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 * Version: $Id: af_inet.c,v 1.127 2000/12/22 19:51:50 davem Exp $
9 *
10 * Authors: Ross Biro, <bir7@leland.Stanford.Edu>
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Changes (see also sock.c)
16 *
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 structures
23 * when accept() ed
24 * Alan Cox : Semantics of SO_LINGER aren't state moved
25 * to close when you look carefully. With
26 * this fixed and the accept bug fixed
27 * some RPC stuff seems happier.
28 * Niibe Yutaka : 4.4BSD style write async I/O
29 * Alan Cox,
30 * Tony Gale : Fixed reuse semantics.
31 * Alan Cox : bind() shouldn't abort existing but dead
32 * sockets. Stops FTP netin:.. I hope.
33 * Alan Cox : bind() works correctly for RAW sockets. Note
34 * that FreeBSD at least was broken in this respect
35 * so be careful with compatibility tests...
36 * Alan Cox : routing cache support
37 * Alan Cox : memzero the socket structure for compactness.
38 * Matt Day : nonblock connect error handler
39 * Alan Cox : Allow large numbers of pending sockets
40 * (eg for big web sites), but only if
41 * specifically application requested.
42 * Alan Cox : New buffering throughout IP. Used dumbly.
43 * Alan Cox : New buffering now used smartly.
44 * Alan Cox : BSD rather than common sense interpretation of
45 * listen.
46 * Germano Caronni : Assorted small races.
47 * Alan Cox : sendmsg/recvmsg basic support.
48 * Alan Cox : Only sendmsg/recvmsg now supported.
49 * Alan Cox : Locked down bind (see security list).
50 * Alan Cox : Loosened bind a little.
51 * Mike McLagan : ADD/DEL DLCI Ioctls
52 * Willy Konynenberg : Transparent proxying support.
53 * David S. Miller : New socket lookup architecture.
54 * Some other random speedups.
55 * Cyrus Durgin : Cleaned up file for kmod hacks.
56 * Andi Kleen : Fix inet_stream_connect TCP race.
57 *
58 * This program is free software; you can redistribute it and/or
59 * modify it under the terms of the GNU General Public License
60 * as published by the Free Software Foundation; either version
61 * 2 of the License, or (at your option) any later version.
62 */
63
64 #include <linux/config.h>
65 #include <linux/errno.h>
66 #include <linux/types.h>
67 #include <linux/socket.h>
68 #include <linux/in.h>
69 #include <linux/kernel.h>
70 #include <linux/major.h>
71 #include <linux/sched.h>
72 #include <linux/timer.h>
73 #include <linux/string.h>
74 #include <linux/sockios.h>
75 #include <linux/net.h>
76 #include <linux/fcntl.h>
77 #include <linux/mm.h>
78 #include <linux/interrupt.h>
79 #include <linux/proc_fs.h>
80 #include <linux/stat.h>
81 #include <linux/init.h>
82 #include <linux/poll.h>
83 #include <linux/netfilter_ipv4.h>
84
85 #include <asm/uaccess.h>
86 #include <asm/system.h>
87
88 #include <linux/smp_lock.h>
89 #include <linux/inet.h>
90 #include <linux/netdevice.h>
91 #include <net/ip.h>
92 #include <net/protocol.h>
93 #include <net/arp.h>
94 #include <net/route.h>
95 #include <net/tcp.h>
96 #include <net/udp.h>
97 #include <linux/skbuff.h>
98 #include <net/sock.h>
99 #include <net/raw.h>
100 #include <net/icmp.h>
101 #include <net/ipip.h>
102 #include <net/inet_common.h>
103 #ifdef CONFIG_IP_MROUTE
104 #include <linux/mroute.h>
105 #endif
106 #include <linux/if_bridge.h>
107 #ifdef CONFIG_KMOD
108 #include <linux/kmod.h>
109 #endif
110 #ifdef CONFIG_NET_DIVERT
111 #include <linux/divert.h>
112 #endif /* CONFIG_NET_DIVERT */
113 #if defined(CONFIG_NET_RADIO) || defined(CONFIG_NET_PCMCIA_RADIO)
114 #include <linux/wireless.h> /* Note : will define WIRELESS_EXT */
115 #endif /* CONFIG_NET_RADIO || CONFIG_NET_PCMCIA_RADIO */
116
117 #define min(a,b) ((a)<(b)?(a):(b))
118
119 struct linux_mib net_statistics[NR_CPUS*2];
120
121 #ifdef INET_REFCNT_DEBUG
122 atomic_t inet_sock_nr;
123 #endif
124
125 extern int raw_get_info(char *, char **, off_t, int);
126 extern int snmp_get_info(char *, char **, off_t, int);
127 extern int netstat_get_info(char *, char **, off_t, int);
128 extern int afinet_get_info(char *, char **, off_t, int);
129 extern int tcp_get_info(char *, char **, off_t, int);
130 extern int udp_get_info(char *, char **, off_t, int);
131 extern void ip_mc_drop_socket(struct sock *sk);
132
133 #ifdef CONFIG_DLCI
134 extern int dlci_ioctl(unsigned int, void*);
135 #endif
136
137 #ifdef CONFIG_DLCI_MODULE
138 int (*dlci_ioctl_hook)(unsigned int, void *) = NULL;
139 #endif
140
141 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
142 int (*br_ioctl_hook)(unsigned long) = NULL;
143 #endif
144
145 /* New destruction routine */
146
147 void inet_sock_destruct(struct sock *sk)
148 {
149 __skb_queue_purge(&sk->receive_queue);
150 __skb_queue_purge(&sk->error_queue);
151
152 if (sk->type == SOCK_STREAM && sk->state != TCP_CLOSE) {
153 printk("Attempt to release TCP socket in state %d %p\n",
154 sk->state,
155 sk);
156 return;
157 }
158 if (!sk->dead) {
159 printk("Attempt to release alive inet socket %p\n", sk);
160 return;
161 }
162
163 BUG_TRAP(atomic_read(&sk->rmem_alloc) == 0);
164 BUG_TRAP(atomic_read(&sk->wmem_alloc) == 0);
165 BUG_TRAP(sk->wmem_queued == 0);
166 BUG_TRAP(sk->forward_alloc == 0);
167
168 if (sk->protinfo.af_inet.opt)
169 kfree(sk->protinfo.af_inet.opt);
170 dst_release(sk->dst_cache);
171 #ifdef INET_REFCNT_DEBUG
172 atomic_dec(&inet_sock_nr);
173 printk(KERN_DEBUG "INET socket %p released, %d are still alive\n", sk, atomic_read(&inet_sock_nr));
174 #endif
175 }
176
177 void inet_sock_release(struct sock *sk)
178 {
179 if (sk->prot->destroy)
180 sk->prot->destroy(sk);
181
182 /* Observation: when inet_sock_release is called, processes have
183 * no access to socket. But net still has.
184 * Step one, detach it from networking:
185 *
186 * A. Remove from hash tables.
187 */
188
189 sk->prot->unhash(sk);
190
191 /* In this point socket cannot receive new packets,
192 * but it is possible that some packets are in flight
193 * because some CPU runs receiver and did hash table lookup
194 * before we unhashed socket. They will achieve receive queue
195 * and will be purged by socket destructor.
196 *
197 * Also we still have packets pending on receive
198 * queue and probably, our own packets waiting in device queues.
199 * sock_destroy will drain receive queue, but transmitted
200 * packets will delay socket destruction until the last reference
201 * will be released.
202 */
203
204 sock_orphan(sk);
205
206 #ifdef INET_REFCNT_DEBUG
207 if (atomic_read(&sk->refcnt) != 1) {
208 printk(KERN_DEBUG "Destruction inet %p delayed, c=%d\n", sk, atomic_read(&sk->refcnt));
209 }
210 #endif
211 sock_put(sk);
212 }
213
214
215 /*
216 * The routines beyond this point handle the behaviour of an AF_INET
217 * socket object. Mostly it punts to the subprotocols of IP to do
218 * the work.
219 */
220
221
222 /*
223 * Set socket options on an inet socket.
224 */
225
226 int inet_setsockopt(struct socket *sock, int level, int optname,
227 char *optval, int optlen)
228 {
229 struct sock *sk=sock->sk;
230
231 return sk->prot->setsockopt(sk,level,optname,optval,optlen);
232 }
233
234 /*
235 * Get a socket option on an AF_INET socket.
236 *
237 * FIX: POSIX 1003.1g is very ambiguous here. It states that
238 * asynchronous errors should be reported by getsockopt. We assume
239 * this means if you specify SO_ERROR (otherwise whats the point of it).
240 */
241
242 int inet_getsockopt(struct socket *sock, int level, int optname,
243 char *optval, int *optlen)
244 {
245 struct sock *sk=sock->sk;
246
247 return sk->prot->getsockopt(sk,level,optname,optval,optlen);
248 }
249
250 /*
251 * Automatically bind an unbound socket.
252 */
253
254 static int inet_autobind(struct sock *sk)
255 {
256 /* We may need to bind the socket. */
257 lock_sock(sk);
258 if (sk->num == 0) {
259 if (sk->prot->get_port(sk, 0) != 0) {
260 release_sock(sk);
261 return -EAGAIN;
262 }
263 sk->sport = htons(sk->num);
264 }
265 release_sock(sk);
266 return 0;
267 }
268
269 /*
270 * Move a socket into listening state.
271 */
272
273 int inet_listen(struct socket *sock, int backlog)
274 {
275 struct sock *sk = sock->sk;
276 unsigned char old_state;
277 int err;
278
279 lock_sock(sk);
280
281 err = -EINVAL;
282 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM)
283 goto out;
284
285 old_state = sk->state;
286 if (!((1<<old_state)&(TCPF_CLOSE|TCPF_LISTEN)))
287 goto out;
288
289 /* Really, if the socket is already in listen state
290 * we can only allow the backlog to be adjusted.
291 */
292 if (old_state != TCP_LISTEN) {
293 err = tcp_listen_start(sk);
294 if (err)
295 goto out;
296 }
297 sk->max_ack_backlog = backlog;
298 err = 0;
299
300 out:
301 release_sock(sk);
302 return err;
303 }
304
305 /*
306 * Create an inet socket.
307 */
308
309 static int inet_create(struct socket *sock, int protocol)
310 {
311 struct sock *sk;
312 struct proto *prot;
313
314 sock->state = SS_UNCONNECTED;
315 sk = sk_alloc(PF_INET, GFP_KERNEL, 1);
316 if (sk == NULL)
317 goto do_oom;
318
319 switch (sock->type) {
320 case SOCK_STREAM:
321 if (protocol && protocol != IPPROTO_TCP)
322 goto free_and_noproto;
323 protocol = IPPROTO_TCP;
324 prot = &tcp_prot;
325 sock->ops = &inet_stream_ops;
326 break;
327 case SOCK_SEQPACKET:
328 goto free_and_badtype;
329 case SOCK_DGRAM:
330 if (protocol && protocol != IPPROTO_UDP)
331 goto free_and_noproto;
332 protocol = IPPROTO_UDP;
333 sk->no_check = UDP_CSUM_DEFAULT;
334 prot=&udp_prot;
335 sock->ops = &inet_dgram_ops;
336 break;
337 case SOCK_RAW:
338 if (!capable(CAP_NET_RAW))
339 goto free_and_badperm;
340 if (!protocol)
341 goto free_and_noproto;
342 prot = &raw_prot;
343 sk->reuse = 1;
344 sk->num = protocol;
345 sock->ops = &inet_dgram_ops;
346 if (protocol == IPPROTO_RAW)
347 sk->protinfo.af_inet.hdrincl = 1;
348 break;
349 default:
350 goto free_and_badtype;
351 }
352
353 if (ipv4_config.no_pmtu_disc)
354 sk->protinfo.af_inet.pmtudisc = IP_PMTUDISC_DONT;
355 else
356 sk->protinfo.af_inet.pmtudisc = IP_PMTUDISC_WANT;
357
358 sock_init_data(sock,sk);
359
360 sk->destruct = inet_sock_destruct;
361
362 sk->zapped = 0;
363 sk->family = PF_INET;
364 sk->protocol = protocol;
365
366 sk->prot = prot;
367 sk->backlog_rcv = prot->backlog_rcv;
368
369 sk->protinfo.af_inet.ttl=sysctl_ip_default_ttl;
370
371 sk->protinfo.af_inet.mc_loop=1;
372 sk->protinfo.af_inet.mc_ttl=1;
373 sk->protinfo.af_inet.mc_index=0;
374 sk->protinfo.af_inet.mc_list=NULL;
375
376 #ifdef INET_REFCNT_DEBUG
377 atomic_inc(&inet_sock_nr);
378 #endif
379
380 if (sk->num) {
381 /* It assumes that any protocol which allows
382 * the user to assign a number at socket
383 * creation time automatically
384 * shares.
385 */
386 sk->sport = htons(sk->num);
387
388 /* Add to protocol hash chains. */
389 sk->prot->hash(sk);
390 }
391
392 if (sk->prot->init) {
393 int err = sk->prot->init(sk);
394 if (err != 0) {
395 inet_sock_release(sk);
396 return(err);
397 }
398 }
399 return(0);
400
401 free_and_badtype:
402 sk_free(sk);
403 return -ESOCKTNOSUPPORT;
404
405 free_and_badperm:
406 sk_free(sk);
407 return -EPERM;
408
409 free_and_noproto:
410 sk_free(sk);
411 return -EPROTONOSUPPORT;
412
413 do_oom:
414 return -ENOBUFS;
415 }
416
417
418 /*
419 * The peer socket should always be NULL (or else). When we call this
420 * function we are destroying the object and from then on nobody
421 * should refer to it.
422 */
423
424 int inet_release(struct socket *sock)
425 {
426 struct sock *sk = sock->sk;
427
428 if (sk) {
429 long timeout;
430
431 /* Applications forget to leave groups before exiting */
432 ip_mc_drop_socket(sk);
433
434 /* If linger is set, we don't return until the close
435 * is complete. Otherwise we return immediately. The
436 * actually closing is done the same either way.
437 *
438 * If the close is due to the process exiting, we never
439 * linger..
440 */
441 timeout = 0;
442 if (sk->linger && !(current->flags & PF_EXITING))
443 timeout = sk->lingertime;
444 sock->sk = NULL;
445 sk->prot->close(sk, timeout);
446 }
447 return(0);
448 }
449
450 /* It is off by default, see below. */
451 int sysctl_ip_nonlocal_bind;
452
453 static int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len)
454 {
455 struct sockaddr_in *addr=(struct sockaddr_in *)uaddr;
456 struct sock *sk=sock->sk;
457 unsigned short snum;
458 int chk_addr_ret;
459 int err;
460
461 /* If the socket has its own bind function then use it. (RAW) */
462 if(sk->prot->bind)
463 return sk->prot->bind(sk, uaddr, addr_len);
464
465 if (addr_len < sizeof(struct sockaddr_in))
466 return -EINVAL;
467
468 chk_addr_ret = inet_addr_type(addr->sin_addr.s_addr);
469
470 /* Not specified by any standard per-se, however it breaks too
471 * many applications when removed. It is unfortunate since
472 * allowing applications to make a non-local bind solves
473 * several problems with systems using dynamic addressing.
474 * (ie. your servers still start up even if your ISDN link
475 * is temporarily down)
476 */
477 if (sysctl_ip_nonlocal_bind == 0 &&
478 sk->protinfo.af_inet.freebind == 0 &&
479 addr->sin_addr.s_addr != INADDR_ANY &&
480 chk_addr_ret != RTN_LOCAL &&
481 chk_addr_ret != RTN_MULTICAST &&
482 chk_addr_ret != RTN_BROADCAST)
483 return -EADDRNOTAVAIL;
484
485 snum = ntohs(addr->sin_port);
486 if (snum && snum < PROT_SOCK && !capable(CAP_NET_BIND_SERVICE))
487 return -EACCES;
488
489 /* We keep a pair of addresses. rcv_saddr is the one
490 * used by hash lookups, and saddr is used for transmit.
491 *
492 * In the BSD API these are the same except where it
493 * would be illegal to use them (multicast/broadcast) in
494 * which case the sending device address is used.
495 */
496 lock_sock(sk);
497
498 /* Check these errors (active socket, double bind). */
499 err = -EINVAL;
500 if ((sk->state != TCP_CLOSE) ||
501 (sk->num != 0))
502 goto out;
503
504 sk->rcv_saddr = sk->saddr = addr->sin_addr.s_addr;
505 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST)
506 sk->saddr = 0; /* Use device */
507
508 /* Make sure we are allowed to bind here. */
509 if (sk->prot->get_port(sk, snum) != 0) {
510 sk->saddr = sk->rcv_saddr = 0;
511 err = -EADDRINUSE;
512 goto out;
513 }
514
515 if (sk->rcv_saddr)
516 sk->userlocks |= SOCK_BINDADDR_LOCK;
517 if (snum)
518 sk->userlocks |= SOCK_BINDPORT_LOCK;
519 sk->sport = htons(sk->num);
520 sk->daddr = 0;
521 sk->dport = 0;
522 sk_dst_reset(sk);
523 err = 0;
524 out:
525 release_sock(sk);
526 return err;
527 }
528
529 int inet_dgram_connect(struct socket *sock, struct sockaddr * uaddr,
530 int addr_len, int flags)
531 {
532 struct sock *sk=sock->sk;
533
534 if (uaddr->sa_family == AF_UNSPEC)
535 return sk->prot->disconnect(sk, flags);
536
537 if (sk->num==0 && inet_autobind(sk) != 0)
538 return -EAGAIN;
539 return sk->prot->connect(sk, (struct sockaddr *)uaddr, addr_len);
540 }
541
542 static long inet_wait_for_connect(struct sock *sk, long timeo)
543 {
544 DECLARE_WAITQUEUE(wait, current);
545
546 __set_current_state(TASK_INTERRUPTIBLE);
547 add_wait_queue(sk->sleep, &wait);
548
549 /* Basic assumption: if someone sets sk->err, he _must_
550 * change state of the socket from TCP_SYN_*.
551 * Connect() does not allow to get error notifications
552 * without closing the socket.
553 */
554 while ((1<<sk->state)&(TCPF_SYN_SENT|TCPF_SYN_RECV)) {
555 release_sock(sk);
556 timeo = schedule_timeout(timeo);
557 lock_sock(sk);
558 if (signal_pending(current) || !timeo)
559 break;
560 set_current_state(TASK_INTERRUPTIBLE);
561 }
562 __set_current_state(TASK_RUNNING);
563 remove_wait_queue(sk->sleep, &wait);
564 return timeo;
565 }
566
567 /*
568 * Connect to a remote host. There is regrettably still a little
569 * TCP 'magic' in here.
570 */
571
572 int inet_stream_connect(struct socket *sock, struct sockaddr * uaddr,
573 int addr_len, int flags)
574 {
575 struct sock *sk=sock->sk;
576 int err;
577 long timeo;
578
579 lock_sock(sk);
580
581 if (uaddr->sa_family == AF_UNSPEC) {
582 err = sk->prot->disconnect(sk, flags);
583 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
584 goto out;
585 }
586
587 switch (sock->state) {
588 default:
589 err = -EINVAL;
590 goto out;
591 case SS_CONNECTED:
592 err = -EISCONN;
593 goto out;
594 case SS_CONNECTING:
595 err = -EALREADY;
596 /* Fall out of switch with err, set for this state */
597 break;
598 case SS_UNCONNECTED:
599 err = -EISCONN;
600 if (sk->state != TCP_CLOSE)
601 goto out;
602
603 err = -EAGAIN;
604 if (sk->num == 0) {
605 if (sk->prot->get_port(sk, 0) != 0)
606 goto out;
607 sk->sport = htons(sk->num);
608 }
609
610 err = sk->prot->connect(sk, uaddr, addr_len);
611 if (err < 0)
612 goto out;
613
614 sock->state = SS_CONNECTING;
615
616 /* Just entered SS_CONNECTING state; the only
617 * difference is that return value in non-blocking
618 * case is EINPROGRESS, rather than EALREADY.
619 */
620 err = -EINPROGRESS;
621 break;
622 }
623
624 timeo = sock_sndtimeo(sk, flags&O_NONBLOCK);
625
626 if ((1<<sk->state)&(TCPF_SYN_SENT|TCPF_SYN_RECV)) {
627 /* Error code is set above */
628 if (!timeo || !inet_wait_for_connect(sk, timeo))
629 goto out;
630
631 err = sock_intr_errno(timeo);
632 if (signal_pending(current))
633 goto out;
634 }
635
636 /* Connection was closed by RST, timeout, ICMP error
637 * or another process disconnected us.
638 */
639 if (sk->state == TCP_CLOSE)
640 goto sock_error;
641
642 /* sk->err may be not zero now, if RECVERR was ordered by user
643 * and error was received after socket entered established state.
644 * Hence, it is handled normally after connect() return successfully.
645 */
646
647 sock->state = SS_CONNECTED;
648 err = 0;
649 out:
650 release_sock(sk);
651 return err;
652
653 sock_error:
654 err = sock_error(sk) ? : -ECONNABORTED;
655 sock->state = SS_UNCONNECTED;
656 if (sk->prot->disconnect(sk, flags))
657 sock->state = SS_DISCONNECTING;
658 goto out;
659 }
660
661 /*
662 * Accept a pending connection. The TCP layer now gives BSD semantics.
663 */
664
665 int inet_accept(struct socket *sock, struct socket *newsock, int flags)
666 {
667 struct sock *sk1 = sock->sk;
668 struct sock *sk2;
669 int err = -EINVAL;
670
671 if((sk2 = sk1->prot->accept(sk1,flags,&err)) == NULL)
672 goto do_err;
673
674 lock_sock(sk2);
675
676 BUG_TRAP((1<<sk2->state)&(TCPF_ESTABLISHED|TCPF_CLOSE_WAIT|TCPF_CLOSE));
677
678 sock_graft(sk2, newsock);
679
680 newsock->state = SS_CONNECTED;
681 release_sock(sk2);
682 return 0;
683
684 do_err:
685 return err;
686 }
687
688
689 /*
690 * This does both peername and sockname.
691 */
692
693 static int inet_getname(struct socket *sock, struct sockaddr *uaddr,
694 int *uaddr_len, int peer)
695 {
696 struct sock *sk = sock->sk;
697 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
698
699 sin->sin_family = AF_INET;
700 if (peer) {
701 if (!sk->dport)
702 return -ENOTCONN;
703 if (((1<<sk->state)&(TCPF_CLOSE|TCPF_SYN_SENT)) && peer == 1)
704 return -ENOTCONN;
705 sin->sin_port = sk->dport;
706 sin->sin_addr.s_addr = sk->daddr;
707 } else {
708 __u32 addr = sk->rcv_saddr;
709 if (!addr)
710 addr = sk->saddr;
711 sin->sin_port = sk->sport;
712 sin->sin_addr.s_addr = addr;
713 }
714 *uaddr_len = sizeof(*sin);
715 return(0);
716 }
717
718
719
720 int inet_recvmsg(struct socket *sock, struct msghdr *msg, int size,
721 int flags, struct scm_cookie *scm)
722 {
723 struct sock *sk = sock->sk;
724 int addr_len = 0;
725 int err;
726
727 err = sk->prot->recvmsg(sk, msg, size, flags&MSG_DONTWAIT,
728 flags&~MSG_DONTWAIT, &addr_len);
729 if (err >= 0)
730 msg->msg_namelen = addr_len;
731 return err;
732 }
733
734
735 int inet_sendmsg(struct socket *sock, struct msghdr *msg, int size,
736 struct scm_cookie *scm)
737 {
738 struct sock *sk = sock->sk;
739
740 /* We may need to bind the socket. */
741 if (sk->num==0 && inet_autobind(sk) != 0)
742 return -EAGAIN;
743
744 return sk->prot->sendmsg(sk, msg, size);
745 }
746
747 int inet_shutdown(struct socket *sock, int how)
748 {
749 struct sock *sk = sock->sk;
750 int err = 0;
751
752 /* This should really check to make sure
753 * the socket is a TCP socket. (WHY AC...)
754 */
755 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and
756 1->2 bit 2 snds.
757 2->3 */
758 if ((how & ~SHUTDOWN_MASK) || how==0) /* MAXINT->0 */
759 return -EINVAL;
760
761 lock_sock(sk);
762 if (sock->state == SS_CONNECTING) {
763 if ((1<<sk->state)&(TCPF_SYN_SENT|TCPF_SYN_RECV|TCPF_CLOSE))
764 sock->state = SS_DISCONNECTING;
765 else
766 sock->state = SS_CONNECTED;
767 }
768
769 switch (sk->state) {
770 case TCP_CLOSE:
771 err = -ENOTCONN;
772 /* Hack to wake up other listeners, who can poll for
773 POLLHUP, even on eg. unconnected UDP sockets -- RR */
774 default:
775 sk->shutdown |= how;
776 if (sk->prot->shutdown)
777 sk->prot->shutdown(sk, how);
778 break;
779
780 /* Remaining two branches are temporary solution for missing
781 * close() in multithreaded environment. It is _not_ a good idea,
782 * but we have no choice until close() is repaired at VFS level.
783 */
784 case TCP_LISTEN:
785 if (!(how & RCV_SHUTDOWN))
786 break;
787 /* Fall through */
788 case TCP_SYN_SENT:
789 err = sk->prot->disconnect(sk, O_NONBLOCK);
790 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED;
791 break;
792 }
793
794 /* Wake up anyone sleeping in poll. */
795 sk->state_change(sk);
796 release_sock(sk);
797 return err;
798 }
799
800 /*
801 * ioctl() calls you can issue on an INET socket. Most of these are
802 * device configuration and stuff and very rarely used. Some ioctls
803 * pass on to the socket itself.
804 *
805 * NOTE: I like the idea of a module for the config stuff. ie ifconfig
806 * loads the devconfigure module does its configuring and unloads it.
807 * There's a good 20K of config code hanging around the kernel.
808 */
809
810 static int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
811 {
812 struct sock *sk = sock->sk;
813 int err;
814 int pid;
815
816 switch(cmd)
817 {
818 case FIOSETOWN:
819 case SIOCSPGRP:
820 err = get_user(pid, (int *) arg);
821 if (err)
822 return err;
823 if (current->pid != pid && current->pgrp != -pid &&
824 !capable(CAP_NET_ADMIN))
825 return -EPERM;
826 sk->proc = pid;
827 return(0);
828 case FIOGETOWN:
829 case SIOCGPGRP:
830 return put_user(sk->proc, (int *)arg);
831 case SIOCGSTAMP:
832 if(sk->stamp.tv_sec==0)
833 return -ENOENT;
834 err = copy_to_user((void *)arg,&sk->stamp,sizeof(struct timeval));
835 if (err)
836 err = -EFAULT;
837 return err;
838 case SIOCADDRT:
839 case SIOCDELRT:
840 case SIOCRTMSG:
841 return(ip_rt_ioctl(cmd,(void *) arg));
842 case SIOCDARP:
843 case SIOCGARP:
844 case SIOCSARP:
845 return(arp_ioctl(cmd,(void *) arg));
846 case SIOCGIFADDR:
847 case SIOCSIFADDR:
848 case SIOCGIFBRDADDR:
849 case SIOCSIFBRDADDR:
850 case SIOCGIFNETMASK:
851 case SIOCSIFNETMASK:
852 case SIOCGIFDSTADDR:
853 case SIOCSIFDSTADDR:
854 case SIOCSIFPFLAGS:
855 case SIOCGIFPFLAGS:
856 case SIOCSIFFLAGS:
857 return(devinet_ioctl(cmd,(void *) arg));
858 case SIOCGIFBR:
859 case SIOCSIFBR:
860 #if defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)
861 #ifdef CONFIG_KMOD
862 if (br_ioctl_hook == NULL)
863 request_module("bridge");
864 #endif
865 if (br_ioctl_hook != NULL)
866 return br_ioctl_hook(arg);
867 #endif
868 case SIOCGIFDIVERT:
869 case SIOCSIFDIVERT:
870 #ifdef CONFIG_NET_DIVERT
871 return(divert_ioctl(cmd, (struct divert_cf *) arg));
872 #else
873 return -ENOPKG;
874 #endif /* CONFIG_NET_DIVERT */
875 return -ENOPKG;
876
877 case SIOCADDDLCI:
878 case SIOCDELDLCI:
879 #ifdef CONFIG_DLCI
880 lock_kernel();
881 err = dlci_ioctl(cmd, (void *) arg);
882 unlock_kernel();
883 return err;
884 #endif
885
886 #ifdef CONFIG_DLCI_MODULE
887
888 #ifdef CONFIG_KMOD
889 if (dlci_ioctl_hook == NULL)
890 request_module("dlci");
891 #endif
892
893 if (dlci_ioctl_hook) {
894 lock_kernel();
895 err = (*dlci_ioctl_hook)(cmd, (void *) arg);
896 unlock_kernel();
897 return err;
898 }
899 #endif
900 return -ENOPKG;
901
902 default:
903 if ((cmd >= SIOCDEVPRIVATE) &&
904 (cmd <= (SIOCDEVPRIVATE + 15)))
905 return(dev_ioctl(cmd,(void *) arg));
906
907 #ifdef WIRELESS_EXT
908 if((cmd >= SIOCIWFIRST) && (cmd <= SIOCIWLAST))
909 return(dev_ioctl(cmd,(void *) arg));
910 #endif /* WIRELESS_EXT */
911
912 if (sk->prot->ioctl==NULL || (err=sk->prot->ioctl(sk, cmd, arg))==-ENOIOCTLCMD)
913 return(dev_ioctl(cmd,(void *) arg));
914 return err;
915 }
916 /*NOTREACHED*/
917 return(0);
918 }
919
920 struct proto_ops inet_stream_ops = {
921 family: PF_INET,
922
923 release: inet_release,
924 bind: inet_bind,
925 connect: inet_stream_connect,
926 socketpair: sock_no_socketpair,
927 accept: inet_accept,
928 getname: inet_getname,
929 poll: tcp_poll,
930 ioctl: inet_ioctl,
931 listen: inet_listen,
932 shutdown: inet_shutdown,
933 setsockopt: inet_setsockopt,
934 getsockopt: inet_getsockopt,
935 sendmsg: inet_sendmsg,
936 recvmsg: inet_recvmsg,
937 mmap: sock_no_mmap
938 };
939
940 struct proto_ops inet_dgram_ops = {
941 family: PF_INET,
942
943 release: inet_release,
944 bind: inet_bind,
945 connect: inet_dgram_connect,
946 socketpair: sock_no_socketpair,
947 accept: sock_no_accept,
948 getname: inet_getname,
949 poll: datagram_poll,
950 ioctl: inet_ioctl,
951 listen: sock_no_listen,
952 shutdown: inet_shutdown,
953 setsockopt: inet_setsockopt,
954 getsockopt: inet_getsockopt,
955 sendmsg: inet_sendmsg,
956 recvmsg: inet_recvmsg,
957 mmap: sock_no_mmap,
958 };
959
960 struct net_proto_family inet_family_ops = {
961 PF_INET,
962 inet_create
963 };
964
965
966 extern void tcp_init(void);
967 extern void tcp_v4_init(struct net_proto_family *);
968
969
970 /*
971 * Called by socket.c on kernel startup.
972 */
973
974 static int __init inet_init(void)
975 {
976 struct sk_buff *dummy_skb;
977 struct inet_protocol *p;
978
979 printk(KERN_INFO "NET4: Linux TCP/IP 1.0 for NET4.0\n");
980
981 if (sizeof(struct inet_skb_parm) > sizeof(dummy_skb->cb))
982 {
983 printk(KERN_CRIT "inet_proto_init: panic\n");
984 return -EINVAL;
985 }
986
987 /*
988 * Tell SOCKET that we are alive...
989 */
990
991 (void) sock_register(&inet_family_ops);
992
993 /*
994 * Add all the protocols.
995 */
996
997 printk(KERN_INFO "IP Protocols: ");
998 for(p = inet_protocol_base; p != NULL;)
999 {
1000 struct inet_protocol *tmp = (struct inet_protocol *) p->next;
1001 inet_add_protocol(p);
1002 printk("%s%s",p->name,tmp?", ":"\n");
1003 p = tmp;
1004 }
1005
1006 /*
1007 * Set the ARP module up
1008 */
1009
1010 arp_init();
1011
1012 /*
1013 * Set the IP module up
1014 */
1015
1016 ip_init();
1017
1018 tcp_v4_init(&inet_family_ops);
1019
1020 /* Setup TCP slab cache for open requests. */
1021 tcp_init();
1022
1023
1024 /*
1025 * Set the ICMP layer up
1026 */
1027
1028 icmp_init(&inet_family_ops);
1029
1030 /* I wish inet_add_protocol had no constructor hook...
1031 I had to move IPIP from net/ipv4/protocol.c :-( --ANK
1032 */
1033 #ifdef CONFIG_NET_IPIP
1034 ipip_init();
1035 #endif
1036 #ifdef CONFIG_NET_IPGRE
1037 ipgre_init();
1038 #endif
1039
1040 /*
1041 * Initialise the multicast router
1042 */
1043 #if defined(CONFIG_IP_MROUTE)
1044 ip_mr_init();
1045 #endif
1046
1047 /*
1048 * Create all the /proc entries.
1049 */
1050 #ifdef CONFIG_PROC_FS
1051 proc_net_create ("raw", 0, raw_get_info);
1052 proc_net_create ("netstat", 0, netstat_get_info);
1053 proc_net_create ("snmp", 0, snmp_get_info);
1054 proc_net_create ("sockstat", 0, afinet_get_info);
1055 proc_net_create ("tcp", 0, tcp_get_info);
1056 proc_net_create ("udp", 0, udp_get_info);
1057 #endif /* CONFIG_PROC_FS */
1058 return 0;
1059 }
1060 module_init(inet_init);
1061
This page was automatically generated by the
LXR engine.
Visit the LXR main site for more
information.