?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
kvm tools, setup: Create private directory
[linux-2.6/next.git] / net / core / sock.c
blob6e819780c23252b3387c1b2daec25a261fa1f2fb
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 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Florian La Roche, <flla@stud.uni-sb.de>
13 * Alan Cox, <A.Cox@swansea.ac.uk>
14 *
15 * Fixes:
16 * Alan Cox : Numerous verify_area() problems
17 * Alan Cox : Connecting on a connecting socket
18 * now returns an error for tcp.
19 * Alan Cox : sock->protocol is set correctly.
20 * and is not sometimes left as 0.
21 * Alan Cox : connect handles icmp errors on a
22 * connect properly. Unfortunately there
23 * is a restart syscall nasty there. I
24 * can't match BSD without hacking the C
25 * library. Ideas urgently sought!
26 * Alan Cox : Disallow bind() to addresses that are
27 * not ours - especially broadcast ones!!
28 * Alan Cox : Socket 1024 _IS_ ok for users. (fencepost)
29 * Alan Cox : sock_wfree/sock_rfree don't destroy sockets,
30 * instead they leave that for the DESTROY timer.
31 * Alan Cox : Clean up error flag in accept
32 * Alan Cox : TCP ack handling is buggy, the DESTROY timer
33 * was buggy. Put a remove_sock() in the handler
34 * for memory when we hit 0. Also altered the timer
35 * code. The ACK stuff can wait and needs major
36 * TCP layer surgery.
37 * Alan Cox : Fixed TCP ack bug, removed remove sock
38 * and fixed timer/inet_bh race.
39 * Alan Cox : Added zapped flag for TCP
40 * Alan Cox : Move kfree_skb into skbuff.c and tidied up surplus code
41 * Alan Cox : for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42 * Alan Cox : kfree_s calls now are kfree_skbmem so we can track skb resources
43 * Alan Cox : Supports socket option broadcast now as does udp. Packet and raw need fixing.
44 * Alan Cox : Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45 * Rick Sladkey : Relaxed UDP rules for matching packets.
46 * C.E.Hawkins : IFF_PROMISC/SIOCGHWADDR support
47 * Pauline Middelink : identd support
48 * Alan Cox : Fixed connect() taking signals I think.
49 * Alan Cox : SO_LINGER supported
50 * Alan Cox : Error reporting fixes
51 * Anonymous : inet_create tidied up (sk->reuse setting)
52 * Alan Cox : inet sockets don't set sk->type!
53 * Alan Cox : Split socket option code
54 * Alan Cox : Callbacks
55 * Alan Cox : Nagle flag for Charles & Johannes stuff
56 * Alex : Removed restriction on inet fioctl
57 * Alan Cox : Splitting INET from NET core
58 * Alan Cox : Fixed bogus SO_TYPE handling in getsockopt()
59 * Adam Caldwell : Missing return in SO_DONTROUTE/SO_DEBUG code
60 * Alan Cox : Split IP from generic code
61 * Alan Cox : New kfree_skbmem()
62 * Alan Cox : Make SO_DEBUG superuser only.
63 * Alan Cox : Allow anyone to clear SO_DEBUG
64 * (compatibility fix)
65 * Alan Cox : Added optimistic memory grabbing for AF_UNIX throughput.
66 * Alan Cox : Allocator for a socket is settable.
67 * Alan Cox : SO_ERROR includes soft errors.
68 * Alan Cox : Allow NULL arguments on some SO_ opts
69 * Alan Cox : Generic socket allocation to make hooks
70 * easier (suggested by Craig Metz).
71 * Michael Pall : SO_ERROR returns positive errno again
72 * Steve Whitehouse: Added default destructor to free
73 * protocol private data.
74 * Steve Whitehouse: Added various other default routines
75 * common to several socket families.
76 * Chris Evans : Call suser() check last on F_SETOWN
77 * Jay Schulist : Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78 * Andi Kleen : Add sock_kmalloc()/sock_kfree_s()
79 * Andi Kleen : Fix write_space callback
80 * Chris Evans : Security fixes - signedness again
81 * Arnaldo C. Melo : cleanups, use skb_queue_purge
82 *
83 * To Fix:
84 *
85 *
86 * This program is free software; you can redistribute it and/or
87 * modify it under the terms of the GNU General Public License
88 * as published by the Free Software Foundation; either version
89 * 2 of the License, or (at your option) any later version.
90 */
91
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
96 #include <linux/in.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113 #include <linux/user_namespace.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/net_namespace.h>
122 #include <net/request_sock.h>
123 #include <net/sock.h>
124 #include <linux/net_tstamp.h>
125 #include <net/xfrm.h>
126 #include <linux/ipsec.h>
127 #include <net/cls_cgroup.h>
128
129 #include <linux/filter.h>
130
131 #ifdef CONFIG_INET
132 #include <net/tcp.h>
133 #endif
134
135 /*
136 * Each address family might have different locking rules, so we have
137 * one slock key per address family:
138 */
139 static struct lock_class_key af_family_keys[AF_MAX];
140 static struct lock_class_key af_family_slock_keys[AF_MAX];
141
142 /*
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
145 * locks is fast):
146 */
147 static const char *const af_family_key_strings[AF_MAX+1] = {
148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
155 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
159 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
160 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
161 "sk_lock-AF_MAX"
162 };
163 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
164 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
165 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
166 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
167 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
168 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
169 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
170 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
171 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
172 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
173 "slock-27" , "slock-28" , "slock-AF_CAN" ,
174 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
175 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
176 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
177 "slock-AF_MAX"
178 };
179 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
180 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
181 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
182 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
183 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
184 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
185 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
186 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
187 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
188 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
189 "clock-27" , "clock-28" , "clock-AF_CAN" ,
190 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
191 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
192 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
193 "clock-AF_MAX"
194 };
195
196 /*
197 * sk_callback_lock locking rules are per-address-family,
198 * so split the lock classes by using a per-AF key:
199 */
200 static struct lock_class_key af_callback_keys[AF_MAX];
201
202 /* Take into consideration the size of the struct sk_buff overhead in the
203 * determination of these values, since that is non-constant across
204 * platforms. This makes socket queueing behavior and performance
205 * not depend upon such differences.
206 */
207 #define _SK_MEM_PACKETS 256
208 #define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
209 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
210 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
211
212 /* Run time adjustable parameters. */
213 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
214 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
215 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
216 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
217
218 /* Maximal space eaten by iovec or ancillary data plus some space */
219 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
220 EXPORT_SYMBOL(sysctl_optmem_max);
221
222 #if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
223 int net_cls_subsys_id = -1;
224 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
225 #endif
226
227 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
228 {
229 struct timeval tv;
230
231 if (optlen < sizeof(tv))
232 return -EINVAL;
233 if (copy_from_user(&tv, optval, sizeof(tv)))
234 return -EFAULT;
235 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
236 return -EDOM;
237
238 if (tv.tv_sec < 0) {
239 static int warned __read_mostly;
240
241 *timeo_p = 0;
242 if (warned < 10 && net_ratelimit()) {
243 warned++;
244 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
245 "tries to set negative timeout\n",
246 current->comm, task_pid_nr(current));
247 }
248 return 0;
249 }
250 *timeo_p = MAX_SCHEDULE_TIMEOUT;
251 if (tv.tv_sec == 0 && tv.tv_usec == 0)
252 return 0;
253 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
254 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
255 return 0;
256 }
257
258 static void sock_warn_obsolete_bsdism(const char *name)
259 {
260 static int warned;
261 static char warncomm[TASK_COMM_LEN];
262 if (strcmp(warncomm, current->comm) && warned < 5) {
263 strcpy(warncomm, current->comm);
264 printk(KERN_WARNING "process `%s' is using obsolete "
265 "%s SO_BSDCOMPAT\n", warncomm, name);
266 warned++;
267 }
268 }
269
270 static void sock_disable_timestamp(struct sock *sk, int flag)
271 {
272 if (sock_flag(sk, flag)) {
273 sock_reset_flag(sk, flag);
274 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
275 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
276 net_disable_timestamp();
277 }
278 }
279 }
280
281
282 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
283 {
284 int err;
285 int skb_len;
286 unsigned long flags;
287 struct sk_buff_head *list = &sk->sk_receive_queue;
288
289 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
290 number of warnings when compiling with -W --ANK
291 */
292 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
293 (unsigned)sk->sk_rcvbuf) {
294 atomic_inc(&sk->sk_drops);
295 return -ENOMEM;
296 }
297
298 err = sk_filter(sk, skb);
299 if (err)
300 return err;
301
302 if (!sk_rmem_schedule(sk, skb->truesize)) {
303 atomic_inc(&sk->sk_drops);
304 return -ENOBUFS;
305 }
306
307 skb->dev = NULL;
308 skb_set_owner_r(skb, sk);
309
310 /* Cache the SKB length before we tack it onto the receive
311 * queue. Once it is added it no longer belongs to us and
312 * may be freed by other threads of control pulling packets
313 * from the queue.
314 */
315 skb_len = skb->len;
316
317 /* we escape from rcu protected region, make sure we dont leak
318 * a norefcounted dst
319 */
320 skb_dst_force(skb);
321
322 spin_lock_irqsave(&list->lock, flags);
323 skb->dropcount = atomic_read(&sk->sk_drops);
324 __skb_queue_tail(list, skb);
325 spin_unlock_irqrestore(&list->lock, flags);
326
327 if (!sock_flag(sk, SOCK_DEAD))
328 sk->sk_data_ready(sk, skb_len);
329 return 0;
330 }
331 EXPORT_SYMBOL(sock_queue_rcv_skb);
332
333 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
334 {
335 int rc = NET_RX_SUCCESS;
336
337 if (sk_filter(sk, skb))
338 goto discard_and_relse;
339
340 skb->dev = NULL;
341
342 if (sk_rcvqueues_full(sk, skb)) {
343 atomic_inc(&sk->sk_drops);
344 goto discard_and_relse;
345 }
346 if (nested)
347 bh_lock_sock_nested(sk);
348 else
349 bh_lock_sock(sk);
350 if (!sock_owned_by_user(sk)) {
351 /*
352 * trylock + unlock semantics:
353 */
354 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
355
356 rc = sk_backlog_rcv(sk, skb);
357
358 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
359 } else if (sk_add_backlog(sk, skb)) {
360 bh_unlock_sock(sk);
361 atomic_inc(&sk->sk_drops);
362 goto discard_and_relse;
363 }
364
365 bh_unlock_sock(sk);
366 out:
367 sock_put(sk);
368 return rc;
369 discard_and_relse:
370 kfree_skb(skb);
371 goto out;
372 }
373 EXPORT_SYMBOL(sk_receive_skb);
374
375 void sk_reset_txq(struct sock *sk)
376 {
377 sk_tx_queue_clear(sk);
378 }
379 EXPORT_SYMBOL(sk_reset_txq);
380
381 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
382 {
383 struct dst_entry *dst = __sk_dst_get(sk);
384
385 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
386 sk_tx_queue_clear(sk);
387 rcu_assign_pointer(sk->sk_dst_cache, NULL);
388 dst_release(dst);
389 return NULL;
390 }
391
392 return dst;
393 }
394 EXPORT_SYMBOL(__sk_dst_check);
395
396 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
397 {
398 struct dst_entry *dst = sk_dst_get(sk);
399
400 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
401 sk_dst_reset(sk);
402 dst_release(dst);
403 return NULL;
404 }
405
406 return dst;
407 }
408 EXPORT_SYMBOL(sk_dst_check);
409
410 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
411 {
412 int ret = -ENOPROTOOPT;
413 #ifdef CONFIG_NETDEVICES
414 struct net *net = sock_net(sk);
415 char devname[IFNAMSIZ];
416 int index;
417
418 /* Sorry... */
419 ret = -EPERM;
420 if (!capable(CAP_NET_RAW))
421 goto out;
422
423 ret = -EINVAL;
424 if (optlen < 0)
425 goto out;
426
427 /* Bind this socket to a particular device like "eth0",
428 * as specified in the passed interface name. If the
429 * name is "" or the option length is zero the socket
430 * is not bound.
431 */
432 if (optlen > IFNAMSIZ - 1)
433 optlen = IFNAMSIZ - 1;
434 memset(devname, 0, sizeof(devname));
435
436 ret = -EFAULT;
437 if (copy_from_user(devname, optval, optlen))
438 goto out;
439
440 index = 0;
441 if (devname[0] != '\0') {
442 struct net_device *dev;
443
444 rcu_read_lock();
445 dev = dev_get_by_name_rcu(net, devname);
446 if (dev)
447 index = dev->ifindex;
448 rcu_read_unlock();
449 ret = -ENODEV;
450 if (!dev)
451 goto out;
452 }
453
454 lock_sock(sk);
455 sk->sk_bound_dev_if = index;
456 sk_dst_reset(sk);
457 release_sock(sk);
458
459 ret = 0;
460
461 out:
462 #endif
463
464 return ret;
465 }
466
467 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
468 {
469 if (valbool)
470 sock_set_flag(sk, bit);
471 else
472 sock_reset_flag(sk, bit);
473 }
474
475 /*
476 * This is meant for all protocols to use and covers goings on
477 * at the socket level. Everything here is generic.
478 */
479
480 int sock_setsockopt(struct socket *sock, int level, int optname,
481 char __user *optval, unsigned int optlen)
482 {
483 struct sock *sk = sock->sk;
484 int val;
485 int valbool;
486 struct linger ling;
487 int ret = 0;
488
489 /*
490 * Options without arguments
491 */
492
493 if (optname == SO_BINDTODEVICE)
494 return sock_bindtodevice(sk, optval, optlen);
495
496 if (optlen < sizeof(int))
497 return -EINVAL;
498
499 if (get_user(val, (int __user *)optval))
500 return -EFAULT;
501
502 valbool = val ? 1 : 0;
503
504 lock_sock(sk);
505
506 switch (optname) {
507 case SO_DEBUG:
508 if (val && !capable(CAP_NET_ADMIN))
509 ret = -EACCES;
510 else
511 sock_valbool_flag(sk, SOCK_DBG, valbool);
512 break;
513 case SO_REUSEADDR:
514 sk->sk_reuse = valbool;
515 break;
516 case SO_TYPE:
517 case SO_PROTOCOL:
518 case SO_DOMAIN:
519 case SO_ERROR:
520 ret = -ENOPROTOOPT;
521 break;
522 case SO_DONTROUTE:
523 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
524 break;
525 case SO_BROADCAST:
526 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
527 break;
528 case SO_SNDBUF:
529 /* Don't error on this BSD doesn't and if you think
530 about it this is right. Otherwise apps have to
531 play 'guess the biggest size' games. RCVBUF/SNDBUF
532 are treated in BSD as hints */
533
534 if (val > sysctl_wmem_max)
535 val = sysctl_wmem_max;
536 set_sndbuf:
537 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
538 if ((val * 2) < SOCK_MIN_SNDBUF)
539 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
540 else
541 sk->sk_sndbuf = val * 2;
542
543 /*
544 * Wake up sending tasks if we
545 * upped the value.
546 */
547 sk->sk_write_space(sk);
548 break;
549
550 case SO_SNDBUFFORCE:
551 if (!capable(CAP_NET_ADMIN)) {
552 ret = -EPERM;
553 break;
554 }
555 goto set_sndbuf;
556
557 case SO_RCVBUF:
558 /* Don't error on this BSD doesn't and if you think
559 about it this is right. Otherwise apps have to
560 play 'guess the biggest size' games. RCVBUF/SNDBUF
561 are treated in BSD as hints */
562
563 if (val > sysctl_rmem_max)
564 val = sysctl_rmem_max;
565 set_rcvbuf:
566 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
567 /*
568 * We double it on the way in to account for
569 * "struct sk_buff" etc. overhead. Applications
570 * assume that the SO_RCVBUF setting they make will
571 * allow that much actual data to be received on that
572 * socket.
573 *
574 * Applications are unaware that "struct sk_buff" and
575 * other overheads allocate from the receive buffer
576 * during socket buffer allocation.
577 *
578 * And after considering the possible alternatives,
579 * returning the value we actually used in getsockopt
580 * is the most desirable behavior.
581 */
582 if ((val * 2) < SOCK_MIN_RCVBUF)
583 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
584 else
585 sk->sk_rcvbuf = val * 2;
586 break;
587
588 case SO_RCVBUFFORCE:
589 if (!capable(CAP_NET_ADMIN)) {
590 ret = -EPERM;
591 break;
592 }
593 goto set_rcvbuf;
594
595 case SO_KEEPALIVE:
596 #ifdef CONFIG_INET
597 if (sk->sk_protocol == IPPROTO_TCP)
598 tcp_set_keepalive(sk, valbool);
599 #endif
600 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
601 break;
602
603 case SO_OOBINLINE:
604 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
605 break;
606
607 case SO_NO_CHECK:
608 sk->sk_no_check = valbool;
609 break;
610
611 case SO_PRIORITY:
612 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
613 sk->sk_priority = val;
614 else
615 ret = -EPERM;
616 break;
617
618 case SO_LINGER:
619 if (optlen < sizeof(ling)) {
620 ret = -EINVAL; /* 1003.1g */
621 break;
622 }
623 if (copy_from_user(&ling, optval, sizeof(ling))) {
624 ret = -EFAULT;
625 break;
626 }
627 if (!ling.l_onoff)
628 sock_reset_flag(sk, SOCK_LINGER);
629 else {
630 #if (BITS_PER_LONG == 32)
631 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
632 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
633 else
634 #endif
635 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
636 sock_set_flag(sk, SOCK_LINGER);
637 }
638 break;
639
640 case SO_BSDCOMPAT:
641 sock_warn_obsolete_bsdism("setsockopt");
642 break;
643
644 case SO_PASSCRED:
645 if (valbool)
646 set_bit(SOCK_PASSCRED, &sock->flags);
647 else
648 clear_bit(SOCK_PASSCRED, &sock->flags);
649 break;
650
651 case SO_TIMESTAMP:
652 case SO_TIMESTAMPNS:
653 if (valbool) {
654 if (optname == SO_TIMESTAMP)
655 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
656 else
657 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
658 sock_set_flag(sk, SOCK_RCVTSTAMP);
659 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
660 } else {
661 sock_reset_flag(sk, SOCK_RCVTSTAMP);
662 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
663 }
664 break;
665
666 case SO_TIMESTAMPING:
667 if (val & ~SOF_TIMESTAMPING_MASK) {
668 ret = -EINVAL;
669 break;
670 }
671 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
672 val & SOF_TIMESTAMPING_TX_HARDWARE);
673 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
674 val & SOF_TIMESTAMPING_TX_SOFTWARE);
675 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
676 val & SOF_TIMESTAMPING_RX_HARDWARE);
677 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
678 sock_enable_timestamp(sk,
679 SOCK_TIMESTAMPING_RX_SOFTWARE);
680 else
681 sock_disable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
684 val & SOF_TIMESTAMPING_SOFTWARE);
685 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
686 val & SOF_TIMESTAMPING_SYS_HARDWARE);
687 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
688 val & SOF_TIMESTAMPING_RAW_HARDWARE);
689 break;
690
691 case SO_RCVLOWAT:
692 if (val < 0)
693 val = INT_MAX;
694 sk->sk_rcvlowat = val ? : 1;
695 break;
696
697 case SO_RCVTIMEO:
698 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
699 break;
700
701 case SO_SNDTIMEO:
702 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
703 break;
704
705 case SO_ATTACH_FILTER:
706 ret = -EINVAL;
707 if (optlen == sizeof(struct sock_fprog)) {
708 struct sock_fprog fprog;
709
710 ret = -EFAULT;
711 if (copy_from_user(&fprog, optval, sizeof(fprog)))
712 break;
713
714 ret = sk_attach_filter(&fprog, sk);
715 }
716 break;
717
718 case SO_DETACH_FILTER:
719 ret = sk_detach_filter(sk);
720 break;
721
722 case SO_PASSSEC:
723 if (valbool)
724 set_bit(SOCK_PASSSEC, &sock->flags);
725 else
726 clear_bit(SOCK_PASSSEC, &sock->flags);
727 break;
728 case SO_MARK:
729 if (!capable(CAP_NET_ADMIN))
730 ret = -EPERM;
731 else
732 sk->sk_mark = val;
733 break;
734
735 /* We implement the SO_SNDLOWAT etc to
736 not be settable (1003.1g 5.3) */
737 case SO_RXQ_OVFL:
738 if (valbool)
739 sock_set_flag(sk, SOCK_RXQ_OVFL);
740 else
741 sock_reset_flag(sk, SOCK_RXQ_OVFL);
742 break;
743 default:
744 ret = -ENOPROTOOPT;
745 break;
746 }
747 release_sock(sk);
748 return ret;
749 }
750 EXPORT_SYMBOL(sock_setsockopt);
751
752
753 void cred_to_ucred(struct pid *pid, const struct cred *cred,
754 struct ucred *ucred)
755 {
756 ucred->pid = pid_vnr(pid);
757 ucred->uid = ucred->gid = -1;
758 if (cred) {
759 struct user_namespace *current_ns = current_user_ns();
760
761 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
762 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
763 }
764 }
765 EXPORT_SYMBOL_GPL(cred_to_ucred);
766
767 int sock_getsockopt(struct socket *sock, int level, int optname,
768 char __user *optval, int __user *optlen)
769 {
770 struct sock *sk = sock->sk;
771
772 union {
773 int val;
774 struct linger ling;
775 struct timeval tm;
776 } v;
777
778 int lv = sizeof(int);
779 int len;
780
781 if (get_user(len, optlen))
782 return -EFAULT;
783 if (len < 0)
784 return -EINVAL;
785
786 memset(&v, 0, sizeof(v));
787
788 switch (optname) {
789 case SO_DEBUG:
790 v.val = sock_flag(sk, SOCK_DBG);
791 break;
792
793 case SO_DONTROUTE:
794 v.val = sock_flag(sk, SOCK_LOCALROUTE);
795 break;
796
797 case SO_BROADCAST:
798 v.val = !!sock_flag(sk, SOCK_BROADCAST);
799 break;
800
801 case SO_SNDBUF:
802 v.val = sk->sk_sndbuf;
803 break;
804
805 case SO_RCVBUF:
806 v.val = sk->sk_rcvbuf;
807 break;
808
809 case SO_REUSEADDR:
810 v.val = sk->sk_reuse;
811 break;
812
813 case SO_KEEPALIVE:
814 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
815 break;
816
817 case SO_TYPE:
818 v.val = sk->sk_type;
819 break;
820
821 case SO_PROTOCOL:
822 v.val = sk->sk_protocol;
823 break;
824
825 case SO_DOMAIN:
826 v.val = sk->sk_family;
827 break;
828
829 case SO_ERROR:
830 v.val = -sock_error(sk);
831 if (v.val == 0)
832 v.val = xchg(&sk->sk_err_soft, 0);
833 break;
834
835 case SO_OOBINLINE:
836 v.val = !!sock_flag(sk, SOCK_URGINLINE);
837 break;
838
839 case SO_NO_CHECK:
840 v.val = sk->sk_no_check;
841 break;
842
843 case SO_PRIORITY:
844 v.val = sk->sk_priority;
845 break;
846
847 case SO_LINGER:
848 lv = sizeof(v.ling);
849 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
850 v.ling.l_linger = sk->sk_lingertime / HZ;
851 break;
852
853 case SO_BSDCOMPAT:
854 sock_warn_obsolete_bsdism("getsockopt");
855 break;
856
857 case SO_TIMESTAMP:
858 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
859 !sock_flag(sk, SOCK_RCVTSTAMPNS);
860 break;
861
862 case SO_TIMESTAMPNS:
863 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
864 break;
865
866 case SO_TIMESTAMPING:
867 v.val = 0;
868 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
869 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
870 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
871 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
872 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
873 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
874 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
875 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
876 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
877 v.val |= SOF_TIMESTAMPING_SOFTWARE;
878 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
879 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
880 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
881 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
882 break;
883
884 case SO_RCVTIMEO:
885 lv = sizeof(struct timeval);
886 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
887 v.tm.tv_sec = 0;
888 v.tm.tv_usec = 0;
889 } else {
890 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
891 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
892 }
893 break;
894
895 case SO_SNDTIMEO:
896 lv = sizeof(struct timeval);
897 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
898 v.tm.tv_sec = 0;
899 v.tm.tv_usec = 0;
900 } else {
901 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
902 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
903 }
904 break;
905
906 case SO_RCVLOWAT:
907 v.val = sk->sk_rcvlowat;
908 break;
909
910 case SO_SNDLOWAT:
911 v.val = 1;
912 break;
913
914 case SO_PASSCRED:
915 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
916 break;
917
918 case SO_PEERCRED:
919 {
920 struct ucred peercred;
921 if (len > sizeof(peercred))
922 len = sizeof(peercred);
923 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
924 if (copy_to_user(optval, &peercred, len))
925 return -EFAULT;
926 goto lenout;
927 }
928
929 case SO_PEERNAME:
930 {
931 char address[128];
932
933 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
934 return -ENOTCONN;
935 if (lv < len)
936 return -EINVAL;
937 if (copy_to_user(optval, address, len))
938 return -EFAULT;
939 goto lenout;
940 }
941
942 /* Dubious BSD thing... Probably nobody even uses it, but
943 * the UNIX standard wants it for whatever reason... -DaveM
944 */
945 case SO_ACCEPTCONN:
946 v.val = sk->sk_state == TCP_LISTEN;
947 break;
948
949 case SO_PASSSEC:
950 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
951 break;
952
953 case SO_PEERSEC:
954 return security_socket_getpeersec_stream(sock, optval, optlen, len);
955
956 case SO_MARK:
957 v.val = sk->sk_mark;
958 break;
959
960 case SO_RXQ_OVFL:
961 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
962 break;
963
964 default:
965 return -ENOPROTOOPT;
966 }
967
968 if (len > lv)
969 len = lv;
970 if (copy_to_user(optval, &v, len))
971 return -EFAULT;
972 lenout:
973 if (put_user(len, optlen))
974 return -EFAULT;
975 return 0;
976 }
977
978 /*
979 * Initialize an sk_lock.
980 *
981 * (We also register the sk_lock with the lock validator.)
982 */
983 static inline void sock_lock_init(struct sock *sk)
984 {
985 sock_lock_init_class_and_name(sk,
986 af_family_slock_key_strings[sk->sk_family],
987 af_family_slock_keys + sk->sk_family,
988 af_family_key_strings[sk->sk_family],
989 af_family_keys + sk->sk_family);
990 }
991
992 /*
993 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
994 * even temporarly, because of RCU lookups. sk_node should also be left as is.
995 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
996 */
997 static void sock_copy(struct sock *nsk, const struct sock *osk)
998 {
999 #ifdef CONFIG_SECURITY_NETWORK
1000 void *sptr = nsk->sk_security;
1001 #endif
1002 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1003
1004 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1005 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1006
1007 #ifdef CONFIG_SECURITY_NETWORK
1008 nsk->sk_security = sptr;
1009 security_sk_clone(osk, nsk);
1010 #endif
1011 }
1012
1013 /*
1014 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1015 * un-modified. Special care is taken when initializing object to zero.
1016 */
1017 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1018 {
1019 if (offsetof(struct sock, sk_node.next) != 0)
1020 memset(sk, 0, offsetof(struct sock, sk_node.next));
1021 memset(&sk->sk_node.pprev, 0,
1022 size - offsetof(struct sock, sk_node.pprev));
1023 }
1024
1025 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1026 {
1027 unsigned long nulls1, nulls2;
1028
1029 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1030 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1031 if (nulls1 > nulls2)
1032 swap(nulls1, nulls2);
1033
1034 if (nulls1 != 0)
1035 memset((char *)sk, 0, nulls1);
1036 memset((char *)sk + nulls1 + sizeof(void *), 0,
1037 nulls2 - nulls1 - sizeof(void *));
1038 memset((char *)sk + nulls2 + sizeof(void *), 0,
1039 size - nulls2 - sizeof(void *));
1040 }
1041 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1042
1043 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1044 int family)
1045 {
1046 struct sock *sk;
1047 struct kmem_cache *slab;
1048
1049 slab = prot->slab;
1050 if (slab != NULL) {
1051 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1052 if (!sk)
1053 return sk;
1054 if (priority & __GFP_ZERO) {
1055 if (prot->clear_sk)
1056 prot->clear_sk(sk, prot->obj_size);
1057 else
1058 sk_prot_clear_nulls(sk, prot->obj_size);
1059 }
1060 } else
1061 sk = kmalloc(prot->obj_size, priority);
1062
1063 if (sk != NULL) {
1064 kmemcheck_annotate_bitfield(sk, flags);
1065
1066 if (security_sk_alloc(sk, family, priority))
1067 goto out_free;
1068
1069 if (!try_module_get(prot->owner))
1070 goto out_free_sec;
1071 sk_tx_queue_clear(sk);
1072 }
1073
1074 return sk;
1075
1076 out_free_sec:
1077 security_sk_free(sk);
1078 out_free:
1079 if (slab != NULL)
1080 kmem_cache_free(slab, sk);
1081 else
1082 kfree(sk);
1083 return NULL;
1084 }
1085
1086 static void sk_prot_free(struct proto *prot, struct sock *sk)
1087 {
1088 struct kmem_cache *slab;
1089 struct module *owner;
1090
1091 owner = prot->owner;
1092 slab = prot->slab;
1093
1094 security_sk_free(sk);
1095 if (slab != NULL)
1096 kmem_cache_free(slab, sk);
1097 else
1098 kfree(sk);
1099 module_put(owner);
1100 }
1101
1102 #ifdef CONFIG_CGROUPS
1103 void sock_update_classid(struct sock *sk)
1104 {
1105 u32 classid;
1106
1107 rcu_read_lock(); /* doing current task, which cannot vanish. */
1108 classid = task_cls_classid(current);
1109 rcu_read_unlock();
1110 if (classid && classid != sk->sk_classid)
1111 sk->sk_classid = classid;
1112 }
1113 EXPORT_SYMBOL(sock_update_classid);
1114 #endif
1115
1116 /**
1117 * sk_alloc - All socket objects are allocated here
1118 * @net: the applicable net namespace
1119 * @family: protocol family
1120 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1121 * @prot: struct proto associated with this new sock instance
1122 */
1123 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1124 struct proto *prot)
1125 {
1126 struct sock *sk;
1127
1128 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1129 if (sk) {
1130 sk->sk_family = family;
1131 /*
1132 * See comment in struct sock definition to understand
1133 * why we need sk_prot_creator -acme
1134 */
1135 sk->sk_prot = sk->sk_prot_creator = prot;
1136 sock_lock_init(sk);
1137 sock_net_set(sk, get_net(net));
1138 atomic_set(&sk->sk_wmem_alloc, 1);
1139
1140 sock_update_classid(sk);
1141 }
1142
1143 return sk;
1144 }
1145 EXPORT_SYMBOL(sk_alloc);
1146
1147 static void __sk_free(struct sock *sk)
1148 {
1149 struct sk_filter *filter;
1150
1151 if (sk->sk_destruct)
1152 sk->sk_destruct(sk);
1153
1154 filter = rcu_dereference_check(sk->sk_filter,
1155 atomic_read(&sk->sk_wmem_alloc) == 0);
1156 if (filter) {
1157 sk_filter_uncharge(sk, filter);
1158 rcu_assign_pointer(sk->sk_filter, NULL);
1159 }
1160
1161 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1162 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1163
1164 if (atomic_read(&sk->sk_omem_alloc))
1165 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1166 __func__, atomic_read(&sk->sk_omem_alloc));
1167
1168 if (sk->sk_peer_cred)
1169 put_cred(sk->sk_peer_cred);
1170 put_pid(sk->sk_peer_pid);
1171 put_net(sock_net(sk));
1172 sk_prot_free(sk->sk_prot_creator, sk);
1173 }
1174
1175 void sk_free(struct sock *sk)
1176 {
1177 /*
1178 * We subtract one from sk_wmem_alloc and can know if
1179 * some packets are still in some tx queue.
1180 * If not null, sock_wfree() will call __sk_free(sk) later
1181 */
1182 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1183 __sk_free(sk);
1184 }
1185 EXPORT_SYMBOL(sk_free);
1186
1187 /*
1188 * Last sock_put should drop reference to sk->sk_net. It has already
1189 * been dropped in sk_change_net. Taking reference to stopping namespace
1190 * is not an option.
1191 * Take reference to a socket to remove it from hash _alive_ and after that
1192 * destroy it in the context of init_net.
1193 */
1194 void sk_release_kernel(struct sock *sk)
1195 {
1196 if (sk == NULL || sk->sk_socket == NULL)
1197 return;
1198
1199 sock_hold(sk);
1200 sock_release(sk->sk_socket);
1201 release_net(sock_net(sk));
1202 sock_net_set(sk, get_net(&init_net));
1203 sock_put(sk);
1204 }
1205 EXPORT_SYMBOL(sk_release_kernel);
1206
1207 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1208 {
1209 struct sock *newsk;
1210
1211 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1212 if (newsk != NULL) {
1213 struct sk_filter *filter;
1214
1215 sock_copy(newsk, sk);
1216
1217 /* SANITY */
1218 get_net(sock_net(newsk));
1219 sk_node_init(&newsk->sk_node);
1220 sock_lock_init(newsk);
1221 bh_lock_sock(newsk);
1222 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
1223 newsk->sk_backlog.len = 0;
1224
1225 atomic_set(&newsk->sk_rmem_alloc, 0);
1226 /*
1227 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1228 */
1229 atomic_set(&newsk->sk_wmem_alloc, 1);
1230 atomic_set(&newsk->sk_omem_alloc, 0);
1231 skb_queue_head_init(&newsk->sk_receive_queue);
1232 skb_queue_head_init(&newsk->sk_write_queue);
1233 #ifdef CONFIG_NET_DMA
1234 skb_queue_head_init(&newsk->sk_async_wait_queue);
1235 #endif
1236
1237 spin_lock_init(&newsk->sk_dst_lock);
1238 rwlock_init(&newsk->sk_callback_lock);
1239 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1240 af_callback_keys + newsk->sk_family,
1241 af_family_clock_key_strings[newsk->sk_family]);
1242
1243 newsk->sk_dst_cache = NULL;
1244 newsk->sk_wmem_queued = 0;
1245 newsk->sk_forward_alloc = 0;
1246 newsk->sk_send_head = NULL;
1247 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1248
1249 sock_reset_flag(newsk, SOCK_DONE);
1250 skb_queue_head_init(&newsk->sk_error_queue);
1251
1252 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1253 if (filter != NULL)
1254 sk_filter_charge(newsk, filter);
1255
1256 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1257 /* It is still raw copy of parent, so invalidate
1258 * destructor and make plain sk_free() */
1259 newsk->sk_destruct = NULL;
1260 sk_free(newsk);
1261 newsk = NULL;
1262 goto out;
1263 }
1264
1265 newsk->sk_err = 0;
1266 newsk->sk_priority = 0;
1267 /*
1268 * Before updating sk_refcnt, we must commit prior changes to memory
1269 * (Documentation/RCU/rculist_nulls.txt for details)
1270 */
1271 smp_wmb();
1272 atomic_set(&newsk->sk_refcnt, 2);
1273
1274 /*
1275 * Increment the counter in the same struct proto as the master
1276 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1277 * is the same as sk->sk_prot->socks, as this field was copied
1278 * with memcpy).
1279 *
1280 * This _changes_ the previous behaviour, where
1281 * tcp_create_openreq_child always was incrementing the
1282 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1283 * to be taken into account in all callers. -acme
1284 */
1285 sk_refcnt_debug_inc(newsk);
1286 sk_set_socket(newsk, NULL);
1287 newsk->sk_wq = NULL;
1288
1289 if (newsk->sk_prot->sockets_allocated)
1290 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1291
1292 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1293 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1294 net_enable_timestamp();
1295 }
1296 out:
1297 return newsk;
1298 }
1299 EXPORT_SYMBOL_GPL(sk_clone);
1300
1301 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1302 {
1303 __sk_dst_set(sk, dst);
1304 sk->sk_route_caps = dst->dev->features;
1305 if (sk->sk_route_caps & NETIF_F_GSO)
1306 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1307 sk->sk_route_caps &= ~sk->sk_route_nocaps;
1308 if (sk_can_gso(sk)) {
1309 if (dst->header_len) {
1310 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1311 } else {
1312 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1313 sk->sk_gso_max_size = dst->dev->gso_max_size;
1314 }
1315 }
1316 }
1317 EXPORT_SYMBOL_GPL(sk_setup_caps);
1318
1319 void __init sk_init(void)
1320 {
1321 if (totalram_pages <= 4096) {
1322 sysctl_wmem_max = 32767;
1323 sysctl_rmem_max = 32767;
1324 sysctl_wmem_default = 32767;
1325 sysctl_rmem_default = 32767;
1326 } else if (totalram_pages >= 131072) {
1327 sysctl_wmem_max = 131071;
1328 sysctl_rmem_max = 131071;
1329 }
1330 }
1331
1332 /*
1333 * Simple resource managers for sockets.
1334 */
1335
1336
1337 /*
1338 * Write buffer destructor automatically called from kfree_skb.
1339 */
1340 void sock_wfree(struct sk_buff *skb)
1341 {
1342 struct sock *sk = skb->sk;
1343 unsigned int len = skb->truesize;
1344
1345 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1346 /*
1347 * Keep a reference on sk_wmem_alloc, this will be released
1348 * after sk_write_space() call
1349 */
1350 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1351 sk->sk_write_space(sk);
1352 len = 1;
1353 }
1354 /*
1355 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1356 * could not do because of in-flight packets
1357 */
1358 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1359 __sk_free(sk);
1360 }
1361 EXPORT_SYMBOL(sock_wfree);
1362
1363 /*
1364 * Read buffer destructor automatically called from kfree_skb.
1365 */
1366 void sock_rfree(struct sk_buff *skb)
1367 {
1368 struct sock *sk = skb->sk;
1369 unsigned int len = skb->truesize;
1370
1371 atomic_sub(len, &sk->sk_rmem_alloc);
1372 sk_mem_uncharge(sk, len);
1373 }
1374 EXPORT_SYMBOL(sock_rfree);
1375
1376
1377 int sock_i_uid(struct sock *sk)
1378 {
1379 int uid;
1380
1381 read_lock_bh(&sk->sk_callback_lock);
1382 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1383 read_unlock_bh(&sk->sk_callback_lock);
1384 return uid;
1385 }
1386 EXPORT_SYMBOL(sock_i_uid);
1387
1388 unsigned long sock_i_ino(struct sock *sk)
1389 {
1390 unsigned long ino;
1391
1392 read_lock_bh(&sk->sk_callback_lock);
1393 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1394 read_unlock_bh(&sk->sk_callback_lock);
1395 return ino;
1396 }
1397 EXPORT_SYMBOL(sock_i_ino);
1398
1399 /*
1400 * Allocate a skb from the socket's send buffer.
1401 */
1402 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1403 gfp_t priority)
1404 {
1405 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1406 struct sk_buff *skb = alloc_skb(size, priority);
1407 if (skb) {
1408 skb_set_owner_w(skb, sk);
1409 return skb;
1410 }
1411 }
1412 return NULL;
1413 }
1414 EXPORT_SYMBOL(sock_wmalloc);
1415
1416 /*
1417 * Allocate a skb from the socket's receive buffer.
1418 */
1419 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1420 gfp_t priority)
1421 {
1422 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1423 struct sk_buff *skb = alloc_skb(size, priority);
1424 if (skb) {
1425 skb_set_owner_r(skb, sk);
1426 return skb;
1427 }
1428 }
1429 return NULL;
1430 }
1431
1432 /*
1433 * Allocate a memory block from the socket's option memory buffer.
1434 */
1435 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1436 {
1437 if ((unsigned)size <= sysctl_optmem_max &&
1438 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1439 void *mem;
1440 /* First do the add, to avoid the race if kmalloc
1441 * might sleep.
1442 */
1443 atomic_add(size, &sk->sk_omem_alloc);
1444 mem = kmalloc(size, priority);
1445 if (mem)
1446 return mem;
1447 atomic_sub(size, &sk->sk_omem_alloc);
1448 }
1449 return NULL;
1450 }
1451 EXPORT_SYMBOL(sock_kmalloc);
1452
1453 /*
1454 * Free an option memory block.
1455 */
1456 void sock_kfree_s(struct sock *sk, void *mem, int size)
1457 {
1458 kfree(mem);
1459 atomic_sub(size, &sk->sk_omem_alloc);
1460 }
1461 EXPORT_SYMBOL(sock_kfree_s);
1462
1463 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1464 I think, these locks should be removed for datagram sockets.
1465 */
1466 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1467 {
1468 DEFINE_WAIT(wait);
1469
1470 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1471 for (;;) {
1472 if (!timeo)
1473 break;
1474 if (signal_pending(current))
1475 break;
1476 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1477 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1478 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1479 break;
1480 if (sk->sk_shutdown & SEND_SHUTDOWN)
1481 break;
1482 if (sk->sk_err)
1483 break;
1484 timeo = schedule_timeout(timeo);
1485 }
1486 finish_wait(sk_sleep(sk), &wait);
1487 return timeo;
1488 }
1489
1490
1491 /*
1492 * Generic send/receive buffer handlers
1493 */
1494
1495 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1496 unsigned long data_len, int noblock,
1497 int *errcode)
1498 {
1499 struct sk_buff *skb;
1500 gfp_t gfp_mask;
1501 long timeo;
1502 int err;
1503
1504 gfp_mask = sk->sk_allocation;
1505 if (gfp_mask & __GFP_WAIT)
1506 gfp_mask |= __GFP_REPEAT;
1507
1508 timeo = sock_sndtimeo(sk, noblock);
1509 while (1) {
1510 err = sock_error(sk);
1511 if (err != 0)
1512 goto failure;
1513
1514 err = -EPIPE;
1515 if (sk->sk_shutdown & SEND_SHUTDOWN)
1516 goto failure;
1517
1518 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1519 skb = alloc_skb(header_len, gfp_mask);
1520 if (skb) {
1521 int npages;
1522 int i;
1523
1524 /* No pages, we're done... */
1525 if (!data_len)
1526 break;
1527
1528 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1529 skb->truesize += data_len;
1530 skb_shinfo(skb)->nr_frags = npages;
1531 for (i = 0; i < npages; i++) {
1532 struct page *page;
1533 skb_frag_t *frag;
1534
1535 page = alloc_pages(sk->sk_allocation, 0);
1536 if (!page) {
1537 err = -ENOBUFS;
1538 skb_shinfo(skb)->nr_frags = i;
1539 kfree_skb(skb);
1540 goto failure;
1541 }
1542
1543 frag = &skb_shinfo(skb)->frags[i];
1544 frag->page = page;
1545 frag->page_offset = 0;
1546 frag->size = (data_len >= PAGE_SIZE ?
1547 PAGE_SIZE :
1548 data_len);
1549 data_len -= PAGE_SIZE;
1550 }
1551
1552 /* Full success... */
1553 break;
1554 }
1555 err = -ENOBUFS;
1556 goto failure;
1557 }
1558 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1559 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1560 err = -EAGAIN;
1561 if (!timeo)
1562 goto failure;
1563 if (signal_pending(current))
1564 goto interrupted;
1565 timeo = sock_wait_for_wmem(sk, timeo);
1566 }
1567
1568 skb_set_owner_w(skb, sk);
1569 return skb;
1570
1571 interrupted:
1572 err = sock_intr_errno(timeo);
1573 failure:
1574 *errcode = err;
1575 return NULL;
1576 }
1577 EXPORT_SYMBOL(sock_alloc_send_pskb);
1578
1579 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1580 int noblock, int *errcode)
1581 {
1582 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1583 }
1584 EXPORT_SYMBOL(sock_alloc_send_skb);
1585
1586 static void __lock_sock(struct sock *sk)
1587 __releases(&sk->sk_lock.slock)
1588 __acquires(&sk->sk_lock.slock)
1589 {
1590 DEFINE_WAIT(wait);
1591
1592 for (;;) {
1593 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1594 TASK_UNINTERRUPTIBLE);
1595 spin_unlock_bh(&sk->sk_lock.slock);
1596 schedule();
1597 spin_lock_bh(&sk->sk_lock.slock);
1598 if (!sock_owned_by_user(sk))
1599 break;
1600 }
1601 finish_wait(&sk->sk_lock.wq, &wait);
1602 }
1603
1604 static void __release_sock(struct sock *sk)
1605 __releases(&sk->sk_lock.slock)
1606 __acquires(&sk->sk_lock.slock)
1607 {
1608 struct sk_buff *skb = sk->sk_backlog.head;
1609
1610 do {
1611 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1612 bh_unlock_sock(sk);
1613
1614 do {
1615 struct sk_buff *next = skb->next;
1616
1617 WARN_ON_ONCE(skb_dst_is_noref(skb));
1618 skb->next = NULL;
1619 sk_backlog_rcv(sk, skb);
1620
1621 /*
1622 * We are in process context here with softirqs
1623 * disabled, use cond_resched_softirq() to preempt.
1624 * This is safe to do because we've taken the backlog
1625 * queue private:
1626 */
1627 cond_resched_softirq();
1628
1629 skb = next;
1630 } while (skb != NULL);
1631
1632 bh_lock_sock(sk);
1633 } while ((skb = sk->sk_backlog.head) != NULL);
1634
1635 /*
1636 * Doing the zeroing here guarantee we can not loop forever
1637 * while a wild producer attempts to flood us.
1638 */
1639 sk->sk_backlog.len = 0;
1640 }
1641
1642 /**
1643 * sk_wait_data - wait for data to arrive at sk_receive_queue
1644 * @sk: sock to wait on
1645 * @timeo: for how long
1646 *
1647 * Now socket state including sk->sk_err is changed only under lock,
1648 * hence we may omit checks after joining wait queue.
1649 * We check receive queue before schedule() only as optimization;
1650 * it is very likely that release_sock() added new data.
1651 */
1652 int sk_wait_data(struct sock *sk, long *timeo)
1653 {
1654 int rc;
1655 DEFINE_WAIT(wait);
1656
1657 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1658 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1659 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1660 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1661 finish_wait(sk_sleep(sk), &wait);
1662 return rc;
1663 }
1664 EXPORT_SYMBOL(sk_wait_data);
1665
1666 /**
1667 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1668 * @sk: socket
1669 * @size: memory size to allocate
1670 * @kind: allocation type
1671 *
1672 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1673 * rmem allocation. This function assumes that protocols which have
1674 * memory_pressure use sk_wmem_queued as write buffer accounting.
1675 */
1676 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1677 {
1678 struct proto *prot = sk->sk_prot;
1679 int amt = sk_mem_pages(size);
1680 long allocated;
1681
1682 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1683 allocated = atomic_long_add_return(amt, prot->memory_allocated);
1684
1685 /* Under limit. */
1686 if (allocated <= prot->sysctl_mem[0]) {
1687 if (prot->memory_pressure && *prot->memory_pressure)
1688 *prot->memory_pressure = 0;
1689 return 1;
1690 }
1691
1692 /* Under pressure. */
1693 if (allocated > prot->sysctl_mem[1])
1694 if (prot->enter_memory_pressure)
1695 prot->enter_memory_pressure(sk);
1696
1697 /* Over hard limit. */
1698 if (allocated > prot->sysctl_mem[2])
1699 goto suppress_allocation;
1700
1701 /* guarantee minimum buffer size under pressure */
1702 if (kind == SK_MEM_RECV) {
1703 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1704 return 1;
1705 } else { /* SK_MEM_SEND */
1706 if (sk->sk_type == SOCK_STREAM) {
1707 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1708 return 1;
1709 } else if (atomic_read(&sk->sk_wmem_alloc) <
1710 prot->sysctl_wmem[0])
1711 return 1;
1712 }
1713
1714 if (prot->memory_pressure) {
1715 int alloc;
1716
1717 if (!*prot->memory_pressure)
1718 return 1;
1719 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1720 if (prot->sysctl_mem[2] > alloc *
1721 sk_mem_pages(sk->sk_wmem_queued +
1722 atomic_read(&sk->sk_rmem_alloc) +
1723 sk->sk_forward_alloc))
1724 return 1;
1725 }
1726
1727 suppress_allocation:
1728
1729 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1730 sk_stream_moderate_sndbuf(sk);
1731
1732 /* Fail only if socket is _under_ its sndbuf.
1733 * In this case we cannot block, so that we have to fail.
1734 */
1735 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1736 return 1;
1737 }
1738
1739 /* Alas. Undo changes. */
1740 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1741 atomic_long_sub(amt, prot->memory_allocated);
1742 return 0;
1743 }
1744 EXPORT_SYMBOL(__sk_mem_schedule);
1745
1746 /**
1747 * __sk_reclaim - reclaim memory_allocated
1748 * @sk: socket
1749 */
1750 void __sk_mem_reclaim(struct sock *sk)
1751 {
1752 struct proto *prot = sk->sk_prot;
1753
1754 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1755 prot->memory_allocated);
1756 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1757
1758 if (prot->memory_pressure && *prot->memory_pressure &&
1759 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1760 *prot->memory_pressure = 0;
1761 }
1762 EXPORT_SYMBOL(__sk_mem_reclaim);
1763
1764
1765 /*
1766 * Set of default routines for initialising struct proto_ops when
1767 * the protocol does not support a particular function. In certain
1768 * cases where it makes no sense for a protocol to have a "do nothing"
1769 * function, some default processing is provided.
1770 */
1771
1772 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1773 {
1774 return -EOPNOTSUPP;
1775 }
1776 EXPORT_SYMBOL(sock_no_bind);
1777
1778 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1779 int len, int flags)
1780 {
1781 return -EOPNOTSUPP;
1782 }
1783 EXPORT_SYMBOL(sock_no_connect);
1784
1785 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1786 {
1787 return -EOPNOTSUPP;
1788 }
1789 EXPORT_SYMBOL(sock_no_socketpair);
1790
1791 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1792 {
1793 return -EOPNOTSUPP;
1794 }
1795 EXPORT_SYMBOL(sock_no_accept);
1796
1797 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1798 int *len, int peer)
1799 {
1800 return -EOPNOTSUPP;
1801 }
1802 EXPORT_SYMBOL(sock_no_getname);
1803
1804 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1805 {
1806 return 0;
1807 }
1808 EXPORT_SYMBOL(sock_no_poll);
1809
1810 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1811 {
1812 return -EOPNOTSUPP;
1813 }
1814 EXPORT_SYMBOL(sock_no_ioctl);
1815
1816 int sock_no_listen(struct socket *sock, int backlog)
1817 {
1818 return -EOPNOTSUPP;
1819 }
1820 EXPORT_SYMBOL(sock_no_listen);
1821
1822 int sock_no_shutdown(struct socket *sock, int how)
1823 {
1824 return -EOPNOTSUPP;
1825 }
1826 EXPORT_SYMBOL(sock_no_shutdown);
1827
1828 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1829 char __user *optval, unsigned int optlen)
1830 {
1831 return -EOPNOTSUPP;
1832 }
1833 EXPORT_SYMBOL(sock_no_setsockopt);
1834
1835 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1836 char __user *optval, int __user *optlen)
1837 {
1838 return -EOPNOTSUPP;
1839 }
1840 EXPORT_SYMBOL(sock_no_getsockopt);
1841
1842 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1843 size_t len)
1844 {
1845 return -EOPNOTSUPP;
1846 }
1847 EXPORT_SYMBOL(sock_no_sendmsg);
1848
1849 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1850 size_t len, int flags)
1851 {
1852 return -EOPNOTSUPP;
1853 }
1854 EXPORT_SYMBOL(sock_no_recvmsg);
1855
1856 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1857 {
1858 /* Mirror missing mmap method error code */
1859 return -ENODEV;
1860 }
1861 EXPORT_SYMBOL(sock_no_mmap);
1862
1863 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1864 {
1865 ssize_t res;
1866 struct msghdr msg = {.msg_flags = flags};
1867 struct kvec iov;
1868 char *kaddr = kmap(page);
1869 iov.iov_base = kaddr + offset;
1870 iov.iov_len = size;
1871 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1872 kunmap(page);
1873 return res;
1874 }
1875 EXPORT_SYMBOL(sock_no_sendpage);
1876
1877 /*
1878 * Default Socket Callbacks
1879 */
1880
1881 static void sock_def_wakeup(struct sock *sk)
1882 {
1883 struct socket_wq *wq;
1884
1885 rcu_read_lock();
1886 wq = rcu_dereference(sk->sk_wq);
1887 if (wq_has_sleeper(wq))
1888 wake_up_interruptible_all(&wq->wait);
1889 rcu_read_unlock();
1890 }
1891
1892 static void sock_def_error_report(struct sock *sk)
1893 {
1894 struct socket_wq *wq;
1895
1896 rcu_read_lock();
1897 wq = rcu_dereference(sk->sk_wq);
1898 if (wq_has_sleeper(wq))
1899 wake_up_interruptible_poll(&wq->wait, POLLERR);
1900 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1901 rcu_read_unlock();
1902 }
1903
1904 static void sock_def_readable(struct sock *sk, int len)
1905 {
1906 struct socket_wq *wq;
1907
1908 rcu_read_lock();
1909 wq = rcu_dereference(sk->sk_wq);
1910 if (wq_has_sleeper(wq))
1911 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1912 POLLRDNORM | POLLRDBAND);
1913 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1914 rcu_read_unlock();
1915 }
1916
1917 static void sock_def_write_space(struct sock *sk)
1918 {
1919 struct socket_wq *wq;
1920
1921 rcu_read_lock();
1922
1923 /* Do not wake up a writer until he can make "significant"
1924 * progress. --DaveM
1925 */
1926 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1927 wq = rcu_dereference(sk->sk_wq);
1928 if (wq_has_sleeper(wq))
1929 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1930 POLLWRNORM | POLLWRBAND);
1931
1932 /* Should agree with poll, otherwise some programs break */
1933 if (sock_writeable(sk))
1934 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1935 }
1936
1937 rcu_read_unlock();
1938 }
1939
1940 static void sock_def_destruct(struct sock *sk)
1941 {
1942 kfree(sk->sk_protinfo);
1943 }
1944
1945 void sk_send_sigurg(struct sock *sk)
1946 {
1947 if (sk->sk_socket && sk->sk_socket->file)
1948 if (send_sigurg(&sk->sk_socket->file->f_owner))
1949 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1950 }
1951 EXPORT_SYMBOL(sk_send_sigurg);
1952
1953 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1954 unsigned long expires)
1955 {
1956 if (!mod_timer(timer, expires))
1957 sock_hold(sk);
1958 }
1959 EXPORT_SYMBOL(sk_reset_timer);
1960
1961 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1962 {
1963 if (timer_pending(timer) && del_timer(timer))
1964 __sock_put(sk);
1965 }
1966 EXPORT_SYMBOL(sk_stop_timer);
1967
1968 void sock_init_data(struct socket *sock, struct sock *sk)
1969 {
1970 skb_queue_head_init(&sk->sk_receive_queue);
1971 skb_queue_head_init(&sk->sk_write_queue);
1972 skb_queue_head_init(&sk->sk_error_queue);
1973 #ifdef CONFIG_NET_DMA
1974 skb_queue_head_init(&sk->sk_async_wait_queue);
1975 #endif
1976
1977 sk->sk_send_head = NULL;
1978
1979 init_timer(&sk->sk_timer);
1980
1981 sk->sk_allocation = GFP_KERNEL;
1982 sk->sk_rcvbuf = sysctl_rmem_default;
1983 sk->sk_sndbuf = sysctl_wmem_default;
1984 sk->sk_state = TCP_CLOSE;
1985 sk_set_socket(sk, sock);
1986
1987 sock_set_flag(sk, SOCK_ZAPPED);
1988
1989 if (sock) {
1990 sk->sk_type = sock->type;
1991 sk->sk_wq = sock->wq;
1992 sock->sk = sk;
1993 } else
1994 sk->sk_wq = NULL;
1995
1996 spin_lock_init(&sk->sk_dst_lock);
1997 rwlock_init(&sk->sk_callback_lock);
1998 lockdep_set_class_and_name(&sk->sk_callback_lock,
1999 af_callback_keys + sk->sk_family,
2000 af_family_clock_key_strings[sk->sk_family]);
2001
2002 sk->sk_state_change = sock_def_wakeup;
2003 sk->sk_data_ready = sock_def_readable;
2004 sk->sk_write_space = sock_def_write_space;
2005 sk->sk_error_report = sock_def_error_report;
2006 sk->sk_destruct = sock_def_destruct;
2007
2008 sk->sk_sndmsg_page = NULL;
2009 sk->sk_sndmsg_off = 0;
2010
2011 sk->sk_peer_pid = NULL;
2012 sk->sk_peer_cred = NULL;
2013 sk->sk_write_pending = 0;
2014 sk->sk_rcvlowat = 1;
2015 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2016 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2017
2018 sk->sk_stamp = ktime_set(-1L, 0);
2019
2020 /*
2021 * Before updating sk_refcnt, we must commit prior changes to memory
2022 * (Documentation/RCU/rculist_nulls.txt for details)
2023 */
2024 smp_wmb();
2025 atomic_set(&sk->sk_refcnt, 1);
2026 atomic_set(&sk->sk_drops, 0);
2027 }
2028 EXPORT_SYMBOL(sock_init_data);
2029
2030 void lock_sock_nested(struct sock *sk, int subclass)
2031 {
2032 might_sleep();
2033 spin_lock_bh(&sk->sk_lock.slock);
2034 if (sk->sk_lock.owned)
2035 __lock_sock(sk);
2036 sk->sk_lock.owned = 1;
2037 spin_unlock(&sk->sk_lock.slock);
2038 /*
2039 * The sk_lock has mutex_lock() semantics here:
2040 */
2041 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2042 local_bh_enable();
2043 }
2044 EXPORT_SYMBOL(lock_sock_nested);
2045
2046 void release_sock(struct sock *sk)
2047 {
2048 /*
2049 * The sk_lock has mutex_unlock() semantics:
2050 */
2051 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2052
2053 spin_lock_bh(&sk->sk_lock.slock);
2054 if (sk->sk_backlog.tail)
2055 __release_sock(sk);
2056 sk->sk_lock.owned = 0;
2057 if (waitqueue_active(&sk->sk_lock.wq))
2058 wake_up(&sk->sk_lock.wq);
2059 spin_unlock_bh(&sk->sk_lock.slock);
2060 }
2061 EXPORT_SYMBOL(release_sock);
2062
2063 /**
2064 * lock_sock_fast - fast version of lock_sock
2065 * @sk: socket
2066 *
2067 * This version should be used for very small section, where process wont block
2068 * return false if fast path is taken
2069 * sk_lock.slock locked, owned = 0, BH disabled
2070 * return true if slow path is taken
2071 * sk_lock.slock unlocked, owned = 1, BH enabled
2072 */
2073 bool lock_sock_fast(struct sock *sk)
2074 {
2075 might_sleep();
2076 spin_lock_bh(&sk->sk_lock.slock);
2077
2078 if (!sk->sk_lock.owned)
2079 /*
2080 * Note : We must disable BH
2081 */
2082 return false;
2083
2084 __lock_sock(sk);
2085 sk->sk_lock.owned = 1;
2086 spin_unlock(&sk->sk_lock.slock);
2087 /*
2088 * The sk_lock has mutex_lock() semantics here:
2089 */
2090 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2091 local_bh_enable();
2092 return true;
2093 }
2094 EXPORT_SYMBOL(lock_sock_fast);
2095
2096 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2097 {
2098 struct timeval tv;
2099 if (!sock_flag(sk, SOCK_TIMESTAMP))
2100 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2101 tv = ktime_to_timeval(sk->sk_stamp);
2102 if (tv.tv_sec == -1)
2103 return -ENOENT;
2104 if (tv.tv_sec == 0) {
2105 sk->sk_stamp = ktime_get_real();
2106 tv = ktime_to_timeval(sk->sk_stamp);
2107 }
2108 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2109 }
2110 EXPORT_SYMBOL(sock_get_timestamp);
2111
2112 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2113 {
2114 struct timespec ts;
2115 if (!sock_flag(sk, SOCK_TIMESTAMP))
2116 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2117 ts = ktime_to_timespec(sk->sk_stamp);
2118 if (ts.tv_sec == -1)
2119 return -ENOENT;
2120 if (ts.tv_sec == 0) {
2121 sk->sk_stamp = ktime_get_real();
2122 ts = ktime_to_timespec(sk->sk_stamp);
2123 }
2124 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2125 }
2126 EXPORT_SYMBOL(sock_get_timestampns);
2127
2128 void sock_enable_timestamp(struct sock *sk, int flag)
2129 {
2130 if (!sock_flag(sk, flag)) {
2131 sock_set_flag(sk, flag);
2132 /*
2133 * we just set one of the two flags which require net
2134 * time stamping, but time stamping might have been on
2135 * already because of the other one
2136 */
2137 if (!sock_flag(sk,
2138 flag == SOCK_TIMESTAMP ?
2139 SOCK_TIMESTAMPING_RX_SOFTWARE :
2140 SOCK_TIMESTAMP))
2141 net_enable_timestamp();
2142 }
2143 }
2144
2145 /*
2146 * Get a socket option on an socket.
2147 *
2148 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2149 * asynchronous errors should be reported by getsockopt. We assume
2150 * this means if you specify SO_ERROR (otherwise whats the point of it).
2151 */
2152 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2153 char __user *optval, int __user *optlen)
2154 {
2155 struct sock *sk = sock->sk;
2156
2157 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2158 }
2159 EXPORT_SYMBOL(sock_common_getsockopt);
2160
2161 #ifdef CONFIG_COMPAT
2162 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2163 char __user *optval, int __user *optlen)
2164 {
2165 struct sock *sk = sock->sk;
2166
2167 if (sk->sk_prot->compat_getsockopt != NULL)
2168 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2169 optval, optlen);
2170 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2171 }
2172 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2173 #endif
2174
2175 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2176 struct msghdr *msg, size_t size, int flags)
2177 {
2178 struct sock *sk = sock->sk;
2179 int addr_len = 0;
2180 int err;
2181
2182 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2183 flags & ~MSG_DONTWAIT, &addr_len);
2184 if (err >= 0)
2185 msg->msg_namelen = addr_len;
2186 return err;
2187 }
2188 EXPORT_SYMBOL(sock_common_recvmsg);
2189
2190 /*
2191 * Set socket options on an inet socket.
2192 */
2193 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2194 char __user *optval, unsigned int optlen)
2195 {
2196 struct sock *sk = sock->sk;
2197
2198 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2199 }
2200 EXPORT_SYMBOL(sock_common_setsockopt);
2201
2202 #ifdef CONFIG_COMPAT
2203 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2204 char __user *optval, unsigned int optlen)
2205 {
2206 struct sock *sk = sock->sk;
2207
2208 if (sk->sk_prot->compat_setsockopt != NULL)
2209 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2210 optval, optlen);
2211 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2212 }
2213 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2214 #endif
2215
2216 void sk_common_release(struct sock *sk)
2217 {
2218 if (sk->sk_prot->destroy)
2219 sk->sk_prot->destroy(sk);
2220
2221 /*
2222 * Observation: when sock_common_release is called, processes have
2223 * no access to socket. But net still has.
2224 * Step one, detach it from networking:
2225 *
2226 * A. Remove from hash tables.
2227 */
2228
2229 sk->sk_prot->unhash(sk);
2230
2231 /*
2232 * In this point socket cannot receive new packets, but it is possible
2233 * that some packets are in flight because some CPU runs receiver and
2234 * did hash table lookup before we unhashed socket. They will achieve
2235 * receive queue and will be purged by socket destructor.
2236 *
2237 * Also we still have packets pending on receive queue and probably,
2238 * our own packets waiting in device queues. sock_destroy will drain
2239 * receive queue, but transmitted packets will delay socket destruction
2240 * until the last reference will be released.
2241 */
2242
2243 sock_orphan(sk);
2244
2245 xfrm_sk_free_policy(sk);
2246
2247 sk_refcnt_debug_release(sk);
2248 sock_put(sk);
2249 }
2250 EXPORT_SYMBOL(sk_common_release);
2251
2252 static DEFINE_RWLOCK(proto_list_lock);
2253 static LIST_HEAD(proto_list);
2254
2255 #ifdef CONFIG_PROC_FS
2256 #define PROTO_INUSE_NR 64 /* should be enough for the first time */
2257 struct prot_inuse {
2258 int val[PROTO_INUSE_NR];
2259 };
2260
2261 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2262
2263 #ifdef CONFIG_NET_NS
2264 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2265 {
2266 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2267 }
2268 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2269
2270 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2271 {
2272 int cpu, idx = prot->inuse_idx;
2273 int res = 0;
2274
2275 for_each_possible_cpu(cpu)
2276 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2277
2278 return res >= 0 ? res : 0;
2279 }
2280 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2281
2282 static int __net_init sock_inuse_init_net(struct net *net)
2283 {
2284 net->core.inuse = alloc_percpu(struct prot_inuse);
2285 return net->core.inuse ? 0 : -ENOMEM;
2286 }
2287
2288 static void __net_exit sock_inuse_exit_net(struct net *net)
2289 {
2290 free_percpu(net->core.inuse);
2291 }
2292
2293 static struct pernet_operations net_inuse_ops = {
2294 .init = sock_inuse_init_net,
2295 .exit = sock_inuse_exit_net,
2296 };
2297
2298 static __init int net_inuse_init(void)
2299 {
2300 if (register_pernet_subsys(&net_inuse_ops))
2301 panic("Cannot initialize net inuse counters");
2302
2303 return 0;
2304 }
2305
2306 core_initcall(net_inuse_init);
2307 #else
2308 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2309
2310 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2311 {
2312 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2313 }
2314 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2315
2316 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2317 {
2318 int cpu, idx = prot->inuse_idx;
2319 int res = 0;
2320
2321 for_each_possible_cpu(cpu)
2322 res += per_cpu(prot_inuse, cpu).val[idx];
2323
2324 return res >= 0 ? res : 0;
2325 }
2326 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2327 #endif
2328
2329 static void assign_proto_idx(struct proto *prot)
2330 {
2331 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2332
2333 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2334 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2335 return;
2336 }
2337
2338 set_bit(prot->inuse_idx, proto_inuse_idx);
2339 }
2340
2341 static void release_proto_idx(struct proto *prot)
2342 {
2343 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2344 clear_bit(prot->inuse_idx, proto_inuse_idx);
2345 }
2346 #else
2347 static inline void assign_proto_idx(struct proto *prot)
2348 {
2349 }
2350
2351 static inline void release_proto_idx(struct proto *prot)
2352 {
2353 }
2354 #endif
2355
2356 int proto_register(struct proto *prot, int alloc_slab)
2357 {
2358 if (alloc_slab) {
2359 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2360 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2361 NULL);
2362
2363 if (prot->slab == NULL) {
2364 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2365 prot->name);
2366 goto out;
2367 }
2368
2369 if (prot->rsk_prot != NULL) {
2370 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2371 if (prot->rsk_prot->slab_name == NULL)
2372 goto out_free_sock_slab;
2373
2374 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2375 prot->rsk_prot->obj_size, 0,
2376 SLAB_HWCACHE_ALIGN, NULL);
2377
2378 if (prot->rsk_prot->slab == NULL) {
2379 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2380 prot->name);
2381 goto out_free_request_sock_slab_name;
2382 }
2383 }
2384
2385 if (prot->twsk_prot != NULL) {
2386 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2387
2388 if (prot->twsk_prot->twsk_slab_name == NULL)
2389 goto out_free_request_sock_slab;
2390
2391 prot->twsk_prot->twsk_slab =
2392 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2393 prot->twsk_prot->twsk_obj_size,
2394 0,
2395 SLAB_HWCACHE_ALIGN |
2396 prot->slab_flags,
2397 NULL);
2398 if (prot->twsk_prot->twsk_slab == NULL)
2399 goto out_free_timewait_sock_slab_name;
2400 }
2401 }
2402
2403 write_lock(&proto_list_lock);
2404 list_add(&prot->node, &proto_list);
2405 assign_proto_idx(prot);
2406 write_unlock(&proto_list_lock);
2407 return 0;
2408
2409 out_free_timewait_sock_slab_name:
2410 kfree(prot->twsk_prot->twsk_slab_name);
2411 out_free_request_sock_slab:
2412 if (prot->rsk_prot && prot->rsk_prot->slab) {
2413 kmem_cache_destroy(prot->rsk_prot->slab);
2414 prot->rsk_prot->slab = NULL;
2415 }
2416 out_free_request_sock_slab_name:
2417 if (prot->rsk_prot)
2418 kfree(prot->rsk_prot->slab_name);
2419 out_free_sock_slab:
2420 kmem_cache_destroy(prot->slab);
2421 prot->slab = NULL;
2422 out:
2423 return -ENOBUFS;
2424 }
2425 EXPORT_SYMBOL(proto_register);
2426
2427 void proto_unregister(struct proto *prot)
2428 {
2429 write_lock(&proto_list_lock);
2430 release_proto_idx(prot);
2431 list_del(&prot->node);
2432 write_unlock(&proto_list_lock);
2433
2434 if (prot->slab != NULL) {
2435 kmem_cache_destroy(prot->slab);
2436 prot->slab = NULL;
2437 }
2438
2439 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2440 kmem_cache_destroy(prot->rsk_prot->slab);
2441 kfree(prot->rsk_prot->slab_name);
2442 prot->rsk_prot->slab = NULL;
2443 }
2444
2445 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2446 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2447 kfree(prot->twsk_prot->twsk_slab_name);
2448 prot->twsk_prot->twsk_slab = NULL;
2449 }
2450 }
2451 EXPORT_SYMBOL(proto_unregister);
2452
2453 #ifdef CONFIG_PROC_FS
2454 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2455 __acquires(proto_list_lock)
2456 {
2457 read_lock(&proto_list_lock);
2458 return seq_list_start_head(&proto_list, *pos);
2459 }
2460
2461 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2462 {
2463 return seq_list_next(v, &proto_list, pos);
2464 }
2465
2466 static void proto_seq_stop(struct seq_file *seq, void *v)
2467 __releases(proto_list_lock)
2468 {
2469 read_unlock(&proto_list_lock);
2470 }
2471
2472 static char proto_method_implemented(const void *method)
2473 {
2474 return method == NULL ? 'n' : 'y';
2475 }
2476
2477 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2478 {
2479 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
2480 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2481 proto->name,
2482 proto->obj_size,
2483 sock_prot_inuse_get(seq_file_net(seq), proto),
2484 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2485 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2486 proto->max_header,
2487 proto->slab == NULL ? "no" : "yes",
2488 module_name(proto->owner),
2489 proto_method_implemented(proto->close),
2490 proto_method_implemented(proto->connect),
2491 proto_method_implemented(proto->disconnect),
2492 proto_method_implemented(proto->accept),
2493 proto_method_implemented(proto->ioctl),
2494 proto_method_implemented(proto->init),
2495 proto_method_implemented(proto->destroy),
2496 proto_method_implemented(proto->shutdown),
2497 proto_method_implemented(proto->setsockopt),
2498 proto_method_implemented(proto->getsockopt),
2499 proto_method_implemented(proto->sendmsg),
2500 proto_method_implemented(proto->recvmsg),
2501 proto_method_implemented(proto->sendpage),
2502 proto_method_implemented(proto->bind),
2503 proto_method_implemented(proto->backlog_rcv),
2504 proto_method_implemented(proto->hash),
2505 proto_method_implemented(proto->unhash),
2506 proto_method_implemented(proto->get_port),
2507 proto_method_implemented(proto->enter_memory_pressure));
2508 }
2509
2510 static int proto_seq_show(struct seq_file *seq, void *v)
2511 {
2512 if (v == &proto_list)
2513 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2514 "protocol",
2515 "size",
2516 "sockets",
2517 "memory",
2518 "press",
2519 "maxhdr",
2520 "slab",
2521 "module",
2522 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2523 else
2524 proto_seq_printf(seq, list_entry(v, struct proto, node));
2525 return 0;
2526 }
2527
2528 static const struct seq_operations proto_seq_ops = {
2529 .start = proto_seq_start,
2530 .next = proto_seq_next,
2531 .stop = proto_seq_stop,
2532 .show = proto_seq_show,
2533 };
2534
2535 static int proto_seq_open(struct inode *inode, struct file *file)
2536 {
2537 return seq_open_net(inode, file, &proto_seq_ops,
2538 sizeof(struct seq_net_private));
2539 }
2540
2541 static const struct file_operations proto_seq_fops = {
2542 .owner = THIS_MODULE,
2543 .open = proto_seq_open,
2544 .read = seq_read,
2545 .llseek = seq_lseek,
2546 .release = seq_release_net,
2547 };
2548
2549 static __net_init int proto_init_net(struct net *net)
2550 {
2551 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2552 return -ENOMEM;
2553
2554 return 0;
2555 }
2556
2557 static __net_exit void proto_exit_net(struct net *net)
2558 {
2559 proc_net_remove(net, "protocols");
2560 }
2561
2562
2563 static __net_initdata struct pernet_operations proto_net_ops = {
2564 .init = proto_init_net,
2565 .exit = proto_exit_net,
2566 };
2567
2568 static int __init proto_init(void)
2569 {
2570 return register_pernet_subsys(&proto_net_ops);
2571 }
2572
2573 subsys_initcall(proto_init);
2574
2575 #endif /* PROC_FS */
linux-next