }
1593
1594 static __net_exit void ipv4_mib_exit_net(struct net *net)
1595 {
1596         snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1597         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1598         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1599         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1600         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1601         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1602         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1603 }
1604
1605 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1606         .init = ipv4_mib_init_net,
1607         .exit = ipv4_mib_exit_net,
1608 };
1609
1610 static int __init init_ipv4_mibs(void)
1611 {
1612         return register_pernet_subsys(&ipv4_mib_ops);
1613 }
1614
1615 static int ipv4_proc_init(void);
1616
1617 /*
1618  *      IP protocol layer initialiser
1619  */
1620
1621 static struct packet_type ip_packet_type __read_mostly = {
1622         .type = cpu_to_be16(ETH_P_IP),
1623         .func = ip_rcv,
1624         .gso_send_check = inet_gso_send_check,
1625         .gso_segment = inet_gso_segment,
1626         .gro_receive = inet_gro_receive,
1627         .gro_complete = inet_gro_complete,
1628 };
1629
1630 static int __init inet_init(void)
1631 {
1632         struct sk_buff *dummy_skb;
1633         struct inet_protosw *q;
1634         struct list_head *r;
1635         int rc = -EINVAL;
1636
1637         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1638
1639         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1640         if (!sysctl_local_reserved_ports)
1641                 goto out;
1642
1643         rc = proto_register(&tcp_prot, 1);
1644         if (rc)
1645                 goto out_free_reserved_ports;
1646
1647         rc = proto_register(&udp_prot, 1);
1648         if (rc)
1649                 goto out_unregister_tcp_proto;
1650
1651         rc = proto_register(&raw_prot, 1);
1652         if (rc)
1653                 goto out_unregister_udp_proto;
1654
1655         rc = proto_register(&ping_prot, 1);
1656         if (rc)
1657                 goto out_unregister_raw_proto;
1658
1659         /*
1660          *      Tell SOCKET that we are alive...
1661          */
1662
1663         (void)sock_register(&inet_family_ops);
1664
1665 #ifdef CONFIG_SYSCTL
1666         ip_static_sysctl_init();
1667 #endif
1668
1669         /*
1670          *      Add all the base protocols.
1671          */
1672
1673         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1674                 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1675         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1676                 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1677         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1678                 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1679 #ifdef CONFIG_IP_MULTICAST
1680         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1681                 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1682 #endif
1683
1684         /* Register the socket-side information for inet_create. */
1685         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1686                 INIT_LIST_HEAD(r);
1687
1688         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1689                 inet_register_protosw(q);
1690
1691         /*
1692          *      Set the ARP module up
1693          */
1694
1695         arp_init();
1696
1697         /*
1698          *      Set the IP module up
1699          */
1700
1701         ip_init();
1702
1703         tcp_v4_init();
1704
1705         /* Setup TCP slab cache for open requests. */
1706         tcp_init();
1707
1708         /* Setup UDP memory threshold */
1709         udp_init();
1710
1711         /* Add UDP-Lite (RFC 3828) */
1712         udplite4_register();
1713
1714         ping_init();
1715
1716         /*
1717          *      Set the ICMP layer up
1718          */
1719
1720         if (icmp_init() < 0)
1721                 panic("Failed to create the ICMP control socket.\n");
1722
1723         /*
1724          *      Initialise the multicast router
1725          */
1726 #if defined(CONFIG_IP_MROUTE)
1727         if (ip_mr_init())
1728                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1729 #endif
1730         /*
1731          *      Initialise per-cpu ipv4 mibs
1732          */
1733
1734         if (init_ipv4_mibs())
1735                 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1736
1737         ipv4_proc_init();
1738
1739         ipfrag_init();
1740
1741         dev_add_pack(&ip_packet_type);
1742
1743         rc = 0;
1744 out:
1745         return rc;
1746 out_unregister_raw_proto:
1747         proto_unregister(&raw_prot);
1748 out_unregister_udp_proto:
1749         proto_unregister(&udp_prot);
1750 out_unregister_tcp_proto:
1751         proto_unregister(&tcp_prot);
1752 out_free_reserved_ports:
1753         kfree(sysctl_local_reserved_ports);
1754         goto out;
1755 }
1756
1757 fs_initcall(inet_init);
1758
1759 /* ------------------------------------------------------------------------ */
1760
1761 #ifdef CONFIG_PROC_FS
1762 static int __init ipv4_proc_init(void)
1763 {
1764         int rc = 0;
1765
1766         if (raw_proc_init())
1767                 goto out_raw;
1768         if (tcp4_proc_init())
1769                 goto out_tcp;
1770         if (udp4_proc_init())
1771                 goto out_udp;
1772         if (ping_proc_init())
1773                 goto out_ping;
1774         if (ip_misc_proc_init())
1775                 goto out_misc;
1776 out:
1777         return rc;
1778 out_misc:
1779         ping_proc_exit();
1780 out_ping:
1781         udp4_proc_exit();
1782 out_udp:
1783         tcp4_proc_exit();
1784 out_tcp:
1785         raw_proc_exit();
1786 out_raw:
1787         rc = -ENOMEM;
1788         goto out;
1789 }
1790
1791 #else /* CONFIG_PROC_FS */
1792 static int __init ipv4_proc_init(void)
1793 {
1794         return 0;
1795 }
1796 #endif /* CONFIG_PROC_FS */
1797
1798 MODULE_ALIAS_NETPROTO(PF_INET);
1799