?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[NET]: make seq_operations const
[linux-2.6.git] / net / core / sock.c
blob73a8018029a82e96726110f0f2c5632c78e9fcc0
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 #include <linux/highmem.h>
115
116 #include <asm/uaccess.h>
117 #include <asm/system.h>
118
119 #include <linux/netdevice.h>
120 #include <net/protocol.h>
121 #include <linux/skbuff.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
126
127 #include <linux/filter.h>
128
129 #ifdef CONFIG_INET
130 #include <net/tcp.h>
131 #endif
132
133 /*
134 * Each address family might have different locking rules, so we have
135 * one slock key per address family:
136 */
137 static struct lock_class_key af_family_keys[AF_MAX];
138 static struct lock_class_key af_family_slock_keys[AF_MAX];
139
140 #ifdef CONFIG_DEBUG_LOCK_ALLOC
141 /*
142 * Make lock validator output more readable. (we pre-construct these
143 * strings build-time, so that runtime initialization of socket
144 * locks is fast):
145 */
146 static const char *af_family_key_strings[AF_MAX+1] = {
147 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
148 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
149 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
150 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
151 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
152 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
153 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
154 "sk_lock-21" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
155 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
156 "sk_lock-27" , "sk_lock-28" , "sk_lock-29" ,
157 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
158 };
159 static const char *af_family_slock_key_strings[AF_MAX+1] = {
160 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
161 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
162 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
163 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
164 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
165 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
166 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
167 "slock-21" , "slock-AF_SNA" , "slock-AF_IRDA" ,
168 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
169 "slock-27" , "slock-28" , "slock-29" ,
170 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_MAX"
171 };
172 #endif
173
174 /*
175 * sk_callback_lock locking rules are per-address-family,
176 * so split the lock classes by using a per-AF key:
177 */
178 static struct lock_class_key af_callback_keys[AF_MAX];
179
180 /* Take into consideration the size of the struct sk_buff overhead in the
181 * determination of these values, since that is non-constant across
182 * platforms. This makes socket queueing behavior and performance
183 * not depend upon such differences.
184 */
185 #define _SK_MEM_PACKETS 256
186 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
187 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
188 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
189
190 /* Run time adjustable parameters. */
191 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
192 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
193 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
194 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
195
196 /* Maximal space eaten by iovec or ancilliary data plus some space */
197 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
198
199 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
200 {
201 struct timeval tv;
202
203 if (optlen < sizeof(tv))
204 return -EINVAL;
205 if (copy_from_user(&tv, optval, sizeof(tv)))
206 return -EFAULT;
207
208 *timeo_p = MAX_SCHEDULE_TIMEOUT;
209 if (tv.tv_sec == 0 && tv.tv_usec == 0)
210 return 0;
211 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
212 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
213 return 0;
214 }
215
216 static void sock_warn_obsolete_bsdism(const char *name)
217 {
218 static int warned;
219 static char warncomm[TASK_COMM_LEN];
220 if (strcmp(warncomm, current->comm) && warned < 5) {
221 strcpy(warncomm, current->comm);
222 printk(KERN_WARNING "process `%s' is using obsolete "
223 "%s SO_BSDCOMPAT\n", warncomm, name);
224 warned++;
225 }
226 }
227
228 static void sock_disable_timestamp(struct sock *sk)
229 {
230 if (sock_flag(sk, SOCK_TIMESTAMP)) {
231 sock_reset_flag(sk, SOCK_TIMESTAMP);
232 net_disable_timestamp();
233 }
234 }
235
236
237 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
238 {
239 int err = 0;
240 int skb_len;
241
242 /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
243 number of warnings when compiling with -W --ANK
244 */
245 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
246 (unsigned)sk->sk_rcvbuf) {
247 err = -ENOMEM;
248 goto out;
249 }
250
251 err = sk_filter(sk, skb);
252 if (err)
253 goto out;
254
255 skb->dev = NULL;
256 skb_set_owner_r(skb, sk);
257
258 /* Cache the SKB length before we tack it onto the receive
259 * queue. Once it is added it no longer belongs to us and
260 * may be freed by other threads of control pulling packets
261 * from the queue.
262 */
263 skb_len = skb->len;
264
265 skb_queue_tail(&sk->sk_receive_queue, skb);
266
267 if (!sock_flag(sk, SOCK_DEAD))
268 sk->sk_data_ready(sk, skb_len);
269 out:
270 return err;
271 }
272 EXPORT_SYMBOL(sock_queue_rcv_skb);
273
274 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
275 {
276 int rc = NET_RX_SUCCESS;
277
278 if (sk_filter(sk, skb))
279 goto discard_and_relse;
280
281 skb->dev = NULL;
282
283 if (nested)
284 bh_lock_sock_nested(sk);
285 else
286 bh_lock_sock(sk);
287 if (!sock_owned_by_user(sk)) {
288 /*
289 * trylock + unlock semantics:
290 */
291 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
292
293 rc = sk->sk_backlog_rcv(sk, skb);
294
295 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
296 } else
297 sk_add_backlog(sk, skb);
298 bh_unlock_sock(sk);
299 out:
300 sock_put(sk);
301 return rc;
302 discard_and_relse:
303 kfree_skb(skb);
304 goto out;
305 }
306 EXPORT_SYMBOL(sk_receive_skb);
307
308 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
309 {
310 struct dst_entry *dst = sk->sk_dst_cache;
311
312 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
313 sk->sk_dst_cache = NULL;
314 dst_release(dst);
315 return NULL;
316 }
317
318 return dst;
319 }
320 EXPORT_SYMBOL(__sk_dst_check);
321
322 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
323 {
324 struct dst_entry *dst = sk_dst_get(sk);
325
326 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
327 sk_dst_reset(sk);
328 dst_release(dst);
329 return NULL;
330 }
331
332 return dst;
333 }
334 EXPORT_SYMBOL(sk_dst_check);
335
336 /*
337 * This is meant for all protocols to use and covers goings on
338 * at the socket level. Everything here is generic.
339 */
340
341 int sock_setsockopt(struct socket *sock, int level, int optname,
342 char __user *optval, int optlen)
343 {
344 struct sock *sk=sock->sk;
345 struct sk_filter *filter;
346 int val;
347 int valbool;
348 struct linger ling;
349 int ret = 0;
350
351 /*
352 * Options without arguments
353 */
354
355 #ifdef SO_DONTLINGER /* Compatibility item... */
356 if (optname == SO_DONTLINGER) {
357 lock_sock(sk);
358 sock_reset_flag(sk, SOCK_LINGER);
359 release_sock(sk);
360 return 0;
361 }
362 #endif
363
364 if (optlen < sizeof(int))
365 return -EINVAL;
366
367 if (get_user(val, (int __user *)optval))
368 return -EFAULT;
369
370 valbool = val?1:0;
371
372 lock_sock(sk);
373
374 switch(optname) {
375 case SO_DEBUG:
376 if (val && !capable(CAP_NET_ADMIN)) {
377 ret = -EACCES;
378 }
379 else if (valbool)
380 sock_set_flag(sk, SOCK_DBG);
381 else
382 sock_reset_flag(sk, SOCK_DBG);
383 break;
384 case SO_REUSEADDR:
385 sk->sk_reuse = valbool;
386 break;
387 case SO_TYPE:
388 case SO_ERROR:
389 ret = -ENOPROTOOPT;
390 break;
391 case SO_DONTROUTE:
392 if (valbool)
393 sock_set_flag(sk, SOCK_LOCALROUTE);
394 else
395 sock_reset_flag(sk, SOCK_LOCALROUTE);
396 break;
397 case SO_BROADCAST:
398 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
399 break;
400 case SO_SNDBUF:
401 /* Don't error on this BSD doesn't and if you think
402 about it this is right. Otherwise apps have to
403 play 'guess the biggest size' games. RCVBUF/SNDBUF
404 are treated in BSD as hints */
405
406 if (val > sysctl_wmem_max)
407 val = sysctl_wmem_max;
408 set_sndbuf:
409 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
410 if ((val * 2) < SOCK_MIN_SNDBUF)
411 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
412 else
413 sk->sk_sndbuf = val * 2;
414
415 /*
416 * Wake up sending tasks if we
417 * upped the value.
418 */
419 sk->sk_write_space(sk);
420 break;
421
422 case SO_SNDBUFFORCE:
423 if (!capable(CAP_NET_ADMIN)) {
424 ret = -EPERM;
425 break;
426 }
427 goto set_sndbuf;
428
429 case SO_RCVBUF:
430 /* Don't error on this BSD doesn't and if you think
431 about it this is right. Otherwise apps have to
432 play 'guess the biggest size' games. RCVBUF/SNDBUF
433 are treated in BSD as hints */
434
435 if (val > sysctl_rmem_max)
436 val = sysctl_rmem_max;
437 set_rcvbuf:
438 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
439 /*
440 * We double it on the way in to account for
441 * "struct sk_buff" etc. overhead. Applications
442 * assume that the SO_RCVBUF setting they make will
443 * allow that much actual data to be received on that
444 * socket.
445 *
446 * Applications are unaware that "struct sk_buff" and
447 * other overheads allocate from the receive buffer
448 * during socket buffer allocation.
449 *
450 * And after considering the possible alternatives,
451 * returning the value we actually used in getsockopt
452 * is the most desirable behavior.
453 */
454 if ((val * 2) < SOCK_MIN_RCVBUF)
455 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
456 else
457 sk->sk_rcvbuf = val * 2;
458 break;
459
460 case SO_RCVBUFFORCE:
461 if (!capable(CAP_NET_ADMIN)) {
462 ret = -EPERM;
463 break;
464 }
465 goto set_rcvbuf;
466
467 case SO_KEEPALIVE:
468 #ifdef CONFIG_INET
469 if (sk->sk_protocol == IPPROTO_TCP)
470 tcp_set_keepalive(sk, valbool);
471 #endif
472 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
473 break;
474
475 case SO_OOBINLINE:
476 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
477 break;
478
479 case SO_NO_CHECK:
480 sk->sk_no_check = valbool;
481 break;
482
483 case SO_PRIORITY:
484 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
485 sk->sk_priority = val;
486 else
487 ret = -EPERM;
488 break;
489
490 case SO_LINGER:
491 if (optlen < sizeof(ling)) {
492 ret = -EINVAL; /* 1003.1g */
493 break;
494 }
495 if (copy_from_user(&ling,optval,sizeof(ling))) {
496 ret = -EFAULT;
497 break;
498 }
499 if (!ling.l_onoff)
500 sock_reset_flag(sk, SOCK_LINGER);
501 else {
502 #if (BITS_PER_LONG == 32)
503 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
504 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
505 else
506 #endif
507 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
508 sock_set_flag(sk, SOCK_LINGER);
509 }
510 break;
511
512 case SO_BSDCOMPAT:
513 sock_warn_obsolete_bsdism("setsockopt");
514 break;
515
516 case SO_PASSCRED:
517 if (valbool)
518 set_bit(SOCK_PASSCRED, &sock->flags);
519 else
520 clear_bit(SOCK_PASSCRED, &sock->flags);
521 break;
522
523 case SO_TIMESTAMP:
524 case SO_TIMESTAMPNS:
525 if (valbool) {
526 if (optname == SO_TIMESTAMP)
527 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
528 else
529 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
530 sock_set_flag(sk, SOCK_RCVTSTAMP);
531 sock_enable_timestamp(sk);
532 } else {
533 sock_reset_flag(sk, SOCK_RCVTSTAMP);
534 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
535 }
536 break;
537
538 case SO_RCVLOWAT:
539 if (val < 0)
540 val = INT_MAX;
541 sk->sk_rcvlowat = val ? : 1;
542 break;
543
544 case SO_RCVTIMEO:
545 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
546 break;
547
548 case SO_SNDTIMEO:
549 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
550 break;
551
552 #ifdef CONFIG_NETDEVICES
553 case SO_BINDTODEVICE:
554 {
555 char devname[IFNAMSIZ];
556
557 /* Sorry... */
558 if (!capable(CAP_NET_RAW)) {
559 ret = -EPERM;
560 break;
561 }
562
563 /* Bind this socket to a particular device like "eth0",
564 * as specified in the passed interface name. If the
565 * name is "" or the option length is zero the socket
566 * is not bound.
567 */
568
569 if (!valbool) {
570 sk->sk_bound_dev_if = 0;
571 } else {
572 if (optlen > IFNAMSIZ - 1)
573 optlen = IFNAMSIZ - 1;
574 memset(devname, 0, sizeof(devname));
575 if (copy_from_user(devname, optval, optlen)) {
576 ret = -EFAULT;
577 break;
578 }
579
580 /* Remove any cached route for this socket. */
581 sk_dst_reset(sk);
582
583 if (devname[0] == '\0') {
584 sk->sk_bound_dev_if = 0;
585 } else {
586 struct net_device *dev = dev_get_by_name(devname);
587 if (!dev) {
588 ret = -ENODEV;
589 break;
590 }
591 sk->sk_bound_dev_if = dev->ifindex;
592 dev_put(dev);
593 }
594 }
595 break;
596 }
597 #endif
598
599
600 case SO_ATTACH_FILTER:
601 ret = -EINVAL;
602 if (optlen == sizeof(struct sock_fprog)) {
603 struct sock_fprog fprog;
604
605 ret = -EFAULT;
606 if (copy_from_user(&fprog, optval, sizeof(fprog)))
607 break;
608
609 ret = sk_attach_filter(&fprog, sk);
610 }
611 break;
612
613 case SO_DETACH_FILTER:
614 rcu_read_lock_bh();
615 filter = rcu_dereference(sk->sk_filter);
616 if (filter) {
617 rcu_assign_pointer(sk->sk_filter, NULL);
618 sk_filter_release(sk, filter);
619 rcu_read_unlock_bh();
620 break;
621 }
622 rcu_read_unlock_bh();
623 ret = -ENONET;
624 break;
625
626 case SO_PASSSEC:
627 if (valbool)
628 set_bit(SOCK_PASSSEC, &sock->flags);
629 else
630 clear_bit(SOCK_PASSSEC, &sock->flags);
631 break;
632
633 /* We implement the SO_SNDLOWAT etc to
634 not be settable (1003.1g 5.3) */
635 default:
636 ret = -ENOPROTOOPT;
637 break;
638 }
639 release_sock(sk);
640 return ret;
641 }
642
643
644 int sock_getsockopt(struct socket *sock, int level, int optname,
645 char __user *optval, int __user *optlen)
646 {
647 struct sock *sk = sock->sk;
648
649 union {
650 int val;
651 struct linger ling;
652 struct timeval tm;
653 } v;
654
655 unsigned int lv = sizeof(int);
656 int len;
657
658 if (get_user(len, optlen))
659 return -EFAULT;
660 if (len < 0)
661 return -EINVAL;
662
663 switch(optname) {
664 case SO_DEBUG:
665 v.val = sock_flag(sk, SOCK_DBG);
666 break;
667
668 case SO_DONTROUTE:
669 v.val = sock_flag(sk, SOCK_LOCALROUTE);
670 break;
671
672 case SO_BROADCAST:
673 v.val = !!sock_flag(sk, SOCK_BROADCAST);
674 break;
675
676 case SO_SNDBUF:
677 v.val = sk->sk_sndbuf;
678 break;
679
680 case SO_RCVBUF:
681 v.val = sk->sk_rcvbuf;
682 break;
683
684 case SO_REUSEADDR:
685 v.val = sk->sk_reuse;
686 break;
687
688 case SO_KEEPALIVE:
689 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
690 break;
691
692 case SO_TYPE:
693 v.val = sk->sk_type;
694 break;
695
696 case SO_ERROR:
697 v.val = -sock_error(sk);
698 if (v.val==0)
699 v.val = xchg(&sk->sk_err_soft, 0);
700 break;
701
702 case SO_OOBINLINE:
703 v.val = !!sock_flag(sk, SOCK_URGINLINE);
704 break;
705
706 case SO_NO_CHECK:
707 v.val = sk->sk_no_check;
708 break;
709
710 case SO_PRIORITY:
711 v.val = sk->sk_priority;
712 break;
713
714 case SO_LINGER:
715 lv = sizeof(v.ling);
716 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
717 v.ling.l_linger = sk->sk_lingertime / HZ;
718 break;
719
720 case SO_BSDCOMPAT:
721 sock_warn_obsolete_bsdism("getsockopt");
722 break;
723
724 case SO_TIMESTAMP:
725 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
726 !sock_flag(sk, SOCK_RCVTSTAMPNS);
727 break;
728
729 case SO_TIMESTAMPNS:
730 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
731 break;
732
733 case SO_RCVTIMEO:
734 lv=sizeof(struct timeval);
735 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
736 v.tm.tv_sec = 0;
737 v.tm.tv_usec = 0;
738 } else {
739 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
740 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
741 }
742 break;
743
744 case SO_SNDTIMEO:
745 lv=sizeof(struct timeval);
746 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
747 v.tm.tv_sec = 0;
748 v.tm.tv_usec = 0;
749 } else {
750 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
751 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
752 }
753 break;
754
755 case SO_RCVLOWAT:
756 v.val = sk->sk_rcvlowat;
757 break;
758
759 case SO_SNDLOWAT:
760 v.val=1;
761 break;
762
763 case SO_PASSCRED:
764 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
765 break;
766
767 case SO_PEERCRED:
768 if (len > sizeof(sk->sk_peercred))
769 len = sizeof(sk->sk_peercred);
770 if (copy_to_user(optval, &sk->sk_peercred, len))
771 return -EFAULT;
772 goto lenout;
773
774 case SO_PEERNAME:
775 {
776 char address[128];
777
778 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
779 return -ENOTCONN;
780 if (lv < len)
781 return -EINVAL;
782 if (copy_to_user(optval, address, len))
783 return -EFAULT;
784 goto lenout;
785 }
786
787 /* Dubious BSD thing... Probably nobody even uses it, but
788 * the UNIX standard wants it for whatever reason... -DaveM
789 */
790 case SO_ACCEPTCONN:
791 v.val = sk->sk_state == TCP_LISTEN;
792 break;
793
794 case SO_PASSSEC:
795 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
796 break;
797
798 case SO_PEERSEC:
799 return security_socket_getpeersec_stream(sock, optval, optlen, len);
800
801 default:
802 return -ENOPROTOOPT;
803 }
804
805 if (len > lv)
806 len = lv;
807 if (copy_to_user(optval, &v, len))
808 return -EFAULT;
809 lenout:
810 if (put_user(len, optlen))
811 return -EFAULT;
812 return 0;
813 }
814
815 /*
816 * Initialize an sk_lock.
817 *
818 * (We also register the sk_lock with the lock validator.)
819 */
820 static inline void sock_lock_init(struct sock *sk)
821 {
822 sock_lock_init_class_and_name(sk,
823 af_family_slock_key_strings[sk->sk_family],
824 af_family_slock_keys + sk->sk_family,
825 af_family_key_strings[sk->sk_family],
826 af_family_keys + sk->sk_family);
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 struct kmem_cache *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 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
917
918 atomic_set(&newsk->sk_rmem_alloc, 0);
919 atomic_set(&newsk->sk_wmem_alloc, 0);
920 atomic_set(&newsk->sk_omem_alloc, 0);
921 skb_queue_head_init(&newsk->sk_receive_queue);
922 skb_queue_head_init(&newsk->sk_write_queue);
923 #ifdef CONFIG_NET_DMA
924 skb_queue_head_init(&newsk->sk_async_wait_queue);
925 #endif
926
927 rwlock_init(&newsk->sk_dst_lock);
928 rwlock_init(&newsk->sk_callback_lock);
929 lockdep_set_class(&newsk->sk_callback_lock,
930 af_callback_keys + newsk->sk_family);
931
932 newsk->sk_dst_cache = NULL;
933 newsk->sk_wmem_queued = 0;
934 newsk->sk_forward_alloc = 0;
935 newsk->sk_send_head = 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 sk->sk_type = sock->type;
1496 sk->sk_sleep = &sock->wait;
1497 sock->sk = sk;
1498 } else
1499 sk->sk_sleep = NULL;
1500
1501 rwlock_init(&sk->sk_dst_lock);
1502 rwlock_init(&sk->sk_callback_lock);
1503 lockdep_set_class(&sk->sk_callback_lock,
1504 af_callback_keys + sk->sk_family);
1505
1506 sk->sk_state_change = sock_def_wakeup;
1507 sk->sk_data_ready = sock_def_readable;
1508 sk->sk_write_space = sock_def_write_space;
1509 sk->sk_error_report = sock_def_error_report;
1510 sk->sk_destruct = sock_def_destruct;
1511
1512 sk->sk_sndmsg_page = NULL;
1513 sk->sk_sndmsg_off = 0;
1514
1515 sk->sk_peercred.pid = 0;
1516 sk->sk_peercred.uid = -1;
1517 sk->sk_peercred.gid = -1;
1518 sk->sk_write_pending = 0;
1519 sk->sk_rcvlowat = 1;
1520 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1521 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1522
1523 sk->sk_stamp = ktime_set(-1L, -1L);
1524
1525 atomic_set(&sk->sk_refcnt, 1);
1526 }
1527
1528 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1529 {
1530 might_sleep();
1531 spin_lock_bh(&sk->sk_lock.slock);
1532 if (sk->sk_lock.owner)
1533 __lock_sock(sk);
1534 sk->sk_lock.owner = (void *)1;
1535 spin_unlock(&sk->sk_lock.slock);
1536 /*
1537 * The sk_lock has mutex_lock() semantics here:
1538 */
1539 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1540 local_bh_enable();
1541 }
1542
1543 EXPORT_SYMBOL(lock_sock_nested);
1544
1545 void fastcall release_sock(struct sock *sk)
1546 {
1547 /*
1548 * The sk_lock has mutex_unlock() semantics:
1549 */
1550 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1551
1552 spin_lock_bh(&sk->sk_lock.slock);
1553 if (sk->sk_backlog.tail)
1554 __release_sock(sk);
1555 sk->sk_lock.owner = NULL;
1556 if (waitqueue_active(&sk->sk_lock.wq))
1557 wake_up(&sk->sk_lock.wq);
1558 spin_unlock_bh(&sk->sk_lock.slock);
1559 }
1560 EXPORT_SYMBOL(release_sock);
1561
1562 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1563 {
1564 struct timeval tv;
1565 if (!sock_flag(sk, SOCK_TIMESTAMP))
1566 sock_enable_timestamp(sk);
1567 tv = ktime_to_timeval(sk->sk_stamp);
1568 if (tv.tv_sec == -1)
1569 return -ENOENT;
1570 if (tv.tv_sec == 0) {
1571 sk->sk_stamp = ktime_get_real();
1572 tv = ktime_to_timeval(sk->sk_stamp);
1573 }
1574 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1575 }
1576 EXPORT_SYMBOL(sock_get_timestamp);
1577
1578 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1579 {
1580 struct timespec ts;
1581 if (!sock_flag(sk, SOCK_TIMESTAMP))
1582 sock_enable_timestamp(sk);
1583 ts = ktime_to_timespec(sk->sk_stamp);
1584 if (ts.tv_sec == -1)
1585 return -ENOENT;
1586 if (ts.tv_sec == 0) {
1587 sk->sk_stamp = ktime_get_real();
1588 ts = ktime_to_timespec(sk->sk_stamp);
1589 }
1590 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1591 }
1592 EXPORT_SYMBOL(sock_get_timestampns);
1593
1594 void sock_enable_timestamp(struct sock *sk)
1595 {
1596 if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1597 sock_set_flag(sk, SOCK_TIMESTAMP);
1598 net_enable_timestamp();
1599 }
1600 }
1601 EXPORT_SYMBOL(sock_enable_timestamp);
1602
1603 /*
1604 * Get a socket option on an socket.
1605 *
1606 * FIX: POSIX 1003.1g is very ambiguous here. It states that
1607 * asynchronous errors should be reported by getsockopt. We assume
1608 * this means if you specify SO_ERROR (otherwise whats the point of it).
1609 */
1610 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1611 char __user *optval, int __user *optlen)
1612 {
1613 struct sock *sk = sock->sk;
1614
1615 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1616 }
1617
1618 EXPORT_SYMBOL(sock_common_getsockopt);
1619
1620 #ifdef CONFIG_COMPAT
1621 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1622 char __user *optval, int __user *optlen)
1623 {
1624 struct sock *sk = sock->sk;
1625
1626 if (sk->sk_prot->compat_getsockopt != NULL)
1627 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1628 optval, optlen);
1629 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1630 }
1631 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1632 #endif
1633
1634 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1635 struct msghdr *msg, size_t size, int flags)
1636 {
1637 struct sock *sk = sock->sk;
1638 int addr_len = 0;
1639 int err;
1640
1641 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1642 flags & ~MSG_DONTWAIT, &addr_len);
1643 if (err >= 0)
1644 msg->msg_namelen = addr_len;
1645 return err;
1646 }
1647
1648 EXPORT_SYMBOL(sock_common_recvmsg);
1649
1650 /*
1651 * Set socket options on an inet socket.
1652 */
1653 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1654 char __user *optval, int optlen)
1655 {
1656 struct sock *sk = sock->sk;
1657
1658 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1659 }
1660
1661 EXPORT_SYMBOL(sock_common_setsockopt);
1662
1663 #ifdef CONFIG_COMPAT
1664 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1665 char __user *optval, int optlen)
1666 {
1667 struct sock *sk = sock->sk;
1668
1669 if (sk->sk_prot->compat_setsockopt != NULL)
1670 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1671 optval, optlen);
1672 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1673 }
1674 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1675 #endif
1676
1677 void sk_common_release(struct sock *sk)
1678 {
1679 if (sk->sk_prot->destroy)
1680 sk->sk_prot->destroy(sk);
1681
1682 /*
1683 * Observation: when sock_common_release is called, processes have
1684 * no access to socket. But net still has.
1685 * Step one, detach it from networking:
1686 *
1687 * A. Remove from hash tables.
1688 */
1689
1690 sk->sk_prot->unhash(sk);
1691
1692 /*
1693 * In this point socket cannot receive new packets, but it is possible
1694 * that some packets are in flight because some CPU runs receiver and
1695 * did hash table lookup before we unhashed socket. They will achieve
1696 * receive queue and will be purged by socket destructor.
1697 *
1698 * Also we still have packets pending on receive queue and probably,
1699 * our own packets waiting in device queues. sock_destroy will drain
1700 * receive queue, but transmitted packets will delay socket destruction
1701 * until the last reference will be released.
1702 */
1703
1704 sock_orphan(sk);
1705
1706 xfrm_sk_free_policy(sk);
1707
1708 sk_refcnt_debug_release(sk);
1709 sock_put(sk);
1710 }
1711
1712 EXPORT_SYMBOL(sk_common_release);
1713
1714 static DEFINE_RWLOCK(proto_list_lock);
1715 static LIST_HEAD(proto_list);
1716
1717 int proto_register(struct proto *prot, int alloc_slab)
1718 {
1719 char *request_sock_slab_name = NULL;
1720 char *timewait_sock_slab_name;
1721 int rc = -ENOBUFS;
1722
1723 if (alloc_slab) {
1724 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1725 SLAB_HWCACHE_ALIGN, NULL, NULL);
1726
1727 if (prot->slab == NULL) {
1728 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1729 prot->name);
1730 goto out;
1731 }
1732
1733 if (prot->rsk_prot != NULL) {
1734 static const char mask[] = "request_sock_%s";
1735
1736 request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1737 if (request_sock_slab_name == NULL)
1738 goto out_free_sock_slab;
1739
1740 sprintf(request_sock_slab_name, mask, prot->name);
1741 prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1742 prot->rsk_prot->obj_size, 0,
1743 SLAB_HWCACHE_ALIGN, NULL, NULL);
1744
1745 if (prot->rsk_prot->slab == NULL) {
1746 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1747 prot->name);
1748 goto out_free_request_sock_slab_name;
1749 }
1750 }
1751
1752 if (prot->twsk_prot != NULL) {
1753 static const char mask[] = "tw_sock_%s";
1754
1755 timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1756
1757 if (timewait_sock_slab_name == NULL)
1758 goto out_free_request_sock_slab;
1759
1760 sprintf(timewait_sock_slab_name, mask, prot->name);
1761 prot->twsk_prot->twsk_slab =
1762 kmem_cache_create(timewait_sock_slab_name,
1763 prot->twsk_prot->twsk_obj_size,
1764 0, SLAB_HWCACHE_ALIGN,
1765 NULL, NULL);
1766 if (prot->twsk_prot->twsk_slab == NULL)
1767 goto out_free_timewait_sock_slab_name;
1768 }
1769 }
1770
1771 write_lock(&proto_list_lock);
1772 list_add(&prot->node, &proto_list);
1773 write_unlock(&proto_list_lock);
1774 rc = 0;
1775 out:
1776 return rc;
1777 out_free_timewait_sock_slab_name:
1778 kfree(timewait_sock_slab_name);
1779 out_free_request_sock_slab:
1780 if (prot->rsk_prot && prot->rsk_prot->slab) {
1781 kmem_cache_destroy(prot->rsk_prot->slab);
1782 prot->rsk_prot->slab = NULL;
1783 }
1784 out_free_request_sock_slab_name:
1785 kfree(request_sock_slab_name);
1786 out_free_sock_slab:
1787 kmem_cache_destroy(prot->slab);
1788 prot->slab = NULL;
1789 goto out;
1790 }
1791
1792 EXPORT_SYMBOL(proto_register);
1793
1794 void proto_unregister(struct proto *prot)
1795 {
1796 write_lock(&proto_list_lock);
1797 list_del(&prot->node);
1798 write_unlock(&proto_list_lock);
1799
1800 if (prot->slab != NULL) {
1801 kmem_cache_destroy(prot->slab);
1802 prot->slab = NULL;
1803 }
1804
1805 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1806 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1807
1808 kmem_cache_destroy(prot->rsk_prot->slab);
1809 kfree(name);
1810 prot->rsk_prot->slab = NULL;
1811 }
1812
1813 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1814 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1815
1816 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1817 kfree(name);
1818 prot->twsk_prot->twsk_slab = NULL;
1819 }
1820 }
1821
1822 EXPORT_SYMBOL(proto_unregister);
1823
1824 #ifdef CONFIG_PROC_FS
1825 static inline struct proto *__proto_head(void)
1826 {
1827 return list_entry(proto_list.next, struct proto, node);
1828 }
1829
1830 static inline struct proto *proto_head(void)
1831 {
1832 return list_empty(&proto_list) ? NULL : __proto_head();
1833 }
1834
1835 static inline struct proto *proto_next(struct proto *proto)
1836 {
1837 return proto->node.next == &proto_list ? NULL :
1838 list_entry(proto->node.next, struct proto, node);
1839 }
1840
1841 static inline struct proto *proto_get_idx(loff_t pos)
1842 {
1843 struct proto *proto;
1844 loff_t i = 0;
1845
1846 list_for_each_entry(proto, &proto_list, node)
1847 if (i++ == pos)
1848 goto out;
1849
1850 proto = NULL;
1851 out:
1852 return proto;
1853 }
1854
1855 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1856 {
1857 read_lock(&proto_list_lock);
1858 return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1859 }
1860
1861 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1862 {
1863 ++*pos;
1864 return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1865 }
1866
1867 static void proto_seq_stop(struct seq_file *seq, void *v)
1868 {
1869 read_unlock(&proto_list_lock);
1870 }
1871
1872 static char proto_method_implemented(const void *method)
1873 {
1874 return method == NULL ? 'n' : 'y';
1875 }
1876
1877 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1878 {
1879 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
1880 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1881 proto->name,
1882 proto->obj_size,
1883 proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1884 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1885 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1886 proto->max_header,
1887 proto->slab == NULL ? "no" : "yes",
1888 module_name(proto->owner),
1889 proto_method_implemented(proto->close),
1890 proto_method_implemented(proto->connect),
1891 proto_method_implemented(proto->disconnect),
1892 proto_method_implemented(proto->accept),
1893 proto_method_implemented(proto->ioctl),
1894 proto_method_implemented(proto->init),
1895 proto_method_implemented(proto->destroy),
1896 proto_method_implemented(proto->shutdown),
1897 proto_method_implemented(proto->setsockopt),
1898 proto_method_implemented(proto->getsockopt),
1899 proto_method_implemented(proto->sendmsg),
1900 proto_method_implemented(proto->recvmsg),
1901 proto_method_implemented(proto->sendpage),
1902 proto_method_implemented(proto->bind),
1903 proto_method_implemented(proto->backlog_rcv),
1904 proto_method_implemented(proto->hash),
1905 proto_method_implemented(proto->unhash),
1906 proto_method_implemented(proto->get_port),
1907 proto_method_implemented(proto->enter_memory_pressure));
1908 }
1909
1910 static int proto_seq_show(struct seq_file *seq, void *v)
1911 {
1912 if (v == SEQ_START_TOKEN)
1913 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1914 "protocol",
1915 "size",
1916 "sockets",
1917 "memory",
1918 "press",
1919 "maxhdr",
1920 "slab",
1921 "module",
1922 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1923 else
1924 proto_seq_printf(seq, v);
1925 return 0;
1926 }
1927
1928 static const struct seq_operations proto_seq_ops = {
1929 .start = proto_seq_start,
1930 .next = proto_seq_next,
1931 .stop = proto_seq_stop,
1932 .show = proto_seq_show,
1933 };
1934
1935 static int proto_seq_open(struct inode *inode, struct file *file)
1936 {
1937 return seq_open(file, &proto_seq_ops);
1938 }
1939
1940 static const struct file_operations proto_seq_fops = {
1941 .owner = THIS_MODULE,
1942 .open = proto_seq_open,
1943 .read = seq_read,
1944 .llseek = seq_lseek,
1945 .release = seq_release,
1946 };
1947
1948 static int __init proto_init(void)
1949 {
1950 /* register /proc/net/protocols */
1951 return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1952 }
1953
1954 subsys_initcall(proto_init);
1955
1956 #endif /* PROC_FS */
1957
1958 EXPORT_SYMBOL(sk_alloc);
1959 EXPORT_SYMBOL(sk_free);
1960 EXPORT_SYMBOL(sk_send_sigurg);
1961 EXPORT_SYMBOL(sock_alloc_send_skb);
1962 EXPORT_SYMBOL(sock_init_data);
1963 EXPORT_SYMBOL(sock_kfree_s);
1964 EXPORT_SYMBOL(sock_kmalloc);
1965 EXPORT_SYMBOL(sock_no_accept);
1966 EXPORT_SYMBOL(sock_no_bind);
1967 EXPORT_SYMBOL(sock_no_connect);
1968 EXPORT_SYMBOL(sock_no_getname);
1969 EXPORT_SYMBOL(sock_no_getsockopt);
1970 EXPORT_SYMBOL(sock_no_ioctl);
1971 EXPORT_SYMBOL(sock_no_listen);
1972 EXPORT_SYMBOL(sock_no_mmap);
1973 EXPORT_SYMBOL(sock_no_poll);
1974 EXPORT_SYMBOL(sock_no_recvmsg);
1975 EXPORT_SYMBOL(sock_no_sendmsg);
1976 EXPORT_SYMBOL(sock_no_sendpage);
1977 EXPORT_SYMBOL(sock_no_setsockopt);
1978 EXPORT_SYMBOL(sock_no_shutdown);
1979 EXPORT_SYMBOL(sock_no_socketpair);
1980 EXPORT_SYMBOL(sock_rfree);
1981 EXPORT_SYMBOL(sock_setsockopt);
1982 EXPORT_SYMBOL(sock_wfree);
1983 EXPORT_SYMBOL(sock_wmalloc);
1984 EXPORT_SYMBOL(sock_i_uid);
1985 EXPORT_SYMBOL(sock_i_ino);
1986 EXPORT_SYMBOL(sysctl_optmem_max);
1987 #ifdef CONFIG_SYSCTL
1988 EXPORT_SYMBOL(sysctl_rmem_max);
1989 EXPORT_SYMBOL(sysctl_wmem_max);
1990 #endif
Linus' kernel tree