1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PF_INET protocol family socket handler.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
12 *
13 * Changes (see also sock.c)
14 *
15 * piggy,
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
23 * structures
24 * when accept() ed
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
30 * Alan Cox,
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
40 * compactness.
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
46 * dumbly.
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
61 *
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
66 */
67
68 #include <linux/err.h>
69 #include <linux/errno.h>
70 #include <linux/types.h>
71 #include <linux/socket.h>
72 #include <linux/in.h>
73 #include <linux/kernel.h>
74 #include <linux/module.h>
75 #include <linux/sched.h>
76 #include <linux/timer.h>
77 #include <linux/string.h>
78 #include <linux/sockios.h>
79 #include <linux/net.h>
80 #include <linux/capability.h>
81 #include <linux/fcntl.h>
82 #include <linux/mm.h>
83 #include <linux/interrupt.h>
84 #include <linux/stat.h>
85 #include <linux/init.h>
86 #include <linux/poll.h>
87 #include <linux/netfilter_ipv4.h>
88 #include <linux/random.h>
89 #include <linux/slab.h>
90
91 #include <asm/uaccess.h>
92 #include <asm/system.h>
93
94 #include <linux/inet.h>
95 #include <linux/igmp.h>
96 #include <linux/inetdevice.h>
97 #include <linux/netdevice.h>
98 #include <net/checksum.h>
99 #include <net/ip.h>
100 #include <net/protocol.h>
101 #include <net/arp.h>
102 #include <net/route.h>
103 #include <net/ip_fib.h>
104 #include <net/inet_connection_sock.h>
105 #include <net/tcp.h>
106 #include <net/udp.h>
107 #include <net/udplite.h>
108 #include <linux/skbuff.h>
109 #include <net/sock.h>
110 #include <net/raw.h>
111 #include <net/icmp.h>
112 #include <net/ipip.h>
113 #include <net/inet_common.h>
114 #include <net/xfrm.h>
115 #include <net/net_namespace.h>
116 #ifdef CONFIG_IP_MROUTE
117 #include <linux/mroute.h>
118 #endif
119
120
121 /* The inetsw table contains everything that inet_create needs to
122 * build a new socket.
123 */
124 static struct list_head inetsw[SOCK_MAX];
125 static DEFINE_SPINLOCK(inetsw_lock);
126
127 struct ipv4_config ipv4_config;
128 EXPORT_SYMBOL(ipv4_config);
129
130 /* New destruction routine */
131
inet_sock_destruct(struct sock * sk)132 void inet_sock_destruct(struct sock *sk)
133 {
134 struct inet_sock *inet = inet_sk(sk);
135
136 __skb_queue_purge(&sk->sk_receive_queue);
137 __skb_queue_purge(&sk->sk_error_queue);
138
139 sk_mem_reclaim(sk);
140
141 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
142 pr_err("Attempt to release TCP socket in state %d %p\n",
143 sk->sk_state, sk);
144 return;
145 }
146 if (!sock_flag(sk, SOCK_DEAD)) {
147 pr_err("Attempt to release alive inet socket %p\n", sk);
148 return;
149 }
150
151 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
152 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
153 WARN_ON(sk->sk_wmem_queued);
154 WARN_ON(sk->sk_forward_alloc);
155
156 kfree(inet->opt);
157 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
158 sk_refcnt_debug_dec(sk);
159 }
160 EXPORT_SYMBOL(inet_sock_destruct);
161
162 /*
163 * The routines beyond this point handle the behaviour of an AF_INET
164 * socket object. Mostly it punts to the subprotocols of IP to do
165 * the work.
166 */
167
168 /*
169 * Automatically bind an unbound socket.
170 */
171
inet_autobind(struct sock * sk)172 static int inet_autobind(struct sock *sk)
173 {
174 struct inet_sock *inet;
175 /* We may need to bind the socket. */
176 lock_sock(sk);
177 inet = inet_sk(sk);
178 if (!inet->inet_num) {
179 if (sk->sk_prot->get_port(sk, 0)) {
180 release_sock(sk);
181 return -EAGAIN;
182 }
183 inet->inet_sport = htons(inet->inet_num);
184 }
185 release_sock(sk);
186 return 0;
187 }
188
189 /*
190 * Move a socket into listening state.
191 */
inet_listen(struct socket * sock,int backlog)192 int inet_listen(struct socket *sock, int backlog)
193 {
194 struct sock *sk = sock->sk;
195 unsigned char old_state;
196 int err;
197
198 lock_sock(sk);
199
200 err = -EINVAL;
201 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
202 goto out;
203
204 old_state = sk->sk_state;
205 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
206 goto out;
207
208 /* Really, if the socket is already in listen state
209 * we can only allow the backlog to be adjusted.
210 */
211 if (old_state != TCP_LISTEN) {
212 err = inet_csk_listen_start(sk, backlog);
213 if (err)
214 goto out;
215 }
216 sk->sk_max_ack_backlog = backlog;
217 err = 0;
218
219 out:
220 release_sock(sk);
221 return err;
222 }
223 EXPORT_SYMBOL(inet_listen);
224
225 u32 inet_ehash_secret __read_mostly;
226 EXPORT_SYMBOL(inet_ehash_secret);
227
228 /*
229 * inet_ehash_secret must be set exactly once
230 */
build_ehash_secret(void)231 void build_ehash_secret(void)
232 {
233 u32 rnd;
234
235 do {
236 get_random_bytes(&rnd, sizeof(rnd));
237 } while (rnd == 0);
238
239 cmpxchg(&inet_ehash_secret, 0, rnd);
240 }
241 EXPORT_SYMBOL(build_ehash_secret);
242
inet_netns_ok(struct net * net,int protocol)243 static inline int inet_netns_ok(struct net *net, int protocol)
244 {
245 int hash;
246 const struct net_protocol *ipprot;
247
248 if (net_eq(net, &init_net))
249 return 1;
250
251 hash = protocol & (MAX_INET_PROTOS - 1);
252 ipprot = rcu_dereference(inet_protos[hash]);
253
254 if (ipprot == NULL)
255 /* raw IP is OK */
256 return 1;
257 return ipprot->netns_ok;
258 }
259
260 /*
261 * Create an inet socket.
262 */
263
inet_create(struct net * net,struct socket * sock,int protocol,int kern)264 static int inet_create(struct net *net, struct socket *sock, int protocol,
265 int kern)
266 {
267 struct sock *sk;
268 struct inet_protosw *answer;
269 struct inet_sock *inet;
270 struct proto *answer_prot;
271 unsigned char answer_flags;
272 char answer_no_check;
273 int try_loading_module = 0;
274 int err;
275
276 if (unlikely(!inet_ehash_secret))
277 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
278 build_ehash_secret();
279
280 sock->state = SS_UNCONNECTED;
281
282 /* Look for the requested type/protocol pair. */
283 lookup_protocol:
284 err = -ESOCKTNOSUPPORT;
285 rcu_read_lock();
286 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
287
288 err = 0;
289 /* Check the non-wild match. */
290 if (protocol == answer->protocol) {
291 if (protocol != IPPROTO_IP)
292 break;
293 } else {
294 /* Check for the two wild cases. */
295 if (IPPROTO_IP == protocol) {
296 protocol = answer->protocol;
297 break;
298 }
299 if (IPPROTO_IP == answer->protocol)
300 break;
301 }
302 err = -EPROTONOSUPPORT;
303 }
304
305 if (unlikely(err)) {
306 if (try_loading_module < 2) {
307 rcu_read_unlock();
308 /*
309 * Be more specific, e.g. net-pf-2-proto-132-type-1
310 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
311 */
312 if (++try_loading_module == 1)
313 request_module("net-pf-%d-proto-%d-type-%d",
314 PF_INET, protocol, sock->type);
315 /*
316 * Fall back to generic, e.g. net-pf-2-proto-132
317 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
318 */
319 else
320 request_module("net-pf-%d-proto-%d",
321 PF_INET, protocol);
322 goto lookup_protocol;
323 } else
324 goto out_rcu_unlock;
325 }
326
327 err = -EPERM;
328 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
329 goto out_rcu_unlock;
330
331 err = -EAFNOSUPPORT;
332 if (!inet_netns_ok(net, protocol))
333 goto out_rcu_unlock;
334
335 sock->ops = answer->ops;
336 answer_prot = answer->prot;
337 answer_no_check = answer->no_check;
338 answer_flags = answer->flags;
339 rcu_read_unlock();
340
341 WARN_ON(answer_prot->slab == NULL);
342
343 err = -ENOBUFS;
344 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
345 if (sk == NULL)
346 goto out;
347
348 err = 0;
349 sk->sk_no_check = answer_no_check;
350 if (INET_PROTOSW_REUSE & answer_flags)
351 sk->sk_reuse = 1;
352
353 inet = inet_sk(sk);
354 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
355
356 inet->nodefrag = 0;
357
358 if (SOCK_RAW == sock->type) {
359 inet->inet_num = protocol;
360 if (IPPROTO_RAW == protocol)
361 inet->hdrincl = 1;
362 }
363
364 if (ipv4_config.no_pmtu_disc)
365 inet->pmtudisc = IP_PMTUDISC_DONT;
366 else
367 inet->pmtudisc = IP_PMTUDISC_WANT;
368
369 inet->inet_id = 0;
370
371 sock_init_data(sock, sk);
372
373 sk->sk_destruct = inet_sock_destruct;
374 sk->sk_protocol = protocol;
375 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
376
377 inet->uc_ttl = -1;
378 inet->mc_loop = 1;
379 inet->mc_ttl = 1;
380 inet->mc_all = 1;
381 inet->mc_index = 0;
382 inet->mc_list = NULL;
383
384 sk_refcnt_debug_inc(sk);
385
386 if (inet->inet_num) {
387 /* It assumes that any protocol which allows
388 * the user to assign a number at socket
389 * creation time automatically
390 * shares.
391 */
392 inet->inet_sport = htons(inet->inet_num);
393 /* Add to protocol hash chains. */
394 sk->sk_prot->hash(sk);
395 }
396
397 if (sk->sk_prot->init) {
398 err = sk->sk_prot->init(sk);
399 if (err)
400 sk_common_release(sk);
401 }
402 out:
403 return err;
404 out_rcu_unlock:
405 rcu_read_unlock();
406 goto out;
407 }
408
409
410 /*
411 * The peer socket should always be NULL (or else). When we call this
412 * function we are destroying the object and from then on nobody
413 * should refer to it.
414 */
inet_release(struct socket * sock)415 int inet_release(struct socket *sock)
416 {
417 struct sock *sk = sock->sk;
418
419 if (sk) {
420 long timeout;
421
422 sock_rps_reset_flow(sk);
423
424 /* Applications forget to leave groups before exiting */
425 ip_mc_drop_socket(sk);
426
427 /* If linger is set, we don't return until the close
428 * is complete. Otherwise we return immediately. The
429 * actually closing is done the same either way.
430 *
431 * If the close is due to the process exiting, we never
432 * linger..
433 */
434 timeout = 0;
435 if (sock_flag(sk, SOCK_LINGER) &&
436 !(current->flags & PF_EXITING))
437 timeout = sk->sk_lingertime;
438 sock->sk = NULL;
439 sk->sk_prot->close(sk, timeout);
440 }
441 return 0;
442 }
443 EXPORT_SYMBOL(inet_release);
444
445 /* It is off by default, see below. */
446 int sysctl_ip_nonlocal_bind __read_mostly;
447 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
448
inet_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)449 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
450 {
451 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
452 struct sock *sk = sock->sk;
453 struct inet_sock *inet = inet_sk(sk);
454 unsigned short snum;
455 int chk_addr_ret;
456 int err;
457
458 /* If the socket has its own bind function then use it. (RAW) */
459 if (sk->sk_prot->bind) {
460 err = sk->sk_prot->bind(sk, uaddr, addr_len);
461 goto out;
462 }
463 err = -EINVAL;
464 if (addr_len < sizeof(struct sockaddr_in))
465 goto out;
466
467 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
468
469 /* Not specified by any standard per-se, however it breaks too
470 * many applications when removed. It is unfortunate since
471 * allowing applications to make a non-local bind solves
472 * several problems with systems using dynamic addressing.
473 * (ie. your servers still start up even if your ISDN link
474 * is temporarily down)
475 */
476 err = -EADDRNOTAVAIL;
477 if (!sysctl_ip_nonlocal_bind &&
478 !(inet->freebind || inet->transparent) &&
479 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
480 chk_addr_ret != RTN_LOCAL &&
481 chk_addr_ret != RTN_MULTICAST &&
482 chk_addr_ret != RTN_BROADCAST)
483 goto out;
484
485 snum = ntohs(addr->sin_port);
486 err = -EACCES;
487 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
488 goto out;
489
490 /* We keep a pair of addresses. rcv_saddr is the one
491 * used by hash lookups, and saddr is used for transmit.
492 *
493 * In the BSD API these are the same except where it
494 * would be illegal to use them (multicast/broadcast) in
495 * which case the sending device address is used.
496 */
497 lock_sock(sk);
498
499 /* Check these errors (active socket, double bind). */
500 err = -EINVAL;
501 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
502 goto out_release_sock;
503
504 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
505 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
506 inet->inet_saddr = 0; /* Use device */
507
508 /* Make sure we are allowed to bind here. */
509 if (sk->sk_prot->get_port(sk, snum)) {
510 inet->inet_saddr = inet->inet_rcv_saddr = 0;
511 err = -EADDRINUSE;
512 goto out_release_sock;
513 }
514
515 if (inet->inet_rcv_saddr)
516 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
517 if (snum)
518 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
519 inet->inet_sport = htons(inet->inet_num);
520 inet->inet_daddr = 0;
521 inet->inet_dport = 0;
522 sk_dst_reset(sk);
523 err = 0;
524 out_release_sock:
525 release_sock(sk);
526 out:
527 return err;
528 }
529 EXPORT_SYMBOL(inet_bind);
530
inet_dgram_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)531 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
532 int addr_len, int flags)
533 {
534 struct sock *sk = sock->sk;
535
536 if (addr_len < sizeof(uaddr->sa_family))
537 return -EINVAL;
538 if (uaddr->sa_family == AF_UNSPEC)
539 return sk->sk_prot->disconnect(sk, flags);
540
541 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
542 return -EAGAIN;
543 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
544 }
545 EXPORT_SYMBOL(inet_dgram_connect);
546
inet_wait_for_connect(struct sock * sk,long timeo)547 static long inet_wait_for_connect(struct sock *sk, long timeo)
548 {
549 DEFINE_WAIT(wait);
550
551 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
552
553 /* Basic assumption: if someone sets sk->sk_err, he _must_
554 * change state of the socket from TCP_SYN_*.
555 * Connect() does not allow to get error notifications
556 * without closing the socket.
557 */
558 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
559 release_sock(sk);
560 timeo = schedule_timeout(timeo);
561 lock_sock(sk);
562 if (signal_pending(current) || !timeo)
563 break;
564 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
565 }
566 finish_wait(sk_sleep(sk), &wait);
567 return timeo;
568 }
569
570 /*
571 * Connect to a remote host. There is regrettably still a little
572 * TCP 'magic' in here.
573 */
inet_stream_connect(struct socket * sock,struct sockaddr * uaddr,int addr_len,int flags)574 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
575 int addr_len, int flags)
576 {
577 struct sock *sk = sock->sk;
578 int err;
579 long timeo;
580
581 if (addr_len < sizeof(uaddr->sa_family))
582 return -EINVAL;
583
584 lock_sock(sk);
585
586 if (uaddr->sa_family == AF_UNSPEC) {
587 err = sk->sk_prot->disconnect(sk, flags);
588 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
589 goto out;
590 }
591
592 switch (sock->state) {
593 default:
594 err = -EINVAL;
595 goto out;
596 case SS_CONNECTED:
597 err = -EISCONN;
598 goto out;
599 case SS_CONNECTING:
600 err = -EALREADY;
601 /* Fall out of switch with err, set for this state */
602 break;
603 case SS_UNCONNECTED:
604 err = -EISCONN;
605 if (sk->sk_state != TCP_CLOSE)
606 goto out;
607
608 err = sk->sk_prot->connect(sk, uaddr, addr_len);
609 if (err < 0)
610 goto out;
611
612 sock->state = SS_CONNECTING;
613
614 /* Just entered SS_CONNECTING state; the only
615 * difference is that return value in non-blocking
616 * case is EINPROGRESS, rather than EALREADY.
617 */
618 err = -EINPROGRESS;
619 break;
620 }
621
622 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
623
624 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
625 /* Error code is set above */
626 if (!timeo || !inet_wait_for_connect(sk, timeo))
627 goto out;
628
629 err = sock_intr_errno(timeo);
630 if (signal_pending(current))
631 goto out;
632 }
633
634 /* Connection was closed by RST, timeout, ICMP error
635 * or another process disconnected us.
636 */
637 if (sk->sk_state == TCP_CLOSE)
638 goto sock_error;
639
640 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
641 * and error was received after socket entered established state.
642 * Hence, it is handled normally after connect() return successfully.
643 */
644
645 sock->state = SS_CONNECTED;
646 err = 0;
647 out:
648 release_sock(sk);
649 return err;
650
651 sock_error:
652 err = sock_error(sk) ? : -ECONNABORTED;
653 sock->state = SS_UNCONNECTED;
654 if (sk->sk_prot->disconnect(sk, flags))
655 sock->state = SS_DISCONNECTING;
656 goto out;
657 }
658 EXPORT_SYMBOL(inet_stream_connect);
659
660 /*
661 * Accept a pending connection. The TCP layer now gives BSD semantics.
662 */
663
inet_accept(struct socket * sock,struct socket * newsock,int flags)664 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
665 {
666 struct sock *sk1 = sock->sk;
667 int err = -EINVAL;
668 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
669
670 if (!sk2)
671 goto do_err;
672
673 lock_sock(sk2);
674
675 WARN_ON(!((1 << sk2->sk_state) &
676 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
677
678 sock_graft(sk2, newsock);
679
680 newsock->state = SS_CONNECTED;
681 err = 0;
682 release_sock(sk2);
683 do_err:
684 return err;
685 }
686 EXPORT_SYMBOL(inet_accept);
687
688
689 /*
690 * This does both peername and sockname.
691 */
inet_getname(struct socket * sock,struct sockaddr * uaddr,int * uaddr_len,int peer)692 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
693 int *uaddr_len, int peer)
694 {
695 struct sock *sk = sock->sk;
696 struct inet_sock *inet = inet_sk(sk);
697 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
698
699 sin->sin_family = AF_INET;
700 if (peer) {
701 if (!inet->inet_dport ||
702 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
703 peer == 1))
704 return -ENOTCONN;
705 sin->sin_port = inet->inet_dport;
706 sin->sin_addr.s_addr = inet->inet_daddr;
707 } else {
708 __be32 addr = inet->inet_rcv_saddr;
709 if (!addr)
710 addr = inet->inet_saddr;
711 sin->sin_port = inet->inet_sport;
712 sin->sin_addr.s_addr = addr;
713 }
714 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
715 *uaddr_len = sizeof(*sin);
716 return 0;
717 }
718 EXPORT_SYMBOL(inet_getname);
719
inet_sendmsg(struct kiocb * iocb,struct socket * sock,struct msghdr * msg,size_t size)720 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
721 size_t size)
722 {
723 struct sock *sk = sock->sk;
724
725 sock_rps_record_flow(sk);
726
727 /* We may need to bind the socket. */
728 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
729 inet_autobind(sk))
730 return -EAGAIN;
731
732 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
733 }
734 EXPORT_SYMBOL(inet_sendmsg);
735
inet_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)736 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
737 size_t size, int flags)
738 {
739 struct sock *sk = sock->sk;
740
741 sock_rps_record_flow(sk);
742
743 /* We may need to bind the socket. */
744 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
745 inet_autobind(sk))
746 return -EAGAIN;
747
748 if (sk->sk_prot->sendpage)
749 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
750 return sock_no_sendpage(sock, page, offset, size, flags);
751 }
752 EXPORT_SYMBOL(inet_sendpage);
753
inet_recvmsg(struct kiocb * iocb,struct socket * sock,struct msghdr * msg,size_t size,int flags)754 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
755 size_t size, int flags)
756 {
757 struct sock *sk = sock->sk;
758 int addr_len = 0;
759 int err;
760
761 sock_rps_record_flow(sk);
762
763 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
764 flags & ~MSG_DONTWAIT, &addr_len);
765 if (err >= 0)
766 msg->msg_namelen = addr_len;
767 return err;
768 }
769 EXPORT_SYMBOL(inet_recvmsg);
770
inet_shutdown(struct socket * sock,int how)771 int inet_shutdown(struct socket *sock, int how)
772 {
773 struct sock *sk = sock->sk;
774 int err = 0;
775
776 /* This should really check to make sure
777 * the socket is a TCP socket. (WHY AC...)
778 */
779 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
780 1->2 bit 2 snds.
781 2->3 */
782 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
783 return -EINVAL;
784
785 lock_sock(sk);
786 if (sock->state == SS_CONNECTING) {
787 if ((1 << sk->sk_state) &
788 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
789 sock->state = SS_DISCONNECTING;
790 else
791 sock->state = SS_CONNECTED;
792 }
793
794 switch (sk->sk_state) {
795 case TCP_CLOSE:
796 err = -ENOTCONN;
797 /* Hack to wake up other listeners, who can poll for
798 POLLHUP, even on eg. unconnected UDP sockets -- RR */
799 default:
800 sk->sk_shutdown |= how;
801 if (sk->sk_prot->shutdown)
802 sk->sk_prot->shutdown(sk, how);
803 break;
804
805 /* Remaining two branches are temporary solution for missing
806 * close() in multithreaded environment. It is _not_ a good idea,
807 * but we have no choice until close() is repaired at VFS level.
808 */
809 case TCP_LISTEN:
810 if (!(how & RCV_SHUTDOWN))
811 break;
812 /* Fall through */
813 case TCP_SYN_SENT:
814 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
815 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
816 break;
817 }
818
819 /* Wake up anyone sleeping in poll. */
820 sk->sk_state_change(sk);
821 release_sock(sk);
822 return err;
823 }
824 EXPORT_SYMBOL(inet_shutdown);
825
826 /*
827 * ioctl() calls you can issue on an INET socket. Most of these are
828 * device configuration and stuff and very rarely used. Some ioctls
829 * pass on to the socket itself.
830 *
831 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
832 * loads the devconfigure module does its configuring and unloads it.
833 * There's a good 20K of config code hanging around the kernel.
834 */
835
inet_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)836 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
837 {
838 struct sock *sk = sock->sk;
839 int err = 0;
840 struct net *net = sock_net(sk);
841
842 switch (cmd) {
843 case SIOCGSTAMP:
844 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
845 break;
846 case SIOCGSTAMPNS:
847 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
848 break;
849 case SIOCADDRT:
850 case SIOCDELRT:
851 case SIOCRTMSG:
852 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
853 break;
854 case SIOCDARP:
855 case SIOCGARP:
856 case SIOCSARP:
857 err = arp_ioctl(net, cmd, (void __user *)arg);
858 break;
859 case SIOCGIFADDR:
860 case SIOCSIFADDR:
861 case SIOCGIFBRDADDR:
862 case SIOCSIFBRDADDR:
863 case SIOCGIFNETMASK:
864 case SIOCSIFNETMASK:
865 case SIOCGIFDSTADDR:
866 case SIOCSIFDSTADDR:
867 case SIOCSIFPFLAGS:
868 case SIOCGIFPFLAGS:
869 case SIOCSIFFLAGS:
870 err = devinet_ioctl(net, cmd, (void __user *)arg);
871 break;
872 default:
873 if (sk->sk_prot->ioctl)
874 err = sk->sk_prot->ioctl(sk, cmd, arg);
875 else
876 err = -ENOIOCTLCMD;
877 break;
878 }
879 return err;
880 }
881 EXPORT_SYMBOL(inet_ioctl);
882
883 #ifdef CONFIG_COMPAT
inet_compat_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)884 int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
885 {
886 struct sock *sk = sock->sk;
887 int err = -ENOIOCTLCMD;
888
889 if (sk->sk_prot->compat_ioctl)
890 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
891
892 return err;
893 }
894 #endif
895
896 const struct proto_ops inet_stream_ops = {
897 .family = PF_INET,
898 .owner = THIS_MODULE,
899 .release = inet_release,
900 .bind = inet_bind,
901 .connect = inet_stream_connect,
902 .socketpair = sock_no_socketpair,
903 .accept = inet_accept,
904 .getname = inet_getname,
905 .poll = tcp_poll,
906 .ioctl = inet_ioctl,
907 .listen = inet_listen,
908 .shutdown = inet_shutdown,
909 .setsockopt = sock_common_setsockopt,
910 .getsockopt = sock_common_getsockopt,
911 .sendmsg = inet_sendmsg,
912 .recvmsg = inet_recvmsg,
913 .mmap = sock_no_mmap,
914 .sendpage = inet_sendpage,
915 .splice_read = tcp_splice_read,
916 #ifdef CONFIG_COMPAT
917 .compat_setsockopt = compat_sock_common_setsockopt,
918 .compat_getsockopt = compat_sock_common_getsockopt,
919 .compat_ioctl = inet_compat_ioctl,
920 #endif
921 };
922 EXPORT_SYMBOL(inet_stream_ops);
923
924 const struct proto_ops inet_dgram_ops = {
925 .family = PF_INET,
926 .owner = THIS_MODULE,
927 .release = inet_release,
928 .bind = inet_bind,
929 .connect = inet_dgram_connect,
930 .socketpair = sock_no_socketpair,
931 .accept = sock_no_accept,
932 .getname = inet_getname,
933 .poll = udp_poll,
934 .ioctl = inet_ioctl,
935 .listen = sock_no_listen,
936 .shutdown = inet_shutdown,
937 .setsockopt = sock_common_setsockopt,
938 .getsockopt = sock_common_getsockopt,
939 .sendmsg = inet_sendmsg,
940 .recvmsg = inet_recvmsg,
941 .mmap = sock_no_mmap,
942 .sendpage = inet_sendpage,
943 #ifdef CONFIG_COMPAT
944 .compat_setsockopt = compat_sock_common_setsockopt,
945 .compat_getsockopt = compat_sock_common_getsockopt,
946 .compat_ioctl = inet_compat_ioctl,
947 #endif
948 };
949 EXPORT_SYMBOL(inet_dgram_ops);
950
951 /*
952 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
953 * udp_poll
954 */
955 static const struct proto_ops inet_sockraw_ops = {
956 .family = PF_INET,
957 .owner = THIS_MODULE,
958 .release = inet_release,
959 .bind = inet_bind,
960 .connect = inet_dgram_connect,
961 .socketpair = sock_no_socketpair,
962 .accept = sock_no_accept,
963 .getname = inet_getname,
964 .poll = datagram_poll,
965 .ioctl = inet_ioctl,
966 .listen = sock_no_listen,
967 .shutdown = inet_shutdown,
968 .setsockopt = sock_common_setsockopt,
969 .getsockopt = sock_common_getsockopt,
970 .sendmsg = inet_sendmsg,
971 .recvmsg = inet_recvmsg,
972 .mmap = sock_no_mmap,
973 .sendpage = inet_sendpage,
974 #ifdef CONFIG_COMPAT
975 .compat_setsockopt = compat_sock_common_setsockopt,
976 .compat_getsockopt = compat_sock_common_getsockopt,
977 .compat_ioctl = inet_compat_ioctl,
978 #endif
979 };
980
981 static const struct net_proto_family inet_family_ops = {
982 .family = PF_INET,
983 .create = inet_create,
984 .owner = THIS_MODULE,
985 };
986
987 /* Upon startup we insert all the elements in inetsw_array[] into
988 * the linked list inetsw.
989 */
990 static struct inet_protosw inetsw_array[] =
991 {
992 {
993 .type = SOCK_STREAM,
994 .protocol = IPPROTO_TCP,
995 .prot = &tcp_prot,
996 .ops = &inet_stream_ops,
997 .no_check = 0,
998 .flags = INET_PROTOSW_PERMANENT |
999 INET_PROTOSW_ICSK,
1000 },
1001
1002 {
1003 .type = SOCK_DGRAM,
1004 .protocol = IPPROTO_UDP,
1005 .prot = &udp_prot,
1006 .ops = &inet_dgram_ops,
1007 .no_check = UDP_CSUM_DEFAULT,
1008 .flags = INET_PROTOSW_PERMANENT,
1009 },
1010
1011
1012 {
1013 .type = SOCK_RAW,
1014 .protocol = IPPROTO_IP, /* wild card */
1015 .prot = &raw_prot,
1016 .ops = &inet_sockraw_ops,
1017 .no_check = UDP_CSUM_DEFAULT,
1018 .flags = INET_PROTOSW_REUSE,
1019 }
1020 };
1021
1022 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1023
inet_register_protosw(struct inet_protosw * p)1024 void inet_register_protosw(struct inet_protosw *p)
1025 {
1026 struct list_head *lh;
1027 struct inet_protosw *answer;
1028 int protocol = p->protocol;
1029 struct list_head *last_perm;
1030
1031 spin_lock_bh(&inetsw_lock);
1032
1033 if (p->type >= SOCK_MAX)
1034 goto out_illegal;
1035
1036 /* If we are trying to override a permanent protocol, bail. */
1037 answer = NULL;
1038 last_perm = &inetsw[p->type];
1039 list_for_each(lh, &inetsw[p->type]) {
1040 answer = list_entry(lh, struct inet_protosw, list);
1041
1042 /* Check only the non-wild match. */
1043 if (INET_PROTOSW_PERMANENT & answer->flags) {
1044 if (protocol == answer->protocol)
1045 break;
1046 last_perm = lh;
1047 }
1048
1049 answer = NULL;
1050 }
1051 if (answer)
1052 goto out_permanent;
1053
1054 /* Add the new entry after the last permanent entry if any, so that
1055 * the new entry does not override a permanent entry when matched with
1056 * a wild-card protocol. But it is allowed to override any existing
1057 * non-permanent entry. This means that when we remove this entry, the
1058 * system automatically returns to the old behavior.
1059 */
1060 list_add_rcu(&p->list, last_perm);
1061 out:
1062 spin_unlock_bh(&inetsw_lock);
1063
1064 return;
1065
1066 out_permanent:
1067 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1068 protocol);
1069 goto out;
1070
1071 out_illegal:
1072 printk(KERN_ERR
1073 "Ignoring attempt to register invalid socket type %d.\n",
1074 p->type);
1075 goto out;
1076 }
1077 EXPORT_SYMBOL(inet_register_protosw);
1078
inet_unregister_protosw(struct inet_protosw * p)1079 void inet_unregister_protosw(struct inet_protosw *p)
1080 {
1081 if (INET_PROTOSW_PERMANENT & p->flags) {
1082 printk(KERN_ERR
1083 "Attempt to unregister permanent protocol %d.\n",
1084 p->protocol);
1085 } else {
1086 spin_lock_bh(&inetsw_lock);
1087 list_del_rcu(&p->list);
1088 spin_unlock_bh(&inetsw_lock);
1089
1090 synchronize_net();
1091 }
1092 }
1093 EXPORT_SYMBOL(inet_unregister_protosw);
1094
1095 /*
1096 * Shall we try to damage output packets if routing dev changes?
1097 */
1098
1099 int sysctl_ip_dynaddr __read_mostly;
1100
inet_sk_reselect_saddr(struct sock * sk)1101 static int inet_sk_reselect_saddr(struct sock *sk)
1102 {
1103 struct inet_sock *inet = inet_sk(sk);
1104 __be32 old_saddr = inet->inet_saddr;
1105 __be32 daddr = inet->inet_daddr;
1106 struct rtable *rt;
1107 __be32 new_saddr;
1108
1109 if (inet->opt && inet->opt->srr)
1110 daddr = inet->opt->faddr;
1111
1112 /* Query new route. */
1113 rt = ip_route_connect(daddr, 0, RT_CONN_FLAGS(sk),
1114 sk->sk_bound_dev_if, sk->sk_protocol,
1115 inet->inet_sport, inet->inet_dport, sk, false);
1116 if (IS_ERR(rt))
1117 return PTR_ERR(rt);
1118
1119 sk_setup_caps(sk, &rt->dst);
1120
1121 new_saddr = rt->rt_src;
1122
1123 if (new_saddr == old_saddr)
1124 return 0;
1125
1126 if (sysctl_ip_dynaddr > 1) {
1127 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1128 __func__, &old_saddr, &new_saddr);
1129 }
1130
1131 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1132
1133 /*
1134 * XXX The only one ugly spot where we need to
1135 * XXX really change the sockets identity after
1136 * XXX it has entered the hashes. -DaveM
1137 *
1138 * Besides that, it does not check for connection
1139 * uniqueness. Wait for troubles.
1140 */
1141 __sk_prot_rehash(sk);
1142 return 0;
1143 }
1144
inet_sk_rebuild_header(struct sock * sk)1145 int inet_sk_rebuild_header(struct sock *sk)
1146 {
1147 struct inet_sock *inet = inet_sk(sk);
1148 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1149 __be32 daddr;
1150 int err;
1151
1152 /* Route is OK, nothing to do. */
1153 if (rt)
1154 return 0;
1155
1156 /* Reroute. */
1157 daddr = inet->inet_daddr;
1158 if (inet->opt && inet->opt->srr)
1159 daddr = inet->opt->faddr;
1160 rt = ip_route_output_ports(sock_net(sk), sk, daddr, inet->inet_saddr,
1161 inet->inet_dport, inet->inet_sport,
1162 sk->sk_protocol, RT_CONN_FLAGS(sk),
1163 sk->sk_bound_dev_if);
1164 if (!IS_ERR(rt)) {
1165 err = 0;
1166 sk_setup_caps(sk, &rt->dst);
1167 } else {
1168 err = PTR_ERR(rt);
1169
1170 /* Routing failed... */
1171 sk->sk_route_caps = 0;
1172 /*
1173 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1174 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1175 */
1176 if (!sysctl_ip_dynaddr ||
1177 sk->sk_state != TCP_SYN_SENT ||
1178 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1179 (err = inet_sk_reselect_saddr(sk)) != 0)
1180 sk->sk_err_soft = -err;
1181 }
1182
1183 return err;
1184 }
1185 EXPORT_SYMBOL(inet_sk_rebuild_header);
1186
inet_gso_send_check(struct sk_buff * skb)1187 static int inet_gso_send_check(struct sk_buff *skb)
1188 {
1189 struct iphdr *iph;
1190 const struct net_protocol *ops;
1191 int proto;
1192 int ihl;
1193 int err = -EINVAL;
1194
1195 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1196 goto out;
1197
1198 iph = ip_hdr(skb);
1199 ihl = iph->ihl * 4;
1200 if (ihl < sizeof(*iph))
1201 goto out;
1202
1203 if (unlikely(!pskb_may_pull(skb, ihl)))
1204 goto out;
1205
1206 __skb_pull(skb, ihl);
1207 skb_reset_transport_header(skb);
1208 iph = ip_hdr(skb);
1209 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1210 err = -EPROTONOSUPPORT;
1211
1212 rcu_read_lock();
1213 ops = rcu_dereference(inet_protos[proto]);
1214 if (likely(ops && ops->gso_send_check))
1215 err = ops->gso_send_check(skb);
1216 rcu_read_unlock();
1217
1218 out:
1219 return err;
1220 }
1221
inet_gso_segment(struct sk_buff * skb,u32 features)1222 static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1223 {
1224 struct sk_buff *segs = ERR_PTR(-EINVAL);
1225 struct iphdr *iph;
1226 const struct net_protocol *ops;
1227 int proto;
1228 int ihl;
1229 int id;
1230 unsigned int offset = 0;
1231
1232 if (!(features & NETIF_F_V4_CSUM))
1233 features &= ~NETIF_F_SG;
1234
1235 if (unlikely(skb_shinfo(skb)->gso_type &
1236 ~(SKB_GSO_TCPV4 |
1237 SKB_GSO_UDP |
1238 SKB_GSO_DODGY |
1239 SKB_GSO_TCP_ECN |
1240 0)))
1241 goto out;
1242
1243 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1244 goto out;
1245
1246 iph = ip_hdr(skb);
1247 ihl = iph->ihl * 4;
1248 if (ihl < sizeof(*iph))
1249 goto out;
1250
1251 if (unlikely(!pskb_may_pull(skb, ihl)))
1252 goto out;
1253
1254 __skb_pull(skb, ihl);
1255 skb_reset_transport_header(skb);
1256 iph = ip_hdr(skb);
1257 id = ntohs(iph->id);
1258 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1259 segs = ERR_PTR(-EPROTONOSUPPORT);
1260
1261 rcu_read_lock();
1262 ops = rcu_dereference(inet_protos[proto]);
1263 if (likely(ops && ops->gso_segment))
1264 segs = ops->gso_segment(skb, features);
1265 rcu_read_unlock();
1266
1267 if (!segs || IS_ERR(segs))
1268 goto out;
1269
1270 skb = segs;
1271 do {
1272 iph = ip_hdr(skb);
1273 if (proto == IPPROTO_UDP) {
1274 iph->id = htons(id);
1275 iph->frag_off = htons(offset >> 3);
1276 if (skb->next != NULL)
1277 iph->frag_off |= htons(IP_MF);
1278 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1279 } else
1280 iph->id = htons(id++);
1281 iph->tot_len = htons(skb->len - skb->mac_len);
1282 iph->check = 0;
1283 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1284 } while ((skb = skb->next));
1285
1286 out:
1287 return segs;
1288 }
1289
inet_gro_receive(struct sk_buff ** head,struct sk_buff * skb)1290 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1291 struct sk_buff *skb)
1292 {
1293 const struct net_protocol *ops;
1294 struct sk_buff **pp = NULL;
1295 struct sk_buff *p;
1296 struct iphdr *iph;
1297 unsigned int hlen;
1298 unsigned int off;
1299 unsigned int id;
1300 int flush = 1;
1301 int proto;
1302
1303 off = skb_gro_offset(skb);
1304 hlen = off + sizeof(*iph);
1305 iph = skb_gro_header_fast(skb, off);
1306 if (skb_gro_header_hard(skb, hlen)) {
1307 iph = skb_gro_header_slow(skb, hlen, off);
1308 if (unlikely(!iph))
1309 goto out;
1310 }
1311
1312 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1313
1314 rcu_read_lock();
1315 ops = rcu_dereference(inet_protos[proto]);
1316 if (!ops || !ops->gro_receive)
1317 goto out_unlock;
1318
1319 if (*(u8 *)iph != 0x45)
1320 goto out_unlock;
1321
1322 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1323 goto out_unlock;
1324
1325 id = ntohl(*(__be32 *)&iph->id);
1326 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1327 id >>= 16;
1328
1329 for (p = *head; p; p = p->next) {
1330 struct iphdr *iph2;
1331
1332 if (!NAPI_GRO_CB(p)->same_flow)
1333 continue;
1334
1335 iph2 = ip_hdr(p);
1336
1337 if ((iph->protocol ^ iph2->protocol) |
1338 (iph->tos ^ iph2->tos) |
1339 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1340 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1341 NAPI_GRO_CB(p)->same_flow = 0;
1342 continue;
1343 }
1344
1345 /* All fields must match except length and checksum. */
1346 NAPI_GRO_CB(p)->flush |=
1347 (iph->ttl ^ iph2->ttl) |
1348 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1349
1350 NAPI_GRO_CB(p)->flush |= flush;
1351 }
1352
1353 NAPI_GRO_CB(skb)->flush |= flush;
1354 skb_gro_pull(skb, sizeof(*iph));
1355 skb_set_transport_header(skb, skb_gro_offset(skb));
1356
1357 pp = ops->gro_receive(head, skb);
1358
1359 out_unlock:
1360 rcu_read_unlock();
1361
1362 out:
1363 NAPI_GRO_CB(skb)->flush |= flush;
1364
1365 return pp;
1366 }
1367
inet_gro_complete(struct sk_buff * skb)1368 static int inet_gro_complete(struct sk_buff *skb)
1369 {
1370 const struct net_protocol *ops;
1371 struct iphdr *iph = ip_hdr(skb);
1372 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1373 int err = -ENOSYS;
1374 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1375
1376 csum_replace2(&iph->check, iph->tot_len, newlen);
1377 iph->tot_len = newlen;
1378
1379 rcu_read_lock();
1380 ops = rcu_dereference(inet_protos[proto]);
1381 if (WARN_ON(!ops || !ops->gro_complete))
1382 goto out_unlock;
1383
1384 err = ops->gro_complete(skb);
1385
1386 out_unlock:
1387 rcu_read_unlock();
1388
1389 return err;
1390 }
1391
inet_ctl_sock_create(struct sock ** sk,unsigned short family,unsigned short type,unsigned char protocol,struct net * net)1392 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1393 unsigned short type, unsigned char protocol,
1394 struct net *net)
1395 {
1396 struct socket *sock;
1397 int rc = sock_create_kern(family, type, protocol, &sock);
1398
1399 if (rc == 0) {
1400 *sk = sock->sk;
1401 (*sk)->sk_allocation = GFP_ATOMIC;
1402 /*
1403 * Unhash it so that IP input processing does not even see it,
1404 * we do not wish this socket to see incoming packets.
1405 */
1406 (*sk)->sk_prot->unhash(*sk);
1407
1408 sk_change_net(*sk, net);
1409 }
1410 return rc;
1411 }
1412 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1413
snmp_fold_field(void __percpu * mib[],int offt)1414 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1415 {
1416 unsigned long res = 0;
1417 int i;
1418
1419 for_each_possible_cpu(i) {
1420 res += *(((unsigned long *) per_cpu_ptr(mib[0], i)) + offt);
1421 res += *(((unsigned long *) per_cpu_ptr(mib[1], i)) + offt);
1422 }
1423 return res;
1424 }
1425 EXPORT_SYMBOL_GPL(snmp_fold_field);
1426
1427 #if BITS_PER_LONG==32
1428
snmp_fold_field64(void __percpu * mib[],int offt,size_t syncp_offset)1429 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1430 {
1431 u64 res = 0;
1432 int cpu;
1433
1434 for_each_possible_cpu(cpu) {
1435 void *bhptr, *userptr;
1436 struct u64_stats_sync *syncp;
1437 u64 v_bh, v_user;
1438 unsigned int start;
1439
1440 /* first mib used by softirq context, we must use _bh() accessors */
1441 bhptr = per_cpu_ptr(SNMP_STAT_BHPTR(mib), cpu);
1442 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1443 do {
1444 start = u64_stats_fetch_begin_bh(syncp);
1445 v_bh = *(((u64 *) bhptr) + offt);
1446 } while (u64_stats_fetch_retry_bh(syncp, start));
1447
1448 /* second mib used in USER context */
1449 userptr = per_cpu_ptr(SNMP_STAT_USRPTR(mib), cpu);
1450 syncp = (struct u64_stats_sync *)(userptr + syncp_offset);
1451 do {
1452 start = u64_stats_fetch_begin(syncp);
1453 v_user = *(((u64 *) userptr) + offt);
1454 } while (u64_stats_fetch_retry(syncp, start));
1455
1456 res += v_bh + v_user;
1457 }
1458 return res;
1459 }
1460 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1461 #endif
1462
snmp_mib_init(void __percpu * ptr[2],size_t mibsize,size_t align)1463 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1464 {
1465 BUG_ON(ptr == NULL);
1466 ptr[0] = __alloc_percpu(mibsize, align);
1467 if (!ptr[0])
1468 goto err0;
1469 ptr[1] = __alloc_percpu(mibsize, align);
1470 if (!ptr[1])
1471 goto err1;
1472 return 0;
1473 err1:
1474 free_percpu(ptr[0]);
1475 ptr[0] = NULL;
1476 err0:
1477 return -ENOMEM;
1478 }
1479 EXPORT_SYMBOL_GPL(snmp_mib_init);
1480
snmp_mib_free(void __percpu * ptr[2])1481 void snmp_mib_free(void __percpu *ptr[2])
1482 {
1483 BUG_ON(ptr == NULL);
1484 free_percpu(ptr[0]);
1485 free_percpu(ptr[1]);
1486 ptr[0] = ptr[1] = NULL;
1487 }
1488 EXPORT_SYMBOL_GPL(snmp_mib_free);
1489
1490 #ifdef CONFIG_IP_MULTICAST
1491 static const struct net_protocol igmp_protocol = {
1492 .handler = igmp_rcv,
1493 .netns_ok = 1,
1494 };
1495 #endif
1496
1497 static const struct net_protocol tcp_protocol = {
1498 .handler = tcp_v4_rcv,
1499 .err_handler = tcp_v4_err,
1500 .gso_send_check = tcp_v4_gso_send_check,
1501 .gso_segment = tcp_tso_segment,
1502 .gro_receive = tcp4_gro_receive,
1503 .gro_complete = tcp4_gro_complete,
1504 .no_policy = 1,
1505 .netns_ok = 1,
1506 };
1507
1508 static const struct net_protocol udp_protocol = {
1509 .handler = udp_rcv,
1510 .err_handler = udp_err,
1511 .gso_send_check = udp4_ufo_send_check,
1512 .gso_segment = udp4_ufo_fragment,
1513 .no_policy = 1,
1514 .netns_ok = 1,
1515 };
1516
1517 static const struct net_protocol icmp_protocol = {
1518 .handler = icmp_rcv,
1519 .no_policy = 1,
1520 .netns_ok = 1,
1521 };
1522
ipv4_mib_init_net(struct net * net)1523 static __net_init int ipv4_mib_init_net(struct net *net)
1524 {
1525 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1526 sizeof(struct tcp_mib),
1527 __alignof__(struct tcp_mib)) < 0)
1528 goto err_tcp_mib;
1529 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1530 sizeof(struct ipstats_mib),
1531 __alignof__(struct ipstats_mib)) < 0)
1532 goto err_ip_mib;
1533 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1534 sizeof(struct linux_mib),
1535 __alignof__(struct linux_mib)) < 0)
1536 goto err_net_mib;
1537 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1538 sizeof(struct udp_mib),
1539 __alignof__(struct udp_mib)) < 0)
1540 goto err_udp_mib;
1541 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1542 sizeof(struct udp_mib),
1543 __alignof__(struct udp_mib)) < 0)
1544 goto err_udplite_mib;
1545 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1546 sizeof(struct icmp_mib),
1547 __alignof__(struct icmp_mib)) < 0)
1548 goto err_icmp_mib;
1549 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1550 sizeof(struct icmpmsg_mib),
1551 __alignof__(struct icmpmsg_mib)) < 0)
1552 goto err_icmpmsg_mib;
1553
1554 tcp_mib_init(net);
1555 return 0;
1556
1557 err_icmpmsg_mib:
1558 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1559 err_icmp_mib:
1560 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1561 err_udplite_mib:
1562 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1563 err_udp_mib:
1564 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1565 err_net_mib:
1566 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1567 err_ip_mib:
1568 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1569 err_tcp_mib:
1570 return -ENOMEM;
1571 }
1572
ipv4_mib_exit_net(struct net * net)1573 static __net_exit void ipv4_mib_exit_net(struct net *net)
1574 {
1575 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1576 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1577 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1578 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1579 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1580 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1581 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1582 }
1583
1584 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1585 .init = ipv4_mib_init_net,
1586 .exit = ipv4_mib_exit_net,
1587 };
1588
init_ipv4_mibs(void)1589 static int __init init_ipv4_mibs(void)
1590 {
1591 return register_pernet_subsys(&ipv4_mib_ops);
1592 }
1593
1594 static int ipv4_proc_init(void);
1595
1596 /*
1597 * IP protocol layer initialiser
1598 */
1599
1600 static struct packet_type ip_packet_type __read_mostly = {
1601 .type = cpu_to_be16(ETH_P_IP),
1602 .func = ip_rcv,
1603 .gso_send_check = inet_gso_send_check,
1604 .gso_segment = inet_gso_segment,
1605 .gro_receive = inet_gro_receive,
1606 .gro_complete = inet_gro_complete,
1607 };
1608
inet_init(void)1609 static int __init inet_init(void)
1610 {
1611 struct sk_buff *dummy_skb;
1612 struct inet_protosw *q;
1613 struct list_head *r;
1614 int rc = -EINVAL;
1615
1616 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1617
1618 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1619 if (!sysctl_local_reserved_ports)
1620 goto out;
1621
1622 rc = proto_register(&tcp_prot, 1);
1623 if (rc)
1624 goto out_free_reserved_ports;
1625
1626 rc = proto_register(&udp_prot, 1);
1627 if (rc)
1628 goto out_unregister_tcp_proto;
1629
1630 rc = proto_register(&raw_prot, 1);
1631 if (rc)
1632 goto out_unregister_udp_proto;
1633
1634 /*
1635 * Tell SOCKET that we are alive...
1636 */
1637
1638 (void)sock_register(&inet_family_ops);
1639
1640 #ifdef CONFIG_SYSCTL
1641 ip_static_sysctl_init();
1642 #endif
1643
1644 /*
1645 * Add all the base protocols.
1646 */
1647
1648 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1649 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1650 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1651 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1652 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1653 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1654 #ifdef CONFIG_IP_MULTICAST
1655 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1656 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1657 #endif
1658
1659 /* Register the socket-side information for inet_create. */
1660 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1661 INIT_LIST_HEAD(r);
1662
1663 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1664 inet_register_protosw(q);
1665
1666 /*
1667 * Set the ARP module up
1668 */
1669
1670 arp_init();
1671
1672 /*
1673 * Set the IP module up
1674 */
1675
1676 ip_init();
1677
1678 tcp_v4_init();
1679
1680 /* Setup TCP slab cache for open requests. */
1681 tcp_init();
1682
1683 /* Setup UDP memory threshold */
1684 udp_init();
1685
1686 /* Add UDP-Lite (RFC 3828) */
1687 udplite4_register();
1688
1689 /*
1690 * Set the ICMP layer up
1691 */
1692
1693 if (icmp_init() < 0)
1694 panic("Failed to create the ICMP control socket.\n");
1695
1696 /*
1697 * Initialise the multicast router
1698 */
1699 #if defined(CONFIG_IP_MROUTE)
1700 if (ip_mr_init())
1701 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1702 #endif
1703 /*
1704 * Initialise per-cpu ipv4 mibs
1705 */
1706
1707 if (init_ipv4_mibs())
1708 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1709
1710 ipv4_proc_init();
1711
1712 ipfrag_init();
1713
1714 dev_add_pack(&ip_packet_type);
1715
1716 rc = 0;
1717 out:
1718 return rc;
1719 out_unregister_udp_proto:
1720 proto_unregister(&udp_prot);
1721 out_unregister_tcp_proto:
1722 proto_unregister(&tcp_prot);
1723 out_free_reserved_ports:
1724 kfree(sysctl_local_reserved_ports);
1725 goto out;
1726 }
1727
1728 fs_initcall(inet_init);
1729
1730 /* ------------------------------------------------------------------------ */
1731
1732 #ifdef CONFIG_PROC_FS
ipv4_proc_init(void)1733 static int __init ipv4_proc_init(void)
1734 {
1735 int rc = 0;
1736
1737 if (raw_proc_init())
1738 goto out_raw;
1739 if (tcp4_proc_init())
1740 goto out_tcp;
1741 if (udp4_proc_init())
1742 goto out_udp;
1743 if (ip_misc_proc_init())
1744 goto out_misc;
1745 out:
1746 return rc;
1747 out_misc:
1748 udp4_proc_exit();
1749 out_udp:
1750 tcp4_proc_exit();
1751 out_tcp:
1752 raw_proc_exit();
1753 out_raw:
1754 rc = -ENOMEM;
1755 goto out;
1756 }
1757
1758 #else /* CONFIG_PROC_FS */
ipv4_proc_init(void)1759 static int __init ipv4_proc_init(void)
1760 {
1761 return 0;
1762 }
1763 #endif /* CONFIG_PROC_FS */
1764
1765 MODULE_ALIAS_NETPROTO(PF_INET);
1766
1767