net: fix a lockdep splat
[linux-2.6.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 EXPORT_SYMBOL_GPL(cred_to_ucred);
766
767 int sock_getsockopt(struct socket *sock, int level, int optname,
768                     char __user *optval, int __user *optlen)
769 {
770         struct sock *sk = sock->sk;
771
772         union {
773                 int val;
774                 struct linger ling;
775                 struct timeval tm;
776         } v;
777
778         int lv = sizeof(int);
779         int len;
780
781         if (get_user(len, optlen))
782                 return -EFAULT;
783         if (len < 0)
784                 return -EINVAL;
785
786         memset(&v, 0, sizeof(v));
787
788         switch (optname) {
789         case SO_DEBUG:
790                 v.val = sock_flag(sk, SOCK_DBG);
791                 break;
792
793         case SO_DONTROUTE:
794                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
795                 break;
796
797         case SO_BROADCAST:
798                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
799                 break;
800
801         case SO_SNDBUF:
802                 v.val = sk->sk_sndbuf;
803                 break;
804
805         case SO_RCVBUF:
806                 v.val = sk->sk_rcvbuf;
807                 break;
808
809         case SO_REUSEADDR:
810                 v.val = sk->sk_reuse;
811                 break;
812
813         case SO_KEEPALIVE:
814                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
815                 break;
816
817         case SO_TYPE:
818                 v.val = sk->sk_type;
819                 break;
820
821         case SO_PROTOCOL:
822                 v.val = sk->sk_protocol;
823                 break;
824
825         case SO_DOMAIN:
826                 v.val = sk->sk_family;
827                 break;
828
829         case SO_ERROR:
830                 v.val = -sock_error(sk);
831                 if (v.val == 0)
832                         v.val = xchg(&sk->sk_err_soft, 0);
833                 break;
834
835         case SO_OOBINLINE:
836                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
837                 break;
838
839         case SO_NO_CHECK:
840                 v.val = sk->sk_no_check;
841                 break;
842
843         case SO_PRIORITY:
844                 v.val = sk->sk_priority;
845                 break;
846
847         case SO_LINGER:
848                 lv              = sizeof(v.ling);
849                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
850                 v.ling.l_linger = sk->sk_lingertime / HZ;
851                 break;
852
853         case SO_BSDCOMPAT:
854                 sock_warn_obsolete_bsdism("getsockopt");
855                 break;
856
857         case SO_TIMESTAMP:
858                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
859                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
860                 break;
861
862         case SO_TIMESTAMPNS:
863                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
864                 break;
865
866         case SO_TIMESTAMPING:
867                 v.val = 0;
868                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
869                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
870                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
871                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
872                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
873                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
874                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
875                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
876                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
877                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
878                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
879                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
880                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
881                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
882                 break;
883
884         case SO_RCVTIMEO:
885                 lv = sizeof(struct timeval);
886                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
887                         v.tm.tv_sec = 0;
888                         v.tm.tv_usec = 0;
889                 } else {
890                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
891                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
892                 }
893                 break;
894
895         case SO_SNDTIMEO:
896                 lv = sizeof(struct timeval);
897                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
898                         v.tm.tv_sec = 0;
899                         v.tm.tv_usec = 0;
900                 } else {
901                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
902                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
903                 }
904                 break;
905
906         case SO_RCVLOWAT:
907                 v.val = sk->sk_rcvlowat;
908                 break;
909
910         case SO_SNDLOWAT:
911                 v.val = 1;
912                 break;
913
914         case SO_PASSCRED:
915                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
916                 break;
917
918         case SO_PEERCRED:
919         {
920                 struct ucred peercred;
921                 if (len > sizeof(peercred))
922                         len = sizeof(peercred);
923                 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
924                 if (copy_to_user(optval, &peercred, len))
925                         return -EFAULT;
926                 goto lenout;
927         }
928
929         case SO_PEERNAME:
930         {
931                 char address[128];
932
933                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
934                         return -ENOTCONN;
935                 if (lv < len)
936                         return -EINVAL;
937                 if (copy_to_user(optval, address, len))
938                         return -EFAULT;
939                 goto lenout;
940         }
941
942         /* Dubious BSD thing... Probably nobody even uses it, but
943          * the UNIX standard wants it for whatever reason... -DaveM
944          */
945         case SO_ACCEPTCONN:
946                 v.val = sk->sk_state == TCP_LISTEN;
947                 break;
948
949         case SO_PASSSEC:
950                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
951                 break;
952
953         case SO_PEERSEC:
954                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
955
956         case SO_MARK:
957                 v.val = sk->sk_mark;
958                 break;
959
960         case SO_RXQ_OVFL:
961                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
962                 break;
963
964         default:
965                 return -ENOPROTOOPT;
966         }
967
968         if (len > lv)
969                 len = lv;
970         if (copy_to_user(optval, &v, len))
971                 return -EFAULT;
972 lenout:
973         if (put_user(len, optlen))
974                 return -EFAULT;
975         return 0;
976 }
977
978 /*
979  * Initialize an sk_lock.
980  *
981  * (We also register the sk_lock with the lock validator.)
982  */
983 static inline void sock_lock_init(struct sock *sk)
984 {
985         sock_lock_init_class_and_name(sk,
986                         af_family_slock_key_strings[sk->sk_family],
987                         af_family_slock_keys + sk->sk_family,
988                         af_family_key_strings[sk->sk_family],
989                         af_family_keys + sk->sk_family);
990 }
991
992 /*
993  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
994  * even temporarly, because of RCU lookups. sk_node should also be left as is.
995  */
996 static void sock_copy(struct sock *nsk, const struct sock *osk)
997 {
998 #ifdef CONFIG_SECURITY_NETWORK
999         void *sptr = nsk->sk_security;
1000 #endif
1001         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
1002                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
1003                      sizeof(osk->sk_tx_queue_mapping));
1004         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
1005                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
1006 #ifdef CONFIG_SECURITY_NETWORK
1007         nsk->sk_security = sptr;
1008         security_sk_clone(osk, nsk);
1009 #endif
1010 }
1011
1012 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1013                 int family)
1014 {
1015         struct sock *sk;
1016         struct kmem_cache *slab;
1017
1018         slab = prot->slab;
1019         if (slab != NULL) {
1020                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1021                 if (!sk)
1022                         return sk;
1023                 if (priority & __GFP_ZERO) {
1024                         /*
1025                          * caches using SLAB_DESTROY_BY_RCU should let
1026                          * sk_node.next un-modified. Special care is taken
1027                          * when initializing object to zero.
1028                          */
1029                         if (offsetof(struct sock, sk_node.next) != 0)
1030                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
1031                         memset(&sk->sk_node.pprev, 0,
1032                                prot->obj_size - offsetof(struct sock,
1033                                                          sk_node.pprev));
1034                 }
1035         }
1036         else
1037                 sk = kmalloc(prot->obj_size, priority);
1038
1039         if (sk != NULL) {
1040                 kmemcheck_annotate_bitfield(sk, flags);
1041
1042                 if (security_sk_alloc(sk, family, priority))
1043                         goto out_free;
1044
1045                 if (!try_module_get(prot->owner))
1046                         goto out_free_sec;
1047                 sk_tx_queue_clear(sk);
1048         }
1049
1050         return sk;
1051
1052 out_free_sec:
1053         security_sk_free(sk);
1054 out_free:
1055         if (slab != NULL)
1056                 kmem_cache_free(slab, sk);
1057         else
1058                 kfree(sk);
1059         return NULL;
1060 }
1061
1062 static void sk_prot_free(struct proto *prot, struct sock *sk)
1063 {
1064         struct kmem_cache *slab;
1065         struct module *owner;
1066
1067         owner = prot->owner;
1068         slab = prot->slab;
1069
1070         security_sk_free(sk);
1071         if (slab != NULL)
1072                 kmem_cache_free(slab, sk);
1073         else
1074                 kfree(sk);
1075         module_put(owner);
1076 }
1077
1078 #ifdef CONFIG_CGROUPS
1079 void sock_update_classid(struct sock *sk)
1080 {
1081         u32 classid = task_cls_classid(current);
1082
1083         if (classid && classid != sk->sk_classid)
1084                 sk->sk_classid = classid;
1085 }
1086 EXPORT_SYMBOL(sock_update_classid);
1087 #endif
1088
1089 /**
1090  *      sk_alloc - All socket objects are allocated here
1091  *      @net: the applicable net namespace
1092  *      @family: protocol family
1093  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1094  *      @prot: struct proto associated with this new sock instance
1095  */
1096 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1097                       struct proto *prot)
1098 {
1099         struct sock *sk;
1100
1101         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1102         if (sk) {
1103                 sk->sk_family = family;
1104                 /*
1105                  * See comment in struct sock definition to understand
1106                  * why we need sk_prot_creator -acme
1107                  */
1108                 sk->sk_prot = sk->sk_prot_creator = prot;
1109                 sock_lock_init(sk);
1110                 sock_net_set(sk, get_net(net));
1111                 atomic_set(&sk->sk_wmem_alloc, 1);
1112
1113                 sock_update_classid(sk);
1114         }
1115
1116         return sk;
1117 }
1118 EXPORT_SYMBOL(sk_alloc);
1119
1120 static void __sk_free(struct sock *sk)
1121 {
1122         struct sk_filter *filter;
1123
1124         if (sk->sk_destruct)
1125                 sk->sk_destruct(sk);
1126
1127         filter = rcu_dereference_check(sk->sk_filter,
1128                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1129         if (filter) {
1130                 sk_filter_uncharge(sk, filter);
1131                 rcu_assign_pointer(sk->sk_filter, NULL);
1132         }
1133
1134         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1135         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1136
1137         if (atomic_read(&sk->sk_omem_alloc))
1138                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1139                        __func__, atomic_read(&sk->sk_omem_alloc));
1140
1141         if (sk->sk_peer_cred)
1142                 put_cred(sk->sk_peer_cred);
1143         put_pid(sk->sk_peer_pid);
1144         put_net(sock_net(sk));
1145         sk_prot_free(sk->sk_prot_creator, sk);
1146 }
1147
1148 void sk_free(struct sock *sk)
1149 {
1150         /*
1151          * We substract one from sk_wmem_alloc and can know if
1152          * some packets are still in some tx queue.
1153          * If not null, sock_wfree() will call __sk_free(sk) later
1154          */
1155         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1156                 __sk_free(sk);
1157 }
1158 EXPORT_SYMBOL(sk_free);
1159
1160 /*
1161  * Last sock_put should drop referrence to sk->sk_net. It has already
1162  * been dropped in sk_change_net. Taking referrence to stopping namespace
1163  * is not an option.
1164  * Take referrence to a socket to remove it from hash _alive_ and after that
1165  * destroy it in the context of init_net.
1166  */
1167 void sk_release_kernel(struct sock *sk)
1168 {
1169         if (sk == NULL || sk->sk_socket == NULL)
1170                 return;
1171
1172         sock_hold(sk);
1173         sock_release(sk->sk_socket);
1174         release_net(sock_net(sk));
1175         sock_net_set(sk, get_net(&init_net));
1176         sock_put(sk);
1177 }
1178 EXPORT_SYMBOL(sk_release_kernel);
1179
1180 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1181 {
1182         struct sock *newsk;
1183
1184         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1185         if (newsk != NULL) {
1186                 struct sk_filter *filter;
1187
1188                 sock_copy(newsk, sk);
1189
1190                 /* SANITY */
1191                 get_net(sock_net(newsk));
1192                 sk_node_init(&newsk->sk_node);
1193                 sock_lock_init(newsk);
1194                 bh_lock_sock(newsk);
1195                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1196                 newsk->sk_backlog.len = 0;
1197
1198                 atomic_set(&newsk->sk_rmem_alloc, 0);
1199                 /*
1200                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1201                  */
1202                 atomic_set(&newsk->sk_wmem_alloc, 1);
1203                 atomic_set(&newsk->sk_omem_alloc, 0);
1204                 skb_queue_head_init(&newsk->sk_receive_queue);
1205                 skb_queue_head_init(&newsk->sk_write_queue);
1206 #ifdef CONFIG_NET_DMA
1207                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1208 #endif
1209
1210                 spin_lock_init(&newsk->sk_dst_lock);
1211                 rwlock_init(&newsk->sk_callback_lock);
1212                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1213                                 af_callback_keys + newsk->sk_family,
1214                                 af_family_clock_key_strings[newsk->sk_family]);
1215
1216                 newsk->sk_dst_cache     = NULL;
1217                 newsk->sk_wmem_queued   = 0;
1218                 newsk->sk_forward_alloc = 0;
1219                 newsk->sk_send_head     = NULL;
1220                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1221
1222                 sock_reset_flag(newsk, SOCK_DONE);
1223                 skb_queue_head_init(&newsk->sk_error_queue);
1224
1225                 filter = newsk->sk_filter;
1226                 if (filter != NULL)
1227                         sk_filter_charge(newsk, filter);
1228
1229                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1230                         /* It is still raw copy of parent, so invalidate
1231                          * destructor and make plain sk_free() */
1232                         newsk->sk_destruct = NULL;
1233                         sk_free(newsk);
1234                         newsk = NULL;
1235                         goto out;
1236                 }
1237
1238                 newsk->sk_err      = 0;
1239                 newsk->sk_priority = 0;
1240                 /*
1241                  * Before updating sk_refcnt, we must commit prior changes to memory
1242                  * (Documentation/RCU/rculist_nulls.txt for details)
1243                  */
1244                 smp_wmb();
1245                 atomic_set(&newsk->sk_refcnt, 2);
1246
1247                 /*
1248                  * Increment the counter in the same struct proto as the master
1249                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1250                  * is the same as sk->sk_prot->socks, as this field was copied
1251                  * with memcpy).
1252                  *
1253                  * This _changes_ the previous behaviour, where
1254                  * tcp_create_openreq_child always was incrementing the
1255                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1256                  * to be taken into account in all callers. -acme
1257                  */
1258                 sk_refcnt_debug_inc(newsk);
1259                 sk_set_socket(newsk, NULL);
1260                 newsk->sk_wq = NULL;
1261
1262                 if (newsk->sk_prot->sockets_allocated)
1263                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1264
1265                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1266                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1267                         net_enable_timestamp();
1268         }
1269 out:
1270         return newsk;
1271 }
1272 EXPORT_SYMBOL_GPL(sk_clone);
1273
1274 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1275 {
1276         __sk_dst_set(sk, dst);
1277         sk->sk_route_caps = dst->dev->features;
1278         if (sk->sk_route_caps & NETIF_F_GSO)
1279                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1280         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1281         if (sk_can_gso(sk)) {
1282                 if (dst->header_len) {
1283                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1284                 } else {
1285                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1286                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1287                 }
1288         }
1289 }
1290 EXPORT_SYMBOL_GPL(sk_setup_caps);
1291
1292 void __init sk_init(void)
1293 {
1294         if (totalram_pages <= 4096) {
1295                 sysctl_wmem_max = 32767;
1296                 sysctl_rmem_max = 32767;
1297                 sysctl_wmem_default = 32767;
1298                 sysctl_rmem_default = 32767;
1299         } else if (totalram_pages >= 131072) {
1300                 sysctl_wmem_max = 131071;
1301                 sysctl_rmem_max = 131071;
1302         }
1303 }
1304
1305 /*
1306  *      Simple resource managers for sockets.
1307  */
1308
1309
1310 /*
1311  * Write buffer destructor automatically called from kfree_skb.
1312  */
1313 void sock_wfree(struct sk_buff *skb)
1314 {
1315         struct sock *sk = skb->sk;
1316         unsigned int len = skb->truesize;
1317
1318         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1319                 /*
1320                  * Keep a reference on sk_wmem_alloc, this will be released
1321                  * after sk_write_space() call
1322                  */
1323                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1324                 sk->sk_write_space(sk);
1325                 len = 1;
1326         }
1327         /*
1328          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1329          * could not do because of in-flight packets
1330          */
1331         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1332                 __sk_free(sk);
1333 }
1334 EXPORT_SYMBOL(sock_wfree);
1335
1336 /*
1337  * Read buffer destructor automatically called from kfree_skb.
1338  */
1339 void sock_rfree(struct sk_buff *skb)
1340 {
1341         struct sock *sk = skb->sk;
1342         unsigned int len = skb->truesize;
1343
1344         atomic_sub(len, &sk->sk_rmem_alloc);
1345         sk_mem_uncharge(sk, len);
1346 }
1347 EXPORT_SYMBOL(sock_rfree);
1348
1349
1350 int sock_i_uid(struct sock *sk)
1351 {
1352         int uid;
1353
1354         read_lock_bh(&sk->sk_callback_lock);
1355         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1356         read_unlock_bh(&sk->sk_callback_lock);
1357         return uid;
1358 }
1359 EXPORT_SYMBOL(sock_i_uid);
1360
1361 unsigned long sock_i_ino(struct sock *sk)
1362 {
1363         unsigned long ino;
1364
1365         read_lock_bh(&sk->sk_callback_lock);
1366         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1367         read_unlock_bh(&sk->sk_callback_lock);
1368         return ino;
1369 }
1370 EXPORT_SYMBOL(sock_i_ino);
1371
1372 /*
1373  * Allocate a skb from the socket's send buffer.
1374  */
1375 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1376                              gfp_t priority)
1377 {
1378         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1379                 struct sk_buff *skb = alloc_skb(size, priority);
1380                 if (skb) {
1381                         skb_set_owner_w(skb, sk);
1382                         return skb;
1383                 }
1384         }
1385         return NULL;
1386 }
1387 EXPORT_SYMBOL(sock_wmalloc);
1388
1389 /*
1390  * Allocate a skb from the socket's receive buffer.
1391  */
1392 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1393                              gfp_t priority)
1394 {
1395         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1396                 struct sk_buff *skb = alloc_skb(size, priority);
1397                 if (skb) {
1398                         skb_set_owner_r(skb, sk);
1399                         return skb;
1400                 }
1401         }
1402         return NULL;
1403 }
1404
1405 /*
1406  * Allocate a memory block from the socket's option memory buffer.
1407  */
1408 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1409 {
1410         if ((unsigned)size <= sysctl_optmem_max &&
1411             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1412                 void *mem;
1413                 /* First do the add, to avoid the race if kmalloc
1414                  * might sleep.
1415                  */
1416                 atomic_add(size, &sk->sk_omem_alloc);
1417                 mem = kmalloc(size, priority);
1418                 if (mem)
1419                         return mem;
1420                 atomic_sub(size, &sk->sk_omem_alloc);
1421         }
1422         return NULL;
1423 }
1424 EXPORT_SYMBOL(sock_kmalloc);
1425
1426 /*
1427  * Free an option memory block.
1428  */
1429 void sock_kfree_s(struct sock *sk, void *mem, int size)
1430 {
1431         kfree(mem);
1432         atomic_sub(size, &sk->sk_omem_alloc);
1433 }
1434 EXPORT_SYMBOL(sock_kfree_s);
1435
1436 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1437    I think, these locks should be removed for datagram sockets.
1438  */
1439 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1440 {
1441         DEFINE_WAIT(wait);
1442
1443         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1444         for (;;) {
1445                 if (!timeo)
1446                         break;
1447                 if (signal_pending(current))
1448                         break;
1449                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1450                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1451                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1452                         break;
1453                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1454                         break;
1455                 if (sk->sk_err)
1456                         break;
1457                 timeo = schedule_timeout(timeo);
1458         }
1459         finish_wait(sk_sleep(sk), &wait);
1460         return timeo;
1461 }
1462
1463
1464 /*
1465  *      Generic send/receive buffer handlers
1466  */
1467
1468 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1469                                      unsigned long data_len, int noblock,
1470                                      int *errcode)
1471 {
1472         struct sk_buff *skb;
1473         gfp_t gfp_mask;
1474         long timeo;
1475         int err;
1476
1477         gfp_mask = sk->sk_allocation;
1478         if (gfp_mask & __GFP_WAIT)
1479                 gfp_mask |= __GFP_REPEAT;
1480
1481         timeo = sock_sndtimeo(sk, noblock);
1482         while (1) {
1483                 err = sock_error(sk);
1484                 if (err != 0)
1485                         goto failure;
1486
1487                 err = -EPIPE;
1488                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1489                         goto failure;
1490
1491                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1492                         skb = alloc_skb(header_len, gfp_mask);
1493                         if (skb) {
1494                                 int npages;
1495                                 int i;
1496
1497                                 /* No pages, we're done... */
1498                                 if (!data_len)
1499                                         break;
1500
1501                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1502                                 skb->truesize += data_len;
1503                                 skb_shinfo(skb)->nr_frags = npages;
1504                                 for (i = 0; i < npages; i++) {
1505                                         struct page *page;
1506                                         skb_frag_t *frag;
1507
1508                                         page = alloc_pages(sk->sk_allocation, 0);
1509                                         if (!page) {
1510                                                 err = -ENOBUFS;
1511                                                 skb_shinfo(skb)->nr_frags = i;
1512                                                 kfree_skb(skb);
1513                                                 goto failure;
1514                                         }
1515
1516                                         frag = &skb_shinfo(skb)->frags[i];
1517                                         frag->page = page;
1518                                         frag->page_offset = 0;
1519                                         frag->size = (data_len >= PAGE_SIZE ?
1520                                                       PAGE_SIZE :
1521                                                       data_len);
1522                                         data_len -= PAGE_SIZE;
1523                                 }
1524
1525                                 /* Full success... */
1526                                 break;
1527                         }
1528                         err = -ENOBUFS;
1529                         goto failure;
1530                 }
1531                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1532                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1533                 err = -EAGAIN;
1534                 if (!timeo)
1535                         goto failure;
1536                 if (signal_pending(current))
1537                         goto interrupted;
1538                 timeo = sock_wait_for_wmem(sk, timeo);
1539         }
1540
1541         skb_set_owner_w(skb, sk);
1542         return skb;
1543
1544 interrupted:
1545         err = sock_intr_errno(timeo);
1546 failure:
1547         *errcode = err;
1548         return NULL;
1549 }
1550 EXPORT_SYMBOL(sock_alloc_send_pskb);
1551
1552 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1553                                     int noblock, int *errcode)
1554 {
1555         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1556 }
1557 EXPORT_SYMBOL(sock_alloc_send_skb);
1558
1559 static void __lock_sock(struct sock *sk)
1560 {
1561         DEFINE_WAIT(wait);
1562
1563         for (;;) {
1564                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1565                                         TASK_UNINTERRUPTIBLE);
1566                 spin_unlock_bh(&sk->sk_lock.slock);
1567                 schedule();
1568                 spin_lock_bh(&sk->sk_lock.slock);
1569                 if (!sock_owned_by_user(sk))
1570                         break;
1571         }
1572         finish_wait(&sk->sk_lock.wq, &wait);
1573 }
1574
1575 static void __release_sock(struct sock *sk)
1576 {
1577         struct sk_buff *skb = sk->sk_backlog.head;
1578
1579         do {
1580                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1581                 bh_unlock_sock(sk);
1582
1583                 do {
1584                         struct sk_buff *next = skb->next;
1585
1586                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1587                         skb->next = NULL;
1588                         sk_backlog_rcv(sk, skb);
1589
1590                         /*
1591                          * We are in process context here with softirqs
1592                          * disabled, use cond_resched_softirq() to preempt.
1593                          * This is safe to do because we've taken the backlog
1594                          * queue private:
1595                          */
1596                         cond_resched_softirq();
1597
1598                         skb = next;
1599                 } while (skb != NULL);
1600
1601                 bh_lock_sock(sk);
1602         } while ((skb = sk->sk_backlog.head) != NULL);
1603
1604         /*
1605          * Doing the zeroing here guarantee we can not loop forever
1606          * while a wild producer attempts to flood us.
1607          */
1608         sk->sk_backlog.len = 0;
1609 }
1610
1611 /**
1612  * sk_wait_data - wait for data to arrive at sk_receive_queue
1613  * @sk:    sock to wait on
1614  * @timeo: for how long
1615  *
1616  * Now socket state including sk->sk_err is changed only under lock,
1617  * hence we may omit checks after joining wait queue.
1618  * We check receive queue before schedule() only as optimization;
1619  * it is very likely that release_sock() added new data.
1620  */
1621 int sk_wait_data(struct sock *sk, long *timeo)
1622 {
1623         int rc;
1624         DEFINE_WAIT(wait);
1625
1626         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1627         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1628         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1629         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1630         finish_wait(sk_sleep(sk), &wait);
1631         return rc;
1632 }
1633 EXPORT_SYMBOL(sk_wait_data);
1634
1635 /**
1636  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1637  *      @sk: socket
1638  *      @size: memory size to allocate
1639  *      @kind: allocation type
1640  *
1641  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1642  *      rmem allocation. This function assumes that protocols which have
1643  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1644  */
1645 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1646 {
1647         struct proto *prot = sk->sk_prot;
1648         int amt = sk_mem_pages(size);
1649         int allocated;
1650
1651         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1652         allocated = atomic_add_return(amt, prot->memory_allocated);
1653
1654         /* Under limit. */
1655         if (allocated <= prot->sysctl_mem[0]) {
1656                 if (prot->memory_pressure && *prot->memory_pressure)
1657                         *prot->memory_pressure = 0;
1658                 return 1;
1659         }
1660
1661         /* Under pressure. */
1662         if (allocated > prot->sysctl_mem[1])
1663                 if (prot->enter_memory_pressure)
1664                         prot->enter_memory_pressure(sk);
1665
1666         /* Over hard limit. */
1667         if (allocated > prot->sysctl_mem[2])
1668                 goto suppress_allocation;
1669
1670         /* guarantee minimum buffer size under pressure */
1671         if (kind == SK_MEM_RECV) {
1672                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1673                         return 1;
1674         } else { /* SK_MEM_SEND */
1675                 if (sk->sk_type == SOCK_STREAM) {
1676                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1677                                 return 1;
1678                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1679                            prot->sysctl_wmem[0])
1680                                 return 1;
1681         }
1682
1683         if (prot->memory_pressure) {
1684                 int alloc;
1685
1686                 if (!*prot->memory_pressure)
1687                         return 1;
1688                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1689                 if (prot->sysctl_mem[2] > alloc *
1690                     sk_mem_pages(sk->sk_wmem_queued +
1691                                  atomic_read(&sk->sk_rmem_alloc) +
1692                                  sk->sk_forward_alloc))
1693                         return 1;
1694         }
1695
1696 suppress_allocation:
1697
1698         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1699                 sk_stream_moderate_sndbuf(sk);
1700
1701                 /* Fail only if socket is _under_ its sndbuf.
1702                  * In this case we cannot block, so that we have to fail.
1703                  */
1704                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1705                         return 1;
1706         }
1707
1708         /* Alas. Undo changes. */
1709         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1710         atomic_sub(amt, prot->memory_allocated);
1711         return 0;
1712 }
1713 EXPORT_SYMBOL(__sk_mem_schedule);
1714
1715 /**
1716  *      __sk_reclaim - reclaim memory_allocated
1717  *      @sk: socket
1718  */
1719 void __sk_mem_reclaim(struct sock *sk)
1720 {
1721         struct proto *prot = sk->sk_prot;
1722
1723         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1724                    prot->memory_allocated);
1725         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1726
1727         if (prot->memory_pressure && *prot->memory_pressure &&
1728             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1729                 *prot->memory_pressure = 0;
1730 }
1731 EXPORT_SYMBOL(__sk_mem_reclaim);
1732
1733
1734 /*
1735  * Set of default routines for initialising struct proto_ops when
1736  * the protocol does not support a particular function. In certain
1737  * cases where it makes no sense for a protocol to have a "do nothing"
1738  * function, some default processing is provided.
1739  */
1740
1741 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1742 {
1743         return -EOPNOTSUPP;
1744 }
1745 EXPORT_SYMBOL(sock_no_bind);
1746
1747 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1748                     int len, int flags)
1749 {
1750         return -EOPNOTSUPP;
1751 }
1752 EXPORT_SYMBOL(sock_no_connect);
1753
1754 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1755 {
1756         return -EOPNOTSUPP;
1757 }
1758 EXPORT_SYMBOL(sock_no_socketpair);
1759
1760 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1761 {
1762         return -EOPNOTSUPP;
1763 }
1764 EXPORT_SYMBOL(sock_no_accept);
1765
1766 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1767                     int *len, int peer)
1768 {
1769         return -EOPNOTSUPP;
1770 }
1771 EXPORT_SYMBOL(sock_no_getname);
1772
1773 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1774 {
1775         return 0;
1776 }
1777 EXPORT_SYMBOL(sock_no_poll);
1778
1779 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1780 {
1781         return -EOPNOTSUPP;
1782 }
1783 EXPORT_SYMBOL(sock_no_ioctl);
1784
1785 int sock_no_listen(struct socket *sock, int backlog)
1786 {
1787         return -EOPNOTSUPP;
1788 }
1789 EXPORT_SYMBOL(sock_no_listen);
1790
1791 int sock_no_shutdown(struct socket *sock, int how)
1792 {
1793         return -EOPNOTSUPP;
1794 }
1795 EXPORT_SYMBOL(sock_no_shutdown);
1796
1797 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1798                     char __user *optval, unsigned int optlen)
1799 {
1800         return -EOPNOTSUPP;
1801 }
1802 EXPORT_SYMBOL(sock_no_setsockopt);
1803
1804 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1805                     char __user *optval, int __user *optlen)
1806 {
1807         return -EOPNOTSUPP;
1808 }
1809 EXPORT_SYMBOL(sock_no_getsockopt);
1810
1811 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1812                     size_t len)
1813 {
1814         return -EOPNOTSUPP;
1815 }
1816 EXPORT_SYMBOL(sock_no_sendmsg);
1817
1818 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1819                     size_t len, int flags)
1820 {
1821         return -EOPNOTSUPP;
1822 }
1823 EXPORT_SYMBOL(sock_no_recvmsg);
1824
1825 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1826 {
1827         /* Mirror missing mmap method error code */
1828         return -ENODEV;
1829 }
1830 EXPORT_SYMBOL(sock_no_mmap);
1831
1832 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1833 {
1834         ssize_t res;
1835         struct msghdr msg = {.msg_flags = flags};
1836         struct kvec iov;
1837         char *kaddr = kmap(page);
1838         iov.iov_base = kaddr + offset;
1839         iov.iov_len = size;
1840         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1841         kunmap(page);
1842         return res;
1843 }
1844 EXPORT_SYMBOL(sock_no_sendpage);
1845
1846 /*
1847  *      Default Socket Callbacks
1848  */
1849
1850 static void sock_def_wakeup(struct sock *sk)
1851 {
1852         struct socket_wq *wq;
1853
1854         rcu_read_lock();
1855         wq = rcu_dereference(sk->sk_wq);
1856         if (wq_has_sleeper(wq))
1857                 wake_up_interruptible_all(&wq->wait);
1858         rcu_read_unlock();
1859 }
1860
1861 static void sock_def_error_report(struct sock *sk)
1862 {
1863         struct socket_wq *wq;
1864
1865         rcu_read_lock();
1866         wq = rcu_dereference(sk->sk_wq);
1867         if (wq_has_sleeper(wq))
1868                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1869         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1870         rcu_read_unlock();
1871 }
1872
1873 static void sock_def_readable(struct sock *sk, int len)
1874 {
1875         struct socket_wq *wq;
1876
1877         rcu_read_lock();
1878         wq = rcu_dereference(sk->sk_wq);
1879         if (wq_has_sleeper(wq))
1880                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
1881                                                 POLLRDNORM | POLLRDBAND);
1882         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1883         rcu_read_unlock();
1884 }
1885
1886 static void sock_def_write_space(struct sock *sk)
1887 {
1888         struct socket_wq *wq;
1889
1890         rcu_read_lock();
1891
1892         /* Do not wake up a writer until he can make "significant"
1893          * progress.  --DaveM
1894          */
1895         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1896                 wq = rcu_dereference(sk->sk_wq);
1897                 if (wq_has_sleeper(wq))
1898                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1899                                                 POLLWRNORM | POLLWRBAND);
1900
1901                 /* Should agree with poll, otherwise some programs break */
1902                 if (sock_writeable(sk))
1903                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1904         }
1905
1906         rcu_read_unlock();
1907 }
1908
1909 static void sock_def_destruct(struct sock *sk)
1910 {
1911         kfree(sk->sk_protinfo);
1912 }
1913
1914 void sk_send_sigurg(struct sock *sk)
1915 {
1916         if (sk->sk_socket && sk->sk_socket->file)
1917                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1918                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1919 }
1920 EXPORT_SYMBOL(sk_send_sigurg);
1921
1922 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1923                     unsigned long expires)
1924 {
1925         if (!mod_timer(timer, expires))
1926                 sock_hold(sk);
1927 }
1928 EXPORT_SYMBOL(sk_reset_timer);
1929
1930 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1931 {
1932         if (timer_pending(timer) && del_timer(timer))
1933                 __sock_put(sk);
1934 }
1935 EXPORT_SYMBOL(sk_stop_timer);
1936
1937 void sock_init_data(struct socket *sock, struct sock *sk)
1938 {
1939         skb_queue_head_init(&sk->sk_receive_queue);
1940         skb_queue_head_init(&sk->sk_write_queue);
1941         skb_queue_head_init(&sk->sk_error_queue);
1942 #ifdef CONFIG_NET_DMA
1943         skb_queue_head_init(&sk->sk_async_wait_queue);
1944 #endif
1945
1946         sk->sk_send_head        =       NULL;
1947
1948         init_timer(&sk->sk_timer);
1949
1950         sk->sk_allocation       =       GFP_KERNEL;
1951         sk->sk_rcvbuf           =       sysctl_rmem_default;
1952         sk->sk_sndbuf           =       sysctl_wmem_default;
1953         sk->sk_state            =       TCP_CLOSE;
1954         sk_set_socket(sk, sock);
1955
1956         sock_set_flag(sk, SOCK_ZAPPED);
1957
1958         if (sock) {
1959                 sk->sk_type     =       sock->type;
1960                 sk->sk_wq       =       sock->wq;
1961                 sock->sk        =       sk;
1962         } else
1963                 sk->sk_wq       =       NULL;
1964
1965         spin_lock_init(&sk->sk_dst_lock);
1966         rwlock_init(&sk->sk_callback_lock);
1967         lockdep_set_class_and_name(&sk->sk_callback_lock,
1968                         af_callback_keys + sk->sk_family,
1969                         af_family_clock_key_strings[sk->sk_family]);
1970
1971         sk->sk_state_change     =       sock_def_wakeup;
1972         sk->sk_data_ready       =       sock_def_readable;
1973         sk->sk_write_space      =       sock_def_write_space;
1974         sk->sk_error_report     =       sock_def_error_report;
1975         sk->sk_destruct         =       sock_def_destruct;
1976
1977         sk->sk_sndmsg_page      =       NULL;
1978         sk->sk_sndmsg_off       =       0;
1979
1980         sk->sk_peer_pid         =       NULL;
1981         sk->sk_peer_cred        =       NULL;
1982         sk->sk_write_pending    =       0;
1983         sk->sk_rcvlowat         =       1;
1984         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1985         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1986
1987         sk->sk_stamp = ktime_set(-1L, 0);
1988
1989         /*
1990          * Before updating sk_refcnt, we must commit prior changes to memory
1991          * (Documentation/RCU/rculist_nulls.txt for details)
1992          */
1993         smp_wmb();
1994         atomic_set(&sk->sk_refcnt, 1);
1995         atomic_set(&sk->sk_drops, 0);
1996 }
1997 EXPORT_SYMBOL(sock_init_data);
1998
1999 void lock_sock_nested(struct sock *sk, int subclass)
2000 {
2001         might_sleep();
2002         spin_lock_bh(&sk->sk_lock.slock);
2003         if (sk->sk_lock.owned)
2004                 __lock_sock(sk);
2005         sk->sk_lock.owned = 1;
2006         spin_unlock(&sk->sk_lock.slock);
2007         /*
2008          * The sk_lock has mutex_lock() semantics here:
2009          */
2010         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2011         local_bh_enable();
2012 }
2013 EXPORT_SYMBOL(lock_sock_nested);
2014
2015 void release_sock(struct sock *sk)
2016 {
2017         /*
2018          * The sk_lock has mutex_unlock() semantics:
2019          */
2020         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2021
2022         spin_lock_bh(&sk->sk_lock.slock);
2023         if (sk->sk_backlog.tail)
2024                 __release_sock(sk);
2025         sk->sk_lock.owned = 0;
2026         if (waitqueue_active(&sk->sk_lock.wq))
2027                 wake_up(&sk->sk_lock.wq);
2028         spin_unlock_bh(&sk->sk_lock.slock);
2029 }
2030 EXPORT_SYMBOL(release_sock);
2031
2032 /**
2033  * lock_sock_fast - fast version of lock_sock
2034  * @sk: socket
2035  *
2036  * This version should be used for very small section, where process wont block
2037  * return false if fast path is taken
2038  *   sk_lock.slock locked, owned = 0, BH disabled
2039  * return true if slow path is taken
2040  *   sk_lock.slock unlocked, owned = 1, BH enabled
2041  */
2042 bool lock_sock_fast(struct sock *sk)
2043 {
2044         might_sleep();
2045         spin_lock_bh(&sk->sk_lock.slock);
2046
2047         if (!sk->sk_lock.owned)
2048                 /*
2049                  * Note : We must disable BH
2050                  */
2051                 return false;
2052
2053         __lock_sock(sk);
2054         sk->sk_lock.owned = 1;
2055         spin_unlock(&sk->sk_lock.slock);
2056         /*
2057          * The sk_lock has mutex_lock() semantics here:
2058          */
2059         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2060         local_bh_enable();
2061         return true;
2062 }
2063 EXPORT_SYMBOL(lock_sock_fast);
2064
2065 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2066 {
2067         struct timeval tv;
2068         if (!sock_flag(sk, SOCK_TIMESTAMP))
2069                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2070         tv = ktime_to_timeval(sk->sk_stamp);
2071         if (tv.tv_sec == -1)
2072                 return -ENOENT;
2073         if (tv.tv_sec == 0) {
2074                 sk->sk_stamp = ktime_get_real();
2075                 tv = ktime_to_timeval(sk->sk_stamp);
2076         }
2077         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2078 }
2079 EXPORT_SYMBOL(sock_get_timestamp);
2080
2081 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2082 {
2083         struct timespec ts;
2084         if (!sock_flag(sk, SOCK_TIMESTAMP))
2085                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2086         ts = ktime_to_timespec(sk->sk_stamp);
2087         if (ts.tv_sec == -1)
2088                 return -ENOENT;
2089         if (ts.tv_sec == 0) {
2090                 sk->sk_stamp = ktime_get_real();
2091                 ts = ktime_to_timespec(sk->sk_stamp);
2092         }
2093         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2094 }
2095 EXPORT_SYMBOL(sock_get_timestampns);
2096
2097 void sock_enable_timestamp(struct sock *sk, int flag)
2098 {
2099         if (!sock_flag(sk, flag)) {
2100                 sock_set_flag(sk, flag);
2101                 /*
2102                  * we just set one of the two flags which require net
2103                  * time stamping, but time stamping might have been on
2104                  * already because of the other one
2105                  */
2106                 if (!sock_flag(sk,
2107                                 flag == SOCK_TIMESTAMP ?
2108                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2109                                 SOCK_TIMESTAMP))
2110                         net_enable_timestamp();
2111         }
2112 }
2113
2114 /*
2115  *      Get a socket option on an socket.
2116  *
2117  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2118  *      asynchronous errors should be reported by getsockopt. We assume
2119  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2120  */
2121 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2122                            char __user *optval, int __user *optlen)
2123 {
2124         struct sock *sk = sock->sk;
2125
2126         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2127 }
2128 EXPORT_SYMBOL(sock_common_getsockopt);
2129
2130 #ifdef CONFIG_COMPAT
2131 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2132                                   char __user *optval, int __user *optlen)
2133 {
2134         struct sock *sk = sock->sk;
2135
2136         if (sk->sk_prot->compat_getsockopt != NULL)
2137                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2138                                                       optval, optlen);
2139         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2140 }
2141 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2142 #endif
2143
2144 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2145                         struct msghdr *msg, size_t size, int flags)
2146 {
2147         struct sock *sk = sock->sk;
2148         int addr_len = 0;
2149         int err;
2150
2151         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2152                                    flags & ~MSG_DONTWAIT, &addr_len);
2153         if (err >= 0)
2154                 msg->msg_namelen = addr_len;
2155         return err;
2156 }
2157 EXPORT_SYMBOL(sock_common_recvmsg);
2158
2159 /*
2160  *      Set socket options on an inet socket.
2161  */
2162 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2163                            char __user *optval, unsigned int optlen)
2164 {
2165         struct sock *sk = sock->sk;
2166
2167         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2168 }
2169 EXPORT_SYMBOL(sock_common_setsockopt);
2170
2171 #ifdef CONFIG_COMPAT
2172 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2173                                   char __user *optval, unsigned int optlen)
2174 {
2175         struct sock *sk = sock->sk;
2176
2177         if (sk->sk_prot->compat_setsockopt != NULL)
2178                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2179                                                       optval, optlen);
2180         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2181 }
2182 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2183 #endif
2184
2185 void sk_common_release(struct sock *sk)
2186 {
2187         if (sk->sk_prot->destroy)
2188                 sk->sk_prot->destroy(sk);
2189
2190         /*
2191          * Observation: when sock_common_release is called, processes have
2192          * no access to socket. But net still has.
2193          * Step one, detach it from networking:
2194          *
2195          * A. Remove from hash tables.
2196          */
2197
2198         sk->sk_prot->unhash(sk);
2199
2200         /*
2201          * In this point socket cannot receive new packets, but it is possible
2202          * that some packets are in flight because some CPU runs receiver and
2203          * did hash table lookup before we unhashed socket. They will achieve
2204          * receive queue and will be purged by socket destructor.
2205          *
2206          * Also we still have packets pending on receive queue and probably,
2207          * our own packets waiting in device queues. sock_destroy will drain
2208          * receive queue, but transmitted packets will delay socket destruction
2209          * until the last reference will be released.
2210          */
2211
2212         sock_orphan(sk);
2213
2214         xfrm_sk_free_policy(sk);
2215
2216         sk_refcnt_debug_release(sk);
2217         sock_put(sk);
2218 }
2219 EXPORT_SYMBOL(sk_common_release);
2220
2221 static DEFINE_RWLOCK(proto_list_lock);
2222 static LIST_HEAD(proto_list);
2223
2224 #ifdef CONFIG_PROC_FS
2225 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2226 struct prot_inuse {
2227         int val[PROTO_INUSE_NR];
2228 };
2229
2230 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2231
2232 #ifdef CONFIG_NET_NS
2233 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2234 {
2235         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2236 }
2237 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2238
2239 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2240 {
2241         int cpu, idx = prot->inuse_idx;
2242         int res = 0;
2243
2244         for_each_possible_cpu(cpu)
2245                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2246
2247         return res >= 0 ? res : 0;
2248 }
2249 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2250
2251 static int __net_init sock_inuse_init_net(struct net *net)
2252 {
2253         net->core.inuse = alloc_percpu(struct prot_inuse);
2254         return net->core.inuse ? 0 : -ENOMEM;
2255 }
2256
2257 static void __net_exit sock_inuse_exit_net(struct net *net)
2258 {
2259         free_percpu(net->core.inuse);
2260 }
2261
2262 static struct pernet_operations net_inuse_ops = {
2263         .init = sock_inuse_init_net,
2264         .exit = sock_inuse_exit_net,
2265 };
2266
2267 static __init int net_inuse_init(void)
2268 {
2269         if (register_pernet_subsys(&net_inuse_ops))
2270                 panic("Cannot initialize net inuse counters");
2271
2272         return 0;
2273 }
2274
2275 core_initcall(net_inuse_init);
2276 #else
2277 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2278
2279 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2280 {
2281         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2282 }
2283 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2284
2285 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2286 {
2287         int cpu, idx = prot->inuse_idx;
2288         int res = 0;
2289
2290         for_each_possible_cpu(cpu)
2291                 res += per_cpu(prot_inuse, cpu).val[idx];
2292
2293         return res >= 0 ? res : 0;
2294 }
2295 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2296 #endif
2297
2298 static void assign_proto_idx(struct proto *prot)
2299 {
2300         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2301
2302         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2303                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2304                 return;
2305         }
2306
2307         set_bit(prot->inuse_idx, proto_inuse_idx);
2308 }
2309
2310 static void release_proto_idx(struct proto *prot)
2311 {
2312         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2313                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2314 }
2315 #else
2316 static inline void assign_proto_idx(struct proto *prot)
2317 {
2318 }
2319
2320 static inline void release_proto_idx(struct proto *prot)
2321 {
2322 }
2323 #endif
2324
2325 int proto_register(struct proto *prot, int alloc_slab)
2326 {
2327         if (alloc_slab) {
2328                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2329                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2330                                         NULL);
2331
2332                 if (prot->slab == NULL) {
2333                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2334                                prot->name);
2335                         goto out;
2336                 }
2337
2338                 if (prot->rsk_prot != NULL) {
2339                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2340                         if (prot->rsk_prot->slab_name == NULL)
2341                                 goto out_free_sock_slab;
2342
2343                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2344                                                                  prot->rsk_prot->obj_size, 0,
2345                                                                  SLAB_HWCACHE_ALIGN, NULL);
2346
2347                         if (prot->rsk_prot->slab == NULL) {
2348                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2349                                        prot->name);
2350                                 goto out_free_request_sock_slab_name;
2351                         }
2352                 }
2353
2354                 if (prot->twsk_prot != NULL) {
2355                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2356
2357                         if (prot->twsk_prot->twsk_slab_name == NULL)
2358                                 goto out_free_request_sock_slab;
2359
2360                         prot->twsk_prot->twsk_slab =
2361                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2362                                                   prot->twsk_prot->twsk_obj_size,
2363                                                   0,
2364                                                   SLAB_HWCACHE_ALIGN |
2365                                                         prot->slab_flags,
2366                                                   NULL);
2367                         if (prot->twsk_prot->twsk_slab == NULL)
2368                                 goto out_free_timewait_sock_slab_name;
2369                 }
2370         }
2371
2372         write_lock(&proto_list_lock);
2373         list_add(&prot->node, &proto_list);
2374         assign_proto_idx(prot);
2375         write_unlock(&proto_list_lock);
2376         return 0;
2377
2378 out_free_timewait_sock_slab_name:
2379         kfree(prot->twsk_prot->twsk_slab_name);
2380 out_free_request_sock_slab:
2381         if (prot->rsk_prot && prot->rsk_prot->slab) {
2382                 kmem_cache_destroy(prot->rsk_prot->slab);
2383                 prot->rsk_prot->slab = NULL;
2384         }
2385 out_free_request_sock_slab_name:
2386         if (prot->rsk_prot)
2387                 kfree(prot->rsk_prot->slab_name);
2388 out_free_sock_slab:
2389         kmem_cache_destroy(prot->slab);
2390         prot->slab = NULL;
2391 out:
2392         return -ENOBUFS;
2393 }
2394 EXPORT_SYMBOL(proto_register);
2395
2396 void proto_unregister(struct proto *prot)
2397 {
2398         write_lock(&proto_list_lock);
2399         release_proto_idx(prot);
2400         list_del(&prot->node);
2401         write_unlock(&proto_list_lock);
2402
2403         if (prot->slab != NULL) {
2404                 kmem_cache_destroy(prot->slab);
2405                 prot->slab = NULL;
2406         }
2407
2408         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2409                 kmem_cache_destroy(prot->rsk_prot->slab);
2410                 kfree(prot->rsk_prot->slab_name);
2411                 prot->rsk_prot->slab = NULL;
2412         }
2413
2414         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2415                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2416                 kfree(prot->twsk_prot->twsk_slab_name);
2417                 prot->twsk_prot->twsk_slab = NULL;
2418         }
2419 }
2420 EXPORT_SYMBOL(proto_unregister);
2421
2422 #ifdef CONFIG_PROC_FS
2423 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2424         __acquires(proto_list_lock)
2425 {
2426         read_lock(&proto_list_lock);
2427         return seq_list_start_head(&proto_list, *pos);
2428 }
2429
2430 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2431 {
2432         return seq_list_next(v, &proto_list, pos);
2433 }
2434
2435 static void proto_seq_stop(struct seq_file *seq, void *v)
2436         __releases(proto_list_lock)
2437 {
2438         read_unlock(&proto_list_lock);
2439 }
2440
2441 static char proto_method_implemented(const void *method)
2442 {
2443         return method == NULL ? 'n' : 'y';
2444 }
2445
2446 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2447 {
2448         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2449                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2450                    proto->name,
2451                    proto->obj_size,
2452                    sock_prot_inuse_get(seq_file_net(seq), proto),
2453                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2454                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2455                    proto->max_header,
2456                    proto->slab == NULL ? "no" : "yes",
2457                    module_name(proto->owner),
2458                    proto_method_implemented(proto->close),
2459                    proto_method_implemented(proto->connect),
2460                    proto_method_implemented(proto->disconnect),
2461                    proto_method_implemented(proto->accept),
2462                    proto_method_implemented(proto->ioctl),
2463                    proto_method_implemented(proto->init),
2464                    proto_method_implemented(proto->destroy),
2465                    proto_method_implemented(proto->shutdown),
2466                    proto_method_implemented(proto->setsockopt),
2467                    proto_method_implemented(proto->getsockopt),
2468                    proto_method_implemented(proto->sendmsg),
2469                    proto_method_implemented(proto->recvmsg),
2470                    proto_method_implemented(proto->sendpage),
2471                    proto_method_implemented(proto->bind),
2472                    proto_method_implemented(proto->backlog_rcv),
2473                    proto_method_implemented(proto->hash),
2474                    proto_method_implemented(proto->unhash),
2475                    proto_method_implemented(proto->get_port),
2476                    proto_method_implemented(proto->enter_memory_pressure));
2477 }
2478
2479 static int proto_seq_show(struct seq_file *seq, void *v)
2480 {
2481         if (v == &proto_list)
2482                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2483                            "protocol",
2484                            "size",
2485                            "sockets",
2486                            "memory",
2487                            "press",
2488                            "maxhdr",
2489                            "slab",
2490                            "module",
2491                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2492         else
2493                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2494         return 0;
2495 }
2496
2497 static const struct seq_operations proto_seq_ops = {
2498         .start  = proto_seq_start,
2499         .next   = proto_seq_next,
2500         .stop   = proto_seq_stop,
2501         .show   = proto_seq_show,
2502 };
2503
2504 static int proto_seq_open(struct inode *inode, struct file *file)
2505 {
2506         return seq_open_net(inode, file, &proto_seq_ops,
2507                             sizeof(struct seq_net_private));
2508 }
2509
2510 static const struct file_operations proto_seq_fops = {
2511         .owner          = THIS_MODULE,
2512         .open           = proto_seq_open,
2513         .read           = seq_read,
2514         .llseek         = seq_lseek,
2515         .release        = seq_release_net,
2516 };
2517
2518 static __net_init int proto_init_net(struct net *net)
2519 {
2520         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2521                 return -ENOMEM;
2522
2523         return 0;
2524 }
2525
2526 static __net_exit void proto_exit_net(struct net *net)
2527 {
2528         proc_net_remove(net, "protocols");
2529 }
2530
2531
2532 static __net_initdata struct pernet_operations proto_net_ops = {
2533         .init = proto_init_net,
2534         .exit = proto_exit_net,
2535 };
2536
2537 static int __init proto_init(void)
2538 {
2539         return register_pernet_subsys(&proto_net_ops);
2540 }
2541
2542 subsys_initcall(proto_init);
2543
2544 #endif /* PROC_FS */