6e22dc973d23305692182f8b830937b96e2ffbce
[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 if (sk_add_backlog_limited(sk, skb)) {
344                 bh_unlock_sock(sk);
345                 atomic_inc(&sk->sk_drops);
346                 goto discard_and_relse;
347         }
348
349         bh_unlock_sock(sk);
350 out:
351         sock_put(sk);
352         return rc;
353 discard_and_relse:
354         kfree_skb(skb);
355         goto out;
356 }
357 EXPORT_SYMBOL(sk_receive_skb);
358
359 void sk_reset_txq(struct sock *sk)
360 {
361         sk_tx_queue_clear(sk);
362 }
363 EXPORT_SYMBOL(sk_reset_txq);
364
365 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
366 {
367         struct dst_entry *dst = sk->sk_dst_cache;
368
369         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
370                 sk_tx_queue_clear(sk);
371                 sk->sk_dst_cache = NULL;
372                 dst_release(dst);
373                 return NULL;
374         }
375
376         return dst;
377 }
378 EXPORT_SYMBOL(__sk_dst_check);
379
380 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
381 {
382         struct dst_entry *dst = sk_dst_get(sk);
383
384         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
385                 sk_dst_reset(sk);
386                 dst_release(dst);
387                 return NULL;
388         }
389
390         return dst;
391 }
392 EXPORT_SYMBOL(sk_dst_check);
393
394 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
395 {
396         int ret = -ENOPROTOOPT;
397 #ifdef CONFIG_NETDEVICES
398         struct net *net = sock_net(sk);
399         char devname[IFNAMSIZ];
400         int index;
401
402         /* Sorry... */
403         ret = -EPERM;
404         if (!capable(CAP_NET_RAW))
405                 goto out;
406
407         ret = -EINVAL;
408         if (optlen < 0)
409                 goto out;
410
411         /* Bind this socket to a particular device like "eth0",
412          * as specified in the passed interface name. If the
413          * name is "" or the option length is zero the socket
414          * is not bound.
415          */
416         if (optlen > IFNAMSIZ - 1)
417                 optlen = IFNAMSIZ - 1;
418         memset(devname, 0, sizeof(devname));
419
420         ret = -EFAULT;
421         if (copy_from_user(devname, optval, optlen))
422                 goto out;
423
424         index = 0;
425         if (devname[0] != '\0') {
426                 struct net_device *dev;
427
428                 rcu_read_lock();
429                 dev = dev_get_by_name_rcu(net, devname);
430                 if (dev)
431                         index = dev->ifindex;
432                 rcu_read_unlock();
433                 ret = -ENODEV;
434                 if (!dev)
435                         goto out;
436         }
437
438         lock_sock(sk);
439         sk->sk_bound_dev_if = index;
440         sk_dst_reset(sk);
441         release_sock(sk);
442
443         ret = 0;
444
445 out:
446 #endif
447
448         return ret;
449 }
450
451 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
452 {
453         if (valbool)
454                 sock_set_flag(sk, bit);
455         else
456                 sock_reset_flag(sk, bit);
457 }
458
459 /*
460  *      This is meant for all protocols to use and covers goings on
461  *      at the socket level. Everything here is generic.
462  */
463
464 int sock_setsockopt(struct socket *sock, int level, int optname,
465                     char __user *optval, unsigned int optlen)
466 {
467         struct sock *sk = sock->sk;
468         int val;
469         int valbool;
470         struct linger ling;
471         int ret = 0;
472
473         /*
474          *      Options without arguments
475          */
476
477         if (optname == SO_BINDTODEVICE)
478                 return sock_bindtodevice(sk, optval, optlen);
479
480         if (optlen < sizeof(int))
481                 return -EINVAL;
482
483         if (get_user(val, (int __user *)optval))
484                 return -EFAULT;
485
486         valbool = val ? 1 : 0;
487
488         lock_sock(sk);
489
490         switch (optname) {
491         case SO_DEBUG:
492                 if (val && !capable(CAP_NET_ADMIN))
493                         ret = -EACCES;
494                 else
495                         sock_valbool_flag(sk, SOCK_DBG, valbool);
496                 break;
497         case SO_REUSEADDR:
498                 sk->sk_reuse = valbool;
499                 break;
500         case SO_TYPE:
501         case SO_PROTOCOL:
502         case SO_DOMAIN:
503         case SO_ERROR:
504                 ret = -ENOPROTOOPT;
505                 break;
506         case SO_DONTROUTE:
507                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
508                 break;
509         case SO_BROADCAST:
510                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
511                 break;
512         case SO_SNDBUF:
513                 /* Don't error on this BSD doesn't and if you think
514                    about it this is right. Otherwise apps have to
515                    play 'guess the biggest size' games. RCVBUF/SNDBUF
516                    are treated in BSD as hints */
517
518                 if (val > sysctl_wmem_max)
519                         val = sysctl_wmem_max;
520 set_sndbuf:
521                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
522                 if ((val * 2) < SOCK_MIN_SNDBUF)
523                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
524                 else
525                         sk->sk_sndbuf = val * 2;
526
527                 /*
528                  *      Wake up sending tasks if we
529                  *      upped the value.
530                  */
531                 sk->sk_write_space(sk);
532                 break;
533
534         case SO_SNDBUFFORCE:
535                 if (!capable(CAP_NET_ADMIN)) {
536                         ret = -EPERM;
537                         break;
538                 }
539                 goto set_sndbuf;
540
541         case SO_RCVBUF:
542                 /* Don't error on this BSD doesn't and if you think
543                    about it this is right. Otherwise apps have to
544                    play 'guess the biggest size' games. RCVBUF/SNDBUF
545                    are treated in BSD as hints */
546
547                 if (val > sysctl_rmem_max)
548                         val = sysctl_rmem_max;
549 set_rcvbuf:
550                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
551                 /*
552                  * We double it on the way in to account for
553                  * "struct sk_buff" etc. overhead.   Applications
554                  * assume that the SO_RCVBUF setting they make will
555                  * allow that much actual data to be received on that
556                  * socket.
557                  *
558                  * Applications are unaware that "struct sk_buff" and
559                  * other overheads allocate from the receive buffer
560                  * during socket buffer allocation.
561                  *
562                  * And after considering the possible alternatives,
563                  * returning the value we actually used in getsockopt
564                  * is the most desirable behavior.
565                  */
566                 if ((val * 2) < SOCK_MIN_RCVBUF)
567                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
568                 else
569                         sk->sk_rcvbuf = val * 2;
570                 break;
571
572         case SO_RCVBUFFORCE:
573                 if (!capable(CAP_NET_ADMIN)) {
574                         ret = -EPERM;
575                         break;
576                 }
577                 goto set_rcvbuf;
578
579         case SO_KEEPALIVE:
580 #ifdef CONFIG_INET
581                 if (sk->sk_protocol == IPPROTO_TCP)
582                         tcp_set_keepalive(sk, valbool);
583 #endif
584                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
585                 break;
586
587         case SO_OOBINLINE:
588                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
589                 break;
590
591         case SO_NO_CHECK:
592                 sk->sk_no_check = valbool;
593                 break;
594
595         case SO_PRIORITY:
596                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
597                         sk->sk_priority = val;
598                 else
599                         ret = -EPERM;
600                 break;
601
602         case SO_LINGER:
603                 if (optlen < sizeof(ling)) {
604                         ret = -EINVAL;  /* 1003.1g */
605                         break;
606                 }
607                 if (copy_from_user(&ling, optval, sizeof(ling))) {
608                         ret = -EFAULT;
609                         break;
610                 }
611                 if (!ling.l_onoff)
612                         sock_reset_flag(sk, SOCK_LINGER);
613                 else {
614 #if (BITS_PER_LONG == 32)
615                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
616                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
617                         else
618 #endif
619                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
620                         sock_set_flag(sk, SOCK_LINGER);
621                 }
622                 break;
623
624         case SO_BSDCOMPAT:
625                 sock_warn_obsolete_bsdism("setsockopt");
626                 break;
627
628         case SO_PASSCRED:
629                 if (valbool)
630                         set_bit(SOCK_PASSCRED, &sock->flags);
631                 else
632                         clear_bit(SOCK_PASSCRED, &sock->flags);
633                 break;
634
635         case SO_TIMESTAMP:
636         case SO_TIMESTAMPNS:
637                 if (valbool)  {
638                         if (optname == SO_TIMESTAMP)
639                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
640                         else
641                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
642                         sock_set_flag(sk, SOCK_RCVTSTAMP);
643                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
644                 } else {
645                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
646                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
647                 }
648                 break;
649
650         case SO_TIMESTAMPING:
651                 if (val & ~SOF_TIMESTAMPING_MASK) {
652                         ret = -EINVAL;
653                         break;
654                 }
655                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
656                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
657                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
658                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
659                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
660                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
661                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
662                         sock_enable_timestamp(sk,
663                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
664                 else
665                         sock_disable_timestamp(sk,
666                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
667                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
668                                   val & SOF_TIMESTAMPING_SOFTWARE);
669                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
670                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
671                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
672                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
673                 break;
674
675         case SO_RCVLOWAT:
676                 if (val < 0)
677                         val = INT_MAX;
678                 sk->sk_rcvlowat = val ? : 1;
679                 break;
680
681         case SO_RCVTIMEO:
682                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
683                 break;
684
685         case SO_SNDTIMEO:
686                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
687                 break;
688
689         case SO_ATTACH_FILTER:
690                 ret = -EINVAL;
691                 if (optlen == sizeof(struct sock_fprog)) {
692                         struct sock_fprog fprog;
693
694                         ret = -EFAULT;
695                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
696                                 break;
697
698                         ret = sk_attach_filter(&fprog, sk);
699                 }
700                 break;
701
702         case SO_DETACH_FILTER:
703                 ret = sk_detach_filter(sk);
704                 break;
705
706         case SO_PASSSEC:
707                 if (valbool)
708                         set_bit(SOCK_PASSSEC, &sock->flags);
709                 else
710                         clear_bit(SOCK_PASSSEC, &sock->flags);
711                 break;
712         case SO_MARK:
713                 if (!capable(CAP_NET_ADMIN))
714                         ret = -EPERM;
715                 else
716                         sk->sk_mark = val;
717                 break;
718
719                 /* We implement the SO_SNDLOWAT etc to
720                    not be settable (1003.1g 5.3) */
721         case SO_RXQ_OVFL:
722                 if (valbool)
723                         sock_set_flag(sk, SOCK_RXQ_OVFL);
724                 else
725                         sock_reset_flag(sk, SOCK_RXQ_OVFL);
726                 break;
727         default:
728                 ret = -ENOPROTOOPT;
729                 break;
730         }
731         release_sock(sk);
732         return ret;
733 }
734 EXPORT_SYMBOL(sock_setsockopt);
735
736
737 int sock_getsockopt(struct socket *sock, int level, int optname,
738                     char __user *optval, int __user *optlen)
739 {
740         struct sock *sk = sock->sk;
741
742         union {
743                 int val;
744                 struct linger ling;
745                 struct timeval tm;
746         } v;
747
748         int lv = sizeof(int);
749         int len;
750
751         if (get_user(len, optlen))
752                 return -EFAULT;
753         if (len < 0)
754                 return -EINVAL;
755
756         memset(&v, 0, sizeof(v));
757
758         switch (optname) {
759         case SO_DEBUG:
760                 v.val = sock_flag(sk, SOCK_DBG);
761                 break;
762
763         case SO_DONTROUTE:
764                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
765                 break;
766
767         case SO_BROADCAST:
768                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
769                 break;
770
771         case SO_SNDBUF:
772                 v.val = sk->sk_sndbuf;
773                 break;
774
775         case SO_RCVBUF:
776                 v.val = sk->sk_rcvbuf;
777                 break;
778
779         case SO_REUSEADDR:
780                 v.val = sk->sk_reuse;
781                 break;
782
783         case SO_KEEPALIVE:
784                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
785                 break;
786
787         case SO_TYPE:
788                 v.val = sk->sk_type;
789                 break;
790
791         case SO_PROTOCOL:
792                 v.val = sk->sk_protocol;
793                 break;
794
795         case SO_DOMAIN:
796                 v.val = sk->sk_family;
797                 break;
798
799         case SO_ERROR:
800                 v.val = -sock_error(sk);
801                 if (v.val == 0)
802                         v.val = xchg(&sk->sk_err_soft, 0);
803                 break;
804
805         case SO_OOBINLINE:
806                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
807                 break;
808
809         case SO_NO_CHECK:
810                 v.val = sk->sk_no_check;
811                 break;
812
813         case SO_PRIORITY:
814                 v.val = sk->sk_priority;
815                 break;
816
817         case SO_LINGER:
818                 lv              = sizeof(v.ling);
819                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
820                 v.ling.l_linger = sk->sk_lingertime / HZ;
821                 break;
822
823         case SO_BSDCOMPAT:
824                 sock_warn_obsolete_bsdism("getsockopt");
825                 break;
826
827         case SO_TIMESTAMP:
828                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
829                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
830                 break;
831
832         case SO_TIMESTAMPNS:
833                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
834                 break;
835
836         case SO_TIMESTAMPING:
837                 v.val = 0;
838                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
839                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
840                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
841                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
842                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
843                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
844                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
845                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
846                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
847                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
848                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
849                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
850                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
851                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
852                 break;
853
854         case SO_RCVTIMEO:
855                 lv = sizeof(struct timeval);
856                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
857                         v.tm.tv_sec = 0;
858                         v.tm.tv_usec = 0;
859                 } else {
860                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
861                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
862                 }
863                 break;
864
865         case SO_SNDTIMEO:
866                 lv = sizeof(struct timeval);
867                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
868                         v.tm.tv_sec = 0;
869                         v.tm.tv_usec = 0;
870                 } else {
871                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
872                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
873                 }
874                 break;
875
876         case SO_RCVLOWAT:
877                 v.val = sk->sk_rcvlowat;
878                 break;
879
880         case SO_SNDLOWAT:
881                 v.val = 1;
882                 break;
883
884         case SO_PASSCRED:
885                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
886                 break;
887
888         case SO_PEERCRED:
889                 if (len > sizeof(sk->sk_peercred))
890                         len = sizeof(sk->sk_peercred);
891                 if (copy_to_user(optval, &sk->sk_peercred, len))
892                         return -EFAULT;
893                 goto lenout;
894
895         case SO_PEERNAME:
896         {
897                 char address[128];
898
899                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
900                         return -ENOTCONN;
901                 if (lv < len)
902                         return -EINVAL;
903                 if (copy_to_user(optval, address, len))
904                         return -EFAULT;
905                 goto lenout;
906         }
907
908         /* Dubious BSD thing... Probably nobody even uses it, but
909          * the UNIX standard wants it for whatever reason... -DaveM
910          */
911         case SO_ACCEPTCONN:
912                 v.val = sk->sk_state == TCP_LISTEN;
913                 break;
914
915         case SO_PASSSEC:
916                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
917                 break;
918
919         case SO_PEERSEC:
920                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
921
922         case SO_MARK:
923                 v.val = sk->sk_mark;
924                 break;
925
926         case SO_RXQ_OVFL:
927                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
928                 break;
929
930         default:
931                 return -ENOPROTOOPT;
932         }
933
934         if (len > lv)
935                 len = lv;
936         if (copy_to_user(optval, &v, len))
937                 return -EFAULT;
938 lenout:
939         if (put_user(len, optlen))
940                 return -EFAULT;
941         return 0;
942 }
943
944 /*
945  * Initialize an sk_lock.
946  *
947  * (We also register the sk_lock with the lock validator.)
948  */
949 static inline void sock_lock_init(struct sock *sk)
950 {
951         sock_lock_init_class_and_name(sk,
952                         af_family_slock_key_strings[sk->sk_family],
953                         af_family_slock_keys + sk->sk_family,
954                         af_family_key_strings[sk->sk_family],
955                         af_family_keys + sk->sk_family);
956 }
957
958 /*
959  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
960  * even temporarly, because of RCU lookups. sk_node should also be left as is.
961  */
962 static void sock_copy(struct sock *nsk, const struct sock *osk)
963 {
964 #ifdef CONFIG_SECURITY_NETWORK
965         void *sptr = nsk->sk_security;
966 #endif
967         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
968                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
969                      sizeof(osk->sk_tx_queue_mapping));
970         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
971                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
972 #ifdef CONFIG_SECURITY_NETWORK
973         nsk->sk_security = sptr;
974         security_sk_clone(osk, nsk);
975 #endif
976 }
977
978 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
979                 int family)
980 {
981         struct sock *sk;
982         struct kmem_cache *slab;
983
984         slab = prot->slab;
985         if (slab != NULL) {
986                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
987                 if (!sk)
988                         return sk;
989                 if (priority & __GFP_ZERO) {
990                         /*
991                          * caches using SLAB_DESTROY_BY_RCU should let
992                          * sk_node.next un-modified. Special care is taken
993                          * when initializing object to zero.
994                          */
995                         if (offsetof(struct sock, sk_node.next) != 0)
996                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
997                         memset(&sk->sk_node.pprev, 0,
998                                prot->obj_size - offsetof(struct sock,
999                                                          sk_node.pprev));
1000                 }
1001         }
1002         else
1003                 sk = kmalloc(prot->obj_size, priority);
1004
1005         if (sk != NULL) {
1006                 kmemcheck_annotate_bitfield(sk, flags);
1007
1008                 if (security_sk_alloc(sk, family, priority))
1009                         goto out_free;
1010
1011                 if (!try_module_get(prot->owner))
1012                         goto out_free_sec;
1013                 sk_tx_queue_clear(sk);
1014         }
1015
1016         return sk;
1017
1018 out_free_sec:
1019         security_sk_free(sk);
1020 out_free:
1021         if (slab != NULL)
1022                 kmem_cache_free(slab, sk);
1023         else
1024                 kfree(sk);
1025         return NULL;
1026 }
1027
1028 static void sk_prot_free(struct proto *prot, struct sock *sk)
1029 {
1030         struct kmem_cache *slab;
1031         struct module *owner;
1032
1033         owner = prot->owner;
1034         slab = prot->slab;
1035
1036         security_sk_free(sk);
1037         if (slab != NULL)
1038                 kmem_cache_free(slab, sk);
1039         else
1040                 kfree(sk);
1041         module_put(owner);
1042 }
1043
1044 /**
1045  *      sk_alloc - All socket objects are allocated here
1046  *      @net: the applicable net namespace
1047  *      @family: protocol family
1048  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1049  *      @prot: struct proto associated with this new sock instance
1050  */
1051 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1052                       struct proto *prot)
1053 {
1054         struct sock *sk;
1055
1056         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1057         if (sk) {
1058                 sk->sk_family = family;
1059                 /*
1060                  * See comment in struct sock definition to understand
1061                  * why we need sk_prot_creator -acme
1062                  */
1063                 sk->sk_prot = sk->sk_prot_creator = prot;
1064                 sock_lock_init(sk);
1065                 sock_net_set(sk, get_net(net));
1066                 atomic_set(&sk->sk_wmem_alloc, 1);
1067         }
1068
1069         return sk;
1070 }
1071 EXPORT_SYMBOL(sk_alloc);
1072
1073 static void __sk_free(struct sock *sk)
1074 {
1075         struct sk_filter *filter;
1076
1077         if (sk->sk_destruct)
1078                 sk->sk_destruct(sk);
1079
1080         filter = rcu_dereference_check(sk->sk_filter,
1081                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1082         if (filter) {
1083                 sk_filter_uncharge(sk, filter);
1084                 rcu_assign_pointer(sk->sk_filter, NULL);
1085         }
1086
1087         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1088         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1089
1090         if (atomic_read(&sk->sk_omem_alloc))
1091                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1092                        __func__, atomic_read(&sk->sk_omem_alloc));
1093
1094         put_net(sock_net(sk));
1095         sk_prot_free(sk->sk_prot_creator, sk);
1096 }
1097
1098 void sk_free(struct sock *sk)
1099 {
1100         /*
1101          * We substract one from sk_wmem_alloc and can know if
1102          * some packets are still in some tx queue.
1103          * If not null, sock_wfree() will call __sk_free(sk) later
1104          */
1105         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1106                 __sk_free(sk);
1107 }
1108 EXPORT_SYMBOL(sk_free);
1109
1110 /*
1111  * Last sock_put should drop referrence to sk->sk_net. It has already
1112  * been dropped in sk_change_net. Taking referrence to stopping namespace
1113  * is not an option.
1114  * Take referrence to a socket to remove it from hash _alive_ and after that
1115  * destroy it in the context of init_net.
1116  */
1117 void sk_release_kernel(struct sock *sk)
1118 {
1119         if (sk == NULL || sk->sk_socket == NULL)
1120                 return;
1121
1122         sock_hold(sk);
1123         sock_release(sk->sk_socket);
1124         release_net(sock_net(sk));
1125         sock_net_set(sk, get_net(&init_net));
1126         sock_put(sk);
1127 }
1128 EXPORT_SYMBOL(sk_release_kernel);
1129
1130 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1131 {
1132         struct sock *newsk;
1133
1134         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1135         if (newsk != NULL) {
1136                 struct sk_filter *filter;
1137
1138                 sock_copy(newsk, sk);
1139
1140                 /* SANITY */
1141                 get_net(sock_net(newsk));
1142                 sk_node_init(&newsk->sk_node);
1143                 sock_lock_init(newsk);
1144                 bh_lock_sock(newsk);
1145                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1146                 newsk->sk_backlog.len = 0;
1147
1148                 atomic_set(&newsk->sk_rmem_alloc, 0);
1149                 /*
1150                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1151                  */
1152                 atomic_set(&newsk->sk_wmem_alloc, 1);
1153                 atomic_set(&newsk->sk_omem_alloc, 0);
1154                 skb_queue_head_init(&newsk->sk_receive_queue);
1155                 skb_queue_head_init(&newsk->sk_write_queue);
1156 #ifdef CONFIG_NET_DMA
1157                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1158 #endif
1159
1160                 rwlock_init(&newsk->sk_dst_lock);
1161                 rwlock_init(&newsk->sk_callback_lock);
1162                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1163                                 af_callback_keys + newsk->sk_family,
1164                                 af_family_clock_key_strings[newsk->sk_family]);
1165
1166                 newsk->sk_dst_cache     = NULL;
1167                 newsk->sk_wmem_queued   = 0;
1168                 newsk->sk_forward_alloc = 0;
1169                 newsk->sk_send_head     = NULL;
1170                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1171
1172                 sock_reset_flag(newsk, SOCK_DONE);
1173                 skb_queue_head_init(&newsk->sk_error_queue);
1174
1175                 filter = newsk->sk_filter;
1176                 if (filter != NULL)
1177                         sk_filter_charge(newsk, filter);
1178
1179                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1180                         /* It is still raw copy of parent, so invalidate
1181                          * destructor and make plain sk_free() */
1182                         newsk->sk_destruct = NULL;
1183                         sk_free(newsk);
1184                         newsk = NULL;
1185                         goto out;
1186                 }
1187
1188                 newsk->sk_err      = 0;
1189                 newsk->sk_priority = 0;
1190                 /*
1191                  * Before updating sk_refcnt, we must commit prior changes to memory
1192                  * (Documentation/RCU/rculist_nulls.txt for details)
1193                  */
1194                 smp_wmb();
1195                 atomic_set(&newsk->sk_refcnt, 2);
1196
1197                 /*
1198                  * Increment the counter in the same struct proto as the master
1199                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1200                  * is the same as sk->sk_prot->socks, as this field was copied
1201                  * with memcpy).
1202                  *
1203                  * This _changes_ the previous behaviour, where
1204                  * tcp_create_openreq_child always was incrementing the
1205                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1206                  * to be taken into account in all callers. -acme
1207                  */
1208                 sk_refcnt_debug_inc(newsk);
1209                 sk_set_socket(newsk, NULL);
1210                 newsk->sk_sleep  = NULL;
1211
1212                 if (newsk->sk_prot->sockets_allocated)
1213                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1214
1215                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1216                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1217                         net_enable_timestamp();
1218         }
1219 out:
1220         return newsk;
1221 }
1222 EXPORT_SYMBOL_GPL(sk_clone);
1223
1224 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1225 {
1226         __sk_dst_set(sk, dst);
1227         sk->sk_route_caps = dst->dev->features;
1228         if (sk->sk_route_caps & NETIF_F_GSO)
1229                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1230         if (sk_can_gso(sk)) {
1231                 if (dst->header_len) {
1232                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1233                 } else {
1234                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1235                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1236                 }
1237         }
1238 }
1239 EXPORT_SYMBOL_GPL(sk_setup_caps);
1240
1241 void __init sk_init(void)
1242 {
1243         if (totalram_pages <= 4096) {
1244                 sysctl_wmem_max = 32767;
1245                 sysctl_rmem_max = 32767;
1246                 sysctl_wmem_default = 32767;
1247                 sysctl_rmem_default = 32767;
1248         } else if (totalram_pages >= 131072) {
1249                 sysctl_wmem_max = 131071;
1250                 sysctl_rmem_max = 131071;
1251         }
1252 }
1253
1254 /*
1255  *      Simple resource managers for sockets.
1256  */
1257
1258
1259 /*
1260  * Write buffer destructor automatically called from kfree_skb.
1261  */
1262 void sock_wfree(struct sk_buff *skb)
1263 {
1264         struct sock *sk = skb->sk;
1265         unsigned int len = skb->truesize;
1266
1267         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1268                 /*
1269                  * Keep a reference on sk_wmem_alloc, this will be released
1270                  * after sk_write_space() call
1271                  */
1272                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1273                 sk->sk_write_space(sk);
1274                 len = 1;
1275         }
1276         /*
1277          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1278          * could not do because of in-flight packets
1279          */
1280         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1281                 __sk_free(sk);
1282 }
1283 EXPORT_SYMBOL(sock_wfree);
1284
1285 /*
1286  * Read buffer destructor automatically called from kfree_skb.
1287  */
1288 void sock_rfree(struct sk_buff *skb)
1289 {
1290         struct sock *sk = skb->sk;
1291
1292         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1293         sk_mem_uncharge(skb->sk, skb->truesize);
1294 }
1295 EXPORT_SYMBOL(sock_rfree);
1296
1297
1298 int sock_i_uid(struct sock *sk)
1299 {
1300         int uid;
1301
1302         read_lock(&sk->sk_callback_lock);
1303         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1304         read_unlock(&sk->sk_callback_lock);
1305         return uid;
1306 }
1307 EXPORT_SYMBOL(sock_i_uid);
1308
1309 unsigned long sock_i_ino(struct sock *sk)
1310 {
1311         unsigned long ino;
1312
1313         read_lock(&sk->sk_callback_lock);
1314         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1315         read_unlock(&sk->sk_callback_lock);
1316         return ino;
1317 }
1318 EXPORT_SYMBOL(sock_i_ino);
1319
1320 /*
1321  * Allocate a skb from the socket's send buffer.
1322  */
1323 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1324                              gfp_t priority)
1325 {
1326         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1327                 struct sk_buff *skb = alloc_skb(size, priority);
1328                 if (skb) {
1329                         skb_set_owner_w(skb, sk);
1330                         return skb;
1331                 }
1332         }
1333         return NULL;
1334 }
1335 EXPORT_SYMBOL(sock_wmalloc);
1336
1337 /*
1338  * Allocate a skb from the socket's receive buffer.
1339  */
1340 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1341                              gfp_t priority)
1342 {
1343         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1344                 struct sk_buff *skb = alloc_skb(size, priority);
1345                 if (skb) {
1346                         skb_set_owner_r(skb, sk);
1347                         return skb;
1348                 }
1349         }
1350         return NULL;
1351 }
1352
1353 /*
1354  * Allocate a memory block from the socket's option memory buffer.
1355  */
1356 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1357 {
1358         if ((unsigned)size <= sysctl_optmem_max &&
1359             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1360                 void *mem;
1361                 /* First do the add, to avoid the race if kmalloc
1362                  * might sleep.
1363                  */
1364                 atomic_add(size, &sk->sk_omem_alloc);
1365                 mem = kmalloc(size, priority);
1366                 if (mem)
1367                         return mem;
1368                 atomic_sub(size, &sk->sk_omem_alloc);
1369         }
1370         return NULL;
1371 }
1372 EXPORT_SYMBOL(sock_kmalloc);
1373
1374 /*
1375  * Free an option memory block.
1376  */
1377 void sock_kfree_s(struct sock *sk, void *mem, int size)
1378 {
1379         kfree(mem);
1380         atomic_sub(size, &sk->sk_omem_alloc);
1381 }
1382 EXPORT_SYMBOL(sock_kfree_s);
1383
1384 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1385    I think, these locks should be removed for datagram sockets.
1386  */
1387 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1388 {
1389         DEFINE_WAIT(wait);
1390
1391         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1392         for (;;) {
1393                 if (!timeo)
1394                         break;
1395                 if (signal_pending(current))
1396                         break;
1397                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1398                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1399                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1400                         break;
1401                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1402                         break;
1403                 if (sk->sk_err)
1404                         break;
1405                 timeo = schedule_timeout(timeo);
1406         }
1407         finish_wait(sk->sk_sleep, &wait);
1408         return timeo;
1409 }
1410
1411
1412 /*
1413  *      Generic send/receive buffer handlers
1414  */
1415
1416 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1417                                      unsigned long data_len, int noblock,
1418                                      int *errcode)
1419 {
1420         struct sk_buff *skb;
1421         gfp_t gfp_mask;
1422         long timeo;
1423         int err;
1424
1425         gfp_mask = sk->sk_allocation;
1426         if (gfp_mask & __GFP_WAIT)
1427                 gfp_mask |= __GFP_REPEAT;
1428
1429         timeo = sock_sndtimeo(sk, noblock);
1430         while (1) {
1431                 err = sock_error(sk);
1432                 if (err != 0)
1433                         goto failure;
1434
1435                 err = -EPIPE;
1436                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1437                         goto failure;
1438
1439                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1440                         skb = alloc_skb(header_len, gfp_mask);
1441                         if (skb) {
1442                                 int npages;
1443                                 int i;
1444
1445                                 /* No pages, we're done... */
1446                                 if (!data_len)
1447                                         break;
1448
1449                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1450                                 skb->truesize += data_len;
1451                                 skb_shinfo(skb)->nr_frags = npages;
1452                                 for (i = 0; i < npages; i++) {
1453                                         struct page *page;
1454                                         skb_frag_t *frag;
1455
1456                                         page = alloc_pages(sk->sk_allocation, 0);
1457                                         if (!page) {
1458                                                 err = -ENOBUFS;
1459                                                 skb_shinfo(skb)->nr_frags = i;
1460                                                 kfree_skb(skb);
1461                                                 goto failure;
1462                                         }
1463
1464                                         frag = &skb_shinfo(skb)->frags[i];
1465                                         frag->page = page;
1466                                         frag->page_offset = 0;
1467                                         frag->size = (data_len >= PAGE_SIZE ?
1468                                                       PAGE_SIZE :
1469                                                       data_len);
1470                                         data_len -= PAGE_SIZE;
1471                                 }
1472
1473                                 /* Full success... */
1474                                 break;
1475                         }
1476                         err = -ENOBUFS;
1477                         goto failure;
1478                 }
1479                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1480                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1481                 err = -EAGAIN;
1482                 if (!timeo)
1483                         goto failure;
1484                 if (signal_pending(current))
1485                         goto interrupted;
1486                 timeo = sock_wait_for_wmem(sk, timeo);
1487         }
1488
1489         skb_set_owner_w(skb, sk);
1490         return skb;
1491
1492 interrupted:
1493         err = sock_intr_errno(timeo);
1494 failure:
1495         *errcode = err;
1496         return NULL;
1497 }
1498 EXPORT_SYMBOL(sock_alloc_send_pskb);
1499
1500 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1501                                     int noblock, int *errcode)
1502 {
1503         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1504 }
1505 EXPORT_SYMBOL(sock_alloc_send_skb);
1506
1507 static void __lock_sock(struct sock *sk)
1508 {
1509         DEFINE_WAIT(wait);
1510
1511         for (;;) {
1512                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1513                                         TASK_UNINTERRUPTIBLE);
1514                 spin_unlock_bh(&sk->sk_lock.slock);
1515                 schedule();
1516                 spin_lock_bh(&sk->sk_lock.slock);
1517                 if (!sock_owned_by_user(sk))
1518                         break;
1519         }
1520         finish_wait(&sk->sk_lock.wq, &wait);
1521 }
1522
1523 static void __release_sock(struct sock *sk)
1524 {
1525         struct sk_buff *skb = sk->sk_backlog.head;
1526
1527         do {
1528                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1529                 bh_unlock_sock(sk);
1530
1531                 do {
1532                         struct sk_buff *next = skb->next;
1533
1534                         skb->next = NULL;
1535                         sk_backlog_rcv(sk, skb);
1536
1537                         /*
1538                          * We are in process context here with softirqs
1539                          * disabled, use cond_resched_softirq() to preempt.
1540                          * This is safe to do because we've taken the backlog
1541                          * queue private:
1542                          */
1543                         cond_resched_softirq();
1544
1545                         skb = next;
1546                 } while (skb != NULL);
1547
1548                 bh_lock_sock(sk);
1549         } while ((skb = sk->sk_backlog.head) != NULL);
1550
1551         /*
1552          * Doing the zeroing here guarantee we can not loop forever
1553          * while a wild producer attempts to flood us.
1554          */
1555         sk->sk_backlog.len = 0;
1556 }
1557
1558 /**
1559  * sk_wait_data - wait for data to arrive at sk_receive_queue
1560  * @sk:    sock to wait on
1561  * @timeo: for how long
1562  *
1563  * Now socket state including sk->sk_err is changed only under lock,
1564  * hence we may omit checks after joining wait queue.
1565  * We check receive queue before schedule() only as optimization;
1566  * it is very likely that release_sock() added new data.
1567  */
1568 int sk_wait_data(struct sock *sk, long *timeo)
1569 {
1570         int rc;
1571         DEFINE_WAIT(wait);
1572
1573         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1574         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1575         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1576         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1577         finish_wait(sk->sk_sleep, &wait);
1578         return rc;
1579 }
1580 EXPORT_SYMBOL(sk_wait_data);
1581
1582 /**
1583  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1584  *      @sk: socket
1585  *      @size: memory size to allocate
1586  *      @kind: allocation type
1587  *
1588  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1589  *      rmem allocation. This function assumes that protocols which have
1590  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1591  */
1592 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1593 {
1594         struct proto *prot = sk->sk_prot;
1595         int amt = sk_mem_pages(size);
1596         int allocated;
1597
1598         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1599         allocated = atomic_add_return(amt, prot->memory_allocated);
1600
1601         /* Under limit. */
1602         if (allocated <= prot->sysctl_mem[0]) {
1603                 if (prot->memory_pressure && *prot->memory_pressure)
1604                         *prot->memory_pressure = 0;
1605                 return 1;
1606         }
1607
1608         /* Under pressure. */
1609         if (allocated > prot->sysctl_mem[1])
1610                 if (prot->enter_memory_pressure)
1611                         prot->enter_memory_pressure(sk);
1612
1613         /* Over hard limit. */
1614         if (allocated > prot->sysctl_mem[2])
1615                 goto suppress_allocation;
1616
1617         /* guarantee minimum buffer size under pressure */
1618         if (kind == SK_MEM_RECV) {
1619                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1620                         return 1;
1621         } else { /* SK_MEM_SEND */
1622                 if (sk->sk_type == SOCK_STREAM) {
1623                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1624                                 return 1;
1625                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1626                            prot->sysctl_wmem[0])
1627                                 return 1;
1628         }
1629
1630         if (prot->memory_pressure) {
1631                 int alloc;
1632
1633                 if (!*prot->memory_pressure)
1634                         return 1;
1635                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1636                 if (prot->sysctl_mem[2] > alloc *
1637                     sk_mem_pages(sk->sk_wmem_queued +
1638                                  atomic_read(&sk->sk_rmem_alloc) +
1639                                  sk->sk_forward_alloc))
1640                         return 1;
1641         }
1642
1643 suppress_allocation:
1644
1645         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1646                 sk_stream_moderate_sndbuf(sk);
1647
1648                 /* Fail only if socket is _under_ its sndbuf.
1649                  * In this case we cannot block, so that we have to fail.
1650                  */
1651                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1652                         return 1;
1653         }
1654
1655         /* Alas. Undo changes. */
1656         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1657         atomic_sub(amt, prot->memory_allocated);
1658         return 0;
1659 }
1660 EXPORT_SYMBOL(__sk_mem_schedule);
1661
1662 /**
1663  *      __sk_reclaim - reclaim memory_allocated
1664  *      @sk: socket
1665  */
1666 void __sk_mem_reclaim(struct sock *sk)
1667 {
1668         struct proto *prot = sk->sk_prot;
1669
1670         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1671                    prot->memory_allocated);
1672         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1673
1674         if (prot->memory_pressure && *prot->memory_pressure &&
1675             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1676                 *prot->memory_pressure = 0;
1677 }
1678 EXPORT_SYMBOL(__sk_mem_reclaim);
1679
1680
1681 /*
1682  * Set of default routines for initialising struct proto_ops when
1683  * the protocol does not support a particular function. In certain
1684  * cases where it makes no sense for a protocol to have a "do nothing"
1685  * function, some default processing is provided.
1686  */
1687
1688 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1689 {
1690         return -EOPNOTSUPP;
1691 }
1692 EXPORT_SYMBOL(sock_no_bind);
1693
1694 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1695                     int len, int flags)
1696 {
1697         return -EOPNOTSUPP;
1698 }
1699 EXPORT_SYMBOL(sock_no_connect);
1700
1701 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1702 {
1703         return -EOPNOTSUPP;
1704 }
1705 EXPORT_SYMBOL(sock_no_socketpair);
1706
1707 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1708 {
1709         return -EOPNOTSUPP;
1710 }
1711 EXPORT_SYMBOL(sock_no_accept);
1712
1713 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1714                     int *len, int peer)
1715 {
1716         return -EOPNOTSUPP;
1717 }
1718 EXPORT_SYMBOL(sock_no_getname);
1719
1720 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1721 {
1722         return 0;
1723 }
1724 EXPORT_SYMBOL(sock_no_poll);
1725
1726 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1727 {
1728         return -EOPNOTSUPP;
1729 }
1730 EXPORT_SYMBOL(sock_no_ioctl);
1731
1732 int sock_no_listen(struct socket *sock, int backlog)
1733 {
1734         return -EOPNOTSUPP;
1735 }
1736 EXPORT_SYMBOL(sock_no_listen);
1737
1738 int sock_no_shutdown(struct socket *sock, int how)
1739 {
1740         return -EOPNOTSUPP;
1741 }
1742 EXPORT_SYMBOL(sock_no_shutdown);
1743
1744 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1745                     char __user *optval, unsigned int optlen)
1746 {
1747         return -EOPNOTSUPP;
1748 }
1749 EXPORT_SYMBOL(sock_no_setsockopt);
1750
1751 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1752                     char __user *optval, int __user *optlen)
1753 {
1754         return -EOPNOTSUPP;
1755 }
1756 EXPORT_SYMBOL(sock_no_getsockopt);
1757
1758 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1759                     size_t len)
1760 {
1761         return -EOPNOTSUPP;
1762 }
1763 EXPORT_SYMBOL(sock_no_sendmsg);
1764
1765 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1766                     size_t len, int flags)
1767 {
1768         return -EOPNOTSUPP;
1769 }
1770 EXPORT_SYMBOL(sock_no_recvmsg);
1771
1772 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1773 {
1774         /* Mirror missing mmap method error code */
1775         return -ENODEV;
1776 }
1777 EXPORT_SYMBOL(sock_no_mmap);
1778
1779 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1780 {
1781         ssize_t res;
1782         struct msghdr msg = {.msg_flags = flags};
1783         struct kvec iov;
1784         char *kaddr = kmap(page);
1785         iov.iov_base = kaddr + offset;
1786         iov.iov_len = size;
1787         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1788         kunmap(page);
1789         return res;
1790 }
1791 EXPORT_SYMBOL(sock_no_sendpage);
1792
1793 /*
1794  *      Default Socket Callbacks
1795  */
1796
1797 static void sock_def_wakeup(struct sock *sk)
1798 {
1799         read_lock(&sk->sk_callback_lock);
1800         if (sk_has_sleeper(sk))
1801                 wake_up_interruptible_all(sk->sk_sleep);
1802         read_unlock(&sk->sk_callback_lock);
1803 }
1804
1805 static void sock_def_error_report(struct sock *sk)
1806 {
1807         read_lock(&sk->sk_callback_lock);
1808         if (sk_has_sleeper(sk))
1809                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1810         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1811         read_unlock(&sk->sk_callback_lock);
1812 }
1813
1814 static void sock_def_readable(struct sock *sk, int len)
1815 {
1816         read_lock(&sk->sk_callback_lock);
1817         if (sk_has_sleeper(sk))
1818                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1819                                                 POLLRDNORM | POLLRDBAND);
1820         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1821         read_unlock(&sk->sk_callback_lock);
1822 }
1823
1824 static void sock_def_write_space(struct sock *sk)
1825 {
1826         read_lock(&sk->sk_callback_lock);
1827
1828         /* Do not wake up a writer until he can make "significant"
1829          * progress.  --DaveM
1830          */
1831         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1832                 if (sk_has_sleeper(sk))
1833                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1834                                                 POLLWRNORM | POLLWRBAND);
1835
1836                 /* Should agree with poll, otherwise some programs break */
1837                 if (sock_writeable(sk))
1838                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1839         }
1840
1841         read_unlock(&sk->sk_callback_lock);
1842 }
1843
1844 static void sock_def_destruct(struct sock *sk)
1845 {
1846         kfree(sk->sk_protinfo);
1847 }
1848
1849 void sk_send_sigurg(struct sock *sk)
1850 {
1851         if (sk->sk_socket && sk->sk_socket->file)
1852                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1853                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1854 }
1855 EXPORT_SYMBOL(sk_send_sigurg);
1856
1857 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1858                     unsigned long expires)
1859 {
1860         if (!mod_timer(timer, expires))
1861                 sock_hold(sk);
1862 }
1863 EXPORT_SYMBOL(sk_reset_timer);
1864
1865 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1866 {
1867         if (timer_pending(timer) && del_timer(timer))
1868                 __sock_put(sk);
1869 }
1870 EXPORT_SYMBOL(sk_stop_timer);
1871
1872 void sock_init_data(struct socket *sock, struct sock *sk)
1873 {
1874         skb_queue_head_init(&sk->sk_receive_queue);
1875         skb_queue_head_init(&sk->sk_write_queue);
1876         skb_queue_head_init(&sk->sk_error_queue);
1877 #ifdef CONFIG_NET_DMA
1878         skb_queue_head_init(&sk->sk_async_wait_queue);
1879 #endif
1880
1881         sk->sk_send_head        =       NULL;
1882
1883         init_timer(&sk->sk_timer);
1884
1885         sk->sk_allocation       =       GFP_KERNEL;
1886         sk->sk_rcvbuf           =       sysctl_rmem_default;
1887         sk->sk_sndbuf           =       sysctl_wmem_default;
1888         sk->sk_backlog.limit    =       sk->sk_rcvbuf << 1;
1889         sk->sk_state            =       TCP_CLOSE;
1890         sk_set_socket(sk, sock);
1891
1892         sock_set_flag(sk, SOCK_ZAPPED);
1893
1894         if (sock) {
1895                 sk->sk_type     =       sock->type;
1896                 sk->sk_sleep    =       &sock->wait;
1897                 sock->sk        =       sk;
1898         } else
1899                 sk->sk_sleep    =       NULL;
1900
1901         rwlock_init(&sk->sk_dst_lock);
1902         rwlock_init(&sk->sk_callback_lock);
1903         lockdep_set_class_and_name(&sk->sk_callback_lock,
1904                         af_callback_keys + sk->sk_family,
1905                         af_family_clock_key_strings[sk->sk_family]);
1906
1907         sk->sk_state_change     =       sock_def_wakeup;
1908         sk->sk_data_ready       =       sock_def_readable;
1909         sk->sk_write_space      =       sock_def_write_space;
1910         sk->sk_error_report     =       sock_def_error_report;
1911         sk->sk_destruct         =       sock_def_destruct;
1912
1913         sk->sk_sndmsg_page      =       NULL;
1914         sk->sk_sndmsg_off       =       0;
1915
1916         sk->sk_peercred.pid     =       0;
1917         sk->sk_peercred.uid     =       -1;
1918         sk->sk_peercred.gid     =       -1;
1919         sk->sk_write_pending    =       0;
1920         sk->sk_rcvlowat         =       1;
1921         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1922         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1923
1924         sk->sk_stamp = ktime_set(-1L, 0);
1925
1926         /*
1927          * Before updating sk_refcnt, we must commit prior changes to memory
1928          * (Documentation/RCU/rculist_nulls.txt for details)
1929          */
1930         smp_wmb();
1931         atomic_set(&sk->sk_refcnt, 1);
1932         atomic_set(&sk->sk_drops, 0);
1933 }
1934 EXPORT_SYMBOL(sock_init_data);
1935
1936 void lock_sock_nested(struct sock *sk, int subclass)
1937 {
1938         might_sleep();
1939         spin_lock_bh(&sk->sk_lock.slock);
1940         if (sk->sk_lock.owned)
1941                 __lock_sock(sk);
1942         sk->sk_lock.owned = 1;
1943         spin_unlock(&sk->sk_lock.slock);
1944         /*
1945          * The sk_lock has mutex_lock() semantics here:
1946          */
1947         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1948         local_bh_enable();
1949 }
1950 EXPORT_SYMBOL(lock_sock_nested);
1951
1952 void release_sock(struct sock *sk)
1953 {
1954         /*
1955          * The sk_lock has mutex_unlock() semantics:
1956          */
1957         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1958
1959         spin_lock_bh(&sk->sk_lock.slock);
1960         if (sk->sk_backlog.tail)
1961                 __release_sock(sk);
1962         sk->sk_lock.owned = 0;
1963         if (waitqueue_active(&sk->sk_lock.wq))
1964                 wake_up(&sk->sk_lock.wq);
1965         spin_unlock_bh(&sk->sk_lock.slock);
1966 }
1967 EXPORT_SYMBOL(release_sock);
1968
1969 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1970 {
1971         struct timeval tv;
1972         if (!sock_flag(sk, SOCK_TIMESTAMP))
1973                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1974         tv = ktime_to_timeval(sk->sk_stamp);
1975         if (tv.tv_sec == -1)
1976                 return -ENOENT;
1977         if (tv.tv_sec == 0) {
1978                 sk->sk_stamp = ktime_get_real();
1979                 tv = ktime_to_timeval(sk->sk_stamp);
1980         }
1981         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1982 }
1983 EXPORT_SYMBOL(sock_get_timestamp);
1984
1985 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1986 {
1987         struct timespec ts;
1988         if (!sock_flag(sk, SOCK_TIMESTAMP))
1989                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1990         ts = ktime_to_timespec(sk->sk_stamp);
1991         if (ts.tv_sec == -1)
1992                 return -ENOENT;
1993         if (ts.tv_sec == 0) {
1994                 sk->sk_stamp = ktime_get_real();
1995                 ts = ktime_to_timespec(sk->sk_stamp);
1996         }
1997         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1998 }
1999 EXPORT_SYMBOL(sock_get_timestampns);
2000
2001 void sock_enable_timestamp(struct sock *sk, int flag)
2002 {
2003         if (!sock_flag(sk, flag)) {
2004                 sock_set_flag(sk, flag);
2005                 /*
2006                  * we just set one of the two flags which require net
2007                  * time stamping, but time stamping might have been on
2008                  * already because of the other one
2009                  */
2010                 if (!sock_flag(sk,
2011                                 flag == SOCK_TIMESTAMP ?
2012                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2013                                 SOCK_TIMESTAMP))
2014                         net_enable_timestamp();
2015         }
2016 }
2017
2018 /*
2019  *      Get a socket option on an socket.
2020  *
2021  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2022  *      asynchronous errors should be reported by getsockopt. We assume
2023  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2024  */
2025 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2026                            char __user *optval, int __user *optlen)
2027 {
2028         struct sock *sk = sock->sk;
2029
2030         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2031 }
2032 EXPORT_SYMBOL(sock_common_getsockopt);
2033
2034 #ifdef CONFIG_COMPAT
2035 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2036                                   char __user *optval, int __user *optlen)
2037 {
2038         struct sock *sk = sock->sk;
2039
2040         if (sk->sk_prot->compat_getsockopt != NULL)
2041                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2042                                                       optval, optlen);
2043         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2044 }
2045 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2046 #endif
2047
2048 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2049                         struct msghdr *msg, size_t size, int flags)
2050 {
2051         struct sock *sk = sock->sk;
2052         int addr_len = 0;
2053         int err;
2054
2055         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2056                                    flags & ~MSG_DONTWAIT, &addr_len);
2057         if (err >= 0)
2058                 msg->msg_namelen = addr_len;
2059         return err;
2060 }
2061 EXPORT_SYMBOL(sock_common_recvmsg);
2062
2063 /*
2064  *      Set socket options on an inet socket.
2065  */
2066 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2067                            char __user *optval, unsigned int optlen)
2068 {
2069         struct sock *sk = sock->sk;
2070
2071         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2072 }
2073 EXPORT_SYMBOL(sock_common_setsockopt);
2074
2075 #ifdef CONFIG_COMPAT
2076 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2077                                   char __user *optval, unsigned int optlen)
2078 {
2079         struct sock *sk = sock->sk;
2080
2081         if (sk->sk_prot->compat_setsockopt != NULL)
2082                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2083                                                       optval, optlen);
2084         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2085 }
2086 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2087 #endif
2088
2089 void sk_common_release(struct sock *sk)
2090 {
2091         if (sk->sk_prot->destroy)
2092                 sk->sk_prot->destroy(sk);
2093
2094         /*
2095          * Observation: when sock_common_release is called, processes have
2096          * no access to socket. But net still has.
2097          * Step one, detach it from networking:
2098          *
2099          * A. Remove from hash tables.
2100          */
2101
2102         sk->sk_prot->unhash(sk);
2103
2104         /*
2105          * In this point socket cannot receive new packets, but it is possible
2106          * that some packets are in flight because some CPU runs receiver and
2107          * did hash table lookup before we unhashed socket. They will achieve
2108          * receive queue and will be purged by socket destructor.
2109          *
2110          * Also we still have packets pending on receive queue and probably,
2111          * our own packets waiting in device queues. sock_destroy will drain
2112          * receive queue, but transmitted packets will delay socket destruction
2113          * until the last reference will be released.
2114          */
2115
2116         sock_orphan(sk);
2117
2118         xfrm_sk_free_policy(sk);
2119
2120         sk_refcnt_debug_release(sk);
2121         sock_put(sk);
2122 }
2123 EXPORT_SYMBOL(sk_common_release);
2124
2125 static DEFINE_RWLOCK(proto_list_lock);
2126 static LIST_HEAD(proto_list);
2127
2128 #ifdef CONFIG_PROC_FS
2129 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2130 struct prot_inuse {
2131         int val[PROTO_INUSE_NR];
2132 };
2133
2134 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2135
2136 #ifdef CONFIG_NET_NS
2137 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2138 {
2139         int cpu = smp_processor_id();
2140         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2141 }
2142 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2143
2144 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2145 {
2146         int cpu, idx = prot->inuse_idx;
2147         int res = 0;
2148
2149         for_each_possible_cpu(cpu)
2150                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2151
2152         return res >= 0 ? res : 0;
2153 }
2154 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2155
2156 static int __net_init sock_inuse_init_net(struct net *net)
2157 {
2158         net->core.inuse = alloc_percpu(struct prot_inuse);
2159         return net->core.inuse ? 0 : -ENOMEM;
2160 }
2161
2162 static void __net_exit sock_inuse_exit_net(struct net *net)
2163 {
2164         free_percpu(net->core.inuse);
2165 }
2166
2167 static struct pernet_operations net_inuse_ops = {
2168         .init = sock_inuse_init_net,
2169         .exit = sock_inuse_exit_net,
2170 };
2171
2172 static __init int net_inuse_init(void)
2173 {
2174         if (register_pernet_subsys(&net_inuse_ops))
2175                 panic("Cannot initialize net inuse counters");
2176
2177         return 0;
2178 }
2179
2180 core_initcall(net_inuse_init);
2181 #else
2182 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2183
2184 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2185 {
2186         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2187 }
2188 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2189
2190 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2191 {
2192         int cpu, idx = prot->inuse_idx;
2193         int res = 0;
2194
2195         for_each_possible_cpu(cpu)
2196                 res += per_cpu(prot_inuse, cpu).val[idx];
2197
2198         return res >= 0 ? res : 0;
2199 }
2200 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2201 #endif
2202
2203 static void assign_proto_idx(struct proto *prot)
2204 {
2205         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2206
2207         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2208                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2209                 return;
2210         }
2211
2212         set_bit(prot->inuse_idx, proto_inuse_idx);
2213 }
2214
2215 static void release_proto_idx(struct proto *prot)
2216 {
2217         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2218                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2219 }
2220 #else
2221 static inline void assign_proto_idx(struct proto *prot)
2222 {
2223 }
2224
2225 static inline void release_proto_idx(struct proto *prot)
2226 {
2227 }
2228 #endif
2229
2230 int proto_register(struct proto *prot, int alloc_slab)
2231 {
2232         if (alloc_slab) {
2233                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2234                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2235                                         NULL);
2236
2237                 if (prot->slab == NULL) {
2238                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2239                                prot->name);
2240                         goto out;
2241                 }
2242
2243                 if (prot->rsk_prot != NULL) {
2244                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2245                         if (prot->rsk_prot->slab_name == NULL)
2246                                 goto out_free_sock_slab;
2247
2248                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2249                                                                  prot->rsk_prot->obj_size, 0,
2250                                                                  SLAB_HWCACHE_ALIGN, NULL);
2251
2252                         if (prot->rsk_prot->slab == NULL) {
2253                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2254                                        prot->name);
2255                                 goto out_free_request_sock_slab_name;
2256                         }
2257                 }
2258
2259                 if (prot->twsk_prot != NULL) {
2260                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2261
2262                         if (prot->twsk_prot->twsk_slab_name == NULL)
2263                                 goto out_free_request_sock_slab;
2264
2265                         prot->twsk_prot->twsk_slab =
2266                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2267                                                   prot->twsk_prot->twsk_obj_size,
2268                                                   0,
2269                                                   SLAB_HWCACHE_ALIGN |
2270                                                         prot->slab_flags,
2271                                                   NULL);
2272                         if (prot->twsk_prot->twsk_slab == NULL)
2273                                 goto out_free_timewait_sock_slab_name;
2274                 }
2275         }
2276
2277         write_lock(&proto_list_lock);
2278         list_add(&prot->node, &proto_list);
2279         assign_proto_idx(prot);
2280         write_unlock(&proto_list_lock);
2281         return 0;
2282
2283 out_free_timewait_sock_slab_name:
2284         kfree(prot->twsk_prot->twsk_slab_name);
2285 out_free_request_sock_slab:
2286         if (prot->rsk_prot && prot->rsk_prot->slab) {
2287                 kmem_cache_destroy(prot->rsk_prot->slab);
2288                 prot->rsk_prot->slab = NULL;
2289         }
2290 out_free_request_sock_slab_name:
2291         kfree(prot->rsk_prot->slab_name);
2292 out_free_sock_slab:
2293         kmem_cache_destroy(prot->slab);
2294         prot->slab = NULL;
2295 out:
2296         return -ENOBUFS;
2297 }
2298 EXPORT_SYMBOL(proto_register);
2299
2300 void proto_unregister(struct proto *prot)
2301 {
2302         write_lock(&proto_list_lock);
2303         release_proto_idx(prot);
2304         list_del(&prot->node);
2305         write_unlock(&proto_list_lock);
2306
2307         if (prot->slab != NULL) {
2308                 kmem_cache_destroy(prot->slab);
2309                 prot->slab = NULL;
2310         }
2311
2312         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2313                 kmem_cache_destroy(prot->rsk_prot->slab);
2314                 kfree(prot->rsk_prot->slab_name);
2315                 prot->rsk_prot->slab = NULL;
2316         }
2317
2318         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2319                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2320                 kfree(prot->twsk_prot->twsk_slab_name);
2321                 prot->twsk_prot->twsk_slab = NULL;
2322         }
2323 }
2324 EXPORT_SYMBOL(proto_unregister);
2325
2326 #ifdef CONFIG_PROC_FS
2327 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2328         __acquires(proto_list_lock)
2329 {
2330         read_lock(&proto_list_lock);
2331         return seq_list_start_head(&proto_list, *pos);
2332 }
2333
2334 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2335 {
2336         return seq_list_next(v, &proto_list, pos);
2337 }
2338
2339 static void proto_seq_stop(struct seq_file *seq, void *v)
2340         __releases(proto_list_lock)
2341 {
2342         read_unlock(&proto_list_lock);
2343 }
2344
2345 static char proto_method_implemented(const void *method)
2346 {
2347         return method == NULL ? 'n' : 'y';
2348 }
2349
2350 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2351 {
2352         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2353                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2354                    proto->name,
2355                    proto->obj_size,
2356                    sock_prot_inuse_get(seq_file_net(seq), proto),
2357                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2358                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2359                    proto->max_header,
2360                    proto->slab == NULL ? "no" : "yes",
2361                    module_name(proto->owner),
2362                    proto_method_implemented(proto->close),
2363                    proto_method_implemented(proto->connect),
2364                    proto_method_implemented(proto->disconnect),
2365                    proto_method_implemented(proto->accept),
2366                    proto_method_implemented(proto->ioctl),
2367                    proto_method_implemented(proto->init),
2368                    proto_method_implemented(proto->destroy),
2369                    proto_method_implemented(proto->shutdown),
2370                    proto_method_implemented(proto->setsockopt),
2371                    proto_method_implemented(proto->getsockopt),
2372                    proto_method_implemented(proto->sendmsg),
2373                    proto_method_implemented(proto->recvmsg),
2374                    proto_method_implemented(proto->sendpage),
2375                    proto_method_implemented(proto->bind),
2376                    proto_method_implemented(proto->backlog_rcv),
2377                    proto_method_implemented(proto->hash),
2378                    proto_method_implemented(proto->unhash),
2379                    proto_method_implemented(proto->get_port),
2380                    proto_method_implemented(proto->enter_memory_pressure));
2381 }
2382
2383 static int proto_seq_show(struct seq_file *seq, void *v)
2384 {
2385         if (v == &proto_list)
2386                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2387                            "protocol",
2388                            "size",
2389                            "sockets",
2390                            "memory",
2391                            "press",
2392                            "maxhdr",
2393                            "slab",
2394                            "module",
2395                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2396         else
2397                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2398         return 0;
2399 }
2400
2401 static const struct seq_operations proto_seq_ops = {
2402         .start  = proto_seq_start,
2403         .next   = proto_seq_next,
2404         .stop   = proto_seq_stop,
2405         .show   = proto_seq_show,
2406 };
2407
2408 static int proto_seq_open(struct inode *inode, struct file *file)
2409 {
2410         return seq_open_net(inode, file, &proto_seq_ops,
2411                             sizeof(struct seq_net_private));
2412 }
2413
2414 static const struct file_operations proto_seq_fops = {
2415         .owner          = THIS_MODULE,
2416         .open           = proto_seq_open,
2417         .read           = seq_read,
2418         .llseek         = seq_lseek,
2419         .release        = seq_release_net,
2420 };
2421
2422 static __net_init int proto_init_net(struct net *net)
2423 {
2424         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2425                 return -ENOMEM;
2426
2427         return 0;
2428 }
2429
2430 static __net_exit void proto_exit_net(struct net *net)
2431 {
2432         proc_net_remove(net, "protocols");
2433 }
2434
2435
2436 static __net_initdata struct pernet_operations proto_net_ops = {
2437         .init = proto_init_net,
2438         .exit = proto_exit_net,
2439 };
2440
2441 static int __init proto_init(void)
2442 {
2443         return register_pernet_subsys(&proto_net_ops);
2444 }
2445
2446 subsys_initcall(proto_init);
2447
2448 #endif /* PROC_FS */