selftests/bpf: Selftest for sys_connect hooks
[muen/linux.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/kmod.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 <linux/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/inet_common.h>
115 #include <net/ip_tunnels.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #include <net/secure_seq.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
121 #endif
122 #include <net/l3mdev.h>
123
124 #include <trace/events/sock.h>
125
126 /* The inetsw table contains everything that inet_create needs to
127  * build a new socket.
128  */
129 static struct list_head inetsw[SOCK_MAX];
130 static DEFINE_SPINLOCK(inetsw_lock);
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(refcount_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, tcp_fastopen;
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                 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
216                  * Note that only TCP sockets (SOCK_STREAM) will reach here.
217                  * Also fastopen backlog may already been set via the option
218                  * because the socket was in TCP_LISTEN state previously but
219                  * was shutdown() rather than close().
220                  */
221                 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
222                 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
223                     (tcp_fastopen & TFO_SERVER_ENABLE) &&
224                     !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
225                         fastopen_queue_tune(sk, backlog);
226                         tcp_fastopen_init_key_once(sock_net(sk));
227                 }
228
229                 err = inet_csk_listen_start(sk, backlog);
230                 if (err)
231                         goto out;
232         }
233         sk->sk_max_ack_backlog = backlog;
234         err = 0;
235
236 out:
237         release_sock(sk);
238         return err;
239 }
240 EXPORT_SYMBOL(inet_listen);
241
242 /*
243  *      Create an inet socket.
244  */
245
246 static int inet_create(struct net *net, struct socket *sock, int protocol,
247                        int kern)
248 {
249         struct sock *sk;
250         struct inet_protosw *answer;
251         struct inet_sock *inet;
252         struct proto *answer_prot;
253         unsigned char answer_flags;
254         int try_loading_module = 0;
255         int err;
256
257         if (protocol < 0 || protocol >= IPPROTO_MAX)
258                 return -EINVAL;
259
260         sock->state = SS_UNCONNECTED;
261
262         /* Look for the requested type/protocol pair. */
263 lookup_protocol:
264         err = -ESOCKTNOSUPPORT;
265         rcu_read_lock();
266         list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
267
268                 err = 0;
269                 /* Check the non-wild match. */
270                 if (protocol == answer->protocol) {
271                         if (protocol != IPPROTO_IP)
272                                 break;
273                 } else {
274                         /* Check for the two wild cases. */
275                         if (IPPROTO_IP == protocol) {
276                                 protocol = answer->protocol;
277                                 break;
278                         }
279                         if (IPPROTO_IP == answer->protocol)
280                                 break;
281                 }
282                 err = -EPROTONOSUPPORT;
283         }
284
285         if (unlikely(err)) {
286                 if (try_loading_module < 2) {
287                         rcu_read_unlock();
288                         /*
289                          * Be more specific, e.g. net-pf-2-proto-132-type-1
290                          * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
291                          */
292                         if (++try_loading_module == 1)
293                                 request_module("net-pf-%d-proto-%d-type-%d",
294                                                PF_INET, protocol, sock->type);
295                         /*
296                          * Fall back to generic, e.g. net-pf-2-proto-132
297                          * (net-pf-PF_INET-proto-IPPROTO_SCTP)
298                          */
299                         else
300                                 request_module("net-pf-%d-proto-%d",
301                                                PF_INET, protocol);
302                         goto lookup_protocol;
303                 } else
304                         goto out_rcu_unlock;
305         }
306
307         err = -EPERM;
308         if (sock->type == SOCK_RAW && !kern &&
309             !ns_capable(net->user_ns, CAP_NET_RAW))
310                 goto out_rcu_unlock;
311
312         sock->ops = answer->ops;
313         answer_prot = answer->prot;
314         answer_flags = answer->flags;
315         rcu_read_unlock();
316
317         WARN_ON(!answer_prot->slab);
318
319         err = -ENOBUFS;
320         sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
321         if (!sk)
322                 goto out;
323
324         err = 0;
325         if (INET_PROTOSW_REUSE & answer_flags)
326                 sk->sk_reuse = SK_CAN_REUSE;
327
328         inet = inet_sk(sk);
329         inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
330
331         inet->nodefrag = 0;
332
333         if (SOCK_RAW == sock->type) {
334                 inet->inet_num = protocol;
335                 if (IPPROTO_RAW == protocol)
336                         inet->hdrincl = 1;
337         }
338
339         if (net->ipv4.sysctl_ip_no_pmtu_disc)
340                 inet->pmtudisc = IP_PMTUDISC_DONT;
341         else
342                 inet->pmtudisc = IP_PMTUDISC_WANT;
343
344         inet->inet_id = 0;
345
346         sock_init_data(sock, sk);
347
348         sk->sk_destruct    = inet_sock_destruct;
349         sk->sk_protocol    = protocol;
350         sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
351
352         inet->uc_ttl    = -1;
353         inet->mc_loop   = 1;
354         inet->mc_ttl    = 1;
355         inet->mc_all    = 1;
356         inet->mc_index  = 0;
357         inet->mc_list   = NULL;
358         inet->rcv_tos   = 0;
359
360         sk_refcnt_debug_inc(sk);
361
362         if (inet->inet_num) {
363                 /* It assumes that any protocol which allows
364                  * the user to assign a number at socket
365                  * creation time automatically
366                  * shares.
367                  */
368                 inet->inet_sport = htons(inet->inet_num);
369                 /* Add to protocol hash chains. */
370                 err = sk->sk_prot->hash(sk);
371                 if (err) {
372                         sk_common_release(sk);
373                         goto out;
374                 }
375         }
376
377         if (sk->sk_prot->init) {
378                 err = sk->sk_prot->init(sk);
379                 if (err) {
380                         sk_common_release(sk);
381                         goto out;
382                 }
383         }
384
385         if (!kern) {
386                 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
387                 if (err) {
388                         sk_common_release(sk);
389                         goto out;
390                 }
391         }
392 out:
393         return err;
394 out_rcu_unlock:
395         rcu_read_unlock();
396         goto out;
397 }
398
399
400 /*
401  *      The peer socket should always be NULL (or else). When we call this
402  *      function we are destroying the object and from then on nobody
403  *      should refer to it.
404  */
405 int inet_release(struct socket *sock)
406 {
407         struct sock *sk = sock->sk;
408
409         if (sk) {
410                 long timeout;
411
412                 /* Applications forget to leave groups before exiting */
413                 ip_mc_drop_socket(sk);
414
415                 /* If linger is set, we don't return until the close
416                  * is complete.  Otherwise we return immediately. The
417                  * actually closing is done the same either way.
418                  *
419                  * If the close is due to the process exiting, we never
420                  * linger..
421                  */
422                 timeout = 0;
423                 if (sock_flag(sk, SOCK_LINGER) &&
424                     !(current->flags & PF_EXITING))
425                         timeout = sk->sk_lingertime;
426                 sock->sk = NULL;
427                 sk->sk_prot->close(sk, timeout);
428         }
429         return 0;
430 }
431 EXPORT_SYMBOL(inet_release);
432
433 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
434 {
435         struct sock *sk = sock->sk;
436         int err;
437
438         /* If the socket has its own bind function then use it. (RAW) */
439         if (sk->sk_prot->bind) {
440                 return sk->sk_prot->bind(sk, uaddr, addr_len);
441         }
442         if (addr_len < sizeof(struct sockaddr_in))
443                 return -EINVAL;
444
445         /* BPF prog is run before any checks are done so that if the prog
446          * changes context in a wrong way it will be caught.
447          */
448         err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
449         if (err)
450                 return err;
451
452         return __inet_bind(sk, uaddr, addr_len, false, true);
453 }
454 EXPORT_SYMBOL(inet_bind);
455
456 int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
457                 bool force_bind_address_no_port, bool with_lock)
458 {
459         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
460         struct inet_sock *inet = inet_sk(sk);
461         struct net *net = sock_net(sk);
462         unsigned short snum;
463         int chk_addr_ret;
464         u32 tb_id = RT_TABLE_LOCAL;
465         int err;
466
467         if (addr->sin_family != AF_INET) {
468                 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
469                  * only if s_addr is INADDR_ANY.
470                  */
471                 err = -EAFNOSUPPORT;
472                 if (addr->sin_family != AF_UNSPEC ||
473                     addr->sin_addr.s_addr != htonl(INADDR_ANY))
474                         goto out;
475         }
476
477         tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
478         chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
479
480         /* Not specified by any standard per-se, however it breaks too
481          * many applications when removed.  It is unfortunate since
482          * allowing applications to make a non-local bind solves
483          * several problems with systems using dynamic addressing.
484          * (ie. your servers still start up even if your ISDN link
485          *  is temporarily down)
486          */
487         err = -EADDRNOTAVAIL;
488         if (!net->ipv4.sysctl_ip_nonlocal_bind &&
489             !(inet->freebind || inet->transparent) &&
490             addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
491             chk_addr_ret != RTN_LOCAL &&
492             chk_addr_ret != RTN_MULTICAST &&
493             chk_addr_ret != RTN_BROADCAST)
494                 goto out;
495
496         snum = ntohs(addr->sin_port);
497         err = -EACCES;
498         if (snum && snum < inet_prot_sock(net) &&
499             !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
500                 goto out;
501
502         /*      We keep a pair of addresses. rcv_saddr is the one
503          *      used by hash lookups, and saddr is used for transmit.
504          *
505          *      In the BSD API these are the same except where it
506          *      would be illegal to use them (multicast/broadcast) in
507          *      which case the sending device address is used.
508          */
509         if (with_lock)
510                 lock_sock(sk);
511
512         /* Check these errors (active socket, double bind). */
513         err = -EINVAL;
514         if (sk->sk_state != TCP_CLOSE || inet->inet_num)
515                 goto out_release_sock;
516
517         inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
518         if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
519                 inet->inet_saddr = 0;  /* Use device */
520
521         /* Make sure we are allowed to bind here. */
522         if ((snum || !(inet->bind_address_no_port ||
523                        force_bind_address_no_port)) &&
524             sk->sk_prot->get_port(sk, snum)) {
525                 inet->inet_saddr = inet->inet_rcv_saddr = 0;
526                 err = -EADDRINUSE;
527                 goto out_release_sock;
528         }
529
530         if (inet->inet_rcv_saddr)
531                 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
532         if (snum)
533                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
534         inet->inet_sport = htons(inet->inet_num);
535         inet->inet_daddr = 0;
536         inet->inet_dport = 0;
537         sk_dst_reset(sk);
538         err = 0;
539 out_release_sock:
540         if (with_lock)
541                 release_sock(sk);
542 out:
543         return err;
544 }
545
546 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
547                        int addr_len, int flags)
548 {
549         struct sock *sk = sock->sk;
550         int err;
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 (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
558                 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
559                 if (err)
560                         return err;
561         }
562
563         if (!inet_sk(sk)->inet_num && inet_autobind(sk))
564                 return -EAGAIN;
565         return sk->sk_prot->connect(sk, uaddr, addr_len);
566 }
567 EXPORT_SYMBOL(inet_dgram_connect);
568
569 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
570 {
571         DEFINE_WAIT_FUNC(wait, woken_wake_function);
572
573         add_wait_queue(sk_sleep(sk), &wait);
574         sk->sk_write_pending += writebias;
575
576         /* Basic assumption: if someone sets sk->sk_err, he _must_
577          * change state of the socket from TCP_SYN_*.
578          * Connect() does not allow to get error notifications
579          * without closing the socket.
580          */
581         while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
582                 release_sock(sk);
583                 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
584                 lock_sock(sk);
585                 if (signal_pending(current) || !timeo)
586                         break;
587         }
588         remove_wait_queue(sk_sleep(sk), &wait);
589         sk->sk_write_pending -= writebias;
590         return timeo;
591 }
592
593 /*
594  *      Connect to a remote host. There is regrettably still a little
595  *      TCP 'magic' in here.
596  */
597 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
598                           int addr_len, int flags, int is_sendmsg)
599 {
600         struct sock *sk = sock->sk;
601         int err;
602         long timeo;
603
604         /*
605          * uaddr can be NULL and addr_len can be 0 if:
606          * sk is a TCP fastopen active socket and
607          * TCP_FASTOPEN_CONNECT sockopt is set and
608          * we already have a valid cookie for this socket.
609          * In this case, user can call write() after connect().
610          * write() will invoke tcp_sendmsg_fastopen() which calls
611          * __inet_stream_connect().
612          */
613         if (uaddr) {
614                 if (addr_len < sizeof(uaddr->sa_family))
615                         return -EINVAL;
616
617                 if (uaddr->sa_family == AF_UNSPEC) {
618                         err = sk->sk_prot->disconnect(sk, flags);
619                         sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
620                         goto out;
621                 }
622         }
623
624         switch (sock->state) {
625         default:
626                 err = -EINVAL;
627                 goto out;
628         case SS_CONNECTED:
629                 err = -EISCONN;
630                 goto out;
631         case SS_CONNECTING:
632                 if (inet_sk(sk)->defer_connect)
633                         err = is_sendmsg ? -EINPROGRESS : -EISCONN;
634                 else
635                         err = -EALREADY;
636                 /* Fall out of switch with err, set for this state */
637                 break;
638         case SS_UNCONNECTED:
639                 err = -EISCONN;
640                 if (sk->sk_state != TCP_CLOSE)
641                         goto out;
642
643                 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
644                         err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
645                         if (err)
646                                 goto out;
647                 }
648
649                 err = sk->sk_prot->connect(sk, uaddr, addr_len);
650                 if (err < 0)
651                         goto out;
652
653                 sock->state = SS_CONNECTING;
654
655                 if (!err && inet_sk(sk)->defer_connect)
656                         goto out;
657
658                 /* Just entered SS_CONNECTING state; the only
659                  * difference is that return value in non-blocking
660                  * case is EINPROGRESS, rather than EALREADY.
661                  */
662                 err = -EINPROGRESS;
663                 break;
664         }
665
666         timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
667
668         if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
669                 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
670                                 tcp_sk(sk)->fastopen_req &&
671                                 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
672
673                 /* Error code is set above */
674                 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
675                         goto out;
676
677                 err = sock_intr_errno(timeo);
678                 if (signal_pending(current))
679                         goto out;
680         }
681
682         /* Connection was closed by RST, timeout, ICMP error
683          * or another process disconnected us.
684          */
685         if (sk->sk_state == TCP_CLOSE)
686                 goto sock_error;
687
688         /* sk->sk_err may be not zero now, if RECVERR was ordered by user
689          * and error was received after socket entered established state.
690          * Hence, it is handled normally after connect() return successfully.
691          */
692
693         sock->state = SS_CONNECTED;
694         err = 0;
695 out:
696         return err;
697
698 sock_error:
699         err = sock_error(sk) ? : -ECONNABORTED;
700         sock->state = SS_UNCONNECTED;
701         if (sk->sk_prot->disconnect(sk, flags))
702                 sock->state = SS_DISCONNECTING;
703         goto out;
704 }
705 EXPORT_SYMBOL(__inet_stream_connect);
706
707 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
708                         int addr_len, int flags)
709 {
710         int err;
711
712         lock_sock(sock->sk);
713         err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
714         release_sock(sock->sk);
715         return err;
716 }
717 EXPORT_SYMBOL(inet_stream_connect);
718
719 /*
720  *      Accept a pending connection. The TCP layer now gives BSD semantics.
721  */
722
723 int inet_accept(struct socket *sock, struct socket *newsock, int flags,
724                 bool kern)
725 {
726         struct sock *sk1 = sock->sk;
727         int err = -EINVAL;
728         struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
729
730         if (!sk2)
731                 goto do_err;
732
733         lock_sock(sk2);
734
735         sock_rps_record_flow(sk2);
736         WARN_ON(!((1 << sk2->sk_state) &
737                   (TCPF_ESTABLISHED | TCPF_SYN_RECV |
738                   TCPF_CLOSE_WAIT | TCPF_CLOSE)));
739
740         sock_graft(sk2, newsock);
741
742         newsock->state = SS_CONNECTED;
743         err = 0;
744         release_sock(sk2);
745 do_err:
746         return err;
747 }
748 EXPORT_SYMBOL(inet_accept);
749
750
751 /*
752  *      This does both peername and sockname.
753  */
754 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
755                         int peer)
756 {
757         struct sock *sk         = sock->sk;
758         struct inet_sock *inet  = inet_sk(sk);
759         DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
760
761         sin->sin_family = AF_INET;
762         if (peer) {
763                 if (!inet->inet_dport ||
764                     (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
765                      peer == 1))
766                         return -ENOTCONN;
767                 sin->sin_port = inet->inet_dport;
768                 sin->sin_addr.s_addr = inet->inet_daddr;
769         } else {
770                 __be32 addr = inet->inet_rcv_saddr;
771                 if (!addr)
772                         addr = inet->inet_saddr;
773                 sin->sin_port = inet->inet_sport;
774                 sin->sin_addr.s_addr = addr;
775         }
776         memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
777         return sizeof(*sin);
778 }
779 EXPORT_SYMBOL(inet_getname);
780
781 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
782 {
783         struct sock *sk = sock->sk;
784
785         sock_rps_record_flow(sk);
786
787         /* We may need to bind the socket. */
788         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
789             inet_autobind(sk))
790                 return -EAGAIN;
791
792         return sk->sk_prot->sendmsg(sk, msg, size);
793 }
794 EXPORT_SYMBOL(inet_sendmsg);
795
796 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
797                       size_t size, int flags)
798 {
799         struct sock *sk = sock->sk;
800
801         sock_rps_record_flow(sk);
802
803         /* We may need to bind the socket. */
804         if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
805             inet_autobind(sk))
806                 return -EAGAIN;
807
808         if (sk->sk_prot->sendpage)
809                 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
810         return sock_no_sendpage(sock, page, offset, size, flags);
811 }
812 EXPORT_SYMBOL(inet_sendpage);
813
814 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
815                  int flags)
816 {
817         struct sock *sk = sock->sk;
818         int addr_len = 0;
819         int err;
820
821         if (likely(!(flags & MSG_ERRQUEUE)))
822                 sock_rps_record_flow(sk);
823
824         err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
825                                    flags & ~MSG_DONTWAIT, &addr_len);
826         if (err >= 0)
827                 msg->msg_namelen = addr_len;
828         return err;
829 }
830 EXPORT_SYMBOL(inet_recvmsg);
831
832 int inet_shutdown(struct socket *sock, int how)
833 {
834         struct sock *sk = sock->sk;
835         int err = 0;
836
837         /* This should really check to make sure
838          * the socket is a TCP socket. (WHY AC...)
839          */
840         how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
841                        1->2 bit 2 snds.
842                        2->3 */
843         if ((how & ~SHUTDOWN_MASK) || !how)     /* MAXINT->0 */
844                 return -EINVAL;
845
846         lock_sock(sk);
847         if (sock->state == SS_CONNECTING) {
848                 if ((1 << sk->sk_state) &
849                     (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
850                         sock->state = SS_DISCONNECTING;
851                 else
852                         sock->state = SS_CONNECTED;
853         }
854
855         switch (sk->sk_state) {
856         case TCP_CLOSE:
857                 err = -ENOTCONN;
858                 /* Hack to wake up other listeners, who can poll for
859                    EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
860                 /* fall through */
861         default:
862                 sk->sk_shutdown |= how;
863                 if (sk->sk_prot->shutdown)
864                         sk->sk_prot->shutdown(sk, how);
865                 break;
866
867         /* Remaining two branches are temporary solution for missing
868          * close() in multithreaded environment. It is _not_ a good idea,
869          * but we have no choice until close() is repaired at VFS level.
870          */
871         case TCP_LISTEN:
872                 if (!(how & RCV_SHUTDOWN))
873                         break;
874                 /* fall through */
875         case TCP_SYN_SENT:
876                 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
877                 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
878                 break;
879         }
880
881         /* Wake up anyone sleeping in poll. */
882         sk->sk_state_change(sk);
883         release_sock(sk);
884         return err;
885 }
886 EXPORT_SYMBOL(inet_shutdown);
887
888 /*
889  *      ioctl() calls you can issue on an INET socket. Most of these are
890  *      device configuration and stuff and very rarely used. Some ioctls
891  *      pass on to the socket itself.
892  *
893  *      NOTE: I like the idea of a module for the config stuff. ie ifconfig
894  *      loads the devconfigure module does its configuring and unloads it.
895  *      There's a good 20K of config code hanging around the kernel.
896  */
897
898 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
899 {
900         struct sock *sk = sock->sk;
901         int err = 0;
902         struct net *net = sock_net(sk);
903         void __user *p = (void __user *)arg;
904         struct ifreq ifr;
905         struct rtentry rt;
906
907         switch (cmd) {
908         case SIOCGSTAMP:
909                 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
910                 break;
911         case SIOCGSTAMPNS:
912                 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
913                 break;
914         case SIOCADDRT:
915         case SIOCDELRT:
916                 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
917                         return -EFAULT;
918                 err = ip_rt_ioctl(net, cmd, &rt);
919                 break;
920         case SIOCRTMSG:
921                 err = -EINVAL;
922                 break;
923         case SIOCDARP:
924         case SIOCGARP:
925         case SIOCSARP:
926                 err = arp_ioctl(net, cmd, (void __user *)arg);
927                 break;
928         case SIOCGIFADDR:
929         case SIOCGIFBRDADDR:
930         case SIOCGIFNETMASK:
931         case SIOCGIFDSTADDR:
932         case SIOCGIFPFLAGS:
933                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
934                         return -EFAULT;
935                 err = devinet_ioctl(net, cmd, &ifr);
936                 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
937                         err = -EFAULT;
938                 break;
939
940         case SIOCSIFADDR:
941         case SIOCSIFBRDADDR:
942         case SIOCSIFNETMASK:
943         case SIOCSIFDSTADDR:
944         case SIOCSIFPFLAGS:
945         case SIOCSIFFLAGS:
946                 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
947                         return -EFAULT;
948                 err = devinet_ioctl(net, cmd, &ifr);
949                 break;
950         default:
951                 if (sk->sk_prot->ioctl)
952                         err = sk->sk_prot->ioctl(sk, cmd, arg);
953                 else
954                         err = -ENOIOCTLCMD;
955                 break;
956         }
957         return err;
958 }
959 EXPORT_SYMBOL(inet_ioctl);
960
961 #ifdef CONFIG_COMPAT
962 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
963 {
964         struct sock *sk = sock->sk;
965         int err = -ENOIOCTLCMD;
966
967         if (sk->sk_prot->compat_ioctl)
968                 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
969
970         return err;
971 }
972 #endif
973
974 const struct proto_ops inet_stream_ops = {
975         .family            = PF_INET,
976         .owner             = THIS_MODULE,
977         .release           = inet_release,
978         .bind              = inet_bind,
979         .connect           = inet_stream_connect,
980         .socketpair        = sock_no_socketpair,
981         .accept            = inet_accept,
982         .getname           = inet_getname,
983         .poll              = tcp_poll,
984         .ioctl             = inet_ioctl,
985         .listen            = inet_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         .splice_read       = tcp_splice_read,
994         .read_sock         = tcp_read_sock,
995         .sendmsg_locked    = tcp_sendmsg_locked,
996         .sendpage_locked   = tcp_sendpage_locked,
997         .peek_len          = tcp_peek_len,
998 #ifdef CONFIG_COMPAT
999         .compat_setsockopt = compat_sock_common_setsockopt,
1000         .compat_getsockopt = compat_sock_common_getsockopt,
1001         .compat_ioctl      = inet_compat_ioctl,
1002 #endif
1003 };
1004 EXPORT_SYMBOL(inet_stream_ops);
1005
1006 const struct proto_ops inet_dgram_ops = {
1007         .family            = PF_INET,
1008         .owner             = THIS_MODULE,
1009         .release           = inet_release,
1010         .bind              = inet_bind,
1011         .connect           = inet_dgram_connect,
1012         .socketpair        = sock_no_socketpair,
1013         .accept            = sock_no_accept,
1014         .getname           = inet_getname,
1015         .poll              = udp_poll,
1016         .ioctl             = inet_ioctl,
1017         .listen            = sock_no_listen,
1018         .shutdown          = inet_shutdown,
1019         .setsockopt        = sock_common_setsockopt,
1020         .getsockopt        = sock_common_getsockopt,
1021         .sendmsg           = inet_sendmsg,
1022         .recvmsg           = inet_recvmsg,
1023         .mmap              = sock_no_mmap,
1024         .sendpage          = inet_sendpage,
1025         .set_peek_off      = sk_set_peek_off,
1026 #ifdef CONFIG_COMPAT
1027         .compat_setsockopt = compat_sock_common_setsockopt,
1028         .compat_getsockopt = compat_sock_common_getsockopt,
1029         .compat_ioctl      = inet_compat_ioctl,
1030 #endif
1031 };
1032 EXPORT_SYMBOL(inet_dgram_ops);
1033
1034 /*
1035  * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1036  * udp_poll
1037  */
1038 static const struct proto_ops inet_sockraw_ops = {
1039         .family            = PF_INET,
1040         .owner             = THIS_MODULE,
1041         .release           = inet_release,
1042         .bind              = inet_bind,
1043         .connect           = inet_dgram_connect,
1044         .socketpair        = sock_no_socketpair,
1045         .accept            = sock_no_accept,
1046         .getname           = inet_getname,
1047         .poll              = datagram_poll,
1048         .ioctl             = inet_ioctl,
1049         .listen            = sock_no_listen,
1050         .shutdown          = inet_shutdown,
1051         .setsockopt        = sock_common_setsockopt,
1052         .getsockopt        = sock_common_getsockopt,
1053         .sendmsg           = inet_sendmsg,
1054         .recvmsg           = inet_recvmsg,
1055         .mmap              = sock_no_mmap,
1056         .sendpage          = inet_sendpage,
1057 #ifdef CONFIG_COMPAT
1058         .compat_setsockopt = compat_sock_common_setsockopt,
1059         .compat_getsockopt = compat_sock_common_getsockopt,
1060         .compat_ioctl      = inet_compat_ioctl,
1061 #endif
1062 };
1063
1064 static const struct net_proto_family inet_family_ops = {
1065         .family = PF_INET,
1066         .create = inet_create,
1067         .owner  = THIS_MODULE,
1068 };
1069
1070 /* Upon startup we insert all the elements in inetsw_array[] into
1071  * the linked list inetsw.
1072  */
1073 static struct inet_protosw inetsw_array[] =
1074 {
1075         {
1076                 .type =       SOCK_STREAM,
1077                 .protocol =   IPPROTO_TCP,
1078                 .prot =       &tcp_prot,
1079                 .ops =        &inet_stream_ops,
1080                 .flags =      INET_PROTOSW_PERMANENT |
1081                               INET_PROTOSW_ICSK,
1082         },
1083
1084         {
1085                 .type =       SOCK_DGRAM,
1086                 .protocol =   IPPROTO_UDP,
1087                 .prot =       &udp_prot,
1088                 .ops =        &inet_dgram_ops,
1089                 .flags =      INET_PROTOSW_PERMANENT,
1090        },
1091
1092        {
1093                 .type =       SOCK_DGRAM,
1094                 .protocol =   IPPROTO_ICMP,
1095                 .prot =       &ping_prot,
1096                 .ops =        &inet_sockraw_ops,
1097                 .flags =      INET_PROTOSW_REUSE,
1098        },
1099
1100        {
1101                .type =       SOCK_RAW,
1102                .protocol =   IPPROTO_IP,        /* wild card */
1103                .prot =       &raw_prot,
1104                .ops =        &inet_sockraw_ops,
1105                .flags =      INET_PROTOSW_REUSE,
1106        }
1107 };
1108
1109 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1110
1111 void inet_register_protosw(struct inet_protosw *p)
1112 {
1113         struct list_head *lh;
1114         struct inet_protosw *answer;
1115         int protocol = p->protocol;
1116         struct list_head *last_perm;
1117
1118         spin_lock_bh(&inetsw_lock);
1119
1120         if (p->type >= SOCK_MAX)
1121                 goto out_illegal;
1122
1123         /* If we are trying to override a permanent protocol, bail. */
1124         last_perm = &inetsw[p->type];
1125         list_for_each(lh, &inetsw[p->type]) {
1126                 answer = list_entry(lh, struct inet_protosw, list);
1127                 /* Check only the non-wild match. */
1128                 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1129                         break;
1130                 if (protocol == answer->protocol)
1131                         goto out_permanent;
1132                 last_perm = lh;
1133         }
1134
1135         /* Add the new entry after the last permanent entry if any, so that
1136          * the new entry does not override a permanent entry when matched with
1137          * a wild-card protocol. But it is allowed to override any existing
1138          * non-permanent entry.  This means that when we remove this entry, the
1139          * system automatically returns to the old behavior.
1140          */
1141         list_add_rcu(&p->list, last_perm);
1142 out:
1143         spin_unlock_bh(&inetsw_lock);
1144
1145         return;
1146
1147 out_permanent:
1148         pr_err("Attempt to override permanent protocol %d\n", protocol);
1149         goto out;
1150
1151 out_illegal:
1152         pr_err("Ignoring attempt to register invalid socket type %d\n",
1153                p->type);
1154         goto out;
1155 }
1156 EXPORT_SYMBOL(inet_register_protosw);
1157
1158 void inet_unregister_protosw(struct inet_protosw *p)
1159 {
1160         if (INET_PROTOSW_PERMANENT & p->flags) {
1161                 pr_err("Attempt to unregister permanent protocol %d\n",
1162                        p->protocol);
1163         } else {
1164                 spin_lock_bh(&inetsw_lock);
1165                 list_del_rcu(&p->list);
1166                 spin_unlock_bh(&inetsw_lock);
1167
1168                 synchronize_net();
1169         }
1170 }
1171 EXPORT_SYMBOL(inet_unregister_protosw);
1172
1173 static int inet_sk_reselect_saddr(struct sock *sk)
1174 {
1175         struct inet_sock *inet = inet_sk(sk);
1176         __be32 old_saddr = inet->inet_saddr;
1177         __be32 daddr = inet->inet_daddr;
1178         struct flowi4 *fl4;
1179         struct rtable *rt;
1180         __be32 new_saddr;
1181         struct ip_options_rcu *inet_opt;
1182
1183         inet_opt = rcu_dereference_protected(inet->inet_opt,
1184                                              lockdep_sock_is_held(sk));
1185         if (inet_opt && inet_opt->opt.srr)
1186                 daddr = inet_opt->opt.faddr;
1187
1188         /* Query new route. */
1189         fl4 = &inet->cork.fl.u.ip4;
1190         rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1191                               sk->sk_bound_dev_if, sk->sk_protocol,
1192                               inet->inet_sport, inet->inet_dport, sk);
1193         if (IS_ERR(rt))
1194                 return PTR_ERR(rt);
1195
1196         sk_setup_caps(sk, &rt->dst);
1197
1198         new_saddr = fl4->saddr;
1199
1200         if (new_saddr == old_saddr)
1201                 return 0;
1202
1203         if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1204                 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1205                         __func__, &old_saddr, &new_saddr);
1206         }
1207
1208         inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1209
1210         /*
1211          * XXX The only one ugly spot where we need to
1212          * XXX really change the sockets identity after
1213          * XXX it has entered the hashes. -DaveM
1214          *
1215          * Besides that, it does not check for connection
1216          * uniqueness. Wait for troubles.
1217          */
1218         return __sk_prot_rehash(sk);
1219 }
1220
1221 int inet_sk_rebuild_header(struct sock *sk)
1222 {
1223         struct inet_sock *inet = inet_sk(sk);
1224         struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1225         __be32 daddr;
1226         struct ip_options_rcu *inet_opt;
1227         struct flowi4 *fl4;
1228         int err;
1229
1230         /* Route is OK, nothing to do. */
1231         if (rt)
1232                 return 0;
1233
1234         /* Reroute. */
1235         rcu_read_lock();
1236         inet_opt = rcu_dereference(inet->inet_opt);
1237         daddr = inet->inet_daddr;
1238         if (inet_opt && inet_opt->opt.srr)
1239                 daddr = inet_opt->opt.faddr;
1240         rcu_read_unlock();
1241         fl4 = &inet->cork.fl.u.ip4;
1242         rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1243                                    inet->inet_dport, inet->inet_sport,
1244                                    sk->sk_protocol, RT_CONN_FLAGS(sk),
1245                                    sk->sk_bound_dev_if);
1246         if (!IS_ERR(rt)) {
1247                 err = 0;
1248                 sk_setup_caps(sk, &rt->dst);
1249         } else {
1250                 err = PTR_ERR(rt);
1251
1252                 /* Routing failed... */
1253                 sk->sk_route_caps = 0;
1254                 /*
1255                  * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1256                  * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1257                  */
1258                 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1259                     sk->sk_state != TCP_SYN_SENT ||
1260                     (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1261                     (err = inet_sk_reselect_saddr(sk)) != 0)
1262                         sk->sk_err_soft = -err;
1263         }
1264
1265         return err;
1266 }
1267 EXPORT_SYMBOL(inet_sk_rebuild_header);
1268
1269 void inet_sk_set_state(struct sock *sk, int state)
1270 {
1271         trace_inet_sock_set_state(sk, sk->sk_state, state);
1272         sk->sk_state = state;
1273 }
1274 EXPORT_SYMBOL(inet_sk_set_state);
1275
1276 void inet_sk_state_store(struct sock *sk, int newstate)
1277 {
1278         trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1279         smp_store_release(&sk->sk_state, newstate);
1280 }
1281
1282 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1283                                  netdev_features_t features)
1284 {
1285         bool udpfrag = false, fixedid = false, gso_partial, encap;
1286         struct sk_buff *segs = ERR_PTR(-EINVAL);
1287         const struct net_offload *ops;
1288         unsigned int offset = 0;
1289         struct iphdr *iph;
1290         int proto, tot_len;
1291         int nhoff;
1292         int ihl;
1293         int id;
1294
1295         skb_reset_network_header(skb);
1296         nhoff = skb_network_header(skb) - skb_mac_header(skb);
1297         if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1298                 goto out;
1299
1300         iph = ip_hdr(skb);
1301         ihl = iph->ihl * 4;
1302         if (ihl < sizeof(*iph))
1303                 goto out;
1304
1305         id = ntohs(iph->id);
1306         proto = iph->protocol;
1307
1308         /* Warning: after this point, iph might be no longer valid */
1309         if (unlikely(!pskb_may_pull(skb, ihl)))
1310                 goto out;
1311         __skb_pull(skb, ihl);
1312
1313         encap = SKB_GSO_CB(skb)->encap_level > 0;
1314         if (encap)
1315                 features &= skb->dev->hw_enc_features;
1316         SKB_GSO_CB(skb)->encap_level += ihl;
1317
1318         skb_reset_transport_header(skb);
1319
1320         segs = ERR_PTR(-EPROTONOSUPPORT);
1321
1322         if (!skb->encapsulation || encap) {
1323                 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1324                 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1325
1326                 /* fixed ID is invalid if DF bit is not set */
1327                 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1328                         goto out;
1329         }
1330
1331         ops = rcu_dereference(inet_offloads[proto]);
1332         if (likely(ops && ops->callbacks.gso_segment))
1333                 segs = ops->callbacks.gso_segment(skb, features);
1334
1335         if (IS_ERR_OR_NULL(segs))
1336                 goto out;
1337
1338         gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1339
1340         skb = segs;
1341         do {
1342                 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1343                 if (udpfrag) {
1344                         iph->frag_off = htons(offset >> 3);
1345                         if (skb->next)
1346                                 iph->frag_off |= htons(IP_MF);
1347                         offset += skb->len - nhoff - ihl;
1348                         tot_len = skb->len - nhoff;
1349                 } else if (skb_is_gso(skb)) {
1350                         if (!fixedid) {
1351                                 iph->id = htons(id);
1352                                 id += skb_shinfo(skb)->gso_segs;
1353                         }
1354
1355                         if (gso_partial)
1356                                 tot_len = skb_shinfo(skb)->gso_size +
1357                                           SKB_GSO_CB(skb)->data_offset +
1358                                           skb->head - (unsigned char *)iph;
1359                         else
1360                                 tot_len = skb->len - nhoff;
1361                 } else {
1362                         if (!fixedid)
1363                                 iph->id = htons(id++);
1364                         tot_len = skb->len - nhoff;
1365                 }
1366                 iph->tot_len = htons(tot_len);
1367                 ip_send_check(iph);
1368                 if (encap)
1369                         skb_reset_inner_headers(skb);
1370                 skb->network_header = (u8 *)iph - skb->head;
1371         } while ((skb = skb->next));
1372
1373 out:
1374         return segs;
1375 }
1376 EXPORT_SYMBOL(inet_gso_segment);
1377
1378 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1379 {
1380         const struct net_offload *ops;
1381         struct sk_buff **pp = NULL;
1382         struct sk_buff *p;
1383         const struct iphdr *iph;
1384         unsigned int hlen;
1385         unsigned int off;
1386         unsigned int id;
1387         int flush = 1;
1388         int proto;
1389
1390         off = skb_gro_offset(skb);
1391         hlen = off + sizeof(*iph);
1392         iph = skb_gro_header_fast(skb, off);
1393         if (skb_gro_header_hard(skb, hlen)) {
1394                 iph = skb_gro_header_slow(skb, hlen, off);
1395                 if (unlikely(!iph))
1396                         goto out;
1397         }
1398
1399         proto = iph->protocol;
1400
1401         rcu_read_lock();
1402         ops = rcu_dereference(inet_offloads[proto]);
1403         if (!ops || !ops->callbacks.gro_receive)
1404                 goto out_unlock;
1405
1406         if (*(u8 *)iph != 0x45)
1407                 goto out_unlock;
1408
1409         if (ip_is_fragment(iph))
1410                 goto out_unlock;
1411
1412         if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1413                 goto out_unlock;
1414
1415         id = ntohl(*(__be32 *)&iph->id);
1416         flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1417         id >>= 16;
1418
1419         for (p = *head; p; p = p->next) {
1420                 struct iphdr *iph2;
1421                 u16 flush_id;
1422
1423                 if (!NAPI_GRO_CB(p)->same_flow)
1424                         continue;
1425
1426                 iph2 = (struct iphdr *)(p->data + off);
1427                 /* The above works because, with the exception of the top
1428                  * (inner most) layer, we only aggregate pkts with the same
1429                  * hdr length so all the hdrs we'll need to verify will start
1430                  * at the same offset.
1431                  */
1432                 if ((iph->protocol ^ iph2->protocol) |
1433                     ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1434                     ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1435                         NAPI_GRO_CB(p)->same_flow = 0;
1436                         continue;
1437                 }
1438
1439                 /* All fields must match except length and checksum. */
1440                 NAPI_GRO_CB(p)->flush |=
1441                         (iph->ttl ^ iph2->ttl) |
1442                         (iph->tos ^ iph2->tos) |
1443                         ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1444
1445                 NAPI_GRO_CB(p)->flush |= flush;
1446
1447                 /* We need to store of the IP ID check to be included later
1448                  * when we can verify that this packet does in fact belong
1449                  * to a given flow.
1450                  */
1451                 flush_id = (u16)(id - ntohs(iph2->id));
1452
1453                 /* This bit of code makes it much easier for us to identify
1454                  * the cases where we are doing atomic vs non-atomic IP ID
1455                  * checks.  Specifically an atomic check can return IP ID
1456                  * values 0 - 0xFFFF, while a non-atomic check can only
1457                  * return 0 or 0xFFFF.
1458                  */
1459                 if (!NAPI_GRO_CB(p)->is_atomic ||
1460                     !(iph->frag_off & htons(IP_DF))) {
1461                         flush_id ^= NAPI_GRO_CB(p)->count;
1462                         flush_id = flush_id ? 0xFFFF : 0;
1463                 }
1464
1465                 /* If the previous IP ID value was based on an atomic
1466                  * datagram we can overwrite the value and ignore it.
1467                  */
1468                 if (NAPI_GRO_CB(skb)->is_atomic)
1469                         NAPI_GRO_CB(p)->flush_id = flush_id;
1470                 else
1471                         NAPI_GRO_CB(p)->flush_id |= flush_id;
1472         }
1473
1474         NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1475         NAPI_GRO_CB(skb)->flush |= flush;
1476         skb_set_network_header(skb, off);
1477         /* The above will be needed by the transport layer if there is one
1478          * immediately following this IP hdr.
1479          */
1480
1481         /* Note : No need to call skb_gro_postpull_rcsum() here,
1482          * as we already checked checksum over ipv4 header was 0
1483          */
1484         skb_gro_pull(skb, sizeof(*iph));
1485         skb_set_transport_header(skb, skb_gro_offset(skb));
1486
1487         pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1488
1489 out_unlock:
1490         rcu_read_unlock();
1491
1492 out:
1493         skb_gro_flush_final(skb, pp, flush);
1494
1495         return pp;
1496 }
1497 EXPORT_SYMBOL(inet_gro_receive);
1498
1499 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1500                                          struct sk_buff *skb)
1501 {
1502         if (NAPI_GRO_CB(skb)->encap_mark) {
1503                 NAPI_GRO_CB(skb)->flush = 1;
1504                 return NULL;
1505         }
1506
1507         NAPI_GRO_CB(skb)->encap_mark = 1;
1508
1509         return inet_gro_receive(head, skb);
1510 }
1511
1512 #define SECONDS_PER_DAY 86400
1513
1514 /* inet_current_timestamp - Return IP network timestamp
1515  *
1516  * Return milliseconds since midnight in network byte order.
1517  */
1518 __be32 inet_current_timestamp(void)
1519 {
1520         u32 secs;
1521         u32 msecs;
1522         struct timespec64 ts;
1523
1524         ktime_get_real_ts64(&ts);
1525
1526         /* Get secs since midnight. */
1527         (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1528         /* Convert to msecs. */
1529         msecs = secs * MSEC_PER_SEC;
1530         /* Convert nsec to msec. */
1531         msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1532
1533         /* Convert to network byte order. */
1534         return htonl(msecs);
1535 }
1536 EXPORT_SYMBOL(inet_current_timestamp);
1537
1538 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1539 {
1540         if (sk->sk_family == AF_INET)
1541                 return ip_recv_error(sk, msg, len, addr_len);
1542 #if IS_ENABLED(CONFIG_IPV6)
1543         if (sk->sk_family == AF_INET6)
1544                 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1545 #endif
1546         return -EINVAL;
1547 }
1548
1549 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1550 {
1551         __be16 newlen = htons(skb->len - nhoff);
1552         struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1553         const struct net_offload *ops;
1554         int proto = iph->protocol;
1555         int err = -ENOSYS;
1556
1557         if (skb->encapsulation) {
1558                 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1559                 skb_set_inner_network_header(skb, nhoff);
1560         }
1561
1562         csum_replace2(&iph->check, iph->tot_len, newlen);
1563         iph->tot_len = newlen;
1564
1565         rcu_read_lock();
1566         ops = rcu_dereference(inet_offloads[proto]);
1567         if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1568                 goto out_unlock;
1569
1570         /* Only need to add sizeof(*iph) to get to the next hdr below
1571          * because any hdr with option will have been flushed in
1572          * inet_gro_receive().
1573          */
1574         err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1575
1576 out_unlock:
1577         rcu_read_unlock();
1578
1579         return err;
1580 }
1581 EXPORT_SYMBOL(inet_gro_complete);
1582
1583 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1584 {
1585         skb->encapsulation = 1;
1586         skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1587         return inet_gro_complete(skb, nhoff);
1588 }
1589
1590 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1591                          unsigned short type, unsigned char protocol,
1592                          struct net *net)
1593 {
1594         struct socket *sock;
1595         int rc = sock_create_kern(net, family, type, protocol, &sock);
1596
1597         if (rc == 0) {
1598                 *sk = sock->sk;
1599                 (*sk)->sk_allocation = GFP_ATOMIC;
1600                 /*
1601                  * Unhash it so that IP input processing does not even see it,
1602                  * we do not wish this socket to see incoming packets.
1603                  */
1604                 (*sk)->sk_prot->unhash(*sk);
1605         }
1606         return rc;
1607 }
1608 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1609
1610 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1611 {
1612         return  *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1613 }
1614 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1615
1616 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1617 {
1618         unsigned long res = 0;
1619         int i;
1620
1621         for_each_possible_cpu(i)
1622                 res += snmp_get_cpu_field(mib, i, offt);
1623         return res;
1624 }
1625 EXPORT_SYMBOL_GPL(snmp_fold_field);
1626
1627 #if BITS_PER_LONG==32
1628
1629 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1630                          size_t syncp_offset)
1631 {
1632         void *bhptr;
1633         struct u64_stats_sync *syncp;
1634         u64 v;
1635         unsigned int start;
1636
1637         bhptr = per_cpu_ptr(mib, cpu);
1638         syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1639         do {
1640                 start = u64_stats_fetch_begin_irq(syncp);
1641                 v = *(((u64 *)bhptr) + offt);
1642         } while (u64_stats_fetch_retry_irq(syncp, start));
1643
1644         return v;
1645 }
1646 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1647
1648 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1649 {
1650         u64 res = 0;
1651         int cpu;
1652
1653         for_each_possible_cpu(cpu) {
1654                 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1655         }
1656         return res;
1657 }
1658 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1659 #endif
1660
1661 #ifdef CONFIG_IP_MULTICAST
1662 static const struct net_protocol igmp_protocol = {
1663         .handler =      igmp_rcv,
1664         .netns_ok =     1,
1665 };
1666 #endif
1667
1668 /* thinking of making this const? Don't.
1669  * early_demux can change based on sysctl.
1670  */
1671 static struct net_protocol tcp_protocol = {
1672         .early_demux    =       tcp_v4_early_demux,
1673         .early_demux_handler =  tcp_v4_early_demux,
1674         .handler        =       tcp_v4_rcv,
1675         .err_handler    =       tcp_v4_err,
1676         .no_policy      =       1,
1677         .netns_ok       =       1,
1678         .icmp_strict_tag_validation = 1,
1679 };
1680
1681 /* thinking of making this const? Don't.
1682  * early_demux can change based on sysctl.
1683  */
1684 static struct net_protocol udp_protocol = {
1685         .early_demux =  udp_v4_early_demux,
1686         .early_demux_handler =  udp_v4_early_demux,
1687         .handler =      udp_rcv,
1688         .err_handler =  udp_err,
1689         .no_policy =    1,
1690         .netns_ok =     1,
1691 };
1692
1693 static const struct net_protocol icmp_protocol = {
1694         .handler =      icmp_rcv,
1695         .err_handler =  icmp_err,
1696         .no_policy =    1,
1697         .netns_ok =     1,
1698 };
1699
1700 static __net_init int ipv4_mib_init_net(struct net *net)
1701 {
1702         int i;
1703
1704         net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1705         if (!net->mib.tcp_statistics)
1706                 goto err_tcp_mib;
1707         net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1708         if (!net->mib.ip_statistics)
1709                 goto err_ip_mib;
1710
1711         for_each_possible_cpu(i) {
1712                 struct ipstats_mib *af_inet_stats;
1713                 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1714                 u64_stats_init(&af_inet_stats->syncp);
1715         }
1716
1717         net->mib.net_statistics = alloc_percpu(struct linux_mib);
1718         if (!net->mib.net_statistics)
1719                 goto err_net_mib;
1720         net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1721         if (!net->mib.udp_statistics)
1722                 goto err_udp_mib;
1723         net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1724         if (!net->mib.udplite_statistics)
1725                 goto err_udplite_mib;
1726         net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1727         if (!net->mib.icmp_statistics)
1728                 goto err_icmp_mib;
1729         net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1730                                               GFP_KERNEL);
1731         if (!net->mib.icmpmsg_statistics)
1732                 goto err_icmpmsg_mib;
1733
1734         tcp_mib_init(net);
1735         return 0;
1736
1737 err_icmpmsg_mib:
1738         free_percpu(net->mib.icmp_statistics);
1739 err_icmp_mib:
1740         free_percpu(net->mib.udplite_statistics);
1741 err_udplite_mib:
1742         free_percpu(net->mib.udp_statistics);
1743 err_udp_mib:
1744         free_percpu(net->mib.net_statistics);
1745 err_net_mib:
1746         free_percpu(net->mib.ip_statistics);
1747 err_ip_mib:
1748         free_percpu(net->mib.tcp_statistics);
1749 err_tcp_mib:
1750         return -ENOMEM;
1751 }
1752
1753 static __net_exit void ipv4_mib_exit_net(struct net *net)
1754 {
1755         kfree(net->mib.icmpmsg_statistics);
1756         free_percpu(net->mib.icmp_statistics);
1757         free_percpu(net->mib.udplite_statistics);
1758         free_percpu(net->mib.udp_statistics);
1759         free_percpu(net->mib.net_statistics);
1760         free_percpu(net->mib.ip_statistics);
1761         free_percpu(net->mib.tcp_statistics);
1762 }
1763
1764 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1765         .init = ipv4_mib_init_net,
1766         .exit = ipv4_mib_exit_net,
1767         .async = true,
1768 };
1769
1770 static int __init init_ipv4_mibs(void)
1771 {
1772         return register_pernet_subsys(&ipv4_mib_ops);
1773 }
1774
1775 static __net_init int inet_init_net(struct net *net)
1776 {
1777         /*
1778          * Set defaults for local port range
1779          */
1780         seqlock_init(&net->ipv4.ip_local_ports.lock);
1781         net->ipv4.ip_local_ports.range[0] =  32768;
1782         net->ipv4.ip_local_ports.range[1] =  60999;
1783
1784         seqlock_init(&net->ipv4.ping_group_range.lock);
1785         /*
1786          * Sane defaults - nobody may create ping sockets.
1787          * Boot scripts should set this to distro-specific group.
1788          */
1789         net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1790         net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1791
1792         /* Default values for sysctl-controlled parameters.
1793          * We set them here, in case sysctl is not compiled.
1794          */
1795         net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1796         net->ipv4.sysctl_ip_dynaddr = 0;
1797         net->ipv4.sysctl_ip_early_demux = 1;
1798         net->ipv4.sysctl_udp_early_demux = 1;
1799         net->ipv4.sysctl_tcp_early_demux = 1;
1800 #ifdef CONFIG_SYSCTL
1801         net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1802 #endif
1803
1804         /* Some igmp sysctl, whose values are always used */
1805         net->ipv4.sysctl_igmp_max_memberships = 20;
1806         net->ipv4.sysctl_igmp_max_msf = 10;
1807         /* IGMP reports for link-local multicast groups are enabled by default */
1808         net->ipv4.sysctl_igmp_llm_reports = 1;
1809         net->ipv4.sysctl_igmp_qrv = 2;
1810
1811         return 0;
1812 }
1813
1814 static __net_exit void inet_exit_net(struct net *net)
1815 {
1816 }
1817
1818 static __net_initdata struct pernet_operations af_inet_ops = {
1819         .init = inet_init_net,
1820         .exit = inet_exit_net,
1821         .async = true,
1822 };
1823
1824 static int __init init_inet_pernet_ops(void)
1825 {
1826         return register_pernet_subsys(&af_inet_ops);
1827 }
1828
1829 static int ipv4_proc_init(void);
1830
1831 /*
1832  *      IP protocol layer initialiser
1833  */
1834
1835 static struct packet_offload ip_packet_offload __read_mostly = {
1836         .type = cpu_to_be16(ETH_P_IP),
1837         .callbacks = {
1838                 .gso_segment = inet_gso_segment,
1839                 .gro_receive = inet_gro_receive,
1840                 .gro_complete = inet_gro_complete,
1841         },
1842 };
1843
1844 static const struct net_offload ipip_offload = {
1845         .callbacks = {
1846                 .gso_segment    = inet_gso_segment,
1847                 .gro_receive    = ipip_gro_receive,
1848                 .gro_complete   = ipip_gro_complete,
1849         },
1850 };
1851
1852 static int __init ipip_offload_init(void)
1853 {
1854         return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1855 }
1856
1857 static int __init ipv4_offload_init(void)
1858 {
1859         /*
1860          * Add offloads
1861          */
1862         if (udpv4_offload_init() < 0)
1863                 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1864         if (tcpv4_offload_init() < 0)
1865                 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1866         if (ipip_offload_init() < 0)
1867                 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1868
1869         dev_add_offload(&ip_packet_offload);
1870         return 0;
1871 }
1872
1873 fs_initcall(ipv4_offload_init);
1874
1875 static struct packet_type ip_packet_type __read_mostly = {
1876         .type = cpu_to_be16(ETH_P_IP),
1877         .func = ip_rcv,
1878 };
1879
1880 static int __init inet_init(void)
1881 {
1882         struct inet_protosw *q;
1883         struct list_head *r;
1884         int rc = -EINVAL;
1885
1886         sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1887
1888         rc = proto_register(&tcp_prot, 1);
1889         if (rc)
1890                 goto out;
1891
1892         rc = proto_register(&udp_prot, 1);
1893         if (rc)
1894                 goto out_unregister_tcp_proto;
1895
1896         rc = proto_register(&raw_prot, 1);
1897         if (rc)
1898                 goto out_unregister_udp_proto;
1899
1900         rc = proto_register(&ping_prot, 1);
1901         if (rc)
1902                 goto out_unregister_raw_proto;
1903
1904         /*
1905          *      Tell SOCKET that we are alive...
1906          */
1907
1908         (void)sock_register(&inet_family_ops);
1909
1910 #ifdef CONFIG_SYSCTL
1911         ip_static_sysctl_init();
1912 #endif
1913
1914         /*
1915          *      Add all the base protocols.
1916          */
1917
1918         if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1919                 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1920         if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1921                 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1922         if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1923                 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1924 #ifdef CONFIG_IP_MULTICAST
1925         if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1926                 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1927 #endif
1928
1929         /* Register the socket-side information for inet_create. */
1930         for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1931                 INIT_LIST_HEAD(r);
1932
1933         for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1934                 inet_register_protosw(q);
1935
1936         /*
1937          *      Set the ARP module up
1938          */
1939
1940         arp_init();
1941
1942         /*
1943          *      Set the IP module up
1944          */
1945
1946         ip_init();
1947
1948         /* Setup TCP slab cache for open requests. */
1949         tcp_init();
1950
1951         /* Setup UDP memory threshold */
1952         udp_init();
1953
1954         /* Add UDP-Lite (RFC 3828) */
1955         udplite4_register();
1956
1957         ping_init();
1958
1959         /*
1960          *      Set the ICMP layer up
1961          */
1962
1963         if (icmp_init() < 0)
1964                 panic("Failed to create the ICMP control socket.\n");
1965
1966         /*
1967          *      Initialise the multicast router
1968          */
1969 #if defined(CONFIG_IP_MROUTE)
1970         if (ip_mr_init())
1971                 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1972 #endif
1973
1974         if (init_inet_pernet_ops())
1975                 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1976         /*
1977          *      Initialise per-cpu ipv4 mibs
1978          */
1979
1980         if (init_ipv4_mibs())
1981                 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1982
1983         ipv4_proc_init();
1984
1985         ipfrag_init();
1986
1987         dev_add_pack(&ip_packet_type);
1988
1989         ip_tunnel_core_init();
1990
1991         rc = 0;
1992 out:
1993         return rc;
1994 out_unregister_raw_proto:
1995         proto_unregister(&raw_prot);
1996 out_unregister_udp_proto:
1997         proto_unregister(&udp_prot);
1998 out_unregister_tcp_proto:
1999         proto_unregister(&tcp_prot);
2000         goto out;
2001 }
2002
2003 fs_initcall(inet_init);
2004
2005 /* ------------------------------------------------------------------------ */
2006
2007 #ifdef CONFIG_PROC_FS
2008 static int __init ipv4_proc_init(void)
2009 {
2010         int rc = 0;
2011
2012         if (raw_proc_init())
2013                 goto out_raw;
2014         if (tcp4_proc_init())
2015                 goto out_tcp;
2016         if (udp4_proc_init())
2017                 goto out_udp;
2018         if (ping_proc_init())
2019                 goto out_ping;
2020         if (ip_misc_proc_init())
2021                 goto out_misc;
2022 out:
2023         return rc;
2024 out_misc:
2025         ping_proc_exit();
2026 out_ping:
2027         udp4_proc_exit();
2028 out_udp:
2029         tcp4_proc_exit();
2030 out_tcp:
2031         raw_proc_exit();
2032 out_raw:
2033         rc = -ENOMEM;
2034         goto out;
2035 }
2036
2037 #else /* CONFIG_PROC_FS */
2038 static int __init ipv4_proc_init(void)
2039 {
2040         return 0;
2041 }
2042 #endif /* CONFIG_PROC_FS */