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