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 */
68 #define pr_fmt(fmt) "IPv4: " fmt
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>
93 #include <asm/uaccess.h>
95 #include <linux/inet.h>
96 #include <linux/igmp.h>
97 #include <linux/inetdevice.h>
98 #include <linux/netdevice.h>
99 #include <net/checksum.h>
100 #include <net/ip.h>
101 #include <net/protocol.h>
102 #include <net/arp.h>
103 #include <net/route.h>
104 #include <net/ip_fib.h>
105 #include <net/inet_connection_sock.h>
106 #include <net/tcp.h>
107 #include <net/udp.h>
108 #include <net/udplite.h>
109 #include <net/ping.h>
110 #include <linux/skbuff.h>
111 #include <net/sock.h>
112 #include <net/raw.h>
113 #include <net/icmp.h>
114 #include <net/ipip.h>
115 #include <net/inet_common.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #ifdef CONFIG_IP_MROUTE
119 #include <linux/mroute.h>
120 #endif
122 #ifdef CONFIG_ANDROID_PARANOID_NETWORK
123 #include <linux/android_aid.h>
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
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);
142 struct ipv4_config ipv4_config;
143 EXPORT_SYMBOL(ipv4_config);
145 /* New destruction routine */
147 void inet_sock_destruct(struct sock *sk)
148 {
149 struct inet_sock *inet = inet_sk(sk);
151 __skb_queue_purge(&sk->sk_receive_queue);
152 __skb_queue_purge(&sk->sk_error_queue);
154 sk_mem_reclaim(sk);
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 }
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);
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);
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 */
184 /*
185 * Automatically bind an unbound socket.
186 */
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 }
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;
214 lock_sock(sk);
216 err = -EINVAL;
217 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
218 goto out;
220 old_state = sk->sk_state;
221 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
222 goto out;
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;
255 out:
256 release_sock(sk);
257 return err;
258 }
259 EXPORT_SYMBOL(inet_listen);
261 u32 inet_ehash_secret __read_mostly;
262 EXPORT_SYMBOL(inet_ehash_secret);
264 u32 ipv6_hash_secret __read_mostly;
265 EXPORT_SYMBOL(ipv6_hash_secret);
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;
275 do {
276 get_random_bytes(&rnd, sizeof(rnd));
277 } while (rnd == 0);
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);
284 static inline int inet_netns_ok(struct net *net, __u8 protocol)
285 {
286 const struct net_protocol *ipprot;
288 if (net_eq(net, &init_net))
289 return 1;
291 ipprot = rcu_dereference(inet_protos[protocol]);
292 if (ipprot == NULL) {
293 /* raw IP is OK */
294 return 1;
295 }
296 return ipprot->netns_ok;
297 }
300 /*
301 * Create an inet socket.
302 */
304 static int inet_create(struct net *net, struct socket *sock, int protocol,
305 int kern)
306 {
307 struct sock *sk;
308 struct inet_protosw *answer;
309 struct inet_sock *inet;
310 struct proto *answer_prot;
311 unsigned char answer_flags;
312 char answer_no_check;
313 int try_loading_module = 0;
314 int err;
316 if (!current_has_network())
317 return -EACCES;
319 if (unlikely(!inet_ehash_secret))
320 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
321 build_ehash_secret();
323 sock->state = SS_UNCONNECTED;
325 /* Look for the requested type/protocol pair. */
326 lookup_protocol:
327 err = -ESOCKTNOSUPPORT;
328 rcu_read_lock();
329 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
331 err = 0;
332 /* Check the non-wild match. */
333 if (protocol == answer->protocol) {
334 if (protocol != IPPROTO_IP)
335 break;
336 } else {
337 /* Check for the two wild cases. */
338 if (IPPROTO_IP == protocol) {
339 protocol = answer->protocol;
340 break;
341 }
342 if (IPPROTO_IP == answer->protocol)
343 break;
344 }
345 err = -EPROTONOSUPPORT;
346 }
348 if (unlikely(err)) {
349 if (try_loading_module < 2) {
350 rcu_read_unlock();
351 /*
352 * Be more specific, e.g. net-pf-2-proto-132-type-1
353 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
354 */
355 if (++try_loading_module == 1)
356 request_module("net-pf-%d-proto-%d-type-%d",
357 PF_INET, protocol, sock->type);
358 /*
359 * Fall back to generic, e.g. net-pf-2-proto-132
360 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
361 */
362 else
363 request_module("net-pf-%d-proto-%d",
364 PF_INET, protocol);
365 goto lookup_protocol;
366 } else
367 goto out_rcu_unlock;
368 }
370 err = -EPERM;
371 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
372 goto out_rcu_unlock;
374 err = -EAFNOSUPPORT;
375 if (!inet_netns_ok(net, protocol))
376 goto out_rcu_unlock;
378 sock->ops = answer->ops;
379 answer_prot = answer->prot;
380 answer_no_check = answer->no_check;
381 answer_flags = answer->flags;
382 rcu_read_unlock();
384 WARN_ON(answer_prot->slab == NULL);
386 err = -ENOBUFS;
387 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
388 if (sk == NULL)
389 goto out;
391 err = 0;
392 sk->sk_no_check = answer_no_check;
393 if (INET_PROTOSW_REUSE & answer_flags)
394 sk->sk_reuse = SK_CAN_REUSE;
396 inet = inet_sk(sk);
397 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
399 inet->nodefrag = 0;
401 if (SOCK_RAW == sock->type) {
402 inet->inet_num = protocol;
403 if (IPPROTO_RAW == protocol)
404 inet->hdrincl = 1;
405 }
407 if (ipv4_config.no_pmtu_disc)
408 inet->pmtudisc = IP_PMTUDISC_DONT;
409 else
410 inet->pmtudisc = IP_PMTUDISC_WANT;
412 inet->inet_id = 0;
414 sock_init_data(sock, sk);
416 sk->sk_destruct = inet_sock_destruct;
417 sk->sk_protocol = protocol;
418 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
420 inet->uc_ttl = -1;
421 inet->mc_loop = 1;
422 inet->mc_ttl = 1;
423 inet->mc_all = 1;
424 inet->mc_index = 0;
425 inet->mc_list = NULL;
426 inet->rcv_tos = 0;
428 sk_refcnt_debug_inc(sk);
430 if (inet->inet_num) {
431 /* It assumes that any protocol which allows
432 * the user to assign a number at socket
433 * creation time automatically
434 * shares.
435 */
436 inet->inet_sport = htons(inet->inet_num);
437 /* Add to protocol hash chains. */
438 sk->sk_prot->hash(sk);
439 }
441 if (sk->sk_prot->init) {
442 err = sk->sk_prot->init(sk);
443 if (err)
444 sk_common_release(sk);
445 }
446 out:
447 return err;
448 out_rcu_unlock:
449 rcu_read_unlock();
450 goto out;
451 }
454 /*
455 * The peer socket should always be NULL (or else). When we call this
456 * function we are destroying the object and from then on nobody
457 * should refer to it.
458 */
459 int inet_release(struct socket *sock)
460 {
461 struct sock *sk = sock->sk;
463 if (sk) {
464 long timeout;
466 sock_rps_reset_flow(sk);
468 /* Applications forget to leave groups before exiting */
469 ip_mc_drop_socket(sk);
471 /* If linger is set, we don't return until the close
472 * is complete. Otherwise we return immediately. The
473 * actually closing is done the same either way.
474 *
475 * If the close is due to the process exiting, we never
476 * linger..
477 */
478 timeout = 0;
479 if (sock_flag(sk, SOCK_LINGER) &&
480 !(current->flags & PF_EXITING))
481 timeout = sk->sk_lingertime;
482 sock->sk = NULL;
483 sk->sk_prot->close(sk, timeout);
484 }
485 return 0;
486 }
487 EXPORT_SYMBOL(inet_release);
489 /* It is off by default, see below. */
490 int sysctl_ip_nonlocal_bind __read_mostly;
491 EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
493 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
494 {
495 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
496 struct sock *sk = sock->sk;
497 struct inet_sock *inet = inet_sk(sk);
498 struct net *net = sock_net(sk);
499 unsigned short snum;
500 int chk_addr_ret;
501 int err;
503 /* If the socket has its own bind function then use it. (RAW) */
504 if (sk->sk_prot->bind) {
505 err = sk->sk_prot->bind(sk, uaddr, addr_len);
506 goto out;
507 }
508 err = -EINVAL;
509 if (addr_len < sizeof(struct sockaddr_in))
510 goto out;
512 if (addr->sin_family != AF_INET) {
513 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
514 * only if s_addr is INADDR_ANY.
515 */
516 err = -EAFNOSUPPORT;
517 if (addr->sin_family != AF_UNSPEC ||
518 addr->sin_addr.s_addr != htonl(INADDR_ANY))
519 goto out;
520 }
522 chk_addr_ret = inet_addr_type(net, addr->sin_addr.s_addr);
524 /* Not specified by any standard per-se, however it breaks too
525 * many applications when removed. It is unfortunate since
526 * allowing applications to make a non-local bind solves
527 * several problems with systems using dynamic addressing.
528 * (ie. your servers still start up even if your ISDN link
529 * is temporarily down)
530 */
531 err = -EADDRNOTAVAIL;
532 if (!sysctl_ip_nonlocal_bind &&
533 !(inet->freebind || inet->transparent) &&
534 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
535 chk_addr_ret != RTN_LOCAL &&
536 chk_addr_ret != RTN_MULTICAST &&
537 chk_addr_ret != RTN_BROADCAST)
538 goto out;
540 snum = ntohs(addr->sin_port);
541 err = -EACCES;
542 if (snum && snum < PROT_SOCK &&
543 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
544 goto out;
546 /* We keep a pair of addresses. rcv_saddr is the one
547 * used by hash lookups, and saddr is used for transmit.
548 *
549 * In the BSD API these are the same except where it
550 * would be illegal to use them (multicast/broadcast) in
551 * which case the sending device address is used.
552 */
553 lock_sock(sk);
555 /* Check these errors (active socket, double bind). */
556 err = -EINVAL;
557 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
558 goto out_release_sock;
560 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
561 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
562 inet->inet_saddr = 0; /* Use device */
564 /* Make sure we are allowed to bind here. */
565 if (sk->sk_prot->get_port(sk, snum)) {
566 inet->inet_saddr = inet->inet_rcv_saddr = 0;
567 err = -EADDRINUSE;
568 goto out_release_sock;
569 }
571 if (inet->inet_rcv_saddr)
572 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
573 if (snum)
574 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
575 inet->inet_sport = htons(inet->inet_num);
576 inet->inet_daddr = 0;
577 inet->inet_dport = 0;
578 sk_dst_reset(sk);
579 err = 0;
580 out_release_sock:
581 release_sock(sk);
582 out:
583 return err;
584 }
585 EXPORT_SYMBOL(inet_bind);
587 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
588 int addr_len, int flags)
589 {
590 struct sock *sk = sock->sk;
592 if (addr_len < sizeof(uaddr->sa_family))
593 return -EINVAL;
594 if (uaddr->sa_family == AF_UNSPEC)
595 return sk->sk_prot->disconnect(sk, flags);
597 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
598 return -EAGAIN;
599 return sk->sk_prot->connect(sk, uaddr, addr_len);
600 }
601 EXPORT_SYMBOL(inet_dgram_connect);
603 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
604 {
605 DEFINE_WAIT(wait);
607 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
608 sk->sk_write_pending += writebias;
610 /* Basic assumption: if someone sets sk->sk_err, he _must_
611 * change state of the socket from TCP_SYN_*.
612 * Connect() does not allow to get error notifications
613 * without closing the socket.
614 */
615 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
616 release_sock(sk);
617 timeo = schedule_timeout(timeo);
618 lock_sock(sk);
619 if (signal_pending(current) || !timeo)
620 break;
621 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
622 }
623 finish_wait(sk_sleep(sk), &wait);
624 sk->sk_write_pending -= writebias;
625 return timeo;
626 }
628 /*
629 * Connect to a remote host. There is regrettably still a little
630 * TCP 'magic' in here.
631 */
632 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
633 int addr_len, int flags)
634 {
635 struct sock *sk = sock->sk;
636 int err;
637 long timeo;
639 if (addr_len < sizeof(uaddr->sa_family))
640 return -EINVAL;
642 if (uaddr->sa_family == AF_UNSPEC) {
643 err = sk->sk_prot->disconnect(sk, flags);
644 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
645 goto out;
646 }
648 switch (sock->state) {
649 default:
650 err = -EINVAL;
651 goto out;
652 case SS_CONNECTED:
653 err = -EISCONN;
654 goto out;
655 case SS_CONNECTING:
656 err = -EALREADY;
657 /* Fall out of switch with err, set for this state */
658 break;
659 case SS_UNCONNECTED:
660 err = -EISCONN;
661 if (sk->sk_state != TCP_CLOSE)
662 goto out;
664 err = sk->sk_prot->connect(sk, uaddr, addr_len);
665 if (err < 0)
666 goto out;
668 sock->state = SS_CONNECTING;
670 /* Just entered SS_CONNECTING state; the only
671 * difference is that return value in non-blocking
672 * case is EINPROGRESS, rather than EALREADY.
673 */
674 err = -EINPROGRESS;
675 break;
676 }
678 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
680 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
681 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
682 tcp_sk(sk)->fastopen_req &&
683 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
685 /* Error code is set above */
686 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
687 goto out;
689 err = sock_intr_errno(timeo);
690 if (signal_pending(current))
691 goto out;
692 }
694 /* Connection was closed by RST, timeout, ICMP error
695 * or another process disconnected us.
696 */
697 if (sk->sk_state == TCP_CLOSE)
698 goto sock_error;
700 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
701 * and error was received after socket entered established state.
702 * Hence, it is handled normally after connect() return successfully.
703 */
705 sock->state = SS_CONNECTED;
706 err = 0;
707 out:
708 return err;
710 sock_error:
711 err = sock_error(sk) ? : -ECONNABORTED;
712 sock->state = SS_UNCONNECTED;
713 if (sk->sk_prot->disconnect(sk, flags))
714 sock->state = SS_DISCONNECTING;
715 goto out;
716 }
717 EXPORT_SYMBOL(__inet_stream_connect);
719 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
720 int addr_len, int flags)
721 {
722 int err;
724 lock_sock(sock->sk);
725 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
726 release_sock(sock->sk);
727 return err;
728 }
729 EXPORT_SYMBOL(inet_stream_connect);
731 /*
732 * Accept a pending connection. The TCP layer now gives BSD semantics.
733 */
735 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
736 {
737 struct sock *sk1 = sock->sk;
738 int err = -EINVAL;
739 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
741 if (!sk2)
742 goto do_err;
744 lock_sock(sk2);
746 sock_rps_record_flow(sk2);
747 WARN_ON(!((1 << sk2->sk_state) &
748 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
749 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
751 sock_graft(sk2, newsock);
753 newsock->state = SS_CONNECTED;
754 err = 0;
755 release_sock(sk2);
756 do_err:
757 return err;
758 }
759 EXPORT_SYMBOL(inet_accept);
762 /*
763 * This does both peername and sockname.
764 */
765 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
766 int *uaddr_len, int peer)
767 {
768 struct sock *sk = sock->sk;
769 struct inet_sock *inet = inet_sk(sk);
770 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
772 sin->sin_family = AF_INET;
773 if (peer) {
774 if (!inet->inet_dport ||
775 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
776 peer == 1))
777 return -ENOTCONN;
778 sin->sin_port = inet->inet_dport;
779 sin->sin_addr.s_addr = inet->inet_daddr;
780 } else {
781 __be32 addr = inet->inet_rcv_saddr;
782 if (!addr)
783 addr = inet->inet_saddr;
784 sin->sin_port = inet->inet_sport;
785 sin->sin_addr.s_addr = addr;
786 }
787 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
788 *uaddr_len = sizeof(*sin);
789 return 0;
790 }
791 EXPORT_SYMBOL(inet_getname);
793 int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
794 size_t size)
795 {
796 struct sock *sk = sock->sk;
798 sock_rps_record_flow(sk);
800 /* We may need to bind the socket. */
801 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
802 inet_autobind(sk))
803 return -EAGAIN;
805 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
806 }
807 EXPORT_SYMBOL(inet_sendmsg);
809 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
810 size_t size, int flags)
811 {
812 struct sock *sk = sock->sk;
814 sock_rps_record_flow(sk);
816 /* We may need to bind the socket. */
817 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
818 inet_autobind(sk))
819 return -EAGAIN;
821 if (sk->sk_prot->sendpage)
822 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
823 return sock_no_sendpage(sock, page, offset, size, flags);
824 }
825 EXPORT_SYMBOL(inet_sendpage);
827 int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
828 size_t size, int flags)
829 {
830 struct sock *sk = sock->sk;
831 int addr_len = 0;
832 int err;
834 sock_rps_record_flow(sk);
836 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
837 flags & ~MSG_DONTWAIT, &addr_len);
838 if (err >= 0)
839 msg->msg_namelen = addr_len;
840 return err;
841 }
842 EXPORT_SYMBOL(inet_recvmsg);
844 int inet_shutdown(struct socket *sock, int how)
845 {
846 struct sock *sk = sock->sk;
847 int err = 0;
849 /* This should really check to make sure
850 * the socket is a TCP socket. (WHY AC...)
851 */
852 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
853 1->2 bit 2 snds.
854 2->3 */
855 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
856 return -EINVAL;
858 lock_sock(sk);
859 if (sock->state == SS_CONNECTING) {
860 if ((1 << sk->sk_state) &
861 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
862 sock->state = SS_DISCONNECTING;
863 else
864 sock->state = SS_CONNECTED;
865 }
867 switch (sk->sk_state) {
868 case TCP_CLOSE:
869 err = -ENOTCONN;
870 /* Hack to wake up other listeners, who can poll for
871 POLLHUP, even on eg. unconnected UDP sockets -- RR */
872 default:
873 sk->sk_shutdown |= how;
874 if (sk->sk_prot->shutdown)
875 sk->sk_prot->shutdown(sk, how);
876 break;
878 /* Remaining two branches are temporary solution for missing
879 * close() in multithreaded environment. It is _not_ a good idea,
880 * but we have no choice until close() is repaired at VFS level.
881 */
882 case TCP_LISTEN:
883 if (!(how & RCV_SHUTDOWN))
884 break;
885 /* Fall through */
886 case TCP_SYN_SENT:
887 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
888 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
889 break;
890 }
892 /* Wake up anyone sleeping in poll. */
893 sk->sk_state_change(sk);
894 release_sock(sk);
895 return err;
896 }
897 EXPORT_SYMBOL(inet_shutdown);
899 /*
900 * ioctl() calls you can issue on an INET socket. Most of these are
901 * device configuration and stuff and very rarely used. Some ioctls
902 * pass on to the socket itself.
903 *
904 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
905 * loads the devconfigure module does its configuring and unloads it.
906 * There's a good 20K of config code hanging around the kernel.
907 */
909 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
910 {
911 struct sock *sk = sock->sk;
912 int err = 0;
913 struct net *net = sock_net(sk);
915 switch (cmd) {
916 case SIOCGSTAMP:
917 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
918 break;
919 case SIOCGSTAMPNS:
920 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
921 break;
922 case SIOCADDRT:
923 case SIOCDELRT:
924 case SIOCRTMSG:
925 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
926 break;
927 case SIOCDARP:
928 case SIOCGARP:
929 case SIOCSARP:
930 err = arp_ioctl(net, cmd, (void __user *)arg);
931 break;
932 case SIOCGIFADDR:
933 case SIOCSIFADDR:
934 case SIOCGIFBRDADDR:
935 case SIOCSIFBRDADDR:
936 case SIOCGIFNETMASK:
937 case SIOCSIFNETMASK:
938 case SIOCGIFDSTADDR:
939 case SIOCSIFDSTADDR:
940 case SIOCSIFPFLAGS:
941 case SIOCGIFPFLAGS:
942 case SIOCSIFFLAGS:
943 case SIOCKILLADDR:
944 err = devinet_ioctl(net, cmd, (void __user *)arg);
945 break;
946 default:
947 if (sk->sk_prot->ioctl)
948 err = sk->sk_prot->ioctl(sk, cmd, arg);
949 else
950 err = -ENOIOCTLCMD;
951 break;
952 }
953 return err;
954 }
955 EXPORT_SYMBOL(inet_ioctl);
957 #ifdef CONFIG_COMPAT
958 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
959 {
960 struct sock *sk = sock->sk;
961 int err = -ENOIOCTLCMD;
963 if (sk->sk_prot->compat_ioctl)
964 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
966 return err;
967 }
968 #endif
970 const struct proto_ops inet_stream_ops = {
971 .family = PF_INET,
972 .owner = THIS_MODULE,
973 .release = inet_release,
974 .bind = inet_bind,
975 .connect = inet_stream_connect,
976 .socketpair = sock_no_socketpair,
977 .accept = inet_accept,
978 .getname = inet_getname,
979 .poll = tcp_poll,
980 .ioctl = inet_ioctl,
981 .listen = inet_listen,
982 .shutdown = inet_shutdown,
983 .setsockopt = sock_common_setsockopt,
984 .getsockopt = sock_common_getsockopt,
985 .sendmsg = inet_sendmsg,
986 .recvmsg = inet_recvmsg,
987 .mmap = sock_no_mmap,
988 .sendpage = inet_sendpage,
989 .splice_read = tcp_splice_read,
990 #ifdef CONFIG_COMPAT
991 .compat_setsockopt = compat_sock_common_setsockopt,
992 .compat_getsockopt = compat_sock_common_getsockopt,
993 .compat_ioctl = inet_compat_ioctl,
994 #endif
995 };
996 EXPORT_SYMBOL(inet_stream_ops);
998 const struct proto_ops inet_dgram_ops = {
999 .family = PF_INET,
1000 .owner = THIS_MODULE,
1001 .release = inet_release,
1002 .bind = inet_bind,
1003 .connect = inet_dgram_connect,
1004 .socketpair = sock_no_socketpair,
1005 .accept = sock_no_accept,
1006 .getname = inet_getname,
1007 .poll = udp_poll,
1008 .ioctl = inet_ioctl,
1009 .listen = sock_no_listen,
1010 .shutdown = inet_shutdown,
1011 .setsockopt = sock_common_setsockopt,
1012 .getsockopt = sock_common_getsockopt,
1013 .sendmsg = inet_sendmsg,
1014 .recvmsg = inet_recvmsg,
1015 .mmap = sock_no_mmap,
1016 .sendpage = inet_sendpage,
1017 #ifdef CONFIG_COMPAT
1018 .compat_setsockopt = compat_sock_common_setsockopt,
1019 .compat_getsockopt = compat_sock_common_getsockopt,
1020 .compat_ioctl = inet_compat_ioctl,
1021 #endif
1022 };
1023 EXPORT_SYMBOL(inet_dgram_ops);
1025 /*
1026 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1027 * udp_poll
1028 */
1029 static const struct proto_ops inet_sockraw_ops = {
1030 .family = PF_INET,
1031 .owner = THIS_MODULE,
1032 .release = inet_release,
1033 .bind = inet_bind,
1034 .connect = inet_dgram_connect,
1035 .socketpair = sock_no_socketpair,
1036 .accept = sock_no_accept,
1037 .getname = inet_getname,
1038 .poll = datagram_poll,
1039 .ioctl = inet_ioctl,
1040 .listen = sock_no_listen,
1041 .shutdown = inet_shutdown,
1042 .setsockopt = sock_common_setsockopt,
1043 .getsockopt = sock_common_getsockopt,
1044 .sendmsg = inet_sendmsg,
1045 .recvmsg = inet_recvmsg,
1046 .mmap = sock_no_mmap,
1047 .sendpage = inet_sendpage,
1048 #ifdef CONFIG_COMPAT
1049 .compat_setsockopt = compat_sock_common_setsockopt,
1050 .compat_getsockopt = compat_sock_common_getsockopt,
1051 .compat_ioctl = inet_compat_ioctl,
1052 #endif
1053 };
1055 static const struct net_proto_family inet_family_ops = {
1056 .family = PF_INET,
1057 .create = inet_create,
1058 .owner = THIS_MODULE,
1059 };
1061 /* Upon startup we insert all the elements in inetsw_array[] into
1062 * the linked list inetsw.
1063 */
1064 static struct inet_protosw inetsw_array[] =
1065 {
1066 {
1067 .type = SOCK_STREAM,
1068 .protocol = IPPROTO_TCP,
1069 .prot = &tcp_prot,
1070 .ops = &inet_stream_ops,
1071 .no_check = 0,
1072 .flags = INET_PROTOSW_PERMANENT |
1073 INET_PROTOSW_ICSK,
1074 },
1076 {
1077 .type = SOCK_DGRAM,
1078 .protocol = IPPROTO_UDP,
1079 .prot = &udp_prot,
1080 .ops = &inet_dgram_ops,
1081 .no_check = UDP_CSUM_DEFAULT,
1082 .flags = INET_PROTOSW_PERMANENT,
1083 },
1085 {
1086 .type = SOCK_DGRAM,
1087 .protocol = IPPROTO_ICMP,
1088 .prot = &ping_prot,
1089 .ops = &inet_dgram_ops,
1090 .no_check = UDP_CSUM_DEFAULT,
1091 .flags = INET_PROTOSW_REUSE,
1092 },
1094 {
1095 .type = SOCK_RAW,
1096 .protocol = IPPROTO_IP, /* wild card */
1097 .prot = &raw_prot,
1098 .ops = &inet_sockraw_ops,
1099 .no_check = UDP_CSUM_DEFAULT,
1100 .flags = INET_PROTOSW_REUSE,
1101 }
1102 };
1104 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1106 void inet_register_protosw(struct inet_protosw *p)
1107 {
1108 struct list_head *lh;
1109 struct inet_protosw *answer;
1110 int protocol = p->protocol;
1111 struct list_head *last_perm;
1113 spin_lock_bh(&inetsw_lock);
1115 if (p->type >= SOCK_MAX)
1116 goto out_illegal;
1118 /* If we are trying to override a permanent protocol, bail. */
1119 answer = NULL;
1120 last_perm = &inetsw[p->type];
1121 list_for_each(lh, &inetsw[p->type]) {
1122 answer = list_entry(lh, struct inet_protosw, list);
1124 /* Check only the non-wild match. */
1125 if (INET_PROTOSW_PERMANENT & answer->flags) {
1126 if (protocol == answer->protocol)
1127 break;
1128 last_perm = lh;
1129 }
1131 answer = NULL;
1132 }
1133 if (answer)
1134 goto out_permanent;
1136 /* Add the new entry after the last permanent entry if any, so that
1137 * the new entry does not override a permanent entry when matched with
1138 * a wild-card protocol. But it is allowed to override any existing
1139 * non-permanent entry. This means that when we remove this entry, the
1140 * system automatically returns to the old behavior.
1141 */
1142 list_add_rcu(&p->list, last_perm);
1143 out:
1144 spin_unlock_bh(&inetsw_lock);
1146 return;
1148 out_permanent:
1149 pr_err("Attempt to override permanent protocol %d\n", protocol);
1150 goto out;
1152 out_illegal:
1153 pr_err("Ignoring attempt to register invalid socket type %d\n",
1154 p->type);
1155 goto out;
1156 }
1157 EXPORT_SYMBOL(inet_register_protosw);
1159 void inet_unregister_protosw(struct inet_protosw *p)
1160 {
1161 if (INET_PROTOSW_PERMANENT & p->flags) {
1162 pr_err("Attempt to unregister permanent protocol %d\n",
1163 p->protocol);
1164 } else {
1165 spin_lock_bh(&inetsw_lock);
1166 list_del_rcu(&p->list);
1167 spin_unlock_bh(&inetsw_lock);
1169 synchronize_net();
1170 }
1171 }
1172 EXPORT_SYMBOL(inet_unregister_protosw);
1174 /*
1175 * Shall we try to damage output packets if routing dev changes?
1176 */
1178 int sysctl_ip_dynaddr __read_mostly;
1180 static int inet_sk_reselect_saddr(struct sock *sk)
1181 {
1182 struct inet_sock *inet = inet_sk(sk);
1183 __be32 old_saddr = inet->inet_saddr;
1184 __be32 daddr = inet->inet_daddr;
1185 struct flowi4 *fl4;
1186 struct rtable *rt;
1187 __be32 new_saddr;
1188 struct ip_options_rcu *inet_opt;
1190 inet_opt = rcu_dereference_protected(inet->inet_opt,
1191 sock_owned_by_user(sk));
1192 if (inet_opt && inet_opt->opt.srr)
1193 daddr = inet_opt->opt.faddr;
1195 /* Query new route. */
1196 fl4 = &inet->cork.fl.u.ip4;
1197 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1198 sk->sk_bound_dev_if, sk->sk_protocol,
1199 inet->inet_sport, inet->inet_dport, sk, false);
1200 if (IS_ERR(rt))
1201 return PTR_ERR(rt);
1203 sk_setup_caps(sk, &rt->dst);
1205 new_saddr = fl4->saddr;
1207 if (new_saddr == old_saddr)
1208 return 0;
1210 if (sysctl_ip_dynaddr > 1) {
1211 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1212 __func__, &old_saddr, &new_saddr);
1213 }
1215 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1217 /*
1218 * XXX The only one ugly spot where we need to
1219 * XXX really change the sockets identity after
1220 * XXX it has entered the hashes. -DaveM
1221 *
1222 * Besides that, it does not check for connection
1223 * uniqueness. Wait for troubles.
1224 */
1225 __sk_prot_rehash(sk);
1226 return 0;
1227 }
1229 int inet_sk_rebuild_header(struct sock *sk)
1230 {
1231 struct inet_sock *inet = inet_sk(sk);
1232 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1233 __be32 daddr;
1234 struct ip_options_rcu *inet_opt;
1235 struct flowi4 *fl4;
1236 int err;
1238 /* Route is OK, nothing to do. */
1239 if (rt)
1240 return 0;
1242 /* Reroute. */
1243 rcu_read_lock();
1244 inet_opt = rcu_dereference(inet->inet_opt);
1245 daddr = inet->inet_daddr;
1246 if (inet_opt && inet_opt->opt.srr)
1247 daddr = inet_opt->opt.faddr;
1248 rcu_read_unlock();
1249 fl4 = &inet->cork.fl.u.ip4;
1250 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1251 inet->inet_dport, inet->inet_sport,
1252 sk->sk_protocol, RT_CONN_FLAGS(sk),
1253 sk->sk_bound_dev_if);
1254 if (!IS_ERR(rt)) {
1255 err = 0;
1256 sk_setup_caps(sk, &rt->dst);
1257 } else {
1258 err = PTR_ERR(rt);
1260 /* Routing failed... */
1261 sk->sk_route_caps = 0;
1262 /*
1263 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1264 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1265 */
1266 if (!sysctl_ip_dynaddr ||
1267 sk->sk_state != TCP_SYN_SENT ||
1268 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1269 (err = inet_sk_reselect_saddr(sk)) != 0)
1270 sk->sk_err_soft = -err;
1271 }
1273 return err;
1274 }
1275 EXPORT_SYMBOL(inet_sk_rebuild_header);
1277 static int inet_gso_send_check(struct sk_buff *skb)
1278 {
1279 const struct net_offload *ops;
1280 const struct iphdr *iph;
1281 int proto;
1282 int ihl;
1283 int err = -EINVAL;
1285 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1286 goto out;
1288 iph = ip_hdr(skb);
1289 ihl = iph->ihl * 4;
1290 if (ihl < sizeof(*iph))
1291 goto out;
1293 if (unlikely(!pskb_may_pull(skb, ihl)))
1294 goto out;
1296 __skb_pull(skb, ihl);
1297 skb_reset_transport_header(skb);
1298 iph = ip_hdr(skb);
1299 proto = iph->protocol;
1300 err = -EPROTONOSUPPORT;
1302 rcu_read_lock();
1303 ops = rcu_dereference(inet_offloads[proto]);
1304 if (likely(ops && ops->callbacks.gso_send_check))
1305 err = ops->callbacks.gso_send_check(skb);
1306 rcu_read_unlock();
1308 out:
1309 return err;
1310 }
1312 static struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1313 netdev_features_t features)
1314 {
1315 struct sk_buff *segs = ERR_PTR(-EINVAL);
1316 const struct net_offload *ops;
1317 struct iphdr *iph;
1318 int proto;
1319 int ihl;
1320 int id;
1321 unsigned int offset = 0;
1323 if (!(features & NETIF_F_V4_CSUM))
1324 features &= ~NETIF_F_SG;
1326 if (unlikely(skb_shinfo(skb)->gso_type &
1327 ~(SKB_GSO_TCPV4 |
1328 SKB_GSO_UDP |
1329 SKB_GSO_DODGY |
1330 SKB_GSO_TCP_ECN |
1331 0)))
1332 goto out;
1334 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1335 goto out;
1337 iph = ip_hdr(skb);
1338 ihl = iph->ihl * 4;
1339 if (ihl < sizeof(*iph))
1340 goto out;
1342 if (unlikely(!pskb_may_pull(skb, ihl)))
1343 goto out;
1345 __skb_pull(skb, ihl);
1346 skb_reset_transport_header(skb);
1347 iph = ip_hdr(skb);
1348 id = ntohs(iph->id);
1349 proto = iph->protocol;
1350 segs = ERR_PTR(-EPROTONOSUPPORT);
1352 rcu_read_lock();
1353 ops = rcu_dereference(inet_offloads[proto]);
1354 if (likely(ops && ops->callbacks.gso_segment))
1355 segs = ops->callbacks.gso_segment(skb, features);
1356 rcu_read_unlock();
1358 if (!segs || IS_ERR(segs))
1359 goto out;
1361 skb = segs;
1362 do {
1363 iph = ip_hdr(skb);
1364 if (proto == IPPROTO_UDP) {
1365 iph->id = htons(id);
1366 iph->frag_off = htons(offset >> 3);
1367 if (skb->next != NULL)
1368 iph->frag_off |= htons(IP_MF);
1369 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1370 } else
1371 iph->id = htons(id++);
1372 iph->tot_len = htons(skb->len - skb->mac_len);
1373 iph->check = 0;
1374 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1375 } while ((skb = skb->next));
1377 out:
1378 return segs;
1379 }
1381 static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1382 struct sk_buff *skb)
1383 {
1384 const struct net_offload *ops;
1385 struct sk_buff **pp = NULL;
1386 struct sk_buff *p;
1387 const struct iphdr *iph;
1388 unsigned int hlen;
1389 unsigned int off;
1390 unsigned int id;
1391 int flush = 1;
1392 int proto;
1394 off = skb_gro_offset(skb);
1395 hlen = off + sizeof(*iph);
1396 iph = skb_gro_header_fast(skb, off);
1397 if (skb_gro_header_hard(skb, hlen)) {
1398 iph = skb_gro_header_slow(skb, hlen, off);
1399 if (unlikely(!iph))
1400 goto out;
1401 }
1403 proto = iph->protocol;
1405 rcu_read_lock();
1406 ops = rcu_dereference(inet_offloads[proto]);
1407 if (!ops || !ops->callbacks.gro_receive)
1408 goto out_unlock;
1410 if (*(u8 *)iph != 0x45)
1411 goto out_unlock;
1413 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1414 goto out_unlock;
1416 id = ntohl(*(__be32 *)&iph->id);
1417 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1418 id >>= 16;
1420 for (p = *head; p; p = p->next) {
1421 struct iphdr *iph2;
1423 if (!NAPI_GRO_CB(p)->same_flow)
1424 continue;
1426 iph2 = ip_hdr(p);
1428 if ((iph->protocol ^ iph2->protocol) |
1429 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1430 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1431 NAPI_GRO_CB(p)->same_flow = 0;
1432 continue;
1433 }
1435 /* All fields must match except length and checksum. */
1436 NAPI_GRO_CB(p)->flush |=
1437 (iph->ttl ^ iph2->ttl) |
1438 (iph->tos ^ iph2->tos) |
1439 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1441 NAPI_GRO_CB(p)->flush |= flush;
1442 }
1444 NAPI_GRO_CB(skb)->flush |= flush;
1445 skb_gro_pull(skb, sizeof(*iph));
1446 skb_set_transport_header(skb, skb_gro_offset(skb));
1448 pp = ops->callbacks.gro_receive(head, skb);
1450 out_unlock:
1451 rcu_read_unlock();
1453 out:
1454 NAPI_GRO_CB(skb)->flush |= flush;
1456 return pp;
1457 }
1459 static int inet_gro_complete(struct sk_buff *skb)
1460 {
1461 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1462 struct iphdr *iph = ip_hdr(skb);
1463 const struct net_offload *ops;
1464 int proto = iph->protocol;
1465 int err = -ENOSYS;
1467 csum_replace2(&iph->check, iph->tot_len, newlen);
1468 iph->tot_len = newlen;
1470 rcu_read_lock();
1471 ops = rcu_dereference(inet_offloads[proto]);
1472 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1473 goto out_unlock;
1475 err = ops->callbacks.gro_complete(skb);
1477 out_unlock:
1478 rcu_read_unlock();
1480 return err;
1481 }
1483 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1484 unsigned short type, unsigned char protocol,
1485 struct net *net)
1486 {
1487 struct socket *sock;
1488 int rc = sock_create_kern(family, type, protocol, &sock);
1490 if (rc == 0) {
1491 *sk = sock->sk;
1492 (*sk)->sk_allocation = GFP_ATOMIC;
1493 /*
1494 * Unhash it so that IP input processing does not even see it,
1495 * we do not wish this socket to see incoming packets.
1496 */
1497 (*sk)->sk_prot->unhash(*sk);
1499 sk_change_net(*sk, net);
1500 }
1501 return rc;
1502 }
1503 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1505 unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1506 {
1507 unsigned long res = 0;
1508 int i, j;
1510 for_each_possible_cpu(i) {
1511 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1512 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1513 }
1514 return res;
1515 }
1516 EXPORT_SYMBOL_GPL(snmp_fold_field);
1518 #if BITS_PER_LONG==32
1520 u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1521 {
1522 u64 res = 0;
1523 int cpu;
1525 for_each_possible_cpu(cpu) {
1526 void *bhptr;
1527 struct u64_stats_sync *syncp;
1528 u64 v;
1529 unsigned int start;
1531 bhptr = per_cpu_ptr(mib[0], cpu);
1532 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1533 do {
1534 start = u64_stats_fetch_begin_bh(syncp);
1535 v = *(((u64 *) bhptr) + offt);
1536 } while (u64_stats_fetch_retry_bh(syncp, start));
1538 res += v;
1539 }
1540 return res;
1541 }
1542 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1543 #endif
1545 int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1546 {
1547 BUG_ON(ptr == NULL);
1548 ptr[0] = __alloc_percpu(mibsize, align);
1549 if (!ptr[0])
1550 return -ENOMEM;
1551 #if SNMP_ARRAY_SZ == 2
1552 ptr[1] = __alloc_percpu(mibsize, align);
1553 if (!ptr[1]) {
1554 free_percpu(ptr[0]);
1555 ptr[0] = NULL;
1556 return -ENOMEM;
1557 }
1558 #endif
1559 return 0;
1560 }
1561 EXPORT_SYMBOL_GPL(snmp_mib_init);
1563 void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1564 {
1565 int i;
1567 BUG_ON(ptr == NULL);
1568 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1569 free_percpu(ptr[i]);
1570 ptr[i] = NULL;
1571 }
1572 }
1573 EXPORT_SYMBOL_GPL(snmp_mib_free);
1575 #ifdef CONFIG_IP_MULTICAST
1576 static const struct net_protocol igmp_protocol = {
1577 .handler = igmp_rcv,
1578 .netns_ok = 1,
1579 };
1580 #endif
1582 static const struct net_protocol tcp_protocol = {
1583 .early_demux = tcp_v4_early_demux,
1584 .handler = tcp_v4_rcv,
1585 .err_handler = tcp_v4_err,
1586 .no_policy = 1,
1587 .netns_ok = 1,
1588 };
1590 static const struct net_offload tcp_offload = {
1591 .callbacks = {
1592 .gso_send_check = tcp_v4_gso_send_check,
1593 .gso_segment = tcp_tso_segment,
1594 .gro_receive = tcp4_gro_receive,
1595 .gro_complete = tcp4_gro_complete,
1596 },
1597 };
1599 static const struct net_protocol udp_protocol = {
1600 .handler = udp_rcv,
1601 .err_handler = udp_err,
1602 .no_policy = 1,
1603 .netns_ok = 1,
1604 };
1606 static const struct net_offload udp_offload = {
1607 .callbacks = {
1608 .gso_send_check = udp4_ufo_send_check,
1609 .gso_segment = udp4_ufo_fragment,
1610 },
1611 };
1613 static const struct net_protocol icmp_protocol = {
1614 .handler = icmp_rcv,
1615 .err_handler = icmp_err,
1616 .no_policy = 1,
1617 .netns_ok = 1,
1618 };
1620 static __net_init int ipv4_mib_init_net(struct net *net)
1621 {
1622 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1623 sizeof(struct tcp_mib),
1624 __alignof__(struct tcp_mib)) < 0)
1625 goto err_tcp_mib;
1626 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1627 sizeof(struct ipstats_mib),
1628 __alignof__(struct ipstats_mib)) < 0)
1629 goto err_ip_mib;
1630 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1631 sizeof(struct linux_mib),
1632 __alignof__(struct linux_mib)) < 0)
1633 goto err_net_mib;
1634 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1635 sizeof(struct udp_mib),
1636 __alignof__(struct udp_mib)) < 0)
1637 goto err_udp_mib;
1638 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1639 sizeof(struct udp_mib),
1640 __alignof__(struct udp_mib)) < 0)
1641 goto err_udplite_mib;
1642 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1643 sizeof(struct icmp_mib),
1644 __alignof__(struct icmp_mib)) < 0)
1645 goto err_icmp_mib;
1646 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1647 GFP_KERNEL);
1648 if (!net->mib.icmpmsg_statistics)
1649 goto err_icmpmsg_mib;
1651 tcp_mib_init(net);
1652 return 0;
1654 err_icmpmsg_mib:
1655 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1656 err_icmp_mib:
1657 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1658 err_udplite_mib:
1659 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1660 err_udp_mib:
1661 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1662 err_net_mib:
1663 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1664 err_ip_mib:
1665 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1666 err_tcp_mib:
1667 return -ENOMEM;
1668 }
1670 static __net_exit void ipv4_mib_exit_net(struct net *net)
1671 {
1672 kfree(net->mib.icmpmsg_statistics);
1673 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1674 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1675 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1676 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1677 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1678 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1679 }
1681 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1682 .init = ipv4_mib_init_net,
1683 .exit = ipv4_mib_exit_net,
1684 };
1686 static int __init init_ipv4_mibs(void)
1687 {
1688 return register_pernet_subsys(&ipv4_mib_ops);
1689 }
1691 static int ipv4_proc_init(void);
1693 /*
1694 * IP protocol layer initialiser
1695 */
1697 static struct packet_offload ip_packet_offload __read_mostly = {
1698 .type = cpu_to_be16(ETH_P_IP),
1699 .callbacks = {
1700 .gso_send_check = inet_gso_send_check,
1701 .gso_segment = inet_gso_segment,
1702 .gro_receive = inet_gro_receive,
1703 .gro_complete = inet_gro_complete,
1704 },
1705 };
1707 static int __init ipv4_offload_init(void)
1708 {
1709 /*
1710 * Add offloads
1711 */
1712 if (inet_add_offload(&udp_offload, IPPROTO_UDP) < 0)
1713 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1714 if (inet_add_offload(&tcp_offload, IPPROTO_TCP) < 0)
1715 pr_crit("%s: Cannot add TCP protocol offlaod\n", __func__);
1717 dev_add_offload(&ip_packet_offload);
1718 return 0;
1719 }
1721 fs_initcall(ipv4_offload_init);
1723 static struct packet_type ip_packet_type __read_mostly = {
1724 .type = cpu_to_be16(ETH_P_IP),
1725 .func = ip_rcv,
1726 };
1728 static int __init inet_init(void)
1729 {
1730 struct sk_buff *dummy_skb;
1731 struct inet_protosw *q;
1732 struct list_head *r;
1733 int rc = -EINVAL;
1735 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1737 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1738 if (!sysctl_local_reserved_ports)
1739 goto out;
1741 rc = proto_register(&tcp_prot, 1);
1742 if (rc)
1743 goto out_free_reserved_ports;
1745 rc = proto_register(&udp_prot, 1);
1746 if (rc)
1747 goto out_unregister_tcp_proto;
1749 rc = proto_register(&raw_prot, 1);
1750 if (rc)
1751 goto out_unregister_udp_proto;
1753 rc = proto_register(&ping_prot, 1);
1754 if (rc)
1755 goto out_unregister_raw_proto;
1757 /*
1758 * Tell SOCKET that we are alive...
1759 */
1761 (void)sock_register(&inet_family_ops);
1763 #ifdef CONFIG_SYSCTL
1764 ip_static_sysctl_init();
1765 #endif
1767 tcp_prot.sysctl_mem = init_net.ipv4.sysctl_tcp_mem;
1769 /*
1770 * Add all the base protocols.
1771 */
1773 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1774 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1775 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1776 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1777 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1778 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1779 #ifdef CONFIG_IP_MULTICAST
1780 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1781 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1782 #endif
1784 /* Register the socket-side information for inet_create. */
1785 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1786 INIT_LIST_HEAD(r);
1788 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1789 inet_register_protosw(q);
1791 /*
1792 * Set the ARP module up
1793 */
1795 arp_init();
1797 /*
1798 * Set the IP module up
1799 */
1801 ip_init();
1803 tcp_v4_init();
1805 /* Setup TCP slab cache for open requests. */
1806 tcp_init();
1808 /* Setup UDP memory threshold */
1809 udp_init();
1811 /* Add UDP-Lite (RFC 3828) */
1812 udplite4_register();
1814 ping_init();
1816 /*
1817 * Set the ICMP layer up
1818 */
1820 if (icmp_init() < 0)
1821 panic("Failed to create the ICMP control socket.\n");
1823 /*
1824 * Initialise the multicast router
1825 */
1826 #if defined(CONFIG_IP_MROUTE)
1827 if (ip_mr_init())
1828 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1829 #endif
1830 /*
1831 * Initialise per-cpu ipv4 mibs
1832 */
1834 if (init_ipv4_mibs())
1835 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1837 ipv4_proc_init();
1839 ipfrag_init();
1841 dev_add_pack(&ip_packet_type);
1843 rc = 0;
1844 out:
1845 return rc;
1846 out_unregister_raw_proto:
1847 proto_unregister(&raw_prot);
1848 out_unregister_udp_proto:
1849 proto_unregister(&udp_prot);
1850 out_unregister_tcp_proto:
1851 proto_unregister(&tcp_prot);
1852 out_free_reserved_ports:
1853 kfree(sysctl_local_reserved_ports);
1854 goto out;
1855 }
1857 fs_initcall(inet_init);
1859 /* ------------------------------------------------------------------------ */
1861 #ifdef CONFIG_PROC_FS
1862 static int __init ipv4_proc_init(void)
1863 {
1864 int rc = 0;
1866 if (raw_proc_init())
1867 goto out_raw;
1868 if (tcp4_proc_init())
1869 goto out_tcp;
1870 if (udp4_proc_init())
1871 goto out_udp;
1872 if (ping_proc_init())
1873 goto out_ping;
1874 if (ip_misc_proc_init())
1875 goto out_misc;
1876 out:
1877 return rc;
1878 out_misc:
1879 ping_proc_exit();
1880 out_ping:
1881 udp4_proc_exit();
1882 out_udp:
1883 tcp4_proc_exit();
1884 out_tcp:
1885 raw_proc_exit();
1886 out_raw:
1887 rc = -ENOMEM;
1888 goto out;
1889 }
1891 #else /* CONFIG_PROC_FS */
1892 static int __init ipv4_proc_init(void)
1893 {
1894 return 0;
1895 }
1896 #endif /* CONFIG_PROC_FS */
1898 MODULE_ALIAS_NETPROTO(PF_INET);