net: Add checking to rcu_dereference() primitives
[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         unsigned 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_check(sk->sk_filter,
1077                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1078         if (filter) {
1079                 sk_filter_uncharge(sk, filter);
1080                 rcu_assign_pointer(sk->sk_filter, NULL);
1081         }
1082
1083         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1084         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1085
1086         if (atomic_read(&sk->sk_omem_alloc))
1087                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1088                        __func__, atomic_read(&sk->sk_omem_alloc));
1089
1090         put_net(sock_net(sk));
1091         sk_prot_free(sk->sk_prot_creator, sk);
1092 }
1093
1094 void sk_free(struct sock *sk)
1095 {
1096         /*
1097          * We substract one from sk_wmem_alloc and can know if
1098          * some packets are still in some tx queue.
1099          * If not null, sock_wfree() will call __sk_free(sk) later
1100          */
1101         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1102                 __sk_free(sk);
1103 }
1104 EXPORT_SYMBOL(sk_free);
1105
1106 /*
1107  * Last sock_put should drop referrence to sk->sk_net. It has already
1108  * been dropped in sk_change_net. Taking referrence to stopping namespace
1109  * is not an option.
1110  * Take referrence to a socket to remove it from hash _alive_ and after that
1111  * destroy it in the context of init_net.
1112  */
1113 void sk_release_kernel(struct sock *sk)
1114 {
1115         if (sk == NULL || sk->sk_socket == NULL)
1116                 return;
1117
1118         sock_hold(sk);
1119         sock_release(sk->sk_socket);
1120         release_net(sock_net(sk));
1121         sock_net_set(sk, get_net(&init_net));
1122         sock_put(sk);
1123 }
1124 EXPORT_SYMBOL(sk_release_kernel);
1125
1126 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1127 {
1128         struct sock *newsk;
1129
1130         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1131         if (newsk != NULL) {
1132                 struct sk_filter *filter;
1133
1134                 sock_copy(newsk, sk);
1135
1136                 /* SANITY */
1137                 get_net(sock_net(newsk));
1138                 sk_node_init(&newsk->sk_node);
1139                 sock_lock_init(newsk);
1140                 bh_lock_sock(newsk);
1141                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1142
1143                 atomic_set(&newsk->sk_rmem_alloc, 0);
1144                 /*
1145                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1146                  */
1147                 atomic_set(&newsk->sk_wmem_alloc, 1);
1148                 atomic_set(&newsk->sk_omem_alloc, 0);
1149                 skb_queue_head_init(&newsk->sk_receive_queue);
1150                 skb_queue_head_init(&newsk->sk_write_queue);
1151 #ifdef CONFIG_NET_DMA
1152                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1153 #endif
1154
1155                 rwlock_init(&newsk->sk_dst_lock);
1156                 rwlock_init(&newsk->sk_callback_lock);
1157                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1158                                 af_callback_keys + newsk->sk_family,
1159                                 af_family_clock_key_strings[newsk->sk_family]);
1160
1161                 newsk->sk_dst_cache     = NULL;
1162                 newsk->sk_wmem_queued   = 0;
1163                 newsk->sk_forward_alloc = 0;
1164                 newsk->sk_send_head     = NULL;
1165                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1166
1167                 sock_reset_flag(newsk, SOCK_DONE);
1168                 skb_queue_head_init(&newsk->sk_error_queue);
1169
1170                 filter = newsk->sk_filter;
1171                 if (filter != NULL)
1172                         sk_filter_charge(newsk, filter);
1173
1174                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1175                         /* It is still raw copy of parent, so invalidate
1176                          * destructor and make plain sk_free() */
1177                         newsk->sk_destruct = NULL;
1178                         sk_free(newsk);
1179                         newsk = NULL;
1180                         goto out;
1181                 }
1182
1183                 newsk->sk_err      = 0;
1184                 newsk->sk_priority = 0;
1185                 /*
1186                  * Before updating sk_refcnt, we must commit prior changes to memory
1187                  * (Documentation/RCU/rculist_nulls.txt for details)
1188                  */
1189                 smp_wmb();
1190                 atomic_set(&newsk->sk_refcnt, 2);
1191
1192                 /*
1193                  * Increment the counter in the same struct proto as the master
1194                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1195                  * is the same as sk->sk_prot->socks, as this field was copied
1196                  * with memcpy).
1197                  *
1198                  * This _changes_ the previous behaviour, where
1199                  * tcp_create_openreq_child always was incrementing the
1200                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1201                  * to be taken into account in all callers. -acme
1202                  */
1203                 sk_refcnt_debug_inc(newsk);
1204                 sk_set_socket(newsk, NULL);
1205                 newsk->sk_sleep  = NULL;
1206
1207                 if (newsk->sk_prot->sockets_allocated)
1208                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1209
1210                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1211                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1212                         net_enable_timestamp();
1213         }
1214 out:
1215         return newsk;
1216 }
1217 EXPORT_SYMBOL_GPL(sk_clone);
1218
1219 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1220 {
1221         __sk_dst_set(sk, dst);
1222         sk->sk_route_caps = dst->dev->features;
1223         if (sk->sk_route_caps & NETIF_F_GSO)
1224                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1225         if (sk_can_gso(sk)) {
1226                 if (dst->header_len) {
1227                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1228                 } else {
1229                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1230                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1231                 }
1232         }
1233 }
1234 EXPORT_SYMBOL_GPL(sk_setup_caps);
1235
1236 void __init sk_init(void)
1237 {
1238         if (totalram_pages <= 4096) {
1239                 sysctl_wmem_max = 32767;
1240                 sysctl_rmem_max = 32767;
1241                 sysctl_wmem_default = 32767;
1242                 sysctl_rmem_default = 32767;
1243         } else if (totalram_pages >= 131072) {
1244                 sysctl_wmem_max = 131071;
1245                 sysctl_rmem_max = 131071;
1246         }
1247 }
1248
1249 /*
1250  *      Simple resource managers for sockets.
1251  */
1252
1253
1254 /*
1255  * Write buffer destructor automatically called from kfree_skb.
1256  */
1257 void sock_wfree(struct sk_buff *skb)
1258 {
1259         struct sock *sk = skb->sk;
1260         unsigned int len = skb->truesize;
1261
1262         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1263                 /*
1264                  * Keep a reference on sk_wmem_alloc, this will be released
1265                  * after sk_write_space() call
1266                  */
1267                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1268                 sk->sk_write_space(sk);
1269                 len = 1;
1270         }
1271         /*
1272          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1273          * could not do because of in-flight packets
1274          */
1275         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1276                 __sk_free(sk);
1277 }
1278 EXPORT_SYMBOL(sock_wfree);
1279
1280 /*
1281  * Read buffer destructor automatically called from kfree_skb.
1282  */
1283 void sock_rfree(struct sk_buff *skb)
1284 {
1285         struct sock *sk = skb->sk;
1286
1287         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1288         sk_mem_uncharge(skb->sk, skb->truesize);
1289 }
1290 EXPORT_SYMBOL(sock_rfree);
1291
1292
1293 int sock_i_uid(struct sock *sk)
1294 {
1295         int uid;
1296
1297         read_lock(&sk->sk_callback_lock);
1298         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1299         read_unlock(&sk->sk_callback_lock);
1300         return uid;
1301 }
1302 EXPORT_SYMBOL(sock_i_uid);
1303
1304 unsigned long sock_i_ino(struct sock *sk)
1305 {
1306         unsigned long ino;
1307
1308         read_lock(&sk->sk_callback_lock);
1309         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1310         read_unlock(&sk->sk_callback_lock);
1311         return ino;
1312 }
1313 EXPORT_SYMBOL(sock_i_ino);
1314
1315 /*
1316  * Allocate a skb from the socket's send buffer.
1317  */
1318 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1319                              gfp_t priority)
1320 {
1321         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1322                 struct sk_buff *skb = alloc_skb(size, priority);
1323                 if (skb) {
1324                         skb_set_owner_w(skb, sk);
1325                         return skb;
1326                 }
1327         }
1328         return NULL;
1329 }
1330 EXPORT_SYMBOL(sock_wmalloc);
1331
1332 /*
1333  * Allocate a skb from the socket's receive buffer.
1334  */
1335 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1336                              gfp_t priority)
1337 {
1338         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1339                 struct sk_buff *skb = alloc_skb(size, priority);
1340                 if (skb) {
1341                         skb_set_owner_r(skb, sk);
1342                         return skb;
1343                 }
1344         }
1345         return NULL;
1346 }
1347
1348 /*
1349  * Allocate a memory block from the socket's option memory buffer.
1350  */
1351 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1352 {
1353         if ((unsigned)size <= sysctl_optmem_max &&
1354             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1355                 void *mem;
1356                 /* First do the add, to avoid the race if kmalloc
1357                  * might sleep.
1358                  */
1359                 atomic_add(size, &sk->sk_omem_alloc);
1360                 mem = kmalloc(size, priority);
1361                 if (mem)
1362                         return mem;
1363                 atomic_sub(size, &sk->sk_omem_alloc);
1364         }
1365         return NULL;
1366 }
1367 EXPORT_SYMBOL(sock_kmalloc);
1368
1369 /*
1370  * Free an option memory block.
1371  */
1372 void sock_kfree_s(struct sock *sk, void *mem, int size)
1373 {
1374         kfree(mem);
1375         atomic_sub(size, &sk->sk_omem_alloc);
1376 }
1377 EXPORT_SYMBOL(sock_kfree_s);
1378
1379 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1380    I think, these locks should be removed for datagram sockets.
1381  */
1382 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1383 {
1384         DEFINE_WAIT(wait);
1385
1386         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1387         for (;;) {
1388                 if (!timeo)
1389                         break;
1390                 if (signal_pending(current))
1391                         break;
1392                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1393                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1394                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1395                         break;
1396                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1397                         break;
1398                 if (sk->sk_err)
1399                         break;
1400                 timeo = schedule_timeout(timeo);
1401         }
1402         finish_wait(sk->sk_sleep, &wait);
1403         return timeo;
1404 }
1405
1406
1407 /*
1408  *      Generic send/receive buffer handlers
1409  */
1410
1411 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1412                                      unsigned long data_len, int noblock,
1413                                      int *errcode)
1414 {
1415         struct sk_buff *skb;
1416         gfp_t gfp_mask;
1417         long timeo;
1418         int err;
1419
1420         gfp_mask = sk->sk_allocation;
1421         if (gfp_mask & __GFP_WAIT)
1422                 gfp_mask |= __GFP_REPEAT;
1423
1424         timeo = sock_sndtimeo(sk, noblock);
1425         while (1) {
1426                 err = sock_error(sk);
1427                 if (err != 0)
1428                         goto failure;
1429
1430                 err = -EPIPE;
1431                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1432                         goto failure;
1433
1434                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1435                         skb = alloc_skb(header_len, gfp_mask);
1436                         if (skb) {
1437                                 int npages;
1438                                 int i;
1439
1440                                 /* No pages, we're done... */
1441                                 if (!data_len)
1442                                         break;
1443
1444                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1445                                 skb->truesize += data_len;
1446                                 skb_shinfo(skb)->nr_frags = npages;
1447                                 for (i = 0; i < npages; i++) {
1448                                         struct page *page;
1449                                         skb_frag_t *frag;
1450
1451                                         page = alloc_pages(sk->sk_allocation, 0);
1452                                         if (!page) {
1453                                                 err = -ENOBUFS;
1454                                                 skb_shinfo(skb)->nr_frags = i;
1455                                                 kfree_skb(skb);
1456                                                 goto failure;
1457                                         }
1458
1459                                         frag = &skb_shinfo(skb)->frags[i];
1460                                         frag->page = page;
1461                                         frag->page_offset = 0;
1462                                         frag->size = (data_len >= PAGE_SIZE ?
1463                                                       PAGE_SIZE :
1464                                                       data_len);
1465                                         data_len -= PAGE_SIZE;
1466                                 }
1467
1468                                 /* Full success... */
1469                                 break;
1470                         }
1471                         err = -ENOBUFS;
1472                         goto failure;
1473                 }
1474                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1475                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1476                 err = -EAGAIN;
1477                 if (!timeo)
1478                         goto failure;
1479                 if (signal_pending(current))
1480                         goto interrupted;
1481                 timeo = sock_wait_for_wmem(sk, timeo);
1482         }
1483
1484         skb_set_owner_w(skb, sk);
1485         return skb;
1486
1487 interrupted:
1488         err = sock_intr_errno(timeo);
1489 failure:
1490         *errcode = err;
1491         return NULL;
1492 }
1493 EXPORT_SYMBOL(sock_alloc_send_pskb);
1494
1495 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1496                                     int noblock, int *errcode)
1497 {
1498         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1499 }
1500 EXPORT_SYMBOL(sock_alloc_send_skb);
1501
1502 static void __lock_sock(struct sock *sk)
1503 {
1504         DEFINE_WAIT(wait);
1505
1506         for (;;) {
1507                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1508                                         TASK_UNINTERRUPTIBLE);
1509                 spin_unlock_bh(&sk->sk_lock.slock);
1510                 schedule();
1511                 spin_lock_bh(&sk->sk_lock.slock);
1512                 if (!sock_owned_by_user(sk))
1513                         break;
1514         }
1515         finish_wait(&sk->sk_lock.wq, &wait);
1516 }
1517
1518 static void __release_sock(struct sock *sk)
1519 {
1520         struct sk_buff *skb = sk->sk_backlog.head;
1521
1522         do {
1523                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1524                 bh_unlock_sock(sk);
1525
1526                 do {
1527                         struct sk_buff *next = skb->next;
1528
1529                         skb->next = NULL;
1530                         sk_backlog_rcv(sk, skb);
1531
1532                         /*
1533                          * We are in process context here with softirqs
1534                          * disabled, use cond_resched_softirq() to preempt.
1535                          * This is safe to do because we've taken the backlog
1536                          * queue private:
1537                          */
1538                         cond_resched_softirq();
1539
1540                         skb = next;
1541                 } while (skb != NULL);
1542
1543                 bh_lock_sock(sk);
1544         } while ((skb = sk->sk_backlog.head) != NULL);
1545 }
1546
1547 /**
1548  * sk_wait_data - wait for data to arrive at sk_receive_queue
1549  * @sk:    sock to wait on
1550  * @timeo: for how long
1551  *
1552  * Now socket state including sk->sk_err is changed only under lock,
1553  * hence we may omit checks after joining wait queue.
1554  * We check receive queue before schedule() only as optimization;
1555  * it is very likely that release_sock() added new data.
1556  */
1557 int sk_wait_data(struct sock *sk, long *timeo)
1558 {
1559         int rc;
1560         DEFINE_WAIT(wait);
1561
1562         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1563         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1564         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1565         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1566         finish_wait(sk->sk_sleep, &wait);
1567         return rc;
1568 }
1569 EXPORT_SYMBOL(sk_wait_data);
1570
1571 /**
1572  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1573  *      @sk: socket
1574  *      @size: memory size to allocate
1575  *      @kind: allocation type
1576  *
1577  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1578  *      rmem allocation. This function assumes that protocols which have
1579  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1580  */
1581 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1582 {
1583         struct proto *prot = sk->sk_prot;
1584         int amt = sk_mem_pages(size);
1585         int allocated;
1586
1587         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1588         allocated = atomic_add_return(amt, prot->memory_allocated);
1589
1590         /* Under limit. */
1591         if (allocated <= prot->sysctl_mem[0]) {
1592                 if (prot->memory_pressure && *prot->memory_pressure)
1593                         *prot->memory_pressure = 0;
1594                 return 1;
1595         }
1596
1597         /* Under pressure. */
1598         if (allocated > prot->sysctl_mem[1])
1599                 if (prot->enter_memory_pressure)
1600                         prot->enter_memory_pressure(sk);
1601
1602         /* Over hard limit. */
1603         if (allocated > prot->sysctl_mem[2])
1604                 goto suppress_allocation;
1605
1606         /* guarantee minimum buffer size under pressure */
1607         if (kind == SK_MEM_RECV) {
1608                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1609                         return 1;
1610         } else { /* SK_MEM_SEND */
1611                 if (sk->sk_type == SOCK_STREAM) {
1612                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1613                                 return 1;
1614                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1615                            prot->sysctl_wmem[0])
1616                                 return 1;
1617         }
1618
1619         if (prot->memory_pressure) {
1620                 int alloc;
1621
1622                 if (!*prot->memory_pressure)
1623                         return 1;
1624                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1625                 if (prot->sysctl_mem[2] > alloc *
1626                     sk_mem_pages(sk->sk_wmem_queued +
1627                                  atomic_read(&sk->sk_rmem_alloc) +
1628                                  sk->sk_forward_alloc))
1629                         return 1;
1630         }
1631
1632 suppress_allocation:
1633
1634         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1635                 sk_stream_moderate_sndbuf(sk);
1636
1637                 /* Fail only if socket is _under_ its sndbuf.
1638                  * In this case we cannot block, so that we have to fail.
1639                  */
1640                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1641                         return 1;
1642         }
1643
1644         /* Alas. Undo changes. */
1645         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1646         atomic_sub(amt, prot->memory_allocated);
1647         return 0;
1648 }
1649 EXPORT_SYMBOL(__sk_mem_schedule);
1650
1651 /**
1652  *      __sk_reclaim - reclaim memory_allocated
1653  *      @sk: socket
1654  */
1655 void __sk_mem_reclaim(struct sock *sk)
1656 {
1657         struct proto *prot = sk->sk_prot;
1658
1659         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1660                    prot->memory_allocated);
1661         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1662
1663         if (prot->memory_pressure && *prot->memory_pressure &&
1664             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1665                 *prot->memory_pressure = 0;
1666 }
1667 EXPORT_SYMBOL(__sk_mem_reclaim);
1668
1669
1670 /*
1671  * Set of default routines for initialising struct proto_ops when
1672  * the protocol does not support a particular function. In certain
1673  * cases where it makes no sense for a protocol to have a "do nothing"
1674  * function, some default processing is provided.
1675  */
1676
1677 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1678 {
1679         return -EOPNOTSUPP;
1680 }
1681 EXPORT_SYMBOL(sock_no_bind);
1682
1683 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1684                     int len, int flags)
1685 {
1686         return -EOPNOTSUPP;
1687 }
1688 EXPORT_SYMBOL(sock_no_connect);
1689
1690 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1691 {
1692         return -EOPNOTSUPP;
1693 }
1694 EXPORT_SYMBOL(sock_no_socketpair);
1695
1696 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1697 {
1698         return -EOPNOTSUPP;
1699 }
1700 EXPORT_SYMBOL(sock_no_accept);
1701
1702 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1703                     int *len, int peer)
1704 {
1705         return -EOPNOTSUPP;
1706 }
1707 EXPORT_SYMBOL(sock_no_getname);
1708
1709 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1710 {
1711         return 0;
1712 }
1713 EXPORT_SYMBOL(sock_no_poll);
1714
1715 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1716 {
1717         return -EOPNOTSUPP;
1718 }
1719 EXPORT_SYMBOL(sock_no_ioctl);
1720
1721 int sock_no_listen(struct socket *sock, int backlog)
1722 {
1723         return -EOPNOTSUPP;
1724 }
1725 EXPORT_SYMBOL(sock_no_listen);
1726
1727 int sock_no_shutdown(struct socket *sock, int how)
1728 {
1729         return -EOPNOTSUPP;
1730 }
1731 EXPORT_SYMBOL(sock_no_shutdown);
1732
1733 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1734                     char __user *optval, unsigned int optlen)
1735 {
1736         return -EOPNOTSUPP;
1737 }
1738 EXPORT_SYMBOL(sock_no_setsockopt);
1739
1740 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1741                     char __user *optval, int __user *optlen)
1742 {
1743         return -EOPNOTSUPP;
1744 }
1745 EXPORT_SYMBOL(sock_no_getsockopt);
1746
1747 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1748                     size_t len)
1749 {
1750         return -EOPNOTSUPP;
1751 }
1752 EXPORT_SYMBOL(sock_no_sendmsg);
1753
1754 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1755                     size_t len, int flags)
1756 {
1757         return -EOPNOTSUPP;
1758 }
1759 EXPORT_SYMBOL(sock_no_recvmsg);
1760
1761 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1762 {
1763         /* Mirror missing mmap method error code */
1764         return -ENODEV;
1765 }
1766 EXPORT_SYMBOL(sock_no_mmap);
1767
1768 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1769 {
1770         ssize_t res;
1771         struct msghdr msg = {.msg_flags = flags};
1772         struct kvec iov;
1773         char *kaddr = kmap(page);
1774         iov.iov_base = kaddr + offset;
1775         iov.iov_len = size;
1776         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1777         kunmap(page);
1778         return res;
1779 }
1780 EXPORT_SYMBOL(sock_no_sendpage);
1781
1782 /*
1783  *      Default Socket Callbacks
1784  */
1785
1786 static void sock_def_wakeup(struct sock *sk)
1787 {
1788         read_lock(&sk->sk_callback_lock);
1789         if (sk_has_sleeper(sk))
1790                 wake_up_interruptible_all(sk->sk_sleep);
1791         read_unlock(&sk->sk_callback_lock);
1792 }
1793
1794 static void sock_def_error_report(struct sock *sk)
1795 {
1796         read_lock(&sk->sk_callback_lock);
1797         if (sk_has_sleeper(sk))
1798                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1799         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1800         read_unlock(&sk->sk_callback_lock);
1801 }
1802
1803 static void sock_def_readable(struct sock *sk, int len)
1804 {
1805         read_lock(&sk->sk_callback_lock);
1806         if (sk_has_sleeper(sk))
1807                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1808                                                 POLLRDNORM | POLLRDBAND);
1809         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1810         read_unlock(&sk->sk_callback_lock);
1811 }
1812
1813 static void sock_def_write_space(struct sock *sk)
1814 {
1815         read_lock(&sk->sk_callback_lock);
1816
1817         /* Do not wake up a writer until he can make "significant"
1818          * progress.  --DaveM
1819          */
1820         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1821                 if (sk_has_sleeper(sk))
1822                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1823                                                 POLLWRNORM | POLLWRBAND);
1824
1825                 /* Should agree with poll, otherwise some programs break */
1826                 if (sock_writeable(sk))
1827                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1828         }
1829
1830         read_unlock(&sk->sk_callback_lock);
1831 }
1832
1833 static void sock_def_destruct(struct sock *sk)
1834 {
1835         kfree(sk->sk_protinfo);
1836 }
1837
1838 void sk_send_sigurg(struct sock *sk)
1839 {
1840         if (sk->sk_socket && sk->sk_socket->file)
1841                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1842                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1843 }
1844 EXPORT_SYMBOL(sk_send_sigurg);
1845
1846 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1847                     unsigned long expires)
1848 {
1849         if (!mod_timer(timer, expires))
1850                 sock_hold(sk);
1851 }
1852 EXPORT_SYMBOL(sk_reset_timer);
1853
1854 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1855 {
1856         if (timer_pending(timer) && del_timer(timer))
1857                 __sock_put(sk);
1858 }
1859 EXPORT_SYMBOL(sk_stop_timer);
1860
1861 void sock_init_data(struct socket *sock, struct sock *sk)
1862 {
1863         skb_queue_head_init(&sk->sk_receive_queue);
1864         skb_queue_head_init(&sk->sk_write_queue);
1865         skb_queue_head_init(&sk->sk_error_queue);
1866 #ifdef CONFIG_NET_DMA
1867         skb_queue_head_init(&sk->sk_async_wait_queue);
1868 #endif
1869
1870         sk->sk_send_head        =       NULL;
1871
1872         init_timer(&sk->sk_timer);
1873
1874         sk->sk_allocation       =       GFP_KERNEL;
1875         sk->sk_rcvbuf           =       sysctl_rmem_default;
1876         sk->sk_sndbuf           =       sysctl_wmem_default;
1877         sk->sk_state            =       TCP_CLOSE;
1878         sk_set_socket(sk, sock);
1879
1880         sock_set_flag(sk, SOCK_ZAPPED);
1881
1882         if (sock) {
1883                 sk->sk_type     =       sock->type;
1884                 sk->sk_sleep    =       &sock->wait;
1885                 sock->sk        =       sk;
1886         } else
1887                 sk->sk_sleep    =       NULL;
1888
1889         rwlock_init(&sk->sk_dst_lock);
1890         rwlock_init(&sk->sk_callback_lock);
1891         lockdep_set_class_and_name(&sk->sk_callback_lock,
1892                         af_callback_keys + sk->sk_family,
1893                         af_family_clock_key_strings[sk->sk_family]);
1894
1895         sk->sk_state_change     =       sock_def_wakeup;
1896         sk->sk_data_ready       =       sock_def_readable;
1897         sk->sk_write_space      =       sock_def_write_space;
1898         sk->sk_error_report     =       sock_def_error_report;
1899         sk->sk_destruct         =       sock_def_destruct;
1900
1901         sk->sk_sndmsg_page      =       NULL;
1902         sk->sk_sndmsg_off       =       0;
1903
1904         sk->sk_peercred.pid     =       0;
1905         sk->sk_peercred.uid     =       -1;
1906         sk->sk_peercred.gid     =       -1;
1907         sk->sk_write_pending    =       0;
1908         sk->sk_rcvlowat         =       1;
1909         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1910         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1911
1912         sk->sk_stamp = ktime_set(-1L, 0);
1913
1914         /*
1915          * Before updating sk_refcnt, we must commit prior changes to memory
1916          * (Documentation/RCU/rculist_nulls.txt for details)
1917          */
1918         smp_wmb();
1919         atomic_set(&sk->sk_refcnt, 1);
1920         atomic_set(&sk->sk_drops, 0);
1921 }
1922 EXPORT_SYMBOL(sock_init_data);
1923
1924 void lock_sock_nested(struct sock *sk, int subclass)
1925 {
1926         might_sleep();
1927         spin_lock_bh(&sk->sk_lock.slock);
1928         if (sk->sk_lock.owned)
1929                 __lock_sock(sk);
1930         sk->sk_lock.owned = 1;
1931         spin_unlock(&sk->sk_lock.slock);
1932         /*
1933          * The sk_lock has mutex_lock() semantics here:
1934          */
1935         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1936         local_bh_enable();
1937 }
1938 EXPORT_SYMBOL(lock_sock_nested);
1939
1940 void release_sock(struct sock *sk)
1941 {
1942         /*
1943          * The sk_lock has mutex_unlock() semantics:
1944          */
1945         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1946
1947         spin_lock_bh(&sk->sk_lock.slock);
1948         if (sk->sk_backlog.tail)
1949                 __release_sock(sk);
1950         sk->sk_lock.owned = 0;
1951         if (waitqueue_active(&sk->sk_lock.wq))
1952                 wake_up(&sk->sk_lock.wq);
1953         spin_unlock_bh(&sk->sk_lock.slock);
1954 }
1955 EXPORT_SYMBOL(release_sock);
1956
1957 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1958 {
1959         struct timeval tv;
1960         if (!sock_flag(sk, SOCK_TIMESTAMP))
1961                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1962         tv = ktime_to_timeval(sk->sk_stamp);
1963         if (tv.tv_sec == -1)
1964                 return -ENOENT;
1965         if (tv.tv_sec == 0) {
1966                 sk->sk_stamp = ktime_get_real();
1967                 tv = ktime_to_timeval(sk->sk_stamp);
1968         }
1969         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1970 }
1971 EXPORT_SYMBOL(sock_get_timestamp);
1972
1973 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1974 {
1975         struct timespec ts;
1976         if (!sock_flag(sk, SOCK_TIMESTAMP))
1977                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1978         ts = ktime_to_timespec(sk->sk_stamp);
1979         if (ts.tv_sec == -1)
1980                 return -ENOENT;
1981         if (ts.tv_sec == 0) {
1982                 sk->sk_stamp = ktime_get_real();
1983                 ts = ktime_to_timespec(sk->sk_stamp);
1984         }
1985         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1986 }
1987 EXPORT_SYMBOL(sock_get_timestampns);
1988
1989 void sock_enable_timestamp(struct sock *sk, int flag)
1990 {
1991         if (!sock_flag(sk, flag)) {
1992                 sock_set_flag(sk, flag);
1993                 /*
1994                  * we just set one of the two flags which require net
1995                  * time stamping, but time stamping might have been on
1996                  * already because of the other one
1997                  */
1998                 if (!sock_flag(sk,
1999                                 flag == SOCK_TIMESTAMP ?
2000                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2001                                 SOCK_TIMESTAMP))
2002                         net_enable_timestamp();
2003         }
2004 }
2005
2006 /*
2007  *      Get a socket option on an socket.
2008  *
2009  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2010  *      asynchronous errors should be reported by getsockopt. We assume
2011  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2012  */
2013 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2014                            char __user *optval, int __user *optlen)
2015 {
2016         struct sock *sk = sock->sk;
2017
2018         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2019 }
2020 EXPORT_SYMBOL(sock_common_getsockopt);
2021
2022 #ifdef CONFIG_COMPAT
2023 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2024                                   char __user *optval, int __user *optlen)
2025 {
2026         struct sock *sk = sock->sk;
2027
2028         if (sk->sk_prot->compat_getsockopt != NULL)
2029                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2030                                                       optval, optlen);
2031         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2032 }
2033 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2034 #endif
2035
2036 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2037                         struct msghdr *msg, size_t size, int flags)
2038 {
2039         struct sock *sk = sock->sk;
2040         int addr_len = 0;
2041         int err;
2042
2043         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2044                                    flags & ~MSG_DONTWAIT, &addr_len);
2045         if (err >= 0)
2046                 msg->msg_namelen = addr_len;
2047         return err;
2048 }
2049 EXPORT_SYMBOL(sock_common_recvmsg);
2050
2051 /*
2052  *      Set socket options on an inet socket.
2053  */
2054 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2055                            char __user *optval, unsigned int optlen)
2056 {
2057         struct sock *sk = sock->sk;
2058
2059         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2060 }
2061 EXPORT_SYMBOL(sock_common_setsockopt);
2062
2063 #ifdef CONFIG_COMPAT
2064 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2065                                   char __user *optval, unsigned int optlen)
2066 {
2067         struct sock *sk = sock->sk;
2068
2069         if (sk->sk_prot->compat_setsockopt != NULL)
2070                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2071                                                       optval, optlen);
2072         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2073 }
2074 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2075 #endif
2076
2077 void sk_common_release(struct sock *sk)
2078 {
2079         if (sk->sk_prot->destroy)
2080                 sk->sk_prot->destroy(sk);
2081
2082         /*
2083          * Observation: when sock_common_release is called, processes have
2084          * no access to socket. But net still has.
2085          * Step one, detach it from networking:
2086          *
2087          * A. Remove from hash tables.
2088          */
2089
2090         sk->sk_prot->unhash(sk);
2091
2092         /*
2093          * In this point socket cannot receive new packets, but it is possible
2094          * that some packets are in flight because some CPU runs receiver and
2095          * did hash table lookup before we unhashed socket. They will achieve
2096          * receive queue and will be purged by socket destructor.
2097          *
2098          * Also we still have packets pending on receive queue and probably,
2099          * our own packets waiting in device queues. sock_destroy will drain
2100          * receive queue, but transmitted packets will delay socket destruction
2101          * until the last reference will be released.
2102          */
2103
2104         sock_orphan(sk);
2105
2106         xfrm_sk_free_policy(sk);
2107
2108         sk_refcnt_debug_release(sk);
2109         sock_put(sk);
2110 }
2111 EXPORT_SYMBOL(sk_common_release);
2112
2113 static DEFINE_RWLOCK(proto_list_lock);
2114 static LIST_HEAD(proto_list);
2115
2116 #ifdef CONFIG_PROC_FS
2117 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2118 struct prot_inuse {
2119         int val[PROTO_INUSE_NR];
2120 };
2121
2122 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2123
2124 #ifdef CONFIG_NET_NS
2125 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2126 {
2127         int cpu = smp_processor_id();
2128         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2129 }
2130 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2131
2132 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2133 {
2134         int cpu, idx = prot->inuse_idx;
2135         int res = 0;
2136
2137         for_each_possible_cpu(cpu)
2138                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2139
2140         return res >= 0 ? res : 0;
2141 }
2142 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2143
2144 static int sock_inuse_init_net(struct net *net)
2145 {
2146         net->core.inuse = alloc_percpu(struct prot_inuse);
2147         return net->core.inuse ? 0 : -ENOMEM;
2148 }
2149
2150 static void sock_inuse_exit_net(struct net *net)
2151 {
2152         free_percpu(net->core.inuse);
2153 }
2154
2155 static struct pernet_operations net_inuse_ops = {
2156         .init = sock_inuse_init_net,
2157         .exit = sock_inuse_exit_net,
2158 };
2159
2160 static __init int net_inuse_init(void)
2161 {
2162         if (register_pernet_subsys(&net_inuse_ops))
2163                 panic("Cannot initialize net inuse counters");
2164
2165         return 0;
2166 }
2167
2168 core_initcall(net_inuse_init);
2169 #else
2170 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2171
2172 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2173 {
2174         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2175 }
2176 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2177
2178 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2179 {
2180         int cpu, idx = prot->inuse_idx;
2181         int res = 0;
2182
2183         for_each_possible_cpu(cpu)
2184                 res += per_cpu(prot_inuse, cpu).val[idx];
2185
2186         return res >= 0 ? res : 0;
2187 }
2188 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2189 #endif
2190
2191 static void assign_proto_idx(struct proto *prot)
2192 {
2193         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2194
2195         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2196                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2197                 return;
2198         }
2199
2200         set_bit(prot->inuse_idx, proto_inuse_idx);
2201 }
2202
2203 static void release_proto_idx(struct proto *prot)
2204 {
2205         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2206                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2207 }
2208 #else
2209 static inline void assign_proto_idx(struct proto *prot)
2210 {
2211 }
2212
2213 static inline void release_proto_idx(struct proto *prot)
2214 {
2215 }
2216 #endif
2217
2218 int proto_register(struct proto *prot, int alloc_slab)
2219 {
2220         if (alloc_slab) {
2221                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2222                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2223                                         NULL);
2224
2225                 if (prot->slab == NULL) {
2226                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2227                                prot->name);
2228                         goto out;
2229                 }
2230
2231                 if (prot->rsk_prot != NULL) {
2232                         static const char mask[] = "request_sock_%s";
2233
2234                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2235                         if (prot->rsk_prot->slab_name == NULL)
2236                                 goto out_free_sock_slab;
2237
2238                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2239                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2240                                                                  prot->rsk_prot->obj_size, 0,
2241                                                                  SLAB_HWCACHE_ALIGN, NULL);
2242
2243                         if (prot->rsk_prot->slab == NULL) {
2244                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2245                                        prot->name);
2246                                 goto out_free_request_sock_slab_name;
2247                         }
2248                 }
2249
2250                 if (prot->twsk_prot != NULL) {
2251                         static const char mask[] = "tw_sock_%s";
2252
2253                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2254
2255                         if (prot->twsk_prot->twsk_slab_name == NULL)
2256                                 goto out_free_request_sock_slab;
2257
2258                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2259                         prot->twsk_prot->twsk_slab =
2260                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2261                                                   prot->twsk_prot->twsk_obj_size,
2262                                                   0,
2263                                                   SLAB_HWCACHE_ALIGN |
2264                                                         prot->slab_flags,
2265                                                   NULL);
2266                         if (prot->twsk_prot->twsk_slab == NULL)
2267                                 goto out_free_timewait_sock_slab_name;
2268                 }
2269         }
2270
2271         write_lock(&proto_list_lock);
2272         list_add(&prot->node, &proto_list);
2273         assign_proto_idx(prot);
2274         write_unlock(&proto_list_lock);
2275         return 0;
2276
2277 out_free_timewait_sock_slab_name:
2278         kfree(prot->twsk_prot->twsk_slab_name);
2279 out_free_request_sock_slab:
2280         if (prot->rsk_prot && prot->rsk_prot->slab) {
2281                 kmem_cache_destroy(prot->rsk_prot->slab);
2282                 prot->rsk_prot->slab = NULL;
2283         }
2284 out_free_request_sock_slab_name:
2285         kfree(prot->rsk_prot->slab_name);
2286 out_free_sock_slab:
2287         kmem_cache_destroy(prot->slab);
2288         prot->slab = NULL;
2289 out:
2290         return -ENOBUFS;
2291 }
2292 EXPORT_SYMBOL(proto_register);
2293
2294 void proto_unregister(struct proto *prot)
2295 {
2296         write_lock(&proto_list_lock);
2297         release_proto_idx(prot);
2298         list_del(&prot->node);
2299         write_unlock(&proto_list_lock);
2300
2301         if (prot->slab != NULL) {
2302                 kmem_cache_destroy(prot->slab);
2303                 prot->slab = NULL;
2304         }
2305
2306         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2307                 kmem_cache_destroy(prot->rsk_prot->slab);
2308                 kfree(prot->rsk_prot->slab_name);
2309                 prot->rsk_prot->slab = NULL;
2310         }
2311
2312         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2313                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2314                 kfree(prot->twsk_prot->twsk_slab_name);
2315                 prot->twsk_prot->twsk_slab = NULL;
2316         }
2317 }
2318 EXPORT_SYMBOL(proto_unregister);
2319
2320 #ifdef CONFIG_PROC_FS
2321 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2322         __acquires(proto_list_lock)
2323 {
2324         read_lock(&proto_list_lock);
2325         return seq_list_start_head(&proto_list, *pos);
2326 }
2327
2328 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2329 {
2330         return seq_list_next(v, &proto_list, pos);
2331 }
2332
2333 static void proto_seq_stop(struct seq_file *seq, void *v)
2334         __releases(proto_list_lock)
2335 {
2336         read_unlock(&proto_list_lock);
2337 }
2338
2339 static char proto_method_implemented(const void *method)
2340 {
2341         return method == NULL ? 'n' : 'y';
2342 }
2343
2344 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2345 {
2346         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2347                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2348                    proto->name,
2349                    proto->obj_size,
2350                    sock_prot_inuse_get(seq_file_net(seq), proto),
2351                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2352                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2353                    proto->max_header,
2354                    proto->slab == NULL ? "no" : "yes",
2355                    module_name(proto->owner),
2356                    proto_method_implemented(proto->close),
2357                    proto_method_implemented(proto->connect),
2358                    proto_method_implemented(proto->disconnect),
2359                    proto_method_implemented(proto->accept),
2360                    proto_method_implemented(proto->ioctl),
2361                    proto_method_implemented(proto->init),
2362                    proto_method_implemented(proto->destroy),
2363                    proto_method_implemented(proto->shutdown),
2364                    proto_method_implemented(proto->setsockopt),
2365                    proto_method_implemented(proto->getsockopt),
2366                    proto_method_implemented(proto->sendmsg),
2367                    proto_method_implemented(proto->recvmsg),
2368                    proto_method_implemented(proto->sendpage),
2369                    proto_method_implemented(proto->bind),
2370                    proto_method_implemented(proto->backlog_rcv),
2371                    proto_method_implemented(proto->hash),
2372                    proto_method_implemented(proto->unhash),
2373                    proto_method_implemented(proto->get_port),
2374                    proto_method_implemented(proto->enter_memory_pressure));
2375 }
2376
2377 static int proto_seq_show(struct seq_file *seq, void *v)
2378 {
2379         if (v == &proto_list)
2380                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2381                            "protocol",
2382                            "size",
2383                            "sockets",
2384                            "memory",
2385                            "press",
2386                            "maxhdr",
2387                            "slab",
2388                            "module",
2389                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2390         else
2391                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2392         return 0;
2393 }
2394
2395 static const struct seq_operations proto_seq_ops = {
2396         .start  = proto_seq_start,
2397         .next   = proto_seq_next,
2398         .stop   = proto_seq_stop,
2399         .show   = proto_seq_show,
2400 };
2401
2402 static int proto_seq_open(struct inode *inode, struct file *file)
2403 {
2404         return seq_open_net(inode, file, &proto_seq_ops,
2405                             sizeof(struct seq_net_private));
2406 }
2407
2408 static const struct file_operations proto_seq_fops = {
2409         .owner          = THIS_MODULE,
2410         .open           = proto_seq_open,
2411         .read           = seq_read,
2412         .llseek         = seq_lseek,
2413         .release        = seq_release_net,
2414 };
2415
2416 static __net_init int proto_init_net(struct net *net)
2417 {
2418         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2419                 return -ENOMEM;
2420
2421         return 0;
2422 }
2423
2424 static __net_exit void proto_exit_net(struct net *net)
2425 {
2426         proc_net_remove(net, "protocols");
2427 }
2428
2429
2430 static __net_initdata struct pernet_operations proto_net_ops = {
2431         .init = proto_init_net,
2432         .exit = proto_exit_net,
2433 };
2434
2435 static int __init proto_init(void)
2436 {
2437         return register_pernet_subsys(&proto_net_ops);
2438 }
2439
2440 subsys_initcall(proto_init);
2441
2442 #endif /* PROC_FS */