projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge tag 'v3.10.95' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ipv4 / af_inet.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PF_INET protocol family socket handler.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
12 *
13 * Changes (see also sock.c)
14 *
15 * piggy,
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
23 * structures
24 * when accept() ed
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
30 * Alan Cox,
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
40 * compactness.
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
46 * dumbly.
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
61 *
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
66 */
67
68 #define pr_fmt(fmt) "IPv4: " fmt
69
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/kernel.h>
76 #include <linux/module.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
84 #include <linux/mm.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
92
93 #include <asm/uaccess.h>
94
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/inet_common.h>
115 #include <net/xfrm.h>
116 #include <net/net_namespace.h>
117 #include <net/secure_seq.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
121
122 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
123 #include <linux/android_aid.h>
124
125 static inline int current_has_network(void)
126 {
127 return in_egroup_p(AID_INET) || capable(CAP_NET_RAW);
128 }
129 #else
130 static inline int current_has_network(void)
131 {
132 return 1;
133 }
134 #endif
135
136 /* The inetsw table contains everything that inet_create needs to
137 * build a new socket.
138 */
139 static struct list_head inetsw[SOCK_MAX];
140 static DEFINE_SPINLOCK(inetsw_lock);
141
142 struct ipv4_config ipv4_config;
143 EXPORT_SYMBOL(ipv4_config);
144
145 /* New destruction routine */
146
147 void inet_sock_destruct(struct sock *sk)
148 {
149 struct inet_sock *inet = inet_sk(sk);
150
151 __skb_queue_purge(&sk->sk_receive_queue);
152 __skb_queue_purge(&sk->sk_error_queue);
153
154 sk_mem_reclaim(sk);
155
156 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
157 pr_err("Attempt to release TCP socket in state %d %p\n",
158 sk->sk_state, sk);
159 return;
160 }
161 if (!sock_flag(sk, SOCK_DEAD)) {
162 pr_err("Attempt to release alive inet socket %p\n", sk);
163 return;
164 }
165
166 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
167 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
168 WARN_ON(sk->sk_wmem_queued);
169 WARN_ON(sk->sk_forward_alloc);
170
171 kfree(rcu_dereference_protected(inet->inet_opt, 1));
172 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
173 dst_release(sk->sk_rx_dst);
174 sk_refcnt_debug_dec(sk);
175 }
176 EXPORT_SYMBOL(inet_sock_destruct);
177
178 /*
179 * The routines beyond this point handle the behaviour of an AF_INET
180 * socket object. Mostly it punts to the subprotocols of IP to do
181 * the work.
182 */
183
184 /*
185 * Automatically bind an unbound socket.
186 */
187
188 static int inet_autobind(struct sock *sk)
189 {
190 struct inet_sock *inet;
191 /* We may need to bind the socket. */
192 lock_sock(sk);
193 inet = inet_sk(sk);
194 if (!inet->inet_num) {
195 if (sk->sk_prot->get_port(sk, 0)) {
196 release_sock(sk);
197 return -EAGAIN;
198 }
199 inet->inet_sport = htons(inet->inet_num);
200 }
201 release_sock(sk);
202 return 0;
203 }
204
205 /*
206 * Move a socket into listening state.
207 */
208 int inet_listen(struct socket *sock, int backlog)
209 {
210 struct sock *sk = sock->sk;
211 unsigned char old_state;
212 int err;
213
214 lock_sock(sk);
215
216 err = -EINVAL;
217 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
218 goto out;
219
220 old_state = sk->sk_state;
221 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
222 goto out;
223
224 /* Really, if the socket is already in listen state
225 * we can only allow the backlog to be adjusted.
226 */
227 if (old_state != TCP_LISTEN) {
228 /* Check special setups for testing purpose to enable TFO w/o
229 * requiring TCP_FASTOPEN sockopt.
230 * Note that only TCP sockets (SOCK_STREAM) will reach here.
231 * Also fastopenq may already been allocated because this
232 * socket was in TCP_LISTEN state previously but was
233 * shutdown() (rather than close()).
234 */
235 if ((sysctl_tcp_fastopen & TFO_SERVER_ENABLE) != 0 &&
236 inet_csk(sk)->icsk_accept_queue.fastopenq == NULL) {
237 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) != 0)
238 err = fastopen_init_queue(sk, backlog);
239 else if ((sysctl_tcp_fastopen &
240 TFO_SERVER_WO_SOCKOPT2) != 0)
241 err = fastopen_init_queue(sk,
242 ((uint)sysctl_tcp_fastopen) >> 16);
243 else
244 err = 0;
245 if (err)
246 goto out;
247 }
248 err = inet_csk_listen_start(sk, backlog);
249 if (err)
250 goto out;
251 }
252 sk->sk_max_ack_backlog = backlog;
253 err = 0;
254
255 out:
256 release_sock(sk);
257 return err;
258 }
259 EXPORT_SYMBOL(inet_listen);
260
261 u32 inet_ehash_secret __read_mostly;
262 EXPORT_SYMBOL(inet_ehash_secret);
263
264 u32 ipv6_hash_secret __read_mostly;
265 EXPORT_SYMBOL(ipv6_hash_secret);
266
267 /*
268 * inet_ehash_secret must be set exactly once, and to a non nul value
269 * ipv6_hash_secret must be set exactly once.
270 */
271 void build_ehash_secret(void)
272 {
273 u32 rnd;
274
275 do {
276 get_random_bytes(&rnd, sizeof(rnd));
277 } while (rnd == 0);
278
279 if (cmpxchg(&inet_ehash_secret, 0, rnd) == 0)
280 get_random_bytes(&ipv6_hash_secret, sizeof(ipv6_hash_secret));
281 }
282 EXPORT_SYMBOL(build_ehash_secret);
283
284 /*
285 * Create an inet socket.
286 */
287
288 static int inet_create(struct net *net, struct socket *sock, int protocol,
289 int kern)
290 {
291 struct sock *sk;
292 struct inet_protosw *answer;
293 struct inet_sock *inet;
294 struct proto *answer_prot;
295 unsigned char answer_flags;
296 char answer_no_check;
297 int try_loading_module = 0;
298 int err;
299
300 if (!current_has_network())
301 return -EACCES;
302
303 if (unlikely(!inet_ehash_secret))
304 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
305 build_ehash_secret();
306
307 if (protocol < 0 || protocol >= IPPROTO_MAX)
308 return -EINVAL;
309
310 sock->state = SS_UNCONNECTED;
311
312 /* Look for the requested type/protocol pair. */
313 lookup_protocol:
314 err = -ESOCKTNOSUPPORT;
315 rcu_read_lock();
316 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
317
318 err = 0;
319 /* Check the non-wild match. */
320 if (protocol == answer->protocol) {
321 if (protocol != IPPROTO_IP)
322 break;
323 } else {
324 /* Check for the two wild cases. */
325 if (IPPROTO_IP == protocol) {
326 protocol = answer->protocol;
327 break;
328 }
329 if (IPPROTO_IP == answer->protocol)
330 break;
331 }
332 err = -EPROTONOSUPPORT;
333 }
334
335 if (unlikely(err)) {
336 if (try_loading_module < 2) {
337 rcu_read_unlock();
338 /*
339 * Be more specific, e.g. net-pf-2-proto-132-type-1
340 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
341 */
342 if (++try_loading_module == 1)
343 request_module("net-pf-%d-proto-%d-type-%d",
344 PF_INET, protocol, sock->type);
345 /*
346 * Fall back to generic, e.g. net-pf-2-proto-132
347 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
348 */
349 else
350 request_module("net-pf-%d-proto-%d",
351 PF_INET, protocol);
352 goto lookup_protocol;
353 } else
354 goto out_rcu_unlock;
355 }
356
357 err = -EPERM;
358 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
359 goto out_rcu_unlock;
360
361 sock->ops = answer->ops;
362 answer_prot = answer->prot;
363 answer_no_check = answer->no_check;
364 answer_flags = answer->flags;
365 rcu_read_unlock();
366
367 WARN_ON(answer_prot->slab == NULL);
368
369 err = -ENOBUFS;
370 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
371 if (sk == NULL)
372 goto out;
373
374 err = 0;
375 sk->sk_no_check = answer_no_check;
376 if (INET_PROTOSW_REUSE & answer_flags)
377 sk->sk_reuse = SK_CAN_REUSE;
378
379 inet = inet_sk(sk);
380 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
381
382 inet->nodefrag = 0;
383
384 if (SOCK_RAW == sock->type) {
385 inet->inet_num = protocol;
386 if (IPPROTO_RAW == protocol)
387 inet->hdrincl = 1;
388 }
389
390 if (ipv4_config.no_pmtu_disc)
391 inet->pmtudisc = IP_PMTUDISC_DONT;
392 else
393 inet->pmtudisc = IP_PMTUDISC_WANT;
394
395 inet->inet_id = 0;
396
397 sock_init_data(sock, sk);
398
399 sk->sk_destruct = inet_sock_destruct;
400 sk->sk_protocol = protocol;
401 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
402
403 inet->uc_ttl = -1;
404 inet->mc_loop = 1;
405 inet->mc_ttl = 1;
406 inet->mc_all = 1;
407 inet->mc_index = 0;
408 inet->mc_list = NULL;
409 inet->rcv_tos = 0;
410
411 sk_refcnt_debug_inc(sk);
412
413 if (inet->inet_num) {
414 /* It assumes that any protocol which allows
415 * the user to assign a number at socket
416 * creation time automatically
417 * shares.
418 */
419 inet->inet_sport = htons(inet->inet_num);
420 /* Add to protocol hash chains. */
421 sk->sk_prot->hash(sk);
422 }
423
424 if (sk->sk_prot->init) {
425 err = sk->sk_prot->init(sk);
426 if (err)
427 sk_common_release(sk);
428 }
429 out:
430 return err;
431 out_rcu_unlock:
432 rcu_read_unlock();
433 goto out;
434 }
435
436
437 /*
438 * The peer socket should always be NULL (or else). When we call this
439 * function we are destroying the object and from then on nobody
440 * should refer to it.
441 */
442 int inet_release(struct socket *sock)
443 {
444 struct sock *sk = sock->sk;
445
446 if (sk) {
447 long timeout;
448
449 sock_rps_reset_flow(sk);
450
451 /* Applications forget to leave groups before exiting */
452 ip_mc_drop_socket(sk);
453
454 /* If linger is set, we don't return until the close
455 * is complete. Otherwise we return immediately. The
456 * actually closing is done the same either way.
457 *
458 * If the close is due to the process exiting, we never
459 * linger..
460 */
461 timeout = 0;
462 if (sock_flag(sk, SOCK_LINGER) &&
463 !(current->flags & PF_EXITING))
464 timeout = sk->sk_lingertime;
465 sock->sk = NULL;
466 sk->sk_prot->close(sk, timeout);
467 }
468 return 0;
469 }
470 EXPORT_SYMBOL(inet_release);
471
472 /* It is off by default, see below. */
473 int sysctl_ip_nonlocal_bind __read_mostly;
474 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
475
476 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
477 {
478 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
479 struct sock *sk = sock->sk;
480 struct inet_sock *inet = inet_sk(sk);
481 struct net *net = sock_net(sk);
482 unsigned short snum;
483 int chk_addr_ret;
484 int err;
485
486 /* If the socket has its own bind function then use it. (RAW) */
487 if (sk->sk_prot->bind) {
488 err = sk->sk_prot->bind(sk, uaddr, addr_len);
489 goto out;
490 }
491 err = -EINVAL;
492 if (addr_len < sizeof(struct sockaddr_in))
493 goto out;
494
495 if (addr->sin_family != AF_INET) {
496 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
497 * only if s_addr is INADDR_ANY.
498 */
499 err = -EAFNOSUPPORT;
500 if (addr->sin_family != AF_UNSPEC ||
501 addr->sin_addr.s_addr != htonl(INADDR_ANY))
502 goto out;
503 }
504
505 chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
506
507 /* Not specified by any standard per-se, however it breaks too
508 * many applications when removed. It is unfortunate since
509 * allowing applications to make a non-local bind solves
510 * several problems with systems using dynamic addressing.
511 * (ie. your servers still start up even if your ISDN link
512 * is temporarily down)
513 */
514 err = -EADDRNOTAVAIL;
515 if (!sysctl_ip_nonlocal_bind &&
516 !(inet->freebind || inet->transparent) &&
517 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
518 chk_addr_ret != RTN_LOCAL &&
519 chk_addr_ret != RTN_MULTICAST &&
520 chk_addr_ret != RTN_BROADCAST)
521 goto out;
522
523 snum = ntohs(addr->sin_port);
524 err = -EACCES;
525 if (snum && snum < PROT_SOCK &&
526 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
527 goto out;
528
529 /* We keep a pair of addresses. rcv_saddr is the one
530 * used by hash lookups, and saddr is used for transmit.
531 *
532 * In the BSD API these are the same except where it
533 * would be illegal to use them (multicast/broadcast) in
534 * which case the sending device address is used.
535 */
536 lock_sock(sk);
537
538 /* Check these errors (active socket, double bind). */
539 err = -EINVAL;
540 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
541 goto out_release_sock;
542
543 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
544 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
545 inet->inet_saddr = 0; /* Use device */
546
547 /* Make sure we are allowed to bind here. */
548 if (sk->sk_prot->get_port(sk, snum)) {
549 inet->inet_saddr = inet->inet_rcv_saddr = 0;
550 err = -EADDRINUSE;
551 goto out_release_sock;
552 }
553
554 if (inet->inet_rcv_saddr)
555 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
556 if (snum)
557 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
558 inet->inet_sport = htons(inet->inet_num);
559 inet->inet_daddr = 0;
560 inet->inet_dport = 0;
561 sk_dst_reset(sk);
562 err = 0;
563 out_release_sock:
564 release_sock(sk);
565 out:
566 return err;
567 }
568 EXPORT_SYMBOL(inet_bind);
569
570 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
571 int addr_len, int flags)
572 {
573 struct sock *sk = sock->sk;
574
575 if (addr_len < sizeof(uaddr->sa_family))
576 return -EINVAL;
577 if (uaddr->sa_family == AF_UNSPEC)
578 return sk->sk_prot->disconnect(sk, flags);
579
580 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
581 return -EAGAIN;
582 return sk->sk_prot->connect(sk, uaddr, addr_len);
583 }
584 EXPORT_SYMBOL(inet_dgram_connect);
585
586 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
587 {
588 DEFINE_WAIT(wait);
589
590 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
591 sk->sk_write_pending += writebias;
592
593 /* Basic assumption: if someone sets sk->sk_err, he _must_
594 * change state of the socket from TCP_SYN_*.
595 * Connect() does not allow to get error notifications
596 * without closing the socket.
597 */
598 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
599 release_sock(sk);
600 timeo = schedule_timeout(timeo);
601 lock_sock(sk);
602 if (signal_pending(current) || !timeo)
603 break;
604 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
605 }
606 finish_wait(sk_sleep(sk), &wait);
607 sk->sk_write_pending -= writebias;
608 return timeo;
609 }
610
611 /*
612 * Connect to a remote host. There is regrettably still a little
613 * TCP 'magic' in here.
614 */
615 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
616 int addr_len, int flags)
617 {
618 struct sock *sk = sock->sk;
619 int err;
620 long timeo;
621
622 if (addr_len < sizeof(uaddr->sa_family))
623 return -EINVAL;
624
625 if (uaddr->sa_family == AF_UNSPEC) {
626 err = sk->sk_prot->disconnect(sk, flags);
627 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
628 goto out;
629 }
630
631 switch (sock->state) {
632 default:
633 err = -EINVAL;
634 goto out;
635 case SS_CONNECTED:
636 err = -EISCONN;
637 goto out;
638 case SS_CONNECTING:
639 err = -EALREADY;
640 /* Fall out of switch with err, set for this state */
641 break;
642 case SS_UNCONNECTED:
643 err = -EISCONN;
644 if (sk->sk_state != TCP_CLOSE)
645 goto out;
646
647 err = sk->sk_prot->connect(sk, uaddr, addr_len);
648 if (err < 0)
649 goto out;
650
651 sock->state = SS_CONNECTING;
652
653 /* Just entered SS_CONNECTING state; the only
654 * difference is that return value in non-blocking
655 * case is EINPROGRESS, rather than EALREADY.
656 */
657 err = -EINPROGRESS;
658 break;
659 }
660
661 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
662
663 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
664 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
665 tcp_sk(sk)->fastopen_req &&
666 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
667
668 /* Error code is set above */
669 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
670 goto out;
671
672 err = sock_intr_errno(timeo);
673 if (signal_pending(current))
674 goto out;
675 }
676
677 /* Connection was closed by RST, timeout, ICMP error
678 * or another process disconnected us.
679 */
680 if (sk->sk_state == TCP_CLOSE)
681 goto sock_error;
682
683 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
684 * and error was received after socket entered established state.
685 * Hence, it is handled normally after connect() return successfully.
686 */
687
688 sock->state = SS_CONNECTED;
689 err = 0;
690 out:
691 return err;
692
693 sock_error:
694 err = sock_error(sk) ? : -ECONNABORTED;
695 sock->state = SS_UNCONNECTED;
696 if (sk->sk_prot->disconnect(sk, flags))
697 sock->state = SS_DISCONNECTING;
698 goto out;
699 }
700 EXPORT_SYMBOL(__inet_stream_connect);
701
702 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
703 int addr_len, int flags)
704 {
705 int err;
706
707 lock_sock(sock->sk);
708 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
709 release_sock(sock->sk);
710 return err;
711 }
712 EXPORT_SYMBOL(inet_stream_connect);
713
714 /*
715 * Accept a pending connection. The TCP layer now gives BSD semantics.
716 */
717
718 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
719 {
720 struct sock *sk1 = sock->sk;
721 int err = -EINVAL;
722 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
723
724 if (!sk2)
725 goto do_err;
726
727 lock_sock(sk2);
728
729 sock_rps_record_flow(sk2);
730 WARN_ON(!((1 << sk2->sk_state) &
731 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
732 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
733
734 sock_graft(sk2, newsock);
735
736 newsock->state = SS_CONNECTED;
737 err = 0;
738 release_sock(sk2);
739 do_err:
740 return err;
741 }
742 EXPORT_SYMBOL(inet_accept);
743
744
745 /*
746 * This does both peername and sockname.
747 */
748 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
749 int *uaddr_len, int peer)
750 {
751 struct sock *sk = sock->sk;
752 struct inet_sock *inet = inet_sk(sk);
753 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
754
755 sin->sin_family = AF_INET;
756 if (peer) {
757 if (!inet->inet_dport ||
758 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
759 peer == 1))
760 return -ENOTCONN;
761 sin->sin_port = inet->inet_dport;
762 sin->sin_addr.s_addr = inet->inet_daddr;
763 } else {
764 __be32 addr = inet->inet_rcv_saddr;
765 if (!addr)
766 addr = inet->inet_saddr;
767 sin->sin_port = inet->inet_sport;
768 sin->sin_addr.s_addr = addr;
769 }
770 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
771 *uaddr_len = sizeof(*sin);
772 return 0;
773 }
774 EXPORT_SYMBOL(inet_getname);
775
776 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
777 size_t size)
778 {
779 struct sock *sk = sock->sk;
780
781 sock_rps_record_flow(sk);
782
783 /* We may need to bind the socket. */
784 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
785 inet_autobind(sk))
786 return -EAGAIN;
787
788 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
789 }
790 EXPORT_SYMBOL(inet_sendmsg);
791
792 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
793 size_t size, int flags)
794 {
795 struct sock *sk = sock->sk;
796
797 sock_rps_record_flow(sk);
798
799 /* We may need to bind the socket. */
800 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
801 inet_autobind(sk))
802 return -EAGAIN;
803
804 if (sk->sk_prot->sendpage)
805 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
806 return sock_no_sendpage(sock, page, offset, size, flags);
807 }
808 EXPORT_SYMBOL(inet_sendpage);
809
810 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
811 size_t size, int flags)
812 {
813 struct sock *sk = sock->sk;
814 int addr_len = 0;
815 int err;
816
817 sock_rps_record_flow(sk);
818
819 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
820 flags & ~MSG_DONTWAIT, &addr_len);
821 if (err >= 0)
822 msg->msg_namelen = addr_len;
823 return err;
824 }
825 EXPORT_SYMBOL(inet_recvmsg);
826
827 int inet_shutdown(struct socket *sock, int how)
828 {
829 struct sock *sk = sock->sk;
830 int err = 0;
831
832 /* This should really check to make sure
833 * the socket is a TCP socket. (WHY AC...)
834 */
835 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
836 1->2 bit 2 snds.
837 2->3 */
838 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
839 return -EINVAL;
840
841 lock_sock(sk);
842 if (sock->state == SS_CONNECTING) {
843 if ((1 << sk->sk_state) &
844 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
845 sock->state = SS_DISCONNECTING;
846 else
847 sock->state = SS_CONNECTED;
848 }
849
850 switch (sk->sk_state) {
851 case TCP_CLOSE:
852 err = -ENOTCONN;
853 /* Hack to wake up other listeners, who can poll for
854 POLLHUP, even on eg. unconnected UDP sockets -- RR */
855 default:
856 sk->sk_shutdown |= how;
857 if (sk->sk_prot->shutdown)
858 sk->sk_prot->shutdown(sk, how);
859 break;
860
861 /* Remaining two branches are temporary solution for missing
862 * close() in multithreaded environment. It is _not_ a good idea,
863 * but we have no choice until close() is repaired at VFS level.
864 */
865 case TCP_LISTEN:
866 if (!(how & RCV_SHUTDOWN))
867 break;
868 /* Fall through */
869 case TCP_SYN_SENT:
870 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
871 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
872 break;
873 }
874
875 /* Wake up anyone sleeping in poll. */
876 sk->sk_state_change(sk);
877 release_sock(sk);
878 return err;
879 }
880 EXPORT_SYMBOL(inet_shutdown);
881
882 /*
883 * ioctl() calls you can issue on an INET socket. Most of these are
884 * device configuration and stuff and very rarely used. Some ioctls
885 * pass on to the socket itself.
886 *
887 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
888 * loads the devconfigure module does its configuring and unloads it.
889 * There's a good 20K of config code hanging around the kernel.
890 */
891
892 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
893 {
894 struct sock *sk = sock->sk;
895 int err = 0;
896 struct net *net = sock_net(sk);
897
898 switch (cmd) {
899 case SIOCGSTAMP:
900 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
901 break;
902 case SIOCGSTAMPNS:
903 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
904 break;
905 case SIOCADDRT:
906 case SIOCDELRT:
907 case SIOCRTMSG:
908 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
909 break;
910 case SIOCDARP:
911 case SIOCGARP:
912 case SIOCSARP:
913 err = arp_ioctl(net, cmd, (void __user *)arg);
914 break;
915 case SIOCGIFADDR:
916 case SIOCSIFADDR:
917 case SIOCGIFBRDADDR:
918 case SIOCSIFBRDADDR:
919 case SIOCGIFNETMASK:
920 case SIOCSIFNETMASK:
921 case SIOCGIFDSTADDR:
922 case SIOCSIFDSTADDR:
923 case SIOCSIFPFLAGS:
924 case SIOCGIFPFLAGS:
925 case SIOCSIFFLAGS:
926 case SIOCKILLADDR:
927 err = devinet_ioctl(net, cmd, (void __user *)arg);
928 break;
929 default:
930 if (sk->sk_prot->ioctl)
931 err = sk->sk_prot->ioctl(sk, cmd, arg);
932 else
933 err = -ENOIOCTLCMD;
934 break;
935 }
936 return err;
937 }
938 EXPORT_SYMBOL(inet_ioctl);
939
940 #ifdef CONFIG_COMPAT
941 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
942 {
943 struct sock *sk = sock->sk;
944 int err = -ENOIOCTLCMD;
945
946 if (sk->sk_prot->compat_ioctl)
947 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
948
949 return err;
950 }
951 #endif
952
953 const struct proto_ops inet_stream_ops = {
954 .family = PF_INET,
955 .owner = THIS_MODULE,
956 .release = inet_release,
957 .bind = inet_bind,
958 .connect = inet_stream_connect,
959 .socketpair = sock_no_socketpair,
960 .accept = inet_accept,
961 .getname = inet_getname,
962 .poll = tcp_poll,
963 .ioctl = inet_ioctl,
964 .listen = inet_listen,
965 .shutdown = inet_shutdown,
966 .setsockopt = sock_common_setsockopt,
967 .getsockopt = sock_common_getsockopt,
968 .sendmsg = inet_sendmsg,
969 .recvmsg = inet_recvmsg,
970 .mmap = sock_no_mmap,
971 .sendpage = inet_sendpage,
972 .splice_read = tcp_splice_read,
973 #ifdef CONFIG_COMPAT
974 .compat_setsockopt = compat_sock_common_setsockopt,
975 .compat_getsockopt = compat_sock_common_getsockopt,
976 .compat_ioctl = inet_compat_ioctl,
977 #endif
978 };
979 EXPORT_SYMBOL(inet_stream_ops);
980
981 const struct proto_ops inet_dgram_ops = {
982 .family = PF_INET,
983 .owner = THIS_MODULE,
984 .release = inet_release,
985 .bind = inet_bind,
986 .connect = inet_dgram_connect,
987 .socketpair = sock_no_socketpair,
988 .accept = sock_no_accept,
989 .getname = inet_getname,
990 .poll = udp_poll,
991 .ioctl = inet_ioctl,
992 .listen = sock_no_listen,
993 .shutdown = inet_shutdown,
994 .setsockopt = sock_common_setsockopt,
995 .getsockopt = sock_common_getsockopt,
996 .sendmsg = inet_sendmsg,
997 .recvmsg = inet_recvmsg,
998 .mmap = sock_no_mmap,
999 .sendpage = inet_sendpage,
1000 #ifdef CONFIG_COMPAT
1001 .compat_setsockopt = compat_sock_common_setsockopt,
1002 .compat_getsockopt = compat_sock_common_getsockopt,
1003 .compat_ioctl = inet_compat_ioctl,
1004 #endif
1005 };
1006 EXPORT_SYMBOL(inet_dgram_ops);
1007
1008 /*
1009 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1010 * udp_poll
1011 */
1012 static const struct proto_ops inet_sockraw_ops = {
1013 .family = PF_INET,
1014 .owner = THIS_MODULE,
1015 .release = inet_release,
1016 .bind = inet_bind,
1017 .connect = inet_dgram_connect,
1018 .socketpair = sock_no_socketpair,
1019 .accept = sock_no_accept,
1020 .getname = inet_getname,
1021 .poll = datagram_poll,
1022 .ioctl = inet_ioctl,
1023 .listen = sock_no_listen,
1024 .shutdown = inet_shutdown,
1025 .setsockopt = sock_common_setsockopt,
1026 .getsockopt = sock_common_getsockopt,
1027 .sendmsg = inet_sendmsg,
1028 .recvmsg = inet_recvmsg,
1029 .mmap = sock_no_mmap,
1030 .sendpage = inet_sendpage,
1031 #ifdef CONFIG_COMPAT
1032 .compat_setsockopt = compat_sock_common_setsockopt,
1033 .compat_getsockopt = compat_sock_common_getsockopt,
1034 .compat_ioctl = inet_compat_ioctl,
1035 #endif
1036 };
1037
1038 static const struct net_proto_family inet_family_ops = {
1039 .family = PF_INET,
1040 .create = inet_create,
1041 .owner = THIS_MODULE,
1042 };
1043
1044 /* Upon startup we insert all the elements in inetsw_array[] into
1045 * the linked list inetsw.
1046 */
1047 static struct inet_protosw inetsw_array[] =
1048 {
1049 {
1050 .type = SOCK_STREAM,
1051 .protocol = IPPROTO_TCP,
1052 .prot = &tcp_prot,
1053 .ops = &inet_stream_ops,
1054 .no_check = 0,
1055 .flags = INET_PROTOSW_PERMANENT |
1056 INET_PROTOSW_ICSK,
1057 },
1058
1059 {
1060 .type = SOCK_DGRAM,
1061 .protocol = IPPROTO_UDP,
1062 .prot = &udp_prot,
1063 .ops = &inet_dgram_ops,
1064 .no_check = UDP_CSUM_DEFAULT,
1065 .flags = INET_PROTOSW_PERMANENT,
1066 },
1067
1068 {
1069 .type = SOCK_DGRAM,
1070 .protocol = IPPROTO_ICMP,
1071 .prot = &ping_prot,
1072 .ops = &inet_dgram_ops,
1073 .no_check = UDP_CSUM_DEFAULT,
1074 .flags = INET_PROTOSW_REUSE,
1075 },
1076
1077 {
1078 .type = SOCK_RAW,
1079 .protocol = IPPROTO_IP, /* wild card */
1080 .prot = &raw_prot,
1081 .ops = &inet_sockraw_ops,
1082 .no_check = UDP_CSUM_DEFAULT,
1083 .flags = INET_PROTOSW_REUSE,
1084 }
1085 };
1086
1087 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1088
1089 void inet_register_protosw(struct inet_protosw *p)
1090 {
1091 struct list_head *lh;
1092 struct inet_protosw *answer;
1093 int protocol = p->protocol;
1094 struct list_head *last_perm;
1095
1096 spin_lock_bh(&inetsw_lock);
1097
1098 if (p->type >= SOCK_MAX)
1099 goto out_illegal;
1100
1101 /* If we are trying to override a permanent protocol, bail. */
1102 answer = NULL;
1103 last_perm = &inetsw[p->type];
1104 list_for_each(lh, &inetsw[p->type]) {
1105 answer = list_entry(lh, struct inet_protosw, list);
1106
1107 /* Check only the non-wild match. */
1108 if (INET_PROTOSW_PERMANENT & answer->flags) {
1109 if (protocol == answer->protocol)
1110 break;
1111 last_perm = lh;
1112 }
1113
1114 answer = NULL;
1115 }
1116 if (answer)
1117 goto out_permanent;
1118
1119 /* Add the new entry after the last permanent entry if any, so that
1120 * the new entry does not override a permanent entry when matched with
1121 * a wild-card protocol. But it is allowed to override any existing
1122 * non-permanent entry. This means that when we remove this entry, the
1123 * system automatically returns to the old behavior.
1124 */
1125 list_add_rcu(&p->list, last_perm);
1126 out:
1127 spin_unlock_bh(&inetsw_lock);
1128
1129 return;
1130
1131 out_permanent:
1132 pr_err("Attempt to override permanent protocol %d\n", protocol);
1133 goto out;
1134
1135 out_illegal:
1136 pr_err("Ignoring attempt to register invalid socket type %d\n",
1137 p->type);
1138 goto out;
1139 }
1140 EXPORT_SYMBOL(inet_register_protosw);
1141
1142 void inet_unregister_protosw(struct inet_protosw *p)
1143 {
1144 if (INET_PROTOSW_PERMANENT & p->flags) {
1145 pr_err("Attempt to unregister permanent protocol %d\n",
1146 p->protocol);
1147 } else {
1148 spin_lock_bh(&inetsw_lock);
1149 list_del_rcu(&p->list);
1150 spin_unlock_bh(&inetsw_lock);
1151
1152 synchronize_net();
1153 }
1154 }
1155 EXPORT_SYMBOL(inet_unregister_protosw);
1156
1157 /*
1158 * Shall we try to damage output packets if routing dev changes?
1159 */
1160
1161 int sysctl_ip_dynaddr __read_mostly;
1162
1163 static int inet_sk_reselect_saddr(struct sock *sk)
1164 {
1165 struct inet_sock *inet = inet_sk(sk);
1166 __be32 old_saddr = inet->inet_saddr;
1167 __be32 daddr = inet->inet_daddr;
1168 struct flowi4 *fl4;
1169 struct rtable *rt;
1170 __be32 new_saddr;
1171 struct ip_options_rcu *inet_opt;
1172
1173 inet_opt = rcu_dereference_protected(inet->inet_opt,
1174 sock_owned_by_user(sk));
1175 if (inet_opt && inet_opt->opt.srr)
1176 daddr = inet_opt->opt.faddr;
1177
1178 /* Query new route. */
1179 fl4 = &inet->cork.fl.u.ip4;
1180 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1181 sk->sk_bound_dev_if, sk->sk_protocol,
1182 inet->inet_sport, inet->inet_dport, sk, false);
1183 if (IS_ERR(rt))
1184 return PTR_ERR(rt);
1185
1186 sk_setup_caps(sk, &rt->dst);
1187
1188 new_saddr = fl4->saddr;
1189
1190 if (new_saddr == old_saddr)
1191 return 0;
1192
1193 if (sysctl_ip_dynaddr > 1) {
1194 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1195 __func__, &old_saddr, &new_saddr);
1196 }
1197
1198 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1199
1200 /*
1201 * XXX The only one ugly spot where we need to
1202 * XXX really change the sockets identity after
1203 * XXX it has entered the hashes. -DaveM
1204 *
1205 * Besides that, it does not check for connection
1206 * uniqueness. Wait for troubles.
1207 */
1208 __sk_prot_rehash(sk);
1209 return 0;
1210 }
1211
1212 int inet_sk_rebuild_header(struct sock *sk)
1213 {
1214 struct inet_sock *inet = inet_sk(sk);
1215 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1216 __be32 daddr;
1217 struct ip_options_rcu *inet_opt;
1218 struct flowi4 *fl4;
1219 int err;
1220
1221 /* Route is OK, nothing to do. */
1222 if (rt)
1223 return 0;
1224
1225 /* Reroute. */
1226 rcu_read_lock();
1227 inet_opt = rcu_dereference(inet->inet_opt);
1228 daddr = inet->inet_daddr;
1229 if (inet_opt && inet_opt->opt.srr)
1230 daddr = inet_opt->opt.faddr;
1231 rcu_read_unlock();
1232 fl4 = &inet->cork.fl.u.ip4;
1233 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1234 inet->inet_dport, inet->inet_sport,
1235 sk->sk_protocol, RT_CONN_FLAGS(sk),
1236 sk->sk_bound_dev_if);
1237 if (!IS_ERR(rt)) {
1238 err = 0;
1239 sk_setup_caps(sk, &rt->dst);
1240 } else {
1241 err = PTR_ERR(rt);
1242
1243 /* Routing failed... */
1244 sk->sk_route_caps = 0;
1245 /*
1246 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1247 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1248 */
1249 if (!sysctl_ip_dynaddr ||
1250 sk->sk_state != TCP_SYN_SENT ||
1251 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1252 (err = inet_sk_reselect_saddr(sk)) != 0)
1253 sk->sk_err_soft = -err;
1254 }
1255
1256 return err;
1257 }
1258 EXPORT_SYMBOL(inet_sk_rebuild_header);
1259
1260 static int inet_gso_send_check(struct sk_buff *skb)
1261 {
1262 const struct net_offload *ops;
1263 const struct iphdr *iph;
1264 int proto;
1265 int ihl;
1266 int err = -EINVAL;
1267
1268 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1269 goto out;
1270
1271 iph = ip_hdr(skb);
1272 ihl = iph->ihl * 4;
1273 if (ihl < sizeof(*iph))
1274 goto out;
1275
1276 if (unlikely(!pskb_may_pull(skb, ihl)))
1277 goto out;
1278
1279 __skb_pull(skb, ihl);
1280 skb_reset_transport_header(skb);
1281 iph = ip_hdr(skb);
1282 proto = iph->protocol;
1283 err = -EPROTONOSUPPORT;
1284
1285 rcu_read_lock();
1286 ops = rcu_dereference(inet_offloads[proto]);
1287 if (likely(ops && ops->callbacks.gso_send_check))
1288 err = ops->callbacks.gso_send_check(skb);
1289 rcu_read_unlock();
1290
1291 out:
1292 return err;
1293 }
1294
1295 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1296 netdev_features_t features)
1297 {
1298 struct sk_buff *segs = ERR_PTR(-EINVAL);
1299 const struct net_offload *ops;
1300 struct iphdr *iph;
1301 int proto;
1302 int ihl;
1303 int id;
1304 unsigned int offset = 0;
1305 bool tunnel;
1306
1307 if (unlikely(skb_shinfo(skb)->gso_type &
1308 ~(SKB_GSO_TCPV4 |
1309 SKB_GSO_UDP |
1310 SKB_GSO_DODGY |
1311 SKB_GSO_TCP_ECN |
1312 SKB_GSO_GRE |
1313 SKB_GSO_TCPV6 |
1314 SKB_GSO_UDP_TUNNEL |
1315 0)))
1316 goto out;
1317
1318 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1319 goto out;
1320
1321 iph = ip_hdr(skb);
1322 ihl = iph->ihl * 4;
1323 if (ihl < sizeof(*iph))
1324 goto out;
1325
1326 if (unlikely(!pskb_may_pull(skb, ihl)))
1327 goto out;
1328
1329 tunnel = !!skb->encapsulation;
1330
1331 __skb_pull(skb, ihl);
1332 skb_reset_transport_header(skb);
1333 iph = ip_hdr(skb);
1334 id = ntohs(iph->id);
1335 proto = iph->protocol;
1336 segs = ERR_PTR(-EPROTONOSUPPORT);
1337
1338 rcu_read_lock();
1339 ops = rcu_dereference(inet_offloads[proto]);
1340 if (likely(ops && ops->callbacks.gso_segment))
1341 segs = ops->callbacks.gso_segment(skb, features);
1342 rcu_read_unlock();
1343
1344 if (IS_ERR_OR_NULL(segs))
1345 goto out;
1346
1347 skb = segs;
1348 do {
1349 iph = ip_hdr(skb);
1350 if (!tunnel && proto == IPPROTO_UDP) {
1351 iph->id = htons(id);
1352 iph->frag_off = htons(offset >> 3);
1353 if (skb->next != NULL)
1354 iph->frag_off |= htons(IP_MF);
1355 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1356 } else {
1357 iph->id = htons(id++);
1358 }
1359 iph->tot_len = htons(skb->len - skb->mac_len);
1360 iph->check = 0;
1361 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1362 } while ((skb = skb->next));
1363
1364 out:
1365 return segs;
1366 }
1367
1368 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1369 struct sk_buff *skb)
1370 {
1371 const struct net_offload *ops;
1372 struct sk_buff **pp = NULL;
1373 struct sk_buff *p;
1374 const struct iphdr *iph;
1375 unsigned int hlen;
1376 unsigned int off;
1377 unsigned int id;
1378 int flush = 1;
1379 int proto;
1380
1381 off = skb_gro_offset(skb);
1382 hlen = off + sizeof(*iph);
1383 iph = skb_gro_header_fast(skb, off);
1384 if (skb_gro_header_hard(skb, hlen)) {
1385 iph = skb_gro_header_slow(skb, hlen, off);
1386 if (unlikely(!iph))
1387 goto out;
1388 }
1389
1390 proto = iph->protocol;
1391
1392 rcu_read_lock();
1393 ops = rcu_dereference(inet_offloads[proto]);
1394 if (!ops || !ops->callbacks.gro_receive)
1395 goto out_unlock;
1396
1397 if (*(u8 *)iph != 0x45)
1398 goto out_unlock;
1399
1400 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1401 goto out_unlock;
1402
1403 id = ntohl(*(__be32 *)&iph->id);
1404 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1405 id >>= 16;
1406
1407 for (p = *head; p; p = p->next) {
1408 struct iphdr *iph2;
1409
1410 if (!NAPI_GRO_CB(p)->same_flow)
1411 continue;
1412
1413 iph2 = ip_hdr(p);
1414
1415 if ((iph->protocol ^ iph2->protocol) |
1416 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1417 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1418 NAPI_GRO_CB(p)->same_flow = 0;
1419 continue;
1420 }
1421
1422 /* All fields must match except length and checksum. */
1423 NAPI_GRO_CB(p)->flush |=
1424 (iph->ttl ^ iph2->ttl) |
1425 (iph->tos ^ iph2->tos) |
1426 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1427
1428 NAPI_GRO_CB(p)->flush |= flush;
1429 }
1430
1431 NAPI_GRO_CB(skb)->flush |= flush;
1432 skb_gro_pull(skb, sizeof(*iph));
1433 skb_set_transport_header(skb, skb_gro_offset(skb));
1434
1435 pp = ops->callbacks.gro_receive(head, skb);
1436
1437 out_unlock:
1438 rcu_read_unlock();
1439
1440 out:
1441 NAPI_GRO_CB(skb)->flush |= flush;
1442
1443 return pp;
1444 }
1445
1446 static int inet_gro_complete(struct sk_buff *skb)
1447 {
1448 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1449 struct iphdr *iph = ip_hdr(skb);
1450 const struct net_offload *ops;
1451 int proto = iph->protocol;
1452 int err = -ENOSYS;
1453
1454 csum_replace2(&iph->check, iph->tot_len, newlen);
1455 iph->tot_len = newlen;
1456
1457 rcu_read_lock();
1458 ops = rcu_dereference(inet_offloads[proto]);
1459 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1460 goto out_unlock;
1461
1462 err = ops->callbacks.gro_complete(skb);
1463
1464 out_unlock:
1465 rcu_read_unlock();
1466
1467 return err;
1468 }
1469
1470 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1471 unsigned short type, unsigned char protocol,
1472 struct net *net)
1473 {
1474 struct socket *sock;
1475 int rc = sock_create_kern(family, type, protocol, &sock);
1476
1477 if (rc == 0) {
1478 *sk = sock->sk;
1479 (*sk)->sk_allocation = GFP_ATOMIC;
1480 /*
1481 * Unhash it so that IP input processing does not even see it,
1482 * we do not wish this socket to see incoming packets.
1483 */
1484 (*sk)->sk_prot->unhash(*sk);
1485
1486 sk_change_net(*sk, net);
1487 }
1488 return rc;
1489 }
1490 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1491
1492 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1493 {
1494 unsigned long res = 0;
1495 int i, j;
1496
1497 for_each_possible_cpu(i) {
1498 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1499 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1500 }
1501 return res;
1502 }
1503 EXPORT_SYMBOL_GPL(snmp_fold_field);
1504
1505 #if BITS_PER_LONG==32
1506
1507 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1508 {
1509 u64 res = 0;
1510 int cpu;
1511
1512 for_each_possible_cpu(cpu) {
1513 void *bhptr;
1514 struct u64_stats_sync *syncp;
1515 u64 v;
1516 unsigned int start;
1517
1518 bhptr = per_cpu_ptr(mib[0], cpu);
1519 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1520 do {
1521 start = u64_stats_fetch_begin_bh(syncp);
1522 v = *(((u64 *) bhptr) + offt);
1523 } while (u64_stats_fetch_retry_bh(syncp, start));
1524
1525 res += v;
1526 }
1527 return res;
1528 }
1529 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1530 #endif
1531
1532 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1533 {
1534 BUG_ON(ptr == NULL);
1535 ptr[0] = __alloc_percpu(mibsize, align);
1536 if (!ptr[0])
1537 return -ENOMEM;
1538 #if SNMP_ARRAY_SZ == 2
1539 ptr[1] = __alloc_percpu(mibsize, align);
1540 if (!ptr[1]) {
1541 free_percpu(ptr[0]);
1542 ptr[0] = NULL;
1543 return -ENOMEM;
1544 }
1545 #endif
1546 return 0;
1547 }
1548 EXPORT_SYMBOL_GPL(snmp_mib_init);
1549
1550 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1551 {
1552 int i;
1553
1554 BUG_ON(ptr == NULL);
1555 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1556 free_percpu(ptr[i]);
1557 ptr[i] = NULL;
1558 }
1559 }
1560 EXPORT_SYMBOL_GPL(snmp_mib_free);
1561
1562 #ifdef CONFIG_IP_MULTICAST
1563 static const struct net_protocol igmp_protocol = {
1564 .handler = igmp_rcv,
1565 .netns_ok = 1,
1566 };
1567 #endif
1568
1569 static const struct net_protocol tcp_protocol = {
1570 .early_demux = tcp_v4_early_demux,
1571 .handler = tcp_v4_rcv,
1572 .err_handler = tcp_v4_err,
1573 .no_policy = 1,
1574 .netns_ok = 1,
1575 };
1576
1577 static const struct net_offload tcp_offload = {
1578 .callbacks = {
1579 .gso_send_check = tcp_v4_gso_send_check,
1580 .gso_segment = tcp_tso_segment,
1581 .gro_receive = tcp4_gro_receive,
1582 .gro_complete = tcp4_gro_complete,
1583 },
1584 };
1585
1586 static const struct net_protocol udp_protocol = {
1587 .handler = udp_rcv,
1588 .err_handler = udp_err,
1589 .no_policy = 1,
1590 .netns_ok = 1,
1591 };
1592
1593 static const struct net_offload udp_offload = {
1594 .callbacks = {
1595 .gso_send_check = udp4_ufo_send_check,
1596 .gso_segment = udp4_ufo_fragment,
1597 },
1598 };
1599
1600 static const struct net_protocol icmp_protocol = {
1601 .handler = icmp_rcv,
1602 .err_handler = icmp_err,
1603 .no_policy = 1,
1604 .netns_ok = 1,
1605 };
1606
1607 static __net_init int ipv4_mib_init_net(struct net *net)
1608 {
1609 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1610 sizeof(struct tcp_mib),
1611 __alignof__(struct tcp_mib)) < 0)
1612 goto err_tcp_mib;
1613 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1614 sizeof(struct ipstats_mib),
1615 __alignof__(struct ipstats_mib)) < 0)
1616 goto err_ip_mib;
1617 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1618 sizeof(struct linux_mib),
1619 __alignof__(struct linux_mib)) < 0)
1620 goto err_net_mib;
1621 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1622 sizeof(struct udp_mib),
1623 __alignof__(struct udp_mib)) < 0)
1624 goto err_udp_mib;
1625 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1626 sizeof(struct udp_mib),
1627 __alignof__(struct udp_mib)) < 0)
1628 goto err_udplite_mib;
1629 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1630 sizeof(struct icmp_mib),
1631 __alignof__(struct icmp_mib)) < 0)
1632 goto err_icmp_mib;
1633 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1634 GFP_KERNEL);
1635 if (!net->mib.icmpmsg_statistics)
1636 goto err_icmpmsg_mib;
1637
1638 tcp_mib_init(net);
1639 return 0;
1640
1641 err_icmpmsg_mib:
1642 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1643 err_icmp_mib:
1644 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1645 err_udplite_mib:
1646 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1647 err_udp_mib:
1648 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1649 err_net_mib:
1650 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1651 err_ip_mib:
1652 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1653 err_tcp_mib:
1654 return -ENOMEM;
1655 }
1656
1657 static __net_exit void ipv4_mib_exit_net(struct net *net)
1658 {
1659 kfree(net->mib.icmpmsg_statistics);
1660 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1661 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1662 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1663 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1664 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1665 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1666 }
1667
1668 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1669 .init = ipv4_mib_init_net,
1670 .exit = ipv4_mib_exit_net,
1671 };
1672
1673 static int __init init_ipv4_mibs(void)
1674 {
1675 return register_pernet_subsys(&ipv4_mib_ops);
1676 }
1677
1678 static int ipv4_proc_init(void);
1679
1680 /*
1681 * IP protocol layer initialiser
1682 */
1683
1684 static struct packet_offload ip_packet_offload __read_mostly = {
1685 .type = cpu_to_be16(ETH_P_IP),
1686 .callbacks = {
1687 .gso_send_check = inet_gso_send_check,
1688 .gso_segment = inet_gso_segment,
1689 .gro_receive = inet_gro_receive,
1690 .gro_complete = inet_gro_complete,
1691 },
1692 };
1693
1694 static int __init ipv4_offload_init(void)
1695 {
1696 /*
1697 * Add offloads
1698 */
1699 if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1700 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1701 if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1702 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1703
1704 dev_add_offload(&ip_packet_offload);
1705 return 0;
1706 }
1707
1708 fs_initcall(ipv4_offload_init);
1709
1710 static struct packet_type ip_packet_type __read_mostly = {
1711 .type = cpu_to_be16(ETH_P_IP),
1712 .func = ip_rcv,
1713 };
1714
1715 static int __init inet_init(void)
1716 {
1717 struct inet_protosw *q;
1718 struct list_head *r;
1719 int rc = -EINVAL;
1720
1721 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > FIELD_SIZEOF(struct sk_buff, cb));
1722
1723 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1724 if (!sysctl_local_reserved_ports)
1725 goto out;
1726
1727 rc = proto_register(&tcp_prot, 1);
1728 if (rc)
1729 goto out_free_reserved_ports;
1730
1731 rc = proto_register(&udp_prot, 1);
1732 if (rc)
1733 goto out_unregister_tcp_proto;
1734
1735 rc = proto_register(&raw_prot, 1);
1736 if (rc)
1737 goto out_unregister_udp_proto;
1738
1739 rc = proto_register(&ping_prot, 1);
1740 if (rc)
1741 goto out_unregister_raw_proto;
1742
1743 /*
1744 * Tell SOCKET that we are alive...
1745 */
1746
1747 (void)sock_register(&inet_family_ops);
1748
1749 #ifdef CONFIG_SYSCTL
1750 ip_static_sysctl_init();
1751 #endif
1752
1753 tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1754
1755 /*
1756 * Add all the base protocols.
1757 */
1758
1759 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1760 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1761 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1762 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1763 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1764 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1765 #ifdef CONFIG_IP_MULTICAST
1766 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1767 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1768 #endif
1769
1770 /* Register the socket-side information for inet_create. */
1771 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1772 INIT_LIST_HEAD(r);
1773
1774 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1775 inet_register_protosw(q);
1776
1777 /*
1778 * Set the ARP module up
1779 */
1780
1781 arp_init();
1782
1783 /*
1784 * Set the IP module up
1785 */
1786
1787 ip_init();
1788
1789 tcp_v4_init();
1790
1791 /* Setup TCP slab cache for open requests. */
1792 tcp_init();
1793
1794 /* Setup UDP memory threshold */
1795 udp_init();
1796
1797 /* Add UDP-Lite (RFC 3828) */
1798 udplite4_register();
1799
1800 ping_init();
1801
1802 /*
1803 * Set the ICMP layer up
1804 */
1805
1806 if (icmp_init() < 0)
1807 panic("Failed to create the ICMP control socket.\n");
1808
1809 /*
1810 * Initialise the multicast router
1811 */
1812 #if defined(CONFIG_IP_MROUTE)
1813 if (ip_mr_init())
1814 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1815 #endif
1816 /*
1817 * Initialise per-cpu ipv4 mibs
1818 */
1819
1820 if (init_ipv4_mibs())
1821 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1822
1823 ipv4_proc_init();
1824
1825 ipfrag_init();
1826
1827 dev_add_pack(&ip_packet_type);
1828
1829 rc = 0;
1830 out:
1831 return rc;
1832 out_unregister_raw_proto:
1833 proto_unregister(&raw_prot);
1834 out_unregister_udp_proto:
1835 proto_unregister(&udp_prot);
1836 out_unregister_tcp_proto:
1837 proto_unregister(&tcp_prot);
1838 out_free_reserved_ports:
1839 kfree(sysctl_local_reserved_ports);
1840 goto out;
1841 }
1842
1843 fs_initcall(inet_init);
1844
1845 /* ------------------------------------------------------------------------ */
1846
1847 #ifdef CONFIG_PROC_FS
1848 static int __init ipv4_proc_init(void)
1849 {
1850 int rc = 0;
1851
1852 if (raw_proc_init())
1853 goto out_raw;
1854 if (tcp4_proc_init())
1855 goto out_tcp;
1856 if (udp4_proc_init())
1857 goto out_udp;
1858 if (ping_proc_init())
1859 goto out_ping;
1860 if (ip_misc_proc_init())
1861 goto out_misc;
1862 out:
1863 return rc;
1864 out_misc:
1865 ping_proc_exit();
1866 out_ping:
1867 udp4_proc_exit();
1868 out_udp:
1869 tcp4_proc_exit();
1870 out_tcp:
1871 raw_proc_exit();
1872 out_raw:
1873 rc = -ENOMEM;
1874 goto out;
1875 }
1876
1877 #else /* CONFIG_PROC_FS */
1878 static int __init ipv4_proc_init(void)
1879 {
1880 return 0;
1881 }
1882 #endif /* CONFIG_PROC_FS */
1883
1884 MODULE_ALIAS_NETPROTO(PF_INET);
1885
kernel source for telekom puls (alcatel/mediatek)