Loading...
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PF_INET protocol family socket handler.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
12 *
13 * Changes (see also sock.c)
14 *
15 * piggy,
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
23 * structures
24 * when accept() ed
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
30 * Alan Cox,
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
40 * compactness.
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
46 * dumbly.
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
61 *
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
66 */
67
68#define pr_fmt(fmt) "IPv4: " fmt
69
70#include <linux/err.h>
71#include <linux/errno.h>
72#include <linux/types.h>
73#include <linux/socket.h>
74#include <linux/in.h>
75#include <linux/kernel.h>
76#include <linux/kmod.h>
77#include <linux/sched.h>
78#include <linux/timer.h>
79#include <linux/string.h>
80#include <linux/sockios.h>
81#include <linux/net.h>
82#include <linux/capability.h>
83#include <linux/fcntl.h>
84#include <linux/mm.h>
85#include <linux/interrupt.h>
86#include <linux/stat.h>
87#include <linux/init.h>
88#include <linux/poll.h>
89#include <linux/netfilter_ipv4.h>
90#include <linux/random.h>
91#include <linux/slab.h>
92
93#include <linux/uaccess.h>
94
95#include <linux/inet.h>
96#include <linux/igmp.h>
97#include <linux/inetdevice.h>
98#include <linux/netdevice.h>
99#include <net/checksum.h>
100#include <net/ip.h>
101#include <net/protocol.h>
102#include <net/arp.h>
103#include <net/route.h>
104#include <net/ip_fib.h>
105#include <net/inet_connection_sock.h>
106#include <net/tcp.h>
107#include <net/udp.h>
108#include <net/udplite.h>
109#include <net/ping.h>
110#include <linux/skbuff.h>
111#include <net/sock.h>
112#include <net/raw.h>
113#include <net/icmp.h>
114#include <net/inet_common.h>
115#include <net/ip_tunnels.h>
116#include <net/xfrm.h>
117#include <net/net_namespace.h>
118#include <net/secure_seq.h>
119#ifdef CONFIG_IP_MROUTE
120#include <linux/mroute.h>
121#endif
122#include <net/l3mdev.h>
123
124#include <trace/events/sock.h>
125
126/* The inetsw table contains everything that inet_create needs to
127 * build a new socket.
128 */
129static struct list_head inetsw[SOCK_MAX];
130static DEFINE_SPINLOCK(inetsw_lock);
131
132/* New destruction routine */
133
134void inet_sock_destruct(struct sock *sk)
135{
136 struct inet_sock *inet = inet_sk(sk);
137
138 __skb_queue_purge(&sk->sk_receive_queue);
139 __skb_queue_purge(&sk->sk_error_queue);
140
141 sk_mem_reclaim(sk);
142
143 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) {
144 pr_err("Attempt to release TCP socket in state %d %p\n",
145 sk->sk_state, sk);
146 return;
147 }
148 if (!sock_flag(sk, SOCK_DEAD)) {
149 pr_err("Attempt to release alive inet socket %p\n", sk);
150 return;
151 }
152
153 WARN_ON(atomic_read(&sk->sk_rmem_alloc));
154 WARN_ON(refcount_read(&sk->sk_wmem_alloc));
155 WARN_ON(sk->sk_wmem_queued);
156 WARN_ON(sk->sk_forward_alloc);
157
158 kfree(rcu_dereference_protected(inet->inet_opt, 1));
159 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1));
160 dst_release(sk->sk_rx_dst);
161 sk_refcnt_debug_dec(sk);
162}
163EXPORT_SYMBOL(inet_sock_destruct);
164
165/*
166 * The routines beyond this point handle the behaviour of an AF_INET
167 * socket object. Mostly it punts to the subprotocols of IP to do
168 * the work.
169 */
170
171/*
172 * Automatically bind an unbound socket.
173 */
174
175static int inet_autobind(struct sock *sk)
176{
177 struct inet_sock *inet;
178 /* We may need to bind the socket. */
179 lock_sock(sk);
180 inet = inet_sk(sk);
181 if (!inet->inet_num) {
182 if (sk->sk_prot->get_port(sk, 0)) {
183 release_sock(sk);
184 return -EAGAIN;
185 }
186 inet->inet_sport = htons(inet->inet_num);
187 }
188 release_sock(sk);
189 return 0;
190}
191
192/*
193 * Move a socket into listening state.
194 */
195int inet_listen(struct socket *sock, int backlog)
196{
197 struct sock *sk = sock->sk;
198 unsigned char old_state;
199 int err, tcp_fastopen;
200
201 lock_sock(sk);
202
203 err = -EINVAL;
204 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
205 goto out;
206
207 old_state = sk->sk_state;
208 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
209 goto out;
210
211 /* Really, if the socket is already in listen state
212 * we can only allow the backlog to be adjusted.
213 */
214 if (old_state != TCP_LISTEN) {
215 /* Enable TFO w/o requiring TCP_FASTOPEN socket option.
216 * Note that only TCP sockets (SOCK_STREAM) will reach here.
217 * Also fastopen backlog may already been set via the option
218 * because the socket was in TCP_LISTEN state previously but
219 * was shutdown() rather than close().
220 */
221 tcp_fastopen = sock_net(sk)->ipv4.sysctl_tcp_fastopen;
222 if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) &&
223 (tcp_fastopen & TFO_SERVER_ENABLE) &&
224 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) {
225 fastopen_queue_tune(sk, backlog);
226 tcp_fastopen_init_key_once(sock_net(sk));
227 }
228
229 err = inet_csk_listen_start(sk, backlog);
230 if (err)
231 goto out;
232 }
233 sk->sk_max_ack_backlog = backlog;
234 err = 0;
235
236out:
237 release_sock(sk);
238 return err;
239}
240EXPORT_SYMBOL(inet_listen);
241
242/*
243 * Create an inet socket.
244 */
245
246static int inet_create(struct net *net, struct socket *sock, int protocol,
247 int kern)
248{
249 struct sock *sk;
250 struct inet_protosw *answer;
251 struct inet_sock *inet;
252 struct proto *answer_prot;
253 unsigned char answer_flags;
254 int try_loading_module = 0;
255 int err;
256
257 if (protocol < 0 || protocol >= IPPROTO_MAX)
258 return -EINVAL;
259
260 sock->state = SS_UNCONNECTED;
261
262 /* Look for the requested type/protocol pair. */
263lookup_protocol:
264 err = -ESOCKTNOSUPPORT;
265 rcu_read_lock();
266 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
267
268 err = 0;
269 /* Check the non-wild match. */
270 if (protocol == answer->protocol) {
271 if (protocol != IPPROTO_IP)
272 break;
273 } else {
274 /* Check for the two wild cases. */
275 if (IPPROTO_IP == protocol) {
276 protocol = answer->protocol;
277 break;
278 }
279 if (IPPROTO_IP == answer->protocol)
280 break;
281 }
282 err = -EPROTONOSUPPORT;
283 }
284
285 if (unlikely(err)) {
286 if (try_loading_module < 2) {
287 rcu_read_unlock();
288 /*
289 * Be more specific, e.g. net-pf-2-proto-132-type-1
290 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
291 */
292 if (++try_loading_module == 1)
293 request_module("net-pf-%d-proto-%d-type-%d",
294 PF_INET, protocol, sock->type);
295 /*
296 * Fall back to generic, e.g. net-pf-2-proto-132
297 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
298 */
299 else
300 request_module("net-pf-%d-proto-%d",
301 PF_INET, protocol);
302 goto lookup_protocol;
303 } else
304 goto out_rcu_unlock;
305 }
306
307 err = -EPERM;
308 if (sock->type == SOCK_RAW && !kern &&
309 !ns_capable(net->user_ns, CAP_NET_RAW))
310 goto out_rcu_unlock;
311
312 sock->ops = answer->ops;
313 answer_prot = answer->prot;
314 answer_flags = answer->flags;
315 rcu_read_unlock();
316
317 WARN_ON(!answer_prot->slab);
318
319 err = -ENOBUFS;
320 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern);
321 if (!sk)
322 goto out;
323
324 err = 0;
325 if (INET_PROTOSW_REUSE & answer_flags)
326 sk->sk_reuse = SK_CAN_REUSE;
327
328 inet = inet_sk(sk);
329 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
330
331 inet->nodefrag = 0;
332
333 if (SOCK_RAW == sock->type) {
334 inet->inet_num = protocol;
335 if (IPPROTO_RAW == protocol)
336 inet->hdrincl = 1;
337 }
338
339 if (net->ipv4.sysctl_ip_no_pmtu_disc)
340 inet->pmtudisc = IP_PMTUDISC_DONT;
341 else
342 inet->pmtudisc = IP_PMTUDISC_WANT;
343
344 inet->inet_id = 0;
345
346 sock_init_data(sock, sk);
347
348 sk->sk_destruct = inet_sock_destruct;
349 sk->sk_protocol = protocol;
350 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
351
352 inet->uc_ttl = -1;
353 inet->mc_loop = 1;
354 inet->mc_ttl = 1;
355 inet->mc_all = 1;
356 inet->mc_index = 0;
357 inet->mc_list = NULL;
358 inet->rcv_tos = 0;
359
360 sk_refcnt_debug_inc(sk);
361
362 if (inet->inet_num) {
363 /* It assumes that any protocol which allows
364 * the user to assign a number at socket
365 * creation time automatically
366 * shares.
367 */
368 inet->inet_sport = htons(inet->inet_num);
369 /* Add to protocol hash chains. */
370 err = sk->sk_prot->hash(sk);
371 if (err) {
372 sk_common_release(sk);
373 goto out;
374 }
375 }
376
377 if (sk->sk_prot->init) {
378 err = sk->sk_prot->init(sk);
379 if (err) {
380 sk_common_release(sk);
381 goto out;
382 }
383 }
384
385 if (!kern) {
386 err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk);
387 if (err) {
388 sk_common_release(sk);
389 goto out;
390 }
391 }
392out:
393 return err;
394out_rcu_unlock:
395 rcu_read_unlock();
396 goto out;
397}
398
399
400/*
401 * The peer socket should always be NULL (or else). When we call this
402 * function we are destroying the object and from then on nobody
403 * should refer to it.
404 */
405int inet_release(struct socket *sock)
406{
407 struct sock *sk = sock->sk;
408
409 if (sk) {
410 long timeout;
411
412 /* Applications forget to leave groups before exiting */
413 ip_mc_drop_socket(sk);
414
415 /* If linger is set, we don't return until the close
416 * is complete. Otherwise we return immediately. The
417 * actually closing is done the same either way.
418 *
419 * If the close is due to the process exiting, we never
420 * linger..
421 */
422 timeout = 0;
423 if (sock_flag(sk, SOCK_LINGER) &&
424 !(current->flags & PF_EXITING))
425 timeout = sk->sk_lingertime;
426 sock->sk = NULL;
427 sk->sk_prot->close(sk, timeout);
428 }
429 return 0;
430}
431EXPORT_SYMBOL(inet_release);
432
433int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
434{
435 struct sock *sk = sock->sk;
436 int err;
437
438 /* If the socket has its own bind function then use it. (RAW) */
439 if (sk->sk_prot->bind) {
440 return sk->sk_prot->bind(sk, uaddr, addr_len);
441 }
442 if (addr_len < sizeof(struct sockaddr_in))
443 return -EINVAL;
444
445 /* BPF prog is run before any checks are done so that if the prog
446 * changes context in a wrong way it will be caught.
447 */
448 err = BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr);
449 if (err)
450 return err;
451
452 return __inet_bind(sk, uaddr, addr_len, false, true);
453}
454EXPORT_SYMBOL(inet_bind);
455
456int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len,
457 bool force_bind_address_no_port, bool with_lock)
458{
459 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
460 struct inet_sock *inet = inet_sk(sk);
461 struct net *net = sock_net(sk);
462 unsigned short snum;
463 int chk_addr_ret;
464 u32 tb_id = RT_TABLE_LOCAL;
465 int err;
466
467 if (addr->sin_family != AF_INET) {
468 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
469 * only if s_addr is INADDR_ANY.
470 */
471 err = -EAFNOSUPPORT;
472 if (addr->sin_family != AF_UNSPEC ||
473 addr->sin_addr.s_addr != htonl(INADDR_ANY))
474 goto out;
475 }
476
477 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id;
478 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id);
479
480 /* Not specified by any standard per-se, however it breaks too
481 * many applications when removed. It is unfortunate since
482 * allowing applications to make a non-local bind solves
483 * several problems with systems using dynamic addressing.
484 * (ie. your servers still start up even if your ISDN link
485 * is temporarily down)
486 */
487 err = -EADDRNOTAVAIL;
488 if (!net->ipv4.sysctl_ip_nonlocal_bind &&
489 !(inet->freebind || inet->transparent) &&
490 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
491 chk_addr_ret != RTN_LOCAL &&
492 chk_addr_ret != RTN_MULTICAST &&
493 chk_addr_ret != RTN_BROADCAST)
494 goto out;
495
496 snum = ntohs(addr->sin_port);
497 err = -EACCES;
498 if (snum && snum < inet_prot_sock(net) &&
499 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE))
500 goto out;
501
502 /* We keep a pair of addresses. rcv_saddr is the one
503 * used by hash lookups, and saddr is used for transmit.
504 *
505 * In the BSD API these are the same except where it
506 * would be illegal to use them (multicast/broadcast) in
507 * which case the sending device address is used.
508 */
509 if (with_lock)
510 lock_sock(sk);
511
512 /* Check these errors (active socket, double bind). */
513 err = -EINVAL;
514 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
515 goto out_release_sock;
516
517 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
518 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
519 inet->inet_saddr = 0; /* Use device */
520
521 /* Make sure we are allowed to bind here. */
522 if (snum || !(inet->bind_address_no_port ||
523 force_bind_address_no_port)) {
524 if (sk->sk_prot->get_port(sk, snum)) {
525 inet->inet_saddr = inet->inet_rcv_saddr = 0;
526 err = -EADDRINUSE;
527 goto out_release_sock;
528 }
529 err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk);
530 if (err) {
531 inet->inet_saddr = inet->inet_rcv_saddr = 0;
532 goto out_release_sock;
533 }
534 }
535
536 if (inet->inet_rcv_saddr)
537 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
538 if (snum)
539 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
540 inet->inet_sport = htons(inet->inet_num);
541 inet->inet_daddr = 0;
542 inet->inet_dport = 0;
543 sk_dst_reset(sk);
544 err = 0;
545out_release_sock:
546 if (with_lock)
547 release_sock(sk);
548out:
549 return err;
550}
551
552int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr,
553 int addr_len, int flags)
554{
555 struct sock *sk = sock->sk;
556 int err;
557
558 if (addr_len < sizeof(uaddr->sa_family))
559 return -EINVAL;
560 if (uaddr->sa_family == AF_UNSPEC)
561 return sk->sk_prot->disconnect(sk, flags);
562
563 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
564 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
565 if (err)
566 return err;
567 }
568
569 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
570 return -EAGAIN;
571 return sk->sk_prot->connect(sk, uaddr, addr_len);
572}
573EXPORT_SYMBOL(inet_dgram_connect);
574
575static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias)
576{
577 DEFINE_WAIT_FUNC(wait, woken_wake_function);
578
579 add_wait_queue(sk_sleep(sk), &wait);
580 sk->sk_write_pending += writebias;
581
582 /* Basic assumption: if someone sets sk->sk_err, he _must_
583 * change state of the socket from TCP_SYN_*.
584 * Connect() does not allow to get error notifications
585 * without closing the socket.
586 */
587 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
588 release_sock(sk);
589 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
590 lock_sock(sk);
591 if (signal_pending(current) || !timeo)
592 break;
593 }
594 remove_wait_queue(sk_sleep(sk), &wait);
595 sk->sk_write_pending -= writebias;
596 return timeo;
597}
598
599/*
600 * Connect to a remote host. There is regrettably still a little
601 * TCP 'magic' in here.
602 */
603int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
604 int addr_len, int flags, int is_sendmsg)
605{
606 struct sock *sk = sock->sk;
607 int err;
608 long timeo;
609
610 /*
611 * uaddr can be NULL and addr_len can be 0 if:
612 * sk is a TCP fastopen active socket and
613 * TCP_FASTOPEN_CONNECT sockopt is set and
614 * we already have a valid cookie for this socket.
615 * In this case, user can call write() after connect().
616 * write() will invoke tcp_sendmsg_fastopen() which calls
617 * __inet_stream_connect().
618 */
619 if (uaddr) {
620 if (addr_len < sizeof(uaddr->sa_family))
621 return -EINVAL;
622
623 if (uaddr->sa_family == AF_UNSPEC) {
624 err = sk->sk_prot->disconnect(sk, flags);
625 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
626 goto out;
627 }
628 }
629
630 switch (sock->state) {
631 default:
632 err = -EINVAL;
633 goto out;
634 case SS_CONNECTED:
635 err = -EISCONN;
636 goto out;
637 case SS_CONNECTING:
638 if (inet_sk(sk)->defer_connect)
639 err = is_sendmsg ? -EINPROGRESS : -EISCONN;
640 else
641 err = -EALREADY;
642 /* Fall out of switch with err, set for this state */
643 break;
644 case SS_UNCONNECTED:
645 err = -EISCONN;
646 if (sk->sk_state != TCP_CLOSE)
647 goto out;
648
649 if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) {
650 err = sk->sk_prot->pre_connect(sk, uaddr, addr_len);
651 if (err)
652 goto out;
653 }
654
655 err = sk->sk_prot->connect(sk, uaddr, addr_len);
656 if (err < 0)
657 goto out;
658
659 sock->state = SS_CONNECTING;
660
661 if (!err && inet_sk(sk)->defer_connect)
662 goto out;
663
664 /* Just entered SS_CONNECTING state; the only
665 * difference is that return value in non-blocking
666 * case is EINPROGRESS, rather than EALREADY.
667 */
668 err = -EINPROGRESS;
669 break;
670 }
671
672 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
673
674 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
675 int writebias = (sk->sk_protocol == IPPROTO_TCP) &&
676 tcp_sk(sk)->fastopen_req &&
677 tcp_sk(sk)->fastopen_req->data ? 1 : 0;
678
679 /* Error code is set above */
680 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias))
681 goto out;
682
683 err = sock_intr_errno(timeo);
684 if (signal_pending(current))
685 goto out;
686 }
687
688 /* Connection was closed by RST, timeout, ICMP error
689 * or another process disconnected us.
690 */
691 if (sk->sk_state == TCP_CLOSE)
692 goto sock_error;
693
694 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
695 * and error was received after socket entered established state.
696 * Hence, it is handled normally after connect() return successfully.
697 */
698
699 sock->state = SS_CONNECTED;
700 err = 0;
701out:
702 return err;
703
704sock_error:
705 err = sock_error(sk) ? : -ECONNABORTED;
706 sock->state = SS_UNCONNECTED;
707 if (sk->sk_prot->disconnect(sk, flags))
708 sock->state = SS_DISCONNECTING;
709 goto out;
710}
711EXPORT_SYMBOL(__inet_stream_connect);
712
713int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
714 int addr_len, int flags)
715{
716 int err;
717
718 lock_sock(sock->sk);
719 err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0);
720 release_sock(sock->sk);
721 return err;
722}
723EXPORT_SYMBOL(inet_stream_connect);
724
725/*
726 * Accept a pending connection. The TCP layer now gives BSD semantics.
727 */
728
729int inet_accept(struct socket *sock, struct socket *newsock, int flags,
730 bool kern)
731{
732 struct sock *sk1 = sock->sk;
733 int err = -EINVAL;
734 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err, kern);
735
736 if (!sk2)
737 goto do_err;
738
739 lock_sock(sk2);
740
741 sock_rps_record_flow(sk2);
742 WARN_ON(!((1 << sk2->sk_state) &
743 (TCPF_ESTABLISHED | TCPF_SYN_RECV |
744 TCPF_CLOSE_WAIT | TCPF_CLOSE)));
745
746 sock_graft(sk2, newsock);
747
748 newsock->state = SS_CONNECTED;
749 err = 0;
750 release_sock(sk2);
751do_err:
752 return err;
753}
754EXPORT_SYMBOL(inet_accept);
755
756
757/*
758 * This does both peername and sockname.
759 */
760int inet_getname(struct socket *sock, struct sockaddr *uaddr,
761 int peer)
762{
763 struct sock *sk = sock->sk;
764 struct inet_sock *inet = inet_sk(sk);
765 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
766
767 sin->sin_family = AF_INET;
768 if (peer) {
769 if (!inet->inet_dport ||
770 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
771 peer == 1))
772 return -ENOTCONN;
773 sin->sin_port = inet->inet_dport;
774 sin->sin_addr.s_addr = inet->inet_daddr;
775 } else {
776 __be32 addr = inet->inet_rcv_saddr;
777 if (!addr)
778 addr = inet->inet_saddr;
779 sin->sin_port = inet->inet_sport;
780 sin->sin_addr.s_addr = addr;
781 }
782 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
783 return sizeof(*sin);
784}
785EXPORT_SYMBOL(inet_getname);
786
787int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size)
788{
789 struct sock *sk = sock->sk;
790
791 sock_rps_record_flow(sk);
792
793 /* We may need to bind the socket. */
794 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
795 inet_autobind(sk))
796 return -EAGAIN;
797
798 return sk->sk_prot->sendmsg(sk, msg, size);
799}
800EXPORT_SYMBOL(inet_sendmsg);
801
802ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
803 size_t size, int flags)
804{
805 struct sock *sk = sock->sk;
806
807 sock_rps_record_flow(sk);
808
809 /* We may need to bind the socket. */
810 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
811 inet_autobind(sk))
812 return -EAGAIN;
813
814 if (sk->sk_prot->sendpage)
815 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
816 return sock_no_sendpage(sock, page, offset, size, flags);
817}
818EXPORT_SYMBOL(inet_sendpage);
819
820int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
821 int flags)
822{
823 struct sock *sk = sock->sk;
824 int addr_len = 0;
825 int err;
826
827 if (likely(!(flags & MSG_ERRQUEUE)))
828 sock_rps_record_flow(sk);
829
830 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
831 flags & ~MSG_DONTWAIT, &addr_len);
832 if (err >= 0)
833 msg->msg_namelen = addr_len;
834 return err;
835}
836EXPORT_SYMBOL(inet_recvmsg);
837
838int inet_shutdown(struct socket *sock, int how)
839{
840 struct sock *sk = sock->sk;
841 int err = 0;
842
843 /* This should really check to make sure
844 * the socket is a TCP socket. (WHY AC...)
845 */
846 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
847 1->2 bit 2 snds.
848 2->3 */
849 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
850 return -EINVAL;
851
852 lock_sock(sk);
853 if (sock->state == SS_CONNECTING) {
854 if ((1 << sk->sk_state) &
855 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
856 sock->state = SS_DISCONNECTING;
857 else
858 sock->state = SS_CONNECTED;
859 }
860
861 switch (sk->sk_state) {
862 case TCP_CLOSE:
863 err = -ENOTCONN;
864 /* Hack to wake up other listeners, who can poll for
865 EPOLLHUP, even on eg. unconnected UDP sockets -- RR */
866 /* fall through */
867 default:
868 sk->sk_shutdown |= how;
869 if (sk->sk_prot->shutdown)
870 sk->sk_prot->shutdown(sk, how);
871 break;
872
873 /* Remaining two branches are temporary solution for missing
874 * close() in multithreaded environment. It is _not_ a good idea,
875 * but we have no choice until close() is repaired at VFS level.
876 */
877 case TCP_LISTEN:
878 if (!(how & RCV_SHUTDOWN))
879 break;
880 /* fall through */
881 case TCP_SYN_SENT:
882 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
883 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
884 break;
885 }
886
887 /* Wake up anyone sleeping in poll. */
888 sk->sk_state_change(sk);
889 release_sock(sk);
890 return err;
891}
892EXPORT_SYMBOL(inet_shutdown);
893
894/*
895 * ioctl() calls you can issue on an INET socket. Most of these are
896 * device configuration and stuff and very rarely used. Some ioctls
897 * pass on to the socket itself.
898 *
899 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
900 * loads the devconfigure module does its configuring and unloads it.
901 * There's a good 20K of config code hanging around the kernel.
902 */
903
904int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
905{
906 struct sock *sk = sock->sk;
907 int err = 0;
908 struct net *net = sock_net(sk);
909 void __user *p = (void __user *)arg;
910 struct ifreq ifr;
911 struct rtentry rt;
912
913 switch (cmd) {
914 case SIOCGSTAMP:
915 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
916 break;
917 case SIOCGSTAMPNS:
918 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
919 break;
920 case SIOCADDRT:
921 case SIOCDELRT:
922 if (copy_from_user(&rt, p, sizeof(struct rtentry)))
923 return -EFAULT;
924 err = ip_rt_ioctl(net, cmd, &rt);
925 break;
926 case SIOCRTMSG:
927 err = -EINVAL;
928 break;
929 case SIOCDARP:
930 case SIOCGARP:
931 case SIOCSARP:
932 err = arp_ioctl(net, cmd, (void __user *)arg);
933 break;
934 case SIOCGIFADDR:
935 case SIOCGIFBRDADDR:
936 case SIOCGIFNETMASK:
937 case SIOCGIFDSTADDR:
938 case SIOCGIFPFLAGS:
939 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
940 return -EFAULT;
941 err = devinet_ioctl(net, cmd, &ifr);
942 if (!err && copy_to_user(p, &ifr, sizeof(struct ifreq)))
943 err = -EFAULT;
944 break;
945
946 case SIOCSIFADDR:
947 case SIOCSIFBRDADDR:
948 case SIOCSIFNETMASK:
949 case SIOCSIFDSTADDR:
950 case SIOCSIFPFLAGS:
951 case SIOCSIFFLAGS:
952 if (copy_from_user(&ifr, p, sizeof(struct ifreq)))
953 return -EFAULT;
954 err = devinet_ioctl(net, cmd, &ifr);
955 break;
956 default:
957 if (sk->sk_prot->ioctl)
958 err = sk->sk_prot->ioctl(sk, cmd, arg);
959 else
960 err = -ENOIOCTLCMD;
961 break;
962 }
963 return err;
964}
965EXPORT_SYMBOL(inet_ioctl);
966
967#ifdef CONFIG_COMPAT
968static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
969{
970 struct sock *sk = sock->sk;
971 int err = -ENOIOCTLCMD;
972
973 if (sk->sk_prot->compat_ioctl)
974 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
975
976 return err;
977}
978#endif
979
980const struct proto_ops inet_stream_ops = {
981 .family = PF_INET,
982 .owner = THIS_MODULE,
983 .release = inet_release,
984 .bind = inet_bind,
985 .connect = inet_stream_connect,
986 .socketpair = sock_no_socketpair,
987 .accept = inet_accept,
988 .getname = inet_getname,
989 .poll = tcp_poll,
990 .ioctl = inet_ioctl,
991 .listen = inet_listen,
992 .shutdown = inet_shutdown,
993 .setsockopt = sock_common_setsockopt,
994 .getsockopt = sock_common_getsockopt,
995 .sendmsg = inet_sendmsg,
996 .recvmsg = inet_recvmsg,
997 .mmap = sock_no_mmap,
998 .sendpage = inet_sendpage,
999 .splice_read = tcp_splice_read,
1000 .read_sock = tcp_read_sock,
1001 .sendmsg_locked = tcp_sendmsg_locked,
1002 .sendpage_locked = tcp_sendpage_locked,
1003 .peek_len = tcp_peek_len,
1004#ifdef CONFIG_COMPAT
1005 .compat_setsockopt = compat_sock_common_setsockopt,
1006 .compat_getsockopt = compat_sock_common_getsockopt,
1007 .compat_ioctl = inet_compat_ioctl,
1008#endif
1009};
1010EXPORT_SYMBOL(inet_stream_ops);
1011
1012const struct proto_ops inet_dgram_ops = {
1013 .family = PF_INET,
1014 .owner = THIS_MODULE,
1015 .release = inet_release,
1016 .bind = inet_bind,
1017 .connect = inet_dgram_connect,
1018 .socketpair = sock_no_socketpair,
1019 .accept = sock_no_accept,
1020 .getname = inet_getname,
1021 .poll = udp_poll,
1022 .ioctl = inet_ioctl,
1023 .listen = sock_no_listen,
1024 .shutdown = inet_shutdown,
1025 .setsockopt = sock_common_setsockopt,
1026 .getsockopt = sock_common_getsockopt,
1027 .sendmsg = inet_sendmsg,
1028 .recvmsg = inet_recvmsg,
1029 .mmap = sock_no_mmap,
1030 .sendpage = inet_sendpage,
1031 .set_peek_off = sk_set_peek_off,
1032#ifdef CONFIG_COMPAT
1033 .compat_setsockopt = compat_sock_common_setsockopt,
1034 .compat_getsockopt = compat_sock_common_getsockopt,
1035 .compat_ioctl = inet_compat_ioctl,
1036#endif
1037};
1038EXPORT_SYMBOL(inet_dgram_ops);
1039
1040/*
1041 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
1042 * udp_poll
1043 */
1044static const struct proto_ops inet_sockraw_ops = {
1045 .family = PF_INET,
1046 .owner = THIS_MODULE,
1047 .release = inet_release,
1048 .bind = inet_bind,
1049 .connect = inet_dgram_connect,
1050 .socketpair = sock_no_socketpair,
1051 .accept = sock_no_accept,
1052 .getname = inet_getname,
1053 .poll = datagram_poll,
1054 .ioctl = inet_ioctl,
1055 .listen = sock_no_listen,
1056 .shutdown = inet_shutdown,
1057 .setsockopt = sock_common_setsockopt,
1058 .getsockopt = sock_common_getsockopt,
1059 .sendmsg = inet_sendmsg,
1060 .recvmsg = inet_recvmsg,
1061 .mmap = sock_no_mmap,
1062 .sendpage = inet_sendpage,
1063#ifdef CONFIG_COMPAT
1064 .compat_setsockopt = compat_sock_common_setsockopt,
1065 .compat_getsockopt = compat_sock_common_getsockopt,
1066 .compat_ioctl = inet_compat_ioctl,
1067#endif
1068};
1069
1070static const struct net_proto_family inet_family_ops = {
1071 .family = PF_INET,
1072 .create = inet_create,
1073 .owner = THIS_MODULE,
1074};
1075
1076/* Upon startup we insert all the elements in inetsw_array[] into
1077 * the linked list inetsw.
1078 */
1079static struct inet_protosw inetsw_array[] =
1080{
1081 {
1082 .type = SOCK_STREAM,
1083 .protocol = IPPROTO_TCP,
1084 .prot = &tcp_prot,
1085 .ops = &inet_stream_ops,
1086 .flags = INET_PROTOSW_PERMANENT |
1087 INET_PROTOSW_ICSK,
1088 },
1089
1090 {
1091 .type = SOCK_DGRAM,
1092 .protocol = IPPROTO_UDP,
1093 .prot = &udp_prot,
1094 .ops = &inet_dgram_ops,
1095 .flags = INET_PROTOSW_PERMANENT,
1096 },
1097
1098 {
1099 .type = SOCK_DGRAM,
1100 .protocol = IPPROTO_ICMP,
1101 .prot = &ping_prot,
1102 .ops = &inet_sockraw_ops,
1103 .flags = INET_PROTOSW_REUSE,
1104 },
1105
1106 {
1107 .type = SOCK_RAW,
1108 .protocol = IPPROTO_IP, /* wild card */
1109 .prot = &raw_prot,
1110 .ops = &inet_sockraw_ops,
1111 .flags = INET_PROTOSW_REUSE,
1112 }
1113};
1114
1115#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1116
1117void inet_register_protosw(struct inet_protosw *p)
1118{
1119 struct list_head *lh;
1120 struct inet_protosw *answer;
1121 int protocol = p->protocol;
1122 struct list_head *last_perm;
1123
1124 spin_lock_bh(&inetsw_lock);
1125
1126 if (p->type >= SOCK_MAX)
1127 goto out_illegal;
1128
1129 /* If we are trying to override a permanent protocol, bail. */
1130 last_perm = &inetsw[p->type];
1131 list_for_each(lh, &inetsw[p->type]) {
1132 answer = list_entry(lh, struct inet_protosw, list);
1133 /* Check only the non-wild match. */
1134 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0)
1135 break;
1136 if (protocol == answer->protocol)
1137 goto out_permanent;
1138 last_perm = lh;
1139 }
1140
1141 /* Add the new entry after the last permanent entry if any, so that
1142 * the new entry does not override a permanent entry when matched with
1143 * a wild-card protocol. But it is allowed to override any existing
1144 * non-permanent entry. This means that when we remove this entry, the
1145 * system automatically returns to the old behavior.
1146 */
1147 list_add_rcu(&p->list, last_perm);
1148out:
1149 spin_unlock_bh(&inetsw_lock);
1150
1151 return;
1152
1153out_permanent:
1154 pr_err("Attempt to override permanent protocol %d\n", protocol);
1155 goto out;
1156
1157out_illegal:
1158 pr_err("Ignoring attempt to register invalid socket type %d\n",
1159 p->type);
1160 goto out;
1161}
1162EXPORT_SYMBOL(inet_register_protosw);
1163
1164void inet_unregister_protosw(struct inet_protosw *p)
1165{
1166 if (INET_PROTOSW_PERMANENT & p->flags) {
1167 pr_err("Attempt to unregister permanent protocol %d\n",
1168 p->protocol);
1169 } else {
1170 spin_lock_bh(&inetsw_lock);
1171 list_del_rcu(&p->list);
1172 spin_unlock_bh(&inetsw_lock);
1173
1174 synchronize_net();
1175 }
1176}
1177EXPORT_SYMBOL(inet_unregister_protosw);
1178
1179static int inet_sk_reselect_saddr(struct sock *sk)
1180{
1181 struct inet_sock *inet = inet_sk(sk);
1182 __be32 old_saddr = inet->inet_saddr;
1183 __be32 daddr = inet->inet_daddr;
1184 struct flowi4 *fl4;
1185 struct rtable *rt;
1186 __be32 new_saddr;
1187 struct ip_options_rcu *inet_opt;
1188
1189 inet_opt = rcu_dereference_protected(inet->inet_opt,
1190 lockdep_sock_is_held(sk));
1191 if (inet_opt && inet_opt->opt.srr)
1192 daddr = inet_opt->opt.faddr;
1193
1194 /* Query new route. */
1195 fl4 = &inet->cork.fl.u.ip4;
1196 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1197 sk->sk_bound_dev_if, sk->sk_protocol,
1198 inet->inet_sport, inet->inet_dport, sk);
1199 if (IS_ERR(rt))
1200 return PTR_ERR(rt);
1201
1202 sk_setup_caps(sk, &rt->dst);
1203
1204 new_saddr = fl4->saddr;
1205
1206 if (new_saddr == old_saddr)
1207 return 0;
1208
1209 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) {
1210 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n",
1211 __func__, &old_saddr, &new_saddr);
1212 }
1213
1214 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1215
1216 /*
1217 * XXX The only one ugly spot where we need to
1218 * XXX really change the sockets identity after
1219 * XXX it has entered the hashes. -DaveM
1220 *
1221 * Besides that, it does not check for connection
1222 * uniqueness. Wait for troubles.
1223 */
1224 return __sk_prot_rehash(sk);
1225}
1226
1227int inet_sk_rebuild_header(struct sock *sk)
1228{
1229 struct inet_sock *inet = inet_sk(sk);
1230 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1231 __be32 daddr;
1232 struct ip_options_rcu *inet_opt;
1233 struct flowi4 *fl4;
1234 int err;
1235
1236 /* Route is OK, nothing to do. */
1237 if (rt)
1238 return 0;
1239
1240 /* Reroute. */
1241 rcu_read_lock();
1242 inet_opt = rcu_dereference(inet->inet_opt);
1243 daddr = inet->inet_daddr;
1244 if (inet_opt && inet_opt->opt.srr)
1245 daddr = inet_opt->opt.faddr;
1246 rcu_read_unlock();
1247 fl4 = &inet->cork.fl.u.ip4;
1248 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1249 inet->inet_dport, inet->inet_sport,
1250 sk->sk_protocol, RT_CONN_FLAGS(sk),
1251 sk->sk_bound_dev_if);
1252 if (!IS_ERR(rt)) {
1253 err = 0;
1254 sk_setup_caps(sk, &rt->dst);
1255 } else {
1256 err = PTR_ERR(rt);
1257
1258 /* Routing failed... */
1259 sk->sk_route_caps = 0;
1260 /*
1261 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1262 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1263 */
1264 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr ||
1265 sk->sk_state != TCP_SYN_SENT ||
1266 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1267 (err = inet_sk_reselect_saddr(sk)) != 0)
1268 sk->sk_err_soft = -err;
1269 }
1270
1271 return err;
1272}
1273EXPORT_SYMBOL(inet_sk_rebuild_header);
1274
1275void inet_sk_set_state(struct sock *sk, int state)
1276{
1277 trace_inet_sock_set_state(sk, sk->sk_state, state);
1278 sk->sk_state = state;
1279}
1280EXPORT_SYMBOL(inet_sk_set_state);
1281
1282void inet_sk_state_store(struct sock *sk, int newstate)
1283{
1284 trace_inet_sock_set_state(sk, sk->sk_state, newstate);
1285 smp_store_release(&sk->sk_state, newstate);
1286}
1287
1288struct sk_buff *inet_gso_segment(struct sk_buff *skb,
1289 netdev_features_t features)
1290{
1291 bool udpfrag = false, fixedid = false, gso_partial, encap;
1292 struct sk_buff *segs = ERR_PTR(-EINVAL);
1293 const struct net_offload *ops;
1294 unsigned int offset = 0;
1295 struct iphdr *iph;
1296 int proto, tot_len;
1297 int nhoff;
1298 int ihl;
1299 int id;
1300
1301 skb_reset_network_header(skb);
1302 nhoff = skb_network_header(skb) - skb_mac_header(skb);
1303 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1304 goto out;
1305
1306 iph = ip_hdr(skb);
1307 ihl = iph->ihl * 4;
1308 if (ihl < sizeof(*iph))
1309 goto out;
1310
1311 id = ntohs(iph->id);
1312 proto = iph->protocol;
1313
1314 /* Warning: after this point, iph might be no longer valid */
1315 if (unlikely(!pskb_may_pull(skb, ihl)))
1316 goto out;
1317 __skb_pull(skb, ihl);
1318
1319 encap = SKB_GSO_CB(skb)->encap_level > 0;
1320 if (encap)
1321 features &= skb->dev->hw_enc_features;
1322 SKB_GSO_CB(skb)->encap_level += ihl;
1323
1324 skb_reset_transport_header(skb);
1325
1326 segs = ERR_PTR(-EPROTONOSUPPORT);
1327
1328 if (!skb->encapsulation || encap) {
1329 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
1330 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID);
1331
1332 /* fixed ID is invalid if DF bit is not set */
1333 if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF)))
1334 goto out;
1335 }
1336
1337 ops = rcu_dereference(inet_offloads[proto]);
1338 if (likely(ops && ops->callbacks.gso_segment))
1339 segs = ops->callbacks.gso_segment(skb, features);
1340
1341 if (IS_ERR_OR_NULL(segs))
1342 goto out;
1343
1344 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
1345
1346 skb = segs;
1347 do {
1348 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff);
1349 if (udpfrag) {
1350 iph->frag_off = htons(offset >> 3);
1351 if (skb->next)
1352 iph->frag_off |= htons(IP_MF);
1353 offset += skb->len - nhoff - ihl;
1354 tot_len = skb->len - nhoff;
1355 } else if (skb_is_gso(skb)) {
1356 if (!fixedid) {
1357 iph->id = htons(id);
1358 id += skb_shinfo(skb)->gso_segs;
1359 }
1360
1361 if (gso_partial)
1362 tot_len = skb_shinfo(skb)->gso_size +
1363 SKB_GSO_CB(skb)->data_offset +
1364 skb->head - (unsigned char *)iph;
1365 else
1366 tot_len = skb->len - nhoff;
1367 } else {
1368 if (!fixedid)
1369 iph->id = htons(id++);
1370 tot_len = skb->len - nhoff;
1371 }
1372 iph->tot_len = htons(tot_len);
1373 ip_send_check(iph);
1374 if (encap)
1375 skb_reset_inner_headers(skb);
1376 skb->network_header = (u8 *)iph - skb->head;
1377 } while ((skb = skb->next));
1378
1379out:
1380 return segs;
1381}
1382EXPORT_SYMBOL(inet_gso_segment);
1383
1384struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
1385{
1386 const struct net_offload *ops;
1387 struct sk_buff **pp = NULL;
1388 struct sk_buff *p;
1389 const struct iphdr *iph;
1390 unsigned int hlen;
1391 unsigned int off;
1392 unsigned int id;
1393 int flush = 1;
1394 int proto;
1395
1396 off = skb_gro_offset(skb);
1397 hlen = off + sizeof(*iph);
1398 iph = skb_gro_header_fast(skb, off);
1399 if (skb_gro_header_hard(skb, hlen)) {
1400 iph = skb_gro_header_slow(skb, hlen, off);
1401 if (unlikely(!iph))
1402 goto out;
1403 }
1404
1405 proto = iph->protocol;
1406
1407 rcu_read_lock();
1408 ops = rcu_dereference(inet_offloads[proto]);
1409 if (!ops || !ops->callbacks.gro_receive)
1410 goto out_unlock;
1411
1412 if (*(u8 *)iph != 0x45)
1413 goto out_unlock;
1414
1415 if (ip_is_fragment(iph))
1416 goto out_unlock;
1417
1418 if (unlikely(ip_fast_csum((u8 *)iph, 5)))
1419 goto out_unlock;
1420
1421 id = ntohl(*(__be32 *)&iph->id);
1422 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF));
1423 id >>= 16;
1424
1425 for (p = *head; p; p = p->next) {
1426 struct iphdr *iph2;
1427 u16 flush_id;
1428
1429 if (!NAPI_GRO_CB(p)->same_flow)
1430 continue;
1431
1432 iph2 = (struct iphdr *)(p->data + off);
1433 /* The above works because, with the exception of the top
1434 * (inner most) layer, we only aggregate pkts with the same
1435 * hdr length so all the hdrs we'll need to verify will start
1436 * at the same offset.
1437 */
1438 if ((iph->protocol ^ iph2->protocol) |
1439 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1440 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1441 NAPI_GRO_CB(p)->same_flow = 0;
1442 continue;
1443 }
1444
1445 /* All fields must match except length and checksum. */
1446 NAPI_GRO_CB(p)->flush |=
1447 (iph->ttl ^ iph2->ttl) |
1448 (iph->tos ^ iph2->tos) |
1449 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF));
1450
1451 NAPI_GRO_CB(p)->flush |= flush;
1452
1453 /* We need to store of the IP ID check to be included later
1454 * when we can verify that this packet does in fact belong
1455 * to a given flow.
1456 */
1457 flush_id = (u16)(id - ntohs(iph2->id));
1458
1459 /* This bit of code makes it much easier for us to identify
1460 * the cases where we are doing atomic vs non-atomic IP ID
1461 * checks. Specifically an atomic check can return IP ID
1462 * values 0 - 0xFFFF, while a non-atomic check can only
1463 * return 0 or 0xFFFF.
1464 */
1465 if (!NAPI_GRO_CB(p)->is_atomic ||
1466 !(iph->frag_off & htons(IP_DF))) {
1467 flush_id ^= NAPI_GRO_CB(p)->count;
1468 flush_id = flush_id ? 0xFFFF : 0;
1469 }
1470
1471 /* If the previous IP ID value was based on an atomic
1472 * datagram we can overwrite the value and ignore it.
1473 */
1474 if (NAPI_GRO_CB(skb)->is_atomic)
1475 NAPI_GRO_CB(p)->flush_id = flush_id;
1476 else
1477 NAPI_GRO_CB(p)->flush_id |= flush_id;
1478 }
1479
1480 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF));
1481 NAPI_GRO_CB(skb)->flush |= flush;
1482 skb_set_network_header(skb, off);
1483 /* The above will be needed by the transport layer if there is one
1484 * immediately following this IP hdr.
1485 */
1486
1487 /* Note : No need to call skb_gro_postpull_rcsum() here,
1488 * as we already checked checksum over ipv4 header was 0
1489 */
1490 skb_gro_pull(skb, sizeof(*iph));
1491 skb_set_transport_header(skb, skb_gro_offset(skb));
1492
1493 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb);
1494
1495out_unlock:
1496 rcu_read_unlock();
1497
1498out:
1499 skb_gro_flush_final(skb, pp, flush);
1500
1501 return pp;
1502}
1503EXPORT_SYMBOL(inet_gro_receive);
1504
1505static struct sk_buff **ipip_gro_receive(struct sk_buff **head,
1506 struct sk_buff *skb)
1507{
1508 if (NAPI_GRO_CB(skb)->encap_mark) {
1509 NAPI_GRO_CB(skb)->flush = 1;
1510 return NULL;
1511 }
1512
1513 NAPI_GRO_CB(skb)->encap_mark = 1;
1514
1515 return inet_gro_receive(head, skb);
1516}
1517
1518#define SECONDS_PER_DAY 86400
1519
1520/* inet_current_timestamp - Return IP network timestamp
1521 *
1522 * Return milliseconds since midnight in network byte order.
1523 */
1524__be32 inet_current_timestamp(void)
1525{
1526 u32 secs;
1527 u32 msecs;
1528 struct timespec64 ts;
1529
1530 ktime_get_real_ts64(&ts);
1531
1532 /* Get secs since midnight. */
1533 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs);
1534 /* Convert to msecs. */
1535 msecs = secs * MSEC_PER_SEC;
1536 /* Convert nsec to msec. */
1537 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC;
1538
1539 /* Convert to network byte order. */
1540 return htonl(msecs);
1541}
1542EXPORT_SYMBOL(inet_current_timestamp);
1543
1544int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len)
1545{
1546 if (sk->sk_family == AF_INET)
1547 return ip_recv_error(sk, msg, len, addr_len);
1548#if IS_ENABLED(CONFIG_IPV6)
1549 if (sk->sk_family == AF_INET6)
1550 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len);
1551#endif
1552 return -EINVAL;
1553}
1554
1555int inet_gro_complete(struct sk_buff *skb, int nhoff)
1556{
1557 __be16 newlen = htons(skb->len - nhoff);
1558 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
1559 const struct net_offload *ops;
1560 int proto = iph->protocol;
1561 int err = -ENOSYS;
1562
1563 if (skb->encapsulation) {
1564 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP));
1565 skb_set_inner_network_header(skb, nhoff);
1566 }
1567
1568 csum_replace2(&iph->check, iph->tot_len, newlen);
1569 iph->tot_len = newlen;
1570
1571 rcu_read_lock();
1572 ops = rcu_dereference(inet_offloads[proto]);
1573 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
1574 goto out_unlock;
1575
1576 /* Only need to add sizeof(*iph) to get to the next hdr below
1577 * because any hdr with option will have been flushed in
1578 * inet_gro_receive().
1579 */
1580 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
1581
1582out_unlock:
1583 rcu_read_unlock();
1584
1585 return err;
1586}
1587EXPORT_SYMBOL(inet_gro_complete);
1588
1589static int ipip_gro_complete(struct sk_buff *skb, int nhoff)
1590{
1591 skb->encapsulation = 1;
1592 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
1593 return inet_gro_complete(skb, nhoff);
1594}
1595
1596int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1597 unsigned short type, unsigned char protocol,
1598 struct net *net)
1599{
1600 struct socket *sock;
1601 int rc = sock_create_kern(net, family, type, protocol, &sock);
1602
1603 if (rc == 0) {
1604 *sk = sock->sk;
1605 (*sk)->sk_allocation = GFP_ATOMIC;
1606 /*
1607 * Unhash it so that IP input processing does not even see it,
1608 * we do not wish this socket to see incoming packets.
1609 */
1610 (*sk)->sk_prot->unhash(*sk);
1611 }
1612 return rc;
1613}
1614EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1615
1616u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt)
1617{
1618 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt);
1619}
1620EXPORT_SYMBOL_GPL(snmp_get_cpu_field);
1621
1622unsigned long snmp_fold_field(void __percpu *mib, int offt)
1623{
1624 unsigned long res = 0;
1625 int i;
1626
1627 for_each_possible_cpu(i)
1628 res += snmp_get_cpu_field(mib, i, offt);
1629 return res;
1630}
1631EXPORT_SYMBOL_GPL(snmp_fold_field);
1632
1633#if BITS_PER_LONG==32
1634
1635u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt,
1636 size_t syncp_offset)
1637{
1638 void *bhptr;
1639 struct u64_stats_sync *syncp;
1640 u64 v;
1641 unsigned int start;
1642
1643 bhptr = per_cpu_ptr(mib, cpu);
1644 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1645 do {
1646 start = u64_stats_fetch_begin_irq(syncp);
1647 v = *(((u64 *)bhptr) + offt);
1648 } while (u64_stats_fetch_retry_irq(syncp, start));
1649
1650 return v;
1651}
1652EXPORT_SYMBOL_GPL(snmp_get_cpu_field64);
1653
1654u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset)
1655{
1656 u64 res = 0;
1657 int cpu;
1658
1659 for_each_possible_cpu(cpu) {
1660 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset);
1661 }
1662 return res;
1663}
1664EXPORT_SYMBOL_GPL(snmp_fold_field64);
1665#endif
1666
1667#ifdef CONFIG_IP_MULTICAST
1668static const struct net_protocol igmp_protocol = {
1669 .handler = igmp_rcv,
1670 .netns_ok = 1,
1671};
1672#endif
1673
1674/* thinking of making this const? Don't.
1675 * early_demux can change based on sysctl.
1676 */
1677static struct net_protocol tcp_protocol = {
1678 .early_demux = tcp_v4_early_demux,
1679 .early_demux_handler = tcp_v4_early_demux,
1680 .handler = tcp_v4_rcv,
1681 .err_handler = tcp_v4_err,
1682 .no_policy = 1,
1683 .netns_ok = 1,
1684 .icmp_strict_tag_validation = 1,
1685};
1686
1687/* thinking of making this const? Don't.
1688 * early_demux can change based on sysctl.
1689 */
1690static struct net_protocol udp_protocol = {
1691 .early_demux = udp_v4_early_demux,
1692 .early_demux_handler = udp_v4_early_demux,
1693 .handler = udp_rcv,
1694 .err_handler = udp_err,
1695 .no_policy = 1,
1696 .netns_ok = 1,
1697};
1698
1699static const struct net_protocol icmp_protocol = {
1700 .handler = icmp_rcv,
1701 .err_handler = icmp_err,
1702 .no_policy = 1,
1703 .netns_ok = 1,
1704};
1705
1706static __net_init int ipv4_mib_init_net(struct net *net)
1707{
1708 int i;
1709
1710 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib);
1711 if (!net->mib.tcp_statistics)
1712 goto err_tcp_mib;
1713 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib);
1714 if (!net->mib.ip_statistics)
1715 goto err_ip_mib;
1716
1717 for_each_possible_cpu(i) {
1718 struct ipstats_mib *af_inet_stats;
1719 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i);
1720 u64_stats_init(&af_inet_stats->syncp);
1721 }
1722
1723 net->mib.net_statistics = alloc_percpu(struct linux_mib);
1724 if (!net->mib.net_statistics)
1725 goto err_net_mib;
1726 net->mib.udp_statistics = alloc_percpu(struct udp_mib);
1727 if (!net->mib.udp_statistics)
1728 goto err_udp_mib;
1729 net->mib.udplite_statistics = alloc_percpu(struct udp_mib);
1730 if (!net->mib.udplite_statistics)
1731 goto err_udplite_mib;
1732 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib);
1733 if (!net->mib.icmp_statistics)
1734 goto err_icmp_mib;
1735 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib),
1736 GFP_KERNEL);
1737 if (!net->mib.icmpmsg_statistics)
1738 goto err_icmpmsg_mib;
1739
1740 tcp_mib_init(net);
1741 return 0;
1742
1743err_icmpmsg_mib:
1744 free_percpu(net->mib.icmp_statistics);
1745err_icmp_mib:
1746 free_percpu(net->mib.udplite_statistics);
1747err_udplite_mib:
1748 free_percpu(net->mib.udp_statistics);
1749err_udp_mib:
1750 free_percpu(net->mib.net_statistics);
1751err_net_mib:
1752 free_percpu(net->mib.ip_statistics);
1753err_ip_mib:
1754 free_percpu(net->mib.tcp_statistics);
1755err_tcp_mib:
1756 return -ENOMEM;
1757}
1758
1759static __net_exit void ipv4_mib_exit_net(struct net *net)
1760{
1761 kfree(net->mib.icmpmsg_statistics);
1762 free_percpu(net->mib.icmp_statistics);
1763 free_percpu(net->mib.udplite_statistics);
1764 free_percpu(net->mib.udp_statistics);
1765 free_percpu(net->mib.net_statistics);
1766 free_percpu(net->mib.ip_statistics);
1767 free_percpu(net->mib.tcp_statistics);
1768}
1769
1770static __net_initdata struct pernet_operations ipv4_mib_ops = {
1771 .init = ipv4_mib_init_net,
1772 .exit = ipv4_mib_exit_net,
1773};
1774
1775static int __init init_ipv4_mibs(void)
1776{
1777 return register_pernet_subsys(&ipv4_mib_ops);
1778}
1779
1780static __net_init int inet_init_net(struct net *net)
1781{
1782 /*
1783 * Set defaults for local port range
1784 */
1785 seqlock_init(&net->ipv4.ip_local_ports.lock);
1786 net->ipv4.ip_local_ports.range[0] = 32768;
1787 net->ipv4.ip_local_ports.range[1] = 60999;
1788
1789 seqlock_init(&net->ipv4.ping_group_range.lock);
1790 /*
1791 * Sane defaults - nobody may create ping sockets.
1792 * Boot scripts should set this to distro-specific group.
1793 */
1794 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1);
1795 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0);
1796
1797 /* Default values for sysctl-controlled parameters.
1798 * We set them here, in case sysctl is not compiled.
1799 */
1800 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL;
1801 net->ipv4.sysctl_ip_dynaddr = 0;
1802 net->ipv4.sysctl_ip_early_demux = 1;
1803 net->ipv4.sysctl_udp_early_demux = 1;
1804 net->ipv4.sysctl_tcp_early_demux = 1;
1805#ifdef CONFIG_SYSCTL
1806 net->ipv4.sysctl_ip_prot_sock = PROT_SOCK;
1807#endif
1808
1809 /* Some igmp sysctl, whose values are always used */
1810 net->ipv4.sysctl_igmp_max_memberships = 20;
1811 net->ipv4.sysctl_igmp_max_msf = 10;
1812 /* IGMP reports for link-local multicast groups are enabled by default */
1813 net->ipv4.sysctl_igmp_llm_reports = 1;
1814 net->ipv4.sysctl_igmp_qrv = 2;
1815
1816 return 0;
1817}
1818
1819static __net_exit void inet_exit_net(struct net *net)
1820{
1821}
1822
1823static __net_initdata struct pernet_operations af_inet_ops = {
1824 .init = inet_init_net,
1825 .exit = inet_exit_net,
1826};
1827
1828static int __init init_inet_pernet_ops(void)
1829{
1830 return register_pernet_subsys(&af_inet_ops);
1831}
1832
1833static int ipv4_proc_init(void);
1834
1835/*
1836 * IP protocol layer initialiser
1837 */
1838
1839static struct packet_offload ip_packet_offload __read_mostly = {
1840 .type = cpu_to_be16(ETH_P_IP),
1841 .callbacks = {
1842 .gso_segment = inet_gso_segment,
1843 .gro_receive = inet_gro_receive,
1844 .gro_complete = inet_gro_complete,
1845 },
1846};
1847
1848static const struct net_offload ipip_offload = {
1849 .callbacks = {
1850 .gso_segment = inet_gso_segment,
1851 .gro_receive = ipip_gro_receive,
1852 .gro_complete = ipip_gro_complete,
1853 },
1854};
1855
1856static int __init ipip_offload_init(void)
1857{
1858 return inet_add_offload(&ipip_offload, IPPROTO_IPIP);
1859}
1860
1861static int __init ipv4_offload_init(void)
1862{
1863 /*
1864 * Add offloads
1865 */
1866 if (udpv4_offload_init() < 0)
1867 pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
1868 if (tcpv4_offload_init() < 0)
1869 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
1870 if (ipip_offload_init() < 0)
1871 pr_crit("%s: Cannot add IPIP protocol offload\n", __func__);
1872
1873 dev_add_offload(&ip_packet_offload);
1874 return 0;
1875}
1876
1877fs_initcall(ipv4_offload_init);
1878
1879static struct packet_type ip_packet_type __read_mostly = {
1880 .type = cpu_to_be16(ETH_P_IP),
1881 .func = ip_rcv,
1882};
1883
1884static int __init inet_init(void)
1885{
1886 struct inet_protosw *q;
1887 struct list_head *r;
1888 int rc = -EINVAL;
1889
1890 sock_skb_cb_check_size(sizeof(struct inet_skb_parm));
1891
1892 rc = proto_register(&tcp_prot, 1);
1893 if (rc)
1894 goto out;
1895
1896 rc = proto_register(&udp_prot, 1);
1897 if (rc)
1898 goto out_unregister_tcp_proto;
1899
1900 rc = proto_register(&raw_prot, 1);
1901 if (rc)
1902 goto out_unregister_udp_proto;
1903
1904 rc = proto_register(&ping_prot, 1);
1905 if (rc)
1906 goto out_unregister_raw_proto;
1907
1908 /*
1909 * Tell SOCKET that we are alive...
1910 */
1911
1912 (void)sock_register(&inet_family_ops);
1913
1914#ifdef CONFIG_SYSCTL
1915 ip_static_sysctl_init();
1916#endif
1917
1918 /*
1919 * Add all the base protocols.
1920 */
1921
1922 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1923 pr_crit("%s: Cannot add ICMP protocol\n", __func__);
1924 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1925 pr_crit("%s: Cannot add UDP protocol\n", __func__);
1926 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1927 pr_crit("%s: Cannot add TCP protocol\n", __func__);
1928#ifdef CONFIG_IP_MULTICAST
1929 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1930 pr_crit("%s: Cannot add IGMP protocol\n", __func__);
1931#endif
1932
1933 /* Register the socket-side information for inet_create. */
1934 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1935 INIT_LIST_HEAD(r);
1936
1937 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1938 inet_register_protosw(q);
1939
1940 /*
1941 * Set the ARP module up
1942 */
1943
1944 arp_init();
1945
1946 /*
1947 * Set the IP module up
1948 */
1949
1950 ip_init();
1951
1952 /* Setup TCP slab cache for open requests. */
1953 tcp_init();
1954
1955 /* Setup UDP memory threshold */
1956 udp_init();
1957
1958 /* Add UDP-Lite (RFC 3828) */
1959 udplite4_register();
1960
1961 ping_init();
1962
1963 /*
1964 * Set the ICMP layer up
1965 */
1966
1967 if (icmp_init() < 0)
1968 panic("Failed to create the ICMP control socket.\n");
1969
1970 /*
1971 * Initialise the multicast router
1972 */
1973#if defined(CONFIG_IP_MROUTE)
1974 if (ip_mr_init())
1975 pr_crit("%s: Cannot init ipv4 mroute\n", __func__);
1976#endif
1977
1978 if (init_inet_pernet_ops())
1979 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__);
1980 /*
1981 * Initialise per-cpu ipv4 mibs
1982 */
1983
1984 if (init_ipv4_mibs())
1985 pr_crit("%s: Cannot init ipv4 mibs\n", __func__);
1986
1987 ipv4_proc_init();
1988
1989 ipfrag_init();
1990
1991 dev_add_pack(&ip_packet_type);
1992
1993 ip_tunnel_core_init();
1994
1995 rc = 0;
1996out:
1997 return rc;
1998out_unregister_raw_proto:
1999 proto_unregister(&raw_prot);
2000out_unregister_udp_proto:
2001 proto_unregister(&udp_prot);
2002out_unregister_tcp_proto:
2003 proto_unregister(&tcp_prot);
2004 goto out;
2005}
2006
2007fs_initcall(inet_init);
2008
2009/* ------------------------------------------------------------------------ */
2010
2011#ifdef CONFIG_PROC_FS
2012static int __init ipv4_proc_init(void)
2013{
2014 int rc = 0;
2015
2016 if (raw_proc_init())
2017 goto out_raw;
2018 if (tcp4_proc_init())
2019 goto out_tcp;
2020 if (udp4_proc_init())
2021 goto out_udp;
2022 if (ping_proc_init())
2023 goto out_ping;
2024 if (ip_misc_proc_init())
2025 goto out_misc;
2026out:
2027 return rc;
2028out_misc:
2029 ping_proc_exit();
2030out_ping:
2031 udp4_proc_exit();
2032out_udp:
2033 tcp4_proc_exit();
2034out_tcp:
2035 raw_proc_exit();
2036out_raw:
2037 rc = -ENOMEM;
2038 goto out;
2039}
2040
2041#else /* CONFIG_PROC_FS */
2042static int __init ipv4_proc_init(void)
2043{
2044 return 0;
2045}
2046#endif /* CONFIG_PROC_FS */
1/*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * PF_INET protocol family socket handler.
7 *
8 * Authors: Ross Biro
9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
10 * Florian La Roche, <flla@stud.uni-sb.de>
11 * Alan Cox, <A.Cox@swansea.ac.uk>
12 *
13 * Changes (see also sock.c)
14 *
15 * piggy,
16 * Karl Knutson : Socket protocol table
17 * A.N.Kuznetsov : Socket death error in accept().
18 * John Richardson : Fix non blocking error in connect()
19 * so sockets that fail to connect
20 * don't return -EINPROGRESS.
21 * Alan Cox : Asynchronous I/O support
22 * Alan Cox : Keep correct socket pointer on sock
23 * structures
24 * when accept() ed
25 * Alan Cox : Semantics of SO_LINGER aren't state
26 * moved to close when you look carefully.
27 * With this fixed and the accept bug fixed
28 * some RPC stuff seems happier.
29 * Niibe Yutaka : 4.4BSD style write async I/O
30 * Alan Cox,
31 * Tony Gale : Fixed reuse semantics.
32 * Alan Cox : bind() shouldn't abort existing but dead
33 * sockets. Stops FTP netin:.. I hope.
34 * Alan Cox : bind() works correctly for RAW sockets.
35 * Note that FreeBSD at least was broken
36 * in this respect so be careful with
37 * compatibility tests...
38 * Alan Cox : routing cache support
39 * Alan Cox : memzero the socket structure for
40 * compactness.
41 * Matt Day : nonblock connect error handler
42 * Alan Cox : Allow large numbers of pending sockets
43 * (eg for big web sites), but only if
44 * specifically application requested.
45 * Alan Cox : New buffering throughout IP. Used
46 * dumbly.
47 * Alan Cox : New buffering now used smartly.
48 * Alan Cox : BSD rather than common sense
49 * interpretation of listen.
50 * Germano Caronni : Assorted small races.
51 * Alan Cox : sendmsg/recvmsg basic support.
52 * Alan Cox : Only sendmsg/recvmsg now supported.
53 * Alan Cox : Locked down bind (see security list).
54 * Alan Cox : Loosened bind a little.
55 * Mike McLagan : ADD/DEL DLCI Ioctls
56 * Willy Konynenberg : Transparent proxying support.
57 * David S. Miller : New socket lookup architecture.
58 * Some other random speedups.
59 * Cyrus Durgin : Cleaned up file for kmod hacks.
60 * Andi Kleen : Fix inet_stream_connect TCP race.
61 *
62 * This program is free software; you can redistribute it and/or
63 * modify it under the terms of the GNU General Public License
64 * as published by the Free Software Foundation; either version
65 * 2 of the License, or (at your option) any later version.
66 */
67
68#include <linux/err.h>
69#include <linux/errno.h>
70#include <linux/types.h>
71#include <linux/socket.h>
72#include <linux/in.h>
73#include <linux/kernel.h>
74#include <linux/module.h>
75#include <linux/sched.h>
76#include <linux/timer.h>
77#include <linux/string.h>
78#include <linux/sockios.h>
79#include <linux/net.h>
80#include <linux/capability.h>
81#include <linux/fcntl.h>
82#include <linux/mm.h>
83#include <linux/interrupt.h>
84#include <linux/stat.h>
85#include <linux/init.h>
86#include <linux/poll.h>
87#include <linux/netfilter_ipv4.h>
88#include <linux/random.h>
89#include <linux/slab.h>
90
91#include <asm/uaccess.h>
92#include <asm/system.h>
93
94#include <linux/inet.h>
95#include <linux/igmp.h>
96#include <linux/inetdevice.h>
97#include <linux/netdevice.h>
98#include <net/checksum.h>
99#include <net/ip.h>
100#include <net/protocol.h>
101#include <net/arp.h>
102#include <net/route.h>
103#include <net/ip_fib.h>
104#include <net/inet_connection_sock.h>
105#include <net/tcp.h>
106#include <net/udp.h>
107#include <net/udplite.h>
108#include <net/ping.h>
109#include <linux/skbuff.h>
110#include <net/sock.h>
111#include <net/raw.h>
112#include <net/icmp.h>
113#include <net/ipip.h>
114#include <net/inet_common.h>
115#include <net/xfrm.h>
116#include <net/net_namespace.h>
117#ifdef CONFIG_IP_MROUTE
118#include <linux/mroute.h>
119#endif
120
121
122/* The inetsw table contains everything that inet_create needs to
123 * build a new socket.
124 */
125static struct list_head inetsw[SOCK_MAX];
126static DEFINE_SPINLOCK(inetsw_lock);
127
128struct ipv4_config ipv4_config;
129EXPORT_SYMBOL(ipv4_config);
130
131/* New destruction routine */
132
133void 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 sk_refcnt_debug_dec(sk);
160}
161EXPORT_SYMBOL(inet_sock_destruct);
162
163/*
164 * The routines beyond this point handle the behaviour of an AF_INET
165 * socket object. Mostly it punts to the subprotocols of IP to do
166 * the work.
167 */
168
169/*
170 * Automatically bind an unbound socket.
171 */
172
173static int inet_autobind(struct sock *sk)
174{
175 struct inet_sock *inet;
176 /* We may need to bind the socket. */
177 lock_sock(sk);
178 inet = inet_sk(sk);
179 if (!inet->inet_num) {
180 if (sk->sk_prot->get_port(sk, 0)) {
181 release_sock(sk);
182 return -EAGAIN;
183 }
184 inet->inet_sport = htons(inet->inet_num);
185 }
186 release_sock(sk);
187 return 0;
188}
189
190/*
191 * Move a socket into listening state.
192 */
193int inet_listen(struct socket *sock, int backlog)
194{
195 struct sock *sk = sock->sk;
196 unsigned char old_state;
197 int err;
198
199 lock_sock(sk);
200
201 err = -EINVAL;
202 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
203 goto out;
204
205 old_state = sk->sk_state;
206 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN)))
207 goto out;
208
209 /* Really, if the socket is already in listen state
210 * we can only allow the backlog to be adjusted.
211 */
212 if (old_state != TCP_LISTEN) {
213 err = inet_csk_listen_start(sk, backlog);
214 if (err)
215 goto out;
216 }
217 sk->sk_max_ack_backlog = backlog;
218 err = 0;
219
220out:
221 release_sock(sk);
222 return err;
223}
224EXPORT_SYMBOL(inet_listen);
225
226u32 inet_ehash_secret __read_mostly;
227EXPORT_SYMBOL(inet_ehash_secret);
228
229/*
230 * inet_ehash_secret must be set exactly once
231 */
232void build_ehash_secret(void)
233{
234 u32 rnd;
235
236 do {
237 get_random_bytes(&rnd, sizeof(rnd));
238 } while (rnd == 0);
239
240 cmpxchg(&inet_ehash_secret, 0, rnd);
241}
242EXPORT_SYMBOL(build_ehash_secret);
243
244static inline int inet_netns_ok(struct net *net, int protocol)
245{
246 int hash;
247 const struct net_protocol *ipprot;
248
249 if (net_eq(net, &init_net))
250 return 1;
251
252 hash = protocol & (MAX_INET_PROTOS - 1);
253 ipprot = rcu_dereference(inet_protos[hash]);
254
255 if (ipprot == NULL)
256 /* raw IP is OK */
257 return 1;
258 return ipprot->netns_ok;
259}
260
261/*
262 * Create an inet socket.
263 */
264
265static int inet_create(struct net *net, struct socket *sock, int protocol,
266 int kern)
267{
268 struct sock *sk;
269 struct inet_protosw *answer;
270 struct inet_sock *inet;
271 struct proto *answer_prot;
272 unsigned char answer_flags;
273 char answer_no_check;
274 int try_loading_module = 0;
275 int err;
276
277 if (unlikely(!inet_ehash_secret))
278 if (sock->type != SOCK_RAW && sock->type != SOCK_DGRAM)
279 build_ehash_secret();
280
281 sock->state = SS_UNCONNECTED;
282
283 /* Look for the requested type/protocol pair. */
284lookup_protocol:
285 err = -ESOCKTNOSUPPORT;
286 rcu_read_lock();
287 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) {
288
289 err = 0;
290 /* Check the non-wild match. */
291 if (protocol == answer->protocol) {
292 if (protocol != IPPROTO_IP)
293 break;
294 } else {
295 /* Check for the two wild cases. */
296 if (IPPROTO_IP == protocol) {
297 protocol = answer->protocol;
298 break;
299 }
300 if (IPPROTO_IP == answer->protocol)
301 break;
302 }
303 err = -EPROTONOSUPPORT;
304 }
305
306 if (unlikely(err)) {
307 if (try_loading_module < 2) {
308 rcu_read_unlock();
309 /*
310 * Be more specific, e.g. net-pf-2-proto-132-type-1
311 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM)
312 */
313 if (++try_loading_module == 1)
314 request_module("net-pf-%d-proto-%d-type-%d",
315 PF_INET, protocol, sock->type);
316 /*
317 * Fall back to generic, e.g. net-pf-2-proto-132
318 * (net-pf-PF_INET-proto-IPPROTO_SCTP)
319 */
320 else
321 request_module("net-pf-%d-proto-%d",
322 PF_INET, protocol);
323 goto lookup_protocol;
324 } else
325 goto out_rcu_unlock;
326 }
327
328 err = -EPERM;
329 if (sock->type == SOCK_RAW && !kern && !capable(CAP_NET_RAW))
330 goto out_rcu_unlock;
331
332 err = -EAFNOSUPPORT;
333 if (!inet_netns_ok(net, protocol))
334 goto out_rcu_unlock;
335
336 sock->ops = answer->ops;
337 answer_prot = answer->prot;
338 answer_no_check = answer->no_check;
339 answer_flags = answer->flags;
340 rcu_read_unlock();
341
342 WARN_ON(answer_prot->slab == NULL);
343
344 err = -ENOBUFS;
345 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot);
346 if (sk == NULL)
347 goto out;
348
349 err = 0;
350 sk->sk_no_check = answer_no_check;
351 if (INET_PROTOSW_REUSE & answer_flags)
352 sk->sk_reuse = 1;
353
354 inet = inet_sk(sk);
355 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0;
356
357 inet->nodefrag = 0;
358
359 if (SOCK_RAW == sock->type) {
360 inet->inet_num = protocol;
361 if (IPPROTO_RAW == protocol)
362 inet->hdrincl = 1;
363 }
364
365 if (ipv4_config.no_pmtu_disc)
366 inet->pmtudisc = IP_PMTUDISC_DONT;
367 else
368 inet->pmtudisc = IP_PMTUDISC_WANT;
369
370 inet->inet_id = 0;
371
372 sock_init_data(sock, sk);
373
374 sk->sk_destruct = inet_sock_destruct;
375 sk->sk_protocol = protocol;
376 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv;
377
378 inet->uc_ttl = -1;
379 inet->mc_loop = 1;
380 inet->mc_ttl = 1;
381 inet->mc_all = 1;
382 inet->mc_index = 0;
383 inet->mc_list = NULL;
384
385 sk_refcnt_debug_inc(sk);
386
387 if (inet->inet_num) {
388 /* It assumes that any protocol which allows
389 * the user to assign a number at socket
390 * creation time automatically
391 * shares.
392 */
393 inet->inet_sport = htons(inet->inet_num);
394 /* Add to protocol hash chains. */
395 sk->sk_prot->hash(sk);
396 }
397
398 if (sk->sk_prot->init) {
399 err = sk->sk_prot->init(sk);
400 if (err)
401 sk_common_release(sk);
402 }
403out:
404 return err;
405out_rcu_unlock:
406 rcu_read_unlock();
407 goto out;
408}
409
410
411/*
412 * The peer socket should always be NULL (or else). When we call this
413 * function we are destroying the object and from then on nobody
414 * should refer to it.
415 */
416int inet_release(struct socket *sock)
417{
418 struct sock *sk = sock->sk;
419
420 if (sk) {
421 long timeout;
422
423 sock_rps_reset_flow(sk);
424
425 /* Applications forget to leave groups before exiting */
426 ip_mc_drop_socket(sk);
427
428 /* If linger is set, we don't return until the close
429 * is complete. Otherwise we return immediately. The
430 * actually closing is done the same either way.
431 *
432 * If the close is due to the process exiting, we never
433 * linger..
434 */
435 timeout = 0;
436 if (sock_flag(sk, SOCK_LINGER) &&
437 !(current->flags & PF_EXITING))
438 timeout = sk->sk_lingertime;
439 sock->sk = NULL;
440 sk->sk_prot->close(sk, timeout);
441 }
442 return 0;
443}
444EXPORT_SYMBOL(inet_release);
445
446/* It is off by default, see below. */
447int sysctl_ip_nonlocal_bind __read_mostly;
448EXPORT_SYMBOL(sysctl_ip_nonlocal_bind);
449
450int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
451{
452 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
453 struct sock *sk = sock->sk;
454 struct inet_sock *inet = inet_sk(sk);
455 unsigned short snum;
456 int chk_addr_ret;
457 int err;
458
459 /* If the socket has its own bind function then use it. (RAW) */
460 if (sk->sk_prot->bind) {
461 err = sk->sk_prot->bind(sk, uaddr, addr_len);
462 goto out;
463 }
464 err = -EINVAL;
465 if (addr_len < sizeof(struct sockaddr_in))
466 goto out;
467
468 if (addr->sin_family != AF_INET) {
469 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET)
470 * only if s_addr is INADDR_ANY.
471 */
472 err = -EAFNOSUPPORT;
473 if (addr->sin_family != AF_UNSPEC ||
474 addr->sin_addr.s_addr != htonl(INADDR_ANY))
475 goto out;
476 }
477
478 chk_addr_ret = inet_addr_type(sock_net(sk), addr->sin_addr.s_addr);
479
480 /* Not specified by any standard per-se, however it breaks too
481 * many applications when removed. It is unfortunate since
482 * allowing applications to make a non-local bind solves
483 * several problems with systems using dynamic addressing.
484 * (ie. your servers still start up even if your ISDN link
485 * is temporarily down)
486 */
487 err = -EADDRNOTAVAIL;
488 if (!sysctl_ip_nonlocal_bind &&
489 !(inet->freebind || inet->transparent) &&
490 addr->sin_addr.s_addr != htonl(INADDR_ANY) &&
491 chk_addr_ret != RTN_LOCAL &&
492 chk_addr_ret != RTN_MULTICAST &&
493 chk_addr_ret != RTN_BROADCAST)
494 goto out;
495
496 snum = ntohs(addr->sin_port);
497 err = -EACCES;
498 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
499 goto out;
500
501 /* We keep a pair of addresses. rcv_saddr is the one
502 * used by hash lookups, and saddr is used for transmit.
503 *
504 * In the BSD API these are the same except where it
505 * would be illegal to use them (multicast/broadcast) in
506 * which case the sending device address is used.
507 */
508 lock_sock(sk);
509
510 /* Check these errors (active socket, double bind). */
511 err = -EINVAL;
512 if (sk->sk_state != TCP_CLOSE || inet->inet_num)
513 goto out_release_sock;
514
515 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr;
516 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
517 inet->inet_saddr = 0; /* Use device */
518
519 /* Make sure we are allowed to bind here. */
520 if (sk->sk_prot->get_port(sk, snum)) {
521 inet->inet_saddr = inet->inet_rcv_saddr = 0;
522 err = -EADDRINUSE;
523 goto out_release_sock;
524 }
525
526 if (inet->inet_rcv_saddr)
527 sk->sk_userlocks |= SOCK_BINDADDR_LOCK;
528 if (snum)
529 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
530 inet->inet_sport = htons(inet->inet_num);
531 inet->inet_daddr = 0;
532 inet->inet_dport = 0;
533 sk_dst_reset(sk);
534 err = 0;
535out_release_sock:
536 release_sock(sk);
537out:
538 return err;
539}
540EXPORT_SYMBOL(inet_bind);
541
542int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
543 int addr_len, int flags)
544{
545 struct sock *sk = sock->sk;
546
547 if (addr_len < sizeof(uaddr->sa_family))
548 return -EINVAL;
549 if (uaddr->sa_family == AF_UNSPEC)
550 return sk->sk_prot->disconnect(sk, flags);
551
552 if (!inet_sk(sk)->inet_num && inet_autobind(sk))
553 return -EAGAIN;
554 return sk->sk_prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
555}
556EXPORT_SYMBOL(inet_dgram_connect);
557
558static long inet_wait_for_connect(struct sock *sk, long timeo)
559{
560 DEFINE_WAIT(wait);
561
562 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
563
564 /* Basic assumption: if someone sets sk->sk_err, he _must_
565 * change state of the socket from TCP_SYN_*.
566 * Connect() does not allow to get error notifications
567 * without closing the socket.
568 */
569 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
570 release_sock(sk);
571 timeo = schedule_timeout(timeo);
572 lock_sock(sk);
573 if (signal_pending(current) || !timeo)
574 break;
575 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
576 }
577 finish_wait(sk_sleep(sk), &wait);
578 return timeo;
579}
580
581/*
582 * Connect to a remote host. There is regrettably still a little
583 * TCP 'magic' in here.
584 */
585int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr,
586 int addr_len, int flags)
587{
588 struct sock *sk = sock->sk;
589 int err;
590 long timeo;
591
592 if (addr_len < sizeof(uaddr->sa_family))
593 return -EINVAL;
594
595 lock_sock(sk);
596
597 if (uaddr->sa_family == AF_UNSPEC) {
598 err = sk->sk_prot->disconnect(sk, flags);
599 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
600 goto out;
601 }
602
603 switch (sock->state) {
604 default:
605 err = -EINVAL;
606 goto out;
607 case SS_CONNECTED:
608 err = -EISCONN;
609 goto out;
610 case SS_CONNECTING:
611 err = -EALREADY;
612 /* Fall out of switch with err, set for this state */
613 break;
614 case SS_UNCONNECTED:
615 err = -EISCONN;
616 if (sk->sk_state != TCP_CLOSE)
617 goto out;
618
619 err = sk->sk_prot->connect(sk, uaddr, addr_len);
620 if (err < 0)
621 goto out;
622
623 sock->state = SS_CONNECTING;
624
625 /* Just entered SS_CONNECTING state; the only
626 * difference is that return value in non-blocking
627 * case is EINPROGRESS, rather than EALREADY.
628 */
629 err = -EINPROGRESS;
630 break;
631 }
632
633 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK);
634
635 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
636 /* Error code is set above */
637 if (!timeo || !inet_wait_for_connect(sk, timeo))
638 goto out;
639
640 err = sock_intr_errno(timeo);
641 if (signal_pending(current))
642 goto out;
643 }
644
645 /* Connection was closed by RST, timeout, ICMP error
646 * or another process disconnected us.
647 */
648 if (sk->sk_state == TCP_CLOSE)
649 goto sock_error;
650
651 /* sk->sk_err may be not zero now, if RECVERR was ordered by user
652 * and error was received after socket entered established state.
653 * Hence, it is handled normally after connect() return successfully.
654 */
655
656 sock->state = SS_CONNECTED;
657 err = 0;
658out:
659 release_sock(sk);
660 return err;
661
662sock_error:
663 err = sock_error(sk) ? : -ECONNABORTED;
664 sock->state = SS_UNCONNECTED;
665 if (sk->sk_prot->disconnect(sk, flags))
666 sock->state = SS_DISCONNECTING;
667 goto out;
668}
669EXPORT_SYMBOL(inet_stream_connect);
670
671/*
672 * Accept a pending connection. The TCP layer now gives BSD semantics.
673 */
674
675int inet_accept(struct socket *sock, struct socket *newsock, int flags)
676{
677 struct sock *sk1 = sock->sk;
678 int err = -EINVAL;
679 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err);
680
681 if (!sk2)
682 goto do_err;
683
684 lock_sock(sk2);
685
686 sock_rps_record_flow(sk2);
687 WARN_ON(!((1 << sk2->sk_state) &
688 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT | TCPF_CLOSE)));
689
690 sock_graft(sk2, newsock);
691
692 newsock->state = SS_CONNECTED;
693 err = 0;
694 release_sock(sk2);
695do_err:
696 return err;
697}
698EXPORT_SYMBOL(inet_accept);
699
700
701/*
702 * This does both peername and sockname.
703 */
704int inet_getname(struct socket *sock, struct sockaddr *uaddr,
705 int *uaddr_len, int peer)
706{
707 struct sock *sk = sock->sk;
708 struct inet_sock *inet = inet_sk(sk);
709 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr);
710
711 sin->sin_family = AF_INET;
712 if (peer) {
713 if (!inet->inet_dport ||
714 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) &&
715 peer == 1))
716 return -ENOTCONN;
717 sin->sin_port = inet->inet_dport;
718 sin->sin_addr.s_addr = inet->inet_daddr;
719 } else {
720 __be32 addr = inet->inet_rcv_saddr;
721 if (!addr)
722 addr = inet->inet_saddr;
723 sin->sin_port = inet->inet_sport;
724 sin->sin_addr.s_addr = addr;
725 }
726 memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
727 *uaddr_len = sizeof(*sin);
728 return 0;
729}
730EXPORT_SYMBOL(inet_getname);
731
732int inet_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
733 size_t size)
734{
735 struct sock *sk = sock->sk;
736
737 sock_rps_record_flow(sk);
738
739 /* We may need to bind the socket. */
740 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
741 inet_autobind(sk))
742 return -EAGAIN;
743
744 return sk->sk_prot->sendmsg(iocb, sk, msg, size);
745}
746EXPORT_SYMBOL(inet_sendmsg);
747
748ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset,
749 size_t size, int flags)
750{
751 struct sock *sk = sock->sk;
752
753 sock_rps_record_flow(sk);
754
755 /* We may need to bind the socket. */
756 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind &&
757 inet_autobind(sk))
758 return -EAGAIN;
759
760 if (sk->sk_prot->sendpage)
761 return sk->sk_prot->sendpage(sk, page, offset, size, flags);
762 return sock_no_sendpage(sock, page, offset, size, flags);
763}
764EXPORT_SYMBOL(inet_sendpage);
765
766int inet_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *msg,
767 size_t size, int flags)
768{
769 struct sock *sk = sock->sk;
770 int addr_len = 0;
771 int err;
772
773 sock_rps_record_flow(sk);
774
775 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
776 flags & ~MSG_DONTWAIT, &addr_len);
777 if (err >= 0)
778 msg->msg_namelen = addr_len;
779 return err;
780}
781EXPORT_SYMBOL(inet_recvmsg);
782
783int inet_shutdown(struct socket *sock, int how)
784{
785 struct sock *sk = sock->sk;
786 int err = 0;
787
788 /* This should really check to make sure
789 * the socket is a TCP socket. (WHY AC...)
790 */
791 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
792 1->2 bit 2 snds.
793 2->3 */
794 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */
795 return -EINVAL;
796
797 lock_sock(sk);
798 if (sock->state == SS_CONNECTING) {
799 if ((1 << sk->sk_state) &
800 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE))
801 sock->state = SS_DISCONNECTING;
802 else
803 sock->state = SS_CONNECTED;
804 }
805
806 switch (sk->sk_state) {
807 case TCP_CLOSE:
808 err = -ENOTCONN;
809 /* Hack to wake up other listeners, who can poll for
810 POLLHUP, even on eg. unconnected UDP sockets -- RR */
811 default:
812 sk->sk_shutdown |= how;
813 if (sk->sk_prot->shutdown)
814 sk->sk_prot->shutdown(sk, how);
815 break;
816
817 /* Remaining two branches are temporary solution for missing
818 * close() in multithreaded environment. It is _not_ a good idea,
819 * but we have no choice until close() is repaired at VFS level.
820 */
821 case TCP_LISTEN:
822 if (!(how & RCV_SHUTDOWN))
823 break;
824 /* Fall through */
825 case TCP_SYN_SENT:
826 err = sk->sk_prot->disconnect(sk, O_NONBLOCK);
827 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
828 break;
829 }
830
831 /* Wake up anyone sleeping in poll. */
832 sk->sk_state_change(sk);
833 release_sock(sk);
834 return err;
835}
836EXPORT_SYMBOL(inet_shutdown);
837
838/*
839 * ioctl() calls you can issue on an INET socket. Most of these are
840 * device configuration and stuff and very rarely used. Some ioctls
841 * pass on to the socket itself.
842 *
843 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
844 * loads the devconfigure module does its configuring and unloads it.
845 * There's a good 20K of config code hanging around the kernel.
846 */
847
848int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
849{
850 struct sock *sk = sock->sk;
851 int err = 0;
852 struct net *net = sock_net(sk);
853
854 switch (cmd) {
855 case SIOCGSTAMP:
856 err = sock_get_timestamp(sk, (struct timeval __user *)arg);
857 break;
858 case SIOCGSTAMPNS:
859 err = sock_get_timestampns(sk, (struct timespec __user *)arg);
860 break;
861 case SIOCADDRT:
862 case SIOCDELRT:
863 case SIOCRTMSG:
864 err = ip_rt_ioctl(net, cmd, (void __user *)arg);
865 break;
866 case SIOCDARP:
867 case SIOCGARP:
868 case SIOCSARP:
869 err = arp_ioctl(net, cmd, (void __user *)arg);
870 break;
871 case SIOCGIFADDR:
872 case SIOCSIFADDR:
873 case SIOCGIFBRDADDR:
874 case SIOCSIFBRDADDR:
875 case SIOCGIFNETMASK:
876 case SIOCSIFNETMASK:
877 case SIOCGIFDSTADDR:
878 case SIOCSIFDSTADDR:
879 case SIOCSIFPFLAGS:
880 case SIOCGIFPFLAGS:
881 case SIOCSIFFLAGS:
882 err = devinet_ioctl(net, cmd, (void __user *)arg);
883 break;
884 default:
885 if (sk->sk_prot->ioctl)
886 err = sk->sk_prot->ioctl(sk, cmd, arg);
887 else
888 err = -ENOIOCTLCMD;
889 break;
890 }
891 return err;
892}
893EXPORT_SYMBOL(inet_ioctl);
894
895#ifdef CONFIG_COMPAT
896int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
897{
898 struct sock *sk = sock->sk;
899 int err = -ENOIOCTLCMD;
900
901 if (sk->sk_prot->compat_ioctl)
902 err = sk->sk_prot->compat_ioctl(sk, cmd, arg);
903
904 return err;
905}
906#endif
907
908const struct proto_ops inet_stream_ops = {
909 .family = PF_INET,
910 .owner = THIS_MODULE,
911 .release = inet_release,
912 .bind = inet_bind,
913 .connect = inet_stream_connect,
914 .socketpair = sock_no_socketpair,
915 .accept = inet_accept,
916 .getname = inet_getname,
917 .poll = tcp_poll,
918 .ioctl = inet_ioctl,
919 .listen = inet_listen,
920 .shutdown = inet_shutdown,
921 .setsockopt = sock_common_setsockopt,
922 .getsockopt = sock_common_getsockopt,
923 .sendmsg = inet_sendmsg,
924 .recvmsg = inet_recvmsg,
925 .mmap = sock_no_mmap,
926 .sendpage = inet_sendpage,
927 .splice_read = tcp_splice_read,
928#ifdef CONFIG_COMPAT
929 .compat_setsockopt = compat_sock_common_setsockopt,
930 .compat_getsockopt = compat_sock_common_getsockopt,
931 .compat_ioctl = inet_compat_ioctl,
932#endif
933};
934EXPORT_SYMBOL(inet_stream_ops);
935
936const struct proto_ops inet_dgram_ops = {
937 .family = PF_INET,
938 .owner = THIS_MODULE,
939 .release = inet_release,
940 .bind = inet_bind,
941 .connect = inet_dgram_connect,
942 .socketpair = sock_no_socketpair,
943 .accept = sock_no_accept,
944 .getname = inet_getname,
945 .poll = udp_poll,
946 .ioctl = inet_ioctl,
947 .listen = sock_no_listen,
948 .shutdown = inet_shutdown,
949 .setsockopt = sock_common_setsockopt,
950 .getsockopt = sock_common_getsockopt,
951 .sendmsg = inet_sendmsg,
952 .recvmsg = inet_recvmsg,
953 .mmap = sock_no_mmap,
954 .sendpage = inet_sendpage,
955#ifdef CONFIG_COMPAT
956 .compat_setsockopt = compat_sock_common_setsockopt,
957 .compat_getsockopt = compat_sock_common_getsockopt,
958 .compat_ioctl = inet_compat_ioctl,
959#endif
960};
961EXPORT_SYMBOL(inet_dgram_ops);
962
963/*
964 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without
965 * udp_poll
966 */
967static const struct proto_ops inet_sockraw_ops = {
968 .family = PF_INET,
969 .owner = THIS_MODULE,
970 .release = inet_release,
971 .bind = inet_bind,
972 .connect = inet_dgram_connect,
973 .socketpair = sock_no_socketpair,
974 .accept = sock_no_accept,
975 .getname = inet_getname,
976 .poll = datagram_poll,
977 .ioctl = inet_ioctl,
978 .listen = sock_no_listen,
979 .shutdown = inet_shutdown,
980 .setsockopt = sock_common_setsockopt,
981 .getsockopt = sock_common_getsockopt,
982 .sendmsg = inet_sendmsg,
983 .recvmsg = inet_recvmsg,
984 .mmap = sock_no_mmap,
985 .sendpage = inet_sendpage,
986#ifdef CONFIG_COMPAT
987 .compat_setsockopt = compat_sock_common_setsockopt,
988 .compat_getsockopt = compat_sock_common_getsockopt,
989 .compat_ioctl = inet_compat_ioctl,
990#endif
991};
992
993static const struct net_proto_family inet_family_ops = {
994 .family = PF_INET,
995 .create = inet_create,
996 .owner = THIS_MODULE,
997};
998
999/* Upon startup we insert all the elements in inetsw_array[] into
1000 * the linked list inetsw.
1001 */
1002static struct inet_protosw inetsw_array[] =
1003{
1004 {
1005 .type = SOCK_STREAM,
1006 .protocol = IPPROTO_TCP,
1007 .prot = &tcp_prot,
1008 .ops = &inet_stream_ops,
1009 .no_check = 0,
1010 .flags = INET_PROTOSW_PERMANENT |
1011 INET_PROTOSW_ICSK,
1012 },
1013
1014 {
1015 .type = SOCK_DGRAM,
1016 .protocol = IPPROTO_UDP,
1017 .prot = &udp_prot,
1018 .ops = &inet_dgram_ops,
1019 .no_check = UDP_CSUM_DEFAULT,
1020 .flags = INET_PROTOSW_PERMANENT,
1021 },
1022
1023 {
1024 .type = SOCK_DGRAM,
1025 .protocol = IPPROTO_ICMP,
1026 .prot = &ping_prot,
1027 .ops = &inet_dgram_ops,
1028 .no_check = UDP_CSUM_DEFAULT,
1029 .flags = INET_PROTOSW_REUSE,
1030 },
1031
1032 {
1033 .type = SOCK_RAW,
1034 .protocol = IPPROTO_IP, /* wild card */
1035 .prot = &raw_prot,
1036 .ops = &inet_sockraw_ops,
1037 .no_check = UDP_CSUM_DEFAULT,
1038 .flags = INET_PROTOSW_REUSE,
1039 }
1040};
1041
1042#define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array)
1043
1044void inet_register_protosw(struct inet_protosw *p)
1045{
1046 struct list_head *lh;
1047 struct inet_protosw *answer;
1048 int protocol = p->protocol;
1049 struct list_head *last_perm;
1050
1051 spin_lock_bh(&inetsw_lock);
1052
1053 if (p->type >= SOCK_MAX)
1054 goto out_illegal;
1055
1056 /* If we are trying to override a permanent protocol, bail. */
1057 answer = NULL;
1058 last_perm = &inetsw[p->type];
1059 list_for_each(lh, &inetsw[p->type]) {
1060 answer = list_entry(lh, struct inet_protosw, list);
1061
1062 /* Check only the non-wild match. */
1063 if (INET_PROTOSW_PERMANENT & answer->flags) {
1064 if (protocol == answer->protocol)
1065 break;
1066 last_perm = lh;
1067 }
1068
1069 answer = NULL;
1070 }
1071 if (answer)
1072 goto out_permanent;
1073
1074 /* Add the new entry after the last permanent entry if any, so that
1075 * the new entry does not override a permanent entry when matched with
1076 * a wild-card protocol. But it is allowed to override any existing
1077 * non-permanent entry. This means that when we remove this entry, the
1078 * system automatically returns to the old behavior.
1079 */
1080 list_add_rcu(&p->list, last_perm);
1081out:
1082 spin_unlock_bh(&inetsw_lock);
1083
1084 return;
1085
1086out_permanent:
1087 printk(KERN_ERR "Attempt to override permanent protocol %d.\n",
1088 protocol);
1089 goto out;
1090
1091out_illegal:
1092 printk(KERN_ERR
1093 "Ignoring attempt to register invalid socket type %d.\n",
1094 p->type);
1095 goto out;
1096}
1097EXPORT_SYMBOL(inet_register_protosw);
1098
1099void inet_unregister_protosw(struct inet_protosw *p)
1100{
1101 if (INET_PROTOSW_PERMANENT & p->flags) {
1102 printk(KERN_ERR
1103 "Attempt to unregister permanent protocol %d.\n",
1104 p->protocol);
1105 } else {
1106 spin_lock_bh(&inetsw_lock);
1107 list_del_rcu(&p->list);
1108 spin_unlock_bh(&inetsw_lock);
1109
1110 synchronize_net();
1111 }
1112}
1113EXPORT_SYMBOL(inet_unregister_protosw);
1114
1115/*
1116 * Shall we try to damage output packets if routing dev changes?
1117 */
1118
1119int sysctl_ip_dynaddr __read_mostly;
1120
1121static int inet_sk_reselect_saddr(struct sock *sk)
1122{
1123 struct inet_sock *inet = inet_sk(sk);
1124 __be32 old_saddr = inet->inet_saddr;
1125 __be32 daddr = inet->inet_daddr;
1126 struct flowi4 *fl4;
1127 struct rtable *rt;
1128 __be32 new_saddr;
1129 struct ip_options_rcu *inet_opt;
1130
1131 inet_opt = rcu_dereference_protected(inet->inet_opt,
1132 sock_owned_by_user(sk));
1133 if (inet_opt && inet_opt->opt.srr)
1134 daddr = inet_opt->opt.faddr;
1135
1136 /* Query new route. */
1137 fl4 = &inet->cork.fl.u.ip4;
1138 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk),
1139 sk->sk_bound_dev_if, sk->sk_protocol,
1140 inet->inet_sport, inet->inet_dport, sk, false);
1141 if (IS_ERR(rt))
1142 return PTR_ERR(rt);
1143
1144 sk_setup_caps(sk, &rt->dst);
1145
1146 new_saddr = fl4->saddr;
1147
1148 if (new_saddr == old_saddr)
1149 return 0;
1150
1151 if (sysctl_ip_dynaddr > 1) {
1152 printk(KERN_INFO "%s(): shifting inet->saddr from %pI4 to %pI4\n",
1153 __func__, &old_saddr, &new_saddr);
1154 }
1155
1156 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr;
1157
1158 /*
1159 * XXX The only one ugly spot where we need to
1160 * XXX really change the sockets identity after
1161 * XXX it has entered the hashes. -DaveM
1162 *
1163 * Besides that, it does not check for connection
1164 * uniqueness. Wait for troubles.
1165 */
1166 __sk_prot_rehash(sk);
1167 return 0;
1168}
1169
1170int inet_sk_rebuild_header(struct sock *sk)
1171{
1172 struct inet_sock *inet = inet_sk(sk);
1173 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0);
1174 __be32 daddr;
1175 struct ip_options_rcu *inet_opt;
1176 struct flowi4 *fl4;
1177 int err;
1178
1179 /* Route is OK, nothing to do. */
1180 if (rt)
1181 return 0;
1182
1183 /* Reroute. */
1184 rcu_read_lock();
1185 inet_opt = rcu_dereference(inet->inet_opt);
1186 daddr = inet->inet_daddr;
1187 if (inet_opt && inet_opt->opt.srr)
1188 daddr = inet_opt->opt.faddr;
1189 rcu_read_unlock();
1190 fl4 = &inet->cork.fl.u.ip4;
1191 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr,
1192 inet->inet_dport, inet->inet_sport,
1193 sk->sk_protocol, RT_CONN_FLAGS(sk),
1194 sk->sk_bound_dev_if);
1195 if (!IS_ERR(rt)) {
1196 err = 0;
1197 sk_setup_caps(sk, &rt->dst);
1198 } else {
1199 err = PTR_ERR(rt);
1200
1201 /* Routing failed... */
1202 sk->sk_route_caps = 0;
1203 /*
1204 * Other protocols have to map its equivalent state to TCP_SYN_SENT.
1205 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme
1206 */
1207 if (!sysctl_ip_dynaddr ||
1208 sk->sk_state != TCP_SYN_SENT ||
1209 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) ||
1210 (err = inet_sk_reselect_saddr(sk)) != 0)
1211 sk->sk_err_soft = -err;
1212 }
1213
1214 return err;
1215}
1216EXPORT_SYMBOL(inet_sk_rebuild_header);
1217
1218static int inet_gso_send_check(struct sk_buff *skb)
1219{
1220 const struct iphdr *iph;
1221 const struct net_protocol *ops;
1222 int proto;
1223 int ihl;
1224 int err = -EINVAL;
1225
1226 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1227 goto out;
1228
1229 iph = ip_hdr(skb);
1230 ihl = iph->ihl * 4;
1231 if (ihl < sizeof(*iph))
1232 goto out;
1233
1234 if (unlikely(!pskb_may_pull(skb, ihl)))
1235 goto out;
1236
1237 __skb_pull(skb, ihl);
1238 skb_reset_transport_header(skb);
1239 iph = ip_hdr(skb);
1240 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1241 err = -EPROTONOSUPPORT;
1242
1243 rcu_read_lock();
1244 ops = rcu_dereference(inet_protos[proto]);
1245 if (likely(ops && ops->gso_send_check))
1246 err = ops->gso_send_check(skb);
1247 rcu_read_unlock();
1248
1249out:
1250 return err;
1251}
1252
1253static struct sk_buff *inet_gso_segment(struct sk_buff *skb, u32 features)
1254{
1255 struct sk_buff *segs = ERR_PTR(-EINVAL);
1256 struct iphdr *iph;
1257 const struct net_protocol *ops;
1258 int proto;
1259 int ihl;
1260 int id;
1261 unsigned int offset = 0;
1262
1263 if (!(features & NETIF_F_V4_CSUM))
1264 features &= ~NETIF_F_SG;
1265
1266 if (unlikely(skb_shinfo(skb)->gso_type &
1267 ~(SKB_GSO_TCPV4 |
1268 SKB_GSO_UDP |
1269 SKB_GSO_DODGY |
1270 SKB_GSO_TCP_ECN |
1271 0)))
1272 goto out;
1273
1274 if (unlikely(!pskb_may_pull(skb, sizeof(*iph))))
1275 goto out;
1276
1277 iph = ip_hdr(skb);
1278 ihl = iph->ihl * 4;
1279 if (ihl < sizeof(*iph))
1280 goto out;
1281
1282 if (unlikely(!pskb_may_pull(skb, ihl)))
1283 goto out;
1284
1285 __skb_pull(skb, ihl);
1286 skb_reset_transport_header(skb);
1287 iph = ip_hdr(skb);
1288 id = ntohs(iph->id);
1289 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1290 segs = ERR_PTR(-EPROTONOSUPPORT);
1291
1292 rcu_read_lock();
1293 ops = rcu_dereference(inet_protos[proto]);
1294 if (likely(ops && ops->gso_segment))
1295 segs = ops->gso_segment(skb, features);
1296 rcu_read_unlock();
1297
1298 if (!segs || IS_ERR(segs))
1299 goto out;
1300
1301 skb = segs;
1302 do {
1303 iph = ip_hdr(skb);
1304 if (proto == IPPROTO_UDP) {
1305 iph->id = htons(id);
1306 iph->frag_off = htons(offset >> 3);
1307 if (skb->next != NULL)
1308 iph->frag_off |= htons(IP_MF);
1309 offset += (skb->len - skb->mac_len - iph->ihl * 4);
1310 } else
1311 iph->id = htons(id++);
1312 iph->tot_len = htons(skb->len - skb->mac_len);
1313 iph->check = 0;
1314 iph->check = ip_fast_csum(skb_network_header(skb), iph->ihl);
1315 } while ((skb = skb->next));
1316
1317out:
1318 return segs;
1319}
1320
1321static struct sk_buff **inet_gro_receive(struct sk_buff **head,
1322 struct sk_buff *skb)
1323{
1324 const struct net_protocol *ops;
1325 struct sk_buff **pp = NULL;
1326 struct sk_buff *p;
1327 const struct iphdr *iph;
1328 unsigned int hlen;
1329 unsigned int off;
1330 unsigned int id;
1331 int flush = 1;
1332 int proto;
1333
1334 off = skb_gro_offset(skb);
1335 hlen = off + sizeof(*iph);
1336 iph = skb_gro_header_fast(skb, off);
1337 if (skb_gro_header_hard(skb, hlen)) {
1338 iph = skb_gro_header_slow(skb, hlen, off);
1339 if (unlikely(!iph))
1340 goto out;
1341 }
1342
1343 proto = iph->protocol & (MAX_INET_PROTOS - 1);
1344
1345 rcu_read_lock();
1346 ops = rcu_dereference(inet_protos[proto]);
1347 if (!ops || !ops->gro_receive)
1348 goto out_unlock;
1349
1350 if (*(u8 *)iph != 0x45)
1351 goto out_unlock;
1352
1353 if (unlikely(ip_fast_csum((u8 *)iph, iph->ihl)))
1354 goto out_unlock;
1355
1356 id = ntohl(*(__be32 *)&iph->id);
1357 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id ^ IP_DF));
1358 id >>= 16;
1359
1360 for (p = *head; p; p = p->next) {
1361 struct iphdr *iph2;
1362
1363 if (!NAPI_GRO_CB(p)->same_flow)
1364 continue;
1365
1366 iph2 = ip_hdr(p);
1367
1368 if ((iph->protocol ^ iph2->protocol) |
1369 (iph->tos ^ iph2->tos) |
1370 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
1371 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
1372 NAPI_GRO_CB(p)->same_flow = 0;
1373 continue;
1374 }
1375
1376 /* All fields must match except length and checksum. */
1377 NAPI_GRO_CB(p)->flush |=
1378 (iph->ttl ^ iph2->ttl) |
1379 ((u16)(ntohs(iph2->id) + NAPI_GRO_CB(p)->count) ^ id);
1380
1381 NAPI_GRO_CB(p)->flush |= flush;
1382 }
1383
1384 NAPI_GRO_CB(skb)->flush |= flush;
1385 skb_gro_pull(skb, sizeof(*iph));
1386 skb_set_transport_header(skb, skb_gro_offset(skb));
1387
1388 pp = ops->gro_receive(head, skb);
1389
1390out_unlock:
1391 rcu_read_unlock();
1392
1393out:
1394 NAPI_GRO_CB(skb)->flush |= flush;
1395
1396 return pp;
1397}
1398
1399static int inet_gro_complete(struct sk_buff *skb)
1400{
1401 const struct net_protocol *ops;
1402 struct iphdr *iph = ip_hdr(skb);
1403 int proto = iph->protocol & (MAX_INET_PROTOS - 1);
1404 int err = -ENOSYS;
1405 __be16 newlen = htons(skb->len - skb_network_offset(skb));
1406
1407 csum_replace2(&iph->check, iph->tot_len, newlen);
1408 iph->tot_len = newlen;
1409
1410 rcu_read_lock();
1411 ops = rcu_dereference(inet_protos[proto]);
1412 if (WARN_ON(!ops || !ops->gro_complete))
1413 goto out_unlock;
1414
1415 err = ops->gro_complete(skb);
1416
1417out_unlock:
1418 rcu_read_unlock();
1419
1420 return err;
1421}
1422
1423int inet_ctl_sock_create(struct sock **sk, unsigned short family,
1424 unsigned short type, unsigned char protocol,
1425 struct net *net)
1426{
1427 struct socket *sock;
1428 int rc = sock_create_kern(family, type, protocol, &sock);
1429
1430 if (rc == 0) {
1431 *sk = sock->sk;
1432 (*sk)->sk_allocation = GFP_ATOMIC;
1433 /*
1434 * Unhash it so that IP input processing does not even see it,
1435 * we do not wish this socket to see incoming packets.
1436 */
1437 (*sk)->sk_prot->unhash(*sk);
1438
1439 sk_change_net(*sk, net);
1440 }
1441 return rc;
1442}
1443EXPORT_SYMBOL_GPL(inet_ctl_sock_create);
1444
1445unsigned long snmp_fold_field(void __percpu *mib[], int offt)
1446{
1447 unsigned long res = 0;
1448 int i, j;
1449
1450 for_each_possible_cpu(i) {
1451 for (j = 0; j < SNMP_ARRAY_SZ; j++)
1452 res += *(((unsigned long *) per_cpu_ptr(mib[j], i)) + offt);
1453 }
1454 return res;
1455}
1456EXPORT_SYMBOL_GPL(snmp_fold_field);
1457
1458#if BITS_PER_LONG==32
1459
1460u64 snmp_fold_field64(void __percpu *mib[], int offt, size_t syncp_offset)
1461{
1462 u64 res = 0;
1463 int cpu;
1464
1465 for_each_possible_cpu(cpu) {
1466 void *bhptr;
1467 struct u64_stats_sync *syncp;
1468 u64 v;
1469 unsigned int start;
1470
1471 bhptr = per_cpu_ptr(mib[0], cpu);
1472 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset);
1473 do {
1474 start = u64_stats_fetch_begin_bh(syncp);
1475 v = *(((u64 *) bhptr) + offt);
1476 } while (u64_stats_fetch_retry_bh(syncp, start));
1477
1478 res += v;
1479 }
1480 return res;
1481}
1482EXPORT_SYMBOL_GPL(snmp_fold_field64);
1483#endif
1484
1485int snmp_mib_init(void __percpu *ptr[2], size_t mibsize, size_t align)
1486{
1487 BUG_ON(ptr == NULL);
1488 ptr[0] = __alloc_percpu(mibsize, align);
1489 if (!ptr[0])
1490 return -ENOMEM;
1491#if SNMP_ARRAY_SZ == 2
1492 ptr[1] = __alloc_percpu(mibsize, align);
1493 if (!ptr[1]) {
1494 free_percpu(ptr[0]);
1495 ptr[0] = NULL;
1496 return -ENOMEM;
1497 }
1498#endif
1499 return 0;
1500}
1501EXPORT_SYMBOL_GPL(snmp_mib_init);
1502
1503void snmp_mib_free(void __percpu *ptr[SNMP_ARRAY_SZ])
1504{
1505 int i;
1506
1507 BUG_ON(ptr == NULL);
1508 for (i = 0; i < SNMP_ARRAY_SZ; i++) {
1509 free_percpu(ptr[i]);
1510 ptr[i] = NULL;
1511 }
1512}
1513EXPORT_SYMBOL_GPL(snmp_mib_free);
1514
1515#ifdef CONFIG_IP_MULTICAST
1516static const struct net_protocol igmp_protocol = {
1517 .handler = igmp_rcv,
1518 .netns_ok = 1,
1519};
1520#endif
1521
1522static const struct net_protocol tcp_protocol = {
1523 .handler = tcp_v4_rcv,
1524 .err_handler = tcp_v4_err,
1525 .gso_send_check = tcp_v4_gso_send_check,
1526 .gso_segment = tcp_tso_segment,
1527 .gro_receive = tcp4_gro_receive,
1528 .gro_complete = tcp4_gro_complete,
1529 .no_policy = 1,
1530 .netns_ok = 1,
1531};
1532
1533static const struct net_protocol udp_protocol = {
1534 .handler = udp_rcv,
1535 .err_handler = udp_err,
1536 .gso_send_check = udp4_ufo_send_check,
1537 .gso_segment = udp4_ufo_fragment,
1538 .no_policy = 1,
1539 .netns_ok = 1,
1540};
1541
1542static const struct net_protocol icmp_protocol = {
1543 .handler = icmp_rcv,
1544 .err_handler = ping_err,
1545 .no_policy = 1,
1546 .netns_ok = 1,
1547};
1548
1549static __net_init int ipv4_mib_init_net(struct net *net)
1550{
1551 if (snmp_mib_init((void __percpu **)net->mib.tcp_statistics,
1552 sizeof(struct tcp_mib),
1553 __alignof__(struct tcp_mib)) < 0)
1554 goto err_tcp_mib;
1555 if (snmp_mib_init((void __percpu **)net->mib.ip_statistics,
1556 sizeof(struct ipstats_mib),
1557 __alignof__(struct ipstats_mib)) < 0)
1558 goto err_ip_mib;
1559 if (snmp_mib_init((void __percpu **)net->mib.net_statistics,
1560 sizeof(struct linux_mib),
1561 __alignof__(struct linux_mib)) < 0)
1562 goto err_net_mib;
1563 if (snmp_mib_init((void __percpu **)net->mib.udp_statistics,
1564 sizeof(struct udp_mib),
1565 __alignof__(struct udp_mib)) < 0)
1566 goto err_udp_mib;
1567 if (snmp_mib_init((void __percpu **)net->mib.udplite_statistics,
1568 sizeof(struct udp_mib),
1569 __alignof__(struct udp_mib)) < 0)
1570 goto err_udplite_mib;
1571 if (snmp_mib_init((void __percpu **)net->mib.icmp_statistics,
1572 sizeof(struct icmp_mib),
1573 __alignof__(struct icmp_mib)) < 0)
1574 goto err_icmp_mib;
1575 if (snmp_mib_init((void __percpu **)net->mib.icmpmsg_statistics,
1576 sizeof(struct icmpmsg_mib),
1577 __alignof__(struct icmpmsg_mib)) < 0)
1578 goto err_icmpmsg_mib;
1579
1580 tcp_mib_init(net);
1581 return 0;
1582
1583err_icmpmsg_mib:
1584 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1585err_icmp_mib:
1586 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1587err_udplite_mib:
1588 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1589err_udp_mib:
1590 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1591err_net_mib:
1592 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1593err_ip_mib:
1594 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1595err_tcp_mib:
1596 return -ENOMEM;
1597}
1598
1599static __net_exit void ipv4_mib_exit_net(struct net *net)
1600{
1601 snmp_mib_free((void __percpu **)net->mib.icmpmsg_statistics);
1602 snmp_mib_free((void __percpu **)net->mib.icmp_statistics);
1603 snmp_mib_free((void __percpu **)net->mib.udplite_statistics);
1604 snmp_mib_free((void __percpu **)net->mib.udp_statistics);
1605 snmp_mib_free((void __percpu **)net->mib.net_statistics);
1606 snmp_mib_free((void __percpu **)net->mib.ip_statistics);
1607 snmp_mib_free((void __percpu **)net->mib.tcp_statistics);
1608}
1609
1610static __net_initdata struct pernet_operations ipv4_mib_ops = {
1611 .init = ipv4_mib_init_net,
1612 .exit = ipv4_mib_exit_net,
1613};
1614
1615static int __init init_ipv4_mibs(void)
1616{
1617 return register_pernet_subsys(&ipv4_mib_ops);
1618}
1619
1620static int ipv4_proc_init(void);
1621
1622/*
1623 * IP protocol layer initialiser
1624 */
1625
1626static struct packet_type ip_packet_type __read_mostly = {
1627 .type = cpu_to_be16(ETH_P_IP),
1628 .func = ip_rcv,
1629 .gso_send_check = inet_gso_send_check,
1630 .gso_segment = inet_gso_segment,
1631 .gro_receive = inet_gro_receive,
1632 .gro_complete = inet_gro_complete,
1633};
1634
1635static int __init inet_init(void)
1636{
1637 struct sk_buff *dummy_skb;
1638 struct inet_protosw *q;
1639 struct list_head *r;
1640 int rc = -EINVAL;
1641
1642 BUILD_BUG_ON(sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb));
1643
1644 sysctl_local_reserved_ports = kzalloc(65536 / 8, GFP_KERNEL);
1645 if (!sysctl_local_reserved_ports)
1646 goto out;
1647
1648 rc = proto_register(&tcp_prot, 1);
1649 if (rc)
1650 goto out_free_reserved_ports;
1651
1652 rc = proto_register(&udp_prot, 1);
1653 if (rc)
1654 goto out_unregister_tcp_proto;
1655
1656 rc = proto_register(&raw_prot, 1);
1657 if (rc)
1658 goto out_unregister_udp_proto;
1659
1660 rc = proto_register(&ping_prot, 1);
1661 if (rc)
1662 goto out_unregister_raw_proto;
1663
1664 /*
1665 * Tell SOCKET that we are alive...
1666 */
1667
1668 (void)sock_register(&inet_family_ops);
1669
1670#ifdef CONFIG_SYSCTL
1671 ip_static_sysctl_init();
1672#endif
1673
1674 /*
1675 * Add all the base protocols.
1676 */
1677
1678 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0)
1679 printk(KERN_CRIT "inet_init: Cannot add ICMP protocol\n");
1680 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0)
1681 printk(KERN_CRIT "inet_init: Cannot add UDP protocol\n");
1682 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0)
1683 printk(KERN_CRIT "inet_init: Cannot add TCP protocol\n");
1684#ifdef CONFIG_IP_MULTICAST
1685 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0)
1686 printk(KERN_CRIT "inet_init: Cannot add IGMP protocol\n");
1687#endif
1688
1689 /* Register the socket-side information for inet_create. */
1690 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r)
1691 INIT_LIST_HEAD(r);
1692
1693 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q)
1694 inet_register_protosw(q);
1695
1696 /*
1697 * Set the ARP module up
1698 */
1699
1700 arp_init();
1701
1702 /*
1703 * Set the IP module up
1704 */
1705
1706 ip_init();
1707
1708 tcp_v4_init();
1709
1710 /* Setup TCP slab cache for open requests. */
1711 tcp_init();
1712
1713 /* Setup UDP memory threshold */
1714 udp_init();
1715
1716 /* Add UDP-Lite (RFC 3828) */
1717 udplite4_register();
1718
1719 ping_init();
1720
1721 /*
1722 * Set the ICMP layer up
1723 */
1724
1725 if (icmp_init() < 0)
1726 panic("Failed to create the ICMP control socket.\n");
1727
1728 /*
1729 * Initialise the multicast router
1730 */
1731#if defined(CONFIG_IP_MROUTE)
1732 if (ip_mr_init())
1733 printk(KERN_CRIT "inet_init: Cannot init ipv4 mroute\n");
1734#endif
1735 /*
1736 * Initialise per-cpu ipv4 mibs
1737 */
1738
1739 if (init_ipv4_mibs())
1740 printk(KERN_CRIT "inet_init: Cannot init ipv4 mibs\n");
1741
1742 ipv4_proc_init();
1743
1744 ipfrag_init();
1745
1746 dev_add_pack(&ip_packet_type);
1747
1748 rc = 0;
1749out:
1750 return rc;
1751out_unregister_raw_proto:
1752 proto_unregister(&raw_prot);
1753out_unregister_udp_proto:
1754 proto_unregister(&udp_prot);
1755out_unregister_tcp_proto:
1756 proto_unregister(&tcp_prot);
1757out_free_reserved_ports:
1758 kfree(sysctl_local_reserved_ports);
1759 goto out;
1760}
1761
1762fs_initcall(inet_init);
1763
1764/* ------------------------------------------------------------------------ */
1765
1766#ifdef CONFIG_PROC_FS
1767static int __init ipv4_proc_init(void)
1768{
1769 int rc = 0;
1770
1771 if (raw_proc_init())
1772 goto out_raw;
1773 if (tcp4_proc_init())
1774 goto out_tcp;
1775 if (udp4_proc_init())
1776 goto out_udp;
1777 if (ping_proc_init())
1778 goto out_ping;
1779 if (ip_misc_proc_init())
1780 goto out_misc;
1781out:
1782 return rc;
1783out_misc:
1784 ping_proc_exit();
1785out_ping:
1786 udp4_proc_exit();
1787out_udp:
1788 tcp4_proc_exit();
1789out_tcp:
1790 raw_proc_exit();
1791out_raw:
1792 rc = -ENOMEM;
1793 goto out;
1794}
1795
1796#else /* CONFIG_PROC_FS */
1797static int __init ipv4_proc_init(void)
1798{
1799 return 0;
1800}
1801#endif /* CONFIG_PROC_FS */
1802
1803MODULE_ALIAS_NETPROTO(PF_INET);
1804