db8335ad7559d1141c079127279b257b74c0587d
[linux-3.10.git] / net / core / sock.c
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" ,
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" ,
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" ,
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 ancilliary 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
766 int sock_getsockopt(struct socket *sock, int level, int optname,
767                     char __user *optval, int __user *optlen)
768 {
769         struct sock *sk = sock->sk;
770
771         union {
772                 int val;
773                 struct linger ling;
774                 struct timeval tm;
775         } v;
776
777         int lv = sizeof(int);
778         int len;
779
780         if (get_user(len, optlen))
781                 return -EFAULT;
782         if (len < 0)
783                 return -EINVAL;
784
785         memset(&v, 0, sizeof(v));
786
787         switch (optname) {
788         case SO_DEBUG:
789                 v.val = sock_flag(sk, SOCK_DBG);
790                 break;
791
792         case SO_DONTROUTE:
793                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
794                 break;
795
796         case SO_BROADCAST:
797                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
798                 break;
799
800         case SO_SNDBUF:
801                 v.val = sk->sk_sndbuf;
802                 break;
803
804         case SO_RCVBUF:
805                 v.val = sk->sk_rcvbuf;
806                 break;
807
808         case SO_REUSEADDR:
809                 v.val = sk->sk_reuse;
810                 break;
811
812         case SO_KEEPALIVE:
813                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
814                 break;
815
816         case SO_TYPE:
817                 v.val = sk->sk_type;
818                 break;
819
820         case SO_PROTOCOL:
821                 v.val = sk->sk_protocol;
822                 break;
823
824         case SO_DOMAIN:
825                 v.val = sk->sk_family;
826                 break;
827
828         case SO_ERROR:
829                 v.val = -sock_error(sk);
830                 if (v.val == 0)
831                         v.val = xchg(&sk->sk_err_soft, 0);
832                 break;
833
834         case SO_OOBINLINE:
835                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
836                 break;
837
838         case SO_NO_CHECK:
839                 v.val = sk->sk_no_check;
840                 break;
841
842         case SO_PRIORITY:
843                 v.val = sk->sk_priority;
844                 break;
845
846         case SO_LINGER:
847                 lv              = sizeof(v.ling);
848                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
849                 v.ling.l_linger = sk->sk_lingertime / HZ;
850                 break;
851
852         case SO_BSDCOMPAT:
853                 sock_warn_obsolete_bsdism("getsockopt");
854                 break;
855
856         case SO_TIMESTAMP:
857                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
858                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
859                 break;
860
861         case SO_TIMESTAMPNS:
862                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
863                 break;
864
865         case SO_TIMESTAMPING:
866                 v.val = 0;
867                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
868                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
869                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
870                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
871                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
872                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
873                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
874                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
875                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
876                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
877                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
878                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
879                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
880                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
881                 break;
882
883         case SO_RCVTIMEO:
884                 lv = sizeof(struct timeval);
885                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
886                         v.tm.tv_sec = 0;
887                         v.tm.tv_usec = 0;
888                 } else {
889                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
890                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
891                 }
892                 break;
893
894         case SO_SNDTIMEO:
895                 lv = sizeof(struct timeval);
896                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
897                         v.tm.tv_sec = 0;
898                         v.tm.tv_usec = 0;
899                 } else {
900                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
901                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
902                 }
903                 break;
904
905         case SO_RCVLOWAT:
906                 v.val = sk->sk_rcvlowat;
907                 break;
908
909         case SO_SNDLOWAT:
910                 v.val = 1;
911                 break;
912
913         case SO_PASSCRED:
914                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
915                 break;
916
917         case SO_PEERCRED:
918                 if (len > sizeof(sk->sk_peercred))
919                         len = sizeof(sk->sk_peercred);
920                 if (copy_to_user(optval, &sk->sk_peercred, len))
921                         return -EFAULT;
922                 goto lenout;
923
924         case SO_PEERNAME:
925         {
926                 char address[128];
927
928                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
929                         return -ENOTCONN;
930                 if (lv < len)
931                         return -EINVAL;
932                 if (copy_to_user(optval, address, len))
933                         return -EFAULT;
934                 goto lenout;
935         }
936
937         /* Dubious BSD thing... Probably nobody even uses it, but
938          * the UNIX standard wants it for whatever reason... -DaveM
939          */
940         case SO_ACCEPTCONN:
941                 v.val = sk->sk_state == TCP_LISTEN;
942                 break;
943
944         case SO_PASSSEC:
945                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
946                 break;
947
948         case SO_PEERSEC:
949                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
950
951         case SO_MARK:
952                 v.val = sk->sk_mark;
953                 break;
954
955         case SO_RXQ_OVFL:
956                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
957                 break;
958
959         default:
960                 return -ENOPROTOOPT;
961         }
962
963         if (len > lv)
964                 len = lv;
965         if (copy_to_user(optval, &v, len))
966                 return -EFAULT;
967 lenout:
968         if (put_user(len, optlen))
969                 return -EFAULT;
970         return 0;
971 }
972
973 /*
974  * Initialize an sk_lock.
975  *
976  * (We also register the sk_lock with the lock validator.)
977  */
978 static inline void sock_lock_init(struct sock *sk)
979 {
980         sock_lock_init_class_and_name(sk,
981                         af_family_slock_key_strings[sk->sk_family],
982                         af_family_slock_keys + sk->sk_family,
983                         af_family_key_strings[sk->sk_family],
984                         af_family_keys + sk->sk_family);
985 }
986
987 /*
988  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
989  * even temporarly, because of RCU lookups. sk_node should also be left as is.
990  */
991 static void sock_copy(struct sock *nsk, const struct sock *osk)
992 {
993 #ifdef CONFIG_SECURITY_NETWORK
994         void *sptr = nsk->sk_security;
995 #endif
996         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
997                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
998                      sizeof(osk->sk_tx_queue_mapping));
999         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
1000                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
1001 #ifdef CONFIG_SECURITY_NETWORK
1002         nsk->sk_security = sptr;
1003         security_sk_clone(osk, nsk);
1004 #endif
1005 }
1006
1007 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1008                 int family)
1009 {
1010         struct sock *sk;
1011         struct kmem_cache *slab;
1012
1013         slab = prot->slab;
1014         if (slab != NULL) {
1015                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1016                 if (!sk)
1017                         return sk;
1018                 if (priority & __GFP_ZERO) {
1019                         /*
1020                          * caches using SLAB_DESTROY_BY_RCU should let
1021                          * sk_node.next un-modified. Special care is taken
1022                          * when initializing object to zero.
1023                          */
1024                         if (offsetof(struct sock, sk_node.next) != 0)
1025                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
1026                         memset(&sk->sk_node.pprev, 0,
1027                                prot->obj_size - offsetof(struct sock,
1028                                                          sk_node.pprev));
1029                 }
1030         }
1031         else
1032                 sk = kmalloc(prot->obj_size, priority);
1033
1034         if (sk != NULL) {
1035                 kmemcheck_annotate_bitfield(sk, flags);
1036
1037                 if (security_sk_alloc(sk, family, priority))
1038                         goto out_free;
1039
1040                 if (!try_module_get(prot->owner))
1041                         goto out_free_sec;
1042                 sk_tx_queue_clear(sk);
1043         }
1044
1045         return sk;
1046
1047 out_free_sec:
1048         security_sk_free(sk);
1049 out_free:
1050         if (slab != NULL)
1051                 kmem_cache_free(slab, sk);
1052         else
1053                 kfree(sk);
1054         return NULL;
1055 }
1056
1057 static void sk_prot_free(struct proto *prot, struct sock *sk)
1058 {
1059         struct kmem_cache *slab;
1060         struct module *owner;
1061
1062         owner = prot->owner;
1063         slab = prot->slab;
1064
1065         security_sk_free(sk);
1066         if (slab != NULL)
1067                 kmem_cache_free(slab, sk);
1068         else
1069                 kfree(sk);
1070         module_put(owner);
1071 }
1072
1073 #ifdef CONFIG_CGROUPS
1074 void sock_update_classid(struct sock *sk)
1075 {
1076         u32 classid = task_cls_classid(current);
1077
1078         if (classid && classid != sk->sk_classid)
1079                 sk->sk_classid = classid;
1080 }
1081 EXPORT_SYMBOL(sock_update_classid);
1082 #endif
1083
1084 /**
1085  *      sk_alloc - All socket objects are allocated here
1086  *      @net: the applicable net namespace
1087  *      @family: protocol family
1088  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1089  *      @prot: struct proto associated with this new sock instance
1090  */
1091 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1092                       struct proto *prot)
1093 {
1094         struct sock *sk;
1095
1096         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1097         if (sk) {
1098                 sk->sk_family = family;
1099                 /*
1100                  * See comment in struct sock definition to understand
1101                  * why we need sk_prot_creator -acme
1102                  */
1103                 sk->sk_prot = sk->sk_prot_creator = prot;
1104                 sock_lock_init(sk);
1105                 sock_net_set(sk, get_net(net));
1106                 atomic_set(&sk->sk_wmem_alloc, 1);
1107
1108                 sock_update_classid(sk);
1109         }
1110
1111         return sk;
1112 }
1113 EXPORT_SYMBOL(sk_alloc);
1114
1115 static void __sk_free(struct sock *sk)
1116 {
1117         struct sk_filter *filter;
1118
1119         if (sk->sk_destruct)
1120                 sk->sk_destruct(sk);
1121
1122         filter = rcu_dereference_check(sk->sk_filter,
1123                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1124         if (filter) {
1125                 sk_filter_uncharge(sk, filter);
1126                 rcu_assign_pointer(sk->sk_filter, NULL);
1127         }
1128
1129         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1130         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1131
1132         if (atomic_read(&sk->sk_omem_alloc))
1133                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1134                        __func__, atomic_read(&sk->sk_omem_alloc));
1135
1136         put_net(sock_net(sk));
1137         sk_prot_free(sk->sk_prot_creator, sk);
1138 }
1139
1140 void sk_free(struct sock *sk)
1141 {
1142         /*
1143          * We substract one from sk_wmem_alloc and can know if
1144          * some packets are still in some tx queue.
1145          * If not null, sock_wfree() will call __sk_free(sk) later
1146          */
1147         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1148                 __sk_free(sk);
1149 }
1150 EXPORT_SYMBOL(sk_free);
1151
1152 /*
1153  * Last sock_put should drop referrence to sk->sk_net. It has already
1154  * been dropped in sk_change_net. Taking referrence to stopping namespace
1155  * is not an option.
1156  * Take referrence to a socket to remove it from hash _alive_ and after that
1157  * destroy it in the context of init_net.
1158  */
1159 void sk_release_kernel(struct sock *sk)
1160 {
1161         if (sk == NULL || sk->sk_socket == NULL)
1162                 return;
1163
1164         sock_hold(sk);
1165         sock_release(sk->sk_socket);
1166         release_net(sock_net(sk));
1167         sock_net_set(sk, get_net(&init_net));
1168         sock_put(sk);
1169 }
1170 EXPORT_SYMBOL(sk_release_kernel);
1171
1172 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1173 {
1174         struct sock *newsk;
1175
1176         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1177         if (newsk != NULL) {
1178                 struct sk_filter *filter;
1179
1180                 sock_copy(newsk, sk);
1181
1182                 /* SANITY */
1183                 get_net(sock_net(newsk));
1184                 sk_node_init(&newsk->sk_node);
1185                 sock_lock_init(newsk);
1186                 bh_lock_sock(newsk);
1187                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1188                 newsk->sk_backlog.len = 0;
1189
1190                 atomic_set(&newsk->sk_rmem_alloc, 0);
1191                 /*
1192                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1193                  */
1194                 atomic_set(&newsk->sk_wmem_alloc, 1);
1195                 atomic_set(&newsk->sk_omem_alloc, 0);
1196                 skb_queue_head_init(&newsk->sk_receive_queue);
1197                 skb_queue_head_init(&newsk->sk_write_queue);
1198 #ifdef CONFIG_NET_DMA
1199                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1200 #endif
1201
1202                 spin_lock_init(&newsk->sk_dst_lock);
1203                 rwlock_init(&newsk->sk_callback_lock);
1204                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1205                                 af_callback_keys + newsk->sk_family,
1206                                 af_family_clock_key_strings[newsk->sk_family]);
1207
1208                 newsk->sk_dst_cache     = NULL;
1209                 newsk->sk_wmem_queued   = 0;
1210                 newsk->sk_forward_alloc = 0;
1211                 newsk->sk_send_head     = NULL;
1212                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1213
1214                 sock_reset_flag(newsk, SOCK_DONE);
1215                 skb_queue_head_init(&newsk->sk_error_queue);
1216
1217                 filter = newsk->sk_filter;
1218                 if (filter != NULL)
1219                         sk_filter_charge(newsk, filter);
1220
1221                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1222                         /* It is still raw copy of parent, so invalidate
1223                          * destructor and make plain sk_free() */
1224                         newsk->sk_destruct = NULL;
1225                         sk_free(newsk);
1226                         newsk = NULL;
1227                         goto out;
1228                 }
1229
1230                 newsk->sk_err      = 0;
1231                 newsk->sk_priority = 0;
1232                 /*
1233                  * Before updating sk_refcnt, we must commit prior changes to memory
1234                  * (Documentation/RCU/rculist_nulls.txt for details)
1235                  */
1236                 smp_wmb();
1237                 atomic_set(&newsk->sk_refcnt, 2);
1238
1239                 /*
1240                  * Increment the counter in the same struct proto as the master
1241                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1242                  * is the same as sk->sk_prot->socks, as this field was copied
1243                  * with memcpy).
1244                  *
1245                  * This _changes_ the previous behaviour, where
1246                  * tcp_create_openreq_child always was incrementing the
1247                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1248                  * to be taken into account in all callers. -acme
1249                  */
1250                 sk_refcnt_debug_inc(newsk);
1251                 sk_set_socket(newsk, NULL);
1252                 newsk->sk_wq = NULL;
1253
1254                 if (newsk->sk_prot->sockets_allocated)
1255                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1256
1257                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1258                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1259                         net_enable_timestamp();
1260         }
1261 out:
1262         return newsk;
1263 }
1264 EXPORT_SYMBOL_GPL(sk_clone);
1265
1266 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1267 {
1268         __sk_dst_set(sk, dst);
1269         sk->sk_route_caps = dst->dev->features;
1270         if (sk->sk_route_caps & NETIF_F_GSO)
1271                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1272         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1273         if (sk_can_gso(sk)) {
1274                 if (dst->header_len) {
1275                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1276                 } else {
1277                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1278                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1279                 }
1280         }
1281 }
1282 EXPORT_SYMBOL_GPL(sk_setup_caps);
1283
1284 void __init sk_init(void)
1285 {
1286         if (totalram_pages <= 4096) {
1287                 sysctl_wmem_max = 32767;
1288                 sysctl_rmem_max = 32767;
1289                 sysctl_wmem_default = 32767;
1290                 sysctl_rmem_default = 32767;
1291         } else if (totalram_pages >= 131072) {
1292                 sysctl_wmem_max = 131071;
1293                 sysctl_rmem_max = 131071;
1294         }
1295 }
1296
1297 /*
1298  *      Simple resource managers for sockets.
1299  */
1300
1301
1302 /*
1303  * Write buffer destructor automatically called from kfree_skb.
1304  */
1305 void sock_wfree(struct sk_buff *skb)
1306 {
1307         struct sock *sk = skb->sk;
1308         unsigned int len = skb->truesize;
1309
1310         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1311                 /*
1312                  * Keep a reference on sk_wmem_alloc, this will be released
1313                  * after sk_write_space() call
1314                  */
1315                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1316                 sk->sk_write_space(sk);
1317                 len = 1;
1318         }
1319         /*
1320          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1321          * could not do because of in-flight packets
1322          */
1323         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1324                 __sk_free(sk);
1325 }
1326 EXPORT_SYMBOL(sock_wfree);
1327
1328 /*
1329  * Read buffer destructor automatically called from kfree_skb.
1330  */
1331 void sock_rfree(struct sk_buff *skb)
1332 {
1333         struct sock *sk = skb->sk;
1334
1335         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1336         sk_mem_uncharge(skb->sk, skb->truesize);
1337 }
1338 EXPORT_SYMBOL(sock_rfree);
1339
1340
1341 int sock_i_uid(struct sock *sk)
1342 {
1343         int uid;
1344
1345         read_lock(&sk->sk_callback_lock);
1346         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1347         read_unlock(&sk->sk_callback_lock);
1348         return uid;
1349 }
1350 EXPORT_SYMBOL(sock_i_uid);
1351
1352 unsigned long sock_i_ino(struct sock *sk)
1353 {
1354         unsigned long ino;
1355
1356         read_lock(&sk->sk_callback_lock);
1357         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1358         read_unlock(&sk->sk_callback_lock);
1359         return ino;
1360 }
1361 EXPORT_SYMBOL(sock_i_ino);
1362
1363 /*
1364  * Allocate a skb from the socket's send buffer.
1365  */
1366 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1367                              gfp_t priority)
1368 {
1369         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1370                 struct sk_buff *skb = alloc_skb(size, priority);
1371                 if (skb) {
1372                         skb_set_owner_w(skb, sk);
1373                         return skb;
1374                 }
1375         }
1376         return NULL;
1377 }
1378 EXPORT_SYMBOL(sock_wmalloc);
1379
1380 /*
1381  * Allocate a skb from the socket's receive buffer.
1382  */
1383 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1384                              gfp_t priority)
1385 {
1386         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1387                 struct sk_buff *skb = alloc_skb(size, priority);
1388                 if (skb) {
1389                         skb_set_owner_r(skb, sk);
1390                         return skb;
1391                 }
1392         }
1393         return NULL;
1394 }
1395
1396 /*
1397  * Allocate a memory block from the socket's option memory buffer.
1398  */
1399 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1400 {
1401         if ((unsigned)size <= sysctl_optmem_max &&
1402             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1403                 void *mem;
1404                 /* First do the add, to avoid the race if kmalloc
1405                  * might sleep.
1406                  */
1407                 atomic_add(size, &sk->sk_omem_alloc);
1408                 mem = kmalloc(size, priority);
1409                 if (mem)
1410                         return mem;
1411                 atomic_sub(size, &sk->sk_omem_alloc);
1412         }
1413         return NULL;
1414 }
1415 EXPORT_SYMBOL(sock_kmalloc);
1416
1417 /*
1418  * Free an option memory block.
1419  */
1420 void sock_kfree_s(struct sock *sk, void *mem, int size)
1421 {
1422         kfree(mem);
1423         atomic_sub(size, &sk->sk_omem_alloc);
1424 }
1425 EXPORT_SYMBOL(sock_kfree_s);
1426
1427 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1428    I think, these locks should be removed for datagram sockets.
1429  */
1430 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1431 {
1432         DEFINE_WAIT(wait);
1433
1434         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1435         for (;;) {
1436                 if (!timeo)
1437                         break;
1438                 if (signal_pending(current))
1439                         break;
1440                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1441                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1442                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1443                         break;
1444                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1445                         break;
1446                 if (sk->sk_err)
1447                         break;
1448                 timeo = schedule_timeout(timeo);
1449         }
1450         finish_wait(sk_sleep(sk), &wait);
1451         return timeo;
1452 }
1453
1454
1455 /*
1456  *      Generic send/receive buffer handlers
1457  */
1458
1459 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1460                                      unsigned long data_len, int noblock,
1461                                      int *errcode)
1462 {
1463         struct sk_buff *skb;
1464         gfp_t gfp_mask;
1465         long timeo;
1466         int err;
1467
1468         gfp_mask = sk->sk_allocation;
1469         if (gfp_mask & __GFP_WAIT)
1470                 gfp_mask |= __GFP_REPEAT;
1471
1472         timeo = sock_sndtimeo(sk, noblock);
1473         while (1) {
1474                 err = sock_error(sk);
1475                 if (err != 0)
1476                         goto failure;
1477
1478                 err = -EPIPE;
1479                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1480                         goto failure;
1481
1482                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1483                         skb = alloc_skb(header_len, gfp_mask);
1484                         if (skb) {
1485                                 int npages;
1486                                 int i;
1487
1488                                 /* No pages, we're done... */
1489                                 if (!data_len)
1490                                         break;
1491
1492                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1493                                 skb->truesize += data_len;
1494                                 skb_shinfo(skb)->nr_frags = npages;
1495                                 for (i = 0; i < npages; i++) {
1496                                         struct page *page;
1497                                         skb_frag_t *frag;
1498
1499                                         page = alloc_pages(sk->sk_allocation, 0);
1500                                         if (!page) {
1501                                                 err = -ENOBUFS;
1502                                                 skb_shinfo(skb)->nr_frags = i;
1503                                                 kfree_skb(skb);
1504                                                 goto failure;
1505                                         }
1506
1507                                         frag = &skb_shinfo(skb)->frags[i];
1508                                         frag->page = page;
1509                                         frag->page_offset = 0;
1510                                         frag->size = (data_len >= PAGE_SIZE ?
1511                                                       PAGE_SIZE :
1512                                                       data_len);
1513                                         data_len -= PAGE_SIZE;
1514                                 }
1515
1516                                 /* Full success... */
1517                                 break;
1518                         }
1519                         err = -ENOBUFS;
1520                         goto failure;
1521                 }
1522                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1523                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1524                 err = -EAGAIN;
1525                 if (!timeo)
1526                         goto failure;
1527                 if (signal_pending(current))
1528                         goto interrupted;
1529                 timeo = sock_wait_for_wmem(sk, timeo);
1530         }
1531
1532         skb_set_owner_w(skb, sk);
1533         return skb;
1534
1535 interrupted:
1536         err = sock_intr_errno(timeo);
1537 failure:
1538         *errcode = err;
1539         return NULL;
1540 }
1541 EXPORT_SYMBOL(sock_alloc_send_pskb);
1542
1543 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1544                                     int noblock, int *errcode)
1545 {
1546         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1547 }
1548 EXPORT_SYMBOL(sock_alloc_send_skb);
1549
1550 static void __lock_sock(struct sock *sk)
1551 {
1552         DEFINE_WAIT(wait);
1553
1554         for (;;) {
1555                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1556                                         TASK_UNINTERRUPTIBLE);
1557                 spin_unlock_bh(&sk->sk_lock.slock);
1558                 schedule();
1559                 spin_lock_bh(&sk->sk_lock.slock);
1560                 if (!sock_owned_by_user(sk))
1561                         break;
1562         }
1563         finish_wait(&sk->sk_lock.wq, &wait);
1564 }
1565
1566 static void __release_sock(struct sock *sk)
1567 {
1568         struct sk_buff *skb = sk->sk_backlog.head;
1569
1570         do {
1571                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1572                 bh_unlock_sock(sk);
1573
1574                 do {
1575                         struct sk_buff *next = skb->next;
1576
1577                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1578                         skb->next = NULL;
1579                         sk_backlog_rcv(sk, skb);
1580
1581                         /*
1582                          * We are in process context here with softirqs
1583                          * disabled, use cond_resched_softirq() to preempt.
1584                          * This is safe to do because we've taken the backlog
1585                          * queue private:
1586                          */
1587                         cond_resched_softirq();
1588
1589                         skb = next;
1590                 } while (skb != NULL);
1591
1592                 bh_lock_sock(sk);
1593         } while ((skb = sk->sk_backlog.head) != NULL);
1594
1595         /*
1596          * Doing the zeroing here guarantee we can not loop forever
1597          * while a wild producer attempts to flood us.
1598          */
1599         sk->sk_backlog.len = 0;
1600 }
1601
1602 /**
1603  * sk_wait_data - wait for data to arrive at sk_receive_queue
1604  * @sk:    sock to wait on
1605  * @timeo: for how long
1606  *
1607  * Now socket state including sk->sk_err is changed only under lock,
1608  * hence we may omit checks after joining wait queue.
1609  * We check receive queue before schedule() only as optimization;
1610  * it is very likely that release_sock() added new data.
1611  */
1612 int sk_wait_data(struct sock *sk, long *timeo)
1613 {
1614         int rc;
1615         DEFINE_WAIT(wait);
1616
1617         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1618         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1619         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1620         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1621         finish_wait(sk_sleep(sk), &wait);
1622         return rc;
1623 }
1624 EXPORT_SYMBOL(sk_wait_data);
1625
1626 /**
1627  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1628  *      @sk: socket
1629  *      @size: memory size to allocate
1630  *      @kind: allocation type
1631  *
1632  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1633  *      rmem allocation. This function assumes that protocols which have
1634  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1635  */
1636 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1637 {
1638         struct proto *prot = sk->sk_prot;
1639         int amt = sk_mem_pages(size);
1640         int allocated;
1641
1642         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1643         allocated = atomic_add_return(amt, prot->memory_allocated);
1644
1645         /* Under limit. */
1646         if (allocated <= prot->sysctl_mem[0]) {
1647                 if (prot->memory_pressure && *prot->memory_pressure)
1648                         *prot->memory_pressure = 0;
1649                 return 1;
1650         }
1651
1652         /* Under pressure. */
1653         if (allocated > prot->sysctl_mem[1])
1654                 if (prot->enter_memory_pressure)
1655                         prot->enter_memory_pressure(sk);
1656
1657         /* Over hard limit. */
1658         if (allocated > prot->sysctl_mem[2])
1659                 goto suppress_allocation;
1660
1661         /* guarantee minimum buffer size under pressure */
1662         if (kind == SK_MEM_RECV) {
1663                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1664                         return 1;
1665         } else { /* SK_MEM_SEND */
1666                 if (sk->sk_type == SOCK_STREAM) {
1667                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1668                                 return 1;
1669                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1670                            prot->sysctl_wmem[0])
1671                                 return 1;
1672         }
1673
1674         if (prot->memory_pressure) {
1675                 int alloc;
1676
1677                 if (!*prot->memory_pressure)
1678                         return 1;
1679                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1680                 if (prot->sysctl_mem[2] > alloc *
1681                     sk_mem_pages(sk->sk_wmem_queued +
1682                                  atomic_read(&sk->sk_rmem_alloc) +
1683                                  sk->sk_forward_alloc))
1684                         return 1;
1685         }
1686
1687 suppress_allocation:
1688
1689         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1690                 sk_stream_moderate_sndbuf(sk);
1691
1692                 /* Fail only if socket is _under_ its sndbuf.
1693                  * In this case we cannot block, so that we have to fail.
1694                  */
1695                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1696                         return 1;
1697         }
1698
1699         /* Alas. Undo changes. */
1700         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1701         atomic_sub(amt, prot->memory_allocated);
1702         return 0;
1703 }
1704 EXPORT_SYMBOL(__sk_mem_schedule);
1705
1706 /**
1707  *      __sk_reclaim - reclaim memory_allocated
1708  *      @sk: socket
1709  */
1710 void __sk_mem_reclaim(struct sock *sk)
1711 {
1712         struct proto *prot = sk->sk_prot;
1713
1714         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1715                    prot->memory_allocated);
1716         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1717
1718         if (prot->memory_pressure && *prot->memory_pressure &&
1719             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1720                 *prot->memory_pressure = 0;
1721 }
1722 EXPORT_SYMBOL(__sk_mem_reclaim);
1723
1724
1725 /*
1726  * Set of default routines for initialising struct proto_ops when
1727  * the protocol does not support a particular function. In certain
1728  * cases where it makes no sense for a protocol to have a "do nothing"
1729  * function, some default processing is provided.
1730  */
1731
1732 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1733 {
1734         return -EOPNOTSUPP;
1735 }
1736 EXPORT_SYMBOL(sock_no_bind);
1737
1738 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1739                     int len, int flags)
1740 {
1741         return -EOPNOTSUPP;
1742 }
1743 EXPORT_SYMBOL(sock_no_connect);
1744
1745 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1746 {
1747         return -EOPNOTSUPP;
1748 }
1749 EXPORT_SYMBOL(sock_no_socketpair);
1750
1751 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1752 {
1753         return -EOPNOTSUPP;
1754 }
1755 EXPORT_SYMBOL(sock_no_accept);
1756
1757 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1758                     int *len, int peer)
1759 {
1760         return -EOPNOTSUPP;
1761 }
1762 EXPORT_SYMBOL(sock_no_getname);
1763
1764 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1765 {
1766         return 0;
1767 }
1768 EXPORT_SYMBOL(sock_no_poll);
1769
1770 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1771 {
1772         return -EOPNOTSUPP;
1773 }
1774 EXPORT_SYMBOL(sock_no_ioctl);
1775
1776 int sock_no_listen(struct socket *sock, int backlog)
1777 {
1778         return -EOPNOTSUPP;
1779 }
1780 EXPORT_SYMBOL(sock_no_listen);
1781
1782 int sock_no_shutdown(struct socket *sock, int how)
1783 {
1784         return -EOPNOTSUPP;
1785 }
1786 EXPORT_SYMBOL(sock_no_shutdown);
1787
1788 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1789                     char __user *optval, unsigned int optlen)
1790 {
1791         return -EOPNOTSUPP;
1792 }
1793 EXPORT_SYMBOL(sock_no_setsockopt);
1794
1795 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1796                     char __user *optval, int __user *optlen)
1797 {
1798         return -EOPNOTSUPP;
1799 }
1800 EXPORT_SYMBOL(sock_no_getsockopt);
1801
1802 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1803                     size_t len)
1804 {
1805         return -EOPNOTSUPP;
1806 }
1807 EXPORT_SYMBOL(sock_no_sendmsg);
1808
1809 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1810                     size_t len, int flags)
1811 {
1812         return -EOPNOTSUPP;
1813 }
1814 EXPORT_SYMBOL(sock_no_recvmsg);
1815
1816 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1817 {
1818         /* Mirror missing mmap method error code */
1819         return -ENODEV;
1820 }
1821 EXPORT_SYMBOL(sock_no_mmap);
1822
1823 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1824 {
1825         ssize_t res;
1826         struct msghdr msg = {.msg_flags = flags};
1827         struct kvec iov;
1828         char *kaddr = kmap(page);
1829         iov.iov_base = kaddr + offset;
1830         iov.iov_len = size;
1831         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1832         kunmap(page);
1833         return res;
1834 }
1835 EXPORT_SYMBOL(sock_no_sendpage);
1836
1837 /*
1838  *      Default Socket Callbacks
1839  */
1840
1841 static void sock_def_wakeup(struct sock *sk)
1842 {
1843         struct socket_wq *wq;
1844
1845         rcu_read_lock();
1846         wq = rcu_dereference(sk->sk_wq);
1847         if (wq_has_sleeper(wq))
1848                 wake_up_interruptible_all(&wq->wait);
1849         rcu_read_unlock();
1850 }
1851
1852 static void sock_def_error_report(struct sock *sk)
1853 {
1854         struct socket_wq *wq;
1855
1856         rcu_read_lock();
1857         wq = rcu_dereference(sk->sk_wq);
1858         if (wq_has_sleeper(wq))
1859                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1860         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1861         rcu_read_unlock();
1862 }
1863
1864 static void sock_def_readable(struct sock *sk, int len)
1865 {
1866         struct socket_wq *wq;
1867
1868         rcu_read_lock();
1869         wq = rcu_dereference(sk->sk_wq);
1870         if (wq_has_sleeper(wq))
1871                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
1872                                                 POLLRDNORM | POLLRDBAND);
1873         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1874         rcu_read_unlock();
1875 }
1876
1877 static void sock_def_write_space(struct sock *sk)
1878 {
1879         struct socket_wq *wq;
1880
1881         rcu_read_lock();
1882
1883         /* Do not wake up a writer until he can make "significant"
1884          * progress.  --DaveM
1885          */
1886         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1887                 wq = rcu_dereference(sk->sk_wq);
1888                 if (wq_has_sleeper(wq))
1889                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1890                                                 POLLWRNORM | POLLWRBAND);
1891
1892                 /* Should agree with poll, otherwise some programs break */
1893                 if (sock_writeable(sk))
1894                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1895         }
1896
1897         rcu_read_unlock();
1898 }
1899
1900 static void sock_def_destruct(struct sock *sk)
1901 {
1902         kfree(sk->sk_protinfo);
1903 }
1904
1905 void sk_send_sigurg(struct sock *sk)
1906 {
1907         if (sk->sk_socket && sk->sk_socket->file)
1908                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1909                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1910 }
1911 EXPORT_SYMBOL(sk_send_sigurg);
1912
1913 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1914                     unsigned long expires)
1915 {
1916         if (!mod_timer(timer, expires))
1917                 sock_hold(sk);
1918 }
1919 EXPORT_SYMBOL(sk_reset_timer);
1920
1921 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1922 {
1923         if (timer_pending(timer) && del_timer(timer))
1924                 __sock_put(sk);
1925 }
1926 EXPORT_SYMBOL(sk_stop_timer);
1927
1928 void sock_init_data(struct socket *sock, struct sock *sk)
1929 {
1930         skb_queue_head_init(&sk->sk_receive_queue);
1931         skb_queue_head_init(&sk->sk_write_queue);
1932         skb_queue_head_init(&sk->sk_error_queue);
1933 #ifdef CONFIG_NET_DMA
1934         skb_queue_head_init(&sk->sk_async_wait_queue);
1935 #endif
1936
1937         sk->sk_send_head        =       NULL;
1938
1939         init_timer(&sk->sk_timer);
1940
1941         sk->sk_allocation       =       GFP_KERNEL;
1942         sk->sk_rcvbuf           =       sysctl_rmem_default;
1943         sk->sk_sndbuf           =       sysctl_wmem_default;
1944         sk->sk_state            =       TCP_CLOSE;
1945         sk_set_socket(sk, sock);
1946
1947         sock_set_flag(sk, SOCK_ZAPPED);
1948
1949         if (sock) {
1950                 sk->sk_type     =       sock->type;
1951                 sk->sk_wq       =       sock->wq;
1952                 sock->sk        =       sk;
1953         } else
1954                 sk->sk_wq       =       NULL;
1955
1956         spin_lock_init(&sk->sk_dst_lock);
1957         rwlock_init(&sk->sk_callback_lock);
1958         lockdep_set_class_and_name(&sk->sk_callback_lock,
1959                         af_callback_keys + sk->sk_family,
1960                         af_family_clock_key_strings[sk->sk_family]);
1961
1962         sk->sk_state_change     =       sock_def_wakeup;
1963         sk->sk_data_ready       =       sock_def_readable;
1964         sk->sk_write_space      =       sock_def_write_space;
1965         sk->sk_error_report     =       sock_def_error_report;
1966         sk->sk_destruct         =       sock_def_destruct;
1967
1968         sk->sk_sndmsg_page      =       NULL;
1969         sk->sk_sndmsg_off       =       0;
1970
1971         sk->sk_peercred.pid     =       0;
1972         sk->sk_peercred.uid     =       -1;
1973         sk->sk_peercred.gid     =       -1;
1974         sk->sk_write_pending    =       0;
1975         sk->sk_rcvlowat         =       1;
1976         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1977         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1978
1979         sk->sk_stamp = ktime_set(-1L, 0);
1980
1981         /*
1982          * Before updating sk_refcnt, we must commit prior changes to memory
1983          * (Documentation/RCU/rculist_nulls.txt for details)
1984          */
1985         smp_wmb();
1986         atomic_set(&sk->sk_refcnt, 1);
1987         atomic_set(&sk->sk_drops, 0);
1988 }
1989 EXPORT_SYMBOL(sock_init_data);
1990
1991 void lock_sock_nested(struct sock *sk, int subclass)
1992 {
1993         might_sleep();
1994         spin_lock_bh(&sk->sk_lock.slock);
1995         if (sk->sk_lock.owned)
1996                 __lock_sock(sk);
1997         sk->sk_lock.owned = 1;
1998         spin_unlock(&sk->sk_lock.slock);
1999         /*
2000          * The sk_lock has mutex_lock() semantics here:
2001          */
2002         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2003         local_bh_enable();
2004 }
2005 EXPORT_SYMBOL(lock_sock_nested);
2006
2007 void release_sock(struct sock *sk)
2008 {
2009         /*
2010          * The sk_lock has mutex_unlock() semantics:
2011          */
2012         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2013
2014         spin_lock_bh(&sk->sk_lock.slock);
2015         if (sk->sk_backlog.tail)
2016                 __release_sock(sk);
2017         sk->sk_lock.owned = 0;
2018         if (waitqueue_active(&sk->sk_lock.wq))
2019                 wake_up(&sk->sk_lock.wq);
2020         spin_unlock_bh(&sk->sk_lock.slock);
2021 }
2022 EXPORT_SYMBOL(release_sock);
2023
2024 /**
2025  * lock_sock_fast - fast version of lock_sock
2026  * @sk: socket
2027  *
2028  * This version should be used for very small section, where process wont block
2029  * return false if fast path is taken
2030  *   sk_lock.slock locked, owned = 0, BH disabled
2031  * return true if slow path is taken
2032  *   sk_lock.slock unlocked, owned = 1, BH enabled
2033  */
2034 bool lock_sock_fast(struct sock *sk)
2035 {
2036         might_sleep();
2037         spin_lock_bh(&sk->sk_lock.slock);
2038
2039         if (!sk->sk_lock.owned)
2040                 /*
2041                  * Note : We must disable BH
2042                  */
2043                 return false;
2044
2045         __lock_sock(sk);
2046         sk->sk_lock.owned = 1;
2047         spin_unlock(&sk->sk_lock.slock);
2048         /*
2049          * The sk_lock has mutex_lock() semantics here:
2050          */
2051         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2052         local_bh_enable();
2053         return true;
2054 }
2055 EXPORT_SYMBOL(lock_sock_fast);
2056
2057 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2058 {
2059         struct timeval tv;
2060         if (!sock_flag(sk, SOCK_TIMESTAMP))
2061                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2062         tv = ktime_to_timeval(sk->sk_stamp);
2063         if (tv.tv_sec == -1)
2064                 return -ENOENT;
2065         if (tv.tv_sec == 0) {
2066                 sk->sk_stamp = ktime_get_real();
2067                 tv = ktime_to_timeval(sk->sk_stamp);
2068         }
2069         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2070 }
2071 EXPORT_SYMBOL(sock_get_timestamp);
2072
2073 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2074 {
2075         struct timespec ts;
2076         if (!sock_flag(sk, SOCK_TIMESTAMP))
2077                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2078         ts = ktime_to_timespec(sk->sk_stamp);
2079         if (ts.tv_sec == -1)
2080                 return -ENOENT;
2081         if (ts.tv_sec == 0) {
2082                 sk->sk_stamp = ktime_get_real();
2083                 ts = ktime_to_timespec(sk->sk_stamp);
2084         }
2085         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2086 }
2087 EXPORT_SYMBOL(sock_get_timestampns);
2088
2089 void sock_enable_timestamp(struct sock *sk, int flag)
2090 {
2091         if (!sock_flag(sk, flag)) {
2092                 sock_set_flag(sk, flag);
2093                 /*
2094                  * we just set one of the two flags which require net
2095                  * time stamping, but time stamping might have been on
2096                  * already because of the other one
2097                  */
2098                 if (!sock_flag(sk,
2099                                 flag == SOCK_TIMESTAMP ?
2100                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2101                                 SOCK_TIMESTAMP))
2102                         net_enable_timestamp();
2103         }
2104 }
2105
2106 /*
2107  *      Get a socket option on an socket.
2108  *
2109  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2110  *      asynchronous errors should be reported by getsockopt. We assume
2111  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2112  */
2113 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2114                            char __user *optval, int __user *optlen)
2115 {
2116         struct sock *sk = sock->sk;
2117
2118         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2119 }
2120 EXPORT_SYMBOL(sock_common_getsockopt);
2121
2122 #ifdef CONFIG_COMPAT
2123 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2124                                   char __user *optval, int __user *optlen)
2125 {
2126         struct sock *sk = sock->sk;
2127
2128         if (sk->sk_prot->compat_getsockopt != NULL)
2129                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2130                                                       optval, optlen);
2131         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2132 }
2133 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2134 #endif
2135
2136 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2137                         struct msghdr *msg, size_t size, int flags)
2138 {
2139         struct sock *sk = sock->sk;
2140         int addr_len = 0;
2141         int err;
2142
2143         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2144                                    flags & ~MSG_DONTWAIT, &addr_len);
2145         if (err >= 0)
2146                 msg->msg_namelen = addr_len;
2147         return err;
2148 }
2149 EXPORT_SYMBOL(sock_common_recvmsg);
2150
2151 /*
2152  *      Set socket options on an inet socket.
2153  */
2154 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2155                            char __user *optval, unsigned int optlen)
2156 {
2157         struct sock *sk = sock->sk;
2158
2159         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2160 }
2161 EXPORT_SYMBOL(sock_common_setsockopt);
2162
2163 #ifdef CONFIG_COMPAT
2164 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2165                                   char __user *optval, unsigned int optlen)
2166 {
2167         struct sock *sk = sock->sk;
2168
2169         if (sk->sk_prot->compat_setsockopt != NULL)
2170                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2171                                                       optval, optlen);
2172         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2173 }
2174 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2175 #endif
2176
2177 void sk_common_release(struct sock *sk)
2178 {
2179         if (sk->sk_prot->destroy)
2180                 sk->sk_prot->destroy(sk);
2181
2182         /*
2183          * Observation: when sock_common_release is called, processes have
2184          * no access to socket. But net still has.
2185          * Step one, detach it from networking:
2186          *
2187          * A. Remove from hash tables.
2188          */
2189
2190         sk->sk_prot->unhash(sk);
2191
2192         /*
2193          * In this point socket cannot receive new packets, but it is possible
2194          * that some packets are in flight because some CPU runs receiver and
2195          * did hash table lookup before we unhashed socket. They will achieve
2196          * receive queue and will be purged by socket destructor.
2197          *
2198          * Also we still have packets pending on receive queue and probably,
2199          * our own packets waiting in device queues. sock_destroy will drain
2200          * receive queue, but transmitted packets will delay socket destruction
2201          * until the last reference will be released.
2202          */
2203
2204         sock_orphan(sk);
2205
2206         xfrm_sk_free_policy(sk);
2207
2208         sk_refcnt_debug_release(sk);
2209         sock_put(sk);
2210 }
2211 EXPORT_SYMBOL(sk_common_release);
2212
2213 static DEFINE_RWLOCK(proto_list_lock);
2214 static LIST_HEAD(proto_list);
2215
2216 #ifdef CONFIG_PROC_FS
2217 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2218 struct prot_inuse {
2219         int val[PROTO_INUSE_NR];
2220 };
2221
2222 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2223
2224 #ifdef CONFIG_NET_NS
2225 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2226 {
2227         int cpu = smp_processor_id();
2228         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2229 }
2230 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2231
2232 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2233 {
2234         int cpu, idx = prot->inuse_idx;
2235         int res = 0;
2236
2237         for_each_possible_cpu(cpu)
2238                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2239
2240         return res >= 0 ? res : 0;
2241 }
2242 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2243
2244 static int __net_init sock_inuse_init_net(struct net *net)
2245 {
2246         net->core.inuse = alloc_percpu(struct prot_inuse);
2247         return net->core.inuse ? 0 : -ENOMEM;
2248 }
2249
2250 static void __net_exit sock_inuse_exit_net(struct net *net)
2251 {
2252         free_percpu(net->core.inuse);
2253 }
2254
2255 static struct pernet_operations net_inuse_ops = {
2256         .init = sock_inuse_init_net,
2257         .exit = sock_inuse_exit_net,
2258 };
2259
2260 static __init int net_inuse_init(void)
2261 {
2262         if (register_pernet_subsys(&net_inuse_ops))
2263                 panic("Cannot initialize net inuse counters");
2264
2265         return 0;
2266 }
2267
2268 core_initcall(net_inuse_init);
2269 #else
2270 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2271
2272 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2273 {
2274         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2275 }
2276 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2277
2278 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2279 {
2280         int cpu, idx = prot->inuse_idx;
2281         int res = 0;
2282
2283         for_each_possible_cpu(cpu)
2284                 res += per_cpu(prot_inuse, cpu).val[idx];
2285
2286         return res >= 0 ? res : 0;
2287 }
2288 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2289 #endif
2290
2291 static void assign_proto_idx(struct proto *prot)
2292 {
2293         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2294
2295         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2296                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2297                 return;
2298         }
2299
2300         set_bit(prot->inuse_idx, proto_inuse_idx);
2301 }
2302
2303 static void release_proto_idx(struct proto *prot)
2304 {
2305         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2306                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2307 }
2308 #else
2309 static inline void assign_proto_idx(struct proto *prot)
2310 {
2311 }
2312
2313 static inline void release_proto_idx(struct proto *prot)
2314 {
2315 }
2316 #endif
2317
2318 int proto_register(struct proto *prot, int alloc_slab)
2319 {
2320         if (alloc_slab) {
2321                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2322                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2323                                         NULL);
2324
2325                 if (prot->slab == NULL) {
2326                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2327                                prot->name);
2328                         goto out;
2329                 }
2330
2331                 if (prot->rsk_prot != NULL) {
2332                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2333                         if (prot->rsk_prot->slab_name == NULL)
2334                                 goto out_free_sock_slab;
2335
2336                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2337                                                                  prot->rsk_prot->obj_size, 0,
2338                                                                  SLAB_HWCACHE_ALIGN, NULL);
2339
2340                         if (prot->rsk_prot->slab == NULL) {
2341                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2342                                        prot->name);
2343                                 goto out_free_request_sock_slab_name;
2344                         }
2345                 }
2346
2347                 if (prot->twsk_prot != NULL) {
2348                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2349
2350                         if (prot->twsk_prot->twsk_slab_name == NULL)
2351                                 goto out_free_request_sock_slab;
2352
2353                         prot->twsk_prot->twsk_slab =
2354                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2355                                                   prot->twsk_prot->twsk_obj_size,
2356                                                   0,
2357                                                   SLAB_HWCACHE_ALIGN |
2358                                                         prot->slab_flags,
2359                                                   NULL);
2360                         if (prot->twsk_prot->twsk_slab == NULL)
2361                                 goto out_free_timewait_sock_slab_name;
2362                 }
2363         }
2364
2365         write_lock(&proto_list_lock);
2366         list_add(&prot->node, &proto_list);
2367         assign_proto_idx(prot);
2368         write_unlock(&proto_list_lock);
2369         return 0;
2370
2371 out_free_timewait_sock_slab_name:
2372         kfree(prot->twsk_prot->twsk_slab_name);
2373 out_free_request_sock_slab:
2374         if (prot->rsk_prot && prot->rsk_prot->slab) {
2375                 kmem_cache_destroy(prot->rsk_prot->slab);
2376                 prot->rsk_prot->slab = NULL;
2377         }
2378 out_free_request_sock_slab_name:
2379         if (prot->rsk_prot)
2380                 kfree(prot->rsk_prot->slab_name);
2381 out_free_sock_slab:
2382         kmem_cache_destroy(prot->slab);
2383         prot->slab = NULL;
2384 out:
2385         return -ENOBUFS;
2386 }
2387 EXPORT_SYMBOL(proto_register);
2388
2389 void proto_unregister(struct proto *prot)
2390 {
2391         write_lock(&proto_list_lock);
2392         release_proto_idx(prot);
2393         list_del(&prot->node);
2394         write_unlock(&proto_list_lock);
2395
2396         if (prot->slab != NULL) {
2397                 kmem_cache_destroy(prot->slab);
2398                 prot->slab = NULL;
2399         }
2400
2401         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2402                 kmem_cache_destroy(prot->rsk_prot->slab);
2403                 kfree(prot->rsk_prot->slab_name);
2404                 prot->rsk_prot->slab = NULL;
2405         }
2406
2407         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2408                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2409                 kfree(prot->twsk_prot->twsk_slab_name);
2410                 prot->twsk_prot->twsk_slab = NULL;
2411         }
2412 }
2413 EXPORT_SYMBOL(proto_unregister);
2414
2415 #ifdef CONFIG_PROC_FS
2416 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2417         __acquires(proto_list_lock)
2418 {
2419         read_lock(&proto_list_lock);
2420         return seq_list_start_head(&proto_list, *pos);
2421 }
2422
2423 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2424 {
2425         return seq_list_next(v, &proto_list, pos);
2426 }
2427
2428 static void proto_seq_stop(struct seq_file *seq, void *v)
2429         __releases(proto_list_lock)
2430 {
2431         read_unlock(&proto_list_lock);
2432 }
2433
2434 static char proto_method_implemented(const void *method)
2435 {
2436         return method == NULL ? 'n' : 'y';
2437 }
2438
2439 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2440 {
2441         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2442                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2443                    proto->name,
2444                    proto->obj_size,
2445                    sock_prot_inuse_get(seq_file_net(seq), proto),
2446                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2447                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2448                    proto->max_header,
2449                    proto->slab == NULL ? "no" : "yes",
2450                    module_name(proto->owner),
2451                    proto_method_implemented(proto->close),
2452                    proto_method_implemented(proto->connect),
2453                    proto_method_implemented(proto->disconnect),
2454                    proto_method_implemented(proto->accept),
2455                    proto_method_implemented(proto->ioctl),
2456                    proto_method_implemented(proto->init),
2457                    proto_method_implemented(proto->destroy),
2458                    proto_method_implemented(proto->shutdown),
2459                    proto_method_implemented(proto->setsockopt),
2460                    proto_method_implemented(proto->getsockopt),
2461                    proto_method_implemented(proto->sendmsg),
2462                    proto_method_implemented(proto->recvmsg),
2463                    proto_method_implemented(proto->sendpage),
2464                    proto_method_implemented(proto->bind),
2465                    proto_method_implemented(proto->backlog_rcv),
2466                    proto_method_implemented(proto->hash),
2467                    proto_method_implemented(proto->unhash),
2468                    proto_method_implemented(proto->get_port),
2469                    proto_method_implemented(proto->enter_memory_pressure));
2470 }
2471
2472 static int proto_seq_show(struct seq_file *seq, void *v)
2473 {
2474         if (v == &proto_list)
2475                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2476                            "protocol",
2477                            "size",
2478                            "sockets",
2479                            "memory",
2480                            "press",
2481                            "maxhdr",
2482                            "slab",
2483                            "module",
2484                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2485         else
2486                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2487         return 0;
2488 }
2489
2490 static const struct seq_operations proto_seq_ops = {
2491         .start  = proto_seq_start,
2492         .next   = proto_seq_next,
2493         .stop   = proto_seq_stop,
2494         .show   = proto_seq_show,
2495 };
2496
2497 static int proto_seq_open(struct inode *inode, struct file *file)
2498 {
2499         return seq_open_net(inode, file, &proto_seq_ops,
2500                             sizeof(struct seq_net_private));
2501 }
2502
2503 static const struct file_operations proto_seq_fops = {
2504         .owner          = THIS_MODULE,
2505         .open           = proto_seq_open,
2506         .read           = seq_read,
2507         .llseek         = seq_lseek,
2508         .release        = seq_release_net,
2509 };
2510
2511 static __net_init int proto_init_net(struct net *net)
2512 {
2513         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2514                 return -ENOMEM;
2515
2516         return 0;
2517 }
2518
2519 static __net_exit void proto_exit_net(struct net *net)
2520 {
2521         proc_net_remove(net, "protocols");
2522 }
2523
2524
2525 static __net_initdata struct pernet_operations proto_net_ops = {
2526         .init = proto_init_net,
2527         .exit = proto_exit_net,
2528 };
2529
2530 static int __init proto_init(void)
2531 {
2532         return register_pernet_subsys(&proto_net_ops);
2533 }
2534
2535 subsys_initcall(proto_init);
2536
2537 #endif /* PROC_FS */