?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 2.6.19-rc6
[cris-mirror.git] / net / core / sock.c
blobee6cd2541d35765642bcc10cb72ff52f5f2d97b7
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 * Generic socket support routines. Memory allocators, socket lock/release
7 * handler for protocols to use and generic option handler.
8 *
9 *
10 * Version: $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
11 *
12 * Authors: Ross Biro
13 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
14 * Florian La Roche, <flla@stud.uni-sb.de>
15 * Alan Cox, <A.Cox@swansea.ac.uk>
16 *
17 * Fixes:
18 * Alan Cox : Numerous verify_area() problems
19 * Alan Cox : Connecting on a connecting socket
20 * now returns an error for tcp.
21 * Alan Cox : sock->protocol is set correctly.
22 * and is not sometimes left as 0.
23 * Alan Cox : connect handles icmp errors on a
24 * connect properly. Unfortunately there
25 * is a restart syscall nasty there. I
26 * can't match BSD without hacking the C
27 * library. Ideas urgently sought!
28 * Alan Cox : Disallow bind() to addresses that are
29 * not ours - especially broadcast ones!!
30 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
31 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
32 * instead they leave that for the DESTROY timer.
33 * Alan Cox : Clean up error flag in accept
34 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
35 * was buggy. Put a remove_sock() in the handler
36 * for memory when we hit 0. Also altered the timer
37 * code. The ACK stuff can wait and needs major
38 * TCP layer surgery.
39 * Alan Cox : Fixed TCP ack bug, removed remove sock
40 * and fixed timer/inet_bh race.
41 * Alan Cox : Added zapped flag for TCP
42 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
43 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
44 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
45 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
46 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
47 * Rick Sladkey : Relaxed UDP rules for matching packets.
48 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
49 * Pauline Middelink : identd support
50 * Alan Cox : Fixed connect() taking signals I think.
51 * Alan Cox : SO_LINGER supported
52 * Alan Cox : Error reporting fixes
53 * Anonymous : inet_create tidied up (sk->reuse setting)
54 * Alan Cox : inet sockets don't set sk->type!
55 * Alan Cox : Split socket option code
56 * Alan Cox : Callbacks
57 * Alan Cox : Nagle flag for Charles & Johannes stuff
58 * Alex : Removed restriction on inet fioctl
59 * Alan Cox : Splitting INET from NET core
60 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
61 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
62 * Alan Cox : Split IP from generic code
63 * Alan Cox : New kfree_skbmem()
64 * Alan Cox : Make SO_DEBUG superuser only.
65 * Alan Cox : Allow anyone to clear SO_DEBUG
66 * (compatibility fix)
67 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
68 * Alan Cox : Allocator for a socket is settable.
69 * Alan Cox : SO_ERROR includes soft errors.
70 * Alan Cox : Allow NULL arguments on some SO_ opts
71 * Alan Cox : Generic socket allocation to make hooks
72 * easier (suggested by Craig Metz).
73 * Michael Pall : SO_ERROR returns positive errno again
74 * Steve Whitehouse: Added default destructor to free
75 * protocol private data.
76 * Steve Whitehouse: Added various other default routines
77 * common to several socket families.
78 * Chris Evans : Call suser() check last on F_SETOWN
79 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
80 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
81 * Andi Kleen : Fix write_space callback
82 * Chris Evans : Security fixes - signedness again
83 * Arnaldo C. Melo : cleanups, use skb_queue_purge
84 *
85 * To Fix:
86 *
87 *
88 * This program is free software; you can redistribute it and/or
89 * modify it under the terms of the GNU General Public License
90 * as published by the Free Software Foundation; either version
91 * 2 of the License, or (at your option) any later version.
92 */
93
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/types.h>
97 #include <linux/socket.h>
98 #include <linux/in.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114
115 #include <asm/uaccess.h>
116 #include <asm/system.h>
117
118 #include <linux/netdevice.h>
119 #include <net/protocol.h>
120 #include <linux/skbuff.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <net/xfrm.h>
124 #include <linux/ipsec.h>
125
126 #include <linux/filter.h>
127
128 #ifdef CONFIG_INET
129 #include <net/tcp.h>
130 #endif
131
132 /*
133 * Each address family might have different locking rules, so we have
134 * one slock key per address family:
135 */
136 static struct lock_class_key af_family_keys[AF_MAX];
137 static struct lock_class_key af_family_slock_keys[AF_MAX];
138
139 #ifdef CONFIG_DEBUG_LOCK_ALLOC
140 /*
141 * Make lock validator output more readable. (we pre-construct these
142 * strings build-time, so that runtime initialization of socket
143 * locks is fast):
144 */
145 static const char *af_family_key_strings[AF_MAX+1] = {
146 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
147 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
148 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
149 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
150 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
151 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
152 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
153 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
154 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
155 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
156 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
157 };
158 static const char *af_family_slock_key_strings[AF_MAX+1] = {
159 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
160 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
161 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
162 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
163 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
164 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
165 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
166 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
167 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
168 "slock-27" , "slock-28" , "slock-29" ,
169 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX"
170 };
171 #endif
172
173 /*
174 * sk_callback_lock locking rules are per-address-family,
175 * so split the lock classes by using a per-AF key:
176 */
177 static struct lock_class_key af_callback_keys[AF_MAX];
178
179 /* Take into consideration the size of the struct sk_buff overhead in the
180 * determination of these values, since that is non-constant across
181 * platforms. This makes socket queueing behavior and performance
182 * not depend upon such differences.
183 */
184 #define _SK_MEM_PACKETS 256
185 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
186 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
187 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
188
189 /* Run time adjustable parameters. */
190 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
191 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
192 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
193 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
194
195 /* Maximal space eaten by iovec or ancilliary data plus some space */
196 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
197
198 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
199 {
200 struct timeval tv;
201
202 if (optlen < sizeof(tv))
203 return -EINVAL;
204 if (copy_from_user(&tv, optval, sizeof(tv)))
205 return -EFAULT;
206
207 *timeo_p = MAX_SCHEDULE_TIMEOUT;
208 if (tv.tv_sec == 0 && tv.tv_usec == 0)
209 return 0;
210 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
211 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
212 return 0;
213 }
214
215 static void sock_warn_obsolete_bsdism(const char *name)
216 {
217 static int warned;
218 static char warncomm[TASK_COMM_LEN];
219 if (strcmp(warncomm, current->comm) && warned < 5) {
220 strcpy(warncomm, current->comm);
221 printk(KERN_WARNING "process `%s' is using obsolete "
222 "%s SO_BSDCOMPAT\n", warncomm, name);
223 warned++;
224 }
225 }
226
227 static void sock_disable_timestamp(struct sock *sk)
228 {
229 if (sock_flag(sk, SOCK_TIMESTAMP)) {
230 sock_reset_flag(sk, SOCK_TIMESTAMP);
231 net_disable_timestamp();
232 }
233 }
234
235
236 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
237 {
238 int err = 0;
239 int skb_len;
240
241 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
242 number of warnings when compiling with -W --ANK
243 */
244 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
245 (unsigned)sk->sk_rcvbuf) {
246 err = -ENOMEM;
247 goto out;
248 }
249
250 err = sk_filter(sk, skb);
251 if (err)
252 goto out;
253
254 skb->dev = NULL;
255 skb_set_owner_r(skb, sk);
256
257 /* Cache the SKB length before we tack it onto the receive
258 * queue. Once it is added it no longer belongs to us and
259 * may be freed by other threads of control pulling packets
260 * from the queue.
261 */
262 skb_len = skb->len;
263
264 skb_queue_tail(&sk->sk_receive_queue, skb);
265
266 if (!sock_flag(sk, SOCK_DEAD))
267 sk->sk_data_ready(sk, skb_len);
268 out:
269 return err;
270 }
271 EXPORT_SYMBOL(sock_queue_rcv_skb);
272
273 int sk_receive_skb(struct sock *sk, struct sk_buff *skb)
274 {
275 int rc = NET_RX_SUCCESS;
276
277 if (sk_filter(sk, skb))
278 goto discard_and_relse;
279
280 skb->dev = NULL;
281
282 bh_lock_sock(sk);
283 if (!sock_owned_by_user(sk)) {
284 /*
285 * trylock + unlock semantics:
286 */
287 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
288
289 rc = sk->sk_backlog_rcv(sk, skb);
290
291 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
292 } else
293 sk_add_backlog(sk, skb);
294 bh_unlock_sock(sk);
295 out:
296 sock_put(sk);
297 return rc;
298 discard_and_relse:
299 kfree_skb(skb);
300 goto out;
301 }
302 EXPORT_SYMBOL(sk_receive_skb);
303
304 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
305 {
306 struct dst_entry *dst = sk->sk_dst_cache;
307
308 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
309 sk->sk_dst_cache = NULL;
310 dst_release(dst);
311 return NULL;
312 }
313
314 return dst;
315 }
316 EXPORT_SYMBOL(__sk_dst_check);
317
318 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
319 {
320 struct dst_entry *dst = sk_dst_get(sk);
321
322 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
323 sk_dst_reset(sk);
324 dst_release(dst);
325 return NULL;
326 }
327
328 return dst;
329 }
330 EXPORT_SYMBOL(sk_dst_check);
331
332 /*
333 * This is meant for all protocols to use and covers goings on
334 * at the socket level. Everything here is generic.
335 */
336
337 int sock_setsockopt(struct socket *sock, int level, int optname,
338 char __user *optval, int optlen)
339 {
340 struct sock *sk=sock->sk;
341 struct sk_filter *filter;
342 int val;
343 int valbool;
344 struct linger ling;
345 int ret = 0;
346
347 /*
348 * Options without arguments
349 */
350
351 #ifdef SO_DONTLINGER /* Compatibility item... */
352 if (optname == SO_DONTLINGER) {
353 lock_sock(sk);
354 sock_reset_flag(sk, SOCK_LINGER);
355 release_sock(sk);
356 return 0;
357 }
358 #endif
359
360 if(optlen<sizeof(int))
361 return(-EINVAL);
362
363 if (get_user(val, (int __user *)optval))
364 return -EFAULT;
365
366 valbool = val?1:0;
367
368 lock_sock(sk);
369
370 switch(optname)
371 {
372 case SO_DEBUG:
373 if(val && !capable(CAP_NET_ADMIN))
374 {
375 ret = -EACCES;
376 }
377 else if (valbool)
378 sock_set_flag(sk, SOCK_DBG);
379 else
380 sock_reset_flag(sk, SOCK_DBG);
381 break;
382 case SO_REUSEADDR:
383 sk->sk_reuse = valbool;
384 break;
385 case SO_TYPE:
386 case SO_ERROR:
387 ret = -ENOPROTOOPT;
388 break;
389 case SO_DONTROUTE:
390 if (valbool)
391 sock_set_flag(sk, SOCK_LOCALROUTE);
392 else
393 sock_reset_flag(sk, SOCK_LOCALROUTE);
394 break;
395 case SO_BROADCAST:
396 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
397 break;
398 case SO_SNDBUF:
399 /* Don't error on this BSD doesn't and if you think
400 about it this is right. Otherwise apps have to
401 play 'guess the biggest size' games. RCVBUF/SNDBUF
402 are treated in BSD as hints */
403
404 if (val > sysctl_wmem_max)
405 val = sysctl_wmem_max;
406 set_sndbuf:
407 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
408 if ((val * 2) < SOCK_MIN_SNDBUF)
409 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
410 else
411 sk->sk_sndbuf = val * 2;
412
413 /*
414 * Wake up sending tasks if we
415 * upped the value.
416 */
417 sk->sk_write_space(sk);
418 break;
419
420 case SO_SNDBUFFORCE:
421 if (!capable(CAP_NET_ADMIN)) {
422 ret = -EPERM;
423 break;
424 }
425 goto set_sndbuf;
426
427 case SO_RCVBUF:
428 /* Don't error on this BSD doesn't and if you think
429 about it this is right. Otherwise apps have to
430 play 'guess the biggest size' games. RCVBUF/SNDBUF
431 are treated in BSD as hints */
432
433 if (val > sysctl_rmem_max)
434 val = sysctl_rmem_max;
435 set_rcvbuf:
436 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
437 /*
438 * We double it on the way in to account for
439 * "struct sk_buff" etc. overhead. Applications
440 * assume that the SO_RCVBUF setting they make will
441 * allow that much actual data to be received on that
442 * socket.
443 *
444 * Applications are unaware that "struct sk_buff" and
445 * other overheads allocate from the receive buffer
446 * during socket buffer allocation.
447 *
448 * And after considering the possible alternatives,
449 * returning the value we actually used in getsockopt
450 * is the most desirable behavior.
451 */
452 if ((val * 2) < SOCK_MIN_RCVBUF)
453 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
454 else
455 sk->sk_rcvbuf = val * 2;
456 break;
457
458 case SO_RCVBUFFORCE:
459 if (!capable(CAP_NET_ADMIN)) {
460 ret = -EPERM;
461 break;
462 }
463 goto set_rcvbuf;
464
465 case SO_KEEPALIVE:
466 #ifdef CONFIG_INET
467 if (sk->sk_protocol == IPPROTO_TCP)
468 tcp_set_keepalive(sk, valbool);
469 #endif
470 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
471 break;
472
473 case SO_OOBINLINE:
474 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
475 break;
476
477 case SO_NO_CHECK:
478 sk->sk_no_check = valbool;
479 break;
480
481 case SO_PRIORITY:
482 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
483 sk->sk_priority = val;
484 else
485 ret = -EPERM;
486 break;
487
488 case SO_LINGER:
489 if(optlen<sizeof(ling)) {
490 ret = -EINVAL; /* 1003.1g */
491 break;
492 }
493 if (copy_from_user(&ling,optval,sizeof(ling))) {
494 ret = -EFAULT;
495 break;
496 }
497 if (!ling.l_onoff)
498 sock_reset_flag(sk, SOCK_LINGER);
499 else {
500 #if (BITS_PER_LONG == 32)
501 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
502 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
503 else
504 #endif
505 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
506 sock_set_flag(sk, SOCK_LINGER);
507 }
508 break;
509
510 case SO_BSDCOMPAT:
511 sock_warn_obsolete_bsdism("setsockopt");
512 break;
513
514 case SO_PASSCRED:
515 if (valbool)
516 set_bit(SOCK_PASSCRED, &sock->flags);
517 else
518 clear_bit(SOCK_PASSCRED, &sock->flags);
519 break;
520
521 case SO_TIMESTAMP:
522 if (valbool) {
523 sock_set_flag(sk, SOCK_RCVTSTAMP);
524 sock_enable_timestamp(sk);
525 } else
526 sock_reset_flag(sk, SOCK_RCVTSTAMP);
527 break;
528
529 case SO_RCVLOWAT:
530 if (val < 0)
531 val = INT_MAX;
532 sk->sk_rcvlowat = val ? : 1;
533 break;
534
535 case SO_RCVTIMEO:
536 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
537 break;
538
539 case SO_SNDTIMEO:
540 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
541 break;
542
543 #ifdef CONFIG_NETDEVICES
544 case SO_BINDTODEVICE:
545 {
546 char devname[IFNAMSIZ];
547
548 /* Sorry... */
549 if (!capable(CAP_NET_RAW)) {
550 ret = -EPERM;
551 break;
552 }
553
554 /* Bind this socket to a particular device like "eth0",
555 * as specified in the passed interface name. If the
556 * name is "" or the option length is zero the socket
557 * is not bound.
558 */
559
560 if (!valbool) {
561 sk->sk_bound_dev_if = 0;
562 } else {
563 if (optlen > IFNAMSIZ - 1)
564 optlen = IFNAMSIZ - 1;
565 memset(devname, 0, sizeof(devname));
566 if (copy_from_user(devname, optval, optlen)) {
567 ret = -EFAULT;
568 break;
569 }
570
571 /* Remove any cached route for this socket. */
572 sk_dst_reset(sk);
573
574 if (devname[0] == '\0') {
575 sk->sk_bound_dev_if = 0;
576 } else {
577 struct net_device *dev = dev_get_by_name(devname);
578 if (!dev) {
579 ret = -ENODEV;
580 break;
581 }
582 sk->sk_bound_dev_if = dev->ifindex;
583 dev_put(dev);
584 }
585 }
586 break;
587 }
588 #endif
589
590
591 case SO_ATTACH_FILTER:
592 ret = -EINVAL;
593 if (optlen == sizeof(struct sock_fprog)) {
594 struct sock_fprog fprog;
595
596 ret = -EFAULT;
597 if (copy_from_user(&fprog, optval, sizeof(fprog)))
598 break;
599
600 ret = sk_attach_filter(&fprog, sk);
601 }
602 break;
603
604 case SO_DETACH_FILTER:
605 rcu_read_lock_bh();
606 filter = rcu_dereference(sk->sk_filter);
607 if (filter) {
608 rcu_assign_pointer(sk->sk_filter, NULL);
609 sk_filter_release(sk, filter);
610 rcu_read_unlock_bh();
611 break;
612 }
613 rcu_read_unlock_bh();
614 ret = -ENONET;
615 break;
616
617 case SO_PASSSEC:
618 if (valbool)
619 set_bit(SOCK_PASSSEC, &sock->flags);
620 else
621 clear_bit(SOCK_PASSSEC, &sock->flags);
622 break;
623
624 /* We implement the SO_SNDLOWAT etc to
625 not be settable (1003.1g 5.3) */
626 default:
627 ret = -ENOPROTOOPT;
628 break;
629 }
630 release_sock(sk);
631 return ret;
632 }
633
634
635 int sock_getsockopt(struct socket *sock, int level, int optname,
636 char __user *optval, int __user *optlen)
637 {
638 struct sock *sk = sock->sk;
639
640 union
641 {
642 int val;
643 struct linger ling;
644 struct timeval tm;
645 } v;
646
647 unsigned int lv = sizeof(int);
648 int len;
649
650 if(get_user(len,optlen))
651 return -EFAULT;
652 if(len < 0)
653 return -EINVAL;
654
655 switch(optname)
656 {
657 case SO_DEBUG:
658 v.val = sock_flag(sk, SOCK_DBG);
659 break;
660
661 case SO_DONTROUTE:
662 v.val = sock_flag(sk, SOCK_LOCALROUTE);
663 break;
664
665 case SO_BROADCAST:
666 v.val = !!sock_flag(sk, SOCK_BROADCAST);
667 break;
668
669 case SO_SNDBUF:
670 v.val = sk->sk_sndbuf;
671 break;
672
673 case SO_RCVBUF:
674 v.val = sk->sk_rcvbuf;
675 break;
676
677 case SO_REUSEADDR:
678 v.val = sk->sk_reuse;
679 break;
680
681 case SO_KEEPALIVE:
682 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
683 break;
684
685 case SO_TYPE:
686 v.val = sk->sk_type;
687 break;
688
689 case SO_ERROR:
690 v.val = -sock_error(sk);
691 if(v.val==0)
692 v.val = xchg(&sk->sk_err_soft, 0);
693 break;
694
695 case SO_OOBINLINE:
696 v.val = !!sock_flag(sk, SOCK_URGINLINE);
697 break;
698
699 case SO_NO_CHECK:
700 v.val = sk->sk_no_check;
701 break;
702
703 case SO_PRIORITY:
704 v.val = sk->sk_priority;
705 break;
706
707 case SO_LINGER:
708 lv = sizeof(v.ling);
709 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
710 v.ling.l_linger = sk->sk_lingertime / HZ;
711 break;
712
713 case SO_BSDCOMPAT:
714 sock_warn_obsolete_bsdism("getsockopt");
715 break;
716
717 case SO_TIMESTAMP:
718 v.val = sock_flag(sk, SOCK_RCVTSTAMP);
719 break;
720
721 case SO_RCVTIMEO:
722 lv=sizeof(struct timeval);
723 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
724 v.tm.tv_sec = 0;
725 v.tm.tv_usec = 0;
726 } else {
727 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
728 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
729 }
730 break;
731
732 case SO_SNDTIMEO:
733 lv=sizeof(struct timeval);
734 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
735 v.tm.tv_sec = 0;
736 v.tm.tv_usec = 0;
737 } else {
738 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
739 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
740 }
741 break;
742
743 case SO_RCVLOWAT:
744 v.val = sk->sk_rcvlowat;
745 break;
746
747 case SO_SNDLOWAT:
748 v.val=1;
749 break;
750
751 case SO_PASSCRED:
752 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
753 break;
754
755 case SO_PEERCRED:
756 if (len > sizeof(sk->sk_peercred))
757 len = sizeof(sk->sk_peercred);
758 if (copy_to_user(optval, &sk->sk_peercred, len))
759 return -EFAULT;
760 goto lenout;
761
762 case SO_PEERNAME:
763 {
764 char address[128];
765
766 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
767 return -ENOTCONN;
768 if (lv < len)
769 return -EINVAL;
770 if (copy_to_user(optval, address, len))
771 return -EFAULT;
772 goto lenout;
773 }
774
775 /* Dubious BSD thing... Probably nobody even uses it, but
776 * the UNIX standard wants it for whatever reason... -DaveM
777 */
778 case SO_ACCEPTCONN:
779 v.val = sk->sk_state == TCP_LISTEN;
780 break;
781
782 case SO_PASSSEC:
783 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
784 break;
785
786 case SO_PEERSEC:
787 return security_socket_getpeersec_stream(sock, optval, optlen, len);
788
789 default:
790 return(-ENOPROTOOPT);
791 }
792 if (len > lv)
793 len = lv;
794 if (copy_to_user(optval, &v, len))
795 return -EFAULT;
796 lenout:
797 if (put_user(len, optlen))
798 return -EFAULT;
799 return 0;
800 }
801
802 /*
803 * Initialize an sk_lock.
804 *
805 * (We also register the sk_lock with the lock validator.)
806 */
807 static void inline sock_lock_init(struct sock *sk)
808 {
809 spin_lock_init(&sk->sk_lock.slock);
810 sk->sk_lock.owner = NULL;
811 init_waitqueue_head(&sk->sk_lock.wq);
812 /*
813 * Make sure we are not reinitializing a held lock:
814 */
815 debug_check_no_locks_freed((void *)&sk->sk_lock, sizeof(sk->sk_lock));
816
817 /*
818 * Mark both the sk_lock and the sk_lock.slock as a
819 * per-address-family lock class:
820 */
821 lockdep_set_class_and_name(&sk->sk_lock.slock,
822 af_family_slock_keys + sk->sk_family,
823 af_family_slock_key_strings[sk->sk_family]);
824 lockdep_init_map(&sk->sk_lock.dep_map,
825 af_family_key_strings[sk->sk_family],
826 af_family_keys + sk->sk_family, 0);
827 }
828
829 /**
830 * sk_alloc - All socket objects are allocated here
831 * @family: protocol family
832 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
833 * @prot: struct proto associated with this new sock instance
834 * @zero_it: if we should zero the newly allocated sock
835 */
836 struct sock *sk_alloc(int family, gfp_t priority,
837 struct proto *prot, int zero_it)
838 {
839 struct sock *sk = NULL;
840 kmem_cache_t *slab = prot->slab;
841
842 if (slab != NULL)
843 sk = kmem_cache_alloc(slab, priority);
844 else
845 sk = kmalloc(prot->obj_size, priority);
846
847 if (sk) {
848 if (zero_it) {
849 memset(sk, 0, prot->obj_size);
850 sk->sk_family = family;
851 /*
852 * See comment in struct sock definition to understand
853 * why we need sk_prot_creator -acme
854 */
855 sk->sk_prot = sk->sk_prot_creator = prot;
856 sock_lock_init(sk);
857 }
858
859 if (security_sk_alloc(sk, family, priority))
860 goto out_free;
861
862 if (!try_module_get(prot->owner))
863 goto out_free;
864 }
865 return sk;
866
867 out_free:
868 if (slab != NULL)
869 kmem_cache_free(slab, sk);
870 else
871 kfree(sk);
872 return NULL;
873 }
874
875 void sk_free(struct sock *sk)
876 {
877 struct sk_filter *filter;
878 struct module *owner = sk->sk_prot_creator->owner;
879
880 if (sk->sk_destruct)
881 sk->sk_destruct(sk);
882
883 filter = rcu_dereference(sk->sk_filter);
884 if (filter) {
885 sk_filter_release(sk, filter);
886 rcu_assign_pointer(sk->sk_filter, NULL);
887 }
888
889 sock_disable_timestamp(sk);
890
891 if (atomic_read(&sk->sk_omem_alloc))
892 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
893 __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
894
895 security_sk_free(sk);
896 if (sk->sk_prot_creator->slab != NULL)
897 kmem_cache_free(sk->sk_prot_creator->slab, sk);
898 else
899 kfree(sk);
900 module_put(owner);
901 }
902
903 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
904 {
905 struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
906
907 if (newsk != NULL) {
908 struct sk_filter *filter;
909
910 sock_copy(newsk, sk);
911
912 /* SANITY */
913 sk_node_init(&newsk->sk_node);
914 sock_lock_init(newsk);
915 bh_lock_sock(newsk);
916
917 atomic_set(&newsk->sk_rmem_alloc, 0);
918 atomic_set(&newsk->sk_wmem_alloc, 0);
919 atomic_set(&newsk->sk_omem_alloc, 0);
920 skb_queue_head_init(&newsk->sk_receive_queue);
921 skb_queue_head_init(&newsk->sk_write_queue);
922 #ifdef CONFIG_NET_DMA
923 skb_queue_head_init(&newsk->sk_async_wait_queue);
924 #endif
925
926 rwlock_init(&newsk->sk_dst_lock);
927 rwlock_init(&newsk->sk_callback_lock);
928 lockdep_set_class(&newsk->sk_callback_lock,
929 af_callback_keys + newsk->sk_family);
930
931 newsk->sk_dst_cache = NULL;
932 newsk->sk_wmem_queued = 0;
933 newsk->sk_forward_alloc = 0;
934 newsk->sk_send_head = NULL;
935 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
936 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
937
938 sock_reset_flag(newsk, SOCK_DONE);
939 skb_queue_head_init(&newsk->sk_error_queue);
940
941 filter = newsk->sk_filter;
942 if (filter != NULL)
943 sk_filter_charge(newsk, filter);
944
945 if (unlikely(xfrm_sk_clone_policy(newsk))) {
946 /* It is still raw copy of parent, so invalidate
947 * destructor and make plain sk_free() */
948 newsk->sk_destruct = NULL;
949 sk_free(newsk);
950 newsk = NULL;
951 goto out;
952 }
953
954 newsk->sk_err = 0;
955 newsk->sk_priority = 0;
956 atomic_set(&newsk->sk_refcnt, 2);
957
958 /*
959 * Increment the counter in the same struct proto as the master
960 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
961 * is the same as sk->sk_prot->socks, as this field was copied
962 * with memcpy).
963 *
964 * This _changes_ the previous behaviour, where
965 * tcp_create_openreq_child always was incrementing the
966 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
967 * to be taken into account in all callers. -acme
968 */
969 sk_refcnt_debug_inc(newsk);
970 newsk->sk_socket = NULL;
971 newsk->sk_sleep = NULL;
972
973 if (newsk->sk_prot->sockets_allocated)
974 atomic_inc(newsk->sk_prot->sockets_allocated);
975 }
976 out:
977 return newsk;
978 }
979
980 EXPORT_SYMBOL_GPL(sk_clone);
981
982 void __init sk_init(void)
983 {
984 if (num_physpages <= 4096) {
985 sysctl_wmem_max = 32767;
986 sysctl_rmem_max = 32767;
987 sysctl_wmem_default = 32767;
988 sysctl_rmem_default = 32767;
989 } else if (num_physpages >= 131072) {
990 sysctl_wmem_max = 131071;
991 sysctl_rmem_max = 131071;
992 }
993 }
994
995 /*
996 * Simple resource managers for sockets.
997 */
998
999
1000 /*
1001 * Write buffer destructor automatically called from kfree_skb.
1002 */
1003 void sock_wfree(struct sk_buff *skb)
1004 {
1005 struct sock *sk = skb->sk;
1006
1007 /* In case it might be waiting for more memory. */
1008 atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1009 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1010 sk->sk_write_space(sk);
1011 sock_put(sk);
1012 }
1013
1014 /*
1015 * Read buffer destructor automatically called from kfree_skb.
1016 */
1017 void sock_rfree(struct sk_buff *skb)
1018 {
1019 struct sock *sk = skb->sk;
1020
1021 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1022 }
1023
1024
1025 int sock_i_uid(struct sock *sk)
1026 {
1027 int uid;
1028
1029 read_lock(&sk->sk_callback_lock);
1030 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1031 read_unlock(&sk->sk_callback_lock);
1032 return uid;
1033 }
1034
1035 unsigned long sock_i_ino(struct sock *sk)
1036 {
1037 unsigned long ino;
1038
1039 read_lock(&sk->sk_callback_lock);
1040 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1041 read_unlock(&sk->sk_callback_lock);
1042 return ino;
1043 }
1044
1045 /*
1046 * Allocate a skb from the socket's send buffer.
1047 */
1048 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1049 gfp_t priority)
1050 {
1051 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1052 struct sk_buff * skb = alloc_skb(size, priority);
1053 if (skb) {
1054 skb_set_owner_w(skb, sk);
1055 return skb;
1056 }
1057 }
1058 return NULL;
1059 }
1060
1061 /*
1062 * Allocate a skb from the socket's receive buffer.
1063 */
1064 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1065 gfp_t priority)
1066 {
1067 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1068 struct sk_buff *skb = alloc_skb(size, priority);
1069 if (skb) {
1070 skb_set_owner_r(skb, sk);
1071 return skb;
1072 }
1073 }
1074 return NULL;
1075 }
1076
1077 /*
1078 * Allocate a memory block from the socket's option memory buffer.
1079 */
1080 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1081 {
1082 if ((unsigned)size <= sysctl_optmem_max &&
1083 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1084 void *mem;
1085 /* First do the add, to avoid the race if kmalloc
1086 * might sleep.
1087 */
1088 atomic_add(size, &sk->sk_omem_alloc);
1089 mem = kmalloc(size, priority);
1090 if (mem)
1091 return mem;
1092 atomic_sub(size, &sk->sk_omem_alloc);
1093 }
1094 return NULL;
1095 }
1096
1097 /*
1098 * Free an option memory block.
1099 */
1100 void sock_kfree_s(struct sock *sk, void *mem, int size)
1101 {
1102 kfree(mem);
1103 atomic_sub(size, &sk->sk_omem_alloc);
1104 }
1105
1106 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1107 I think, these locks should be removed for datagram sockets.
1108 */
1109 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1110 {
1111 DEFINE_WAIT(wait);
1112
1113 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1114 for (;;) {
1115 if (!timeo)
1116 break;
1117 if (signal_pending(current))
1118 break;
1119 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1120 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1121 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1122 break;
1123 if (sk->sk_shutdown & SEND_SHUTDOWN)
1124 break;
1125 if (sk->sk_err)
1126 break;
1127 timeo = schedule_timeout(timeo);
1128 }
1129 finish_wait(sk->sk_sleep, &wait);
1130 return timeo;
1131 }
1132
1133
1134 /*
1135 * Generic send/receive buffer handlers
1136 */
1137
1138 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1139 unsigned long header_len,
1140 unsigned long data_len,
1141 int noblock, int *errcode)
1142 {
1143 struct sk_buff *skb;
1144 gfp_t gfp_mask;
1145 long timeo;
1146 int err;
1147
1148 gfp_mask = sk->sk_allocation;
1149 if (gfp_mask & __GFP_WAIT)
1150 gfp_mask |= __GFP_REPEAT;
1151
1152 timeo = sock_sndtimeo(sk, noblock);
1153 while (1) {
1154 err = sock_error(sk);
1155 if (err != 0)
1156 goto failure;
1157
1158 err = -EPIPE;
1159 if (sk->sk_shutdown & SEND_SHUTDOWN)
1160 goto failure;
1161
1162 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1163 skb = alloc_skb(header_len, gfp_mask);
1164 if (skb) {
1165 int npages;
1166 int i;
1167
1168 /* No pages, we're done... */
1169 if (!data_len)
1170 break;
1171
1172 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1173 skb->truesize += data_len;
1174 skb_shinfo(skb)->nr_frags = npages;
1175 for (i = 0; i < npages; i++) {
1176 struct page *page;
1177 skb_frag_t *frag;
1178
1179 page = alloc_pages(sk->sk_allocation, 0);
1180 if (!page) {
1181 err = -ENOBUFS;
1182 skb_shinfo(skb)->nr_frags = i;
1183 kfree_skb(skb);
1184 goto failure;
1185 }
1186
1187 frag = &skb_shinfo(skb)->frags[i];
1188 frag->page = page;
1189 frag->page_offset = 0;
1190 frag->size = (data_len >= PAGE_SIZE ?
1191 PAGE_SIZE :
1192 data_len);
1193 data_len -= PAGE_SIZE;
1194 }
1195
1196 /* Full success... */
1197 break;
1198 }
1199 err = -ENOBUFS;
1200 goto failure;
1201 }
1202 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1203 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1204 err = -EAGAIN;
1205 if (!timeo)
1206 goto failure;
1207 if (signal_pending(current))
1208 goto interrupted;
1209 timeo = sock_wait_for_wmem(sk, timeo);
1210 }
1211
1212 skb_set_owner_w(skb, sk);
1213 return skb;
1214
1215 interrupted:
1216 err = sock_intr_errno(timeo);
1217 failure:
1218 *errcode = err;
1219 return NULL;
1220 }
1221
1222 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1223 int noblock, int *errcode)
1224 {
1225 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1226 }
1227
1228 static void __lock_sock(struct sock *sk)
1229 {
1230 DEFINE_WAIT(wait);
1231
1232 for(;;) {
1233 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1234 TASK_UNINTERRUPTIBLE);
1235 spin_unlock_bh(&sk->sk_lock.slock);
1236 schedule();
1237 spin_lock_bh(&sk->sk_lock.slock);
1238 if(!sock_owned_by_user(sk))
1239 break;
1240 }
1241 finish_wait(&sk->sk_lock.wq, &wait);
1242 }
1243
1244 static void __release_sock(struct sock *sk)
1245 {
1246 struct sk_buff *skb = sk->sk_backlog.head;
1247
1248 do {
1249 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1250 bh_unlock_sock(sk);
1251
1252 do {
1253 struct sk_buff *next = skb->next;
1254
1255 skb->next = NULL;
1256 sk->sk_backlog_rcv(sk, skb);
1257
1258 /*
1259 * We are in process context here with softirqs
1260 * disabled, use cond_resched_softirq() to preempt.
1261 * This is safe to do because we've taken the backlog
1262 * queue private:
1263 */
1264 cond_resched_softirq();
1265
1266 skb = next;
1267 } while (skb != NULL);
1268
1269 bh_lock_sock(sk);
1270 } while((skb = sk->sk_backlog.head) != NULL);
1271 }
1272
1273 /**
1274 * sk_wait_data - wait for data to arrive at sk_receive_queue
1275 * @sk: sock to wait on
1276 * @timeo: for how long
1277 *
1278 * Now socket state including sk->sk_err is changed only under lock,
1279 * hence we may omit checks after joining wait queue.
1280 * We check receive queue before schedule() only as optimization;
1281 * it is very likely that release_sock() added new data.
1282 */
1283 int sk_wait_data(struct sock *sk, long *timeo)
1284 {
1285 int rc;
1286 DEFINE_WAIT(wait);
1287
1288 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1289 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1290 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1291 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1292 finish_wait(sk->sk_sleep, &wait);
1293 return rc;
1294 }
1295
1296 EXPORT_SYMBOL(sk_wait_data);
1297
1298 /*
1299 * Set of default routines for initialising struct proto_ops when
1300 * the protocol does not support a particular function. In certain
1301 * cases where it makes no sense for a protocol to have a "do nothing"
1302 * function, some default processing is provided.
1303 */
1304
1305 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1306 {
1307 return -EOPNOTSUPP;
1308 }
1309
1310 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1311 int len, int flags)
1312 {
1313 return -EOPNOTSUPP;
1314 }
1315
1316 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1317 {
1318 return -EOPNOTSUPP;
1319 }
1320
1321 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1322 {
1323 return -EOPNOTSUPP;
1324 }
1325
1326 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1327 int *len, int peer)
1328 {
1329 return -EOPNOTSUPP;
1330 }
1331
1332 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1333 {
1334 return 0;
1335 }
1336
1337 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1338 {
1339 return -EOPNOTSUPP;
1340 }
1341
1342 int sock_no_listen(struct socket *sock, int backlog)
1343 {
1344 return -EOPNOTSUPP;
1345 }
1346
1347 int sock_no_shutdown(struct socket *sock, int how)
1348 {
1349 return -EOPNOTSUPP;
1350 }
1351
1352 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1353 char __user *optval, int optlen)
1354 {
1355 return -EOPNOTSUPP;
1356 }
1357
1358 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1359 char __user *optval, int __user *optlen)
1360 {
1361 return -EOPNOTSUPP;
1362 }
1363
1364 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1365 size_t len)
1366 {
1367 return -EOPNOTSUPP;
1368 }
1369
1370 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1371 size_t len, int flags)
1372 {
1373 return -EOPNOTSUPP;
1374 }
1375
1376 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1377 {
1378 /* Mirror missing mmap method error code */
1379 return -ENODEV;
1380 }
1381
1382 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1383 {
1384 ssize_t res;
1385 struct msghdr msg = {.msg_flags = flags};
1386 struct kvec iov;
1387 char *kaddr = kmap(page);
1388 iov.iov_base = kaddr + offset;
1389 iov.iov_len = size;
1390 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1391 kunmap(page);
1392 return res;
1393 }
1394
1395 /*
1396 * Default Socket Callbacks
1397 */
1398
1399 static void sock_def_wakeup(struct sock *sk)
1400 {
1401 read_lock(&sk->sk_callback_lock);
1402 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1403 wake_up_interruptible_all(sk->sk_sleep);
1404 read_unlock(&sk->sk_callback_lock);
1405 }
1406
1407 static void sock_def_error_report(struct sock *sk)
1408 {
1409 read_lock(&sk->sk_callback_lock);
1410 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1411 wake_up_interruptible(sk->sk_sleep);
1412 sk_wake_async(sk,0,POLL_ERR);
1413 read_unlock(&sk->sk_callback_lock);
1414 }
1415
1416 static void sock_def_readable(struct sock *sk, int len)
1417 {
1418 read_lock(&sk->sk_callback_lock);
1419 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1420 wake_up_interruptible(sk->sk_sleep);
1421 sk_wake_async(sk,1,POLL_IN);
1422 read_unlock(&sk->sk_callback_lock);
1423 }
1424
1425 static void sock_def_write_space(struct sock *sk)
1426 {
1427 read_lock(&sk->sk_callback_lock);
1428
1429 /* Do not wake up a writer until he can make "significant"
1430 * progress. --DaveM
1431 */
1432 if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1433 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1434 wake_up_interruptible(sk->sk_sleep);
1435
1436 /* Should agree with poll, otherwise some programs break */
1437 if (sock_writeable(sk))
1438 sk_wake_async(sk, 2, POLL_OUT);
1439 }
1440
1441 read_unlock(&sk->sk_callback_lock);
1442 }
1443
1444 static void sock_def_destruct(struct sock *sk)
1445 {
1446 kfree(sk->sk_protinfo);
1447 }
1448
1449 void sk_send_sigurg(struct sock *sk)
1450 {
1451 if (sk->sk_socket && sk->sk_socket->file)
1452 if (send_sigurg(&sk->sk_socket->file->f_owner))
1453 sk_wake_async(sk, 3, POLL_PRI);
1454 }
1455
1456 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1457 unsigned long expires)
1458 {
1459 if (!mod_timer(timer, expires))
1460 sock_hold(sk);
1461 }
1462
1463 EXPORT_SYMBOL(sk_reset_timer);
1464
1465 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1466 {
1467 if (timer_pending(timer) && del_timer(timer))
1468 __sock_put(sk);
1469 }
1470
1471 EXPORT_SYMBOL(sk_stop_timer);
1472
1473 void sock_init_data(struct socket *sock, struct sock *sk)
1474 {
1475 skb_queue_head_init(&sk->sk_receive_queue);
1476 skb_queue_head_init(&sk->sk_write_queue);
1477 skb_queue_head_init(&sk->sk_error_queue);
1478 #ifdef CONFIG_NET_DMA
1479 skb_queue_head_init(&sk->sk_async_wait_queue);
1480 #endif
1481
1482 sk->sk_send_head = NULL;
1483
1484 init_timer(&sk->sk_timer);
1485
1486 sk->sk_allocation = GFP_KERNEL;
1487 sk->sk_rcvbuf = sysctl_rmem_default;
1488 sk->sk_sndbuf = sysctl_wmem_default;
1489 sk->sk_state = TCP_CLOSE;
1490 sk->sk_socket = sock;
1491
1492 sock_set_flag(sk, SOCK_ZAPPED);
1493
1494 if(sock)
1495 {
1496 sk->sk_type = sock->type;
1497 sk->sk_sleep = &sock->wait;
1498 sock->sk = sk;
1499 } else
1500 sk->sk_sleep = NULL;
1501
1502 rwlock_init(&sk->sk_dst_lock);
1503 rwlock_init(&sk->sk_callback_lock);
1504 lockdep_set_class(&sk->sk_callback_lock,
1505 af_callback_keys + sk->sk_family);
1506
1507 sk->sk_state_change = sock_def_wakeup;
1508 sk->sk_data_ready = sock_def_readable;
1509 sk->sk_write_space = sock_def_write_space;
1510 sk->sk_error_report = sock_def_error_report;
1511 sk->sk_destruct = sock_def_destruct;
1512
1513 sk->sk_sndmsg_page = NULL;
1514 sk->sk_sndmsg_off = 0;
1515
1516 sk->sk_peercred.pid = 0;
1517 sk->sk_peercred.uid = -1;
1518 sk->sk_peercred.gid = -1;
1519 sk->sk_write_pending = 0;
1520 sk->sk_rcvlowat = 1;
1521 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1522 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1523
1524 sk->sk_stamp.tv_sec = -1L;
1525 sk->sk_stamp.tv_usec = -1L;
1526
1527 atomic_set(&sk->sk_refcnt, 1);
1528 }
1529
1530 void fastcall lock_sock(struct sock *sk)
1531 {
1532 might_sleep();
1533 spin_lock_bh(&sk->sk_lock.slock);
1534 if (sk->sk_lock.owner)
1535 __lock_sock(sk);
1536 sk->sk_lock.owner = (void *)1;
1537 spin_unlock(&sk->sk_lock.slock);
1538 /*
1539 * The sk_lock has mutex_lock() semantics here:
1540 */
1541 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1542 local_bh_enable();
1543 }
1544
1545 EXPORT_SYMBOL(lock_sock);
1546
1547 void fastcall release_sock(struct sock *sk)
1548 {
1549 /*
1550 * The sk_lock has mutex_unlock() semantics:
1551 */
1552 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1553
1554 spin_lock_bh(&sk->sk_lock.slock);
1555 if (sk->sk_backlog.tail)
1556 __release_sock(sk);
1557 sk->sk_lock.owner = NULL;
1558 if (waitqueue_active(&sk->sk_lock.wq))
1559 wake_up(&sk->sk_lock.wq);
1560 spin_unlock_bh(&sk->sk_lock.slock);
1561 }
1562 EXPORT_SYMBOL(release_sock);
1563
1564 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1565 {
1566 if (!sock_flag(sk, SOCK_TIMESTAMP))
1567 sock_enable_timestamp(sk);
1568 if (sk->sk_stamp.tv_sec == -1)
1569 return -ENOENT;
1570 if (sk->sk_stamp.tv_sec == 0)
1571 do_gettimeofday(&sk->sk_stamp);
1572 return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1573 -EFAULT : 0;
1574 }
1575 EXPORT_SYMBOL(sock_get_timestamp);
1576
1577 void sock_enable_timestamp(struct sock *sk)
1578 {
1579 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1580 sock_set_flag(sk, SOCK_TIMESTAMP);
1581 net_enable_timestamp();
1582 }
1583 }
1584 EXPORT_SYMBOL(sock_enable_timestamp);
1585
1586 /*
1587 * Get a socket option on an socket.
1588 *
1589 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1590 * asynchronous errors should be reported by getsockopt. We assume
1591 * this means if you specify SO_ERROR (otherwise whats the point of it).
1592 */
1593 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1594 char __user *optval, int __user *optlen)
1595 {
1596 struct sock *sk = sock->sk;
1597
1598 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1599 }
1600
1601 EXPORT_SYMBOL(sock_common_getsockopt);
1602
1603 #ifdef CONFIG_COMPAT
1604 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1605 char __user *optval, int __user *optlen)
1606 {
1607 struct sock *sk = sock->sk;
1608
1609 if (sk->sk_prot->compat_setsockopt != NULL)
1610 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1611 optval, optlen);
1612 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1613 }
1614 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1615 #endif
1616
1617 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1618 struct msghdr *msg, size_t size, int flags)
1619 {
1620 struct sock *sk = sock->sk;
1621 int addr_len = 0;
1622 int err;
1623
1624 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1625 flags & ~MSG_DONTWAIT, &addr_len);
1626 if (err >= 0)
1627 msg->msg_namelen = addr_len;
1628 return err;
1629 }
1630
1631 EXPORT_SYMBOL(sock_common_recvmsg);
1632
1633 /*
1634 * Set socket options on an inet socket.
1635 */
1636 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1637 char __user *optval, int optlen)
1638 {
1639 struct sock *sk = sock->sk;
1640
1641 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1642 }
1643
1644 EXPORT_SYMBOL(sock_common_setsockopt);
1645
1646 #ifdef CONFIG_COMPAT
1647 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1648 char __user *optval, int optlen)
1649 {
1650 struct sock *sk = sock->sk;
1651
1652 if (sk->sk_prot->compat_setsockopt != NULL)
1653 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1654 optval, optlen);
1655 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1656 }
1657 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1658 #endif
1659
1660 void sk_common_release(struct sock *sk)
1661 {
1662 if (sk->sk_prot->destroy)
1663 sk->sk_prot->destroy(sk);
1664
1665 /*
1666 * Observation: when sock_common_release is called, processes have
1667 * no access to socket. But net still has.
1668 * Step one, detach it from networking:
1669 *
1670 * A. Remove from hash tables.
1671 */
1672
1673 sk->sk_prot->unhash(sk);
1674
1675 /*
1676 * In this point socket cannot receive new packets, but it is possible
1677 * that some packets are in flight because some CPU runs receiver and
1678 * did hash table lookup before we unhashed socket. They will achieve
1679 * receive queue and will be purged by socket destructor.
1680 *
1681 * Also we still have packets pending on receive queue and probably,
1682 * our own packets waiting in device queues. sock_destroy will drain
1683 * receive queue, but transmitted packets will delay socket destruction
1684 * until the last reference will be released.
1685 */
1686
1687 sock_orphan(sk);
1688
1689 xfrm_sk_free_policy(sk);
1690
1691 sk_refcnt_debug_release(sk);
1692 sock_put(sk);
1693 }
1694
1695 EXPORT_SYMBOL(sk_common_release);
1696
1697 static DEFINE_RWLOCK(proto_list_lock);
1698 static LIST_HEAD(proto_list);
1699
1700 int proto_register(struct proto *prot, int alloc_slab)
1701 {
1702 char *request_sock_slab_name = NULL;
1703 char *timewait_sock_slab_name;
1704 int rc = -ENOBUFS;
1705
1706 if (alloc_slab) {
1707 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1708 SLAB_HWCACHE_ALIGN, NULL, NULL);
1709
1710 if (prot->slab == NULL) {
1711 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1712 prot->name);
1713 goto out;
1714 }
1715
1716 if (prot->rsk_prot != NULL) {
1717 static const char mask[] = "request_sock_%s";
1718
1719 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1720 if (request_sock_slab_name == NULL)
1721 goto out_free_sock_slab;
1722
1723 sprintf(request_sock_slab_name, mask, prot->name);
1724 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1725 prot->rsk_prot->obj_size, 0,
1726 SLAB_HWCACHE_ALIGN, NULL, NULL);
1727
1728 if (prot->rsk_prot->slab == NULL) {
1729 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1730 prot->name);
1731 goto out_free_request_sock_slab_name;
1732 }
1733 }
1734
1735 if (prot->twsk_prot != NULL) {
1736 static const char mask[] = "tw_sock_%s";
1737
1738 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1739
1740 if (timewait_sock_slab_name == NULL)
1741 goto out_free_request_sock_slab;
1742
1743 sprintf(timewait_sock_slab_name, mask, prot->name);
1744 prot->twsk_prot->twsk_slab =
1745 kmem_cache_create(timewait_sock_slab_name,
1746 prot->twsk_prot->twsk_obj_size,
1747 0, SLAB_HWCACHE_ALIGN,
1748 NULL, NULL);
1749 if (prot->twsk_prot->twsk_slab == NULL)
1750 goto out_free_timewait_sock_slab_name;
1751 }
1752 }
1753
1754 write_lock(&proto_list_lock);
1755 list_add(&prot->node, &proto_list);
1756 write_unlock(&proto_list_lock);
1757 rc = 0;
1758 out:
1759 return rc;
1760 out_free_timewait_sock_slab_name:
1761 kfree(timewait_sock_slab_name);
1762 out_free_request_sock_slab:
1763 if (prot->rsk_prot && prot->rsk_prot->slab) {
1764 kmem_cache_destroy(prot->rsk_prot->slab);
1765 prot->rsk_prot->slab = NULL;
1766 }
1767 out_free_request_sock_slab_name:
1768 kfree(request_sock_slab_name);
1769 out_free_sock_slab:
1770 kmem_cache_destroy(prot->slab);
1771 prot->slab = NULL;
1772 goto out;
1773 }
1774
1775 EXPORT_SYMBOL(proto_register);
1776
1777 void proto_unregister(struct proto *prot)
1778 {
1779 write_lock(&proto_list_lock);
1780 list_del(&prot->node);
1781 write_unlock(&proto_list_lock);
1782
1783 if (prot->slab != NULL) {
1784 kmem_cache_destroy(prot->slab);
1785 prot->slab = NULL;
1786 }
1787
1788 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1789 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1790
1791 kmem_cache_destroy(prot->rsk_prot->slab);
1792 kfree(name);
1793 prot->rsk_prot->slab = NULL;
1794 }
1795
1796 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1797 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1798
1799 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1800 kfree(name);
1801 prot->twsk_prot->twsk_slab = NULL;
1802 }
1803 }
1804
1805 EXPORT_SYMBOL(proto_unregister);
1806
1807 #ifdef CONFIG_PROC_FS
1808 static inline struct proto *__proto_head(void)
1809 {
1810 return list_entry(proto_list.next, struct proto, node);
1811 }
1812
1813 static inline struct proto *proto_head(void)
1814 {
1815 return list_empty(&proto_list) ? NULL : __proto_head();
1816 }
1817
1818 static inline struct proto *proto_next(struct proto *proto)
1819 {
1820 return proto->node.next == &proto_list ? NULL :
1821 list_entry(proto->node.next, struct proto, node);
1822 }
1823
1824 static inline struct proto *proto_get_idx(loff_t pos)
1825 {
1826 struct proto *proto;
1827 loff_t i = 0;
1828
1829 list_for_each_entry(proto, &proto_list, node)
1830 if (i++ == pos)
1831 goto out;
1832
1833 proto = NULL;
1834 out:
1835 return proto;
1836 }
1837
1838 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1839 {
1840 read_lock(&proto_list_lock);
1841 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1842 }
1843
1844 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1845 {
1846 ++*pos;
1847 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1848 }
1849
1850 static void proto_seq_stop(struct seq_file *seq, void *v)
1851 {
1852 read_unlock(&proto_list_lock);
1853 }
1854
1855 static char proto_method_implemented(const void *method)
1856 {
1857 return method == NULL ? 'n' : 'y';
1858 }
1859
1860 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1861 {
1862 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1863 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1864 proto->name,
1865 proto->obj_size,
1866 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1867 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1868 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1869 proto->max_header,
1870 proto->slab == NULL ? "no" : "yes",
1871 module_name(proto->owner),
1872 proto_method_implemented(proto->close),
1873 proto_method_implemented(proto->connect),
1874 proto_method_implemented(proto->disconnect),
1875 proto_method_implemented(proto->accept),
1876 proto_method_implemented(proto->ioctl),
1877 proto_method_implemented(proto->init),
1878 proto_method_implemented(proto->destroy),
1879 proto_method_implemented(proto->shutdown),
1880 proto_method_implemented(proto->setsockopt),
1881 proto_method_implemented(proto->getsockopt),
1882 proto_method_implemented(proto->sendmsg),
1883 proto_method_implemented(proto->recvmsg),
1884 proto_method_implemented(proto->sendpage),
1885 proto_method_implemented(proto->bind),
1886 proto_method_implemented(proto->backlog_rcv),
1887 proto_method_implemented(proto->hash),
1888 proto_method_implemented(proto->unhash),
1889 proto_method_implemented(proto->get_port),
1890 proto_method_implemented(proto->enter_memory_pressure));
1891 }
1892
1893 static int proto_seq_show(struct seq_file *seq, void *v)
1894 {
1895 if (v == SEQ_START_TOKEN)
1896 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1897 "protocol",
1898 "size",
1899 "sockets",
1900 "memory",
1901 "press",
1902 "maxhdr",
1903 "slab",
1904 "module",
1905 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1906 else
1907 proto_seq_printf(seq, v);
1908 return 0;
1909 }
1910
1911 static struct seq_operations proto_seq_ops = {
1912 .start = proto_seq_start,
1913 .next = proto_seq_next,
1914 .stop = proto_seq_stop,
1915 .show = proto_seq_show,
1916 };
1917
1918 static int proto_seq_open(struct inode *inode, struct file *file)
1919 {
1920 return seq_open(file, &proto_seq_ops);
1921 }
1922
1923 static struct file_operations proto_seq_fops = {
1924 .owner = THIS_MODULE,
1925 .open = proto_seq_open,
1926 .read = seq_read,
1927 .llseek = seq_lseek,
1928 .release = seq_release,
1929 };
1930
1931 static int __init proto_init(void)
1932 {
1933 /* register /proc/net/protocols */
1934 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1935 }
1936
1937 subsys_initcall(proto_init);
1938
1939 #endif /* PROC_FS */
1940
1941 EXPORT_SYMBOL(sk_alloc);
1942 EXPORT_SYMBOL(sk_free);
1943 EXPORT_SYMBOL(sk_send_sigurg);
1944 EXPORT_SYMBOL(sock_alloc_send_skb);
1945 EXPORT_SYMBOL(sock_init_data);
1946 EXPORT_SYMBOL(sock_kfree_s);
1947 EXPORT_SYMBOL(sock_kmalloc);
1948 EXPORT_SYMBOL(sock_no_accept);
1949 EXPORT_SYMBOL(sock_no_bind);
1950 EXPORT_SYMBOL(sock_no_connect);
1951 EXPORT_SYMBOL(sock_no_getname);
1952 EXPORT_SYMBOL(sock_no_getsockopt);
1953 EXPORT_SYMBOL(sock_no_ioctl);
1954 EXPORT_SYMBOL(sock_no_listen);
1955 EXPORT_SYMBOL(sock_no_mmap);
1956 EXPORT_SYMBOL(sock_no_poll);
1957 EXPORT_SYMBOL(sock_no_recvmsg);
1958 EXPORT_SYMBOL(sock_no_sendmsg);
1959 EXPORT_SYMBOL(sock_no_sendpage);
1960 EXPORT_SYMBOL(sock_no_setsockopt);
1961 EXPORT_SYMBOL(sock_no_shutdown);
1962 EXPORT_SYMBOL(sock_no_socketpair);
1963 EXPORT_SYMBOL(sock_rfree);
1964 EXPORT_SYMBOL(sock_setsockopt);
1965 EXPORT_SYMBOL(sock_wfree);
1966 EXPORT_SYMBOL(sock_wmalloc);
1967 EXPORT_SYMBOL(sock_i_uid);
1968 EXPORT_SYMBOL(sock_i_ino);
1969 EXPORT_SYMBOL(sysctl_optmem_max);
1970 #ifdef CONFIG_SYSCTL
1971 EXPORT_SYMBOL(sysctl_rmem_max);
1972 EXPORT_SYMBOL(sysctl_wmem_max);
1973 #endif
CRIS linux kernel tree