04e35eb2e736144a554dcd02354f38d943ba4e3e
[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 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 EXPORT_SYMBOL(sk_free);
1035
1036 /*
1037  * Last sock_put should drop referrence to sk->sk_net. It has already
1038  * been dropped in sk_change_net. Taking referrence to stopping namespace
1039  * is not an option.
1040  * Take referrence to a socket to remove it from hash _alive_ and after that
1041  * destroy it in the context of init_net.
1042  */
1043 void sk_release_kernel(struct sock *sk)
1044 {
1045         if (sk == NULL || sk->sk_socket == NULL)
1046                 return;
1047
1048         sock_hold(sk);
1049         sock_release(sk->sk_socket);
1050         release_net(sock_net(sk));
1051         sock_net_set(sk, get_net(&init_net));
1052         sock_put(sk);
1053 }
1054 EXPORT_SYMBOL(sk_release_kernel);
1055
1056 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1057 {
1058         struct sock *newsk;
1059
1060         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1061         if (newsk != NULL) {
1062                 struct sk_filter *filter;
1063
1064                 sock_copy(newsk, sk);
1065
1066                 /* SANITY */
1067                 get_net(sock_net(newsk));
1068                 sk_node_init(&newsk->sk_node);
1069                 sock_lock_init(newsk);
1070                 bh_lock_sock(newsk);
1071                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1072
1073                 atomic_set(&newsk->sk_rmem_alloc, 0);
1074                 atomic_set(&newsk->sk_wmem_alloc, 0);
1075                 atomic_set(&newsk->sk_omem_alloc, 0);
1076                 skb_queue_head_init(&newsk->sk_receive_queue);
1077                 skb_queue_head_init(&newsk->sk_write_queue);
1078 #ifdef CONFIG_NET_DMA
1079                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1080 #endif
1081
1082                 rwlock_init(&newsk->sk_dst_lock);
1083                 rwlock_init(&newsk->sk_callback_lock);
1084                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1085                                 af_callback_keys + newsk->sk_family,
1086                                 af_family_clock_key_strings[newsk->sk_family]);
1087
1088                 newsk->sk_dst_cache     = NULL;
1089                 newsk->sk_wmem_queued   = 0;
1090                 newsk->sk_forward_alloc = 0;
1091                 newsk->sk_send_head     = NULL;
1092                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1093
1094                 sock_reset_flag(newsk, SOCK_DONE);
1095                 skb_queue_head_init(&newsk->sk_error_queue);
1096
1097                 filter = newsk->sk_filter;
1098                 if (filter != NULL)
1099                         sk_filter_charge(newsk, filter);
1100
1101                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1102                         /* It is still raw copy of parent, so invalidate
1103                          * destructor and make plain sk_free() */
1104                         newsk->sk_destruct = NULL;
1105                         sk_free(newsk);
1106                         newsk = NULL;
1107                         goto out;
1108                 }
1109
1110                 newsk->sk_err      = 0;
1111                 newsk->sk_priority = 0;
1112                 atomic_set(&newsk->sk_refcnt, 2);
1113
1114                 /*
1115                  * Increment the counter in the same struct proto as the master
1116                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1117                  * is the same as sk->sk_prot->socks, as this field was copied
1118                  * with memcpy).
1119                  *
1120                  * This _changes_ the previous behaviour, where
1121                  * tcp_create_openreq_child always was incrementing the
1122                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1123                  * to be taken into account in all callers. -acme
1124                  */
1125                 sk_refcnt_debug_inc(newsk);
1126                 sk_set_socket(newsk, NULL);
1127                 newsk->sk_sleep  = NULL;
1128
1129                 if (newsk->sk_prot->sockets_allocated)
1130                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1131         }
1132 out:
1133         return newsk;
1134 }
1135 EXPORT_SYMBOL_GPL(sk_clone);
1136
1137 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1138 {
1139         __sk_dst_set(sk, dst);
1140         sk->sk_route_caps = dst->dev->features;
1141         if (sk->sk_route_caps & NETIF_F_GSO)
1142                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1143         if (sk_can_gso(sk)) {
1144                 if (dst->header_len) {
1145                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1146                 } else {
1147                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1148                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1149                 }
1150         }
1151 }
1152 EXPORT_SYMBOL_GPL(sk_setup_caps);
1153
1154 void __init sk_init(void)
1155 {
1156         if (num_physpages <= 4096) {
1157                 sysctl_wmem_max = 32767;
1158                 sysctl_rmem_max = 32767;
1159                 sysctl_wmem_default = 32767;
1160                 sysctl_rmem_default = 32767;
1161         } else if (num_physpages >= 131072) {
1162                 sysctl_wmem_max = 131071;
1163                 sysctl_rmem_max = 131071;
1164         }
1165 }
1166
1167 /*
1168  *      Simple resource managers for sockets.
1169  */
1170
1171
1172 /*
1173  * Write buffer destructor automatically called from kfree_skb.
1174  */
1175 void sock_wfree(struct sk_buff *skb)
1176 {
1177         struct sock *sk = skb->sk;
1178
1179         /* In case it might be waiting for more memory. */
1180         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1181         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1182                 sk->sk_write_space(sk);
1183         sock_put(sk);
1184 }
1185 EXPORT_SYMBOL(sock_wfree);
1186
1187 /*
1188  * Read buffer destructor automatically called from kfree_skb.
1189  */
1190 void sock_rfree(struct sk_buff *skb)
1191 {
1192         struct sock *sk = skb->sk;
1193
1194         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1195         sk_mem_uncharge(skb->sk, skb->truesize);
1196 }
1197 EXPORT_SYMBOL(sock_rfree);
1198
1199
1200 int sock_i_uid(struct sock *sk)
1201 {
1202         int uid;
1203
1204         read_lock(&sk->sk_callback_lock);
1205         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1206         read_unlock(&sk->sk_callback_lock);
1207         return uid;
1208 }
1209 EXPORT_SYMBOL(sock_i_uid);
1210
1211 unsigned long sock_i_ino(struct sock *sk)
1212 {
1213         unsigned long ino;
1214
1215         read_lock(&sk->sk_callback_lock);
1216         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1217         read_unlock(&sk->sk_callback_lock);
1218         return ino;
1219 }
1220 EXPORT_SYMBOL(sock_i_ino);
1221
1222 /*
1223  * Allocate a skb from the socket's send buffer.
1224  */
1225 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1226                              gfp_t priority)
1227 {
1228         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1229                 struct sk_buff *skb = alloc_skb(size, priority);
1230                 if (skb) {
1231                         skb_set_owner_w(skb, sk);
1232                         return skb;
1233                 }
1234         }
1235         return NULL;
1236 }
1237 EXPORT_SYMBOL(sock_wmalloc);
1238
1239 /*
1240  * Allocate a skb from the socket's receive buffer.
1241  */
1242 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1243                              gfp_t priority)
1244 {
1245         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1246                 struct sk_buff *skb = alloc_skb(size, priority);
1247                 if (skb) {
1248                         skb_set_owner_r(skb, sk);
1249                         return skb;
1250                 }
1251         }
1252         return NULL;
1253 }
1254
1255 /*
1256  * Allocate a memory block from the socket's option memory buffer.
1257  */
1258 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1259 {
1260         if ((unsigned)size <= sysctl_optmem_max &&
1261             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1262                 void *mem;
1263                 /* First do the add, to avoid the race if kmalloc
1264                  * might sleep.
1265                  */
1266                 atomic_add(size, &sk->sk_omem_alloc);
1267                 mem = kmalloc(size, priority);
1268                 if (mem)
1269                         return mem;
1270                 atomic_sub(size, &sk->sk_omem_alloc);
1271         }
1272         return NULL;
1273 }
1274 EXPORT_SYMBOL(sock_kmalloc);
1275
1276 /*
1277  * Free an option memory block.
1278  */
1279 void sock_kfree_s(struct sock *sk, void *mem, int size)
1280 {
1281         kfree(mem);
1282         atomic_sub(size, &sk->sk_omem_alloc);
1283 }
1284 EXPORT_SYMBOL(sock_kfree_s);
1285
1286 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1287    I think, these locks should be removed for datagram sockets.
1288  */
1289 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1290 {
1291         DEFINE_WAIT(wait);
1292
1293         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1294         for (;;) {
1295                 if (!timeo)
1296                         break;
1297                 if (signal_pending(current))
1298                         break;
1299                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1300                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1301                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1302                         break;
1303                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1304                         break;
1305                 if (sk->sk_err)
1306                         break;
1307                 timeo = schedule_timeout(timeo);
1308         }
1309         finish_wait(sk->sk_sleep, &wait);
1310         return timeo;
1311 }
1312
1313
1314 /*
1315  *      Generic send/receive buffer handlers
1316  */
1317
1318 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1319                                      unsigned long data_len, int noblock,
1320                                      int *errcode)
1321 {
1322         struct sk_buff *skb;
1323         gfp_t gfp_mask;
1324         long timeo;
1325         int err;
1326
1327         gfp_mask = sk->sk_allocation;
1328         if (gfp_mask & __GFP_WAIT)
1329                 gfp_mask |= __GFP_REPEAT;
1330
1331         timeo = sock_sndtimeo(sk, noblock);
1332         while (1) {
1333                 err = sock_error(sk);
1334                 if (err != 0)
1335                         goto failure;
1336
1337                 err = -EPIPE;
1338                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1339                         goto failure;
1340
1341                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1342                         skb = alloc_skb(header_len, gfp_mask);
1343                         if (skb) {
1344                                 int npages;
1345                                 int i;
1346
1347                                 /* No pages, we're done... */
1348                                 if (!data_len)
1349                                         break;
1350
1351                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1352                                 skb->truesize += data_len;
1353                                 skb_shinfo(skb)->nr_frags = npages;
1354                                 for (i = 0; i < npages; i++) {
1355                                         struct page *page;
1356                                         skb_frag_t *frag;
1357
1358                                         page = alloc_pages(sk->sk_allocation, 0);
1359                                         if (!page) {
1360                                                 err = -ENOBUFS;
1361                                                 skb_shinfo(skb)->nr_frags = i;
1362                                                 kfree_skb(skb);
1363                                                 goto failure;
1364                                         }
1365
1366                                         frag = &skb_shinfo(skb)->frags[i];
1367                                         frag->page = page;
1368                                         frag->page_offset = 0;
1369                                         frag->size = (data_len >= PAGE_SIZE ?
1370                                                       PAGE_SIZE :
1371                                                       data_len);
1372                                         data_len -= PAGE_SIZE;
1373                                 }
1374
1375                                 /* Full success... */
1376                                 break;
1377                         }
1378                         err = -ENOBUFS;
1379                         goto failure;
1380                 }
1381                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1382                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1383                 err = -EAGAIN;
1384                 if (!timeo)
1385                         goto failure;
1386                 if (signal_pending(current))
1387                         goto interrupted;
1388                 timeo = sock_wait_for_wmem(sk, timeo);
1389         }
1390
1391         skb_set_owner_w(skb, sk);
1392         return skb;
1393
1394 interrupted:
1395         err = sock_intr_errno(timeo);
1396 failure:
1397         *errcode = err;
1398         return NULL;
1399 }
1400 EXPORT_SYMBOL(sock_alloc_send_pskb);
1401
1402 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1403                                     int noblock, int *errcode)
1404 {
1405         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1406 }
1407 EXPORT_SYMBOL(sock_alloc_send_skb);
1408
1409 static void __lock_sock(struct sock *sk)
1410 {
1411         DEFINE_WAIT(wait);
1412
1413         for (;;) {
1414                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1415                                         TASK_UNINTERRUPTIBLE);
1416                 spin_unlock_bh(&sk->sk_lock.slock);
1417                 schedule();
1418                 spin_lock_bh(&sk->sk_lock.slock);
1419                 if (!sock_owned_by_user(sk))
1420                         break;
1421         }
1422         finish_wait(&sk->sk_lock.wq, &wait);
1423 }
1424
1425 static void __release_sock(struct sock *sk)
1426 {
1427         struct sk_buff *skb = sk->sk_backlog.head;
1428
1429         do {
1430                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1431                 bh_unlock_sock(sk);
1432
1433                 do {
1434                         struct sk_buff *next = skb->next;
1435
1436                         skb->next = NULL;
1437                         sk_backlog_rcv(sk, skb);
1438
1439                         /*
1440                          * We are in process context here with softirqs
1441                          * disabled, use cond_resched_softirq() to preempt.
1442                          * This is safe to do because we've taken the backlog
1443                          * queue private:
1444                          */
1445                         cond_resched_softirq();
1446
1447                         skb = next;
1448                 } while (skb != NULL);
1449
1450                 bh_lock_sock(sk);
1451         } while ((skb = sk->sk_backlog.head) != NULL);
1452 }
1453
1454 /**
1455  * sk_wait_data - wait for data to arrive at sk_receive_queue
1456  * @sk:    sock to wait on
1457  * @timeo: for how long
1458  *
1459  * Now socket state including sk->sk_err is changed only under lock,
1460  * hence we may omit checks after joining wait queue.
1461  * We check receive queue before schedule() only as optimization;
1462  * it is very likely that release_sock() added new data.
1463  */
1464 int sk_wait_data(struct sock *sk, long *timeo)
1465 {
1466         int rc;
1467         DEFINE_WAIT(wait);
1468
1469         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1470         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1471         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1472         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1473         finish_wait(sk->sk_sleep, &wait);
1474         return rc;
1475 }
1476 EXPORT_SYMBOL(sk_wait_data);
1477
1478 /**
1479  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1480  *      @sk: socket
1481  *      @size: memory size to allocate
1482  *      @kind: allocation type
1483  *
1484  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1485  *      rmem allocation. This function assumes that protocols which have
1486  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1487  */
1488 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1489 {
1490         struct proto *prot = sk->sk_prot;
1491         int amt = sk_mem_pages(size);
1492         int allocated;
1493
1494         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1495         allocated = atomic_add_return(amt, prot->memory_allocated);
1496
1497         /* Under limit. */
1498         if (allocated <= prot->sysctl_mem[0]) {
1499                 if (prot->memory_pressure && *prot->memory_pressure)
1500                         *prot->memory_pressure = 0;
1501                 return 1;
1502         }
1503
1504         /* Under pressure. */
1505         if (allocated > prot->sysctl_mem[1])
1506                 if (prot->enter_memory_pressure)
1507                         prot->enter_memory_pressure(sk);
1508
1509         /* Over hard limit. */
1510         if (allocated > prot->sysctl_mem[2])
1511                 goto suppress_allocation;
1512
1513         /* guarantee minimum buffer size under pressure */
1514         if (kind == SK_MEM_RECV) {
1515                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1516                         return 1;
1517         } else { /* SK_MEM_SEND */
1518                 if (sk->sk_type == SOCK_STREAM) {
1519                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1520                                 return 1;
1521                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1522                            prot->sysctl_wmem[0])
1523                                 return 1;
1524         }
1525
1526         if (prot->memory_pressure) {
1527                 int alloc;
1528
1529                 if (!*prot->memory_pressure)
1530                         return 1;
1531                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1532                 if (prot->sysctl_mem[2] > alloc *
1533                     sk_mem_pages(sk->sk_wmem_queued +
1534                                  atomic_read(&sk->sk_rmem_alloc) +
1535                                  sk->sk_forward_alloc))
1536                         return 1;
1537         }
1538
1539 suppress_allocation:
1540
1541         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1542                 sk_stream_moderate_sndbuf(sk);
1543
1544                 /* Fail only if socket is _under_ its sndbuf.
1545                  * In this case we cannot block, so that we have to fail.
1546                  */
1547                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1548                         return 1;
1549         }
1550
1551         /* Alas. Undo changes. */
1552         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1553         atomic_sub(amt, prot->memory_allocated);
1554         return 0;
1555 }
1556 EXPORT_SYMBOL(__sk_mem_schedule);
1557
1558 /**
1559  *      __sk_reclaim - reclaim memory_allocated
1560  *      @sk: socket
1561  */
1562 void __sk_mem_reclaim(struct sock *sk)
1563 {
1564         struct proto *prot = sk->sk_prot;
1565
1566         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1567                    prot->memory_allocated);
1568         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1569
1570         if (prot->memory_pressure && *prot->memory_pressure &&
1571             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1572                 *prot->memory_pressure = 0;
1573 }
1574 EXPORT_SYMBOL(__sk_mem_reclaim);
1575
1576
1577 /*
1578  * Set of default routines for initialising struct proto_ops when
1579  * the protocol does not support a particular function. In certain
1580  * cases where it makes no sense for a protocol to have a "do nothing"
1581  * function, some default processing is provided.
1582  */
1583
1584 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1585 {
1586         return -EOPNOTSUPP;
1587 }
1588 EXPORT_SYMBOL(sock_no_bind);
1589
1590 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1591                     int len, int flags)
1592 {
1593         return -EOPNOTSUPP;
1594 }
1595 EXPORT_SYMBOL(sock_no_connect);
1596
1597 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1598 {
1599         return -EOPNOTSUPP;
1600 }
1601 EXPORT_SYMBOL(sock_no_socketpair);
1602
1603 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1604 {
1605         return -EOPNOTSUPP;
1606 }
1607 EXPORT_SYMBOL(sock_no_accept);
1608
1609 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1610                     int *len, int peer)
1611 {
1612         return -EOPNOTSUPP;
1613 }
1614 EXPORT_SYMBOL(sock_no_getname);
1615
1616 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1617 {
1618         return 0;
1619 }
1620 EXPORT_SYMBOL(sock_no_poll);
1621
1622 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1623 {
1624         return -EOPNOTSUPP;
1625 }
1626 EXPORT_SYMBOL(sock_no_ioctl);
1627
1628 int sock_no_listen(struct socket *sock, int backlog)
1629 {
1630         return -EOPNOTSUPP;
1631 }
1632 EXPORT_SYMBOL(sock_no_listen);
1633
1634 int sock_no_shutdown(struct socket *sock, int how)
1635 {
1636         return -EOPNOTSUPP;
1637 }
1638 EXPORT_SYMBOL(sock_no_shutdown);
1639
1640 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1641                     char __user *optval, int optlen)
1642 {
1643         return -EOPNOTSUPP;
1644 }
1645 EXPORT_SYMBOL(sock_no_setsockopt);
1646
1647 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1648                     char __user *optval, int __user *optlen)
1649 {
1650         return -EOPNOTSUPP;
1651 }
1652 EXPORT_SYMBOL(sock_no_getsockopt);
1653
1654 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1655                     size_t len)
1656 {
1657         return -EOPNOTSUPP;
1658 }
1659 EXPORT_SYMBOL(sock_no_sendmsg);
1660
1661 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1662                     size_t len, int flags)
1663 {
1664         return -EOPNOTSUPP;
1665 }
1666 EXPORT_SYMBOL(sock_no_recvmsg);
1667
1668 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1669 {
1670         /* Mirror missing mmap method error code */
1671         return -ENODEV;
1672 }
1673 EXPORT_SYMBOL(sock_no_mmap);
1674
1675 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1676 {
1677         ssize_t res;
1678         struct msghdr msg = {.msg_flags = flags};
1679         struct kvec iov;
1680         char *kaddr = kmap(page);
1681         iov.iov_base = kaddr + offset;
1682         iov.iov_len = size;
1683         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1684         kunmap(page);
1685         return res;
1686 }
1687 EXPORT_SYMBOL(sock_no_sendpage);
1688
1689 /*
1690  *      Default Socket Callbacks
1691  */
1692
1693 static void sock_def_wakeup(struct sock *sk)
1694 {
1695         read_lock(&sk->sk_callback_lock);
1696         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1697                 wake_up_interruptible_all(sk->sk_sleep);
1698         read_unlock(&sk->sk_callback_lock);
1699 }
1700
1701 static void sock_def_error_report(struct sock *sk)
1702 {
1703         read_lock(&sk->sk_callback_lock);
1704         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1705                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1706         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1707         read_unlock(&sk->sk_callback_lock);
1708 }
1709
1710 static void sock_def_readable(struct sock *sk, int len)
1711 {
1712         read_lock(&sk->sk_callback_lock);
1713         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1714                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1715                                                 POLLRDNORM | POLLRDBAND);
1716         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1717         read_unlock(&sk->sk_callback_lock);
1718 }
1719
1720 static void sock_def_write_space(struct sock *sk)
1721 {
1722         read_lock(&sk->sk_callback_lock);
1723
1724         /* Do not wake up a writer until he can make "significant"
1725          * progress.  --DaveM
1726          */
1727         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1728                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1729                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1730                                                 POLLWRNORM | POLLWRBAND);
1731
1732                 /* Should agree with poll, otherwise some programs break */
1733                 if (sock_writeable(sk))
1734                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1735         }
1736
1737         read_unlock(&sk->sk_callback_lock);
1738 }
1739
1740 static void sock_def_destruct(struct sock *sk)
1741 {
1742         kfree(sk->sk_protinfo);
1743 }
1744
1745 void sk_send_sigurg(struct sock *sk)
1746 {
1747         if (sk->sk_socket && sk->sk_socket->file)
1748                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1749                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1750 }
1751 EXPORT_SYMBOL(sk_send_sigurg);
1752
1753 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1754                     unsigned long expires)
1755 {
1756         if (!mod_timer(timer, expires))
1757                 sock_hold(sk);
1758 }
1759 EXPORT_SYMBOL(sk_reset_timer);
1760
1761 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1762 {
1763         if (timer_pending(timer) && del_timer(timer))
1764                 __sock_put(sk);
1765 }
1766 EXPORT_SYMBOL(sk_stop_timer);
1767
1768 void sock_init_data(struct socket *sock, struct sock *sk)
1769 {
1770         skb_queue_head_init(&sk->sk_receive_queue);
1771         skb_queue_head_init(&sk->sk_write_queue);
1772         skb_queue_head_init(&sk->sk_error_queue);
1773 #ifdef CONFIG_NET_DMA
1774         skb_queue_head_init(&sk->sk_async_wait_queue);
1775 #endif
1776
1777         sk->sk_send_head        =       NULL;
1778
1779         init_timer(&sk->sk_timer);
1780
1781         sk->sk_allocation       =       GFP_KERNEL;
1782         sk->sk_rcvbuf           =       sysctl_rmem_default;
1783         sk->sk_sndbuf           =       sysctl_wmem_default;
1784         sk->sk_state            =       TCP_CLOSE;
1785         sk_set_socket(sk, sock);
1786
1787         sock_set_flag(sk, SOCK_ZAPPED);
1788
1789         if (sock) {
1790                 sk->sk_type     =       sock->type;
1791                 sk->sk_sleep    =       &sock->wait;
1792                 sock->sk        =       sk;
1793         } else
1794                 sk->sk_sleep    =       NULL;
1795
1796         rwlock_init(&sk->sk_dst_lock);
1797         rwlock_init(&sk->sk_callback_lock);
1798         lockdep_set_class_and_name(&sk->sk_callback_lock,
1799                         af_callback_keys + sk->sk_family,
1800                         af_family_clock_key_strings[sk->sk_family]);
1801
1802         sk->sk_state_change     =       sock_def_wakeup;
1803         sk->sk_data_ready       =       sock_def_readable;
1804         sk->sk_write_space      =       sock_def_write_space;
1805         sk->sk_error_report     =       sock_def_error_report;
1806         sk->sk_destruct         =       sock_def_destruct;
1807
1808         sk->sk_sndmsg_page      =       NULL;
1809         sk->sk_sndmsg_off       =       0;
1810
1811         sk->sk_peercred.pid     =       0;
1812         sk->sk_peercred.uid     =       -1;
1813         sk->sk_peercred.gid     =       -1;
1814         sk->sk_write_pending    =       0;
1815         sk->sk_rcvlowat         =       1;
1816         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1817         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1818
1819         sk->sk_stamp = ktime_set(-1L, 0);
1820
1821         atomic_set(&sk->sk_refcnt, 1);
1822         atomic_set(&sk->sk_drops, 0);
1823 }
1824 EXPORT_SYMBOL(sock_init_data);
1825
1826 void lock_sock_nested(struct sock *sk, int subclass)
1827 {
1828         might_sleep();
1829         spin_lock_bh(&sk->sk_lock.slock);
1830         if (sk->sk_lock.owned)
1831                 __lock_sock(sk);
1832         sk->sk_lock.owned = 1;
1833         spin_unlock(&sk->sk_lock.slock);
1834         /*
1835          * The sk_lock has mutex_lock() semantics here:
1836          */
1837         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1838         local_bh_enable();
1839 }
1840 EXPORT_SYMBOL(lock_sock_nested);
1841
1842 void release_sock(struct sock *sk)
1843 {
1844         /*
1845          * The sk_lock has mutex_unlock() semantics:
1846          */
1847         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1848
1849         spin_lock_bh(&sk->sk_lock.slock);
1850         if (sk->sk_backlog.tail)
1851                 __release_sock(sk);
1852         sk->sk_lock.owned = 0;
1853         if (waitqueue_active(&sk->sk_lock.wq))
1854                 wake_up(&sk->sk_lock.wq);
1855         spin_unlock_bh(&sk->sk_lock.slock);
1856 }
1857 EXPORT_SYMBOL(release_sock);
1858
1859 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1860 {
1861         struct timeval tv;
1862         if (!sock_flag(sk, SOCK_TIMESTAMP))
1863                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1864         tv = ktime_to_timeval(sk->sk_stamp);
1865         if (tv.tv_sec == -1)
1866                 return -ENOENT;
1867         if (tv.tv_sec == 0) {
1868                 sk->sk_stamp = ktime_get_real();
1869                 tv = ktime_to_timeval(sk->sk_stamp);
1870         }
1871         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1872 }
1873 EXPORT_SYMBOL(sock_get_timestamp);
1874
1875 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1876 {
1877         struct timespec ts;
1878         if (!sock_flag(sk, SOCK_TIMESTAMP))
1879                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1880         ts = ktime_to_timespec(sk->sk_stamp);
1881         if (ts.tv_sec == -1)
1882                 return -ENOENT;
1883         if (ts.tv_sec == 0) {
1884                 sk->sk_stamp = ktime_get_real();
1885                 ts = ktime_to_timespec(sk->sk_stamp);
1886         }
1887         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1888 }
1889 EXPORT_SYMBOL(sock_get_timestampns);
1890
1891 void sock_enable_timestamp(struct sock *sk, int flag)
1892 {
1893         if (!sock_flag(sk, flag)) {
1894                 sock_set_flag(sk, flag);
1895                 /*
1896                  * we just set one of the two flags which require net
1897                  * time stamping, but time stamping might have been on
1898                  * already because of the other one
1899                  */
1900                 if (!sock_flag(sk,
1901                                 flag == SOCK_TIMESTAMP ?
1902                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
1903                                 SOCK_TIMESTAMP))
1904                         net_enable_timestamp();
1905         }
1906 }
1907
1908 /*
1909  *      Get a socket option on an socket.
1910  *
1911  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1912  *      asynchronous errors should be reported by getsockopt. We assume
1913  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1914  */
1915 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1916                            char __user *optval, int __user *optlen)
1917 {
1918         struct sock *sk = sock->sk;
1919
1920         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1921 }
1922 EXPORT_SYMBOL(sock_common_getsockopt);
1923
1924 #ifdef CONFIG_COMPAT
1925 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1926                                   char __user *optval, int __user *optlen)
1927 {
1928         struct sock *sk = sock->sk;
1929
1930         if (sk->sk_prot->compat_getsockopt != NULL)
1931                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1932                                                       optval, optlen);
1933         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1934 }
1935 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1936 #endif
1937
1938 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1939                         struct msghdr *msg, size_t size, int flags)
1940 {
1941         struct sock *sk = sock->sk;
1942         int addr_len = 0;
1943         int err;
1944
1945         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1946                                    flags & ~MSG_DONTWAIT, &addr_len);
1947         if (err >= 0)
1948                 msg->msg_namelen = addr_len;
1949         return err;
1950 }
1951 EXPORT_SYMBOL(sock_common_recvmsg);
1952
1953 /*
1954  *      Set socket options on an inet socket.
1955  */
1956 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1957                            char __user *optval, int optlen)
1958 {
1959         struct sock *sk = sock->sk;
1960
1961         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1962 }
1963 EXPORT_SYMBOL(sock_common_setsockopt);
1964
1965 #ifdef CONFIG_COMPAT
1966 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1967                                   char __user *optval, int optlen)
1968 {
1969         struct sock *sk = sock->sk;
1970
1971         if (sk->sk_prot->compat_setsockopt != NULL)
1972                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1973                                                       optval, optlen);
1974         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1975 }
1976 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1977 #endif
1978
1979 void sk_common_release(struct sock *sk)
1980 {
1981         if (sk->sk_prot->destroy)
1982                 sk->sk_prot->destroy(sk);
1983
1984         /*
1985          * Observation: when sock_common_release is called, processes have
1986          * no access to socket. But net still has.
1987          * Step one, detach it from networking:
1988          *
1989          * A. Remove from hash tables.
1990          */
1991
1992         sk->sk_prot->unhash(sk);
1993
1994         /*
1995          * In this point socket cannot receive new packets, but it is possible
1996          * that some packets are in flight because some CPU runs receiver and
1997          * did hash table lookup before we unhashed socket. They will achieve
1998          * receive queue and will be purged by socket destructor.
1999          *
2000          * Also we still have packets pending on receive queue and probably,
2001          * our own packets waiting in device queues. sock_destroy will drain
2002          * receive queue, but transmitted packets will delay socket destruction
2003          * until the last reference will be released.
2004          */
2005
2006         sock_orphan(sk);
2007
2008         xfrm_sk_free_policy(sk);
2009
2010         sk_refcnt_debug_release(sk);
2011         sock_put(sk);
2012 }
2013 EXPORT_SYMBOL(sk_common_release);
2014
2015 static DEFINE_RWLOCK(proto_list_lock);
2016 static LIST_HEAD(proto_list);
2017
2018 #ifdef CONFIG_PROC_FS
2019 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2020 struct prot_inuse {
2021         int val[PROTO_INUSE_NR];
2022 };
2023
2024 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2025
2026 #ifdef CONFIG_NET_NS
2027 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2028 {
2029         int cpu = smp_processor_id();
2030         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2031 }
2032 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2033
2034 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2035 {
2036         int cpu, idx = prot->inuse_idx;
2037         int res = 0;
2038
2039         for_each_possible_cpu(cpu)
2040                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2041
2042         return res >= 0 ? res : 0;
2043 }
2044 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2045
2046 static int sock_inuse_init_net(struct net *net)
2047 {
2048         net->core.inuse = alloc_percpu(struct prot_inuse);
2049         return net->core.inuse ? 0 : -ENOMEM;
2050 }
2051
2052 static void sock_inuse_exit_net(struct net *net)
2053 {
2054         free_percpu(net->core.inuse);
2055 }
2056
2057 static struct pernet_operations net_inuse_ops = {
2058         .init = sock_inuse_init_net,
2059         .exit = sock_inuse_exit_net,
2060 };
2061
2062 static __init int net_inuse_init(void)
2063 {
2064         if (register_pernet_subsys(&net_inuse_ops))
2065                 panic("Cannot initialize net inuse counters");
2066
2067         return 0;
2068 }
2069
2070 core_initcall(net_inuse_init);
2071 #else
2072 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2073
2074 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2075 {
2076         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2077 }
2078 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2079
2080 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2081 {
2082         int cpu, idx = prot->inuse_idx;
2083         int res = 0;
2084
2085         for_each_possible_cpu(cpu)
2086                 res += per_cpu(prot_inuse, cpu).val[idx];
2087
2088         return res >= 0 ? res : 0;
2089 }
2090 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2091 #endif
2092
2093 static void assign_proto_idx(struct proto *prot)
2094 {
2095         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2096
2097         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2098                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2099                 return;
2100         }
2101
2102         set_bit(prot->inuse_idx, proto_inuse_idx);
2103 }
2104
2105 static void release_proto_idx(struct proto *prot)
2106 {
2107         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2108                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2109 }
2110 #else
2111 static inline void assign_proto_idx(struct proto *prot)
2112 {
2113 }
2114
2115 static inline void release_proto_idx(struct proto *prot)
2116 {
2117 }
2118 #endif
2119
2120 int proto_register(struct proto *prot, int alloc_slab)
2121 {
2122         if (alloc_slab) {
2123                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2124                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2125                                         NULL);
2126
2127                 if (prot->slab == NULL) {
2128                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2129                                prot->name);
2130                         goto out;
2131                 }
2132
2133                 if (prot->rsk_prot != NULL) {
2134                         static const char mask[] = "request_sock_%s";
2135
2136                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2137                         if (prot->rsk_prot->slab_name == NULL)
2138                                 goto out_free_sock_slab;
2139
2140                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2141                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2142                                                                  prot->rsk_prot->obj_size, 0,
2143                                                                  SLAB_HWCACHE_ALIGN, NULL);
2144
2145                         if (prot->rsk_prot->slab == NULL) {
2146                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2147                                        prot->name);
2148                                 goto out_free_request_sock_slab_name;
2149                         }
2150                 }
2151
2152                 if (prot->twsk_prot != NULL) {
2153                         static const char mask[] = "tw_sock_%s";
2154
2155                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2156
2157                         if (prot->twsk_prot->twsk_slab_name == NULL)
2158                                 goto out_free_request_sock_slab;
2159
2160                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2161                         prot->twsk_prot->twsk_slab =
2162                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2163                                                   prot->twsk_prot->twsk_obj_size,
2164                                                   0,
2165                                                   SLAB_HWCACHE_ALIGN |
2166                                                         prot->slab_flags,
2167                                                   NULL);
2168                         if (prot->twsk_prot->twsk_slab == NULL)
2169                                 goto out_free_timewait_sock_slab_name;
2170                 }
2171         }
2172
2173         write_lock(&proto_list_lock);
2174         list_add(&prot->node, &proto_list);
2175         assign_proto_idx(prot);
2176         write_unlock(&proto_list_lock);
2177         return 0;
2178
2179 out_free_timewait_sock_slab_name:
2180         kfree(prot->twsk_prot->twsk_slab_name);
2181 out_free_request_sock_slab:
2182         if (prot->rsk_prot && prot->rsk_prot->slab) {
2183                 kmem_cache_destroy(prot->rsk_prot->slab);
2184                 prot->rsk_prot->slab = NULL;
2185         }
2186 out_free_request_sock_slab_name:
2187         kfree(prot->rsk_prot->slab_name);
2188 out_free_sock_slab:
2189         kmem_cache_destroy(prot->slab);
2190         prot->slab = NULL;
2191 out:
2192         return -ENOBUFS;
2193 }
2194 EXPORT_SYMBOL(proto_register);
2195
2196 void proto_unregister(struct proto *prot)
2197 {
2198         write_lock(&proto_list_lock);
2199         release_proto_idx(prot);
2200         list_del(&prot->node);
2201         write_unlock(&proto_list_lock);
2202
2203         if (prot->slab != NULL) {
2204                 kmem_cache_destroy(prot->slab);
2205                 prot->slab = NULL;
2206         }
2207
2208         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2209                 kmem_cache_destroy(prot->rsk_prot->slab);
2210                 kfree(prot->rsk_prot->slab_name);
2211                 prot->rsk_prot->slab = NULL;
2212         }
2213
2214         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2215                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2216                 kfree(prot->twsk_prot->twsk_slab_name);
2217                 prot->twsk_prot->twsk_slab = NULL;
2218         }
2219 }
2220 EXPORT_SYMBOL(proto_unregister);
2221
2222 #ifdef CONFIG_PROC_FS
2223 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2224         __acquires(proto_list_lock)
2225 {
2226         read_lock(&proto_list_lock);
2227         return seq_list_start_head(&proto_list, *pos);
2228 }
2229
2230 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2231 {
2232         return seq_list_next(v, &proto_list, pos);
2233 }
2234
2235 static void proto_seq_stop(struct seq_file *seq, void *v)
2236         __releases(proto_list_lock)
2237 {
2238         read_unlock(&proto_list_lock);
2239 }
2240
2241 static char proto_method_implemented(const void *method)
2242 {
2243         return method == NULL ? 'n' : 'y';
2244 }
2245
2246 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2247 {
2248         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2249                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2250                    proto->name,
2251                    proto->obj_size,
2252                    sock_prot_inuse_get(seq_file_net(seq), proto),
2253                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2254                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2255                    proto->max_header,
2256                    proto->slab == NULL ? "no" : "yes",
2257                    module_name(proto->owner),
2258                    proto_method_implemented(proto->close),
2259                    proto_method_implemented(proto->connect),
2260                    proto_method_implemented(proto->disconnect),
2261                    proto_method_implemented(proto->accept),
2262                    proto_method_implemented(proto->ioctl),
2263                    proto_method_implemented(proto->init),
2264                    proto_method_implemented(proto->destroy),
2265                    proto_method_implemented(proto->shutdown),
2266                    proto_method_implemented(proto->setsockopt),
2267                    proto_method_implemented(proto->getsockopt),
2268                    proto_method_implemented(proto->sendmsg),
2269                    proto_method_implemented(proto->recvmsg),
2270                    proto_method_implemented(proto->sendpage),
2271                    proto_method_implemented(proto->bind),
2272                    proto_method_implemented(proto->backlog_rcv),
2273                    proto_method_implemented(proto->hash),
2274                    proto_method_implemented(proto->unhash),
2275                    proto_method_implemented(proto->get_port),
2276                    proto_method_implemented(proto->enter_memory_pressure));
2277 }
2278
2279 static int proto_seq_show(struct seq_file *seq, void *v)
2280 {
2281         if (v == &proto_list)
2282                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2283                            "protocol",
2284                            "size",
2285                            "sockets",
2286                            "memory",
2287                            "press",
2288                            "maxhdr",
2289                            "slab",
2290                            "module",
2291                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2292         else
2293                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2294         return 0;
2295 }
2296
2297 static const struct seq_operations proto_seq_ops = {
2298         .start  = proto_seq_start,
2299         .next   = proto_seq_next,
2300         .stop   = proto_seq_stop,
2301         .show   = proto_seq_show,
2302 };
2303
2304 static int proto_seq_open(struct inode *inode, struct file *file)
2305 {
2306         return seq_open_net(inode, file, &proto_seq_ops,
2307                             sizeof(struct seq_net_private));
2308 }
2309
2310 static const struct file_operations proto_seq_fops = {
2311         .owner          = THIS_MODULE,
2312         .open           = proto_seq_open,
2313         .read           = seq_read,
2314         .llseek         = seq_lseek,
2315         .release        = seq_release_net,
2316 };
2317
2318 static __net_init int proto_init_net(struct net *net)
2319 {
2320         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2321                 return -ENOMEM;
2322
2323         return 0;
2324 }
2325
2326 static __net_exit void proto_exit_net(struct net *net)
2327 {
2328         proc_net_remove(net, "protocols");
2329 }
2330
2331
2332 static __net_initdata struct pernet_operations proto_net_ops = {
2333         .init = proto_init_net,
2334         .exit = proto_exit_net,
2335 };
2336
2337 static int __init proto_init(void)
2338 {
2339         return register_pernet_subsys(&proto_net_ops);
2340 }
2341
2342 subsys_initcall(proto_init);
2343
2344 #endif /* PROC_FS */