]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - net/ipv4/af_inet.c
net: Fix possible wrong checksum generation.
[~shefty/rdma-dev.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         dst_release(sk->sk_rx_dst);
161         sk_refcnt_debug_dec(sk);
162 }
163 EXPORT_SYMBOL(inet_sock_destruct);
164
165 /*
166  *      The routines beyond this point handle the behaviour of an AF_INET
167  *      socket object. Mostly it punts to the subprotocols of IP to do
168  *      the work.
169  */
170
171 /*
172  *      Automatically bind an unbound socket.
173  */
174
175 static int inet_autobind(struct sock *sk)
176 {
177         struct inet_sock *inet;
178         /* We may need to bind the socket. */
179         lock_sock(sk);
180         inet = inet_sk(sk);
181         if (!inet->inet_num) {
182                 if (sk->sk_prot->get_port(sk, 0)) {
183                         release_sock(sk);
184                         return -EAGAIN;
185                 }
186                 inet->inet_sport = htons(inet->inet_num);
187         }
188         release_sock(sk);
189         return 0;
190 }
191
192 /*
193  *      Move a socket into listening state.
194  */
195 int inet_listen(struct socket *sock, int backlog)
196 {
197         struct sock *sk = sock->sk;
198         unsigned char old_state;
199         int err;
200
201         lock_sock(sk);
202
203         err = -EINVAL;
204         if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205                 goto out;
206
207         old_state = sk->sk_state;
208         if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209                 goto out;
210
211         /* Really, if the socket is already in listen state
212          * we can only allow the backlog to be adjusted.
213          */
214         if (old_state != TCP_LISTEN) {
215                 /* Check special setups for testing purpose to enable TFO w/o
216                  * requiring TCP_FASTOPEN sockopt.
217                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
218                  * Also fastopenq may already been allocated because this
219                  * socket was in TCP_LISTEN state previously but was
220                  * shutdown() (rather than close()).
221                  */
222                 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
223                     inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
224                         if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
225                                 err = fastopen_init_queue(sk, backlog);
226                         else if ((sysctl_tcp_fastopen &
227                                   TFO_SERVER_WO_SOCKOPT2) != 0)
228                                 err = fastopen_init_queue(sk,
229                                     ((uint)sysctl_tcp_fastopen) >> 16);
230                         else
231                                 err = 0;
232                         if (err)
233                                 goto out;
234                 }
235                 err = inet_csk_listen_start(sk, backlog);
236                 if (err)
237                         goto out;
238         }
239         sk->sk_max_ack_backlog = backlog;
240         err = 0;
241
242 out:
243         release_sock(sk);
244         return err;
245 }
246 EXPORT_SYMBOL(inet_listen);
247
248 u32 inet_ehash_secret __read_mostly;
249 EXPORT_SYMBOL(inet_ehash_secret);
250
251 /*
252  * inet_ehash_secret must be set exactly once
253  */
254 void build_ehash_secret(void)
255 {
256         u32 rnd;
257
258         do {
259                 get_random_bytes(&rnd, sizeof(rnd));
260         } while (rnd == 0);
261
262         cmpxchg(&inet_ehash_secret, 0, rnd);
263 }
264 EXPORT_SYMBOL(build_ehash_secret);
265
266 /*
267  *      Create an inet socket.
268  */
269
270 static int inet_create(struct net *net, struct socket *sock, int protocol,
271                        int kern)
272 {
273         struct sock *sk;
274         struct inet_protosw *answer;
275         struct inet_sock *inet;
276         struct proto *answer_prot;
277         unsigned char answer_flags;
278         char answer_no_check;
279         int try_loading_module = 0;
280         int err;
281
282         if (unlikely(!inet_ehash_secret))
283                 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
284                         build_ehash_secret();
285
286         sock->state = SS_UNCONNECTED;
287
288         /* Look for the requested type/protocol pair. */
289 lookup_protocol:
290         err = -ESOCKTNOSUPPORT;
291         rcu_read_lock();
292         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
293
294                 err = 0;
295                 /* Check the non-wild match. */
296                 if (protocol == answer->protocol) {
297                         if (protocol != IPPROTO_IP)
298                                 break;
299                 } else {
300                         /* Check for the two wild cases. */
301                         if (IPPROTO_IP == protocol) {
302                                 protocol = answer->protocol;
303                                 break;
304                         }
305                         if (IPPROTO_IP == answer->protocol)
306                                 break;
307                 }
308                 err = -EPROTONOSUPPORT;
309         }
310
311         if (unlikely(err)) {
312                 if (try_loading_module < 2) {
313                         rcu_read_unlock();
314                         /*
315                          * Be more specific, e.g. net-pf-2-proto-132-type-1
316                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
317                          */
318                         if (++try_loading_module == 1)
319                                 request_module("net-pf-%d-proto-%d-type-%d",
320                                                PF_INET, protocol, sock->type);
321                         /*
322                          * Fall back to generic, e.g. net-pf-2-proto-132
323                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
324                          */
325                         else
326                                 request_module("net-pf-%d-proto-%d",
327                                                PF_INET, protocol);
328                         goto lookup_protocol;
329                 } else
330                         goto out_rcu_unlock;
331         }
332
333         err = -EPERM;
334         if (sock->type == SOCK_RAW && !kern &&
335             !ns_capable(net->user_ns, CAP_NET_RAW))
336                 goto out_rcu_unlock;
337
338         sock->ops = answer->ops;
339         answer_prot = answer->prot;
340         answer_no_check = answer->no_check;
341         answer_flags = answer->flags;
342         rcu_read_unlock();
343
344         WARN_ON(answer_prot->slab == NULL);
345
346         err = -ENOBUFS;
347         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
348         if (sk == NULL)
349                 goto out;
350
351         err = 0;
352         sk->sk_no_check = answer_no_check;
353         if (INET_PROTOSW_REUSE & answer_flags)
354                 sk->sk_reuse = SK_CAN_REUSE;
355
356         inet = inet_sk(sk);
357         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
358
359         inet->nodefrag = 0;
360
361         if (SOCK_RAW == sock->type) {
362                 inet->inet_num = protocol;
363                 if (IPPROTO_RAW == protocol)
364                         inet->hdrincl = 1;
365         }
366
367         if (ipv4_config.no_pmtu_disc)
368                 inet->pmtudisc = IP_PMTUDISC_DONT;
369         else
370                 inet->pmtudisc = IP_PMTUDISC_WANT;
371
372         inet->inet_id = 0;
373
374         sock_init_data(sock, sk);
375
376         sk->sk_destruct    = inet_sock_destruct;
377         sk->sk_protocol    = protocol;
378         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
379
380         inet->uc_ttl    = -1;
381         inet->mc_loop   = 1;
382         inet->mc_ttl    = 1;
383         inet->mc_all    = 1;
384         inet->mc_index  = 0;
385         inet->mc_list   = NULL;
386         inet->rcv_tos   = 0;
387
388         sk_refcnt_debug_inc(sk);
389
390         if (inet->inet_num) {
391                 /* It assumes that any protocol which allows
392                  * the user to assign a number at socket
393                  * creation time automatically
394                  * shares.
395                  */
396                 inet->inet_sport = htons(inet->inet_num);
397                 /* Add to protocol hash chains. */
398                 sk->sk_prot->hash(sk);
399         }
400
401         if (sk->sk_prot->init) {
402                 err = sk->sk_prot->init(sk);
403                 if (err)
404                         sk_common_release(sk);
405         }
406 out:
407         return err;
408 out_rcu_unlock:
409         rcu_read_unlock();
410         goto out;
411 }
412
413
414 /*
415  *      The peer socket should always be NULL (or else). When we call this
416  *      function we are destroying the object and from then on nobody
417  *      should refer to it.
418  */
419 int inet_release(struct socket *sock)
420 {
421         struct sock *sk = sock->sk;
422
423         if (sk) {
424                 long timeout;
425
426                 sock_rps_reset_flow(sk);
427
428                 /* Applications forget to leave groups before exiting */
429                 ip_mc_drop_socket(sk);
430
431                 /* If linger is set, we don't return until the close
432                  * is complete.  Otherwise we return immediately. The
433                  * actually closing is done the same either way.
434                  *
435                  * If the close is due to the process exiting, we never
436                  * linger..
437                  */
438                 timeout = 0;
439                 if (sock_flag(sk, SOCK_LINGER) &&
440                     !(current->flags & PF_EXITING))
441                         timeout = sk->sk_lingertime;
442                 sock->sk = NULL;
443                 sk->sk_prot->close(sk, timeout);
444         }
445         return 0;
446 }
447 EXPORT_SYMBOL(inet_release);
448
449 /* It is off by default, see below. */
450 int sysctl_ip_nonlocal_bind __read_mostly;
451 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
452
453 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
454 {
455         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
456         struct sock *sk = sock->sk;
457         struct inet_sock *inet = inet_sk(sk);
458         struct net *net = sock_net(sk);
459         unsigned short snum;
460         int chk_addr_ret;
461         int err;
462
463         /* If the socket has its own bind function then use it. (RAW) */
464         if (sk->sk_prot->bind) {
465                 err = sk->sk_prot->bind(sk, uaddr, addr_len);
466                 goto out;
467         }
468         err = -EINVAL;
469         if (addr_len < sizeof(struct sockaddr_in))
470                 goto out;
471
472         if (addr->sin_family != AF_INET) {
473                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
474                  * only if s_addr is INADDR_ANY.
475                  */
476                 err = -EAFNOSUPPORT;
477                 if (addr->sin_family != AF_UNSPEC ||
478                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
479                         goto out;
480         }
481
482         chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
483
484         /* Not specified by any standard per-se, however it breaks too
485          * many applications when removed.  It is unfortunate since
486          * allowing applications to make a non-local bind solves
487          * several problems with systems using dynamic addressing.
488          * (ie. your servers still start up even if your ISDN link
489          *  is temporarily down)
490          */
491         err = -EADDRNOTAVAIL;
492         if (!sysctl_ip_nonlocal_bind &&
493             !(inet->freebind || inet->transparent) &&
494             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
495             chk_addr_ret != RTN_LOCAL &&
496             chk_addr_ret != RTN_MULTICAST &&
497             chk_addr_ret != RTN_BROADCAST)
498                 goto out;
499
500         snum = ntohs(addr->sin_port);
501         err = -EACCES;
502         if (snum && snum < PROT_SOCK &&
503             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
504                 goto out;
505
506         /*      We keep a pair of addresses. rcv_saddr is the one
507          *      used by hash lookups, and saddr is used for transmit.
508          *
509          *      In the BSD API these are the same except where it
510          *      would be illegal to use them (multicast/broadcast) in
511          *      which case the sending device address is used.
512          */
513         lock_sock(sk);
514
515         /* Check these errors (active socket, double bind). */
516         err = -EINVAL;
517         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
518                 goto out_release_sock;
519
520         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
521         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
522                 inet->inet_saddr = 0;  /* Use device */
523
524         /* Make sure we are allowed to bind here. */
525         if (sk->sk_prot->get_port(sk, snum)) {
526                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
527                 err = -EADDRINUSE;
528                 goto out_release_sock;
529         }
530
531         if (inet->inet_rcv_saddr)
532                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
533         if (snum)
534                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
535         inet->inet_sport = htons(inet->inet_num);
536         inet->inet_daddr = 0;
537         inet->inet_dport = 0;
538         sk_dst_reset(sk);
539         err = 0;
540 out_release_sock:
541         release_sock(sk);
542 out:
543         return err;
544 }
545 EXPORT_SYMBOL(inet_bind);
546
547 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
548                        int addr_len, int flags)
549 {
550         struct sock *sk = sock->sk;
551
552         if (addr_len < sizeof(uaddr->sa_family))
553                 return -EINVAL;
554         if (uaddr->sa_family == AF_UNSPEC)
555                 return sk->sk_prot->disconnect(sk, flags);
556
557         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
558                 return -EAGAIN;
559         return sk->sk_prot->connect(sk, uaddr, addr_len);
560 }
561 EXPORT_SYMBOL(inet_dgram_connect);
562
563 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
564 {
565         DEFINE_WAIT(wait);
566
567         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
568         sk->sk_write_pending += writebias;
569
570         /* Basic assumption: if someone sets sk->sk_err, he _must_
571          * change state of the socket from TCP_SYN_*.
572          * Connect() does not allow to get error notifications
573          * without closing the socket.
574          */
575         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
576                 release_sock(sk);
577                 timeo = schedule_timeout(timeo);
578                 lock_sock(sk);
579                 if (signal_pending(current) || !timeo)
580                         break;
581                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
582         }
583         finish_wait(sk_sleep(sk), &wait);
584         sk->sk_write_pending -= writebias;
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         if (uaddr->sa_family == AF_UNSPEC) {
603                 err = sk->sk_prot->disconnect(sk, flags);
604                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
605                 goto out;
606         }
607
608         switch (sock->state) {
609         default:
610                 err = -EINVAL;
611                 goto out;
612         case SS_CONNECTED:
613                 err = -EISCONN;
614                 goto out;
615         case SS_CONNECTING:
616                 err = -EALREADY;
617                 /* Fall out of switch with err, set for this state */
618                 break;
619         case SS_UNCONNECTED:
620                 err = -EISCONN;
621                 if (sk->sk_state != TCP_CLOSE)
622                         goto out;
623
624                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
625                 if (err < 0)
626                         goto out;
627
628                 sock->state = SS_CONNECTING;
629
630                 /* Just entered SS_CONNECTING state; the only
631                  * difference is that return value in non-blocking
632                  * case is EINPROGRESS, rather than EALREADY.
633                  */
634                 err = -EINPROGRESS;
635                 break;
636         }
637
638         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
639
640         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
641                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
642                                 tcp_sk(sk)->fastopen_req &&
643                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
644
645                 /* Error code is set above */
646                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
647                         goto out;
648
649                 err = sock_intr_errno(timeo);
650                 if (signal_pending(current))
651                         goto out;
652         }
653
654         /* Connection was closed by RST, timeout, ICMP error
655          * or another process disconnected us.
656          */
657         if (sk->sk_state == TCP_CLOSE)
658                 goto sock_error;
659
660         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
661          * and error was received after socket entered established state.
662          * Hence, it is handled normally after connect() return successfully.
663          */
664
665         sock->state = SS_CONNECTED;
666         err = 0;
667 out:
668         return err;
669
670 sock_error:
671         err = sock_error(sk) ? : -ECONNABORTED;
672         sock->state = SS_UNCONNECTED;
673         if (sk->sk_prot->disconnect(sk, flags))
674                 sock->state = SS_DISCONNECTING;
675         goto out;
676 }
677 EXPORT_SYMBOL(__inet_stream_connect);
678
679 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
680                         int addr_len, int flags)
681 {
682         int err;
683
684         lock_sock(sock->sk);
685         err = __inet_stream_connect(sock, uaddr, addr_len, flags);
686         release_sock(sock->sk);
687         return err;
688 }
689 EXPORT_SYMBOL(inet_stream_connect);
690
691 /*
692  *      Accept a pending connection. The TCP layer now gives BSD semantics.
693  */
694
695 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
696 {
697         struct sock *sk1 = sock->sk;
698         int err = -EINVAL;
699         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
700
701         if (!sk2)
702                 goto do_err;
703
704         lock_sock(sk2);
705
706         sock_rps_record_flow(sk2);
707         WARN_ON(!((1 << sk2->sk_state) &
708                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
709                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
710
711         sock_graft(sk2, newsock);
712
713         newsock->state = SS_CONNECTED;
714         err = 0;
715         release_sock(sk2);
716 do_err:
717         return err;
718 }
719 EXPORT_SYMBOL(inet_accept);
720
721
722 /*
723  *      This does both peername and sockname.
724  */
725 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
726                         int *uaddr_len, int peer)
727 {
728         struct sock *sk         = sock->sk;
729         struct inet_sock *inet  = inet_sk(sk);
730         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
731
732         sin->sin_family = AF_INET;
733         if (peer) {
734                 if (!inet->inet_dport ||
735                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
736                      peer == 1))
737                         return -ENOTCONN;
738                 sin->sin_port = inet->inet_dport;
739                 sin->sin_addr.s_addr = inet->inet_daddr;
740         } else {
741                 __be32 addr = inet->inet_rcv_saddr;
742                 if (!addr)
743                         addr = inet->inet_saddr;
744                 sin->sin_port = inet->inet_sport;
745                 sin->sin_addr.s_addr = addr;
746         }
747         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
748         *uaddr_len = sizeof(*sin);
749         return 0;
750 }
751 EXPORT_SYMBOL(inet_getname);
752
753 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
754                  size_t size)
755 {
756         struct sock *sk = sock->sk;
757
758         sock_rps_record_flow(sk);
759
760         /* We may need to bind the socket. */
761         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
762             inet_autobind(sk))
763                 return -EAGAIN;
764
765         return sk->sk_prot->sendmsg(iocb, sk, msg, size);
766 }
767 EXPORT_SYMBOL(inet_sendmsg);
768
769 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
770                       size_t size, int flags)
771 {
772         struct sock *sk = sock->sk;
773
774         sock_rps_record_flow(sk);
775
776         /* We may need to bind the socket. */
777         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
778             inet_autobind(sk))
779                 return -EAGAIN;
780
781         if (sk->sk_prot->sendpage)
782                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
783         return sock_no_sendpage(sock, page, offset, size, flags);
784 }
785 EXPORT_SYMBOL(inet_sendpage);
786
787 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
788                  size_t size, int flags)
789 {
790         struct sock *sk = sock->sk;
791         int addr_len = 0;
792         int err;
793
794         sock_rps_record_flow(sk);
795
796         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
797                                    flags & ~MSG_DONTWAIT, &addr_len);
798         if (err >= 0)
799                 msg->msg_namelen = addr_len;
800         return err;
801 }
802 EXPORT_SYMBOL(inet_recvmsg);
803
804 int inet_shutdown(struct socket *sock, int how)
805 {
806         struct sock *sk = sock->sk;
807         int err = 0;
808
809         /* This should really check to make sure
810          * the socket is a TCP socket. (WHY AC...)
811          */
812         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
813                        1->2 bit 2 snds.
814                        2->3 */
815         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
816                 return -EINVAL;
817
818         lock_sock(sk);
819         if (sock->state == SS_CONNECTING) {
820                 if ((1 << sk->sk_state) &
821                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
822                         sock->state = SS_DISCONNECTING;
823                 else
824                         sock->state = SS_CONNECTED;
825         }
826
827         switch (sk->sk_state) {
828         case TCP_CLOSE:
829                 err = -ENOTCONN;
830                 /* Hack to wake up other listeners, who can poll for
831                    POLLHUP, even on eg. unconnected UDP sockets -- RR */
832         default:
833                 sk->sk_shutdown |= how;
834                 if (sk->sk_prot->shutdown)
835                         sk->sk_prot->shutdown(sk, how);
836                 break;
837
838         /* Remaining two branches are temporary solution for missing
839          * close() in multithreaded environment. It is _not_ a good idea,
840          * but we have no choice until close() is repaired at VFS level.
841          */
842         case TCP_LISTEN:
843                 if (!(how & RCV_SHUTDOWN))
844                         break;
845                 /* Fall through */
846         case TCP_SYN_SENT:
847                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
848                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
849                 break;
850         }
851
852         /* Wake up anyone sleeping in poll. */
853         sk->sk_state_change(sk);
854         release_sock(sk);
855         return err;
856 }
857 EXPORT_SYMBOL(inet_shutdown);
858
859 /*
860  *      ioctl() calls you can issue on an INET socket. Most of these are
861  *      device configuration and stuff and very rarely used. Some ioctls
862  *      pass on to the socket itself.
863  *
864  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
865  *      loads the devconfigure module does its configuring and unloads it.
866  *      There's a good 20K of config code hanging around the kernel.
867  */
868
869 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
870 {
871         struct sock *sk = sock->sk;
872         int err = 0;
873         struct net *net = sock_net(sk);
874
875         switch (cmd) {
876         case SIOCGSTAMP:
877                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
878                 break;
879         case SIOCGSTAMPNS:
880                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
881                 break;
882         case SIOCADDRT:
883         case SIOCDELRT:
884         case SIOCRTMSG:
885                 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
886                 break;
887         case SIOCDARP:
888         case SIOCGARP:
889         case SIOCSARP:
890                 err = arp_ioctl(net, cmd, (void __user *)arg);
891                 break;
892         case SIOCGIFADDR:
893         case SIOCSIFADDR:
894         case SIOCGIFBRDADDR:
895         case SIOCSIFBRDADDR:
896         case SIOCGIFNETMASK:
897         case SIOCSIFNETMASK:
898         case SIOCGIFDSTADDR:
899         case SIOCSIFDSTADDR:
900         case SIOCSIFPFLAGS:
901         case SIOCGIFPFLAGS:
902         case SIOCSIFFLAGS:
903                 err = devinet_ioctl(net, cmd, (void __user *)arg);
904                 break;
905         default:
906                 if (sk->sk_prot->ioctl)
907                         err = sk->sk_prot->ioctl(sk, cmd, arg);
908                 else
909                         err = -ENOIOCTLCMD;
910                 break;
911         }
912         return err;
913 }
914 EXPORT_SYMBOL(inet_ioctl);
915
916 #ifdef CONFIG_COMPAT
917 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
918 {
919         struct sock *sk = sock->sk;
920         int err = -ENOIOCTLCMD;
921
922         if (sk->sk_prot->compat_ioctl)
923                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
924
925         return err;
926 }
927 #endif
928
929 const struct proto_ops inet_stream_ops = {
930         .family            = PF_INET,
931         .owner             = THIS_MODULE,
932         .release           = inet_release,
933         .bind              = inet_bind,
934         .connect           = inet_stream_connect,
935         .socketpair        = sock_no_socketpair,
936         .accept            = inet_accept,
937         .getname           = inet_getname,
938         .poll              = tcp_poll,
939         .ioctl             = inet_ioctl,
940         .listen            = inet_listen,
941         .shutdown          = inet_shutdown,
942         .setsockopt        = sock_common_setsockopt,
943         .getsockopt        = sock_common_getsockopt,
944         .sendmsg           = inet_sendmsg,
945         .recvmsg           = inet_recvmsg,
946         .mmap              = sock_no_mmap,
947         .sendpage          = inet_sendpage,
948         .splice_read       = tcp_splice_read,
949 #ifdef CONFIG_COMPAT
950         .compat_setsockopt = compat_sock_common_setsockopt,
951         .compat_getsockopt = compat_sock_common_getsockopt,
952         .compat_ioctl      = inet_compat_ioctl,
953 #endif
954 };
955 EXPORT_SYMBOL(inet_stream_ops);
956
957 const struct proto_ops inet_dgram_ops = {
958         .family            = PF_INET,
959         .owner             = THIS_MODULE,
960         .release           = inet_release,
961         .bind              = inet_bind,
962         .connect           = inet_dgram_connect,
963         .socketpair        = sock_no_socketpair,
964         .accept            = sock_no_accept,
965         .getname           = inet_getname,
966         .poll              = udp_poll,
967         .ioctl             = inet_ioctl,
968         .listen            = sock_no_listen,
969         .shutdown          = inet_shutdown,
970         .setsockopt        = sock_common_setsockopt,
971         .getsockopt        = sock_common_getsockopt,
972         .sendmsg           = inet_sendmsg,
973         .recvmsg           = inet_recvmsg,
974         .mmap              = sock_no_mmap,
975         .sendpage          = inet_sendpage,
976 #ifdef CONFIG_COMPAT
977         .compat_setsockopt = compat_sock_common_setsockopt,
978         .compat_getsockopt = compat_sock_common_getsockopt,
979         .compat_ioctl      = inet_compat_ioctl,
980 #endif
981 };
982 EXPORT_SYMBOL(inet_dgram_ops);
983
984 /*
985  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
986  * udp_poll
987  */
988 static const struct proto_ops inet_sockraw_ops = {
989         .family            = PF_INET,
990         .owner             = THIS_MODULE,
991         .release           = inet_release,
992         .bind              = inet_bind,
993         .connect           = inet_dgram_connect,
994         .socketpair        = sock_no_socketpair,
995         .accept            = sock_no_accept,
996         .getname           = inet_getname,
997         .poll              = datagram_poll,
998         .ioctl             = inet_ioctl,
999         .listen            = sock_no_listen,
1000         .shutdown          = inet_shutdown,
1001         .setsockopt        = sock_common_setsockopt,
1002         .getsockopt        = sock_common_getsockopt,
1003         .sendmsg           = inet_sendmsg,
1004         .recvmsg           = inet_recvmsg,
1005         .mmap              = sock_no_mmap,
1006         .sendpage          = inet_sendpage,
1007 #ifdef CONFIG_COMPAT
1008         .compat_setsockopt = compat_sock_common_setsockopt,
1009         .compat_getsockopt = compat_sock_common_getsockopt,
1010         .compat_ioctl      = inet_compat_ioctl,
1011 #endif
1012 };
1013
1014 static const struct net_proto_family inet_family_ops = {
1015         .family = PF_INET,
1016         .create = inet_create,
1017         .owner  = THIS_MODULE,
1018 };
1019
1020 /* Upon startup we insert all the elements in inetsw_array[] into
1021  * the linked list inetsw.
1022  */
1023 static struct inet_protosw inetsw_array[] =
1024 {
1025         {
1026                 .type =       SOCK_STREAM,
1027                 .protocol =   IPPROTO_TCP,
1028                 .prot =       &tcp_prot,
1029                 .ops =        &inet_stream_ops,
1030                 .no_check =   0,
1031                 .flags =      INET_PROTOSW_PERMANENT |
1032                               INET_PROTOSW_ICSK,
1033         },
1034
1035         {
1036                 .type =       SOCK_DGRAM,
1037                 .protocol =   IPPROTO_UDP,
1038                 .prot =       &udp_prot,
1039                 .ops =        &inet_dgram_ops,
1040                 .no_check =   UDP_CSUM_DEFAULT,
1041                 .flags =      INET_PROTOSW_PERMANENT,
1042        },
1043
1044        {
1045                 .type =       SOCK_DGRAM,
1046                 .protocol =   IPPROTO_ICMP,
1047                 .prot =       &ping_prot,
1048                 .ops =        &inet_dgram_ops,
1049                 .no_check =   UDP_CSUM_DEFAULT,
1050                 .flags =      INET_PROTOSW_REUSE,
1051        },
1052
1053        {
1054                .type =       SOCK_RAW,
1055                .protocol =   IPPROTO_IP,        /* wild card */
1056                .prot =       &raw_prot,
1057                .ops =        &inet_sockraw_ops,
1058                .no_check =   UDP_CSUM_DEFAULT,
1059                .flags =      INET_PROTOSW_REUSE,
1060        }
1061 };
1062
1063 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1064
1065 void inet_register_protosw(struct inet_protosw *p)
1066 {
1067         struct list_head *lh;
1068         struct inet_protosw *answer;
1069         int protocol = p->protocol;
1070         struct list_head *last_perm;
1071
1072         spin_lock_bh(&inetsw_lock);
1073
1074         if (p->type >= SOCK_MAX)
1075                 goto out_illegal;
1076
1077         /* If we are trying to override a permanent protocol, bail. */
1078         answer = NULL;
1079         last_perm = &inetsw[p->type];
1080         list_for_each(lh, &inetsw[p->type]) {
1081                 answer = list_entry(lh, struct inet_protosw, list);
1082
1083                 /* Check only the non-wild match. */
1084                 if (INET_PROTOSW_PERMANENT & answer->flags) {
1085                         if (protocol == answer->protocol)
1086                                 break;
1087                         last_perm = lh;
1088                 }
1089
1090                 answer = NULL;
1091         }
1092         if (answer)
1093                 goto out_permanent;
1094
1095         /* Add the new entry after the last permanent entry if any, so that
1096          * the new entry does not override a permanent entry when matched with
1097          * a wild-card protocol. But it is allowed to override any existing
1098          * non-permanent entry.  This means that when we remove this entry, the
1099          * system automatically returns to the old behavior.
1100          */
1101         list_add_rcu(&p->list, last_perm);
1102 out:
1103         spin_unlock_bh(&inetsw_lock);
1104
1105         return;
1106
1107 out_permanent:
1108         pr_err("Attempt to override permanent protocol %d\n", protocol);
1109         goto out;
1110
1111 out_illegal:
1112         pr_err("Ignoring attempt to register invalid socket type %d\n",
1113                p->type);
1114         goto out;
1115 }
1116 EXPORT_SYMBOL(inet_register_protosw);
1117
1118 void inet_unregister_protosw(struct inet_protosw *p)
1119 {
1120         if (INET_PROTOSW_PERMANENT & p->flags) {
1121                 pr_err("Attempt to unregister permanent protocol %d\n",
1122                        p->protocol);
1123         } else {
1124                 spin_lock_bh(&inetsw_lock);
1125                 list_del_rcu(&p->list);
1126                 spin_unlock_bh(&inetsw_lock);
1127
1128                 synchronize_net();
1129         }
1130 }
1131 EXPORT_SYMBOL(inet_unregister_protosw);
1132
1133 /*
1134  *      Shall we try to damage output packets if routing dev changes?
1135  */
1136
1137 int sysctl_ip_dynaddr __read_mostly;
1138
1139 static int inet_sk_reselect_saddr(struct sock *sk)
1140 {
1141         struct inet_sock *inet = inet_sk(sk);
1142         __be32 old_saddr = inet->inet_saddr;
1143         __be32 daddr = inet->inet_daddr;
1144         struct flowi4 *fl4;
1145         struct rtable *rt;
1146         __be32 new_saddr;
1147         struct ip_options_rcu *inet_opt;
1148
1149         inet_opt = rcu_dereference_protected(inet->inet_opt,
1150                                              sock_owned_by_user(sk));
1151         if (inet_opt && inet_opt->opt.srr)
1152                 daddr = inet_opt->opt.faddr;
1153
1154         /* Query new route. */
1155         fl4 = &inet->cork.fl.u.ip4;
1156         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1157                               sk->sk_bound_dev_if, sk->sk_protocol,
1158                               inet->inet_sport, inet->inet_dport, sk, false);
1159         if (IS_ERR(rt))
1160                 return PTR_ERR(rt);
1161
1162         sk_setup_caps(sk, &rt->dst);
1163
1164         new_saddr = fl4->saddr;
1165
1166         if (new_saddr == old_saddr)
1167                 return 0;
1168
1169         if (sysctl_ip_dynaddr > 1) {
1170                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1171                         __func__, &old_saddr, &new_saddr);
1172         }
1173
1174         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1175
1176         /*
1177          * XXX The only one ugly spot where we need to
1178          * XXX really change the sockets identity after
1179          * XXX it has entered the hashes. -DaveM
1180          *
1181          * Besides that, it does not check for connection
1182          * uniqueness. Wait for troubles.
1183          */
1184         __sk_prot_rehash(sk);
1185         return 0;
1186 }
1187
1188 int inet_sk_rebuild_header(struct sock *sk)
1189 {
1190         struct inet_sock *inet = inet_sk(sk);
1191         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1192         __be32 daddr;
1193         struct ip_options_rcu *inet_opt;
1194         struct flowi4 *fl4;
1195         int err;
1196
1197         /* Route is OK, nothing to do. */
1198         if (rt)
1199                 return 0;
1200
1201         /* Reroute. */
1202         rcu_read_lock();
1203         inet_opt = rcu_dereference(inet->inet_opt);
1204         daddr = inet->inet_daddr;
1205         if (inet_opt && inet_opt->opt.srr)
1206                 daddr = inet_opt->opt.faddr;
1207         rcu_read_unlock();
1208         fl4 = &inet->cork.fl.u.ip4;
1209         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1210                                    inet->inet_dport, inet->inet_sport,
1211                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1212                                    sk->sk_bound_dev_if);
1213         if (!IS_ERR(rt)) {
1214                 err = 0;
1215                 sk_setup_caps(sk, &rt->dst);
1216         } else {
1217                 err = PTR_ERR(rt);
1218
1219                 /* Routing failed... */
1220                 sk->sk_route_caps = 0;
1221                 /*
1222                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1223                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1224                  */
1225                 if (!sysctl_ip_dynaddr ||
1226                     sk->sk_state != TCP_SYN_SENT ||
1227                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1228                     (err = inet_sk_reselect_saddr(sk)) != 0)
1229                         sk->sk_err_soft = -err;
1230         }
1231
1232         return err;
1233 }
1234 EXPORT_SYMBOL(inet_sk_rebuild_header);
1235
1236 static int inet_gso_send_check(struct sk_buff *skb)
1237 {
1238         const struct net_offload *ops;
1239         const struct iphdr *iph;
1240         int proto;
1241         int ihl;
1242         int err = -EINVAL;
1243
1244         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1245                 goto out;
1246
1247         iph = ip_hdr(skb);
1248         ihl = iph->ihl * 4;
1249         if (ihl < sizeof(*iph))
1250                 goto out;
1251
1252         if (unlikely(!pskb_may_pull(skb, ihl)))
1253                 goto out;
1254
1255         __skb_pull(skb, ihl);
1256         skb_reset_transport_header(skb);
1257         iph = ip_hdr(skb);
1258         proto = iph->protocol;
1259         err = -EPROTONOSUPPORT;
1260
1261         rcu_read_lock();
1262         ops = rcu_dereference(inet_offloads[proto]);
1263         if (likely(ops && ops->callbacks.gso_send_check))
1264                 err = ops->callbacks.gso_send_check(skb);
1265         rcu_read_unlock();
1266
1267 out:
1268         return err;
1269 }
1270
1271 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1272         netdev_features_t features)
1273 {
1274         struct sk_buff *segs = ERR_PTR(-EINVAL);
1275         const struct net_offload *ops;
1276         struct iphdr *iph;
1277         int proto;
1278         int ihl;
1279         int id;
1280         unsigned int offset = 0;
1281
1282         if (!(features & NETIF_F_V4_CSUM))
1283                 features &= ~NETIF_F_SG;
1284
1285         if (unlikely(skb_shinfo(skb)->gso_type &
1286                      ~(SKB_GSO_TCPV4 |
1287                        SKB_GSO_UDP |
1288                        SKB_GSO_DODGY |
1289                        SKB_GSO_TCP_ECN |
1290                        0)))
1291                 goto out;
1292
1293         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1294                 goto out;
1295
1296         iph = ip_hdr(skb);
1297         ihl = iph->ihl * 4;
1298         if (ihl < sizeof(*iph))
1299                 goto out;
1300
1301         if (unlikely(!pskb_may_pull(skb, ihl)))
1302                 goto out;
1303
1304         __skb_pull(skb, ihl);
1305         skb_reset_transport_header(skb);
1306         iph = ip_hdr(skb);
1307         id = ntohs(iph->id);
1308         proto = iph->protocol;
1309         segs = ERR_PTR(-EPROTONOSUPPORT);
1310
1311         rcu_read_lock();
1312         ops = rcu_dereference(inet_offloads[proto]);
1313         if (likely(ops && ops->callbacks.gso_segment))
1314                 segs = ops->callbacks.gso_segment(skb, features);
1315         rcu_read_unlock();
1316
1317         if (IS_ERR_OR_NULL(segs))
1318                 goto out;
1319
1320         skb = segs;
1321         do {
1322                 iph = ip_hdr(skb);
1323                 if (proto == IPPROTO_UDP) {
1324                         iph->id = htons(id);
1325                         iph->frag_off = htons(offset >> 3);
1326                         if (skb->next != NULL)
1327                                 iph->frag_off |= htons(IP_MF);
1328                         offset += (skb->len - skb->mac_len - iph->ihl * 4);
1329                 } else
1330                         iph->id = htons(id++);
1331                 iph->tot_len = htons(skb->len - skb->mac_len);
1332                 iph->check = 0;
1333                 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1334         } while ((skb = skb->next));
1335
1336 out:
1337         return segs;
1338 }
1339
1340 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1341                                          struct sk_buff *skb)
1342 {
1343         const struct net_offload *ops;
1344         struct sk_buff **pp = NULL;
1345         struct sk_buff *p;
1346         const struct iphdr *iph;
1347         unsigned int hlen;
1348         unsigned int off;
1349         unsigned int id;
1350         int flush = 1;
1351         int proto;
1352
1353         off = skb_gro_offset(skb);
1354         hlen = off + sizeof(*iph);
1355         iph = skb_gro_header_fast(skb, off);
1356         if (skb_gro_header_hard(skb, hlen)) {
1357                 iph = skb_gro_header_slow(skb, hlen, off);
1358                 if (unlikely(!iph))
1359                         goto out;
1360         }
1361
1362         proto = iph->protocol;
1363
1364         rcu_read_lock();
1365         ops = rcu_dereference(inet_offloads[proto]);
1366         if (!ops || !ops->callbacks.gro_receive)
1367                 goto out_unlock;
1368
1369         if (*(u8 *)iph != 0x45)
1370                 goto out_unlock;
1371
1372         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1373                 goto out_unlock;
1374
1375         id = ntohl(*(__be32 *)&iph->id);
1376         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1377         id >>= 16;
1378
1379         for (p = *head; p; p = p->next) {
1380                 struct iphdr *iph2;
1381
1382                 if (!NAPI_GRO_CB(p)->same_flow)
1383                         continue;
1384
1385                 iph2 = ip_hdr(p);
1386
1387                 if ((iph->protocol ^ iph2->protocol) |
1388                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1389                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1390                         NAPI_GRO_CB(p)->same_flow = 0;
1391                         continue;
1392                 }
1393
1394                 /* All fields must match except length and checksum. */
1395                 NAPI_GRO_CB(p)->flush |=
1396                         (iph->ttl ^ iph2->ttl) |
1397                         (iph->tos ^ iph2->tos) |
1398                         ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1399
1400                 NAPI_GRO_CB(p)->flush |= flush;
1401         }
1402
1403         NAPI_GRO_CB(skb)->flush |= flush;
1404         skb_gro_pull(skb, sizeof(*iph));
1405         skb_set_transport_header(skb, skb_gro_offset(skb));
1406
1407         pp = ops->callbacks.gro_receive(head, skb);
1408
1409 out_unlock:
1410         rcu_read_unlock();
1411
1412 out:
1413         NAPI_GRO_CB(skb)->flush |= flush;
1414
1415         return pp;
1416 }
1417
1418 static int inet_gro_complete(struct sk_buff *skb)
1419 {
1420         __be16 newlen = htons(skb->len - skb_network_offset(skb));
1421         struct iphdr *iph = ip_hdr(skb);
1422         const struct net_offload *ops;
1423         int proto = iph->protocol;
1424         int err = -ENOSYS;
1425
1426         csum_replace2(&iph->check, iph->tot_len, newlen);
1427         iph->tot_len = newlen;
1428
1429         rcu_read_lock();
1430         ops = rcu_dereference(inet_offloads[proto]);
1431         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1432                 goto out_unlock;
1433
1434         err = ops->callbacks.gro_complete(skb);
1435
1436 out_unlock:
1437         rcu_read_unlock();
1438
1439         return err;
1440 }
1441
1442 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1443                          unsigned short type, unsigned char protocol,
1444                          struct net *net)
1445 {
1446         struct socket *sock;
1447         int rc = sock_create_kern(family, type, protocol, &sock);
1448
1449         if (rc == 0) {
1450                 *sk = sock->sk;
1451                 (*sk)->sk_allocation = GFP_ATOMIC;
1452                 /*
1453                  * Unhash it so that IP input processing does not even see it,
1454                  * we do not wish this socket to see incoming packets.
1455                  */
1456                 (*sk)->sk_prot->unhash(*sk);
1457
1458                 sk_change_net(*sk, net);
1459         }
1460         return rc;
1461 }
1462 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1463
1464 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1465 {
1466         unsigned long res = 0;
1467         int i, j;
1468
1469         for_each_possible_cpu(i) {
1470                 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1471                         res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1472         }
1473         return res;
1474 }
1475 EXPORT_SYMBOL_GPL(snmp_fold_field);
1476
1477 #if BITS_PER_LONG==32
1478
1479 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1480 {
1481         u64 res = 0;
1482         int cpu;
1483
1484         for_each_possible_cpu(cpu) {
1485                 void *bhptr;
1486                 struct u64_stats_sync *syncp;
1487                 u64 v;
1488                 unsigned int start;
1489
1490                 bhptr = per_cpu_ptr(mib[0], cpu);
1491                 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1492                 do {
1493                         start = u64_stats_fetch_begin_bh(syncp);
1494                         v = *(((u64 *) bhptr) + offt);
1495                 } while (u64_stats_fetch_retry_bh(syncp, start));
1496
1497                 res += v;
1498         }
1499         return res;
1500 }
1501 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1502 #endif
1503
1504 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1505 {
1506         BUG_ON(ptr == NULL);
1507         ptr[0] = __alloc_percpu(mibsize, align);
1508         if (!ptr[0])
1509                 return -ENOMEM;
1510 #if SNMP_ARRAY_SZ == 2
1511         ptr[1] = __alloc_percpu(mibsize, align);
1512         if (!ptr[1]) {
1513                 free_percpu(ptr[0]);
1514                 ptr[0] = NULL;
1515                 return -ENOMEM;
1516         }
1517 #endif
1518         return 0;
1519 }
1520 EXPORT_SYMBOL_GPL(snmp_mib_init);
1521
1522 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1523 {
1524         int i;
1525
1526         BUG_ON(ptr == NULL);
1527         for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1528                 free_percpu(ptr[i]);
1529                 ptr[i] = NULL;
1530         }
1531 }
1532 EXPORT_SYMBOL_GPL(snmp_mib_free);
1533
1534 #ifdef CONFIG_IP_MULTICAST
1535 static const struct net_protocol igmp_protocol = {
1536         .handler =      igmp_rcv,
1537         .netns_ok =     1,
1538 };
1539 #endif
1540
1541 static const struct net_protocol tcp_protocol = {
1542         .early_demux    =       tcp_v4_early_demux,
1543         .handler        =       tcp_v4_rcv,
1544         .err_handler    =       tcp_v4_err,
1545         .no_policy      =       1,
1546         .netns_ok       =       1,
1547 };
1548
1549 static const struct net_offload tcp_offload = {
1550         .callbacks = {
1551                 .gso_send_check =       tcp_v4_gso_send_check,
1552                 .gso_segment    =       tcp_tso_segment,
1553                 .gro_receive    =       tcp4_gro_receive,
1554                 .gro_complete   =       tcp4_gro_complete,
1555         },
1556 };
1557
1558 static const struct net_protocol udp_protocol = {
1559         .handler =      udp_rcv,
1560         .err_handler =  udp_err,
1561         .no_policy =    1,
1562         .netns_ok =     1,
1563 };
1564
1565 static const struct net_offload udp_offload = {
1566         .callbacks = {
1567                 .gso_send_check = udp4_ufo_send_check,
1568                 .gso_segment = udp4_ufo_fragment,
1569         },
1570 };
1571
1572 static const struct net_protocol icmp_protocol = {
1573         .handler =      icmp_rcv,
1574         .err_handler =  ping_err,
1575         .no_policy =    1,
1576         .netns_ok =     1,
1577 };
1578
1579 static __net_init int ipv4_mib_init_net(struct net *net)
1580 {
1581         if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1582                           sizeof(struct tcp_mib),
1583                           __alignof__(struct tcp_mib)) < 0)
1584                 goto err_tcp_mib;
1585         if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1586                           sizeof(struct ipstats_mib),
1587                           __alignof__(struct ipstats_mib)) < 0)
1588                 goto err_ip_mib;
1589         if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1590                           sizeof(struct linux_mib),
1591                           __alignof__(struct linux_mib)) < 0)
1592                 goto err_net_mib;
1593         if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1594                           sizeof(struct udp_mib),
1595                           __alignof__(struct udp_mib)) < 0)
1596                 goto err_udp_mib;
1597         if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1598                           sizeof(struct udp_mib),
1599                           __alignof__(struct udp_mib)) < 0)
1600                 goto err_udplite_mib;
1601         if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1602                           sizeof(struct icmp_mib),
1603                           __alignof__(struct icmp_mib)) < 0)
1604                 goto err_icmp_mib;
1605         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1606                                               GFP_KERNEL);
1607         if (!net->mib.icmpmsg_statistics)
1608                 goto err_icmpmsg_mib;
1609
1610         tcp_mib_init(net);
1611         return 0;
1612
1613 err_icmpmsg_mib:
1614         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1615 err_icmp_mib:
1616         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1617 err_udplite_mib:
1618         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1619 err_udp_mib:
1620         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1621 err_net_mib:
1622         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1623 err_ip_mib:
1624         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1625 err_tcp_mib:
1626         return -ENOMEM;
1627 }
1628
1629 static __net_exit void ipv4_mib_exit_net(struct net *net)
1630 {
1631         kfree(net->mib.icmpmsg_statistics);
1632         snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1633         snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1634         snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1635         snmp_mib_free((void __percpu **)net->mib.net_statistics);
1636         snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1637         snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1638 }
1639
1640 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1641         .init = ipv4_mib_init_net,
1642         .exit = ipv4_mib_exit_net,
1643 };
1644
1645 static int __init init_ipv4_mibs(void)
1646 {
1647         return register_pernet_subsys(&ipv4_mib_ops);
1648 }
1649
1650 static int ipv4_proc_init(void);
1651
1652 /*
1653  *      IP protocol layer initialiser
1654  */
1655
1656 static struct packet_offload ip_packet_offload __read_mostly = {
1657         .type = cpu_to_be16(ETH_P_IP),
1658         .callbacks = {
1659                 .gso_send_check = inet_gso_send_check,
1660                 .gso_segment = inet_gso_segment,
1661                 .gro_receive = inet_gro_receive,
1662                 .gro_complete = inet_gro_complete,
1663         },
1664 };
1665
1666 static int __init ipv4_offload_init(void)
1667 {
1668         /*
1669          * Add offloads
1670          */
1671         if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1672                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1673         if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1674                 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1675
1676         dev_add_offload(&ip_packet_offload);
1677         return 0;
1678 }
1679
1680 fs_initcall(ipv4_offload_init);
1681
1682 static struct packet_type ip_packet_type __read_mostly = {
1683         .type = cpu_to_be16(ETH_P_IP),
1684         .func = ip_rcv,
1685 };
1686
1687 static int __init inet_init(void)
1688 {
1689         struct inet_protosw *q;
1690         struct list_head *r;
1691         int rc = -EINVAL;
1692
1693         BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1694
1695         sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1696         if (!sysctl_local_reserved_ports)
1697                 goto out;
1698
1699         rc = proto_register(&tcp_prot, 1);
1700         if (rc)
1701                 goto out_free_reserved_ports;
1702
1703         rc = proto_register(&udp_prot, 1);
1704         if (rc)
1705                 goto out_unregister_tcp_proto;
1706
1707         rc = proto_register(&raw_prot, 1);
1708         if (rc)
1709                 goto out_unregister_udp_proto;
1710
1711         rc = proto_register(&ping_prot, 1);
1712         if (rc)
1713                 goto out_unregister_raw_proto;
1714
1715         /*
1716          *      Tell SOCKET that we are alive...
1717          */
1718
1719         (void)sock_register(&inet_family_ops);
1720
1721 #ifdef CONFIG_SYSCTL
1722         ip_static_sysctl_init();
1723 #endif
1724
1725         tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1726
1727         /*
1728          *      Add all the base protocols.
1729          */
1730
1731         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1732                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1733         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1734                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1735         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1736                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1737 #ifdef CONFIG_IP_MULTICAST
1738         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1739                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1740 #endif
1741
1742         /* Register the socket-side information for inet_create. */
1743         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1744                 INIT_LIST_HEAD(r);
1745
1746         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1747                 inet_register_protosw(q);
1748
1749         /*
1750          *      Set the ARP module up
1751          */
1752
1753         arp_init();
1754
1755         /*
1756          *      Set the IP module up
1757          */
1758
1759         ip_init();
1760
1761         tcp_v4_init();
1762
1763         /* Setup TCP slab cache for open requests. */
1764         tcp_init();
1765
1766         /* Setup UDP memory threshold */
1767         udp_init();
1768
1769         /* Add UDP-Lite (RFC 3828) */
1770         udplite4_register();
1771
1772         ping_init();
1773
1774         /*
1775          *      Set the ICMP layer up
1776          */
1777
1778         if (icmp_init() < 0)
1779                 panic("Failed to create the ICMP control socket.\n");
1780
1781         /*
1782          *      Initialise the multicast router
1783          */
1784 #if defined(CONFIG_IP_MROUTE)
1785         if (ip_mr_init())
1786                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1787 #endif
1788         /*
1789          *      Initialise per-cpu ipv4 mibs
1790          */
1791
1792         if (init_ipv4_mibs())
1793                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1794
1795         ipv4_proc_init();
1796
1797         ipfrag_init();
1798
1799         dev_add_pack(&ip_packet_type);
1800
1801         rc = 0;
1802 out:
1803         return rc;
1804 out_unregister_raw_proto:
1805         proto_unregister(&raw_prot);
1806 out_unregister_udp_proto:
1807         proto_unregister(&udp_prot);
1808 out_unregister_tcp_proto:
1809         proto_unregister(&tcp_prot);
1810 out_free_reserved_ports:
1811         kfree(sysctl_local_reserved_ports);
1812         goto out;
1813 }
1814
1815 fs_initcall(inet_init);
1816
1817 /* ------------------------------------------------------------------------ */
1818
1819 #ifdef CONFIG_PROC_FS
1820 static int __init ipv4_proc_init(void)
1821 {
1822         int rc = 0;
1823
1824         if (raw_proc_init())
1825                 goto out_raw;
1826         if (tcp4_proc_init())
1827                 goto out_tcp;
1828         if (udp4_proc_init())
1829                 goto out_udp;
1830         if (ping_proc_init())
1831                 goto out_ping;
1832         if (ip_misc_proc_init())
1833                 goto out_misc;
1834 out:
1835         return rc;
1836 out_misc:
1837         ping_proc_exit();
1838 out_ping:
1839         udp4_proc_exit();
1840 out_udp:
1841         tcp4_proc_exit();
1842 out_tcp:
1843         raw_proc_exit();
1844 out_raw:
1845         rc = -ENOMEM;
1846         goto out;
1847 }
1848
1849 #else /* CONFIG_PROC_FS */
1850 static int __init ipv4_proc_init(void)
1851 {
1852         return 0;
1853 }
1854 #endif /* CONFIG_PROC_FS */
1855
1856 MODULE_ALIAS_NETPROTO(PF_INET);
1857