net: spread __net_init, __net_exit
[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
114 #include <asm/uaccess.h>
115 #include <asm/system.h>
116
117 #include <linux/netdevice.h>
118 #include <net/protocol.h>
119 #include <linux/skbuff.h>
120 #include <net/net_namespace.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <linux/net_tstamp.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
126
127 #include <linux/filter.h>
128
129 #ifdef CONFIG_INET
130 #include <net/tcp.h>
131 #endif
132
133 /*
134  * Each address family might have different locking rules, so we have
135  * one slock key per address family:
136  */
137 static struct lock_class_key af_family_keys[AF_MAX];
138 static struct lock_class_key af_family_slock_keys[AF_MAX];
139
140 /*
141  * Make lock validator output more readable. (we pre-construct these
142  * strings build-time, so that runtime initialization of socket
143  * locks is fast):
144  */
145 static const char *const af_family_key_strings[AF_MAX+1] = {
146   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
147   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
148   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
149   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
150   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
151   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
152   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
153   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
154   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
155   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
156   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
157   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
158   "sk_lock-AF_IEEE802154",
159   "sk_lock-AF_MAX"
160 };
161 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
162   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
163   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
164   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
165   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
166   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
167   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
168   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
169   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
170   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
171   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
172   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
173   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
174   "slock-AF_IEEE802154",
175   "slock-AF_MAX"
176 };
177 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
178   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
179   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
180   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
181   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
182   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
183   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
184   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
185   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
186   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
187   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
188   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
189   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
190   "clock-AF_IEEE802154",
191   "clock-AF_MAX"
192 };
193
194 /*
195  * sk_callback_lock locking rules are per-address-family,
196  * so split the lock classes by using a per-AF key:
197  */
198 static struct lock_class_key af_callback_keys[AF_MAX];
199
200 /* Take into consideration the size of the struct sk_buff overhead in the
201  * determination of these values, since that is non-constant across
202  * platforms.  This makes socket queueing behavior and performance
203  * not depend upon such differences.
204  */
205 #define _SK_MEM_PACKETS         256
206 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
207 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
208 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
209
210 /* Run time adjustable parameters. */
211 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
212 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
213 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
214 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
215
216 /* Maximal space eaten by iovec or ancilliary data plus some space */
217 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
218 EXPORT_SYMBOL(sysctl_optmem_max);
219
220 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
221 {
222         struct timeval tv;
223
224         if (optlen < sizeof(tv))
225                 return -EINVAL;
226         if (copy_from_user(&tv, optval, sizeof(tv)))
227                 return -EFAULT;
228         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
229                 return -EDOM;
230
231         if (tv.tv_sec < 0) {
232                 static int warned __read_mostly;
233
234                 *timeo_p = 0;
235                 if (warned < 10 && net_ratelimit()) {
236                         warned++;
237                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
238                                "tries to set negative timeout\n",
239                                 current->comm, task_pid_nr(current));
240                 }
241                 return 0;
242         }
243         *timeo_p = MAX_SCHEDULE_TIMEOUT;
244         if (tv.tv_sec == 0 && tv.tv_usec == 0)
245                 return 0;
246         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
247                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
248         return 0;
249 }
250
251 static void sock_warn_obsolete_bsdism(const char *name)
252 {
253         static int warned;
254         static char warncomm[TASK_COMM_LEN];
255         if (strcmp(warncomm, current->comm) && warned < 5) {
256                 strcpy(warncomm,  current->comm);
257                 printk(KERN_WARNING "process `%s' is using obsolete "
258                        "%s SO_BSDCOMPAT\n", warncomm, name);
259                 warned++;
260         }
261 }
262
263 static void sock_disable_timestamp(struct sock *sk, int flag)
264 {
265         if (sock_flag(sk, flag)) {
266                 sock_reset_flag(sk, flag);
267                 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
268                     !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
269                         net_disable_timestamp();
270                 }
271         }
272 }
273
274
275 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
276 {
277         int err;
278         int skb_len;
279         unsigned long flags;
280         struct sk_buff_head *list = &sk->sk_receive_queue;
281
282         /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
283            number of warnings when compiling with -W --ANK
284          */
285         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
286             (unsigned)sk->sk_rcvbuf) {
287                 atomic_inc(&sk->sk_drops);
288                 return -ENOMEM;
289         }
290
291         err = sk_filter(sk, skb);
292         if (err)
293                 return err;
294
295         if (!sk_rmem_schedule(sk, skb->truesize)) {
296                 atomic_inc(&sk->sk_drops);
297                 return -ENOBUFS;
298         }
299
300         skb->dev = NULL;
301         skb_set_owner_r(skb, sk);
302
303         /* Cache the SKB length before we tack it onto the receive
304          * queue.  Once it is added it no longer belongs to us and
305          * may be freed by other threads of control pulling packets
306          * from the queue.
307          */
308         skb_len = skb->len;
309
310         spin_lock_irqsave(&list->lock, flags);
311         skb->dropcount = atomic_read(&sk->sk_drops);
312         __skb_queue_tail(list, skb);
313         spin_unlock_irqrestore(&list->lock, flags);
314
315         if (!sock_flag(sk, SOCK_DEAD))
316                 sk->sk_data_ready(sk, skb_len);
317         return 0;
318 }
319 EXPORT_SYMBOL(sock_queue_rcv_skb);
320
321 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
322 {
323         int rc = NET_RX_SUCCESS;
324
325         if (sk_filter(sk, skb))
326                 goto discard_and_relse;
327
328         skb->dev = NULL;
329
330         if (nested)
331                 bh_lock_sock_nested(sk);
332         else
333                 bh_lock_sock(sk);
334         if (!sock_owned_by_user(sk)) {
335                 /*
336                  * trylock + unlock semantics:
337                  */
338                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
339
340                 rc = sk_backlog_rcv(sk, skb);
341
342                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
343         } else
344                 sk_add_backlog(sk, skb);
345         bh_unlock_sock(sk);
346 out:
347         sock_put(sk);
348         return rc;
349 discard_and_relse:
350         kfree_skb(skb);
351         goto out;
352 }
353 EXPORT_SYMBOL(sk_receive_skb);
354
355 void sk_reset_txq(struct sock *sk)
356 {
357         sk_tx_queue_clear(sk);
358 }
359 EXPORT_SYMBOL(sk_reset_txq);
360
361 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
362 {
363         struct dst_entry *dst = sk->sk_dst_cache;
364
365         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
366                 sk_tx_queue_clear(sk);
367                 sk->sk_dst_cache = NULL;
368                 dst_release(dst);
369                 return NULL;
370         }
371
372         return dst;
373 }
374 EXPORT_SYMBOL(__sk_dst_check);
375
376 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
377 {
378         struct dst_entry *dst = sk_dst_get(sk);
379
380         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
381                 sk_dst_reset(sk);
382                 dst_release(dst);
383                 return NULL;
384         }
385
386         return dst;
387 }
388 EXPORT_SYMBOL(sk_dst_check);
389
390 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
391 {
392         int ret = -ENOPROTOOPT;
393 #ifdef CONFIG_NETDEVICES
394         struct net *net = sock_net(sk);
395         char devname[IFNAMSIZ];
396         int index;
397
398         /* Sorry... */
399         ret = -EPERM;
400         if (!capable(CAP_NET_RAW))
401                 goto out;
402
403         ret = -EINVAL;
404         if (optlen < 0)
405                 goto out;
406
407         /* Bind this socket to a particular device like "eth0",
408          * as specified in the passed interface name. If the
409          * name is "" or the option length is zero the socket
410          * is not bound.
411          */
412         if (optlen > IFNAMSIZ - 1)
413                 optlen = IFNAMSIZ - 1;
414         memset(devname, 0, sizeof(devname));
415
416         ret = -EFAULT;
417         if (copy_from_user(devname, optval, optlen))
418                 goto out;
419
420         index = 0;
421         if (devname[0] != '\0') {
422                 struct net_device *dev;
423
424                 rcu_read_lock();
425                 dev = dev_get_by_name_rcu(net, devname);
426                 if (dev)
427                         index = dev->ifindex;
428                 rcu_read_unlock();
429                 ret = -ENODEV;
430                 if (!dev)
431                         goto out;
432         }
433
434         lock_sock(sk);
435         sk->sk_bound_dev_if = index;
436         sk_dst_reset(sk);
437         release_sock(sk);
438
439         ret = 0;
440
441 out:
442 #endif
443
444         return ret;
445 }
446
447 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
448 {
449         if (valbool)
450                 sock_set_flag(sk, bit);
451         else
452                 sock_reset_flag(sk, bit);
453 }
454
455 /*
456  *      This is meant for all protocols to use and covers goings on
457  *      at the socket level. Everything here is generic.
458  */
459
460 int sock_setsockopt(struct socket *sock, int level, int optname,
461                     char __user *optval, unsigned int optlen)
462 {
463         struct sock *sk = sock->sk;
464         int val;
465         int valbool;
466         struct linger ling;
467         int ret = 0;
468
469         /*
470          *      Options without arguments
471          */
472
473         if (optname == SO_BINDTODEVICE)
474                 return sock_bindtodevice(sk, optval, optlen);
475
476         if (optlen < sizeof(int))
477                 return -EINVAL;
478
479         if (get_user(val, (int __user *)optval))
480                 return -EFAULT;
481
482         valbool = val ? 1 : 0;
483
484         lock_sock(sk);
485
486         switch (optname) {
487         case SO_DEBUG:
488                 if (val && !capable(CAP_NET_ADMIN))
489                         ret = -EACCES;
490                 else
491                         sock_valbool_flag(sk, SOCK_DBG, valbool);
492                 break;
493         case SO_REUSEADDR:
494                 sk->sk_reuse = valbool;
495                 break;
496         case SO_TYPE:
497         case SO_PROTOCOL:
498         case SO_DOMAIN:
499         case SO_ERROR:
500                 ret = -ENOPROTOOPT;
501                 break;
502         case SO_DONTROUTE:
503                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
504                 break;
505         case SO_BROADCAST:
506                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
507                 break;
508         case SO_SNDBUF:
509                 /* Don't error on this BSD doesn't and if you think
510                    about it this is right. Otherwise apps have to
511                    play 'guess the biggest size' games. RCVBUF/SNDBUF
512                    are treated in BSD as hints */
513
514                 if (val > sysctl_wmem_max)
515                         val = sysctl_wmem_max;
516 set_sndbuf:
517                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
518                 if ((val * 2) < SOCK_MIN_SNDBUF)
519                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
520                 else
521                         sk->sk_sndbuf = val * 2;
522
523                 /*
524                  *      Wake up sending tasks if we
525                  *      upped the value.
526                  */
527                 sk->sk_write_space(sk);
528                 break;
529
530         case SO_SNDBUFFORCE:
531                 if (!capable(CAP_NET_ADMIN)) {
532                         ret = -EPERM;
533                         break;
534                 }
535                 goto set_sndbuf;
536
537         case SO_RCVBUF:
538                 /* Don't error on this BSD doesn't and if you think
539                    about it this is right. Otherwise apps have to
540                    play 'guess the biggest size' games. RCVBUF/SNDBUF
541                    are treated in BSD as hints */
542
543                 if (val > sysctl_rmem_max)
544                         val = sysctl_rmem_max;
545 set_rcvbuf:
546                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
547                 /*
548                  * We double it on the way in to account for
549                  * "struct sk_buff" etc. overhead.   Applications
550                  * assume that the SO_RCVBUF setting they make will
551                  * allow that much actual data to be received on that
552                  * socket.
553                  *
554                  * Applications are unaware that "struct sk_buff" and
555                  * other overheads allocate from the receive buffer
556                  * during socket buffer allocation.
557                  *
558                  * And after considering the possible alternatives,
559                  * returning the value we actually used in getsockopt
560                  * is the most desirable behavior.
561                  */
562                 if ((val * 2) < SOCK_MIN_RCVBUF)
563                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
564                 else
565                         sk->sk_rcvbuf = val * 2;
566                 break;
567
568         case SO_RCVBUFFORCE:
569                 if (!capable(CAP_NET_ADMIN)) {
570                         ret = -EPERM;
571                         break;
572                 }
573                 goto set_rcvbuf;
574
575         case SO_KEEPALIVE:
576 #ifdef CONFIG_INET
577                 if (sk->sk_protocol == IPPROTO_TCP)
578                         tcp_set_keepalive(sk, valbool);
579 #endif
580                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
581                 break;
582
583         case SO_OOBINLINE:
584                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
585                 break;
586
587         case SO_NO_CHECK:
588                 sk->sk_no_check = valbool;
589                 break;
590
591         case SO_PRIORITY:
592                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
593                         sk->sk_priority = val;
594                 else
595                         ret = -EPERM;
596                 break;
597
598         case SO_LINGER:
599                 if (optlen < sizeof(ling)) {
600                         ret = -EINVAL;  /* 1003.1g */
601                         break;
602                 }
603                 if (copy_from_user(&ling, optval, sizeof(ling))) {
604                         ret = -EFAULT;
605                         break;
606                 }
607                 if (!ling.l_onoff)
608                         sock_reset_flag(sk, SOCK_LINGER);
609                 else {
610 #if (BITS_PER_LONG == 32)
611                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
612                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
613                         else
614 #endif
615                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
616                         sock_set_flag(sk, SOCK_LINGER);
617                 }
618                 break;
619
620         case SO_BSDCOMPAT:
621                 sock_warn_obsolete_bsdism("setsockopt");
622                 break;
623
624         case SO_PASSCRED:
625                 if (valbool)
626                         set_bit(SOCK_PASSCRED, &sock->flags);
627                 else
628                         clear_bit(SOCK_PASSCRED, &sock->flags);
629                 break;
630
631         case SO_TIMESTAMP:
632         case SO_TIMESTAMPNS:
633                 if (valbool)  {
634                         if (optname == SO_TIMESTAMP)
635                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
636                         else
637                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
638                         sock_set_flag(sk, SOCK_RCVTSTAMP);
639                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
640                 } else {
641                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
642                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
643                 }
644                 break;
645
646         case SO_TIMESTAMPING:
647                 if (val & ~SOF_TIMESTAMPING_MASK) {
648                         ret = -EINVAL;
649                         break;
650                 }
651                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
652                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
653                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
654                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
655                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
656                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
657                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
658                         sock_enable_timestamp(sk,
659                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
660                 else
661                         sock_disable_timestamp(sk,
662                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
663                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
664                                   val & SOF_TIMESTAMPING_SOFTWARE);
665                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
666                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
667                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
668                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
669                 break;
670
671         case SO_RCVLOWAT:
672                 if (val < 0)
673                         val = INT_MAX;
674                 sk->sk_rcvlowat = val ? : 1;
675                 break;
676
677         case SO_RCVTIMEO:
678                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
679                 break;
680
681         case SO_SNDTIMEO:
682                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
683                 break;
684
685         case SO_ATTACH_FILTER:
686                 ret = -EINVAL;
687                 if (optlen == sizeof(struct sock_fprog)) {
688                         struct sock_fprog fprog;
689
690                         ret = -EFAULT;
691                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
692                                 break;
693
694                         ret = sk_attach_filter(&fprog, sk);
695                 }
696                 break;
697
698         case SO_DETACH_FILTER:
699                 ret = sk_detach_filter(sk);
700                 break;
701
702         case SO_PASSSEC:
703                 if (valbool)
704                         set_bit(SOCK_PASSSEC, &sock->flags);
705                 else
706                         clear_bit(SOCK_PASSSEC, &sock->flags);
707                 break;
708         case SO_MARK:
709                 if (!capable(CAP_NET_ADMIN))
710                         ret = -EPERM;
711                 else
712                         sk->sk_mark = val;
713                 break;
714
715                 /* We implement the SO_SNDLOWAT etc to
716                    not be settable (1003.1g 5.3) */
717         case SO_RXQ_OVFL:
718                 if (valbool)
719                         sock_set_flag(sk, SOCK_RXQ_OVFL);
720                 else
721                         sock_reset_flag(sk, SOCK_RXQ_OVFL);
722                 break;
723         default:
724                 ret = -ENOPROTOOPT;
725                 break;
726         }
727         release_sock(sk);
728         return ret;
729 }
730 EXPORT_SYMBOL(sock_setsockopt);
731
732
733 int sock_getsockopt(struct socket *sock, int level, int optname,
734                     char __user *optval, int __user *optlen)
735 {
736         struct sock *sk = sock->sk;
737
738         union {
739                 int val;
740                 struct linger ling;
741                 struct timeval tm;
742         } v;
743
744         int lv = sizeof(int);
745         int len;
746
747         if (get_user(len, optlen))
748                 return -EFAULT;
749         if (len < 0)
750                 return -EINVAL;
751
752         memset(&v, 0, sizeof(v));
753
754         switch (optname) {
755         case SO_DEBUG:
756                 v.val = sock_flag(sk, SOCK_DBG);
757                 break;
758
759         case SO_DONTROUTE:
760                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
761                 break;
762
763         case SO_BROADCAST:
764                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
765                 break;
766
767         case SO_SNDBUF:
768                 v.val = sk->sk_sndbuf;
769                 break;
770
771         case SO_RCVBUF:
772                 v.val = sk->sk_rcvbuf;
773                 break;
774
775         case SO_REUSEADDR:
776                 v.val = sk->sk_reuse;
777                 break;
778
779         case SO_KEEPALIVE:
780                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
781                 break;
782
783         case SO_TYPE:
784                 v.val = sk->sk_type;
785                 break;
786
787         case SO_PROTOCOL:
788                 v.val = sk->sk_protocol;
789                 break;
790
791         case SO_DOMAIN:
792                 v.val = sk->sk_family;
793                 break;
794
795         case SO_ERROR:
796                 v.val = -sock_error(sk);
797                 if (v.val == 0)
798                         v.val = xchg(&sk->sk_err_soft, 0);
799                 break;
800
801         case SO_OOBINLINE:
802                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
803                 break;
804
805         case SO_NO_CHECK:
806                 v.val = sk->sk_no_check;
807                 break;
808
809         case SO_PRIORITY:
810                 v.val = sk->sk_priority;
811                 break;
812
813         case SO_LINGER:
814                 lv              = sizeof(v.ling);
815                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
816                 v.ling.l_linger = sk->sk_lingertime / HZ;
817                 break;
818
819         case SO_BSDCOMPAT:
820                 sock_warn_obsolete_bsdism("getsockopt");
821                 break;
822
823         case SO_TIMESTAMP:
824                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
825                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
826                 break;
827
828         case SO_TIMESTAMPNS:
829                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
830                 break;
831
832         case SO_TIMESTAMPING:
833                 v.val = 0;
834                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
835                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
836                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
837                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
838                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
839                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
840                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
841                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
842                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
843                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
844                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
845                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
846                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
847                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
848                 break;
849
850         case SO_RCVTIMEO:
851                 lv = sizeof(struct timeval);
852                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
853                         v.tm.tv_sec = 0;
854                         v.tm.tv_usec = 0;
855                 } else {
856                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
857                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
858                 }
859                 break;
860
861         case SO_SNDTIMEO:
862                 lv = sizeof(struct timeval);
863                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
864                         v.tm.tv_sec = 0;
865                         v.tm.tv_usec = 0;
866                 } else {
867                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
868                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
869                 }
870                 break;
871
872         case SO_RCVLOWAT:
873                 v.val = sk->sk_rcvlowat;
874                 break;
875
876         case SO_SNDLOWAT:
877                 v.val = 1;
878                 break;
879
880         case SO_PASSCRED:
881                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
882                 break;
883
884         case SO_PEERCRED:
885                 if (len > sizeof(sk->sk_peercred))
886                         len = sizeof(sk->sk_peercred);
887                 if (copy_to_user(optval, &sk->sk_peercred, len))
888                         return -EFAULT;
889                 goto lenout;
890
891         case SO_PEERNAME:
892         {
893                 char address[128];
894
895                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
896                         return -ENOTCONN;
897                 if (lv < len)
898                         return -EINVAL;
899                 if (copy_to_user(optval, address, len))
900                         return -EFAULT;
901                 goto lenout;
902         }
903
904         /* Dubious BSD thing... Probably nobody even uses it, but
905          * the UNIX standard wants it for whatever reason... -DaveM
906          */
907         case SO_ACCEPTCONN:
908                 v.val = sk->sk_state == TCP_LISTEN;
909                 break;
910
911         case SO_PASSSEC:
912                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
913                 break;
914
915         case SO_PEERSEC:
916                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
917
918         case SO_MARK:
919                 v.val = sk->sk_mark;
920                 break;
921
922         case SO_RXQ_OVFL:
923                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
924                 break;
925
926         default:
927                 return -ENOPROTOOPT;
928         }
929
930         if (len > lv)
931                 len = lv;
932         if (copy_to_user(optval, &v, len))
933                 return -EFAULT;
934 lenout:
935         if (put_user(len, optlen))
936                 return -EFAULT;
937         return 0;
938 }
939
940 /*
941  * Initialize an sk_lock.
942  *
943  * (We also register the sk_lock with the lock validator.)
944  */
945 static inline void sock_lock_init(struct sock *sk)
946 {
947         sock_lock_init_class_and_name(sk,
948                         af_family_slock_key_strings[sk->sk_family],
949                         af_family_slock_keys + sk->sk_family,
950                         af_family_key_strings[sk->sk_family],
951                         af_family_keys + sk->sk_family);
952 }
953
954 /*
955  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
956  * even temporarly, because of RCU lookups. sk_node should also be left as is.
957  */
958 static void sock_copy(struct sock *nsk, const struct sock *osk)
959 {
960 #ifdef CONFIG_SECURITY_NETWORK
961         void *sptr = nsk->sk_security;
962 #endif
963         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
964                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
965                      sizeof(osk->sk_tx_queue_mapping));
966         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
967                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
968 #ifdef CONFIG_SECURITY_NETWORK
969         nsk->sk_security = sptr;
970         security_sk_clone(osk, nsk);
971 #endif
972 }
973
974 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
975                 int family)
976 {
977         struct sock *sk;
978         struct kmem_cache *slab;
979
980         slab = prot->slab;
981         if (slab != NULL) {
982                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
983                 if (!sk)
984                         return sk;
985                 if (priority & __GFP_ZERO) {
986                         /*
987                          * caches using SLAB_DESTROY_BY_RCU should let
988                          * sk_node.next un-modified. Special care is taken
989                          * when initializing object to zero.
990                          */
991                         if (offsetof(struct sock, sk_node.next) != 0)
992                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
993                         memset(&sk->sk_node.pprev, 0,
994                                prot->obj_size - offsetof(struct sock,
995                                                          sk_node.pprev));
996                 }
997         }
998         else
999                 sk = kmalloc(prot->obj_size, priority);
1000
1001         if (sk != NULL) {
1002                 kmemcheck_annotate_bitfield(sk, flags);
1003
1004                 if (security_sk_alloc(sk, family, priority))
1005                         goto out_free;
1006
1007                 if (!try_module_get(prot->owner))
1008                         goto out_free_sec;
1009                 sk_tx_queue_clear(sk);
1010         }
1011
1012         return sk;
1013
1014 out_free_sec:
1015         security_sk_free(sk);
1016 out_free:
1017         if (slab != NULL)
1018                 kmem_cache_free(slab, sk);
1019         else
1020                 kfree(sk);
1021         return NULL;
1022 }
1023
1024 static void sk_prot_free(struct proto *prot, struct sock *sk)
1025 {
1026         struct kmem_cache *slab;
1027         struct module *owner;
1028
1029         owner = prot->owner;
1030         slab = prot->slab;
1031
1032         security_sk_free(sk);
1033         if (slab != NULL)
1034                 kmem_cache_free(slab, sk);
1035         else
1036                 kfree(sk);
1037         module_put(owner);
1038 }
1039
1040 /**
1041  *      sk_alloc - All socket objects are allocated here
1042  *      @net: the applicable net namespace
1043  *      @family: protocol family
1044  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1045  *      @prot: struct proto associated with this new sock instance
1046  */
1047 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1048                       struct proto *prot)
1049 {
1050         struct sock *sk;
1051
1052         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1053         if (sk) {
1054                 sk->sk_family = family;
1055                 /*
1056                  * See comment in struct sock definition to understand
1057                  * why we need sk_prot_creator -acme
1058                  */
1059                 sk->sk_prot = sk->sk_prot_creator = prot;
1060                 sock_lock_init(sk);
1061                 sock_net_set(sk, get_net(net));
1062                 atomic_set(&sk->sk_wmem_alloc, 1);
1063         }
1064
1065         return sk;
1066 }
1067 EXPORT_SYMBOL(sk_alloc);
1068
1069 static void __sk_free(struct sock *sk)
1070 {
1071         struct sk_filter *filter;
1072
1073         if (sk->sk_destruct)
1074                 sk->sk_destruct(sk);
1075
1076         filter = rcu_dereference(sk->sk_filter);
1077         if (filter) {
1078                 sk_filter_uncharge(sk, filter);
1079                 rcu_assign_pointer(sk->sk_filter, NULL);
1080         }
1081
1082         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1083         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1084
1085         if (atomic_read(&sk->sk_omem_alloc))
1086                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1087                        __func__, atomic_read(&sk->sk_omem_alloc));
1088
1089         put_net(sock_net(sk));
1090         sk_prot_free(sk->sk_prot_creator, sk);
1091 }
1092
1093 void sk_free(struct sock *sk)
1094 {
1095         /*
1096          * We substract one from sk_wmem_alloc and can know if
1097          * some packets are still in some tx queue.
1098          * If not null, sock_wfree() will call __sk_free(sk) later
1099          */
1100         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1101                 __sk_free(sk);
1102 }
1103 EXPORT_SYMBOL(sk_free);
1104
1105 /*
1106  * Last sock_put should drop referrence to sk->sk_net. It has already
1107  * been dropped in sk_change_net. Taking referrence to stopping namespace
1108  * is not an option.
1109  * Take referrence to a socket to remove it from hash _alive_ and after that
1110  * destroy it in the context of init_net.
1111  */
1112 void sk_release_kernel(struct sock *sk)
1113 {
1114         if (sk == NULL || sk->sk_socket == NULL)
1115                 return;
1116
1117         sock_hold(sk);
1118         sock_release(sk->sk_socket);
1119         release_net(sock_net(sk));
1120         sock_net_set(sk, get_net(&init_net));
1121         sock_put(sk);
1122 }
1123 EXPORT_SYMBOL(sk_release_kernel);
1124
1125 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1126 {
1127         struct sock *newsk;
1128
1129         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1130         if (newsk != NULL) {
1131                 struct sk_filter *filter;
1132
1133                 sock_copy(newsk, sk);
1134
1135                 /* SANITY */
1136                 get_net(sock_net(newsk));
1137                 sk_node_init(&newsk->sk_node);
1138                 sock_lock_init(newsk);
1139                 bh_lock_sock(newsk);
1140                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1141
1142                 atomic_set(&newsk->sk_rmem_alloc, 0);
1143                 /*
1144                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1145                  */
1146                 atomic_set(&newsk->sk_wmem_alloc, 1);
1147                 atomic_set(&newsk->sk_omem_alloc, 0);
1148                 skb_queue_head_init(&newsk->sk_receive_queue);
1149                 skb_queue_head_init(&newsk->sk_write_queue);
1150 #ifdef CONFIG_NET_DMA
1151                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1152 #endif
1153
1154                 rwlock_init(&newsk->sk_dst_lock);
1155                 rwlock_init(&newsk->sk_callback_lock);
1156                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1157                                 af_callback_keys + newsk->sk_family,
1158                                 af_family_clock_key_strings[newsk->sk_family]);
1159
1160                 newsk->sk_dst_cache     = NULL;
1161                 newsk->sk_wmem_queued   = 0;
1162                 newsk->sk_forward_alloc = 0;
1163                 newsk->sk_send_head     = NULL;
1164                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1165
1166                 sock_reset_flag(newsk, SOCK_DONE);
1167                 skb_queue_head_init(&newsk->sk_error_queue);
1168
1169                 filter = newsk->sk_filter;
1170                 if (filter != NULL)
1171                         sk_filter_charge(newsk, filter);
1172
1173                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1174                         /* It is still raw copy of parent, so invalidate
1175                          * destructor and make plain sk_free() */
1176                         newsk->sk_destruct = NULL;
1177                         sk_free(newsk);
1178                         newsk = NULL;
1179                         goto out;
1180                 }
1181
1182                 newsk->sk_err      = 0;
1183                 newsk->sk_priority = 0;
1184                 /*
1185                  * Before updating sk_refcnt, we must commit prior changes to memory
1186                  * (Documentation/RCU/rculist_nulls.txt for details)
1187                  */
1188                 smp_wmb();
1189                 atomic_set(&newsk->sk_refcnt, 2);
1190
1191                 /*
1192                  * Increment the counter in the same struct proto as the master
1193                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1194                  * is the same as sk->sk_prot->socks, as this field was copied
1195                  * with memcpy).
1196                  *
1197                  * This _changes_ the previous behaviour, where
1198                  * tcp_create_openreq_child always was incrementing the
1199                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1200                  * to be taken into account in all callers. -acme
1201                  */
1202                 sk_refcnt_debug_inc(newsk);
1203                 sk_set_socket(newsk, NULL);
1204                 newsk->sk_sleep  = NULL;
1205
1206                 if (newsk->sk_prot->sockets_allocated)
1207                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1208
1209                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1210                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1211                         net_enable_timestamp();
1212         }
1213 out:
1214         return newsk;
1215 }
1216 EXPORT_SYMBOL_GPL(sk_clone);
1217
1218 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1219 {
1220         __sk_dst_set(sk, dst);
1221         sk->sk_route_caps = dst->dev->features;
1222         if (sk->sk_route_caps & NETIF_F_GSO)
1223                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1224         if (sk_can_gso(sk)) {
1225                 if (dst->header_len) {
1226                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1227                 } else {
1228                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1229                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1230                 }
1231         }
1232 }
1233 EXPORT_SYMBOL_GPL(sk_setup_caps);
1234
1235 void __init sk_init(void)
1236 {
1237         if (totalram_pages <= 4096) {
1238                 sysctl_wmem_max = 32767;
1239                 sysctl_rmem_max = 32767;
1240                 sysctl_wmem_default = 32767;
1241                 sysctl_rmem_default = 32767;
1242         } else if (totalram_pages >= 131072) {
1243                 sysctl_wmem_max = 131071;
1244                 sysctl_rmem_max = 131071;
1245         }
1246 }
1247
1248 /*
1249  *      Simple resource managers for sockets.
1250  */
1251
1252
1253 /*
1254  * Write buffer destructor automatically called from kfree_skb.
1255  */
1256 void sock_wfree(struct sk_buff *skb)
1257 {
1258         struct sock *sk = skb->sk;
1259         unsigned int len = skb->truesize;
1260
1261         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1262                 /*
1263                  * Keep a reference on sk_wmem_alloc, this will be released
1264                  * after sk_write_space() call
1265                  */
1266                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1267                 sk->sk_write_space(sk);
1268                 len = 1;
1269         }
1270         /*
1271          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1272          * could not do because of in-flight packets
1273          */
1274         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1275                 __sk_free(sk);
1276 }
1277 EXPORT_SYMBOL(sock_wfree);
1278
1279 /*
1280  * Read buffer destructor automatically called from kfree_skb.
1281  */
1282 void sock_rfree(struct sk_buff *skb)
1283 {
1284         struct sock *sk = skb->sk;
1285
1286         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1287         sk_mem_uncharge(skb->sk, skb->truesize);
1288 }
1289 EXPORT_SYMBOL(sock_rfree);
1290
1291
1292 int sock_i_uid(struct sock *sk)
1293 {
1294         int uid;
1295
1296         read_lock(&sk->sk_callback_lock);
1297         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1298         read_unlock(&sk->sk_callback_lock);
1299         return uid;
1300 }
1301 EXPORT_SYMBOL(sock_i_uid);
1302
1303 unsigned long sock_i_ino(struct sock *sk)
1304 {
1305         unsigned long ino;
1306
1307         read_lock(&sk->sk_callback_lock);
1308         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1309         read_unlock(&sk->sk_callback_lock);
1310         return ino;
1311 }
1312 EXPORT_SYMBOL(sock_i_ino);
1313
1314 /*
1315  * Allocate a skb from the socket's send buffer.
1316  */
1317 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1318                              gfp_t priority)
1319 {
1320         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1321                 struct sk_buff *skb = alloc_skb(size, priority);
1322                 if (skb) {
1323                         skb_set_owner_w(skb, sk);
1324                         return skb;
1325                 }
1326         }
1327         return NULL;
1328 }
1329 EXPORT_SYMBOL(sock_wmalloc);
1330
1331 /*
1332  * Allocate a skb from the socket's receive buffer.
1333  */
1334 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1335                              gfp_t priority)
1336 {
1337         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1338                 struct sk_buff *skb = alloc_skb(size, priority);
1339                 if (skb) {
1340                         skb_set_owner_r(skb, sk);
1341                         return skb;
1342                 }
1343         }
1344         return NULL;
1345 }
1346
1347 /*
1348  * Allocate a memory block from the socket's option memory buffer.
1349  */
1350 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1351 {
1352         if ((unsigned)size <= sysctl_optmem_max &&
1353             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1354                 void *mem;
1355                 /* First do the add, to avoid the race if kmalloc
1356                  * might sleep.
1357                  */
1358                 atomic_add(size, &sk->sk_omem_alloc);
1359                 mem = kmalloc(size, priority);
1360                 if (mem)
1361                         return mem;
1362                 atomic_sub(size, &sk->sk_omem_alloc);
1363         }
1364         return NULL;
1365 }
1366 EXPORT_SYMBOL(sock_kmalloc);
1367
1368 /*
1369  * Free an option memory block.
1370  */
1371 void sock_kfree_s(struct sock *sk, void *mem, int size)
1372 {
1373         kfree(mem);
1374         atomic_sub(size, &sk->sk_omem_alloc);
1375 }
1376 EXPORT_SYMBOL(sock_kfree_s);
1377
1378 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1379    I think, these locks should be removed for datagram sockets.
1380  */
1381 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1382 {
1383         DEFINE_WAIT(wait);
1384
1385         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1386         for (;;) {
1387                 if (!timeo)
1388                         break;
1389                 if (signal_pending(current))
1390                         break;
1391                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1392                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1393                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1394                         break;
1395                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1396                         break;
1397                 if (sk->sk_err)
1398                         break;
1399                 timeo = schedule_timeout(timeo);
1400         }
1401         finish_wait(sk->sk_sleep, &wait);
1402         return timeo;
1403 }
1404
1405
1406 /*
1407  *      Generic send/receive buffer handlers
1408  */
1409
1410 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1411                                      unsigned long data_len, int noblock,
1412                                      int *errcode)
1413 {
1414         struct sk_buff *skb;
1415         gfp_t gfp_mask;
1416         long timeo;
1417         int err;
1418
1419         gfp_mask = sk->sk_allocation;
1420         if (gfp_mask & __GFP_WAIT)
1421                 gfp_mask |= __GFP_REPEAT;
1422
1423         timeo = sock_sndtimeo(sk, noblock);
1424         while (1) {
1425                 err = sock_error(sk);
1426                 if (err != 0)
1427                         goto failure;
1428
1429                 err = -EPIPE;
1430                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1431                         goto failure;
1432
1433                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1434                         skb = alloc_skb(header_len, gfp_mask);
1435                         if (skb) {
1436                                 int npages;
1437                                 int i;
1438
1439                                 /* No pages, we're done... */
1440                                 if (!data_len)
1441                                         break;
1442
1443                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1444                                 skb->truesize += data_len;
1445                                 skb_shinfo(skb)->nr_frags = npages;
1446                                 for (i = 0; i < npages; i++) {
1447                                         struct page *page;
1448                                         skb_frag_t *frag;
1449
1450                                         page = alloc_pages(sk->sk_allocation, 0);
1451                                         if (!page) {
1452                                                 err = -ENOBUFS;
1453                                                 skb_shinfo(skb)->nr_frags = i;
1454                                                 kfree_skb(skb);
1455                                                 goto failure;
1456                                         }
1457
1458                                         frag = &skb_shinfo(skb)->frags[i];
1459                                         frag->page = page;
1460                                         frag->page_offset = 0;
1461                                         frag->size = (data_len >= PAGE_SIZE ?
1462                                                       PAGE_SIZE :
1463                                                       data_len);
1464                                         data_len -= PAGE_SIZE;
1465                                 }
1466
1467                                 /* Full success... */
1468                                 break;
1469                         }
1470                         err = -ENOBUFS;
1471                         goto failure;
1472                 }
1473                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1474                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1475                 err = -EAGAIN;
1476                 if (!timeo)
1477                         goto failure;
1478                 if (signal_pending(current))
1479                         goto interrupted;
1480                 timeo = sock_wait_for_wmem(sk, timeo);
1481         }
1482
1483         skb_set_owner_w(skb, sk);
1484         return skb;
1485
1486 interrupted:
1487         err = sock_intr_errno(timeo);
1488 failure:
1489         *errcode = err;
1490         return NULL;
1491 }
1492 EXPORT_SYMBOL(sock_alloc_send_pskb);
1493
1494 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1495                                     int noblock, int *errcode)
1496 {
1497         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1498 }
1499 EXPORT_SYMBOL(sock_alloc_send_skb);
1500
1501 static void __lock_sock(struct sock *sk)
1502 {
1503         DEFINE_WAIT(wait);
1504
1505         for (;;) {
1506                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1507                                         TASK_UNINTERRUPTIBLE);
1508                 spin_unlock_bh(&sk->sk_lock.slock);
1509                 schedule();
1510                 spin_lock_bh(&sk->sk_lock.slock);
1511                 if (!sock_owned_by_user(sk))
1512                         break;
1513         }
1514         finish_wait(&sk->sk_lock.wq, &wait);
1515 }
1516
1517 static void __release_sock(struct sock *sk)
1518 {
1519         struct sk_buff *skb = sk->sk_backlog.head;
1520
1521         do {
1522                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1523                 bh_unlock_sock(sk);
1524
1525                 do {
1526                         struct sk_buff *next = skb->next;
1527
1528                         skb->next = NULL;
1529                         sk_backlog_rcv(sk, skb);
1530
1531                         /*
1532                          * We are in process context here with softirqs
1533                          * disabled, use cond_resched_softirq() to preempt.
1534                          * This is safe to do because we've taken the backlog
1535                          * queue private:
1536                          */
1537                         cond_resched_softirq();
1538
1539                         skb = next;
1540                 } while (skb != NULL);
1541
1542                 bh_lock_sock(sk);
1543         } while ((skb = sk->sk_backlog.head) != NULL);
1544 }
1545
1546 /**
1547  * sk_wait_data - wait for data to arrive at sk_receive_queue
1548  * @sk:    sock to wait on
1549  * @timeo: for how long
1550  *
1551  * Now socket state including sk->sk_err is changed only under lock,
1552  * hence we may omit checks after joining wait queue.
1553  * We check receive queue before schedule() only as optimization;
1554  * it is very likely that release_sock() added new data.
1555  */
1556 int sk_wait_data(struct sock *sk, long *timeo)
1557 {
1558         int rc;
1559         DEFINE_WAIT(wait);
1560
1561         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1562         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1563         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1564         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1565         finish_wait(sk->sk_sleep, &wait);
1566         return rc;
1567 }
1568 EXPORT_SYMBOL(sk_wait_data);
1569
1570 /**
1571  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1572  *      @sk: socket
1573  *      @size: memory size to allocate
1574  *      @kind: allocation type
1575  *
1576  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1577  *      rmem allocation. This function assumes that protocols which have
1578  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1579  */
1580 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1581 {
1582         struct proto *prot = sk->sk_prot;
1583         int amt = sk_mem_pages(size);
1584         int allocated;
1585
1586         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1587         allocated = atomic_add_return(amt, prot->memory_allocated);
1588
1589         /* Under limit. */
1590         if (allocated <= prot->sysctl_mem[0]) {
1591                 if (prot->memory_pressure && *prot->memory_pressure)
1592                         *prot->memory_pressure = 0;
1593                 return 1;
1594         }
1595
1596         /* Under pressure. */
1597         if (allocated > prot->sysctl_mem[1])
1598                 if (prot->enter_memory_pressure)
1599                         prot->enter_memory_pressure(sk);
1600
1601         /* Over hard limit. */
1602         if (allocated > prot->sysctl_mem[2])
1603                 goto suppress_allocation;
1604
1605         /* guarantee minimum buffer size under pressure */
1606         if (kind == SK_MEM_RECV) {
1607                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1608                         return 1;
1609         } else { /* SK_MEM_SEND */
1610                 if (sk->sk_type == SOCK_STREAM) {
1611                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1612                                 return 1;
1613                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1614                            prot->sysctl_wmem[0])
1615                                 return 1;
1616         }
1617
1618         if (prot->memory_pressure) {
1619                 int alloc;
1620
1621                 if (!*prot->memory_pressure)
1622                         return 1;
1623                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1624                 if (prot->sysctl_mem[2] > alloc *
1625                     sk_mem_pages(sk->sk_wmem_queued +
1626                                  atomic_read(&sk->sk_rmem_alloc) +
1627                                  sk->sk_forward_alloc))
1628                         return 1;
1629         }
1630
1631 suppress_allocation:
1632
1633         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1634                 sk_stream_moderate_sndbuf(sk);
1635
1636                 /* Fail only if socket is _under_ its sndbuf.
1637                  * In this case we cannot block, so that we have to fail.
1638                  */
1639                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1640                         return 1;
1641         }
1642
1643         /* Alas. Undo changes. */
1644         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1645         atomic_sub(amt, prot->memory_allocated);
1646         return 0;
1647 }
1648 EXPORT_SYMBOL(__sk_mem_schedule);
1649
1650 /**
1651  *      __sk_reclaim - reclaim memory_allocated
1652  *      @sk: socket
1653  */
1654 void __sk_mem_reclaim(struct sock *sk)
1655 {
1656         struct proto *prot = sk->sk_prot;
1657
1658         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1659                    prot->memory_allocated);
1660         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1661
1662         if (prot->memory_pressure && *prot->memory_pressure &&
1663             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1664                 *prot->memory_pressure = 0;
1665 }
1666 EXPORT_SYMBOL(__sk_mem_reclaim);
1667
1668
1669 /*
1670  * Set of default routines for initialising struct proto_ops when
1671  * the protocol does not support a particular function. In certain
1672  * cases where it makes no sense for a protocol to have a "do nothing"
1673  * function, some default processing is provided.
1674  */
1675
1676 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1677 {
1678         return -EOPNOTSUPP;
1679 }
1680 EXPORT_SYMBOL(sock_no_bind);
1681
1682 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1683                     int len, int flags)
1684 {
1685         return -EOPNOTSUPP;
1686 }
1687 EXPORT_SYMBOL(sock_no_connect);
1688
1689 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1690 {
1691         return -EOPNOTSUPP;
1692 }
1693 EXPORT_SYMBOL(sock_no_socketpair);
1694
1695 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1696 {
1697         return -EOPNOTSUPP;
1698 }
1699 EXPORT_SYMBOL(sock_no_accept);
1700
1701 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1702                     int *len, int peer)
1703 {
1704         return -EOPNOTSUPP;
1705 }
1706 EXPORT_SYMBOL(sock_no_getname);
1707
1708 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1709 {
1710         return 0;
1711 }
1712 EXPORT_SYMBOL(sock_no_poll);
1713
1714 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1715 {
1716         return -EOPNOTSUPP;
1717 }
1718 EXPORT_SYMBOL(sock_no_ioctl);
1719
1720 int sock_no_listen(struct socket *sock, int backlog)
1721 {
1722         return -EOPNOTSUPP;
1723 }
1724 EXPORT_SYMBOL(sock_no_listen);
1725
1726 int sock_no_shutdown(struct socket *sock, int how)
1727 {
1728         return -EOPNOTSUPP;
1729 }
1730 EXPORT_SYMBOL(sock_no_shutdown);
1731
1732 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1733                     char __user *optval, unsigned int optlen)
1734 {
1735         return -EOPNOTSUPP;
1736 }
1737 EXPORT_SYMBOL(sock_no_setsockopt);
1738
1739 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1740                     char __user *optval, int __user *optlen)
1741 {
1742         return -EOPNOTSUPP;
1743 }
1744 EXPORT_SYMBOL(sock_no_getsockopt);
1745
1746 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1747                     size_t len)
1748 {
1749         return -EOPNOTSUPP;
1750 }
1751 EXPORT_SYMBOL(sock_no_sendmsg);
1752
1753 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1754                     size_t len, int flags)
1755 {
1756         return -EOPNOTSUPP;
1757 }
1758 EXPORT_SYMBOL(sock_no_recvmsg);
1759
1760 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1761 {
1762         /* Mirror missing mmap method error code */
1763         return -ENODEV;
1764 }
1765 EXPORT_SYMBOL(sock_no_mmap);
1766
1767 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1768 {
1769         ssize_t res;
1770         struct msghdr msg = {.msg_flags = flags};
1771         struct kvec iov;
1772         char *kaddr = kmap(page);
1773         iov.iov_base = kaddr + offset;
1774         iov.iov_len = size;
1775         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1776         kunmap(page);
1777         return res;
1778 }
1779 EXPORT_SYMBOL(sock_no_sendpage);
1780
1781 /*
1782  *      Default Socket Callbacks
1783  */
1784
1785 static void sock_def_wakeup(struct sock *sk)
1786 {
1787         read_lock(&sk->sk_callback_lock);
1788         if (sk_has_sleeper(sk))
1789                 wake_up_interruptible_all(sk->sk_sleep);
1790         read_unlock(&sk->sk_callback_lock);
1791 }
1792
1793 static void sock_def_error_report(struct sock *sk)
1794 {
1795         read_lock(&sk->sk_callback_lock);
1796         if (sk_has_sleeper(sk))
1797                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1798         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1799         read_unlock(&sk->sk_callback_lock);
1800 }
1801
1802 static void sock_def_readable(struct sock *sk, int len)
1803 {
1804         read_lock(&sk->sk_callback_lock);
1805         if (sk_has_sleeper(sk))
1806                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1807                                                 POLLRDNORM | POLLRDBAND);
1808         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1809         read_unlock(&sk->sk_callback_lock);
1810 }
1811
1812 static void sock_def_write_space(struct sock *sk)
1813 {
1814         read_lock(&sk->sk_callback_lock);
1815
1816         /* Do not wake up a writer until he can make "significant"
1817          * progress.  --DaveM
1818          */
1819         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1820                 if (sk_has_sleeper(sk))
1821                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1822                                                 POLLWRNORM | POLLWRBAND);
1823
1824                 /* Should agree with poll, otherwise some programs break */
1825                 if (sock_writeable(sk))
1826                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1827         }
1828
1829         read_unlock(&sk->sk_callback_lock);
1830 }
1831
1832 static void sock_def_destruct(struct sock *sk)
1833 {
1834         kfree(sk->sk_protinfo);
1835 }
1836
1837 void sk_send_sigurg(struct sock *sk)
1838 {
1839         if (sk->sk_socket && sk->sk_socket->file)
1840                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1841                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1842 }
1843 EXPORT_SYMBOL(sk_send_sigurg);
1844
1845 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1846                     unsigned long expires)
1847 {
1848         if (!mod_timer(timer, expires))
1849                 sock_hold(sk);
1850 }
1851 EXPORT_SYMBOL(sk_reset_timer);
1852
1853 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1854 {
1855         if (timer_pending(timer) && del_timer(timer))
1856                 __sock_put(sk);
1857 }
1858 EXPORT_SYMBOL(sk_stop_timer);
1859
1860 void sock_init_data(struct socket *sock, struct sock *sk)
1861 {
1862         skb_queue_head_init(&sk->sk_receive_queue);
1863         skb_queue_head_init(&sk->sk_write_queue);
1864         skb_queue_head_init(&sk->sk_error_queue);
1865 #ifdef CONFIG_NET_DMA
1866         skb_queue_head_init(&sk->sk_async_wait_queue);
1867 #endif
1868
1869         sk->sk_send_head        =       NULL;
1870
1871         init_timer(&sk->sk_timer);
1872
1873         sk->sk_allocation       =       GFP_KERNEL;
1874         sk->sk_rcvbuf           =       sysctl_rmem_default;
1875         sk->sk_sndbuf           =       sysctl_wmem_default;
1876         sk->sk_state            =       TCP_CLOSE;
1877         sk_set_socket(sk, sock);
1878
1879         sock_set_flag(sk, SOCK_ZAPPED);
1880
1881         if (sock) {
1882                 sk->sk_type     =       sock->type;
1883                 sk->sk_sleep    =       &sock->wait;
1884                 sock->sk        =       sk;
1885         } else
1886                 sk->sk_sleep    =       NULL;
1887
1888         rwlock_init(&sk->sk_dst_lock);
1889         rwlock_init(&sk->sk_callback_lock);
1890         lockdep_set_class_and_name(&sk->sk_callback_lock,
1891                         af_callback_keys + sk->sk_family,
1892                         af_family_clock_key_strings[sk->sk_family]);
1893
1894         sk->sk_state_change     =       sock_def_wakeup;
1895         sk->sk_data_ready       =       sock_def_readable;
1896         sk->sk_write_space      =       sock_def_write_space;
1897         sk->sk_error_report     =       sock_def_error_report;
1898         sk->sk_destruct         =       sock_def_destruct;
1899
1900         sk->sk_sndmsg_page      =       NULL;
1901         sk->sk_sndmsg_off       =       0;
1902
1903         sk->sk_peercred.pid     =       0;
1904         sk->sk_peercred.uid     =       -1;
1905         sk->sk_peercred.gid     =       -1;
1906         sk->sk_write_pending    =       0;
1907         sk->sk_rcvlowat         =       1;
1908         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1909         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1910
1911         sk->sk_stamp = ktime_set(-1L, 0);
1912
1913         /*
1914          * Before updating sk_refcnt, we must commit prior changes to memory
1915          * (Documentation/RCU/rculist_nulls.txt for details)
1916          */
1917         smp_wmb();
1918         atomic_set(&sk->sk_refcnt, 1);
1919         atomic_set(&sk->sk_drops, 0);
1920 }
1921 EXPORT_SYMBOL(sock_init_data);
1922
1923 void lock_sock_nested(struct sock *sk, int subclass)
1924 {
1925         might_sleep();
1926         spin_lock_bh(&sk->sk_lock.slock);
1927         if (sk->sk_lock.owned)
1928                 __lock_sock(sk);
1929         sk->sk_lock.owned = 1;
1930         spin_unlock(&sk->sk_lock.slock);
1931         /*
1932          * The sk_lock has mutex_lock() semantics here:
1933          */
1934         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1935         local_bh_enable();
1936 }
1937 EXPORT_SYMBOL(lock_sock_nested);
1938
1939 void release_sock(struct sock *sk)
1940 {
1941         /*
1942          * The sk_lock has mutex_unlock() semantics:
1943          */
1944         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1945
1946         spin_lock_bh(&sk->sk_lock.slock);
1947         if (sk->sk_backlog.tail)
1948                 __release_sock(sk);
1949         sk->sk_lock.owned = 0;
1950         if (waitqueue_active(&sk->sk_lock.wq))
1951                 wake_up(&sk->sk_lock.wq);
1952         spin_unlock_bh(&sk->sk_lock.slock);
1953 }
1954 EXPORT_SYMBOL(release_sock);
1955
1956 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1957 {
1958         struct timeval tv;
1959         if (!sock_flag(sk, SOCK_TIMESTAMP))
1960                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1961         tv = ktime_to_timeval(sk->sk_stamp);
1962         if (tv.tv_sec == -1)
1963                 return -ENOENT;
1964         if (tv.tv_sec == 0) {
1965                 sk->sk_stamp = ktime_get_real();
1966                 tv = ktime_to_timeval(sk->sk_stamp);
1967         }
1968         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1969 }
1970 EXPORT_SYMBOL(sock_get_timestamp);
1971
1972 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1973 {
1974         struct timespec ts;
1975         if (!sock_flag(sk, SOCK_TIMESTAMP))
1976                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1977         ts = ktime_to_timespec(sk->sk_stamp);
1978         if (ts.tv_sec == -1)
1979                 return -ENOENT;
1980         if (ts.tv_sec == 0) {
1981                 sk->sk_stamp = ktime_get_real();
1982                 ts = ktime_to_timespec(sk->sk_stamp);
1983         }
1984         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1985 }
1986 EXPORT_SYMBOL(sock_get_timestampns);
1987
1988 void sock_enable_timestamp(struct sock *sk, int flag)
1989 {
1990         if (!sock_flag(sk, flag)) {
1991                 sock_set_flag(sk, flag);
1992                 /*
1993                  * we just set one of the two flags which require net
1994                  * time stamping, but time stamping might have been on
1995                  * already because of the other one
1996                  */
1997                 if (!sock_flag(sk,
1998                                 flag == SOCK_TIMESTAMP ?
1999                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2000                                 SOCK_TIMESTAMP))
2001                         net_enable_timestamp();
2002         }
2003 }
2004
2005 /*
2006  *      Get a socket option on an socket.
2007  *
2008  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2009  *      asynchronous errors should be reported by getsockopt. We assume
2010  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2011  */
2012 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2013                            char __user *optval, int __user *optlen)
2014 {
2015         struct sock *sk = sock->sk;
2016
2017         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2018 }
2019 EXPORT_SYMBOL(sock_common_getsockopt);
2020
2021 #ifdef CONFIG_COMPAT
2022 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2023                                   char __user *optval, int __user *optlen)
2024 {
2025         struct sock *sk = sock->sk;
2026
2027         if (sk->sk_prot->compat_getsockopt != NULL)
2028                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2029                                                       optval, optlen);
2030         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2031 }
2032 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2033 #endif
2034
2035 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2036                         struct msghdr *msg, size_t size, int flags)
2037 {
2038         struct sock *sk = sock->sk;
2039         int addr_len = 0;
2040         int err;
2041
2042         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2043                                    flags & ~MSG_DONTWAIT, &addr_len);
2044         if (err >= 0)
2045                 msg->msg_namelen = addr_len;
2046         return err;
2047 }
2048 EXPORT_SYMBOL(sock_common_recvmsg);
2049
2050 /*
2051  *      Set socket options on an inet socket.
2052  */
2053 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2054                            char __user *optval, unsigned int optlen)
2055 {
2056         struct sock *sk = sock->sk;
2057
2058         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2059 }
2060 EXPORT_SYMBOL(sock_common_setsockopt);
2061
2062 #ifdef CONFIG_COMPAT
2063 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2064                                   char __user *optval, unsigned int optlen)
2065 {
2066         struct sock *sk = sock->sk;
2067
2068         if (sk->sk_prot->compat_setsockopt != NULL)
2069                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2070                                                       optval, optlen);
2071         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2072 }
2073 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2074 #endif
2075
2076 void sk_common_release(struct sock *sk)
2077 {
2078         if (sk->sk_prot->destroy)
2079                 sk->sk_prot->destroy(sk);
2080
2081         /*
2082          * Observation: when sock_common_release is called, processes have
2083          * no access to socket. But net still has.
2084          * Step one, detach it from networking:
2085          *
2086          * A. Remove from hash tables.
2087          */
2088
2089         sk->sk_prot->unhash(sk);
2090
2091         /*
2092          * In this point socket cannot receive new packets, but it is possible
2093          * that some packets are in flight because some CPU runs receiver and
2094          * did hash table lookup before we unhashed socket. They will achieve
2095          * receive queue and will be purged by socket destructor.
2096          *
2097          * Also we still have packets pending on receive queue and probably,
2098          * our own packets waiting in device queues. sock_destroy will drain
2099          * receive queue, but transmitted packets will delay socket destruction
2100          * until the last reference will be released.
2101          */
2102
2103         sock_orphan(sk);
2104
2105         xfrm_sk_free_policy(sk);
2106
2107         sk_refcnt_debug_release(sk);
2108         sock_put(sk);
2109 }
2110 EXPORT_SYMBOL(sk_common_release);
2111
2112 static DEFINE_RWLOCK(proto_list_lock);
2113 static LIST_HEAD(proto_list);
2114
2115 #ifdef CONFIG_PROC_FS
2116 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2117 struct prot_inuse {
2118         int val[PROTO_INUSE_NR];
2119 };
2120
2121 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2122
2123 #ifdef CONFIG_NET_NS
2124 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2125 {
2126         int cpu = smp_processor_id();
2127         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2128 }
2129 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2130
2131 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2132 {
2133         int cpu, idx = prot->inuse_idx;
2134         int res = 0;
2135
2136         for_each_possible_cpu(cpu)
2137                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2138
2139         return res >= 0 ? res : 0;
2140 }
2141 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2142
2143 static int __net_init sock_inuse_init_net(struct net *net)
2144 {
2145         net->core.inuse = alloc_percpu(struct prot_inuse);
2146         return net->core.inuse ? 0 : -ENOMEM;
2147 }
2148
2149 static void __net_exit sock_inuse_exit_net(struct net *net)
2150 {
2151         free_percpu(net->core.inuse);
2152 }
2153
2154 static struct pernet_operations net_inuse_ops = {
2155         .init = sock_inuse_init_net,
2156         .exit = sock_inuse_exit_net,
2157 };
2158
2159 static __init int net_inuse_init(void)
2160 {
2161         if (register_pernet_subsys(&net_inuse_ops))
2162                 panic("Cannot initialize net inuse counters");
2163
2164         return 0;
2165 }
2166
2167 core_initcall(net_inuse_init);
2168 #else
2169 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2170
2171 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2172 {
2173         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2174 }
2175 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2176
2177 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2178 {
2179         int cpu, idx = prot->inuse_idx;
2180         int res = 0;
2181
2182         for_each_possible_cpu(cpu)
2183                 res += per_cpu(prot_inuse, cpu).val[idx];
2184
2185         return res >= 0 ? res : 0;
2186 }
2187 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2188 #endif
2189
2190 static void assign_proto_idx(struct proto *prot)
2191 {
2192         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2193
2194         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2195                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2196                 return;
2197         }
2198
2199         set_bit(prot->inuse_idx, proto_inuse_idx);
2200 }
2201
2202 static void release_proto_idx(struct proto *prot)
2203 {
2204         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2205                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2206 }
2207 #else
2208 static inline void assign_proto_idx(struct proto *prot)
2209 {
2210 }
2211
2212 static inline void release_proto_idx(struct proto *prot)
2213 {
2214 }
2215 #endif
2216
2217 int proto_register(struct proto *prot, int alloc_slab)
2218 {
2219         if (alloc_slab) {
2220                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2221                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2222                                         NULL);
2223
2224                 if (prot->slab == NULL) {
2225                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2226                                prot->name);
2227                         goto out;
2228                 }
2229
2230                 if (prot->rsk_prot != NULL) {
2231                         static const char mask[] = "request_sock_%s";
2232
2233                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2234                         if (prot->rsk_prot->slab_name == NULL)
2235                                 goto out_free_sock_slab;
2236
2237                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2238                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2239                                                                  prot->rsk_prot->obj_size, 0,
2240                                                                  SLAB_HWCACHE_ALIGN, NULL);
2241
2242                         if (prot->rsk_prot->slab == NULL) {
2243                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2244                                        prot->name);
2245                                 goto out_free_request_sock_slab_name;
2246                         }
2247                 }
2248
2249                 if (prot->twsk_prot != NULL) {
2250                         static const char mask[] = "tw_sock_%s";
2251
2252                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2253
2254                         if (prot->twsk_prot->twsk_slab_name == NULL)
2255                                 goto out_free_request_sock_slab;
2256
2257                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2258                         prot->twsk_prot->twsk_slab =
2259                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2260                                                   prot->twsk_prot->twsk_obj_size,
2261                                                   0,
2262                                                   SLAB_HWCACHE_ALIGN |
2263                                                         prot->slab_flags,
2264                                                   NULL);
2265                         if (prot->twsk_prot->twsk_slab == NULL)
2266                                 goto out_free_timewait_sock_slab_name;
2267                 }
2268         }
2269
2270         write_lock(&proto_list_lock);
2271         list_add(&prot->node, &proto_list);
2272         assign_proto_idx(prot);
2273         write_unlock(&proto_list_lock);
2274         return 0;
2275
2276 out_free_timewait_sock_slab_name:
2277         kfree(prot->twsk_prot->twsk_slab_name);
2278 out_free_request_sock_slab:
2279         if (prot->rsk_prot && prot->rsk_prot->slab) {
2280                 kmem_cache_destroy(prot->rsk_prot->slab);
2281                 prot->rsk_prot->slab = NULL;
2282         }
2283 out_free_request_sock_slab_name:
2284         kfree(prot->rsk_prot->slab_name);
2285 out_free_sock_slab:
2286         kmem_cache_destroy(prot->slab);
2287         prot->slab = NULL;
2288 out:
2289         return -ENOBUFS;
2290 }
2291 EXPORT_SYMBOL(proto_register);
2292
2293 void proto_unregister(struct proto *prot)
2294 {
2295         write_lock(&proto_list_lock);
2296         release_proto_idx(prot);
2297         list_del(&prot->node);
2298         write_unlock(&proto_list_lock);
2299
2300         if (prot->slab != NULL) {
2301                 kmem_cache_destroy(prot->slab);
2302                 prot->slab = NULL;
2303         }
2304
2305         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2306                 kmem_cache_destroy(prot->rsk_prot->slab);
2307                 kfree(prot->rsk_prot->slab_name);
2308                 prot->rsk_prot->slab = NULL;
2309         }
2310
2311         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2312                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2313                 kfree(prot->twsk_prot->twsk_slab_name);
2314                 prot->twsk_prot->twsk_slab = NULL;
2315         }
2316 }
2317 EXPORT_SYMBOL(proto_unregister);
2318
2319 #ifdef CONFIG_PROC_FS
2320 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2321         __acquires(proto_list_lock)
2322 {
2323         read_lock(&proto_list_lock);
2324         return seq_list_start_head(&proto_list, *pos);
2325 }
2326
2327 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2328 {
2329         return seq_list_next(v, &proto_list, pos);
2330 }
2331
2332 static void proto_seq_stop(struct seq_file *seq, void *v)
2333         __releases(proto_list_lock)
2334 {
2335         read_unlock(&proto_list_lock);
2336 }
2337
2338 static char proto_method_implemented(const void *method)
2339 {
2340         return method == NULL ? 'n' : 'y';
2341 }
2342
2343 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2344 {
2345         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2346                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2347                    proto->name,
2348                    proto->obj_size,
2349                    sock_prot_inuse_get(seq_file_net(seq), proto),
2350                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2351                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2352                    proto->max_header,
2353                    proto->slab == NULL ? "no" : "yes",
2354                    module_name(proto->owner),
2355                    proto_method_implemented(proto->close),
2356                    proto_method_implemented(proto->connect),
2357                    proto_method_implemented(proto->disconnect),
2358                    proto_method_implemented(proto->accept),
2359                    proto_method_implemented(proto->ioctl),
2360                    proto_method_implemented(proto->init),
2361                    proto_method_implemented(proto->destroy),
2362                    proto_method_implemented(proto->shutdown),
2363                    proto_method_implemented(proto->setsockopt),
2364                    proto_method_implemented(proto->getsockopt),
2365                    proto_method_implemented(proto->sendmsg),
2366                    proto_method_implemented(proto->recvmsg),
2367                    proto_method_implemented(proto->sendpage),
2368                    proto_method_implemented(proto->bind),
2369                    proto_method_implemented(proto->backlog_rcv),
2370                    proto_method_implemented(proto->hash),
2371                    proto_method_implemented(proto->unhash),
2372                    proto_method_implemented(proto->get_port),
2373                    proto_method_implemented(proto->enter_memory_pressure));
2374 }
2375
2376 static int proto_seq_show(struct seq_file *seq, void *v)
2377 {
2378         if (v == &proto_list)
2379                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2380                            "protocol",
2381                            "size",
2382                            "sockets",
2383                            "memory",
2384                            "press",
2385                            "maxhdr",
2386                            "slab",
2387                            "module",
2388                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2389         else
2390                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2391         return 0;
2392 }
2393
2394 static const struct seq_operations proto_seq_ops = {
2395         .start  = proto_seq_start,
2396         .next   = proto_seq_next,
2397         .stop   = proto_seq_stop,
2398         .show   = proto_seq_show,
2399 };
2400
2401 static int proto_seq_open(struct inode *inode, struct file *file)
2402 {
2403         return seq_open_net(inode, file, &proto_seq_ops,
2404                             sizeof(struct seq_net_private));
2405 }
2406
2407 static const struct file_operations proto_seq_fops = {
2408         .owner          = THIS_MODULE,
2409         .open           = proto_seq_open,
2410         .read           = seq_read,
2411         .llseek         = seq_lseek,
2412         .release        = seq_release_net,
2413 };
2414
2415 static __net_init int proto_init_net(struct net *net)
2416 {
2417         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2418                 return -ENOMEM;
2419
2420         return 0;
2421 }
2422
2423 static __net_exit void proto_exit_net(struct net *net)
2424 {
2425         proc_net_remove(net, "protocols");
2426 }
2427
2428
2429 static __net_initdata struct pernet_operations proto_net_ops = {
2430         .init = proto_init_net,
2431         .exit = proto_exit_net,
2432 };
2433
2434 static int __init proto_init(void)
2435 {
2436         return register_pernet_subsys(&proto_net_ops);
2437 }
2438
2439 subsys_initcall(proto_init);
2440
2441 #endif /* PROC_FS */