net: Introduce sk_route_nocaps
[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 (sk_rcvqueues_full(sk, skb)) {
331                 atomic_inc(&sk->sk_drops);
332                 goto discard_and_relse;
333         }
334         if (nested)
335                 bh_lock_sock_nested(sk);
336         else
337                 bh_lock_sock(sk);
338         if (!sock_owned_by_user(sk)) {
339                 /*
340                  * trylock + unlock semantics:
341                  */
342                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
343
344                 rc = sk_backlog_rcv(sk, skb);
345
346                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
347         } else if (sk_add_backlog(sk, skb)) {
348                 bh_unlock_sock(sk);
349                 atomic_inc(&sk->sk_drops);
350                 goto discard_and_relse;
351         }
352
353         bh_unlock_sock(sk);
354 out:
355         sock_put(sk);
356         return rc;
357 discard_and_relse:
358         kfree_skb(skb);
359         goto out;
360 }
361 EXPORT_SYMBOL(sk_receive_skb);
362
363 void sk_reset_txq(struct sock *sk)
364 {
365         sk_tx_queue_clear(sk);
366 }
367 EXPORT_SYMBOL(sk_reset_txq);
368
369 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
370 {
371         struct dst_entry *dst = __sk_dst_get(sk);
372
373         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
374                 sk_tx_queue_clear(sk);
375                 rcu_assign_pointer(sk->sk_dst_cache, NULL);
376                 dst_release(dst);
377                 return NULL;
378         }
379
380         return dst;
381 }
382 EXPORT_SYMBOL(__sk_dst_check);
383
384 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
385 {
386         struct dst_entry *dst = sk_dst_get(sk);
387
388         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
389                 sk_dst_reset(sk);
390                 dst_release(dst);
391                 return NULL;
392         }
393
394         return dst;
395 }
396 EXPORT_SYMBOL(sk_dst_check);
397
398 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
399 {
400         int ret = -ENOPROTOOPT;
401 #ifdef CONFIG_NETDEVICES
402         struct net *net = sock_net(sk);
403         char devname[IFNAMSIZ];
404         int index;
405
406         /* Sorry... */
407         ret = -EPERM;
408         if (!capable(CAP_NET_RAW))
409                 goto out;
410
411         ret = -EINVAL;
412         if (optlen < 0)
413                 goto out;
414
415         /* Bind this socket to a particular device like "eth0",
416          * as specified in the passed interface name. If the
417          * name is "" or the option length is zero the socket
418          * is not bound.
419          */
420         if (optlen > IFNAMSIZ - 1)
421                 optlen = IFNAMSIZ - 1;
422         memset(devname, 0, sizeof(devname));
423
424         ret = -EFAULT;
425         if (copy_from_user(devname, optval, optlen))
426                 goto out;
427
428         index = 0;
429         if (devname[0] != '\0') {
430                 struct net_device *dev;
431
432                 rcu_read_lock();
433                 dev = dev_get_by_name_rcu(net, devname);
434                 if (dev)
435                         index = dev->ifindex;
436                 rcu_read_unlock();
437                 ret = -ENODEV;
438                 if (!dev)
439                         goto out;
440         }
441
442         lock_sock(sk);
443         sk->sk_bound_dev_if = index;
444         sk_dst_reset(sk);
445         release_sock(sk);
446
447         ret = 0;
448
449 out:
450 #endif
451
452         return ret;
453 }
454
455 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
456 {
457         if (valbool)
458                 sock_set_flag(sk, bit);
459         else
460                 sock_reset_flag(sk, bit);
461 }
462
463 /*
464  *      This is meant for all protocols to use and covers goings on
465  *      at the socket level. Everything here is generic.
466  */
467
468 int sock_setsockopt(struct socket *sock, int level, int optname,
469                     char __user *optval, unsigned int optlen)
470 {
471         struct sock *sk = sock->sk;
472         int val;
473         int valbool;
474         struct linger ling;
475         int ret = 0;
476
477         /*
478          *      Options without arguments
479          */
480
481         if (optname == SO_BINDTODEVICE)
482                 return sock_bindtodevice(sk, optval, optlen);
483
484         if (optlen < sizeof(int))
485                 return -EINVAL;
486
487         if (get_user(val, (int __user *)optval))
488                 return -EFAULT;
489
490         valbool = val ? 1 : 0;
491
492         lock_sock(sk);
493
494         switch (optname) {
495         case SO_DEBUG:
496                 if (val && !capable(CAP_NET_ADMIN))
497                         ret = -EACCES;
498                 else
499                         sock_valbool_flag(sk, SOCK_DBG, valbool);
500                 break;
501         case SO_REUSEADDR:
502                 sk->sk_reuse = valbool;
503                 break;
504         case SO_TYPE:
505         case SO_PROTOCOL:
506         case SO_DOMAIN:
507         case SO_ERROR:
508                 ret = -ENOPROTOOPT;
509                 break;
510         case SO_DONTROUTE:
511                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
512                 break;
513         case SO_BROADCAST:
514                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
515                 break;
516         case SO_SNDBUF:
517                 /* Don't error on this BSD doesn't and if you think
518                    about it this is right. Otherwise apps have to
519                    play 'guess the biggest size' games. RCVBUF/SNDBUF
520                    are treated in BSD as hints */
521
522                 if (val > sysctl_wmem_max)
523                         val = sysctl_wmem_max;
524 set_sndbuf:
525                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
526                 if ((val * 2) < SOCK_MIN_SNDBUF)
527                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
528                 else
529                         sk->sk_sndbuf = val * 2;
530
531                 /*
532                  *      Wake up sending tasks if we
533                  *      upped the value.
534                  */
535                 sk->sk_write_space(sk);
536                 break;
537
538         case SO_SNDBUFFORCE:
539                 if (!capable(CAP_NET_ADMIN)) {
540                         ret = -EPERM;
541                         break;
542                 }
543                 goto set_sndbuf;
544
545         case SO_RCVBUF:
546                 /* Don't error on this BSD doesn't and if you think
547                    about it this is right. Otherwise apps have to
548                    play 'guess the biggest size' games. RCVBUF/SNDBUF
549                    are treated in BSD as hints */
550
551                 if (val > sysctl_rmem_max)
552                         val = sysctl_rmem_max;
553 set_rcvbuf:
554                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
555                 /*
556                  * We double it on the way in to account for
557                  * "struct sk_buff" etc. overhead.   Applications
558                  * assume that the SO_RCVBUF setting they make will
559                  * allow that much actual data to be received on that
560                  * socket.
561                  *
562                  * Applications are unaware that "struct sk_buff" and
563                  * other overheads allocate from the receive buffer
564                  * during socket buffer allocation.
565                  *
566                  * And after considering the possible alternatives,
567                  * returning the value we actually used in getsockopt
568                  * is the most desirable behavior.
569                  */
570                 if ((val * 2) < SOCK_MIN_RCVBUF)
571                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
572                 else
573                         sk->sk_rcvbuf = val * 2;
574                 break;
575
576         case SO_RCVBUFFORCE:
577                 if (!capable(CAP_NET_ADMIN)) {
578                         ret = -EPERM;
579                         break;
580                 }
581                 goto set_rcvbuf;
582
583         case SO_KEEPALIVE:
584 #ifdef CONFIG_INET
585                 if (sk->sk_protocol == IPPROTO_TCP)
586                         tcp_set_keepalive(sk, valbool);
587 #endif
588                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
589                 break;
590
591         case SO_OOBINLINE:
592                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
593                 break;
594
595         case SO_NO_CHECK:
596                 sk->sk_no_check = valbool;
597                 break;
598
599         case SO_PRIORITY:
600                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
601                         sk->sk_priority = val;
602                 else
603                         ret = -EPERM;
604                 break;
605
606         case SO_LINGER:
607                 if (optlen < sizeof(ling)) {
608                         ret = -EINVAL;  /* 1003.1g */
609                         break;
610                 }
611                 if (copy_from_user(&ling, optval, sizeof(ling))) {
612                         ret = -EFAULT;
613                         break;
614                 }
615                 if (!ling.l_onoff)
616                         sock_reset_flag(sk, SOCK_LINGER);
617                 else {
618 #if (BITS_PER_LONG == 32)
619                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
620                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
621                         else
622 #endif
623                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
624                         sock_set_flag(sk, SOCK_LINGER);
625                 }
626                 break;
627
628         case SO_BSDCOMPAT:
629                 sock_warn_obsolete_bsdism("setsockopt");
630                 break;
631
632         case SO_PASSCRED:
633                 if (valbool)
634                         set_bit(SOCK_PASSCRED, &sock->flags);
635                 else
636                         clear_bit(SOCK_PASSCRED, &sock->flags);
637                 break;
638
639         case SO_TIMESTAMP:
640         case SO_TIMESTAMPNS:
641                 if (valbool)  {
642                         if (optname == SO_TIMESTAMP)
643                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
644                         else
645                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
646                         sock_set_flag(sk, SOCK_RCVTSTAMP);
647                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
648                 } else {
649                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
650                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
651                 }
652                 break;
653
654         case SO_TIMESTAMPING:
655                 if (val & ~SOF_TIMESTAMPING_MASK) {
656                         ret = -EINVAL;
657                         break;
658                 }
659                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
660                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
661                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
662                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
663                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
664                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
665                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
666                         sock_enable_timestamp(sk,
667                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
668                 else
669                         sock_disable_timestamp(sk,
670                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
671                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
672                                   val & SOF_TIMESTAMPING_SOFTWARE);
673                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
674                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
675                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
676                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
677                 break;
678
679         case SO_RCVLOWAT:
680                 if (val < 0)
681                         val = INT_MAX;
682                 sk->sk_rcvlowat = val ? : 1;
683                 break;
684
685         case SO_RCVTIMEO:
686                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
687                 break;
688
689         case SO_SNDTIMEO:
690                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
691                 break;
692
693         case SO_ATTACH_FILTER:
694                 ret = -EINVAL;
695                 if (optlen == sizeof(struct sock_fprog)) {
696                         struct sock_fprog fprog;
697
698                         ret = -EFAULT;
699                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
700                                 break;
701
702                         ret = sk_attach_filter(&fprog, sk);
703                 }
704                 break;
705
706         case SO_DETACH_FILTER:
707                 ret = sk_detach_filter(sk);
708                 break;
709
710         case SO_PASSSEC:
711                 if (valbool)
712                         set_bit(SOCK_PASSSEC, &sock->flags);
713                 else
714                         clear_bit(SOCK_PASSSEC, &sock->flags);
715                 break;
716         case SO_MARK:
717                 if (!capable(CAP_NET_ADMIN))
718                         ret = -EPERM;
719                 else
720                         sk->sk_mark = val;
721                 break;
722
723                 /* We implement the SO_SNDLOWAT etc to
724                    not be settable (1003.1g 5.3) */
725         case SO_RXQ_OVFL:
726                 if (valbool)
727                         sock_set_flag(sk, SOCK_RXQ_OVFL);
728                 else
729                         sock_reset_flag(sk, SOCK_RXQ_OVFL);
730                 break;
731         default:
732                 ret = -ENOPROTOOPT;
733                 break;
734         }
735         release_sock(sk);
736         return ret;
737 }
738 EXPORT_SYMBOL(sock_setsockopt);
739
740
741 int sock_getsockopt(struct socket *sock, int level, int optname,
742                     char __user *optval, int __user *optlen)
743 {
744         struct sock *sk = sock->sk;
745
746         union {
747                 int val;
748                 struct linger ling;
749                 struct timeval tm;
750         } v;
751
752         int lv = sizeof(int);
753         int len;
754
755         if (get_user(len, optlen))
756                 return -EFAULT;
757         if (len < 0)
758                 return -EINVAL;
759
760         memset(&v, 0, sizeof(v));
761
762         switch (optname) {
763         case SO_DEBUG:
764                 v.val = sock_flag(sk, SOCK_DBG);
765                 break;
766
767         case SO_DONTROUTE:
768                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
769                 break;
770
771         case SO_BROADCAST:
772                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
773                 break;
774
775         case SO_SNDBUF:
776                 v.val = sk->sk_sndbuf;
777                 break;
778
779         case SO_RCVBUF:
780                 v.val = sk->sk_rcvbuf;
781                 break;
782
783         case SO_REUSEADDR:
784                 v.val = sk->sk_reuse;
785                 break;
786
787         case SO_KEEPALIVE:
788                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
789                 break;
790
791         case SO_TYPE:
792                 v.val = sk->sk_type;
793                 break;
794
795         case SO_PROTOCOL:
796                 v.val = sk->sk_protocol;
797                 break;
798
799         case SO_DOMAIN:
800                 v.val = sk->sk_family;
801                 break;
802
803         case SO_ERROR:
804                 v.val = -sock_error(sk);
805                 if (v.val == 0)
806                         v.val = xchg(&sk->sk_err_soft, 0);
807                 break;
808
809         case SO_OOBINLINE:
810                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
811                 break;
812
813         case SO_NO_CHECK:
814                 v.val = sk->sk_no_check;
815                 break;
816
817         case SO_PRIORITY:
818                 v.val = sk->sk_priority;
819                 break;
820
821         case SO_LINGER:
822                 lv              = sizeof(v.ling);
823                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
824                 v.ling.l_linger = sk->sk_lingertime / HZ;
825                 break;
826
827         case SO_BSDCOMPAT:
828                 sock_warn_obsolete_bsdism("getsockopt");
829                 break;
830
831         case SO_TIMESTAMP:
832                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
833                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
834                 break;
835
836         case SO_TIMESTAMPNS:
837                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
838                 break;
839
840         case SO_TIMESTAMPING:
841                 v.val = 0;
842                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
843                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
844                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
845                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
846                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
847                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
848                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
849                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
850                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
851                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
852                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
853                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
854                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
855                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
856                 break;
857
858         case SO_RCVTIMEO:
859                 lv = sizeof(struct timeval);
860                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
861                         v.tm.tv_sec = 0;
862                         v.tm.tv_usec = 0;
863                 } else {
864                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
865                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
866                 }
867                 break;
868
869         case SO_SNDTIMEO:
870                 lv = sizeof(struct timeval);
871                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
872                         v.tm.tv_sec = 0;
873                         v.tm.tv_usec = 0;
874                 } else {
875                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
876                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
877                 }
878                 break;
879
880         case SO_RCVLOWAT:
881                 v.val = sk->sk_rcvlowat;
882                 break;
883
884         case SO_SNDLOWAT:
885                 v.val = 1;
886                 break;
887
888         case SO_PASSCRED:
889                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
890                 break;
891
892         case SO_PEERCRED:
893                 if (len > sizeof(sk->sk_peercred))
894                         len = sizeof(sk->sk_peercred);
895                 if (copy_to_user(optval, &sk->sk_peercred, len))
896                         return -EFAULT;
897                 goto lenout;
898
899         case SO_PEERNAME:
900         {
901                 char address[128];
902
903                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
904                         return -ENOTCONN;
905                 if (lv < len)
906                         return -EINVAL;
907                 if (copy_to_user(optval, address, len))
908                         return -EFAULT;
909                 goto lenout;
910         }
911
912         /* Dubious BSD thing... Probably nobody even uses it, but
913          * the UNIX standard wants it for whatever reason... -DaveM
914          */
915         case SO_ACCEPTCONN:
916                 v.val = sk->sk_state == TCP_LISTEN;
917                 break;
918
919         case SO_PASSSEC:
920                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
921                 break;
922
923         case SO_PEERSEC:
924                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
925
926         case SO_MARK:
927                 v.val = sk->sk_mark;
928                 break;
929
930         case SO_RXQ_OVFL:
931                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
932                 break;
933
934         default:
935                 return -ENOPROTOOPT;
936         }
937
938         if (len > lv)
939                 len = lv;
940         if (copy_to_user(optval, &v, len))
941                 return -EFAULT;
942 lenout:
943         if (put_user(len, optlen))
944                 return -EFAULT;
945         return 0;
946 }
947
948 /*
949  * Initialize an sk_lock.
950  *
951  * (We also register the sk_lock with the lock validator.)
952  */
953 static inline void sock_lock_init(struct sock *sk)
954 {
955         sock_lock_init_class_and_name(sk,
956                         af_family_slock_key_strings[sk->sk_family],
957                         af_family_slock_keys + sk->sk_family,
958                         af_family_key_strings[sk->sk_family],
959                         af_family_keys + sk->sk_family);
960 }
961
962 /*
963  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
964  * even temporarly, because of RCU lookups. sk_node should also be left as is.
965  */
966 static void sock_copy(struct sock *nsk, const struct sock *osk)
967 {
968 #ifdef CONFIG_SECURITY_NETWORK
969         void *sptr = nsk->sk_security;
970 #endif
971         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
972                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
973                      sizeof(osk->sk_tx_queue_mapping));
974         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
975                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
976 #ifdef CONFIG_SECURITY_NETWORK
977         nsk->sk_security = sptr;
978         security_sk_clone(osk, nsk);
979 #endif
980 }
981
982 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
983                 int family)
984 {
985         struct sock *sk;
986         struct kmem_cache *slab;
987
988         slab = prot->slab;
989         if (slab != NULL) {
990                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
991                 if (!sk)
992                         return sk;
993                 if (priority & __GFP_ZERO) {
994                         /*
995                          * caches using SLAB_DESTROY_BY_RCU should let
996                          * sk_node.next un-modified. Special care is taken
997                          * when initializing object to zero.
998                          */
999                         if (offsetof(struct sock, sk_node.next) != 0)
1000                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
1001                         memset(&sk->sk_node.pprev, 0,
1002                                prot->obj_size - offsetof(struct sock,
1003                                                          sk_node.pprev));
1004                 }
1005         }
1006         else
1007                 sk = kmalloc(prot->obj_size, priority);
1008
1009         if (sk != NULL) {
1010                 kmemcheck_annotate_bitfield(sk, flags);
1011
1012                 if (security_sk_alloc(sk, family, priority))
1013                         goto out_free;
1014
1015                 if (!try_module_get(prot->owner))
1016                         goto out_free_sec;
1017                 sk_tx_queue_clear(sk);
1018         }
1019
1020         return sk;
1021
1022 out_free_sec:
1023         security_sk_free(sk);
1024 out_free:
1025         if (slab != NULL)
1026                 kmem_cache_free(slab, sk);
1027         else
1028                 kfree(sk);
1029         return NULL;
1030 }
1031
1032 static void sk_prot_free(struct proto *prot, struct sock *sk)
1033 {
1034         struct kmem_cache *slab;
1035         struct module *owner;
1036
1037         owner = prot->owner;
1038         slab = prot->slab;
1039
1040         security_sk_free(sk);
1041         if (slab != NULL)
1042                 kmem_cache_free(slab, sk);
1043         else
1044                 kfree(sk);
1045         module_put(owner);
1046 }
1047
1048 /**
1049  *      sk_alloc - All socket objects are allocated here
1050  *      @net: the applicable net namespace
1051  *      @family: protocol family
1052  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1053  *      @prot: struct proto associated with this new sock instance
1054  */
1055 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1056                       struct proto *prot)
1057 {
1058         struct sock *sk;
1059
1060         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1061         if (sk) {
1062                 sk->sk_family = family;
1063                 /*
1064                  * See comment in struct sock definition to understand
1065                  * why we need sk_prot_creator -acme
1066                  */
1067                 sk->sk_prot = sk->sk_prot_creator = prot;
1068                 sock_lock_init(sk);
1069                 sock_net_set(sk, get_net(net));
1070                 atomic_set(&sk->sk_wmem_alloc, 1);
1071         }
1072
1073         return sk;
1074 }
1075 EXPORT_SYMBOL(sk_alloc);
1076
1077 static void __sk_free(struct sock *sk)
1078 {
1079         struct sk_filter *filter;
1080
1081         if (sk->sk_destruct)
1082                 sk->sk_destruct(sk);
1083
1084         filter = rcu_dereference_check(sk->sk_filter,
1085                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1086         if (filter) {
1087                 sk_filter_uncharge(sk, filter);
1088                 rcu_assign_pointer(sk->sk_filter, NULL);
1089         }
1090
1091         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1092         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1093
1094         if (atomic_read(&sk->sk_omem_alloc))
1095                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1096                        __func__, atomic_read(&sk->sk_omem_alloc));
1097
1098         put_net(sock_net(sk));
1099         sk_prot_free(sk->sk_prot_creator, sk);
1100 }
1101
1102 void sk_free(struct sock *sk)
1103 {
1104         /*
1105          * We substract one from sk_wmem_alloc and can know if
1106          * some packets are still in some tx queue.
1107          * If not null, sock_wfree() will call __sk_free(sk) later
1108          */
1109         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1110                 __sk_free(sk);
1111 }
1112 EXPORT_SYMBOL(sk_free);
1113
1114 /*
1115  * Last sock_put should drop referrence to sk->sk_net. It has already
1116  * been dropped in sk_change_net. Taking referrence to stopping namespace
1117  * is not an option.
1118  * Take referrence to a socket to remove it from hash _alive_ and after that
1119  * destroy it in the context of init_net.
1120  */
1121 void sk_release_kernel(struct sock *sk)
1122 {
1123         if (sk == NULL || sk->sk_socket == NULL)
1124                 return;
1125
1126         sock_hold(sk);
1127         sock_release(sk->sk_socket);
1128         release_net(sock_net(sk));
1129         sock_net_set(sk, get_net(&init_net));
1130         sock_put(sk);
1131 }
1132 EXPORT_SYMBOL(sk_release_kernel);
1133
1134 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1135 {
1136         struct sock *newsk;
1137
1138         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1139         if (newsk != NULL) {
1140                 struct sk_filter *filter;
1141
1142                 sock_copy(newsk, sk);
1143
1144                 /* SANITY */
1145                 get_net(sock_net(newsk));
1146                 sk_node_init(&newsk->sk_node);
1147                 sock_lock_init(newsk);
1148                 bh_lock_sock(newsk);
1149                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1150                 newsk->sk_backlog.len = 0;
1151
1152                 atomic_set(&newsk->sk_rmem_alloc, 0);
1153                 /*
1154                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1155                  */
1156                 atomic_set(&newsk->sk_wmem_alloc, 1);
1157                 atomic_set(&newsk->sk_omem_alloc, 0);
1158                 skb_queue_head_init(&newsk->sk_receive_queue);
1159                 skb_queue_head_init(&newsk->sk_write_queue);
1160 #ifdef CONFIG_NET_DMA
1161                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1162 #endif
1163
1164                 spin_lock_init(&newsk->sk_dst_lock);
1165                 rwlock_init(&newsk->sk_callback_lock);
1166                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1167                                 af_callback_keys + newsk->sk_family,
1168                                 af_family_clock_key_strings[newsk->sk_family]);
1169
1170                 newsk->sk_dst_cache     = NULL;
1171                 newsk->sk_wmem_queued   = 0;
1172                 newsk->sk_forward_alloc = 0;
1173                 newsk->sk_send_head     = NULL;
1174                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1175
1176                 sock_reset_flag(newsk, SOCK_DONE);
1177                 skb_queue_head_init(&newsk->sk_error_queue);
1178
1179                 filter = newsk->sk_filter;
1180                 if (filter != NULL)
1181                         sk_filter_charge(newsk, filter);
1182
1183                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1184                         /* It is still raw copy of parent, so invalidate
1185                          * destructor and make plain sk_free() */
1186                         newsk->sk_destruct = NULL;
1187                         sk_free(newsk);
1188                         newsk = NULL;
1189                         goto out;
1190                 }
1191
1192                 newsk->sk_err      = 0;
1193                 newsk->sk_priority = 0;
1194                 /*
1195                  * Before updating sk_refcnt, we must commit prior changes to memory
1196                  * (Documentation/RCU/rculist_nulls.txt for details)
1197                  */
1198                 smp_wmb();
1199                 atomic_set(&newsk->sk_refcnt, 2);
1200
1201                 /*
1202                  * Increment the counter in the same struct proto as the master
1203                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1204                  * is the same as sk->sk_prot->socks, as this field was copied
1205                  * with memcpy).
1206                  *
1207                  * This _changes_ the previous behaviour, where
1208                  * tcp_create_openreq_child always was incrementing the
1209                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1210                  * to be taken into account in all callers. -acme
1211                  */
1212                 sk_refcnt_debug_inc(newsk);
1213                 sk_set_socket(newsk, NULL);
1214                 newsk->sk_wq = NULL;
1215
1216                 if (newsk->sk_prot->sockets_allocated)
1217                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1218
1219                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1220                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1221                         net_enable_timestamp();
1222         }
1223 out:
1224         return newsk;
1225 }
1226 EXPORT_SYMBOL_GPL(sk_clone);
1227
1228 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1229 {
1230         __sk_dst_set(sk, dst);
1231         sk->sk_route_caps = dst->dev->features;
1232         if (sk->sk_route_caps & NETIF_F_GSO)
1233                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1234         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1235         if (sk_can_gso(sk)) {
1236                 if (dst->header_len) {
1237                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1238                 } else {
1239                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1240                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1241                 }
1242         }
1243 }
1244 EXPORT_SYMBOL_GPL(sk_setup_caps);
1245
1246 void __init sk_init(void)
1247 {
1248         if (totalram_pages <= 4096) {
1249                 sysctl_wmem_max = 32767;
1250                 sysctl_rmem_max = 32767;
1251                 sysctl_wmem_default = 32767;
1252                 sysctl_rmem_default = 32767;
1253         } else if (totalram_pages >= 131072) {
1254                 sysctl_wmem_max = 131071;
1255                 sysctl_rmem_max = 131071;
1256         }
1257 }
1258
1259 /*
1260  *      Simple resource managers for sockets.
1261  */
1262
1263
1264 /*
1265  * Write buffer destructor automatically called from kfree_skb.
1266  */
1267 void sock_wfree(struct sk_buff *skb)
1268 {
1269         struct sock *sk = skb->sk;
1270         unsigned int len = skb->truesize;
1271
1272         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1273                 /*
1274                  * Keep a reference on sk_wmem_alloc, this will be released
1275                  * after sk_write_space() call
1276                  */
1277                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1278                 sk->sk_write_space(sk);
1279                 len = 1;
1280         }
1281         /*
1282          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1283          * could not do because of in-flight packets
1284          */
1285         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1286                 __sk_free(sk);
1287 }
1288 EXPORT_SYMBOL(sock_wfree);
1289
1290 /*
1291  * Read buffer destructor automatically called from kfree_skb.
1292  */
1293 void sock_rfree(struct sk_buff *skb)
1294 {
1295         struct sock *sk = skb->sk;
1296
1297         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1298         sk_mem_uncharge(skb->sk, skb->truesize);
1299 }
1300 EXPORT_SYMBOL(sock_rfree);
1301
1302
1303 int sock_i_uid(struct sock *sk)
1304 {
1305         int uid;
1306
1307         read_lock(&sk->sk_callback_lock);
1308         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1309         read_unlock(&sk->sk_callback_lock);
1310         return uid;
1311 }
1312 EXPORT_SYMBOL(sock_i_uid);
1313
1314 unsigned long sock_i_ino(struct sock *sk)
1315 {
1316         unsigned long ino;
1317
1318         read_lock(&sk->sk_callback_lock);
1319         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1320         read_unlock(&sk->sk_callback_lock);
1321         return ino;
1322 }
1323 EXPORT_SYMBOL(sock_i_ino);
1324
1325 /*
1326  * Allocate a skb from the socket's send buffer.
1327  */
1328 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1329                              gfp_t priority)
1330 {
1331         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1332                 struct sk_buff *skb = alloc_skb(size, priority);
1333                 if (skb) {
1334                         skb_set_owner_w(skb, sk);
1335                         return skb;
1336                 }
1337         }
1338         return NULL;
1339 }
1340 EXPORT_SYMBOL(sock_wmalloc);
1341
1342 /*
1343  * Allocate a skb from the socket's receive buffer.
1344  */
1345 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1346                              gfp_t priority)
1347 {
1348         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1349                 struct sk_buff *skb = alloc_skb(size, priority);
1350                 if (skb) {
1351                         skb_set_owner_r(skb, sk);
1352                         return skb;
1353                 }
1354         }
1355         return NULL;
1356 }
1357
1358 /*
1359  * Allocate a memory block from the socket's option memory buffer.
1360  */
1361 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1362 {
1363         if ((unsigned)size <= sysctl_optmem_max &&
1364             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1365                 void *mem;
1366                 /* First do the add, to avoid the race if kmalloc
1367                  * might sleep.
1368                  */
1369                 atomic_add(size, &sk->sk_omem_alloc);
1370                 mem = kmalloc(size, priority);
1371                 if (mem)
1372                         return mem;
1373                 atomic_sub(size, &sk->sk_omem_alloc);
1374         }
1375         return NULL;
1376 }
1377 EXPORT_SYMBOL(sock_kmalloc);
1378
1379 /*
1380  * Free an option memory block.
1381  */
1382 void sock_kfree_s(struct sock *sk, void *mem, int size)
1383 {
1384         kfree(mem);
1385         atomic_sub(size, &sk->sk_omem_alloc);
1386 }
1387 EXPORT_SYMBOL(sock_kfree_s);
1388
1389 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1390    I think, these locks should be removed for datagram sockets.
1391  */
1392 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1393 {
1394         DEFINE_WAIT(wait);
1395
1396         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1397         for (;;) {
1398                 if (!timeo)
1399                         break;
1400                 if (signal_pending(current))
1401                         break;
1402                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1403                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1404                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1405                         break;
1406                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1407                         break;
1408                 if (sk->sk_err)
1409                         break;
1410                 timeo = schedule_timeout(timeo);
1411         }
1412         finish_wait(sk_sleep(sk), &wait);
1413         return timeo;
1414 }
1415
1416
1417 /*
1418  *      Generic send/receive buffer handlers
1419  */
1420
1421 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1422                                      unsigned long data_len, int noblock,
1423                                      int *errcode)
1424 {
1425         struct sk_buff *skb;
1426         gfp_t gfp_mask;
1427         long timeo;
1428         int err;
1429
1430         gfp_mask = sk->sk_allocation;
1431         if (gfp_mask & __GFP_WAIT)
1432                 gfp_mask |= __GFP_REPEAT;
1433
1434         timeo = sock_sndtimeo(sk, noblock);
1435         while (1) {
1436                 err = sock_error(sk);
1437                 if (err != 0)
1438                         goto failure;
1439
1440                 err = -EPIPE;
1441                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1442                         goto failure;
1443
1444                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1445                         skb = alloc_skb(header_len, gfp_mask);
1446                         if (skb) {
1447                                 int npages;
1448                                 int i;
1449
1450                                 /* No pages, we're done... */
1451                                 if (!data_len)
1452                                         break;
1453
1454                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1455                                 skb->truesize += data_len;
1456                                 skb_shinfo(skb)->nr_frags = npages;
1457                                 for (i = 0; i < npages; i++) {
1458                                         struct page *page;
1459                                         skb_frag_t *frag;
1460
1461                                         page = alloc_pages(sk->sk_allocation, 0);
1462                                         if (!page) {
1463                                                 err = -ENOBUFS;
1464                                                 skb_shinfo(skb)->nr_frags = i;
1465                                                 kfree_skb(skb);
1466                                                 goto failure;
1467                                         }
1468
1469                                         frag = &skb_shinfo(skb)->frags[i];
1470                                         frag->page = page;
1471                                         frag->page_offset = 0;
1472                                         frag->size = (data_len >= PAGE_SIZE ?
1473                                                       PAGE_SIZE :
1474                                                       data_len);
1475                                         data_len -= PAGE_SIZE;
1476                                 }
1477
1478                                 /* Full success... */
1479                                 break;
1480                         }
1481                         err = -ENOBUFS;
1482                         goto failure;
1483                 }
1484                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1485                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1486                 err = -EAGAIN;
1487                 if (!timeo)
1488                         goto failure;
1489                 if (signal_pending(current))
1490                         goto interrupted;
1491                 timeo = sock_wait_for_wmem(sk, timeo);
1492         }
1493
1494         skb_set_owner_w(skb, sk);
1495         return skb;
1496
1497 interrupted:
1498         err = sock_intr_errno(timeo);
1499 failure:
1500         *errcode = err;
1501         return NULL;
1502 }
1503 EXPORT_SYMBOL(sock_alloc_send_pskb);
1504
1505 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1506                                     int noblock, int *errcode)
1507 {
1508         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1509 }
1510 EXPORT_SYMBOL(sock_alloc_send_skb);
1511
1512 static void __lock_sock(struct sock *sk)
1513 {
1514         DEFINE_WAIT(wait);
1515
1516         for (;;) {
1517                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1518                                         TASK_UNINTERRUPTIBLE);
1519                 spin_unlock_bh(&sk->sk_lock.slock);
1520                 schedule();
1521                 spin_lock_bh(&sk->sk_lock.slock);
1522                 if (!sock_owned_by_user(sk))
1523                         break;
1524         }
1525         finish_wait(&sk->sk_lock.wq, &wait);
1526 }
1527
1528 static void __release_sock(struct sock *sk)
1529 {
1530         struct sk_buff *skb = sk->sk_backlog.head;
1531
1532         do {
1533                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1534                 bh_unlock_sock(sk);
1535
1536                 do {
1537                         struct sk_buff *next = skb->next;
1538
1539                         skb->next = NULL;
1540                         sk_backlog_rcv(sk, skb);
1541
1542                         /*
1543                          * We are in process context here with softirqs
1544                          * disabled, use cond_resched_softirq() to preempt.
1545                          * This is safe to do because we've taken the backlog
1546                          * queue private:
1547                          */
1548                         cond_resched_softirq();
1549
1550                         skb = next;
1551                 } while (skb != NULL);
1552
1553                 bh_lock_sock(sk);
1554         } while ((skb = sk->sk_backlog.head) != NULL);
1555
1556         /*
1557          * Doing the zeroing here guarantee we can not loop forever
1558          * while a wild producer attempts to flood us.
1559          */
1560         sk->sk_backlog.len = 0;
1561 }
1562
1563 /**
1564  * sk_wait_data - wait for data to arrive at sk_receive_queue
1565  * @sk:    sock to wait on
1566  * @timeo: for how long
1567  *
1568  * Now socket state including sk->sk_err is changed only under lock,
1569  * hence we may omit checks after joining wait queue.
1570  * We check receive queue before schedule() only as optimization;
1571  * it is very likely that release_sock() added new data.
1572  */
1573 int sk_wait_data(struct sock *sk, long *timeo)
1574 {
1575         int rc;
1576         DEFINE_WAIT(wait);
1577
1578         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1579         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1580         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1581         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1582         finish_wait(sk_sleep(sk), &wait);
1583         return rc;
1584 }
1585 EXPORT_SYMBOL(sk_wait_data);
1586
1587 /**
1588  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1589  *      @sk: socket
1590  *      @size: memory size to allocate
1591  *      @kind: allocation type
1592  *
1593  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1594  *      rmem allocation. This function assumes that protocols which have
1595  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1596  */
1597 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1598 {
1599         struct proto *prot = sk->sk_prot;
1600         int amt = sk_mem_pages(size);
1601         int allocated;
1602
1603         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1604         allocated = atomic_add_return(amt, prot->memory_allocated);
1605
1606         /* Under limit. */
1607         if (allocated <= prot->sysctl_mem[0]) {
1608                 if (prot->memory_pressure && *prot->memory_pressure)
1609                         *prot->memory_pressure = 0;
1610                 return 1;
1611         }
1612
1613         /* Under pressure. */
1614         if (allocated > prot->sysctl_mem[1])
1615                 if (prot->enter_memory_pressure)
1616                         prot->enter_memory_pressure(sk);
1617
1618         /* Over hard limit. */
1619         if (allocated > prot->sysctl_mem[2])
1620                 goto suppress_allocation;
1621
1622         /* guarantee minimum buffer size under pressure */
1623         if (kind == SK_MEM_RECV) {
1624                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1625                         return 1;
1626         } else { /* SK_MEM_SEND */
1627                 if (sk->sk_type == SOCK_STREAM) {
1628                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1629                                 return 1;
1630                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1631                            prot->sysctl_wmem[0])
1632                                 return 1;
1633         }
1634
1635         if (prot->memory_pressure) {
1636                 int alloc;
1637
1638                 if (!*prot->memory_pressure)
1639                         return 1;
1640                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1641                 if (prot->sysctl_mem[2] > alloc *
1642                     sk_mem_pages(sk->sk_wmem_queued +
1643                                  atomic_read(&sk->sk_rmem_alloc) +
1644                                  sk->sk_forward_alloc))
1645                         return 1;
1646         }
1647
1648 suppress_allocation:
1649
1650         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1651                 sk_stream_moderate_sndbuf(sk);
1652
1653                 /* Fail only if socket is _under_ its sndbuf.
1654                  * In this case we cannot block, so that we have to fail.
1655                  */
1656                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1657                         return 1;
1658         }
1659
1660         /* Alas. Undo changes. */
1661         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1662         atomic_sub(amt, prot->memory_allocated);
1663         return 0;
1664 }
1665 EXPORT_SYMBOL(__sk_mem_schedule);
1666
1667 /**
1668  *      __sk_reclaim - reclaim memory_allocated
1669  *      @sk: socket
1670  */
1671 void __sk_mem_reclaim(struct sock *sk)
1672 {
1673         struct proto *prot = sk->sk_prot;
1674
1675         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1676                    prot->memory_allocated);
1677         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1678
1679         if (prot->memory_pressure && *prot->memory_pressure &&
1680             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1681                 *prot->memory_pressure = 0;
1682 }
1683 EXPORT_SYMBOL(__sk_mem_reclaim);
1684
1685
1686 /*
1687  * Set of default routines for initialising struct proto_ops when
1688  * the protocol does not support a particular function. In certain
1689  * cases where it makes no sense for a protocol to have a "do nothing"
1690  * function, some default processing is provided.
1691  */
1692
1693 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1694 {
1695         return -EOPNOTSUPP;
1696 }
1697 EXPORT_SYMBOL(sock_no_bind);
1698
1699 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1700                     int len, int flags)
1701 {
1702         return -EOPNOTSUPP;
1703 }
1704 EXPORT_SYMBOL(sock_no_connect);
1705
1706 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1707 {
1708         return -EOPNOTSUPP;
1709 }
1710 EXPORT_SYMBOL(sock_no_socketpair);
1711
1712 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1713 {
1714         return -EOPNOTSUPP;
1715 }
1716 EXPORT_SYMBOL(sock_no_accept);
1717
1718 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1719                     int *len, int peer)
1720 {
1721         return -EOPNOTSUPP;
1722 }
1723 EXPORT_SYMBOL(sock_no_getname);
1724
1725 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1726 {
1727         return 0;
1728 }
1729 EXPORT_SYMBOL(sock_no_poll);
1730
1731 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1732 {
1733         return -EOPNOTSUPP;
1734 }
1735 EXPORT_SYMBOL(sock_no_ioctl);
1736
1737 int sock_no_listen(struct socket *sock, int backlog)
1738 {
1739         return -EOPNOTSUPP;
1740 }
1741 EXPORT_SYMBOL(sock_no_listen);
1742
1743 int sock_no_shutdown(struct socket *sock, int how)
1744 {
1745         return -EOPNOTSUPP;
1746 }
1747 EXPORT_SYMBOL(sock_no_shutdown);
1748
1749 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1750                     char __user *optval, unsigned int optlen)
1751 {
1752         return -EOPNOTSUPP;
1753 }
1754 EXPORT_SYMBOL(sock_no_setsockopt);
1755
1756 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1757                     char __user *optval, int __user *optlen)
1758 {
1759         return -EOPNOTSUPP;
1760 }
1761 EXPORT_SYMBOL(sock_no_getsockopt);
1762
1763 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1764                     size_t len)
1765 {
1766         return -EOPNOTSUPP;
1767 }
1768 EXPORT_SYMBOL(sock_no_sendmsg);
1769
1770 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1771                     size_t len, int flags)
1772 {
1773         return -EOPNOTSUPP;
1774 }
1775 EXPORT_SYMBOL(sock_no_recvmsg);
1776
1777 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1778 {
1779         /* Mirror missing mmap method error code */
1780         return -ENODEV;
1781 }
1782 EXPORT_SYMBOL(sock_no_mmap);
1783
1784 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1785 {
1786         ssize_t res;
1787         struct msghdr msg = {.msg_flags = flags};
1788         struct kvec iov;
1789         char *kaddr = kmap(page);
1790         iov.iov_base = kaddr + offset;
1791         iov.iov_len = size;
1792         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1793         kunmap(page);
1794         return res;
1795 }
1796 EXPORT_SYMBOL(sock_no_sendpage);
1797
1798 /*
1799  *      Default Socket Callbacks
1800  */
1801
1802 static void sock_def_wakeup(struct sock *sk)
1803 {
1804         struct socket_wq *wq;
1805
1806         rcu_read_lock();
1807         wq = rcu_dereference(sk->sk_wq);
1808         if (wq_has_sleeper(wq))
1809                 wake_up_interruptible_all(&wq->wait);
1810         rcu_read_unlock();
1811 }
1812
1813 static void sock_def_error_report(struct sock *sk)
1814 {
1815         struct socket_wq *wq;
1816
1817         rcu_read_lock();
1818         wq = rcu_dereference(sk->sk_wq);
1819         if (wq_has_sleeper(wq))
1820                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1821         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1822         rcu_read_unlock();
1823 }
1824
1825 static void sock_def_readable(struct sock *sk, int len)
1826 {
1827         struct socket_wq *wq;
1828
1829         rcu_read_lock();
1830         wq = rcu_dereference(sk->sk_wq);
1831         if (wq_has_sleeper(wq))
1832                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
1833                                                 POLLRDNORM | POLLRDBAND);
1834         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1835         rcu_read_unlock();
1836 }
1837
1838 static void sock_def_write_space(struct sock *sk)
1839 {
1840         struct socket_wq *wq;
1841
1842         rcu_read_lock();
1843
1844         /* Do not wake up a writer until he can make "significant"
1845          * progress.  --DaveM
1846          */
1847         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1848                 wq = rcu_dereference(sk->sk_wq);
1849                 if (wq_has_sleeper(wq))
1850                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1851                                                 POLLWRNORM | POLLWRBAND);
1852
1853                 /* Should agree with poll, otherwise some programs break */
1854                 if (sock_writeable(sk))
1855                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1856         }
1857
1858         rcu_read_unlock();
1859 }
1860
1861 static void sock_def_destruct(struct sock *sk)
1862 {
1863         kfree(sk->sk_protinfo);
1864 }
1865
1866 void sk_send_sigurg(struct sock *sk)
1867 {
1868         if (sk->sk_socket && sk->sk_socket->file)
1869                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1870                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1871 }
1872 EXPORT_SYMBOL(sk_send_sigurg);
1873
1874 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1875                     unsigned long expires)
1876 {
1877         if (!mod_timer(timer, expires))
1878                 sock_hold(sk);
1879 }
1880 EXPORT_SYMBOL(sk_reset_timer);
1881
1882 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1883 {
1884         if (timer_pending(timer) && del_timer(timer))
1885                 __sock_put(sk);
1886 }
1887 EXPORT_SYMBOL(sk_stop_timer);
1888
1889 void sock_init_data(struct socket *sock, struct sock *sk)
1890 {
1891         skb_queue_head_init(&sk->sk_receive_queue);
1892         skb_queue_head_init(&sk->sk_write_queue);
1893         skb_queue_head_init(&sk->sk_error_queue);
1894 #ifdef CONFIG_NET_DMA
1895         skb_queue_head_init(&sk->sk_async_wait_queue);
1896 #endif
1897
1898         sk->sk_send_head        =       NULL;
1899
1900         init_timer(&sk->sk_timer);
1901
1902         sk->sk_allocation       =       GFP_KERNEL;
1903         sk->sk_rcvbuf           =       sysctl_rmem_default;
1904         sk->sk_sndbuf           =       sysctl_wmem_default;
1905         sk->sk_state            =       TCP_CLOSE;
1906         sk_set_socket(sk, sock);
1907
1908         sock_set_flag(sk, SOCK_ZAPPED);
1909
1910         if (sock) {
1911                 sk->sk_type     =       sock->type;
1912                 sk->sk_wq       =       sock->wq;
1913                 sock->sk        =       sk;
1914         } else
1915                 sk->sk_wq       =       NULL;
1916
1917         spin_lock_init(&sk->sk_dst_lock);
1918         rwlock_init(&sk->sk_callback_lock);
1919         lockdep_set_class_and_name(&sk->sk_callback_lock,
1920                         af_callback_keys + sk->sk_family,
1921                         af_family_clock_key_strings[sk->sk_family]);
1922
1923         sk->sk_state_change     =       sock_def_wakeup;
1924         sk->sk_data_ready       =       sock_def_readable;
1925         sk->sk_write_space      =       sock_def_write_space;
1926         sk->sk_error_report     =       sock_def_error_report;
1927         sk->sk_destruct         =       sock_def_destruct;
1928
1929         sk->sk_sndmsg_page      =       NULL;
1930         sk->sk_sndmsg_off       =       0;
1931
1932         sk->sk_peercred.pid     =       0;
1933         sk->sk_peercred.uid     =       -1;
1934         sk->sk_peercred.gid     =       -1;
1935         sk->sk_write_pending    =       0;
1936         sk->sk_rcvlowat         =       1;
1937         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1938         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1939
1940         sk->sk_stamp = ktime_set(-1L, 0);
1941
1942         /*
1943          * Before updating sk_refcnt, we must commit prior changes to memory
1944          * (Documentation/RCU/rculist_nulls.txt for details)
1945          */
1946         smp_wmb();
1947         atomic_set(&sk->sk_refcnt, 1);
1948         atomic_set(&sk->sk_drops, 0);
1949 }
1950 EXPORT_SYMBOL(sock_init_data);
1951
1952 void lock_sock_nested(struct sock *sk, int subclass)
1953 {
1954         might_sleep();
1955         spin_lock_bh(&sk->sk_lock.slock);
1956         if (sk->sk_lock.owned)
1957                 __lock_sock(sk);
1958         sk->sk_lock.owned = 1;
1959         spin_unlock(&sk->sk_lock.slock);
1960         /*
1961          * The sk_lock has mutex_lock() semantics here:
1962          */
1963         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1964         local_bh_enable();
1965 }
1966 EXPORT_SYMBOL(lock_sock_nested);
1967
1968 void release_sock(struct sock *sk)
1969 {
1970         /*
1971          * The sk_lock has mutex_unlock() semantics:
1972          */
1973         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1974
1975         spin_lock_bh(&sk->sk_lock.slock);
1976         if (sk->sk_backlog.tail)
1977                 __release_sock(sk);
1978         sk->sk_lock.owned = 0;
1979         if (waitqueue_active(&sk->sk_lock.wq))
1980                 wake_up(&sk->sk_lock.wq);
1981         spin_unlock_bh(&sk->sk_lock.slock);
1982 }
1983 EXPORT_SYMBOL(release_sock);
1984
1985 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1986 {
1987         struct timeval tv;
1988         if (!sock_flag(sk, SOCK_TIMESTAMP))
1989                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1990         tv = ktime_to_timeval(sk->sk_stamp);
1991         if (tv.tv_sec == -1)
1992                 return -ENOENT;
1993         if (tv.tv_sec == 0) {
1994                 sk->sk_stamp = ktime_get_real();
1995                 tv = ktime_to_timeval(sk->sk_stamp);
1996         }
1997         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1998 }
1999 EXPORT_SYMBOL(sock_get_timestamp);
2000
2001 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2002 {
2003         struct timespec ts;
2004         if (!sock_flag(sk, SOCK_TIMESTAMP))
2005                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2006         ts = ktime_to_timespec(sk->sk_stamp);
2007         if (ts.tv_sec == -1)
2008                 return -ENOENT;
2009         if (ts.tv_sec == 0) {
2010                 sk->sk_stamp = ktime_get_real();
2011                 ts = ktime_to_timespec(sk->sk_stamp);
2012         }
2013         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2014 }
2015 EXPORT_SYMBOL(sock_get_timestampns);
2016
2017 void sock_enable_timestamp(struct sock *sk, int flag)
2018 {
2019         if (!sock_flag(sk, flag)) {
2020                 sock_set_flag(sk, flag);
2021                 /*
2022                  * we just set one of the two flags which require net
2023                  * time stamping, but time stamping might have been on
2024                  * already because of the other one
2025                  */
2026                 if (!sock_flag(sk,
2027                                 flag == SOCK_TIMESTAMP ?
2028                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2029                                 SOCK_TIMESTAMP))
2030                         net_enable_timestamp();
2031         }
2032 }
2033
2034 /*
2035  *      Get a socket option on an socket.
2036  *
2037  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2038  *      asynchronous errors should be reported by getsockopt. We assume
2039  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2040  */
2041 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2042                            char __user *optval, int __user *optlen)
2043 {
2044         struct sock *sk = sock->sk;
2045
2046         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2047 }
2048 EXPORT_SYMBOL(sock_common_getsockopt);
2049
2050 #ifdef CONFIG_COMPAT
2051 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2052                                   char __user *optval, int __user *optlen)
2053 {
2054         struct sock *sk = sock->sk;
2055
2056         if (sk->sk_prot->compat_getsockopt != NULL)
2057                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2058                                                       optval, optlen);
2059         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2060 }
2061 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2062 #endif
2063
2064 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2065                         struct msghdr *msg, size_t size, int flags)
2066 {
2067         struct sock *sk = sock->sk;
2068         int addr_len = 0;
2069         int err;
2070
2071         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2072                                    flags & ~MSG_DONTWAIT, &addr_len);
2073         if (err >= 0)
2074                 msg->msg_namelen = addr_len;
2075         return err;
2076 }
2077 EXPORT_SYMBOL(sock_common_recvmsg);
2078
2079 /*
2080  *      Set socket options on an inet socket.
2081  */
2082 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2083                            char __user *optval, unsigned int optlen)
2084 {
2085         struct sock *sk = sock->sk;
2086
2087         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2088 }
2089 EXPORT_SYMBOL(sock_common_setsockopt);
2090
2091 #ifdef CONFIG_COMPAT
2092 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2093                                   char __user *optval, unsigned int optlen)
2094 {
2095         struct sock *sk = sock->sk;
2096
2097         if (sk->sk_prot->compat_setsockopt != NULL)
2098                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2099                                                       optval, optlen);
2100         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2101 }
2102 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2103 #endif
2104
2105 void sk_common_release(struct sock *sk)
2106 {
2107         if (sk->sk_prot->destroy)
2108                 sk->sk_prot->destroy(sk);
2109
2110         /*
2111          * Observation: when sock_common_release is called, processes have
2112          * no access to socket. But net still has.
2113          * Step one, detach it from networking:
2114          *
2115          * A. Remove from hash tables.
2116          */
2117
2118         sk->sk_prot->unhash(sk);
2119
2120         /*
2121          * In this point socket cannot receive new packets, but it is possible
2122          * that some packets are in flight because some CPU runs receiver and
2123          * did hash table lookup before we unhashed socket. They will achieve
2124          * receive queue and will be purged by socket destructor.
2125          *
2126          * Also we still have packets pending on receive queue and probably,
2127          * our own packets waiting in device queues. sock_destroy will drain
2128          * receive queue, but transmitted packets will delay socket destruction
2129          * until the last reference will be released.
2130          */
2131
2132         sock_orphan(sk);
2133
2134         xfrm_sk_free_policy(sk);
2135
2136         sk_refcnt_debug_release(sk);
2137         sock_put(sk);
2138 }
2139 EXPORT_SYMBOL(sk_common_release);
2140
2141 static DEFINE_RWLOCK(proto_list_lock);
2142 static LIST_HEAD(proto_list);
2143
2144 #ifdef CONFIG_PROC_FS
2145 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2146 struct prot_inuse {
2147         int val[PROTO_INUSE_NR];
2148 };
2149
2150 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2151
2152 #ifdef CONFIG_NET_NS
2153 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2154 {
2155         int cpu = smp_processor_id();
2156         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2157 }
2158 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2159
2160 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2161 {
2162         int cpu, idx = prot->inuse_idx;
2163         int res = 0;
2164
2165         for_each_possible_cpu(cpu)
2166                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2167
2168         return res >= 0 ? res : 0;
2169 }
2170 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2171
2172 static int __net_init sock_inuse_init_net(struct net *net)
2173 {
2174         net->core.inuse = alloc_percpu(struct prot_inuse);
2175         return net->core.inuse ? 0 : -ENOMEM;
2176 }
2177
2178 static void __net_exit sock_inuse_exit_net(struct net *net)
2179 {
2180         free_percpu(net->core.inuse);
2181 }
2182
2183 static struct pernet_operations net_inuse_ops = {
2184         .init = sock_inuse_init_net,
2185         .exit = sock_inuse_exit_net,
2186 };
2187
2188 static __init int net_inuse_init(void)
2189 {
2190         if (register_pernet_subsys(&net_inuse_ops))
2191                 panic("Cannot initialize net inuse counters");
2192
2193         return 0;
2194 }
2195
2196 core_initcall(net_inuse_init);
2197 #else
2198 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2199
2200 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2201 {
2202         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2203 }
2204 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2205
2206 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2207 {
2208         int cpu, idx = prot->inuse_idx;
2209         int res = 0;
2210
2211         for_each_possible_cpu(cpu)
2212                 res += per_cpu(prot_inuse, cpu).val[idx];
2213
2214         return res >= 0 ? res : 0;
2215 }
2216 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2217 #endif
2218
2219 static void assign_proto_idx(struct proto *prot)
2220 {
2221         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2222
2223         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2224                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2225                 return;
2226         }
2227
2228         set_bit(prot->inuse_idx, proto_inuse_idx);
2229 }
2230
2231 static void release_proto_idx(struct proto *prot)
2232 {
2233         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2234                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2235 }
2236 #else
2237 static inline void assign_proto_idx(struct proto *prot)
2238 {
2239 }
2240
2241 static inline void release_proto_idx(struct proto *prot)
2242 {
2243 }
2244 #endif
2245
2246 int proto_register(struct proto *prot, int alloc_slab)
2247 {
2248         if (alloc_slab) {
2249                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2250                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2251                                         NULL);
2252
2253                 if (prot->slab == NULL) {
2254                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2255                                prot->name);
2256                         goto out;
2257                 }
2258
2259                 if (prot->rsk_prot != NULL) {
2260                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2261                         if (prot->rsk_prot->slab_name == NULL)
2262                                 goto out_free_sock_slab;
2263
2264                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2265                                                                  prot->rsk_prot->obj_size, 0,
2266                                                                  SLAB_HWCACHE_ALIGN, NULL);
2267
2268                         if (prot->rsk_prot->slab == NULL) {
2269                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2270                                        prot->name);
2271                                 goto out_free_request_sock_slab_name;
2272                         }
2273                 }
2274
2275                 if (prot->twsk_prot != NULL) {
2276                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2277
2278                         if (prot->twsk_prot->twsk_slab_name == NULL)
2279                                 goto out_free_request_sock_slab;
2280
2281                         prot->twsk_prot->twsk_slab =
2282                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2283                                                   prot->twsk_prot->twsk_obj_size,
2284                                                   0,
2285                                                   SLAB_HWCACHE_ALIGN |
2286                                                         prot->slab_flags,
2287                                                   NULL);
2288                         if (prot->twsk_prot->twsk_slab == NULL)
2289                                 goto out_free_timewait_sock_slab_name;
2290                 }
2291         }
2292
2293         write_lock(&proto_list_lock);
2294         list_add(&prot->node, &proto_list);
2295         assign_proto_idx(prot);
2296         write_unlock(&proto_list_lock);
2297         return 0;
2298
2299 out_free_timewait_sock_slab_name:
2300         kfree(prot->twsk_prot->twsk_slab_name);
2301 out_free_request_sock_slab:
2302         if (prot->rsk_prot && prot->rsk_prot->slab) {
2303                 kmem_cache_destroy(prot->rsk_prot->slab);
2304                 prot->rsk_prot->slab = NULL;
2305         }
2306 out_free_request_sock_slab_name:
2307         if (prot->rsk_prot)
2308                 kfree(prot->rsk_prot->slab_name);
2309 out_free_sock_slab:
2310         kmem_cache_destroy(prot->slab);
2311         prot->slab = NULL;
2312 out:
2313         return -ENOBUFS;
2314 }
2315 EXPORT_SYMBOL(proto_register);
2316
2317 void proto_unregister(struct proto *prot)
2318 {
2319         write_lock(&proto_list_lock);
2320         release_proto_idx(prot);
2321         list_del(&prot->node);
2322         write_unlock(&proto_list_lock);
2323
2324         if (prot->slab != NULL) {
2325                 kmem_cache_destroy(prot->slab);
2326                 prot->slab = NULL;
2327         }
2328
2329         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2330                 kmem_cache_destroy(prot->rsk_prot->slab);
2331                 kfree(prot->rsk_prot->slab_name);
2332                 prot->rsk_prot->slab = NULL;
2333         }
2334
2335         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2336                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2337                 kfree(prot->twsk_prot->twsk_slab_name);
2338                 prot->twsk_prot->twsk_slab = NULL;
2339         }
2340 }
2341 EXPORT_SYMBOL(proto_unregister);
2342
2343 #ifdef CONFIG_PROC_FS
2344 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2345         __acquires(proto_list_lock)
2346 {
2347         read_lock(&proto_list_lock);
2348         return seq_list_start_head(&proto_list, *pos);
2349 }
2350
2351 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2352 {
2353         return seq_list_next(v, &proto_list, pos);
2354 }
2355
2356 static void proto_seq_stop(struct seq_file *seq, void *v)
2357         __releases(proto_list_lock)
2358 {
2359         read_unlock(&proto_list_lock);
2360 }
2361
2362 static char proto_method_implemented(const void *method)
2363 {
2364         return method == NULL ? 'n' : 'y';
2365 }
2366
2367 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2368 {
2369         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2370                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2371                    proto->name,
2372                    proto->obj_size,
2373                    sock_prot_inuse_get(seq_file_net(seq), proto),
2374                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2375                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2376                    proto->max_header,
2377                    proto->slab == NULL ? "no" : "yes",
2378                    module_name(proto->owner),
2379                    proto_method_implemented(proto->close),
2380                    proto_method_implemented(proto->connect),
2381                    proto_method_implemented(proto->disconnect),
2382                    proto_method_implemented(proto->accept),
2383                    proto_method_implemented(proto->ioctl),
2384                    proto_method_implemented(proto->init),
2385                    proto_method_implemented(proto->destroy),
2386                    proto_method_implemented(proto->shutdown),
2387                    proto_method_implemented(proto->setsockopt),
2388                    proto_method_implemented(proto->getsockopt),
2389                    proto_method_implemented(proto->sendmsg),
2390                    proto_method_implemented(proto->recvmsg),
2391                    proto_method_implemented(proto->sendpage),
2392                    proto_method_implemented(proto->bind),
2393                    proto_method_implemented(proto->backlog_rcv),
2394                    proto_method_implemented(proto->hash),
2395                    proto_method_implemented(proto->unhash),
2396                    proto_method_implemented(proto->get_port),
2397                    proto_method_implemented(proto->enter_memory_pressure));
2398 }
2399
2400 static int proto_seq_show(struct seq_file *seq, void *v)
2401 {
2402         if (v == &proto_list)
2403                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2404                            "protocol",
2405                            "size",
2406                            "sockets",
2407                            "memory",
2408                            "press",
2409                            "maxhdr",
2410                            "slab",
2411                            "module",
2412                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2413         else
2414                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2415         return 0;
2416 }
2417
2418 static const struct seq_operations proto_seq_ops = {
2419         .start  = proto_seq_start,
2420         .next   = proto_seq_next,
2421         .stop   = proto_seq_stop,
2422         .show   = proto_seq_show,
2423 };
2424
2425 static int proto_seq_open(struct inode *inode, struct file *file)
2426 {
2427         return seq_open_net(inode, file, &proto_seq_ops,
2428                             sizeof(struct seq_net_private));
2429 }
2430
2431 static const struct file_operations proto_seq_fops = {
2432         .owner          = THIS_MODULE,
2433         .open           = proto_seq_open,
2434         .read           = seq_read,
2435         .llseek         = seq_lseek,
2436         .release        = seq_release_net,
2437 };
2438
2439 static __net_init int proto_init_net(struct net *net)
2440 {
2441         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2442                 return -ENOMEM;
2443
2444         return 0;
2445 }
2446
2447 static __net_exit void proto_exit_net(struct net *net)
2448 {
2449         proc_net_remove(net, "protocols");
2450 }
2451
2452
2453 static __net_initdata struct pernet_operations proto_net_ops = {
2454         .init = proto_init_net,
2455         .exit = proto_exit_net,
2456 };
2457
2458 static int __init proto_init(void)
2459 {
2460         return register_pernet_subsys(&proto_net_ops);
2461 }
2462
2463 subsys_initcall(proto_init);
2464
2465 #endif /* PROC_FS */