ipv6: use a stronger hash for tcp
[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 #define pr_fmt(fmt) "IPv4: " fmt
69
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
84 #include <linux/mm.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
92
93 #include <asm/uaccess.h>
94
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/ipip.h>
115 #include <net/inet_common.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
121
122
123 /* The inetsw table contains everything that inet_create needs to
124  * build a new socket.
125  */
126 static struct list_head inetsw[SOCK_MAX];
127 static DEFINE_SPINLOCK(inetsw_lock);
128
129 struct ipv4_config ipv4_config;
130 EXPORT_SYMBOL(ipv4_config);
131
132 /* New destruction routine */
133
134 void inet_sock_destruct(struct sock *sk)
135 {
136         struct inet_sock *inet = inet_sk(sk);
137
138         __skb_queue_purge(&sk->sk_receive_queue);
139         __skb_queue_purge(&sk->sk_error_queue);
140
141         sk_mem_reclaim(sk);
142
143         if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144                 pr_err("Attempt to release TCP socket in state %d %p\n",
145                        sk->sk_state, sk);
146                 return;
147         }
148         if (!sock_flag(sk, SOCK_DEAD)) {
149                 pr_err("Attempt to release alive inet socket %p\n", sk);
150                 return;
151         }
152
153         WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154         WARN_ON(atomic_read(&sk->sk_wmem_alloc));
155         WARN_ON(sk->sk_wmem_queued);
156         WARN_ON(sk->sk_forward_alloc);
157
158         kfree(rcu_dereference_protected(inet->inet_opt, 1));
159         dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
160         sk_refcnt_debug_dec(sk);
161 }
162 EXPORT_SYMBOL(inet_sock_destruct);
163
164 /*
165  *      The routines beyond this point handle the behaviour of an AF_INET
166  *      socket object. Mostly it punts to the subprotocols of IP to do
167  *      the work.
168  */
169
170 /*
171  *      Automatically bind an unbound socket.
172  */
173
174 static int inet_autobind(struct sock *sk)
175 {
176         struct inet_sock *inet;
177         /* We may need to bind the socket. */
178         lock_sock(sk);
179         inet = inet_sk(sk);
180         if (!inet->inet_num) {
181                 if (sk->sk_prot->get_port(sk, 0)) {
182                         release_sock(sk);
183                         return -EAGAIN;
184                 }
185                 inet->inet_sport = htons(inet->inet_num);
186         }
187         release_sock(sk);
188         return 0;
189 }
190
191 /*
192  *      Move a socket into listening state.
193  */
194 int inet_listen(struct socket *sock, int backlog)
195 {
196         struct sock *sk = sock->sk;
197         unsigned char old_state;
198         int err;
199
200         lock_sock(sk);
201
202         err = -EINVAL;
203         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204                 goto out;
205
206         old_state = sk->sk_state;
207         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208                 goto out;
209
210         /* Really, if the socket is already in listen state
211          * we can only allow the backlog to be adjusted.
212          */
213         if (old_state != TCP_LISTEN) {
214                 err = inet_csk_listen_start(sk, backlog);
215                 if (err)
216                         goto out;
217         }
218         sk->sk_max_ack_backlog = backlog;
219         err = 0;
220
221 out:
222         release_sock(sk);
223         return err;
224 }
225 EXPORT_SYMBOL(inet_listen);
226
227 u32 inet_ehash_secret __read_mostly;
228 EXPORT_SYMBOL(inet_ehash_secret);
229
230 u32 ipv6_hash_secret __read_mostly;
231 EXPORT_SYMBOL(ipv6_hash_secret);
232
233 /*
234  * inet_ehash_secret must be set exactly once, and to a non nul value
235  * ipv6_hash_secret must be set exactly once.
236  */
237 void build_ehash_secret(void)
238 {
239         u32 rnd;
240
241         do {
242                 get_random_bytes(&rnd, sizeof(rnd));
243         } while (rnd == 0);
244
245         if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
246                 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
247 }
248 EXPORT_SYMBOL(build_ehash_secret);
249
250 static inline int inet_netns_ok(struct net *net, int protocol)
251 {
252         int hash;
253         const struct net_protocol *ipprot;
254
255         if (net_eq(net, &init_net))
256                 return 1;
257
258         hash = protocol & (MAX_INET_PROTOS - 1);
259         ipprot = rcu_dereference(inet_protos[hash]);
260
261         if (ipprot == NULL)
262                 /* raw IP is OK */
263                 return 1;
264         return ipprot->netns_ok;
265 }
266
267 /*
268  *      Create an inet socket.
269  */
270
271 static int inet_create(struct net *net, struct socket *sock, int protocol,
272                        int kern)
273 {
274         struct sock *sk;
275         struct inet_protosw *answer;
276         struct inet_sock *inet;
277         struct proto *answer_prot;
278         unsigned char answer_flags;
279         char answer_no_check;
280         int try_loading_module = 0;
281         int err;
282
283         if (unlikely(!inet_ehash_secret))
284                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
285                         build_ehash_secret();
286
287         sock->state = SS_UNCONNECTED;
288
289         /* Look for the requested type/protocol pair. */
290 lookup_protocol:
291         err = -ESOCKTNOSUPPORT;
292         rcu_read_lock();
293         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
294
295                 err = 0;
296                 /* Check the non-wild match. */
297                 if (protocol == answer->protocol) {
298                         if (protocol != IPPROTO_IP)
299                                 break;
300                 } else {
301                         /* Check for the two wild cases. */
302                         if (IPPROTO_IP == protocol) {
303                                 protocol = answer->protocol;
304                                 break;
305                         }
306                         if (IPPROTO_IP == answer->protocol)
307                                 break;
308                 }
309                 err = -EPROTONOSUPPORT;
310         }
311
312         if (unlikely(err)) {
313                 if (try_loading_module < 2) {
314                         rcu_read_unlock();
315                         /*
316                          * Be more specific, e.g. net-pf-2-proto-132-type-1
317                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
318                          */
319                         if (++try_loading_module == 1)
320                                 request_module("net-pf-%d-proto-%d-type-%d",
321                                                PF_INET, protocol, sock->type);
322                         /*
323                          * Fall back to generic, e.g. net-pf-2-proto-132
324                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
325                          */
326                         else
327                                 request_module("net-pf-%d-proto-%d",
328                                                PF_INET, protocol);
329                         goto lookup_protocol;
330                 } else
331                         goto out_rcu_unlock;
332         }
333
334         err = -EPERM;
335         if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
336                 goto out_rcu_unlock;
337
338         err = -EAFNOSUPPORT;
339         if (!inet_netns_ok(net, protocol))
340                 goto out_rcu_unlock;
341
342         sock->ops = answer->ops;
343         answer_prot = answer->prot;
344         answer_no_check = answer->no_check;
345         answer_flags = answer->flags;
346         rcu_read_unlock();
347
348         WARN_ON(answer_prot->slab == NULL);
349
350         err = -ENOBUFS;
351         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
352         if (sk == NULL)
353                 goto out;
354
355         err = 0;
356         sk->sk_no_check = answer_no_check;
357         if (INET_PROTOSW_REUSE & answer_flags)
358                 sk->sk_reuse = 1;
359
360         inet = inet_sk(sk);
361         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
362
363         inet->nodefrag = 0;
364
365         if (SOCK_RAW == sock->type) {
366                 inet->inet_num = protocol;
367                 if (IPPROTO_RAW == protocol)
368                         inet->hdrincl = 1;
369         }
370
371         if (ipv4_config.no_pmtu_disc)
372                 inet->pmtudisc = IP_PMTUDISC_DONT;
373         else
374                 inet->pmtudisc = IP_PMTUDISC_WANT;
375
376         inet->inet_id = 0;
377
378         sock_init_data(sock, sk);
379
380         sk->sk_destruct    = inet_sock_destruct;
381         sk->sk_protocol    = protocol;
382         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
383
384         inet->uc_ttl    = -1;
385         inet->mc_loop   = 1;
386         inet->mc_ttl    = 1;
387         inet->mc_all    = 1;
388         inet->mc_index  = 0;
389         inet->mc_list   = NULL;
390         inet->rcv_tos   = 0;
391
392         sk_refcnt_debug_inc(sk);
393
394         if (inet->inet_num) {
395                 /* It assumes that any protocol which allows
396                  * the user to assign a number at socket
397                  * creation time automatically
398                  * shares.
399                  */
400                 inet->inet_sport = htons(inet->inet_num);
401                 /* Add to protocol hash chains. */
402                 sk->sk_prot->hash(sk);
403         }
404
405         if (sk->sk_prot->init) {
406                 err = sk->sk_prot->init(sk);
407                 if (err)
408                         sk_common_release(sk);
409         }
410 out:
411         return err;
412 out_rcu_unlock:
413         rcu_read_unlock();
414         goto out;
415 }
416
417
418 /*
419  *      The peer socket should always be NULL (or else). When we call this
420  *      function we are destroying the object and from then on nobody
421  *      should refer to it.
422  */
423 int inet_release(struct socket *sock)
424 {
425         struct sock *sk = sock->sk;
426
427         if (sk) {
428                 long timeout;
429
430                 sock_rps_reset_flow(sk);
431
432                 /* Applications forget to leave groups before exiting */
433                 ip_mc_drop_socket(sk);
434
435                 /* If linger is set, we don't return until the close
436                  * is complete.  Otherwise we return immediately. The
437                  * actually closing is done the same either way.
438                  *
439                  * If the close is due to the process exiting, we never
440                  * linger..
441                  */
442                 timeout = 0;
443                 if (sock_flag(sk, SOCK_LINGER) &&
444                     !(current->flags & PF_EXITING))
445                         timeout = sk->sk_lingertime;
446                 sock->sk = NULL;
447                 sk->sk_prot->close(sk, timeout);
448         }
449         return 0;
450 }
451 EXPORT_SYMBOL(inet_release);
452
453 /* It is off by default, see below. */
454 int sysctl_ip_nonlocal_bind __read_mostly;
455 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
456
457 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
458 {
459         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
460         struct sock *sk = sock->sk;
461         struct inet_sock *inet = inet_sk(sk);
462         unsigned short snum;
463         int chk_addr_ret;
464         int err;
465
466         /* If the socket has its own bind function then use it. (RAW) */
467         if (sk->sk_prot->bind) {
468                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
469                 goto out;
470         }
471         err = -EINVAL;
472         if (addr_len < sizeof(struct sockaddr_in))
473                 goto out;
474
475         if (addr->sin_family != AF_INET) {
476                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
477                  * only if s_addr is INADDR_ANY.
478                  */
479                 err = -EAFNOSUPPORT;
480                 if (addr->sin_family != AF_UNSPEC ||
481                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
482                         goto out;
483         }
484
485         chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
486
487         /* Not specified by any standard per-se, however it breaks too
488          * many applications when removed.  It is unfortunate since
489          * allowing applications to make a non-local bind solves
490          * several problems with systems using dynamic addressing.
491          * (ie. your servers still start up even if your ISDN link
492          *  is temporarily down)
493          */
494         err = -EADDRNOTAVAIL;
495         if (!sysctl_ip_nonlocal_bind &&
496             !(inet->freebind || inet->transparent) &&
497             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
498             chk_addr_ret != RTN_LOCAL &&
499             chk_addr_ret != RTN_MULTICAST &&
500             chk_addr_ret != RTN_BROADCAST)
501                 goto out;
502
503         snum = ntohs(addr->sin_port);
504         err = -EACCES;
505         if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
506                 goto out;
507
508         /*      We keep a pair of addresses. rcv_saddr is the one
509          *      used by hash lookups, and saddr is used for transmit.
510          *
511          *      In the BSD API these are the same except where it
512          *      would be illegal to use them (multicast/broadcast) in
513          *      which case the sending device address is used.
514          */
515         lock_sock(sk);
516
517         /* Check these errors (active socket, double bind). */
518         err = -EINVAL;
519         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
520                 goto out_release_sock;
521
522         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
523         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
524                 inet->inet_saddr = 0;  /* Use device */
525
526         /* Make sure we are allowed to bind here. */
527         if (sk->sk_prot->get_port(sk, snum)) {
528                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
529                 err = -EADDRINUSE;
530                 goto out_release_sock;
531         }
532
533         if (inet->inet_rcv_saddr)
534                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
535         if (snum)
536                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
537         inet->inet_sport = htons(inet->inet_num);
538         inet->inet_daddr = 0;
539         inet->inet_dport = 0;
540         sk_dst_reset(sk);
541         err = 0;
542 out_release_sock:
543         release_sock(sk);
544 out:
545         return err;
546 }
547 EXPORT_SYMBOL(inet_bind);
548
549 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
550                        int addr_len, int flags)
551 {
552         struct sock *sk = sock->sk;
553
554         if (addr_len < sizeof(uaddr->sa_family))
555                 return -EINVAL;
556         if (uaddr->sa_family == AF_UNSPEC)
557                 return sk->sk_prot->disconnect(sk, flags);
558
559         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
560                 return -EAGAIN;
561         return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
562 }
563 EXPORT_SYMBOL(inet_dgram_connect);
564
565 static long inet_wait_for_connect(struct sock *sk, long timeo)
566 {
567         DEFINE_WAIT(wait);
568
569         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
570
571         /* Basic assumption: if someone sets sk->sk_err, he _must_
572          * change state of the socket from TCP_SYN_*.
573          * Connect() does not allow to get error notifications
574          * without closing the socket.
575          */
576         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
577                 release_sock(sk);
578                 timeo = schedule_timeout(timeo);
579                 lock_sock(sk);
580                 if (signal_pending(current) || !timeo)
581                         break;
582                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
583         }
584         finish_wait(sk_sleep(sk), &wait);
585         return timeo;
586 }
587
588 /*
589  *      Connect to a remote host. There is regrettably still a little
590  *      TCP 'magic' in here.
591  */
592 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
593                         int addr_len, int flags)
594 {
595         struct sock *sk = sock->sk;
596         int err;
597         long timeo;
598
599         if (addr_len < sizeof(uaddr->sa_family))
600                 return -EINVAL;
601
602         lock_sock(sk);
603
604         if (uaddr->sa_family == AF_UNSPEC) {
605                 err = sk->sk_prot->disconnect(sk, flags);
606                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
607                 goto out;
608         }
609
610         switch (sock->state) {
611         default:
612                 err = -EINVAL;
613                 goto out;
614         case SS_CONNECTED:
615                 err = -EISCONN;
616                 goto out;
617         case SS_CONNECTING:
618                 err = -EALREADY;
619                 /* Fall out of switch with err, set for this state */
620                 break;
621         case SS_UNCONNECTED:
622                 err = -EISCONN;
623                 if (sk->sk_state != TCP_CLOSE)
624                         goto out;
625
626                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
627                 if (err < 0)
628                         goto out;
629
630                 sock->state = SS_CONNECTING;
631
632                 /* Just entered SS_CONNECTING state; the only
633                  * difference is that return value in non-blocking
634                  * case is EINPROGRESS, rather than EALREADY.
635                  */
636                 err = -EINPROGRESS;
637                 break;
638         }
639
640         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
641
642         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
643                 /* Error code is set above */
644                 if (!timeo || !inet_wait_for_connect(sk, timeo))
645                         goto out;
646
647                 err = sock_intr_errno(timeo);
648                 if (signal_pending(current))
649                         goto out;
650         }
651
652         /* Connection was closed by RST, timeout, ICMP error
653          * or another process disconnected us.
654          */
655         if (sk->sk_state == TCP_CLOSE)
656                 goto sock_error;
657
658         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
659          * and error was received after socket entered established state.
660          * Hence, it is handled normally after connect() return successfully.
661          */
662
663         sock->state = SS_CONNECTED;
664         err = 0;
665 out:
666         release_sock(sk);
667         return err;
668
669 sock_error:
670         err = sock_error(sk) ? : -ECONNABORTED;
671         sock->state = SS_UNCONNECTED;
672         if (sk->sk_prot->disconnect(sk, flags))
673                 sock->state = SS_DISCONNECTING;
674         goto out;
675 }
676 EXPORT_SYMBOL(inet_stream_connect);
677
678 /*
679  *      Accept a pending connection. The TCP layer now gives BSD semantics.
680  */
681
682 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
683 {
684         struct sock *sk1 = sock->sk;
685         int err = -EINVAL;
686         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
687
688         if (!sk2)
689                 goto do_err;
690
691         lock_sock(sk2);
692
693         sock_rps_record_flow(sk2);
694         WARN_ON(!((1 << sk2->sk_state) &
695                   (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
696
697         sock_graft(sk2, newsock);
698
699         newsock->state = SS_CONNECTED;
700         err = 0;
701         release_sock(sk2);
702 do_err:
703         return err;
704 }
705 EXPORT_SYMBOL(inet_accept);
706
707
708 /*
709  *      This does both peername and sockname.
710  */
711 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
712                         int *uaddr_len, int peer)
713 {
714         struct sock *sk         = sock->sk;
715         struct inet_sock *inet  = inet_sk(sk);
716         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
717
718         sin->sin_family = AF_INET;
719         if (peer) {
720                 if (!inet->inet_dport ||
721                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
722                      peer == 1))
723                         return -ENOTCONN;
724                 sin->sin_port = inet->inet_dport;
725                 sin->sin_addr.s_addr = inet->inet_daddr;
726         } else {
727                 __be32 addr = inet->inet_rcv_saddr;
728                 if (!addr)
729                         addr = inet->inet_saddr;
730                 sin->sin_port = inet->inet_sport;
731                 sin->sin_addr.s_addr = addr;
732         }
733         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
734         *uaddr_len = sizeof(*sin);
735         return 0;
736 }
737 EXPORT_SYMBOL(inet_getname);
738
739 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
740                  size_t size)
741 {
742         struct sock *sk = sock->sk;
743
744         sock_rps_record_flow(sk);
745
746         /* We may need to bind the socket. */
747         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
748             inet_autobind(sk))
749                 return -EAGAIN;
750
751         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
752 }
753 EXPORT_SYMBOL(inet_sendmsg);
754
755 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
756                       size_t size, int flags)
757 {
758         struct sock *sk = sock->sk;
759
760         sock_rps_record_flow(sk);
761
762         /* We may need to bind the socket. */
763         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
764             inet_autobind(sk))
765                 return -EAGAIN;
766
767         if (sk->sk_prot->sendpage)
768                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
769         return sock_no_sendpage(sock, page, offset, size, flags);
770 }
771 EXPORT_SYMBOL(inet_sendpage);
772
773 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
774                  size_t size, int flags)
775 {
776         struct sock *sk = sock->sk;
777         int addr_len = 0;
778         int err;
779
780         sock_rps_record_flow(sk);
781
782         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
783                                    flags & ~MSG_DONTWAIT, &addr_len);
784         if (err >= 0)
785                 msg->msg_namelen = addr_len;
786         return err;
787 }
788 EXPORT_SYMBOL(inet_recvmsg);
789
790 int inet_shutdown(struct socket *sock, int how)
791 {
792         struct sock *sk = sock->sk;
793         int err = 0;
794
795         /* This should really check to make sure
796          * the socket is a TCP socket. (WHY AC...)
797          */
798         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
799                        1->2 bit 2 snds.
800                        2->3 */
801         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
802                 return -EINVAL;
803
804         lock_sock(sk);
805         if (sock->state == SS_CONNECTING) {
806                 if ((1 << sk->sk_state) &
807                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
808                         sock->state = SS_DISCONNECTING;
809                 else
810                         sock->state = SS_CONNECTED;
811         }
812
813         switch (sk->sk_state) {
814         case TCP_CLOSE:
815                 err = -ENOTCONN;
816                 /* Hack to wake up other listeners, who can poll for
817                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
818         default:
819                 sk->sk_shutdown |= how;
820                 if (sk->sk_prot->shutdown)
821                         sk->sk_prot->shutdown(sk, how);
822                 break;
823
824         /* Remaining two branches are temporary solution for missing
825          * close() in multithreaded environment. It is _not_ a good idea,
826          * but we have no choice until close() is repaired at VFS level.
827          */
828         case TCP_LISTEN:
829                 if (!(how & RCV_SHUTDOWN))
830                         break;
831                 /* Fall through */
832         case TCP_SYN_SENT:
833                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
834                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
835                 break;
836         }
837
838         /* Wake up anyone sleeping in poll. */
839         sk->sk_state_change(sk);
840         release_sock(sk);
841         return err;
842 }
843 EXPORT_SYMBOL(inet_shutdown);
844
845 /*
846  *      ioctl() calls you can issue on an INET socket. Most of these are
847  *      device configuration and stuff and very rarely used. Some ioctls
848  *      pass on to the socket itself.
849  *
850  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
851  *      loads the devconfigure module does its configuring and unloads it.
852  *      There's a good 20K of config code hanging around the kernel.
853  */
854
855 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
856 {
857         struct sock *sk = sock->sk;
858         int err = 0;
859         struct net *net = sock_net(sk);
860
861         switch (cmd) {
862         case SIOCGSTAMP:
863                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
864                 break;
865         case SIOCGSTAMPNS:
866                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
867                 break;
868         case SIOCADDRT:
869         case SIOCDELRT:
870         case SIOCRTMSG:
871                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
872                 break;
873         case SIOCDARP:
874         case SIOCGARP:
875         case SIOCSARP:
876                 err = arp_ioctl(net, cmd, (void __user *)arg);
877                 break;
878         case SIOCGIFADDR:
879         case SIOCSIFADDR:
880         case SIOCGIFBRDADDR:
881         case SIOCSIFBRDADDR:
882         case SIOCGIFNETMASK:
883         case SIOCSIFNETMASK:
884         case SIOCGIFDSTADDR:
885         case SIOCSIFDSTADDR:
886         case SIOCSIFPFLAGS:
887         case SIOCGIFPFLAGS:
888         case SIOCSIFFLAGS:
889                 err = devinet_ioctl(net, cmd, (void __user *)arg);
890                 break;
891         default:
892                 if (sk->sk_prot->ioctl)
893                         err = sk->sk_prot->ioctl(sk, cmd, arg);
894                 else
895                         err = -ENOIOCTLCMD;
896                 break;
897         }
898         return err;
899 }
900 EXPORT_SYMBOL(inet_ioctl);
901
902 #ifdef CONFIG_COMPAT
903 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
904 {
905         struct sock *sk = sock->sk;
906         int err = -ENOIOCTLCMD;
907
908         if (sk->sk_prot->compat_ioctl)
909                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
910
911         return err;
912 }
913 #endif
914
915 const struct proto_ops inet_stream_ops = {
916         .family            = PF_INET,
917         .owner             = THIS_MODULE,
918         .release           = inet_release,
919         .bind              = inet_bind,
920         .connect           = inet_stream_connect,
921         .socketpair        = sock_no_socketpair,
922         .accept            = inet_accept,
923         .getname           = inet_getname,
924         .poll              = tcp_poll,
925         .ioctl             = inet_ioctl,
926         .listen            = inet_listen,
927         .shutdown          = inet_shutdown,
928         .setsockopt        = sock_common_setsockopt,
929         .getsockopt        = sock_common_getsockopt,
930         .sendmsg           = inet_sendmsg,
931         .recvmsg           = inet_recvmsg,
932         .mmap              = sock_no_mmap,
933         .sendpage          = inet_sendpage,
934         .splice_read       = tcp_splice_read,
935 #ifdef CONFIG_COMPAT
936         .compat_setsockopt = compat_sock_common_setsockopt,
937         .compat_getsockopt = compat_sock_common_getsockopt,
938         .compat_ioctl      = inet_compat_ioctl,
939 #endif
940 };
941 EXPORT_SYMBOL(inet_stream_ops);
942
943 const struct proto_ops inet_dgram_ops = {
944         .family            = PF_INET,
945         .owner             = THIS_MODULE,
946         .release           = inet_release,
947         .bind              = inet_bind,
948         .connect           = inet_dgram_connect,
949         .socketpair        = sock_no_socketpair,
950         .accept            = sock_no_accept,
951         .getname           = inet_getname,
952         .poll              = udp_poll,
953         .ioctl             = inet_ioctl,
954         .listen            = sock_no_listen,
955         .shutdown          = inet_shutdown,
956         .setsockopt        = sock_common_setsockopt,
957         .getsockopt        = sock_common_getsockopt,
958         .sendmsg           = inet_sendmsg,
959         .recvmsg           = inet_recvmsg,
960         .mmap              = sock_no_mmap,
961         .sendpage          = inet_sendpage,
962 #ifdef CONFIG_COMPAT
963         .compat_setsockopt = compat_sock_common_setsockopt,
964         .compat_getsockopt = compat_sock_common_getsockopt,
965         .compat_ioctl      = inet_compat_ioctl,
966 #endif
967 };
968 EXPORT_SYMBOL(inet_dgram_ops);
969
970 /*
971  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
972  * udp_poll
973  */
974 static const struct proto_ops inet_sockraw_ops = {
975         .family            = PF_INET,
976         .owner             = THIS_MODULE,
977         .release           = inet_release,
978         .bind              = inet_bind,
979         .connect           = inet_dgram_connect,
980         .socketpair        = sock_no_socketpair,
981         .accept            = sock_no_accept,
982         .getname           = inet_getname,
983         .poll              = datagram_poll,
984         .ioctl             = inet_ioctl,
985         .listen            = sock_no_listen,
986         .shutdown          = inet_shutdown,
987         .setsockopt        = sock_common_setsockopt,
988         .getsockopt        = sock_common_getsockopt,
989         .sendmsg           = inet_sendmsg,
990         .recvmsg           = inet_recvmsg,
991         .mmap              = sock_no_mmap,
992         .sendpage          = inet_sendpage,
993 #ifdef CONFIG_COMPAT
994         .compat_setsockopt = compat_sock_common_setsockopt,
995         .compat_getsockopt = compat_sock_common_getsockopt,
996         .compat_ioctl      = inet_compat_ioctl,
997 #endif
998 };
999
1000 static const struct net_proto_family inet_family_ops = {
1001         .family = PF_INET,
1002         .create = inet_create,
1003         .owner  = THIS_MODULE,
1004 };
1005
1006 /* Upon startup we insert all the elements in inetsw_array[] into
1007  * the linked list inetsw.
1008  */
1009 static struct inet_protosw inetsw_array[] =
1010 {
1011         {
1012                 .type =       SOCK_STREAM,
1013                 .protocol =   IPPROTO_TCP,
1014                 .prot =       &tcp_prot,
1015                 .ops =        &inet_stream_ops,
1016                 .no_check =   0,
1017                 .flags =      INET_PROTOSW_PERMANENT |
1018                               INET_PROTOSW_ICSK,
1019         },
1020
1021         {
1022                 .type =       SOCK_DGRAM,
1023                 .protocol =   IPPROTO_UDP,
1024                 .prot =       &udp_prot,
1025                 .ops =        &inet_dgram_ops,
1026                 .no_check =   UDP_CSUM_DEFAULT,
1027                 .flags =      INET_PROTOSW_PERMANENT,
1028        },
1029
1030        {
1031                 .type =       SOCK_DGRAM,
1032                 .protocol =   IPPROTO_ICMP,
1033                 .prot =       &ping_prot,
1034                 .ops =        &inet_dgram_ops,
1035                 .no_check =   UDP_CSUM_DEFAULT,
1036                 .flags =      INET_PROTOSW_REUSE,
1037        },
1038
1039        {
1040                .type =       SOCK_RAW,
1041                .protocol =   IPPROTO_IP,        /* wild card */
1042                .prot =       &raw_prot,
1043                .ops =        &inet_sockraw_ops,
1044                .no_check =   UDP_CSUM_DEFAULT,
1045                .flags =      INET_PROTOSW_REUSE,
1046        }
1047 };
1048
1049 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1050
1051 void inet_register_protosw(struct inet_protosw *p)
1052 {
1053         struct list_head *lh;
1054         struct inet_protosw *answer;
1055         int protocol = p->protocol;
1056         struct list_head *last_perm;
1057
1058         spin_lock_bh(&inetsw_lock);
1059
1060         if (p->type >= SOCK_MAX)
1061                 goto out_illegal;
1062
1063         /* If we are trying to override a permanent protocol, bail. */
1064         answer = NULL;
1065         last_perm = &inetsw[p->type];
1066         list_for_each(lh, &inetsw[p->type]) {
1067                 answer = list_entry(lh, struct inet_protosw, list);
1068
1069                 /* Check only the non-wild match. */
1070                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1071                         if (protocol == answer->protocol)
1072                                 break;
1073                         last_perm = lh;
1074                 }
1075
1076                 answer = NULL;
1077         }
1078         if (answer)
1079                 goto out_permanent;
1080
1081         /* Add the new entry after the last permanent entry if any, so that
1082          * the new entry does not override a permanent entry when matched with
1083          * a wild-card protocol. But it is allowed to override any existing
1084          * non-permanent entry.  This means that when we remove this entry, the
1085          * system automatically returns to the old behavior.
1086          */
1087         list_add_rcu(&p->list, last_perm);
1088 out:
1089         spin_unlock_bh(&inetsw_lock);
1090
1091         return;
1092
1093 out_permanent:
1094         pr_err("Attempt to override permanent protocol %d\n", protocol);
1095         goto out;
1096
1097 out_illegal:
1098         pr_err("Ignoring attempt to register invalid socket type %d\n",
1099                p->type);
1100         goto out;
1101 }
1102 EXPORT_SYMBOL(inet_register_protosw);
1103
1104 void inet_unregister_protosw(struct inet_protosw *p)
1105 {
1106         if (INET_PROTOSW_PERMANENT & p->flags) {
1107                 pr_err("Attempt to unregister permanent protocol %d\n",
1108                        p->protocol);
1109         } else {
1110                 spin_lock_bh(&inetsw_lock);
1111                 list_del_rcu(&p->list);
1112                 spin_unlock_bh(&inetsw_lock);
1113
1114                 synchronize_net();
1115         }
1116 }
1117 EXPORT_SYMBOL(inet_unregister_protosw);
1118
1119 /*
1120  *      Shall we try to damage output packets if routing dev changes?
1121  */
1122
1123 int sysctl_ip_dynaddr __read_mostly;
1124
1125 static int inet_sk_reselect_saddr(struct sock *sk)
1126 {
1127         struct inet_sock *inet = inet_sk(sk);
1128         __be32 old_saddr = inet->inet_saddr;
1129         __be32 daddr = inet->inet_daddr;
1130         struct flowi4 *fl4;
1131         struct rtable *rt;
1132         __be32 new_saddr;
1133         struct ip_options_rcu *inet_opt;
1134
1135         inet_opt = rcu_dereference_protected(inet->inet_opt,
1136                                              sock_owned_by_user(sk));
1137         if (inet_opt && inet_opt->opt.srr)
1138                 daddr = inet_opt->opt.faddr;
1139
1140         /* Query new route. */
1141         fl4 = &inet->cork.fl.u.ip4;
1142         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1143                               sk->sk_bound_dev_if, sk->sk_protocol,
1144                               inet->inet_sport, inet->inet_dport, sk, false);
1145         if (IS_ERR(rt))
1146                 return PTR_ERR(rt);
1147
1148         sk_setup_caps(sk, &rt->dst);
1149
1150         new_saddr = fl4->saddr;
1151
1152         if (new_saddr == old_saddr)
1153                 return 0;
1154
1155         if (sysctl_ip_dynaddr > 1) {
1156                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1157                         __func__, &old_saddr, &new_saddr);
1158         }
1159
1160         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1161
1162         /*
1163          * XXX The only one ugly spot where we need to
1164          * XXX really change the sockets identity after
1165          * XXX it has entered the hashes. -DaveM
1166          *
1167          * Besides that, it does not check for connection
1168          * uniqueness. Wait for troubles.
1169          */
1170         __sk_prot_rehash(sk);
1171         return 0;
1172 }
1173
1174 int inet_sk_rebuild_header(struct sock *sk)
1175 {
1176         struct inet_sock *inet = inet_sk(sk);
1177         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1178         __be32 daddr;
1179         struct ip_options_rcu *inet_opt;
1180         struct flowi4 *fl4;
1181         int err;
1182
1183         /* Route is OK, nothing to do. */
1184         if (rt)
1185                 return 0;
1186
1187         /* Reroute. */
1188         rcu_read_lock();
1189         inet_opt = rcu_dereference(inet->inet_opt);
1190         daddr = inet->inet_daddr;
1191         if (inet_opt && inet_opt->opt.srr)
1192                 daddr = inet_opt->opt.faddr;
1193         rcu_read_unlock();
1194         fl4 = &inet->cork.fl.u.ip4;
1195         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1196                                    inet->inet_dport, inet->inet_sport,
1197                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1198                                    sk->sk_bound_dev_if);
1199         if (!IS_ERR(rt)) {
1200                 err = 0;
1201                 sk_setup_caps(sk, &rt->dst);
1202         } else {
1203                 err = PTR_ERR(rt);
1204
1205                 /* Routing failed... */
1206                 sk->sk_route_caps = 0;
1207                 /*
1208                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1209                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1210                  */
1211                 if (!sysctl_ip_dynaddr ||
1212                     sk->sk_state != TCP_SYN_SENT ||
1213                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1214                     (err = inet_sk_reselect_saddr(sk)) != 0)
1215                         sk->sk_err_soft = -err;
1216         }
1217
1218         return err;
1219 }
1220 EXPORT_SYMBOL(inet_sk_rebuild_header);
1221
1222 static int inet_gso_send_check(struct sk_buff *skb)
1223 {
1224         const struct iphdr *iph;
1225         const struct net_protocol *ops;
1226         int proto;
1227         int ihl;
1228         int err = -EINVAL;
1229
1230         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1231                 goto out;
1232
1233         iph = ip_hdr(skb);
1234         ihl = iph->ihl * 4;
1235         if (ihl < sizeof(*iph))
1236                 goto out;
1237
1238         if (unlikely(!pskb_may_pull(skb, ihl)))
1239                 goto out;
1240
1241         __skb_pull(skb, ihl);
1242         skb_reset_transport_header(skb);
1243         iph = ip_hdr(skb);
1244         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1245         err = -EPROTONOSUPPORT;
1246
1247         rcu_read_lock();
1248         ops = rcu_dereference(inet_protos[proto]);
1249         if (likely(ops && ops->gso_send_check))
1250                 err = ops->gso_send_check(skb);
1251         rcu_read_unlock();
1252
1253 out:
1254         return err;
1255 }
1256
1257 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1258         netdev_features_t features)
1259 {
1260         struct sk_buff *segs = ERR_PTR(-EINVAL);
1261         struct iphdr *iph;
1262         const struct net_protocol *ops;
1263         int proto;
1264         int ihl;
1265         int id;
1266         unsigned int offset = 0;
1267
1268         if (!(features & NETIF_F_V4_CSUM))
1269                 features &= ~NETIF_F_SG;
1270
1271         if (unlikely(skb_shinfo(skb)->gso_type &
1272                      ~(SKB_GSO_TCPV4 |
1273                        SKB_GSO_UDP |
1274                        SKB_GSO_DODGY |
1275                        SKB_GSO_TCP_ECN |
1276                        0)))
1277                 goto out;
1278
1279         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1280                 goto out;
1281
1282         iph = ip_hdr(skb);
1283         ihl = iph->ihl * 4;
1284         if (ihl < sizeof(*iph))
1285                 goto out;
1286
1287         if (unlikely(!pskb_may_pull(skb, ihl)))
1288                 goto out;
1289
1290         __skb_pull(skb, ihl);
1291         skb_reset_transport_header(skb);
1292         iph = ip_hdr(skb);
1293         id = ntohs(iph->id);
1294         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1295         segs = ERR_PTR(-EPROTONOSUPPORT);
1296
1297         rcu_read_lock();
1298         ops = rcu_dereference(inet_protos[proto]);
1299         if (likely(ops && ops->gso_segment))
1300                 segs = ops->gso_segment(skb, features);
1301         rcu_read_unlock();
1302
1303         if (!segs || IS_ERR(segs))
1304                 goto out;
1305
1306         skb = segs;
1307         do {
1308                 iph = ip_hdr(skb);
1309                 if (proto == IPPROTO_UDP) {
1310                         iph->id = htons(id);
1311                         iph->frag_off = htons(offset >> 3);
1312                         if (skb->next != NULL)
1313                                 iph->frag_off |= htons(IP_MF);
1314                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1315                 } else
1316                         iph->id = htons(id++);
1317                 iph->tot_len = htons(skb->len - skb->mac_len);
1318                 iph->check = 0;
1319                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1320         } while ((skb = skb->next));
1321
1322 out:
1323         return segs;
1324 }
1325
1326 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1327                                          struct sk_buff *skb)
1328 {
1329         const struct net_protocol *ops;
1330         struct sk_buff **pp = NULL;
1331         struct sk_buff *p;
1332         const struct iphdr *iph;
1333         unsigned int hlen;
1334         unsigned int off;
1335         unsigned int id;
1336         int flush = 1;
1337         int proto;
1338
1339         off = skb_gro_offset(skb);
1340         hlen = off + sizeof(*iph);
1341         iph = skb_gro_header_fast(skb, off);
1342         if (skb_gro_header_hard(skb, hlen)) {
1343                 iph = skb_gro_header_slow(skb, hlen, off);
1344                 if (unlikely(!iph))
1345                         goto out;
1346         }
1347
1348         proto = iph->protocol & (MAX_INET_PROTOS - 1);
1349
1350         rcu_read_lock();
1351         ops = rcu_dereference(inet_protos[proto]);
1352         if (!ops || !ops->gro_receive)
1353                 goto out_unlock;
1354
1355         if (*(u8 *)iph != 0x45)
1356                 goto out_unlock;
1357
1358         if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1359                 goto out_unlock;
1360
1361         id = ntohl(*(__be32 *)&iph->id);
1362         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1363         id >>= 16;
1364
1365         for (p = *head; p; p = p->next) {
1366                 struct iphdr *iph2;
1367
1368                 if (!NAPI_GRO_CB(p)->same_flow)
1369                         continue;
1370
1371                 iph2 = ip_hdr(p);
1372
1373                 if ((iph->protocol ^ iph2->protocol) |
1374                     (iph->tos ^ iph2->tos) |
1375                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1376                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1377                         NAPI_GRO_CB(p)->same_flow = 0;
1378                         continue;
1379                 }
1380
1381                 /* All fields must match except length and checksum. */
1382                 NAPI_GRO_CB(p)->flush |=
1383                         (iph->ttl ^ iph2->ttl) |
1384                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1385
1386                 NAPI_GRO_CB(p)->flush |= flush;
1387         }
1388
1389         NAPI_GRO_CB(skb)->flush |= flush;
1390         skb_gro_pull(skb, sizeof(*iph));
1391         skb_set_transport_header(skb, skb_gro_offset(skb));
1392
1393         pp = ops->gro_receive(head, skb);
1394
1395 out_unlock:
1396         rcu_read_unlock();
1397
1398 out:
1399         NAPI_GRO_CB(skb)->flush |= flush;
1400
1401         return pp;
1402 }
1403
1404 static int inet_gro_complete(struct sk_buff *skb)
1405 {
1406         const struct net_protocol *ops;
1407         struct iphdr *iph = ip_hdr(skb);
1408         int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1409         int err = -ENOSYS;
1410         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1411
1412         csum_replace2(&iph->check, iph->tot_len, newlen);
1413         iph->tot_len = newlen;
1414
1415         rcu_read_lock();
1416         ops = rcu_dereference(inet_protos[proto]);
1417         if (WARN_ON(!ops || !ops->gro_complete))
1418                 goto out_unlock;
1419
1420         err = ops->gro_complete(skb);
1421
1422 out_unlock:
1423         rcu_read_unlock();
1424
1425         return err;
1426 }
1427
1428 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1429                          unsigned short type, unsigned char protocol,
1430                          struct net *net)
1431 {
1432         struct socket *sock;
1433         int rc = sock_create_kern(family, type, protocol, &sock);
1434
1435         if (rc == 0) {
1436                 *sk = sock->sk;
1437                 (*sk)->sk_allocation = GFP_ATOMIC;
1438                 /*
1439                  * Unhash it so that IP input processing does not even see it,
1440                  * we do not wish this socket to see incoming packets.
1441                  */
1442                 (*sk)->sk_prot->unhash(*sk);
1443
1444                 sk_change_net(*sk, net);
1445         }
1446         return rc;
1447 }
1448 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1449
1450 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1451 {
1452         unsigned long res = 0;
1453         int i, j;
1454
1455         for_each_possible_cpu(i) {
1456                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1457                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1458         }
1459         return res;
1460 }
1461 EXPORT_SYMBOL_GPL(snmp_fold_field);
1462
1463 #if BITS_PER_LONG==32
1464
1465 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1466 {
1467         u64 res = 0;
1468         int cpu;
1469
1470         for_each_possible_cpu(cpu) {
1471                 void *bhptr;
1472                 struct u64_stats_sync *syncp;
1473                 u64 v;
1474                 unsigned int start;
1475
1476                 bhptr = per_cpu_ptr(mib[0], cpu);
1477                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1478                 do {
1479                         start = u64_stats_fetch_begin_bh(syncp);
1480                         v = *(((u64 *) bhptr) + offt);
1481                 } while (u64_stats_fetch_retry_bh(syncp, start));
1482
1483                 res += v;
1484         }
1485         return res;
1486 }
1487 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1488 #endif
1489
1490 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1491 {
1492         BUG_ON(ptr == NULL);
1493         ptr[0] = __alloc_percpu(mibsize, align);
1494         if (!ptr[0])
1495                 return -ENOMEM;
1496 #if SNMP_ARRAY_SZ == 2
1497         ptr[1] = __alloc_percpu(mibsize, align);
1498         if (!ptr[1]) {
1499                 free_percpu(ptr[0]);
1500                 ptr[0] = NULL;
1501                 return -ENOMEM;
1502         }
1503 #endif
1504         return 0;
1505 }
1506 EXPORT_SYMBOL_GPL(snmp_mib_init);
1507
1508 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1509 {
1510         int i;
1511
1512         BUG_ON(ptr == NULL);
1513         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1514                 free_percpu(ptr[i]);
1515                 ptr[i] = NULL;
1516         }
1517 }
1518 EXPORT_SYMBOL_GPL(snmp_mib_free);
1519
1520 #ifdef CONFIG_IP_MULTICAST
1521 static const struct net_protocol igmp_protocol = {
1522         .handler =      igmp_rcv,
1523         .netns_ok =     1,
1524 };
1525 #endif
1526
1527 static const struct net_protocol tcp_protocol = {
1528         .handler =      tcp_v4_rcv,
1529         .err_handler =  tcp_v4_err,
1530         .gso_send_check = tcp_v4_gso_send_check,
1531         .gso_segment =  tcp_tso_segment,
1532         .gro_receive =  tcp4_gro_receive,
1533         .gro_complete = tcp4_gro_complete,
1534         .no_policy =    1,
1535         .netns_ok =     1,
1536 };
1537
1538 static const struct net_protocol udp_protocol = {
1539         .handler =      udp_rcv,
1540         .err_handler =  udp_err,
1541         .gso_send_check = udp4_ufo_send_check,
1542         .gso_segment = udp4_ufo_fragment,
1543         .no_policy =    1,
1544         .netns_ok =     1,
1545 };
1546
1547 static const struct net_protocol icmp_protocol = {
1548         .handler =      icmp_rcv,
1549         .err_handler =  ping_err,
1550         .no_policy =    1,
1551         .netns_ok =     1,
1552 };
1553
1554 static __net_init int ipv4_mib_init_net(struct net *net)
1555 {
1556         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1557                           sizeof(struct tcp_mib),
1558                           __alignof__(struct tcp_mib)) < 0)
1559                 goto err_tcp_mib;
1560         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1561                           sizeof(struct ipstats_mib),
1562                           __alignof__(struct ipstats_mib)) < 0)
1563                 goto err_ip_mib;
1564         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1565                           sizeof(struct linux_mib),
1566                           __alignof__(struct linux_mib)) < 0)
1567                 goto err_net_mib;
1568         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1569                           sizeof(struct udp_mib),
1570                           __alignof__(struct udp_mib)) < 0)
1571                 goto err_udp_mib;
1572         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1573                           sizeof(struct udp_mib),
1574                           __alignof__(struct udp_mib)) < 0)
1575                 goto err_udplite_mib;
1576         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1577                           sizeof(struct icmp_mib),
1578                           __alignof__(struct icmp_mib)) < 0)
1579                 goto err_icmp_mib;
1580         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1581                                               GFP_KERNEL);
1582         if (!net->mib.icmpmsg_statistics)
1583                 goto err_icmpmsg_mib;
1584
1585         tcp_mib_init(net);
1586         return 0;
1587
1588 err_icmpmsg_mib:
1589         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1590 err_icmp_mib:
1591         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1592 err_udplite_mib:
1593         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1594 err_udp_mib:
1595         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1596 err_net_mib:
1597         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1598 err_ip_mib:
1599         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1600 err_tcp_mib:
1601         return -ENOMEM;
1602 }
1603
1604 static __net_exit void ipv4_mib_exit_net(struct net *net)
1605 {
1606         kfree(net->mib.icmpmsg_statistics);
1607         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1608         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1609         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1610         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1611         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1612         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1613 }
1614
1615 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1616         .init = ipv4_mib_init_net,
1617         .exit = ipv4_mib_exit_net,
1618 };
1619
1620 static int __init init_ipv4_mibs(void)
1621 {
1622         return register_pernet_subsys(&ipv4_mib_ops);
1623 }
1624
1625 static int ipv4_proc_init(void);
1626
1627 /*
1628  *      IP protocol layer initialiser
1629  */
1630
1631 static struct packet_type ip_packet_type __read_mostly = {
1632         .type = cpu_to_be16(ETH_P_IP),
1633         .func = ip_rcv,
1634         .gso_send_check = inet_gso_send_check,
1635         .gso_segment = inet_gso_segment,
1636         .gro_receive = inet_gro_receive,
1637         .gro_complete = inet_gro_complete,
1638 };
1639
1640 static int __init inet_init(void)
1641 {
1642         struct sk_buff *dummy_skb;
1643         struct inet_protosw *q;
1644         struct list_head *r;
1645         int rc = -EINVAL;
1646
1647         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1648
1649         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1650         if (!sysctl_local_reserved_ports)
1651                 goto out;
1652
1653         rc = proto_register(&tcp_prot, 1);
1654         if (rc)
1655                 goto out_free_reserved_ports;
1656
1657         rc = proto_register(&udp_prot, 1);
1658         if (rc)
1659                 goto out_unregister_tcp_proto;
1660
1661         rc = proto_register(&raw_prot, 1);
1662         if (rc)
1663                 goto out_unregister_udp_proto;
1664
1665         rc = proto_register(&ping_prot, 1);
1666         if (rc)
1667                 goto out_unregister_raw_proto;
1668
1669         /*
1670          *      Tell SOCKET that we are alive...
1671          */
1672
1673         (void)sock_register(&inet_family_ops);
1674
1675 #ifdef CONFIG_SYSCTL
1676         ip_static_sysctl_init();
1677 #endif
1678
1679         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1680
1681         /*
1682          *      Add all the base protocols.
1683          */
1684
1685         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1686                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1687         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1688                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1689         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1690                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1691 #ifdef CONFIG_IP_MULTICAST
1692         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1693                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1694 #endif
1695
1696         /* Register the socket-side information for inet_create. */
1697         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1698                 INIT_LIST_HEAD(r);
1699
1700         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1701                 inet_register_protosw(q);
1702
1703         /*
1704          *      Set the ARP module up
1705          */
1706
1707         arp_init();
1708
1709         /*
1710          *      Set the IP module up
1711          */
1712
1713         ip_init();
1714
1715         tcp_v4_init();
1716
1717         /* Setup TCP slab cache for open requests. */
1718         tcp_init();
1719
1720         /* Setup UDP memory threshold */
1721         udp_init();
1722
1723         /* Add UDP-Lite (RFC 3828) */
1724         udplite4_register();
1725
1726         ping_init();
1727
1728         /*
1729          *      Set the ICMP layer up
1730          */
1731
1732         if (icmp_init() < 0)
1733                 panic("Failed to create the ICMP control socket.\n");
1734
1735         /*
1736          *      Initialise the multicast router
1737          */
1738 #if defined(CONFIG_IP_MROUTE)
1739         if (ip_mr_init())
1740                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1741 #endif
1742         /*
1743          *      Initialise per-cpu ipv4 mibs
1744          */
1745
1746         if (init_ipv4_mibs())
1747                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1748
1749         ipv4_proc_init();
1750
1751         ipfrag_init();
1752
1753         dev_add_pack(&ip_packet_type);
1754
1755         rc = 0;
1756 out:
1757         return rc;
1758 out_unregister_raw_proto:
1759         proto_unregister(&raw_prot);
1760 out_unregister_udp_proto:
1761         proto_unregister(&udp_prot);
1762 out_unregister_tcp_proto:
1763         proto_unregister(&tcp_prot);
1764 out_free_reserved_ports:
1765         kfree(sysctl_local_reserved_ports);
1766         goto out;
1767 }
1768
1769 fs_initcall(inet_init);
1770
1771 /* ------------------------------------------------------------------------ */
1772
1773 #ifdef CONFIG_PROC_FS
1774 static int __init ipv4_proc_init(void)
1775 {
1776         int rc = 0;
1777
1778         if (raw_proc_init())
1779                 goto out_raw;
1780         if (tcp4_proc_init())
1781                 goto out_tcp;
1782         if (udp4_proc_init())
1783                 goto out_udp;
1784         if (ping_proc_init())
1785                 goto out_ping;
1786         if (ip_misc_proc_init())
1787                 goto out_misc;
1788 out:
1789         return rc;
1790 out_misc:
1791         ping_proc_exit();
1792 out_ping:
1793         udp4_proc_exit();
1794 out_udp:
1795         tcp4_proc_exit();
1796 out_tcp:
1797         raw_proc_exit();
1798 out_raw:
1799         rc = -ENOMEM;
1800         goto out;
1801 }
1802
1803 #else /* CONFIG_PROC_FS */
1804 static int __init ipv4_proc_init(void)
1805 {
1806         return 0;
1807 }
1808 #endif /* CONFIG_PROC_FS */
1809
1810 MODULE_ALIAS_NETPROTO(PF_INET);
1811