projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
[deliverable/linux.git] / net / ipv4 / af_inet.c
1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PF_INET protocol family socket handler.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
12 *
13 * Changes (see also sock.c)
14 *
15 * piggy,
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
23 * structures
24 * when accept() ed
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
30 * Alan Cox,
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
40 * compactness.
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
46 * dumbly.
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
61 *
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
66 */
67
68 #define pr_fmt(fmt) "IPv4: " fmt
69
70 #include <linux/err.h>
71 #include <linux/errno.h>
72 #include <linux/types.h>
73 #include <linux/socket.h>
74 #include <linux/in.h>
75 #include <linux/kernel.h>
76 #include <linux/kmod.h>
77 #include <linux/sched.h>
78 #include <linux/timer.h>
79 #include <linux/string.h>
80 #include <linux/sockios.h>
81 #include <linux/net.h>
82 #include <linux/capability.h>
83 #include <linux/fcntl.h>
84 #include <linux/mm.h>
85 #include <linux/interrupt.h>
86 #include <linux/stat.h>
87 #include <linux/init.h>
88 #include <linux/poll.h>
89 #include <linux/netfilter_ipv4.h>
90 #include <linux/random.h>
91 #include <linux/slab.h>
92
93 #include <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/ip_tunnels.h>
116 #include <net/xfrm.h>
117 #include <net/net_namespace.h>
118 #include <net/secure_seq.h>
119 #ifdef CONFIG_IP_MROUTE
120 #include <linux/mroute.h>
121 #endif
122 #include <net/l3mdev.h>
123
124
125 /* The inetsw table contains everything that inet_create needs to
126 * build a new socket.
127 */
128 static struct list_head inetsw[SOCK_MAX];
129 static DEFINE_SPINLOCK(inetsw_lock);
130
131 /* New destruction routine */
132
133 void inet_sock_destruct(struct sock *sk)
134 {
135 struct inet_sock *inet = inet_sk(sk);
136
137 __skb_queue_purge(&sk->sk_receive_queue);
138 __skb_queue_purge(&sk->sk_error_queue);
139
140 sk_mem_reclaim(sk);
141
142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
143 pr_err("Attempt to release TCP socket in state %d %p\n",
144 sk->sk_state, sk);
145 return;
146 }
147 if (!sock_flag(sk, SOCK_DEAD)) {
148 pr_err("Attempt to release alive inet socket %p\n", sk);
149 return;
150 }
151
152 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
153 WARN_ON(atomic_read(&sk->sk_wmem_alloc));
154 WARN_ON(sk->sk_wmem_queued);
155 WARN_ON(sk->sk_forward_alloc);
156
157 kfree(rcu_dereference_protected(inet->inet_opt, 1));
158 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
159 dst_release(sk->sk_rx_dst);
160 sk_refcnt_debug_dec(sk);
161 }
162 EXPORT_SYMBOL(inet_sock_destruct);
163
164 /*
165 * The routines beyond this point handle the behaviour of an AF_INET
166 * socket object. Mostly it punts to the subprotocols of IP to do
167 * the work.
168 */
169
170 /*
171 * Automatically bind an unbound socket.
172 */
173
174 static int inet_autobind(struct sock *sk)
175 {
176 struct inet_sock *inet;
177 /* We may need to bind the socket. */
178 lock_sock(sk);
179 inet = inet_sk(sk);
180 if (!inet->inet_num) {
181 if (sk->sk_prot->get_port(sk, 0)) {
182 release_sock(sk);
183 return -EAGAIN;
184 }
185 inet->inet_sport = htons(inet->inet_num);
186 }
187 release_sock(sk);
188 return 0;
189 }
190
191 /*
192 * Move a socket into listening state.
193 */
194 int inet_listen(struct socket *sock, int backlog)
195 {
196 struct sock *sk = sock->sk;
197 unsigned char old_state;
198 int err;
199
200 lock_sock(sk);
201
202 err = -EINVAL;
203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
204 goto out;
205
206 old_state = sk->sk_state;
207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
208 goto out;
209
210 /* Really, if the socket is already in listen state
211 * we can only allow the backlog to be adjusted.
212 */
213 if (old_state != TCP_LISTEN) {
214 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
215 * Note that only TCP sockets (SOCK_STREAM) will reach here.
216 * Also fastopen backlog may already been set via the option
217 * because the socket was in TCP_LISTEN state previously but
218 * was shutdown() rather than close().
219 */
220 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
221 (sysctl_tcp_fastopen & TFO_SERVER_ENABLE) &&
222 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
223 fastopen_queue_tune(sk, backlog);
224 tcp_fastopen_init_key_once(true);
225 }
226
227 err = inet_csk_listen_start(sk, backlog);
228 if (err)
229 goto out;
230 }
231 sk->sk_max_ack_backlog = backlog;
232 err = 0;
233
234 out:
235 release_sock(sk);
236 return err;
237 }
238 EXPORT_SYMBOL(inet_listen);
239
240 /*
241 * Create an inet socket.
242 */
243
244 static int inet_create(struct net *net, struct socket *sock, int protocol,
245 int kern)
246 {
247 struct sock *sk;
248 struct inet_protosw *answer;
249 struct inet_sock *inet;
250 struct proto *answer_prot;
251 unsigned char answer_flags;
252 int try_loading_module = 0;
253 int err;
254
255 if (protocol < 0 || protocol >= IPPROTO_MAX)
256 return -EINVAL;
257
258 sock->state = SS_UNCONNECTED;
259
260 /* Look for the requested type/protocol pair. */
261 lookup_protocol:
262 err = -ESOCKTNOSUPPORT;
263 rcu_read_lock();
264 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
265
266 err = 0;
267 /* Check the non-wild match. */
268 if (protocol == answer->protocol) {
269 if (protocol != IPPROTO_IP)
270 break;
271 } else {
272 /* Check for the two wild cases. */
273 if (IPPROTO_IP == protocol) {
274 protocol = answer->protocol;
275 break;
276 }
277 if (IPPROTO_IP == answer->protocol)
278 break;
279 }
280 err = -EPROTONOSUPPORT;
281 }
282
283 if (unlikely(err)) {
284 if (try_loading_module < 2) {
285 rcu_read_unlock();
286 /*
287 * Be more specific, e.g. net-pf-2-proto-132-type-1
288 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
289 */
290 if (++try_loading_module == 1)
291 request_module("net-pf-%d-proto-%d-type-%d",
292 PF_INET, protocol, sock->type);
293 /*
294 * Fall back to generic, e.g. net-pf-2-proto-132
295 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
296 */
297 else
298 request_module("net-pf-%d-proto-%d",
299 PF_INET, protocol);
300 goto lookup_protocol;
301 } else
302 goto out_rcu_unlock;
303 }
304
305 err = -EPERM;
306 if (sock->type == SOCK_RAW && !kern &&
307 !ns_capable(net->user_ns, CAP_NET_RAW))
308 goto out_rcu_unlock;
309
310 sock->ops = answer->ops;
311 answer_prot = answer->prot;
312 answer_flags = answer->flags;
313 rcu_read_unlock();
314
315 WARN_ON(!answer_prot->slab);
316
317 err = -ENOBUFS;
318 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
319 if (!sk)
320 goto out;
321
322 err = 0;
323 if (INET_PROTOSW_REUSE & answer_flags)
324 sk->sk_reuse = SK_CAN_REUSE;
325
326 inet = inet_sk(sk);
327 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
328
329 inet->nodefrag = 0;
330
331 if (SOCK_RAW == sock->type) {
332 inet->inet_num = protocol;
333 if (IPPROTO_RAW == protocol)
334 inet->hdrincl = 1;
335 }
336
337 if (net->ipv4.sysctl_ip_no_pmtu_disc)
338 inet->pmtudisc = IP_PMTUDISC_DONT;
339 else
340 inet->pmtudisc = IP_PMTUDISC_WANT;
341
342 inet->inet_id = 0;
343
344 sock_init_data(sock, sk);
345
346 sk->sk_destruct = inet_sock_destruct;
347 sk->sk_protocol = protocol;
348 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
349
350 inet->uc_ttl = -1;
351 inet->mc_loop = 1;
352 inet->mc_ttl = 1;
353 inet->mc_all = 1;
354 inet->mc_index = 0;
355 inet->mc_list = NULL;
356 inet->rcv_tos = 0;
357
358 sk_refcnt_debug_inc(sk);
359
360 if (inet->inet_num) {
361 /* It assumes that any protocol which allows
362 * the user to assign a number at socket
363 * creation time automatically
364 * shares.
365 */
366 inet->inet_sport = htons(inet->inet_num);
367 /* Add to protocol hash chains. */
368 err = sk->sk_prot->hash(sk);
369 if (err) {
370 sk_common_release(sk);
371 goto out;
372 }
373 }
374
375 if (sk->sk_prot->init) {
376 err = sk->sk_prot->init(sk);
377 if (err)
378 sk_common_release(sk);
379 }
380 out:
381 return err;
382 out_rcu_unlock:
383 rcu_read_unlock();
384 goto out;
385 }
386
387
388 /*
389 * The peer socket should always be NULL (or else). When we call this
390 * function we are destroying the object and from then on nobody
391 * should refer to it.
392 */
393 int inet_release(struct socket *sock)
394 {
395 struct sock *sk = sock->sk;
396
397 if (sk) {
398 long timeout;
399
400 /* Applications forget to leave groups before exiting */
401 ip_mc_drop_socket(sk);
402
403 /* If linger is set, we don't return until the close
404 * is complete. Otherwise we return immediately. The
405 * actually closing is done the same either way.
406 *
407 * If the close is due to the process exiting, we never
408 * linger..
409 */
410 timeout = 0;
411 if (sock_flag(sk, SOCK_LINGER) &&
412 !(current->flags & PF_EXITING))
413 timeout = sk->sk_lingertime;
414 sock->sk = NULL;
415 sk->sk_prot->close(sk, timeout);
416 }
417 return 0;
418 }
419 EXPORT_SYMBOL(inet_release);
420
421 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
422 {
423 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
424 struct sock *sk = sock->sk;
425 struct inet_sock *inet = inet_sk(sk);
426 struct net *net = sock_net(sk);
427 unsigned short snum;
428 int chk_addr_ret;
429 u32 tb_id = RT_TABLE_LOCAL;
430 int err;
431
432 /* If the socket has its own bind function then use it. (RAW) */
433 if (sk->sk_prot->bind) {
434 err = sk->sk_prot->bind(sk, uaddr, addr_len);
435 goto out;
436 }
437 err = -EINVAL;
438 if (addr_len < sizeof(struct sockaddr_in))
439 goto out;
440
441 if (addr->sin_family != AF_INET) {
442 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
443 * only if s_addr is INADDR_ANY.
444 */
445 err = -EAFNOSUPPORT;
446 if (addr->sin_family != AF_UNSPEC ||
447 addr->sin_addr.s_addr != htonl(INADDR_ANY))
448 goto out;
449 }
450
451 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
452 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
453
454 /* Not specified by any standard per-se, however it breaks too
455 * many applications when removed. It is unfortunate since
456 * allowing applications to make a non-local bind solves
457 * several problems with systems using dynamic addressing.
458 * (ie. your servers still start up even if your ISDN link
459 * is temporarily down)
460 */
461 err = -EADDRNOTAVAIL;
462 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
463 !(inet->freebind || inet->transparent) &&
464 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
465 chk_addr_ret != RTN_LOCAL &&
466 chk_addr_ret != RTN_MULTICAST &&
467 chk_addr_ret != RTN_BROADCAST)
468 goto out;
469
470 snum = ntohs(addr->sin_port);
471 err = -EACCES;
472 if (snum && snum < PROT_SOCK &&
473 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
474 goto out;
475
476 /* We keep a pair of addresses. rcv_saddr is the one
477 * used by hash lookups, and saddr is used for transmit.
478 *
479 * In the BSD API these are the same except where it
480 * would be illegal to use them (multicast/broadcast) in
481 * which case the sending device address is used.
482 */
483 lock_sock(sk);
484
485 /* Check these errors (active socket, double bind). */
486 err = -EINVAL;
487 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
488 goto out_release_sock;
489
490 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
491 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
492 inet->inet_saddr = 0; /* Use device */
493
494 /* Make sure we are allowed to bind here. */
495 if ((snum || !inet->bind_address_no_port) &&
496 sk->sk_prot->get_port(sk, snum)) {
497 inet->inet_saddr = inet->inet_rcv_saddr = 0;
498 err = -EADDRINUSE;
499 goto out_release_sock;
500 }
501
502 if (inet->inet_rcv_saddr)
503 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
504 if (snum)
505 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
506 inet->inet_sport = htons(inet->inet_num);
507 inet->inet_daddr = 0;
508 inet->inet_dport = 0;
509 sk_dst_reset(sk);
510 err = 0;
511 out_release_sock:
512 release_sock(sk);
513 out:
514 return err;
515 }
516 EXPORT_SYMBOL(inet_bind);
517
518 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
519 int addr_len, int flags)
520 {
521 struct sock *sk = sock->sk;
522
523 if (addr_len < sizeof(uaddr->sa_family))
524 return -EINVAL;
525 if (uaddr->sa_family == AF_UNSPEC)
526 return sk->sk_prot->disconnect(sk, flags);
527
528 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
529 return -EAGAIN;
530 return sk->sk_prot->connect(sk, uaddr, addr_len);
531 }
532 EXPORT_SYMBOL(inet_dgram_connect);
533
534 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
535 {
536 DEFINE_WAIT(wait);
537
538 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
539 sk->sk_write_pending += writebias;
540
541 /* Basic assumption: if someone sets sk->sk_err, he _must_
542 * change state of the socket from TCP_SYN_*.
543 * Connect() does not allow to get error notifications
544 * without closing the socket.
545 */
546 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
547 release_sock(sk);
548 timeo = schedule_timeout(timeo);
549 lock_sock(sk);
550 if (signal_pending(current) || !timeo)
551 break;
552 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
553 }
554 finish_wait(sk_sleep(sk), &wait);
555 sk->sk_write_pending -= writebias;
556 return timeo;
557 }
558
559 /*
560 * Connect to a remote host. There is regrettably still a little
561 * TCP 'magic' in here.
562 */
563 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
564 int addr_len, int flags)
565 {
566 struct sock *sk = sock->sk;
567 int err;
568 long timeo;
569
570 if (addr_len < sizeof(uaddr->sa_family))
571 return -EINVAL;
572
573 if (uaddr->sa_family == AF_UNSPEC) {
574 err = sk->sk_prot->disconnect(sk, flags);
575 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
576 goto out;
577 }
578
579 switch (sock->state) {
580 default:
581 err = -EINVAL;
582 goto out;
583 case SS_CONNECTED:
584 err = -EISCONN;
585 goto out;
586 case SS_CONNECTING:
587 err = -EALREADY;
588 /* Fall out of switch with err, set for this state */
589 break;
590 case SS_UNCONNECTED:
591 err = -EISCONN;
592 if (sk->sk_state != TCP_CLOSE)
593 goto out;
594
595 err = sk->sk_prot->connect(sk, uaddr, addr_len);
596 if (err < 0)
597 goto out;
598
599 sock->state = SS_CONNECTING;
600
601 /* Just entered SS_CONNECTING state; the only
602 * difference is that return value in non-blocking
603 * case is EINPROGRESS, rather than EALREADY.
604 */
605 err = -EINPROGRESS;
606 break;
607 }
608
609 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
610
611 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
612 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
613 tcp_sk(sk)->fastopen_req &&
614 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
615
616 /* Error code is set above */
617 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
618 goto out;
619
620 err = sock_intr_errno(timeo);
621 if (signal_pending(current))
622 goto out;
623 }
624
625 /* Connection was closed by RST, timeout, ICMP error
626 * or another process disconnected us.
627 */
628 if (sk->sk_state == TCP_CLOSE)
629 goto sock_error;
630
631 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
632 * and error was received after socket entered established state.
633 * Hence, it is handled normally after connect() return successfully.
634 */
635
636 sock->state = SS_CONNECTED;
637 err = 0;
638 out:
639 return err;
640
641 sock_error:
642 err = sock_error(sk) ? : -ECONNABORTED;
643 sock->state = SS_UNCONNECTED;
644 if (sk->sk_prot->disconnect(sk, flags))
645 sock->state = SS_DISCONNECTING;
646 goto out;
647 }
648 EXPORT_SYMBOL(__inet_stream_connect);
649
650 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
651 int addr_len, int flags)
652 {
653 int err;
654
655 lock_sock(sock->sk);
656 err = __inet_stream_connect(sock, uaddr, addr_len, flags);
657 release_sock(sock->sk);
658 return err;
659 }
660 EXPORT_SYMBOL(inet_stream_connect);
661
662 /*
663 * Accept a pending connection. The TCP layer now gives BSD semantics.
664 */
665
666 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
667 {
668 struct sock *sk1 = sock->sk;
669 int err = -EINVAL;
670 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
671
672 if (!sk2)
673 goto do_err;
674
675 lock_sock(sk2);
676
677 sock_rps_record_flow(sk2);
678 WARN_ON(!((1 << sk2->sk_state) &
679 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
680 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
681
682 sock_graft(sk2, newsock);
683
684 newsock->state = SS_CONNECTED;
685 err = 0;
686 release_sock(sk2);
687 do_err:
688 return err;
689 }
690 EXPORT_SYMBOL(inet_accept);
691
692
693 /*
694 * This does both peername and sockname.
695 */
696 int inet_getname(struct socket *sock, struct sockaddr *uaddr,
697 int *uaddr_len, int peer)
698 {
699 struct sock *sk = sock->sk;
700 struct inet_sock *inet = inet_sk(sk);
701 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
702
703 sin->sin_family = AF_INET;
704 if (peer) {
705 if (!inet->inet_dport ||
706 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
707 peer == 1))
708 return -ENOTCONN;
709 sin->sin_port = inet->inet_dport;
710 sin->sin_addr.s_addr = inet->inet_daddr;
711 } else {
712 __be32 addr = inet->inet_rcv_saddr;
713 if (!addr)
714 addr = inet->inet_saddr;
715 sin->sin_port = inet->inet_sport;
716 sin->sin_addr.s_addr = addr;
717 }
718 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
719 *uaddr_len = sizeof(*sin);
720 return 0;
721 }
722 EXPORT_SYMBOL(inet_getname);
723
724 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
725 {
726 struct sock *sk = sock->sk;
727
728 sock_rps_record_flow(sk);
729
730 /* We may need to bind the socket. */
731 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
732 inet_autobind(sk))
733 return -EAGAIN;
734
735 return sk->sk_prot->sendmsg(sk, msg, size);
736 }
737 EXPORT_SYMBOL(inet_sendmsg);
738
739 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
740 size_t size, int flags)
741 {
742 struct sock *sk = sock->sk;
743
744 sock_rps_record_flow(sk);
745
746 /* We may need to bind the socket. */
747 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
748 inet_autobind(sk))
749 return -EAGAIN;
750
751 if (sk->sk_prot->sendpage)
752 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
753 return sock_no_sendpage(sock, page, offset, size, flags);
754 }
755 EXPORT_SYMBOL(inet_sendpage);
756
757 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
758 int flags)
759 {
760 struct sock *sk = sock->sk;
761 int addr_len = 0;
762 int err;
763
764 sock_rps_record_flow(sk);
765
766 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
767 flags & ~MSG_DONTWAIT, &addr_len);
768 if (err >= 0)
769 msg->msg_namelen = addr_len;
770 return err;
771 }
772 EXPORT_SYMBOL(inet_recvmsg);
773
774 int inet_shutdown(struct socket *sock, int how)
775 {
776 struct sock *sk = sock->sk;
777 int err = 0;
778
779 /* This should really check to make sure
780 * the socket is a TCP socket. (WHY AC...)
781 */
782 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
783 1->2 bit 2 snds.
784 2->3 */
785 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
786 return -EINVAL;
787
788 lock_sock(sk);
789 if (sock->state == SS_CONNECTING) {
790 if ((1 << sk->sk_state) &
791 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
792 sock->state = SS_DISCONNECTING;
793 else
794 sock->state = SS_CONNECTED;
795 }
796
797 switch (sk->sk_state) {
798 case TCP_CLOSE:
799 err = -ENOTCONN;
800 /* Hack to wake up other listeners, who can poll for
801 POLLHUP, even on eg. unconnected UDP sockets -- RR */
802 default:
803 sk->sk_shutdown |= how;
804 if (sk->sk_prot->shutdown)
805 sk->sk_prot->shutdown(sk, how);
806 break;
807
808 /* Remaining two branches are temporary solution for missing
809 * close() in multithreaded environment. It is _not_ a good idea,
810 * but we have no choice until close() is repaired at VFS level.
811 */
812 case TCP_LISTEN:
813 if (!(how & RCV_SHUTDOWN))
814 break;
815 /* Fall through */
816 case TCP_SYN_SENT:
817 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
818 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
819 break;
820 }
821
822 /* Wake up anyone sleeping in poll. */
823 sk->sk_state_change(sk);
824 release_sock(sk);
825 return err;
826 }
827 EXPORT_SYMBOL(inet_shutdown);
828
829 /*
830 * ioctl() calls you can issue on an INET socket. Most of these are
831 * device configuration and stuff and very rarely used. Some ioctls
832 * pass on to the socket itself.
833 *
834 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
835 * loads the devconfigure module does its configuring and unloads it.
836 * There's a good 20K of config code hanging around the kernel.
837 */
838
839 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
840 {
841 struct sock *sk = sock->sk;
842 int err = 0;
843 struct net *net = sock_net(sk);
844
845 switch (cmd) {
846 case SIOCGSTAMP:
847 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
848 break;
849 case SIOCGSTAMPNS:
850 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
851 break;
852 case SIOCADDRT:
853 case SIOCDELRT:
854 case SIOCRTMSG:
855 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
856 break;
857 case SIOCDARP:
858 case SIOCGARP:
859 case SIOCSARP:
860 err = arp_ioctl(net, cmd, (void __user *)arg);
861 break;
862 case SIOCGIFADDR:
863 case SIOCSIFADDR:
864 case SIOCGIFBRDADDR:
865 case SIOCSIFBRDADDR:
866 case SIOCGIFNETMASK:
867 case SIOCSIFNETMASK:
868 case SIOCGIFDSTADDR:
869 case SIOCSIFDSTADDR:
870 case SIOCSIFPFLAGS:
871 case SIOCGIFPFLAGS:
872 case SIOCSIFFLAGS:
873 err = devinet_ioctl(net, cmd, (void __user *)arg);
874 break;
875 default:
876 if (sk->sk_prot->ioctl)
877 err = sk->sk_prot->ioctl(sk, cmd, arg);
878 else
879 err = -ENOIOCTLCMD;
880 break;
881 }
882 return err;
883 }
884 EXPORT_SYMBOL(inet_ioctl);
885
886 #ifdef CONFIG_COMPAT
887 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
888 {
889 struct sock *sk = sock->sk;
890 int err = -ENOIOCTLCMD;
891
892 if (sk->sk_prot->compat_ioctl)
893 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
894
895 return err;
896 }
897 #endif
898
899 const struct proto_ops inet_stream_ops = {
900 .family = PF_INET,
901 .owner = THIS_MODULE,
902 .release = inet_release,
903 .bind = inet_bind,
904 .connect = inet_stream_connect,
905 .socketpair = sock_no_socketpair,
906 .accept = inet_accept,
907 .getname = inet_getname,
908 .poll = tcp_poll,
909 .ioctl = inet_ioctl,
910 .listen = inet_listen,
911 .shutdown = inet_shutdown,
912 .setsockopt = sock_common_setsockopt,
913 .getsockopt = sock_common_getsockopt,
914 .sendmsg = inet_sendmsg,
915 .recvmsg = inet_recvmsg,
916 .mmap = sock_no_mmap,
917 .sendpage = inet_sendpage,
918 .splice_read = tcp_splice_read,
919 .read_sock = tcp_read_sock,
920 .peek_len = tcp_peek_len,
921 #ifdef CONFIG_COMPAT
922 .compat_setsockopt = compat_sock_common_setsockopt,
923 .compat_getsockopt = compat_sock_common_getsockopt,
924 .compat_ioctl = inet_compat_ioctl,
925 #endif
926 };
927 EXPORT_SYMBOL(inet_stream_ops);
928
929 const struct proto_ops inet_dgram_ops = {
930 .family = PF_INET,
931 .owner = THIS_MODULE,
932 .release = inet_release,
933 .bind = inet_bind,
934 .connect = inet_dgram_connect,
935 .socketpair = sock_no_socketpair,
936 .accept = sock_no_accept,
937 .getname = inet_getname,
938 .poll = udp_poll,
939 .ioctl = inet_ioctl,
940 .listen = sock_no_listen,
941 .shutdown = inet_shutdown,
942 .setsockopt = sock_common_setsockopt,
943 .getsockopt = sock_common_getsockopt,
944 .sendmsg = inet_sendmsg,
945 .recvmsg = inet_recvmsg,
946 .mmap = sock_no_mmap,
947 .sendpage = inet_sendpage,
948 .set_peek_off = sk_set_peek_off,
949 #ifdef CONFIG_COMPAT
950 .compat_setsockopt = compat_sock_common_setsockopt,
951 .compat_getsockopt = compat_sock_common_getsockopt,
952 .compat_ioctl = inet_compat_ioctl,
953 #endif
954 };
955 EXPORT_SYMBOL(inet_dgram_ops);
956
957 /*
958 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
959 * udp_poll
960 */
961 static const struct proto_ops inet_sockraw_ops = {
962 .family = PF_INET,
963 .owner = THIS_MODULE,
964 .release = inet_release,
965 .bind = inet_bind,
966 .connect = inet_dgram_connect,
967 .socketpair = sock_no_socketpair,
968 .accept = sock_no_accept,
969 .getname = inet_getname,
970 .poll = datagram_poll,
971 .ioctl = inet_ioctl,
972 .listen = sock_no_listen,
973 .shutdown = inet_shutdown,
974 .setsockopt = sock_common_setsockopt,
975 .getsockopt = sock_common_getsockopt,
976 .sendmsg = inet_sendmsg,
977 .recvmsg = inet_recvmsg,
978 .mmap = sock_no_mmap,
979 .sendpage = inet_sendpage,
980 #ifdef CONFIG_COMPAT
981 .compat_setsockopt = compat_sock_common_setsockopt,
982 .compat_getsockopt = compat_sock_common_getsockopt,
983 .compat_ioctl = inet_compat_ioctl,
984 #endif
985 };
986
987 static const struct net_proto_family inet_family_ops = {
988 .family = PF_INET,
989 .create = inet_create,
990 .owner = THIS_MODULE,
991 };
992
993 /* Upon startup we insert all the elements in inetsw_array[] into
994 * the linked list inetsw.
995 */
996 static struct inet_protosw inetsw_array[] =
997 {
998 {
999 .type = SOCK_STREAM,
1000 .protocol = IPPROTO_TCP,
1001 .prot = &tcp_prot,
1002 .ops = &inet_stream_ops,
1003 .flags = INET_PROTOSW_PERMANENT |
1004 INET_PROTOSW_ICSK,
1005 },
1006
1007 {
1008 .type = SOCK_DGRAM,
1009 .protocol = IPPROTO_UDP,
1010 .prot = &udp_prot,
1011 .ops = &inet_dgram_ops,
1012 .flags = INET_PROTOSW_PERMANENT,
1013 },
1014
1015 {
1016 .type = SOCK_DGRAM,
1017 .protocol = IPPROTO_ICMP,
1018 .prot = &ping_prot,
1019 .ops = &inet_dgram_ops,
1020 .flags = INET_PROTOSW_REUSE,
1021 },
1022
1023 {
1024 .type = SOCK_RAW,
1025 .protocol = IPPROTO_IP, /* wild card */
1026 .prot = &raw_prot,
1027 .ops = &inet_sockraw_ops,
1028 .flags = INET_PROTOSW_REUSE,
1029 }
1030 };
1031
1032 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1033
1034 void inet_register_protosw(struct inet_protosw *p)
1035 {
1036 struct list_head *lh;
1037 struct inet_protosw *answer;
1038 int protocol = p->protocol;
1039 struct list_head *last_perm;
1040
1041 spin_lock_bh(&inetsw_lock);
1042
1043 if (p->type >= SOCK_MAX)
1044 goto out_illegal;
1045
1046 /* If we are trying to override a permanent protocol, bail. */
1047 last_perm = &inetsw[p->type];
1048 list_for_each(lh, &inetsw[p->type]) {
1049 answer = list_entry(lh, struct inet_protosw, list);
1050 /* Check only the non-wild match. */
1051 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1052 break;
1053 if (protocol == answer->protocol)
1054 goto out_permanent;
1055 last_perm = lh;
1056 }
1057
1058 /* Add the new entry after the last permanent entry if any, so that
1059 * the new entry does not override a permanent entry when matched with
1060 * a wild-card protocol. But it is allowed to override any existing
1061 * non-permanent entry. This means that when we remove this entry, the
1062 * system automatically returns to the old behavior.
1063 */
1064 list_add_rcu(&p->list, last_perm);
1065 out:
1066 spin_unlock_bh(&inetsw_lock);
1067
1068 return;
1069
1070 out_permanent:
1071 pr_err("Attempt to override permanent protocol %d\n", protocol);
1072 goto out;
1073
1074 out_illegal:
1075 pr_err("Ignoring attempt to register invalid socket type %d\n",
1076 p->type);
1077 goto out;
1078 }
1079 EXPORT_SYMBOL(inet_register_protosw);
1080
1081 void inet_unregister_protosw(struct inet_protosw *p)
1082 {
1083 if (INET_PROTOSW_PERMANENT & p->flags) {
1084 pr_err("Attempt to unregister permanent protocol %d\n",
1085 p->protocol);
1086 } else {
1087 spin_lock_bh(&inetsw_lock);
1088 list_del_rcu(&p->list);
1089 spin_unlock_bh(&inetsw_lock);
1090
1091 synchronize_net();
1092 }
1093 }
1094 EXPORT_SYMBOL(inet_unregister_protosw);
1095
1096 static int inet_sk_reselect_saddr(struct sock *sk)
1097 {
1098 struct inet_sock *inet = inet_sk(sk);
1099 __be32 old_saddr = inet->inet_saddr;
1100 __be32 daddr = inet->inet_daddr;
1101 struct flowi4 *fl4;
1102 struct rtable *rt;
1103 __be32 new_saddr;
1104 struct ip_options_rcu *inet_opt;
1105
1106 inet_opt = rcu_dereference_protected(inet->inet_opt,
1107 lockdep_sock_is_held(sk));
1108 if (inet_opt && inet_opt->opt.srr)
1109 daddr = inet_opt->opt.faddr;
1110
1111 /* Query new route. */
1112 fl4 = &inet->cork.fl.u.ip4;
1113 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1114 sk->sk_bound_dev_if, sk->sk_protocol,
1115 inet->inet_sport, inet->inet_dport, sk);
1116 if (IS_ERR(rt))
1117 return PTR_ERR(rt);
1118
1119 sk_setup_caps(sk, &rt->dst);
1120
1121 new_saddr = fl4->saddr;
1122
1123 if (new_saddr == old_saddr)
1124 return 0;
1125
1126 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1127 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1128 __func__, &old_saddr, &new_saddr);
1129 }
1130
1131 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1132
1133 /*
1134 * XXX The only one ugly spot where we need to
1135 * XXX really change the sockets identity after
1136 * XXX it has entered the hashes. -DaveM
1137 *
1138 * Besides that, it does not check for connection
1139 * uniqueness. Wait for troubles.
1140 */
1141 return __sk_prot_rehash(sk);
1142 }
1143
1144 int inet_sk_rebuild_header(struct sock *sk)
1145 {
1146 struct inet_sock *inet = inet_sk(sk);
1147 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1148 __be32 daddr;
1149 struct ip_options_rcu *inet_opt;
1150 struct flowi4 *fl4;
1151 int err;
1152
1153 /* Route is OK, nothing to do. */
1154 if (rt)
1155 return 0;
1156
1157 /* Reroute. */
1158 rcu_read_lock();
1159 inet_opt = rcu_dereference(inet->inet_opt);
1160 daddr = inet->inet_daddr;
1161 if (inet_opt && inet_opt->opt.srr)
1162 daddr = inet_opt->opt.faddr;
1163 rcu_read_unlock();
1164 fl4 = &inet->cork.fl.u.ip4;
1165 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1166 inet->inet_dport, inet->inet_sport,
1167 sk->sk_protocol, RT_CONN_FLAGS(sk),
1168 sk->sk_bound_dev_if);
1169 if (!IS_ERR(rt)) {
1170 err = 0;
1171 sk_setup_caps(sk, &rt->dst);
1172 } else {
1173 err = PTR_ERR(rt);
1174
1175 /* Routing failed... */
1176 sk->sk_route_caps = 0;
1177 /*
1178 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1179 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1180 */
1181 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1182 sk->sk_state != TCP_SYN_SENT ||
1183 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1184 (err = inet_sk_reselect_saddr(sk)) != 0)
1185 sk->sk_err_soft = -err;
1186 }
1187
1188 return err;
1189 }
1190 EXPORT_SYMBOL(inet_sk_rebuild_header);
1191
1192 struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1193 netdev_features_t features)
1194 {
1195 bool udpfrag = false, fixedid = false, encap;
1196 struct sk_buff *segs = ERR_PTR(-EINVAL);
1197 const struct net_offload *ops;
1198 unsigned int offset = 0;
1199 struct iphdr *iph;
1200 int proto, tot_len;
1201 int nhoff;
1202 int ihl;
1203 int id;
1204
1205 skb_reset_network_header(skb);
1206 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1207 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1208 goto out;
1209
1210 iph = ip_hdr(skb);
1211 ihl = iph->ihl * 4;
1212 if (ihl < sizeof(*iph))
1213 goto out;
1214
1215 id = ntohs(iph->id);
1216 proto = iph->protocol;
1217
1218 /* Warning: after this point, iph might be no longer valid */
1219 if (unlikely(!pskb_may_pull(skb, ihl)))
1220 goto out;
1221 __skb_pull(skb, ihl);
1222
1223 encap = SKB_GSO_CB(skb)->encap_level > 0;
1224 if (encap)
1225 features &= skb->dev->hw_enc_features;
1226 SKB_GSO_CB(skb)->encap_level += ihl;
1227
1228 skb_reset_transport_header(skb);
1229
1230 segs = ERR_PTR(-EPROTONOSUPPORT);
1231
1232 if (!skb->encapsulation || encap) {
1233 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1234 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1235
1236 /* fixed ID is invalid if DF bit is not set */
1237 if (fixedid && !(iph->frag_off & htons(IP_DF)))
1238 goto out;
1239 }
1240
1241 ops = rcu_dereference(inet_offloads[proto]);
1242 if (likely(ops && ops->callbacks.gso_segment))
1243 segs = ops->callbacks.gso_segment(skb, features);
1244
1245 if (IS_ERR_OR_NULL(segs))
1246 goto out;
1247
1248 skb = segs;
1249 do {
1250 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1251 if (udpfrag) {
1252 iph->frag_off = htons(offset >> 3);
1253 if (skb->next)
1254 iph->frag_off |= htons(IP_MF);
1255 offset += skb->len - nhoff - ihl;
1256 tot_len = skb->len - nhoff;
1257 } else if (skb_is_gso(skb)) {
1258 if (!fixedid) {
1259 iph->id = htons(id);
1260 id += skb_shinfo(skb)->gso_segs;
1261 }
1262 tot_len = skb_shinfo(skb)->gso_size +
1263 SKB_GSO_CB(skb)->data_offset +
1264 skb->head - (unsigned char *)iph;
1265 } else {
1266 if (!fixedid)
1267 iph->id = htons(id++);
1268 tot_len = skb->len - nhoff;
1269 }
1270 iph->tot_len = htons(tot_len);
1271 ip_send_check(iph);
1272 if (encap)
1273 skb_reset_inner_headers(skb);
1274 skb->network_header = (u8 *)iph - skb->head;
1275 } while ((skb = skb->next));
1276
1277 out:
1278 return segs;
1279 }
1280 EXPORT_SYMBOL(inet_gso_segment);
1281
1282 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1283 {
1284 const struct net_offload *ops;
1285 struct sk_buff **pp = NULL;
1286 struct sk_buff *p;
1287 const struct iphdr *iph;
1288 unsigned int hlen;
1289 unsigned int off;
1290 unsigned int id;
1291 int flush = 1;
1292 int proto;
1293
1294 off = skb_gro_offset(skb);
1295 hlen = off + sizeof(*iph);
1296 iph = skb_gro_header_fast(skb, off);
1297 if (skb_gro_header_hard(skb, hlen)) {
1298 iph = skb_gro_header_slow(skb, hlen, off);
1299 if (unlikely(!iph))
1300 goto out;
1301 }
1302
1303 proto = iph->protocol;
1304
1305 rcu_read_lock();
1306 ops = rcu_dereference(inet_offloads[proto]);
1307 if (!ops || !ops->callbacks.gro_receive)
1308 goto out_unlock;
1309
1310 if (*(u8 *)iph != 0x45)
1311 goto out_unlock;
1312
1313 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1314 goto out_unlock;
1315
1316 id = ntohl(*(__be32 *)&iph->id);
1317 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1318 id >>= 16;
1319
1320 for (p = *head; p; p = p->next) {
1321 struct iphdr *iph2;
1322 u16 flush_id;
1323
1324 if (!NAPI_GRO_CB(p)->same_flow)
1325 continue;
1326
1327 iph2 = (struct iphdr *)(p->data + off);
1328 /* The above works because, with the exception of the top
1329 * (inner most) layer, we only aggregate pkts with the same
1330 * hdr length so all the hdrs we'll need to verify will start
1331 * at the same offset.
1332 */
1333 if ((iph->protocol ^ iph2->protocol) |
1334 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1335 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1336 NAPI_GRO_CB(p)->same_flow = 0;
1337 continue;
1338 }
1339
1340 /* All fields must match except length and checksum. */
1341 NAPI_GRO_CB(p)->flush |=
1342 (iph->ttl ^ iph2->ttl) |
1343 (iph->tos ^ iph2->tos) |
1344 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1345
1346 NAPI_GRO_CB(p)->flush |= flush;
1347
1348 /* We need to store of the IP ID check to be included later
1349 * when we can verify that this packet does in fact belong
1350 * to a given flow.
1351 */
1352 flush_id = (u16)(id - ntohs(iph2->id));
1353
1354 /* This bit of code makes it much easier for us to identify
1355 * the cases where we are doing atomic vs non-atomic IP ID
1356 * checks. Specifically an atomic check can return IP ID
1357 * values 0 - 0xFFFF, while a non-atomic check can only
1358 * return 0 or 0xFFFF.
1359 */
1360 if (!NAPI_GRO_CB(p)->is_atomic ||
1361 !(iph->frag_off & htons(IP_DF))) {
1362 flush_id ^= NAPI_GRO_CB(p)->count;
1363 flush_id = flush_id ? 0xFFFF : 0;
1364 }
1365
1366 /* If the previous IP ID value was based on an atomic
1367 * datagram we can overwrite the value and ignore it.
1368 */
1369 if (NAPI_GRO_CB(skb)->is_atomic)
1370 NAPI_GRO_CB(p)->flush_id = flush_id;
1371 else
1372 NAPI_GRO_CB(p)->flush_id |= flush_id;
1373 }
1374
1375 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1376 NAPI_GRO_CB(skb)->flush |= flush;
1377 skb_set_network_header(skb, off);
1378 /* The above will be needed by the transport layer if there is one
1379 * immediately following this IP hdr.
1380 */
1381
1382 /* Note : No need to call skb_gro_postpull_rcsum() here,
1383 * as we already checked checksum over ipv4 header was 0
1384 */
1385 skb_gro_pull(skb, sizeof(*iph));
1386 skb_set_transport_header(skb, skb_gro_offset(skb));
1387
1388 pp = ops->callbacks.gro_receive(head, skb);
1389
1390 out_unlock:
1391 rcu_read_unlock();
1392
1393 out:
1394 NAPI_GRO_CB(skb)->flush |= flush;
1395
1396 return pp;
1397 }
1398 EXPORT_SYMBOL(inet_gro_receive);
1399
1400 static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1401 struct sk_buff *skb)
1402 {
1403 if (NAPI_GRO_CB(skb)->encap_mark) {
1404 NAPI_GRO_CB(skb)->flush = 1;
1405 return NULL;
1406 }
1407
1408 NAPI_GRO_CB(skb)->encap_mark = 1;
1409
1410 return inet_gro_receive(head, skb);
1411 }
1412
1413 #define SECONDS_PER_DAY 86400
1414
1415 /* inet_current_timestamp - Return IP network timestamp
1416 *
1417 * Return milliseconds since midnight in network byte order.
1418 */
1419 __be32 inet_current_timestamp(void)
1420 {
1421 u32 secs;
1422 u32 msecs;
1423 struct timespec64 ts;
1424
1425 ktime_get_real_ts64(&ts);
1426
1427 /* Get secs since midnight. */
1428 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1429 /* Convert to msecs. */
1430 msecs = secs * MSEC_PER_SEC;
1431 /* Convert nsec to msec. */
1432 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1433
1434 /* Convert to network byte order. */
1435 return htonl(msecs);
1436 }
1437 EXPORT_SYMBOL(inet_current_timestamp);
1438
1439 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1440 {
1441 if (sk->sk_family == AF_INET)
1442 return ip_recv_error(sk, msg, len, addr_len);
1443 #if IS_ENABLED(CONFIG_IPV6)
1444 if (sk->sk_family == AF_INET6)
1445 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1446 #endif
1447 return -EINVAL;
1448 }
1449
1450 int inet_gro_complete(struct sk_buff *skb, int nhoff)
1451 {
1452 __be16 newlen = htons(skb->len - nhoff);
1453 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1454 const struct net_offload *ops;
1455 int proto = iph->protocol;
1456 int err = -ENOSYS;
1457
1458 if (skb->encapsulation)
1459 skb_set_inner_network_header(skb, nhoff);
1460
1461 csum_replace2(&iph->check, iph->tot_len, newlen);
1462 iph->tot_len = newlen;
1463
1464 rcu_read_lock();
1465 ops = rcu_dereference(inet_offloads[proto]);
1466 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1467 goto out_unlock;
1468
1469 /* Only need to add sizeof(*iph) to get to the next hdr below
1470 * because any hdr with option will have been flushed in
1471 * inet_gro_receive().
1472 */
1473 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1474
1475 out_unlock:
1476 rcu_read_unlock();
1477
1478 return err;
1479 }
1480 EXPORT_SYMBOL(inet_gro_complete);
1481
1482 static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1483 {
1484 skb->encapsulation = 1;
1485 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1486 return inet_gro_complete(skb, nhoff);
1487 }
1488
1489 int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1490 unsigned short type, unsigned char protocol,
1491 struct net *net)
1492 {
1493 struct socket *sock;
1494 int rc = sock_create_kern(net, family, type, protocol, &sock);
1495
1496 if (rc == 0) {
1497 *sk = sock->sk;
1498 (*sk)->sk_allocation = GFP_ATOMIC;
1499 /*
1500 * Unhash it so that IP input processing does not even see it,
1501 * we do not wish this socket to see incoming packets.
1502 */
1503 (*sk)->sk_prot->unhash(*sk);
1504 }
1505 return rc;
1506 }
1507 EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1508
1509 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1510 {
1511 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1512 }
1513 EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1514
1515 unsigned long snmp_fold_field(void __percpu *mib, int offt)
1516 {
1517 unsigned long res = 0;
1518 int i;
1519
1520 for_each_possible_cpu(i)
1521 res += snmp_get_cpu_field(mib, i, offt);
1522 return res;
1523 }
1524 EXPORT_SYMBOL_GPL(snmp_fold_field);
1525
1526 #if BITS_PER_LONG==32
1527
1528 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1529 size_t syncp_offset)
1530 {
1531 void *bhptr;
1532 struct u64_stats_sync *syncp;
1533 u64 v;
1534 unsigned int start;
1535
1536 bhptr = per_cpu_ptr(mib, cpu);
1537 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1538 do {
1539 start = u64_stats_fetch_begin_irq(syncp);
1540 v = *(((u64 *)bhptr) + offt);
1541 } while (u64_stats_fetch_retry_irq(syncp, start));
1542
1543 return v;
1544 }
1545 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1546
1547 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1548 {
1549 u64 res = 0;
1550 int cpu;
1551
1552 for_each_possible_cpu(cpu) {
1553 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1554 }
1555 return res;
1556 }
1557 EXPORT_SYMBOL_GPL(snmp_fold_field64);
1558 #endif
1559
1560 #ifdef CONFIG_IP_MULTICAST
1561 static const struct net_protocol igmp_protocol = {
1562 .handler = igmp_rcv,
1563 .netns_ok = 1,
1564 };
1565 #endif
1566
1567 static const struct net_protocol tcp_protocol = {
1568 .early_demux = tcp_v4_early_demux,
1569 .handler = tcp_v4_rcv,
1570 .err_handler = tcp_v4_err,
1571 .no_policy = 1,
1572 .netns_ok = 1,
1573 .icmp_strict_tag_validation = 1,
1574 };
1575
1576 static const struct net_protocol udp_protocol = {
1577 .early_demux = udp_v4_early_demux,
1578 .handler = udp_rcv,
1579 .err_handler = udp_err,
1580 .no_policy = 1,
1581 .netns_ok = 1,
1582 };
1583
1584 static const struct net_protocol icmp_protocol = {
1585 .handler = icmp_rcv,
1586 .err_handler = icmp_err,
1587 .no_policy = 1,
1588 .netns_ok = 1,
1589 };
1590
1591 static __net_init int ipv4_mib_init_net(struct net *net)
1592 {
1593 int i;
1594
1595 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1596 if (!net->mib.tcp_statistics)
1597 goto err_tcp_mib;
1598 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1599 if (!net->mib.ip_statistics)
1600 goto err_ip_mib;
1601
1602 for_each_possible_cpu(i) {
1603 struct ipstats_mib *af_inet_stats;
1604 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1605 u64_stats_init(&af_inet_stats->syncp);
1606 }
1607
1608 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1609 if (!net->mib.net_statistics)
1610 goto err_net_mib;
1611 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1612 if (!net->mib.udp_statistics)
1613 goto err_udp_mib;
1614 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1615 if (!net->mib.udplite_statistics)
1616 goto err_udplite_mib;
1617 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1618 if (!net->mib.icmp_statistics)
1619 goto err_icmp_mib;
1620 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1621 GFP_KERNEL);
1622 if (!net->mib.icmpmsg_statistics)
1623 goto err_icmpmsg_mib;
1624
1625 tcp_mib_init(net);
1626 return 0;
1627
1628 err_icmpmsg_mib:
1629 free_percpu(net->mib.icmp_statistics);
1630 err_icmp_mib:
1631 free_percpu(net->mib.udplite_statistics);
1632 err_udplite_mib:
1633 free_percpu(net->mib.udp_statistics);
1634 err_udp_mib:
1635 free_percpu(net->mib.net_statistics);
1636 err_net_mib:
1637 free_percpu(net->mib.ip_statistics);
1638 err_ip_mib:
1639 free_percpu(net->mib.tcp_statistics);
1640 err_tcp_mib:
1641 return -ENOMEM;
1642 }
1643
1644 static __net_exit void ipv4_mib_exit_net(struct net *net)
1645 {
1646 kfree(net->mib.icmpmsg_statistics);
1647 free_percpu(net->mib.icmp_statistics);
1648 free_percpu(net->mib.udplite_statistics);
1649 free_percpu(net->mib.udp_statistics);
1650 free_percpu(net->mib.net_statistics);
1651 free_percpu(net->mib.ip_statistics);
1652 free_percpu(net->mib.tcp_statistics);
1653 }
1654
1655 static __net_initdata struct pernet_operations ipv4_mib_ops = {
1656 .init = ipv4_mib_init_net,
1657 .exit = ipv4_mib_exit_net,
1658 };
1659
1660 static int __init init_ipv4_mibs(void)
1661 {
1662 return register_pernet_subsys(&ipv4_mib_ops);
1663 }
1664
1665 static __net_init int inet_init_net(struct net *net)
1666 {
1667 /*
1668 * Set defaults for local port range
1669 */
1670 seqlock_init(&net->ipv4.ip_local_ports.lock);
1671 net->ipv4.ip_local_ports.range[0] = 32768;
1672 net->ipv4.ip_local_ports.range[1] = 60999;
1673
1674 seqlock_init(&net->ipv4.ping_group_range.lock);
1675 /*
1676 * Sane defaults - nobody may create ping sockets.
1677 * Boot scripts should set this to distro-specific group.
1678 */
1679 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1680 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1681
1682 /* Default values for sysctl-controlled parameters.
1683 * We set them here, in case sysctl is not compiled.
1684 */
1685 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1686 net->ipv4.sysctl_ip_dynaddr = 0;
1687 net->ipv4.sysctl_ip_early_demux = 1;
1688
1689 return 0;
1690 }
1691
1692 static __net_exit void inet_exit_net(struct net *net)
1693 {
1694 }
1695
1696 static __net_initdata struct pernet_operations af_inet_ops = {
1697 .init = inet_init_net,
1698 .exit = inet_exit_net,
1699 };
1700
1701 static int __init init_inet_pernet_ops(void)
1702 {
1703 return register_pernet_subsys(&af_inet_ops);
1704 }
1705
1706 static int ipv4_proc_init(void);
1707
1708 /*
1709 * IP protocol layer initialiser
1710 */
1711
1712 static struct packet_offload ip_packet_offload __read_mostly = {
1713 .type = cpu_to_be16(ETH_P_IP),
1714 .callbacks = {
1715 .gso_segment = inet_gso_segment,
1716 .gro_receive = inet_gro_receive,
1717 .gro_complete = inet_gro_complete,
1718 },
1719 };
1720
1721 static const struct net_offload ipip_offload = {
1722 .callbacks = {
1723 .gso_segment = inet_gso_segment,
1724 .gro_receive = ipip_gro_receive,
1725 .gro_complete = ipip_gro_complete,
1726 },
1727 };
1728
1729 static int __init ipv4_offload_init(void)
1730 {
1731 /*
1732 * Add offloads
1733 */
1734 if (udpv4_offload_init() < 0)
1735 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1736 if (tcpv4_offload_init() < 0)
1737 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1738
1739 dev_add_offload(&ip_packet_offload);
1740 inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1741 return 0;
1742 }
1743
1744 fs_initcall(ipv4_offload_init);
1745
1746 static struct packet_type ip_packet_type __read_mostly = {
1747 .type = cpu_to_be16(ETH_P_IP),
1748 .func = ip_rcv,
1749 };
1750
1751 static int __init inet_init(void)
1752 {
1753 struct inet_protosw *q;
1754 struct list_head *r;
1755 int rc = -EINVAL;
1756
1757 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1758
1759 rc = proto_register(&tcp_prot, 1);
1760 if (rc)
1761 goto out;
1762
1763 rc = proto_register(&udp_prot, 1);
1764 if (rc)
1765 goto out_unregister_tcp_proto;
1766
1767 rc = proto_register(&raw_prot, 1);
1768 if (rc)
1769 goto out_unregister_udp_proto;
1770
1771 rc = proto_register(&ping_prot, 1);
1772 if (rc)
1773 goto out_unregister_raw_proto;
1774
1775 /*
1776 * Tell SOCKET that we are alive...
1777 */
1778
1779 (void)sock_register(&inet_family_ops);
1780
1781 #ifdef CONFIG_SYSCTL
1782 ip_static_sysctl_init();
1783 #endif
1784
1785 /*
1786 * Add all the base protocols.
1787 */
1788
1789 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1790 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1791 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1792 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1793 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1794 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1795 #ifdef CONFIG_IP_MULTICAST
1796 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1797 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1798 #endif
1799
1800 /* Register the socket-side information for inet_create. */
1801 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1802 INIT_LIST_HEAD(r);
1803
1804 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1805 inet_register_protosw(q);
1806
1807 /*
1808 * Set the ARP module up
1809 */
1810
1811 arp_init();
1812
1813 /*
1814 * Set the IP module up
1815 */
1816
1817 ip_init();
1818
1819 tcp_v4_init();
1820
1821 /* Setup TCP slab cache for open requests. */
1822 tcp_init();
1823
1824 /* Setup UDP memory threshold */
1825 udp_init();
1826
1827 /* Add UDP-Lite (RFC 3828) */
1828 udplite4_register();
1829
1830 ping_init();
1831
1832 /*
1833 * Set the ICMP layer up
1834 */
1835
1836 if (icmp_init() < 0)
1837 panic("Failed to create the ICMP control socket.\n");
1838
1839 /*
1840 * Initialise the multicast router
1841 */
1842 #if defined(CONFIG_IP_MROUTE)
1843 if (ip_mr_init())
1844 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1845 #endif
1846
1847 if (init_inet_pernet_ops())
1848 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1849 /*
1850 * Initialise per-cpu ipv4 mibs
1851 */
1852
1853 if (init_ipv4_mibs())
1854 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1855
1856 ipv4_proc_init();
1857
1858 ipfrag_init();
1859
1860 dev_add_pack(&ip_packet_type);
1861
1862 ip_tunnel_core_init();
1863
1864 rc = 0;
1865 out:
1866 return rc;
1867 out_unregister_raw_proto:
1868 proto_unregister(&raw_prot);
1869 out_unregister_udp_proto:
1870 proto_unregister(&udp_prot);
1871 out_unregister_tcp_proto:
1872 proto_unregister(&tcp_prot);
1873 goto out;
1874 }
1875
1876 fs_initcall(inet_init);
1877
1878 /* ------------------------------------------------------------------------ */
1879
1880 #ifdef CONFIG_PROC_FS
1881 static int __init ipv4_proc_init(void)
1882 {
1883 int rc = 0;
1884
1885 if (raw_proc_init())
1886 goto out_raw;
1887 if (tcp4_proc_init())
1888 goto out_tcp;
1889 if (udp4_proc_init())
1890 goto out_udp;
1891 if (ping_proc_init())
1892 goto out_ping;
1893 if (ip_misc_proc_init())
1894 goto out_misc;
1895 out:
1896 return rc;
1897 out_misc:
1898 ping_proc_exit();
1899 out_ping:
1900 udp4_proc_exit();
1901 out_udp:
1902 tcp4_proc_exit();
1903 out_tcp:
1904 raw_proc_exit();
1905 out_raw:
1906 rc = -ENOMEM;
1907 goto out;
1908 }
1909
1910 #else /* CONFIG_PROC_FS */
1911 static int __init ipv4_proc_init(void)
1912 {
1913 return 0;
1914 }
1915 #endif /* CONFIG_PROC_FS */
Linux kernel - Deliverables
This page took 0.067274 seconds and 6 git commands to generate.