Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
[linux-2.6] / 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_MAX"
159 };
160 static const char *af_family_slock_key_strings[AF_MAX+1] = {
161   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
162   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
163   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
164   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
165   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
166   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
167   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
168   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
169   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
170   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
171   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
172   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
173   "slock-AF_MAX"
174 };
175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
176   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
177   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
178   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
179   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
180   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
181   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
182   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
183   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
184   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
185   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
186   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
187   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
188   "clock-AF_MAX"
189 };
190
191 /*
192  * sk_callback_lock locking rules are per-address-family,
193  * so split the lock classes by using a per-AF key:
194  */
195 static struct lock_class_key af_callback_keys[AF_MAX];
196
197 /* Take into consideration the size of the struct sk_buff overhead in the
198  * determination of these values, since that is non-constant across
199  * platforms.  This makes socket queueing behavior and performance
200  * not depend upon such differences.
201  */
202 #define _SK_MEM_PACKETS         256
203 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
204 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
206
207 /* Run time adjustable parameters. */
208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
212
213 /* Maximal space eaten by iovec or ancilliary data plus some space */
214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
215
216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
217 {
218         struct timeval tv;
219
220         if (optlen < sizeof(tv))
221                 return -EINVAL;
222         if (copy_from_user(&tv, optval, sizeof(tv)))
223                 return -EFAULT;
224         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
225                 return -EDOM;
226
227         if (tv.tv_sec < 0) {
228                 static int warned __read_mostly;
229
230                 *timeo_p = 0;
231                 if (warned < 10 && net_ratelimit()) {
232                         warned++;
233                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
234                                "tries to set negative timeout\n",
235                                 current->comm, task_pid_nr(current));
236                 }
237                 return 0;
238         }
239         *timeo_p = MAX_SCHEDULE_TIMEOUT;
240         if (tv.tv_sec == 0 && tv.tv_usec == 0)
241                 return 0;
242         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
243                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
244         return 0;
245 }
246
247 static void sock_warn_obsolete_bsdism(const char *name)
248 {
249         static int warned;
250         static char warncomm[TASK_COMM_LEN];
251         if (strcmp(warncomm, current->comm) && warned < 5) {
252                 strcpy(warncomm,  current->comm);
253                 printk(KERN_WARNING "process `%s' is using obsolete "
254                        "%s SO_BSDCOMPAT\n", warncomm, name);
255                 warned++;
256         }
257 }
258
259 static void sock_disable_timestamp(struct sock *sk, int flag)
260 {
261         if (sock_flag(sk, flag)) {
262                 sock_reset_flag(sk, flag);
263                 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
264                     !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
265                         net_disable_timestamp();
266                 }
267         }
268 }
269
270
271 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
272 {
273         int err = 0;
274         int skb_len;
275
276         /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
277            number of warnings when compiling with -W --ANK
278          */
279         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
280             (unsigned)sk->sk_rcvbuf) {
281                 err = -ENOMEM;
282                 goto out;
283         }
284
285         err = sk_filter(sk, skb);
286         if (err)
287                 goto out;
288
289         if (!sk_rmem_schedule(sk, skb->truesize)) {
290                 err = -ENOBUFS;
291                 goto out;
292         }
293
294         skb->dev = NULL;
295         skb_set_owner_r(skb, sk);
296
297         /* Cache the SKB length before we tack it onto the receive
298          * queue.  Once it is added it no longer belongs to us and
299          * may be freed by other threads of control pulling packets
300          * from the queue.
301          */
302         skb_len = skb->len;
303
304         skb_queue_tail(&sk->sk_receive_queue, skb);
305
306         if (!sock_flag(sk, SOCK_DEAD))
307                 sk->sk_data_ready(sk, skb_len);
308 out:
309         return err;
310 }
311 EXPORT_SYMBOL(sock_queue_rcv_skb);
312
313 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
314 {
315         int rc = NET_RX_SUCCESS;
316
317         if (sk_filter(sk, skb))
318                 goto discard_and_relse;
319
320         skb->dev = NULL;
321
322         if (nested)
323                 bh_lock_sock_nested(sk);
324         else
325                 bh_lock_sock(sk);
326         if (!sock_owned_by_user(sk)) {
327                 /*
328                  * trylock + unlock semantics:
329                  */
330                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
331
332                 rc = sk_backlog_rcv(sk, skb);
333
334                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
335         } else
336                 sk_add_backlog(sk, skb);
337         bh_unlock_sock(sk);
338 out:
339         sock_put(sk);
340         return rc;
341 discard_and_relse:
342         kfree_skb(skb);
343         goto out;
344 }
345 EXPORT_SYMBOL(sk_receive_skb);
346
347 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
348 {
349         struct dst_entry *dst = sk->sk_dst_cache;
350
351         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
352                 sk->sk_dst_cache = NULL;
353                 dst_release(dst);
354                 return NULL;
355         }
356
357         return dst;
358 }
359 EXPORT_SYMBOL(__sk_dst_check);
360
361 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
362 {
363         struct dst_entry *dst = sk_dst_get(sk);
364
365         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
366                 sk_dst_reset(sk);
367                 dst_release(dst);
368                 return NULL;
369         }
370
371         return dst;
372 }
373 EXPORT_SYMBOL(sk_dst_check);
374
375 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
376 {
377         int ret = -ENOPROTOOPT;
378 #ifdef CONFIG_NETDEVICES
379         struct net *net = sock_net(sk);
380         char devname[IFNAMSIZ];
381         int index;
382
383         /* Sorry... */
384         ret = -EPERM;
385         if (!capable(CAP_NET_RAW))
386                 goto out;
387
388         ret = -EINVAL;
389         if (optlen < 0)
390                 goto out;
391
392         /* Bind this socket to a particular device like "eth0",
393          * as specified in the passed interface name. If the
394          * name is "" or the option length is zero the socket
395          * is not bound.
396          */
397         if (optlen > IFNAMSIZ - 1)
398                 optlen = IFNAMSIZ - 1;
399         memset(devname, 0, sizeof(devname));
400
401         ret = -EFAULT;
402         if (copy_from_user(devname, optval, optlen))
403                 goto out;
404
405         if (devname[0] == '\0') {
406                 index = 0;
407         } else {
408                 struct net_device *dev = dev_get_by_name(net, devname);
409
410                 ret = -ENODEV;
411                 if (!dev)
412                         goto out;
413
414                 index = dev->ifindex;
415                 dev_put(dev);
416         }
417
418         lock_sock(sk);
419         sk->sk_bound_dev_if = index;
420         sk_dst_reset(sk);
421         release_sock(sk);
422
423         ret = 0;
424
425 out:
426 #endif
427
428         return ret;
429 }
430
431 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
432 {
433         if (valbool)
434                 sock_set_flag(sk, bit);
435         else
436                 sock_reset_flag(sk, bit);
437 }
438
439 /*
440  *      This is meant for all protocols to use and covers goings on
441  *      at the socket level. Everything here is generic.
442  */
443
444 int sock_setsockopt(struct socket *sock, int level, int optname,
445                     char __user *optval, int optlen)
446 {
447         struct sock *sk=sock->sk;
448         int val;
449         int valbool;
450         struct linger ling;
451         int ret = 0;
452
453         /*
454          *      Options without arguments
455          */
456
457         if (optname == SO_BINDTODEVICE)
458                 return sock_bindtodevice(sk, optval, optlen);
459
460         if (optlen < sizeof(int))
461                 return -EINVAL;
462
463         if (get_user(val, (int __user *)optval))
464                 return -EFAULT;
465
466         valbool = val?1:0;
467
468         lock_sock(sk);
469
470         switch(optname) {
471         case SO_DEBUG:
472                 if (val && !capable(CAP_NET_ADMIN)) {
473                         ret = -EACCES;
474                 } else
475                         sock_valbool_flag(sk, SOCK_DBG, valbool);
476                 break;
477         case SO_REUSEADDR:
478                 sk->sk_reuse = valbool;
479                 break;
480         case SO_TYPE:
481         case SO_ERROR:
482                 ret = -ENOPROTOOPT;
483                 break;
484         case SO_DONTROUTE:
485                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
486                 break;
487         case SO_BROADCAST:
488                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
489                 break;
490         case SO_SNDBUF:
491                 /* Don't error on this BSD doesn't and if you think
492                    about it this is right. Otherwise apps have to
493                    play 'guess the biggest size' games. RCVBUF/SNDBUF
494                    are treated in BSD as hints */
495
496                 if (val > sysctl_wmem_max)
497                         val = sysctl_wmem_max;
498 set_sndbuf:
499                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
500                 if ((val * 2) < SOCK_MIN_SNDBUF)
501                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
502                 else
503                         sk->sk_sndbuf = val * 2;
504
505                 /*
506                  *      Wake up sending tasks if we
507                  *      upped the value.
508                  */
509                 sk->sk_write_space(sk);
510                 break;
511
512         case SO_SNDBUFFORCE:
513                 if (!capable(CAP_NET_ADMIN)) {
514                         ret = -EPERM;
515                         break;
516                 }
517                 goto set_sndbuf;
518
519         case SO_RCVBUF:
520                 /* Don't error on this BSD doesn't and if you think
521                    about it this is right. Otherwise apps have to
522                    play 'guess the biggest size' games. RCVBUF/SNDBUF
523                    are treated in BSD as hints */
524
525                 if (val > sysctl_rmem_max)
526                         val = sysctl_rmem_max;
527 set_rcvbuf:
528                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
529                 /*
530                  * We double it on the way in to account for
531                  * "struct sk_buff" etc. overhead.   Applications
532                  * assume that the SO_RCVBUF setting they make will
533                  * allow that much actual data to be received on that
534                  * socket.
535                  *
536                  * Applications are unaware that "struct sk_buff" and
537                  * other overheads allocate from the receive buffer
538                  * during socket buffer allocation.
539                  *
540                  * And after considering the possible alternatives,
541                  * returning the value we actually used in getsockopt
542                  * is the most desirable behavior.
543                  */
544                 if ((val * 2) < SOCK_MIN_RCVBUF)
545                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
546                 else
547                         sk->sk_rcvbuf = val * 2;
548                 break;
549
550         case SO_RCVBUFFORCE:
551                 if (!capable(CAP_NET_ADMIN)) {
552                         ret = -EPERM;
553                         break;
554                 }
555                 goto set_rcvbuf;
556
557         case SO_KEEPALIVE:
558 #ifdef CONFIG_INET
559                 if (sk->sk_protocol == IPPROTO_TCP)
560                         tcp_set_keepalive(sk, valbool);
561 #endif
562                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
563                 break;
564
565         case SO_OOBINLINE:
566                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
567                 break;
568
569         case SO_NO_CHECK:
570                 sk->sk_no_check = valbool;
571                 break;
572
573         case SO_PRIORITY:
574                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
575                         sk->sk_priority = val;
576                 else
577                         ret = -EPERM;
578                 break;
579
580         case SO_LINGER:
581                 if (optlen < sizeof(ling)) {
582                         ret = -EINVAL;  /* 1003.1g */
583                         break;
584                 }
585                 if (copy_from_user(&ling,optval,sizeof(ling))) {
586                         ret = -EFAULT;
587                         break;
588                 }
589                 if (!ling.l_onoff)
590                         sock_reset_flag(sk, SOCK_LINGER);
591                 else {
592 #if (BITS_PER_LONG == 32)
593                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
594                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
595                         else
596 #endif
597                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
598                         sock_set_flag(sk, SOCK_LINGER);
599                 }
600                 break;
601
602         case SO_BSDCOMPAT:
603                 sock_warn_obsolete_bsdism("setsockopt");
604                 break;
605
606         case SO_PASSCRED:
607                 if (valbool)
608                         set_bit(SOCK_PASSCRED, &sock->flags);
609                 else
610                         clear_bit(SOCK_PASSCRED, &sock->flags);
611                 break;
612
613         case SO_TIMESTAMP:
614         case SO_TIMESTAMPNS:
615                 if (valbool)  {
616                         if (optname == SO_TIMESTAMP)
617                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
618                         else
619                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
620                         sock_set_flag(sk, SOCK_RCVTSTAMP);
621                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
622                 } else {
623                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
624                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
625                 }
626                 break;
627
628         case SO_TIMESTAMPING:
629                 if (val & ~SOF_TIMESTAMPING_MASK) {
630                         ret = EINVAL;
631                         break;
632                 }
633                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
634                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
635                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
636                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
637                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
638                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
639                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
640                         sock_enable_timestamp(sk,
641                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
642                 else
643                         sock_disable_timestamp(sk,
644                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
645                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
646                                   val & SOF_TIMESTAMPING_SOFTWARE);
647                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
648                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
649                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
650                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
651                 break;
652
653         case SO_RCVLOWAT:
654                 if (val < 0)
655                         val = INT_MAX;
656                 sk->sk_rcvlowat = val ? : 1;
657                 break;
658
659         case SO_RCVTIMEO:
660                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
661                 break;
662
663         case SO_SNDTIMEO:
664                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
665                 break;
666
667         case SO_ATTACH_FILTER:
668                 ret = -EINVAL;
669                 if (optlen == sizeof(struct sock_fprog)) {
670                         struct sock_fprog fprog;
671
672                         ret = -EFAULT;
673                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
674                                 break;
675
676                         ret = sk_attach_filter(&fprog, sk);
677                 }
678                 break;
679
680         case SO_DETACH_FILTER:
681                 ret = sk_detach_filter(sk);
682                 break;
683
684         case SO_PASSSEC:
685                 if (valbool)
686                         set_bit(SOCK_PASSSEC, &sock->flags);
687                 else
688                         clear_bit(SOCK_PASSSEC, &sock->flags);
689                 break;
690         case SO_MARK:
691                 if (!capable(CAP_NET_ADMIN))
692                         ret = -EPERM;
693                 else {
694                         sk->sk_mark = val;
695                 }
696                 break;
697
698                 /* We implement the SO_SNDLOWAT etc to
699                    not be settable (1003.1g 5.3) */
700         default:
701                 ret = -ENOPROTOOPT;
702                 break;
703         }
704         release_sock(sk);
705         return ret;
706 }
707
708
709 int sock_getsockopt(struct socket *sock, int level, int optname,
710                     char __user *optval, int __user *optlen)
711 {
712         struct sock *sk = sock->sk;
713
714         union {
715                 int val;
716                 struct linger ling;
717                 struct timeval tm;
718         } v;
719
720         unsigned int lv = sizeof(int);
721         int len;
722
723         if (get_user(len, optlen))
724                 return -EFAULT;
725         if (len < 0)
726                 return -EINVAL;
727
728         memset(&v, 0, sizeof(v));
729
730         switch(optname) {
731         case SO_DEBUG:
732                 v.val = sock_flag(sk, SOCK_DBG);
733                 break;
734
735         case SO_DONTROUTE:
736                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
737                 break;
738
739         case SO_BROADCAST:
740                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
741                 break;
742
743         case SO_SNDBUF:
744                 v.val = sk->sk_sndbuf;
745                 break;
746
747         case SO_RCVBUF:
748                 v.val = sk->sk_rcvbuf;
749                 break;
750
751         case SO_REUSEADDR:
752                 v.val = sk->sk_reuse;
753                 break;
754
755         case SO_KEEPALIVE:
756                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
757                 break;
758
759         case SO_TYPE:
760                 v.val = sk->sk_type;
761                 break;
762
763         case SO_ERROR:
764                 v.val = -sock_error(sk);
765                 if (v.val==0)
766                         v.val = xchg(&sk->sk_err_soft, 0);
767                 break;
768
769         case SO_OOBINLINE:
770                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
771                 break;
772
773         case SO_NO_CHECK:
774                 v.val = sk->sk_no_check;
775                 break;
776
777         case SO_PRIORITY:
778                 v.val = sk->sk_priority;
779                 break;
780
781         case SO_LINGER:
782                 lv              = sizeof(v.ling);
783                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
784                 v.ling.l_linger = sk->sk_lingertime / HZ;
785                 break;
786
787         case SO_BSDCOMPAT:
788                 sock_warn_obsolete_bsdism("getsockopt");
789                 break;
790
791         case SO_TIMESTAMP:
792                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
793                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
794                 break;
795
796         case SO_TIMESTAMPNS:
797                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
798                 break;
799
800         case SO_TIMESTAMPING:
801                 v.val = 0;
802                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
803                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
804                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
805                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
806                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
807                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
808                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
809                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
810                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
811                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
812                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
813                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
814                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
815                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
816                 break;
817
818         case SO_RCVTIMEO:
819                 lv=sizeof(struct timeval);
820                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
821                         v.tm.tv_sec = 0;
822                         v.tm.tv_usec = 0;
823                 } else {
824                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
825                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
826                 }
827                 break;
828
829         case SO_SNDTIMEO:
830                 lv=sizeof(struct timeval);
831                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
832                         v.tm.tv_sec = 0;
833                         v.tm.tv_usec = 0;
834                 } else {
835                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
836                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
837                 }
838                 break;
839
840         case SO_RCVLOWAT:
841                 v.val = sk->sk_rcvlowat;
842                 break;
843
844         case SO_SNDLOWAT:
845                 v.val=1;
846                 break;
847
848         case SO_PASSCRED:
849                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
850                 break;
851
852         case SO_PEERCRED:
853                 if (len > sizeof(sk->sk_peercred))
854                         len = sizeof(sk->sk_peercred);
855                 if (copy_to_user(optval, &sk->sk_peercred, len))
856                         return -EFAULT;
857                 goto lenout;
858
859         case SO_PEERNAME:
860         {
861                 char address[128];
862
863                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
864                         return -ENOTCONN;
865                 if (lv < len)
866                         return -EINVAL;
867                 if (copy_to_user(optval, address, len))
868                         return -EFAULT;
869                 goto lenout;
870         }
871
872         /* Dubious BSD thing... Probably nobody even uses it, but
873          * the UNIX standard wants it for whatever reason... -DaveM
874          */
875         case SO_ACCEPTCONN:
876                 v.val = sk->sk_state == TCP_LISTEN;
877                 break;
878
879         case SO_PASSSEC:
880                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
881                 break;
882
883         case SO_PEERSEC:
884                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
885
886         case SO_MARK:
887                 v.val = sk->sk_mark;
888                 break;
889
890         default:
891                 return -ENOPROTOOPT;
892         }
893
894         if (len > lv)
895                 len = lv;
896         if (copy_to_user(optval, &v, len))
897                 return -EFAULT;
898 lenout:
899         if (put_user(len, optlen))
900                 return -EFAULT;
901         return 0;
902 }
903
904 /*
905  * Initialize an sk_lock.
906  *
907  * (We also register the sk_lock with the lock validator.)
908  */
909 static inline void sock_lock_init(struct sock *sk)
910 {
911         sock_lock_init_class_and_name(sk,
912                         af_family_slock_key_strings[sk->sk_family],
913                         af_family_slock_keys + sk->sk_family,
914                         af_family_key_strings[sk->sk_family],
915                         af_family_keys + sk->sk_family);
916 }
917
918 static void sock_copy(struct sock *nsk, const struct sock *osk)
919 {
920 #ifdef CONFIG_SECURITY_NETWORK
921         void *sptr = nsk->sk_security;
922 #endif
923
924         memcpy(nsk, osk, osk->sk_prot->obj_size);
925 #ifdef CONFIG_SECURITY_NETWORK
926         nsk->sk_security = sptr;
927         security_sk_clone(osk, nsk);
928 #endif
929 }
930
931 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
932                 int family)
933 {
934         struct sock *sk;
935         struct kmem_cache *slab;
936
937         slab = prot->slab;
938         if (slab != NULL)
939                 sk = kmem_cache_alloc(slab, priority);
940         else
941                 sk = kmalloc(prot->obj_size, priority);
942
943         if (sk != NULL) {
944                 if (security_sk_alloc(sk, family, priority))
945                         goto out_free;
946
947                 if (!try_module_get(prot->owner))
948                         goto out_free_sec;
949         }
950
951         return sk;
952
953 out_free_sec:
954         security_sk_free(sk);
955 out_free:
956         if (slab != NULL)
957                 kmem_cache_free(slab, sk);
958         else
959                 kfree(sk);
960         return NULL;
961 }
962
963 static void sk_prot_free(struct proto *prot, struct sock *sk)
964 {
965         struct kmem_cache *slab;
966         struct module *owner;
967
968         owner = prot->owner;
969         slab = prot->slab;
970
971         security_sk_free(sk);
972         if (slab != NULL)
973                 kmem_cache_free(slab, sk);
974         else
975                 kfree(sk);
976         module_put(owner);
977 }
978
979 /**
980  *      sk_alloc - All socket objects are allocated here
981  *      @net: the applicable net namespace
982  *      @family: protocol family
983  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
984  *      @prot: struct proto associated with this new sock instance
985  */
986 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
987                       struct proto *prot)
988 {
989         struct sock *sk;
990
991         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
992         if (sk) {
993                 sk->sk_family = family;
994                 /*
995                  * See comment in struct sock definition to understand
996                  * why we need sk_prot_creator -acme
997                  */
998                 sk->sk_prot = sk->sk_prot_creator = prot;
999                 sock_lock_init(sk);
1000                 sock_net_set(sk, get_net(net));
1001         }
1002
1003         return sk;
1004 }
1005
1006 void sk_free(struct sock *sk)
1007 {
1008         struct sk_filter *filter;
1009
1010         if (sk->sk_destruct)
1011                 sk->sk_destruct(sk);
1012
1013         filter = rcu_dereference(sk->sk_filter);
1014         if (filter) {
1015                 sk_filter_uncharge(sk, filter);
1016                 rcu_assign_pointer(sk->sk_filter, NULL);
1017         }
1018
1019         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1020         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1021
1022         if (atomic_read(&sk->sk_omem_alloc))
1023                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1024                        __func__, atomic_read(&sk->sk_omem_alloc));
1025
1026         put_net(sock_net(sk));
1027         sk_prot_free(sk->sk_prot_creator, sk);
1028 }
1029
1030 /*
1031  * Last sock_put should drop referrence to sk->sk_net. It has already
1032  * been dropped in sk_change_net. Taking referrence to stopping namespace
1033  * is not an option.
1034  * Take referrence to a socket to remove it from hash _alive_ and after that
1035  * destroy it in the context of init_net.
1036  */
1037 void sk_release_kernel(struct sock *sk)
1038 {
1039         if (sk == NULL || sk->sk_socket == NULL)
1040                 return;
1041
1042         sock_hold(sk);
1043         sock_release(sk->sk_socket);
1044         release_net(sock_net(sk));
1045         sock_net_set(sk, get_net(&init_net));
1046         sock_put(sk);
1047 }
1048 EXPORT_SYMBOL(sk_release_kernel);
1049
1050 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1051 {
1052         struct sock *newsk;
1053
1054         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1055         if (newsk != NULL) {
1056                 struct sk_filter *filter;
1057
1058                 sock_copy(newsk, sk);
1059
1060                 /* SANITY */
1061                 get_net(sock_net(newsk));
1062                 sk_node_init(&newsk->sk_node);
1063                 sock_lock_init(newsk);
1064                 bh_lock_sock(newsk);
1065                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1066
1067                 atomic_set(&newsk->sk_rmem_alloc, 0);
1068                 atomic_set(&newsk->sk_wmem_alloc, 0);
1069                 atomic_set(&newsk->sk_omem_alloc, 0);
1070                 skb_queue_head_init(&newsk->sk_receive_queue);
1071                 skb_queue_head_init(&newsk->sk_write_queue);
1072 #ifdef CONFIG_NET_DMA
1073                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1074 #endif
1075
1076                 rwlock_init(&newsk->sk_dst_lock);
1077                 rwlock_init(&newsk->sk_callback_lock);
1078                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1079                                 af_callback_keys + newsk->sk_family,
1080                                 af_family_clock_key_strings[newsk->sk_family]);
1081
1082                 newsk->sk_dst_cache     = NULL;
1083                 newsk->sk_wmem_queued   = 0;
1084                 newsk->sk_forward_alloc = 0;
1085                 newsk->sk_send_head     = NULL;
1086                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1087
1088                 sock_reset_flag(newsk, SOCK_DONE);
1089                 skb_queue_head_init(&newsk->sk_error_queue);
1090
1091                 filter = newsk->sk_filter;
1092                 if (filter != NULL)
1093                         sk_filter_charge(newsk, filter);
1094
1095                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1096                         /* It is still raw copy of parent, so invalidate
1097                          * destructor and make plain sk_free() */
1098                         newsk->sk_destruct = NULL;
1099                         sk_free(newsk);
1100                         newsk = NULL;
1101                         goto out;
1102                 }
1103
1104                 newsk->sk_err      = 0;
1105                 newsk->sk_priority = 0;
1106                 atomic_set(&newsk->sk_refcnt, 2);
1107
1108                 /*
1109                  * Increment the counter in the same struct proto as the master
1110                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1111                  * is the same as sk->sk_prot->socks, as this field was copied
1112                  * with memcpy).
1113                  *
1114                  * This _changes_ the previous behaviour, where
1115                  * tcp_create_openreq_child always was incrementing the
1116                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1117                  * to be taken into account in all callers. -acme
1118                  */
1119                 sk_refcnt_debug_inc(newsk);
1120                 sk_set_socket(newsk, NULL);
1121                 newsk->sk_sleep  = NULL;
1122
1123                 if (newsk->sk_prot->sockets_allocated)
1124                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1125         }
1126 out:
1127         return newsk;
1128 }
1129
1130 EXPORT_SYMBOL_GPL(sk_clone);
1131
1132 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1133 {
1134         __sk_dst_set(sk, dst);
1135         sk->sk_route_caps = dst->dev->features;
1136         if (sk->sk_route_caps & NETIF_F_GSO)
1137                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1138         if (sk_can_gso(sk)) {
1139                 if (dst->header_len) {
1140                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1141                 } else {
1142                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1143                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1144                 }
1145         }
1146 }
1147 EXPORT_SYMBOL_GPL(sk_setup_caps);
1148
1149 void __init sk_init(void)
1150 {
1151         if (num_physpages <= 4096) {
1152                 sysctl_wmem_max = 32767;
1153                 sysctl_rmem_max = 32767;
1154                 sysctl_wmem_default = 32767;
1155                 sysctl_rmem_default = 32767;
1156         } else if (num_physpages >= 131072) {
1157                 sysctl_wmem_max = 131071;
1158                 sysctl_rmem_max = 131071;
1159         }
1160 }
1161
1162 /*
1163  *      Simple resource managers for sockets.
1164  */
1165
1166
1167 /*
1168  * Write buffer destructor automatically called from kfree_skb.
1169  */
1170 void sock_wfree(struct sk_buff *skb)
1171 {
1172         struct sock *sk = skb->sk;
1173
1174         /* In case it might be waiting for more memory. */
1175         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1176         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1177                 sk->sk_write_space(sk);
1178         sock_put(sk);
1179 }
1180
1181 /*
1182  * Read buffer destructor automatically called from kfree_skb.
1183  */
1184 void sock_rfree(struct sk_buff *skb)
1185 {
1186         struct sock *sk = skb->sk;
1187
1188         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1189         sk_mem_uncharge(skb->sk, skb->truesize);
1190 }
1191
1192
1193 int sock_i_uid(struct sock *sk)
1194 {
1195         int uid;
1196
1197         read_lock(&sk->sk_callback_lock);
1198         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1199         read_unlock(&sk->sk_callback_lock);
1200         return uid;
1201 }
1202
1203 unsigned long sock_i_ino(struct sock *sk)
1204 {
1205         unsigned long ino;
1206
1207         read_lock(&sk->sk_callback_lock);
1208         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1209         read_unlock(&sk->sk_callback_lock);
1210         return ino;
1211 }
1212
1213 /*
1214  * Allocate a skb from the socket's send buffer.
1215  */
1216 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1217                              gfp_t priority)
1218 {
1219         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1220                 struct sk_buff * skb = alloc_skb(size, priority);
1221                 if (skb) {
1222                         skb_set_owner_w(skb, sk);
1223                         return skb;
1224                 }
1225         }
1226         return NULL;
1227 }
1228
1229 /*
1230  * Allocate a skb from the socket's receive buffer.
1231  */
1232 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1233                              gfp_t priority)
1234 {
1235         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1236                 struct sk_buff *skb = alloc_skb(size, priority);
1237                 if (skb) {
1238                         skb_set_owner_r(skb, sk);
1239                         return skb;
1240                 }
1241         }
1242         return NULL;
1243 }
1244
1245 /*
1246  * Allocate a memory block from the socket's option memory buffer.
1247  */
1248 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1249 {
1250         if ((unsigned)size <= sysctl_optmem_max &&
1251             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1252                 void *mem;
1253                 /* First do the add, to avoid the race if kmalloc
1254                  * might sleep.
1255                  */
1256                 atomic_add(size, &sk->sk_omem_alloc);
1257                 mem = kmalloc(size, priority);
1258                 if (mem)
1259                         return mem;
1260                 atomic_sub(size, &sk->sk_omem_alloc);
1261         }
1262         return NULL;
1263 }
1264
1265 /*
1266  * Free an option memory block.
1267  */
1268 void sock_kfree_s(struct sock *sk, void *mem, int size)
1269 {
1270         kfree(mem);
1271         atomic_sub(size, &sk->sk_omem_alloc);
1272 }
1273
1274 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1275    I think, these locks should be removed for datagram sockets.
1276  */
1277 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1278 {
1279         DEFINE_WAIT(wait);
1280
1281         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1282         for (;;) {
1283                 if (!timeo)
1284                         break;
1285                 if (signal_pending(current))
1286                         break;
1287                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1288                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1289                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1290                         break;
1291                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1292                         break;
1293                 if (sk->sk_err)
1294                         break;
1295                 timeo = schedule_timeout(timeo);
1296         }
1297         finish_wait(sk->sk_sleep, &wait);
1298         return timeo;
1299 }
1300
1301
1302 /*
1303  *      Generic send/receive buffer handlers
1304  */
1305
1306 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1307                                      unsigned long data_len, int noblock,
1308                                      int *errcode)
1309 {
1310         struct sk_buff *skb;
1311         gfp_t gfp_mask;
1312         long timeo;
1313         int err;
1314
1315         gfp_mask = sk->sk_allocation;
1316         if (gfp_mask & __GFP_WAIT)
1317                 gfp_mask |= __GFP_REPEAT;
1318
1319         timeo = sock_sndtimeo(sk, noblock);
1320         while (1) {
1321                 err = sock_error(sk);
1322                 if (err != 0)
1323                         goto failure;
1324
1325                 err = -EPIPE;
1326                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1327                         goto failure;
1328
1329                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1330                         skb = alloc_skb(header_len, gfp_mask);
1331                         if (skb) {
1332                                 int npages;
1333                                 int i;
1334
1335                                 /* No pages, we're done... */
1336                                 if (!data_len)
1337                                         break;
1338
1339                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1340                                 skb->truesize += data_len;
1341                                 skb_shinfo(skb)->nr_frags = npages;
1342                                 for (i = 0; i < npages; i++) {
1343                                         struct page *page;
1344                                         skb_frag_t *frag;
1345
1346                                         page = alloc_pages(sk->sk_allocation, 0);
1347                                         if (!page) {
1348                                                 err = -ENOBUFS;
1349                                                 skb_shinfo(skb)->nr_frags = i;
1350                                                 kfree_skb(skb);
1351                                                 goto failure;
1352                                         }
1353
1354                                         frag = &skb_shinfo(skb)->frags[i];
1355                                         frag->page = page;
1356                                         frag->page_offset = 0;
1357                                         frag->size = (data_len >= PAGE_SIZE ?
1358                                                       PAGE_SIZE :
1359                                                       data_len);
1360                                         data_len -= PAGE_SIZE;
1361                                 }
1362
1363                                 /* Full success... */
1364                                 break;
1365                         }
1366                         err = -ENOBUFS;
1367                         goto failure;
1368                 }
1369                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1370                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1371                 err = -EAGAIN;
1372                 if (!timeo)
1373                         goto failure;
1374                 if (signal_pending(current))
1375                         goto interrupted;
1376                 timeo = sock_wait_for_wmem(sk, timeo);
1377         }
1378
1379         skb_set_owner_w(skb, sk);
1380         return skb;
1381
1382 interrupted:
1383         err = sock_intr_errno(timeo);
1384 failure:
1385         *errcode = err;
1386         return NULL;
1387 }
1388 EXPORT_SYMBOL(sock_alloc_send_pskb);
1389
1390 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1391                                     int noblock, int *errcode)
1392 {
1393         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1394 }
1395
1396 static void __lock_sock(struct sock *sk)
1397 {
1398         DEFINE_WAIT(wait);
1399
1400         for (;;) {
1401                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1402                                         TASK_UNINTERRUPTIBLE);
1403                 spin_unlock_bh(&sk->sk_lock.slock);
1404                 schedule();
1405                 spin_lock_bh(&sk->sk_lock.slock);
1406                 if (!sock_owned_by_user(sk))
1407                         break;
1408         }
1409         finish_wait(&sk->sk_lock.wq, &wait);
1410 }
1411
1412 static void __release_sock(struct sock *sk)
1413 {
1414         struct sk_buff *skb = sk->sk_backlog.head;
1415
1416         do {
1417                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1418                 bh_unlock_sock(sk);
1419
1420                 do {
1421                         struct sk_buff *next = skb->next;
1422
1423                         skb->next = NULL;
1424                         sk_backlog_rcv(sk, skb);
1425
1426                         /*
1427                          * We are in process context here with softirqs
1428                          * disabled, use cond_resched_softirq() to preempt.
1429                          * This is safe to do because we've taken the backlog
1430                          * queue private:
1431                          */
1432                         cond_resched_softirq();
1433
1434                         skb = next;
1435                 } while (skb != NULL);
1436
1437                 bh_lock_sock(sk);
1438         } while ((skb = sk->sk_backlog.head) != NULL);
1439 }
1440
1441 /**
1442  * sk_wait_data - wait for data to arrive at sk_receive_queue
1443  * @sk:    sock to wait on
1444  * @timeo: for how long
1445  *
1446  * Now socket state including sk->sk_err is changed only under lock,
1447  * hence we may omit checks after joining wait queue.
1448  * We check receive queue before schedule() only as optimization;
1449  * it is very likely that release_sock() added new data.
1450  */
1451 int sk_wait_data(struct sock *sk, long *timeo)
1452 {
1453         int rc;
1454         DEFINE_WAIT(wait);
1455
1456         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1457         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1458         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1459         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1460         finish_wait(sk->sk_sleep, &wait);
1461         return rc;
1462 }
1463
1464 EXPORT_SYMBOL(sk_wait_data);
1465
1466 /**
1467  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1468  *      @sk: socket
1469  *      @size: memory size to allocate
1470  *      @kind: allocation type
1471  *
1472  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1473  *      rmem allocation. This function assumes that protocols which have
1474  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1475  */
1476 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1477 {
1478         struct proto *prot = sk->sk_prot;
1479         int amt = sk_mem_pages(size);
1480         int allocated;
1481
1482         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1483         allocated = atomic_add_return(amt, prot->memory_allocated);
1484
1485         /* Under limit. */
1486         if (allocated <= prot->sysctl_mem[0]) {
1487                 if (prot->memory_pressure && *prot->memory_pressure)
1488                         *prot->memory_pressure = 0;
1489                 return 1;
1490         }
1491
1492         /* Under pressure. */
1493         if (allocated > prot->sysctl_mem[1])
1494                 if (prot->enter_memory_pressure)
1495                         prot->enter_memory_pressure(sk);
1496
1497         /* Over hard limit. */
1498         if (allocated > prot->sysctl_mem[2])
1499                 goto suppress_allocation;
1500
1501         /* guarantee minimum buffer size under pressure */
1502         if (kind == SK_MEM_RECV) {
1503                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1504                         return 1;
1505         } else { /* SK_MEM_SEND */
1506                 if (sk->sk_type == SOCK_STREAM) {
1507                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1508                                 return 1;
1509                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1510                            prot->sysctl_wmem[0])
1511                                 return 1;
1512         }
1513
1514         if (prot->memory_pressure) {
1515                 int alloc;
1516
1517                 if (!*prot->memory_pressure)
1518                         return 1;
1519                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1520                 if (prot->sysctl_mem[2] > alloc *
1521                     sk_mem_pages(sk->sk_wmem_queued +
1522                                  atomic_read(&sk->sk_rmem_alloc) +
1523                                  sk->sk_forward_alloc))
1524                         return 1;
1525         }
1526
1527 suppress_allocation:
1528
1529         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1530                 sk_stream_moderate_sndbuf(sk);
1531
1532                 /* Fail only if socket is _under_ its sndbuf.
1533                  * In this case we cannot block, so that we have to fail.
1534                  */
1535                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1536                         return 1;
1537         }
1538
1539         /* Alas. Undo changes. */
1540         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1541         atomic_sub(amt, prot->memory_allocated);
1542         return 0;
1543 }
1544
1545 EXPORT_SYMBOL(__sk_mem_schedule);
1546
1547 /**
1548  *      __sk_reclaim - reclaim memory_allocated
1549  *      @sk: socket
1550  */
1551 void __sk_mem_reclaim(struct sock *sk)
1552 {
1553         struct proto *prot = sk->sk_prot;
1554
1555         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1556                    prot->memory_allocated);
1557         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1558
1559         if (prot->memory_pressure && *prot->memory_pressure &&
1560             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1561                 *prot->memory_pressure = 0;
1562 }
1563
1564 EXPORT_SYMBOL(__sk_mem_reclaim);
1565
1566
1567 /*
1568  * Set of default routines for initialising struct proto_ops when
1569  * the protocol does not support a particular function. In certain
1570  * cases where it makes no sense for a protocol to have a "do nothing"
1571  * function, some default processing is provided.
1572  */
1573
1574 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1575 {
1576         return -EOPNOTSUPP;
1577 }
1578
1579 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1580                     int len, int flags)
1581 {
1582         return -EOPNOTSUPP;
1583 }
1584
1585 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1586 {
1587         return -EOPNOTSUPP;
1588 }
1589
1590 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1591 {
1592         return -EOPNOTSUPP;
1593 }
1594
1595 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1596                     int *len, int peer)
1597 {
1598         return -EOPNOTSUPP;
1599 }
1600
1601 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1602 {
1603         return 0;
1604 }
1605
1606 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1607 {
1608         return -EOPNOTSUPP;
1609 }
1610
1611 int sock_no_listen(struct socket *sock, int backlog)
1612 {
1613         return -EOPNOTSUPP;
1614 }
1615
1616 int sock_no_shutdown(struct socket *sock, int how)
1617 {
1618         return -EOPNOTSUPP;
1619 }
1620
1621 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1622                     char __user *optval, int optlen)
1623 {
1624         return -EOPNOTSUPP;
1625 }
1626
1627 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1628                     char __user *optval, int __user *optlen)
1629 {
1630         return -EOPNOTSUPP;
1631 }
1632
1633 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1634                     size_t len)
1635 {
1636         return -EOPNOTSUPP;
1637 }
1638
1639 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1640                     size_t len, int flags)
1641 {
1642         return -EOPNOTSUPP;
1643 }
1644
1645 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1646 {
1647         /* Mirror missing mmap method error code */
1648         return -ENODEV;
1649 }
1650
1651 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1652 {
1653         ssize_t res;
1654         struct msghdr msg = {.msg_flags = flags};
1655         struct kvec iov;
1656         char *kaddr = kmap(page);
1657         iov.iov_base = kaddr + offset;
1658         iov.iov_len = size;
1659         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1660         kunmap(page);
1661         return res;
1662 }
1663
1664 /*
1665  *      Default Socket Callbacks
1666  */
1667
1668 static void sock_def_wakeup(struct sock *sk)
1669 {
1670         read_lock(&sk->sk_callback_lock);
1671         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1672                 wake_up_interruptible_all(sk->sk_sleep);
1673         read_unlock(&sk->sk_callback_lock);
1674 }
1675
1676 static void sock_def_error_report(struct sock *sk)
1677 {
1678         read_lock(&sk->sk_callback_lock);
1679         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1680                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1681         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1682         read_unlock(&sk->sk_callback_lock);
1683 }
1684
1685 static void sock_def_readable(struct sock *sk, int len)
1686 {
1687         read_lock(&sk->sk_callback_lock);
1688         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1689                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1690                                                 POLLRDNORM | POLLRDBAND);
1691         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1692         read_unlock(&sk->sk_callback_lock);
1693 }
1694
1695 static void sock_def_write_space(struct sock *sk)
1696 {
1697         read_lock(&sk->sk_callback_lock);
1698
1699         /* Do not wake up a writer until he can make "significant"
1700          * progress.  --DaveM
1701          */
1702         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1703                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1704                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1705                                                 POLLWRNORM | POLLWRBAND);
1706
1707                 /* Should agree with poll, otherwise some programs break */
1708                 if (sock_writeable(sk))
1709                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1710         }
1711
1712         read_unlock(&sk->sk_callback_lock);
1713 }
1714
1715 static void sock_def_destruct(struct sock *sk)
1716 {
1717         kfree(sk->sk_protinfo);
1718 }
1719
1720 void sk_send_sigurg(struct sock *sk)
1721 {
1722         if (sk->sk_socket && sk->sk_socket->file)
1723                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1724                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1725 }
1726
1727 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1728                     unsigned long expires)
1729 {
1730         if (!mod_timer(timer, expires))
1731                 sock_hold(sk);
1732 }
1733
1734 EXPORT_SYMBOL(sk_reset_timer);
1735
1736 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1737 {
1738         if (timer_pending(timer) && del_timer(timer))
1739                 __sock_put(sk);
1740 }
1741
1742 EXPORT_SYMBOL(sk_stop_timer);
1743
1744 void sock_init_data(struct socket *sock, struct sock *sk)
1745 {
1746         skb_queue_head_init(&sk->sk_receive_queue);
1747         skb_queue_head_init(&sk->sk_write_queue);
1748         skb_queue_head_init(&sk->sk_error_queue);
1749 #ifdef CONFIG_NET_DMA
1750         skb_queue_head_init(&sk->sk_async_wait_queue);
1751 #endif
1752
1753         sk->sk_send_head        =       NULL;
1754
1755         init_timer(&sk->sk_timer);
1756
1757         sk->sk_allocation       =       GFP_KERNEL;
1758         sk->sk_rcvbuf           =       sysctl_rmem_default;
1759         sk->sk_sndbuf           =       sysctl_wmem_default;
1760         sk->sk_state            =       TCP_CLOSE;
1761         sk_set_socket(sk, sock);
1762
1763         sock_set_flag(sk, SOCK_ZAPPED);
1764
1765         if (sock) {
1766                 sk->sk_type     =       sock->type;
1767                 sk->sk_sleep    =       &sock->wait;
1768                 sock->sk        =       sk;
1769         } else
1770                 sk->sk_sleep    =       NULL;
1771
1772         rwlock_init(&sk->sk_dst_lock);
1773         rwlock_init(&sk->sk_callback_lock);
1774         lockdep_set_class_and_name(&sk->sk_callback_lock,
1775                         af_callback_keys + sk->sk_family,
1776                         af_family_clock_key_strings[sk->sk_family]);
1777
1778         sk->sk_state_change     =       sock_def_wakeup;
1779         sk->sk_data_ready       =       sock_def_readable;
1780         sk->sk_write_space      =       sock_def_write_space;
1781         sk->sk_error_report     =       sock_def_error_report;
1782         sk->sk_destruct         =       sock_def_destruct;
1783
1784         sk->sk_sndmsg_page      =       NULL;
1785         sk->sk_sndmsg_off       =       0;
1786
1787         sk->sk_peercred.pid     =       0;
1788         sk->sk_peercred.uid     =       -1;
1789         sk->sk_peercred.gid     =       -1;
1790         sk->sk_write_pending    =       0;
1791         sk->sk_rcvlowat         =       1;
1792         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1793         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1794
1795         sk->sk_stamp = ktime_set(-1L, 0);
1796
1797         atomic_set(&sk->sk_refcnt, 1);
1798         atomic_set(&sk->sk_drops, 0);
1799 }
1800
1801 void lock_sock_nested(struct sock *sk, int subclass)
1802 {
1803         might_sleep();
1804         spin_lock_bh(&sk->sk_lock.slock);
1805         if (sk->sk_lock.owned)
1806                 __lock_sock(sk);
1807         sk->sk_lock.owned = 1;
1808         spin_unlock(&sk->sk_lock.slock);
1809         /*
1810          * The sk_lock has mutex_lock() semantics here:
1811          */
1812         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1813         local_bh_enable();
1814 }
1815
1816 EXPORT_SYMBOL(lock_sock_nested);
1817
1818 void release_sock(struct sock *sk)
1819 {
1820         /*
1821          * The sk_lock has mutex_unlock() semantics:
1822          */
1823         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1824
1825         spin_lock_bh(&sk->sk_lock.slock);
1826         if (sk->sk_backlog.tail)
1827                 __release_sock(sk);
1828         sk->sk_lock.owned = 0;
1829         if (waitqueue_active(&sk->sk_lock.wq))
1830                 wake_up(&sk->sk_lock.wq);
1831         spin_unlock_bh(&sk->sk_lock.slock);
1832 }
1833 EXPORT_SYMBOL(release_sock);
1834
1835 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1836 {
1837         struct timeval tv;
1838         if (!sock_flag(sk, SOCK_TIMESTAMP))
1839                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1840         tv = ktime_to_timeval(sk->sk_stamp);
1841         if (tv.tv_sec == -1)
1842                 return -ENOENT;
1843         if (tv.tv_sec == 0) {
1844                 sk->sk_stamp = ktime_get_real();
1845                 tv = ktime_to_timeval(sk->sk_stamp);
1846         }
1847         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1848 }
1849 EXPORT_SYMBOL(sock_get_timestamp);
1850
1851 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1852 {
1853         struct timespec ts;
1854         if (!sock_flag(sk, SOCK_TIMESTAMP))
1855                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1856         ts = ktime_to_timespec(sk->sk_stamp);
1857         if (ts.tv_sec == -1)
1858                 return -ENOENT;
1859         if (ts.tv_sec == 0) {
1860                 sk->sk_stamp = ktime_get_real();
1861                 ts = ktime_to_timespec(sk->sk_stamp);
1862         }
1863         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1864 }
1865 EXPORT_SYMBOL(sock_get_timestampns);
1866
1867 void sock_enable_timestamp(struct sock *sk, int flag)
1868 {
1869         if (!sock_flag(sk, flag)) {
1870                 sock_set_flag(sk, flag);
1871                 /*
1872                  * we just set one of the two flags which require net
1873                  * time stamping, but time stamping might have been on
1874                  * already because of the other one
1875                  */
1876                 if (!sock_flag(sk,
1877                                 flag == SOCK_TIMESTAMP ?
1878                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
1879                                 SOCK_TIMESTAMP))
1880                         net_enable_timestamp();
1881         }
1882 }
1883
1884 /*
1885  *      Get a socket option on an socket.
1886  *
1887  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1888  *      asynchronous errors should be reported by getsockopt. We assume
1889  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1890  */
1891 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1892                            char __user *optval, int __user *optlen)
1893 {
1894         struct sock *sk = sock->sk;
1895
1896         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1897 }
1898
1899 EXPORT_SYMBOL(sock_common_getsockopt);
1900
1901 #ifdef CONFIG_COMPAT
1902 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1903                                   char __user *optval, int __user *optlen)
1904 {
1905         struct sock *sk = sock->sk;
1906
1907         if (sk->sk_prot->compat_getsockopt != NULL)
1908                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1909                                                       optval, optlen);
1910         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1911 }
1912 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1913 #endif
1914
1915 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1916                         struct msghdr *msg, size_t size, int flags)
1917 {
1918         struct sock *sk = sock->sk;
1919         int addr_len = 0;
1920         int err;
1921
1922         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1923                                    flags & ~MSG_DONTWAIT, &addr_len);
1924         if (err >= 0)
1925                 msg->msg_namelen = addr_len;
1926         return err;
1927 }
1928
1929 EXPORT_SYMBOL(sock_common_recvmsg);
1930
1931 /*
1932  *      Set socket options on an inet socket.
1933  */
1934 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1935                            char __user *optval, int optlen)
1936 {
1937         struct sock *sk = sock->sk;
1938
1939         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1940 }
1941
1942 EXPORT_SYMBOL(sock_common_setsockopt);
1943
1944 #ifdef CONFIG_COMPAT
1945 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1946                                   char __user *optval, int optlen)
1947 {
1948         struct sock *sk = sock->sk;
1949
1950         if (sk->sk_prot->compat_setsockopt != NULL)
1951                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1952                                                       optval, optlen);
1953         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1954 }
1955 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1956 #endif
1957
1958 void sk_common_release(struct sock *sk)
1959 {
1960         if (sk->sk_prot->destroy)
1961                 sk->sk_prot->destroy(sk);
1962
1963         /*
1964          * Observation: when sock_common_release is called, processes have
1965          * no access to socket. But net still has.
1966          * Step one, detach it from networking:
1967          *
1968          * A. Remove from hash tables.
1969          */
1970
1971         sk->sk_prot->unhash(sk);
1972
1973         /*
1974          * In this point socket cannot receive new packets, but it is possible
1975          * that some packets are in flight because some CPU runs receiver and
1976          * did hash table lookup before we unhashed socket. They will achieve
1977          * receive queue and will be purged by socket destructor.
1978          *
1979          * Also we still have packets pending on receive queue and probably,
1980          * our own packets waiting in device queues. sock_destroy will drain
1981          * receive queue, but transmitted packets will delay socket destruction
1982          * until the last reference will be released.
1983          */
1984
1985         sock_orphan(sk);
1986
1987         xfrm_sk_free_policy(sk);
1988
1989         sk_refcnt_debug_release(sk);
1990         sock_put(sk);
1991 }
1992
1993 EXPORT_SYMBOL(sk_common_release);
1994
1995 static DEFINE_RWLOCK(proto_list_lock);
1996 static LIST_HEAD(proto_list);
1997
1998 #ifdef CONFIG_PROC_FS
1999 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2000 struct prot_inuse {
2001         int val[PROTO_INUSE_NR];
2002 };
2003
2004 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2005
2006 #ifdef CONFIG_NET_NS
2007 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2008 {
2009         int cpu = smp_processor_id();
2010         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2011 }
2012 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2013
2014 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2015 {
2016         int cpu, idx = prot->inuse_idx;
2017         int res = 0;
2018
2019         for_each_possible_cpu(cpu)
2020                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2021
2022         return res >= 0 ? res : 0;
2023 }
2024 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2025
2026 static int sock_inuse_init_net(struct net *net)
2027 {
2028         net->core.inuse = alloc_percpu(struct prot_inuse);
2029         return net->core.inuse ? 0 : -ENOMEM;
2030 }
2031
2032 static void sock_inuse_exit_net(struct net *net)
2033 {
2034         free_percpu(net->core.inuse);
2035 }
2036
2037 static struct pernet_operations net_inuse_ops = {
2038         .init = sock_inuse_init_net,
2039         .exit = sock_inuse_exit_net,
2040 };
2041
2042 static __init int net_inuse_init(void)
2043 {
2044         if (register_pernet_subsys(&net_inuse_ops))
2045                 panic("Cannot initialize net inuse counters");
2046
2047         return 0;
2048 }
2049
2050 core_initcall(net_inuse_init);
2051 #else
2052 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2053
2054 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2055 {
2056         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2057 }
2058 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2059
2060 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2061 {
2062         int cpu, idx = prot->inuse_idx;
2063         int res = 0;
2064
2065         for_each_possible_cpu(cpu)
2066                 res += per_cpu(prot_inuse, cpu).val[idx];
2067
2068         return res >= 0 ? res : 0;
2069 }
2070 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2071 #endif
2072
2073 static void assign_proto_idx(struct proto *prot)
2074 {
2075         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2076
2077         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2078                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2079                 return;
2080         }
2081
2082         set_bit(prot->inuse_idx, proto_inuse_idx);
2083 }
2084
2085 static void release_proto_idx(struct proto *prot)
2086 {
2087         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2088                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2089 }
2090 #else
2091 static inline void assign_proto_idx(struct proto *prot)
2092 {
2093 }
2094
2095 static inline void release_proto_idx(struct proto *prot)
2096 {
2097 }
2098 #endif
2099
2100 int proto_register(struct proto *prot, int alloc_slab)
2101 {
2102         if (alloc_slab) {
2103                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2104                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2105                                         NULL);
2106
2107                 if (prot->slab == NULL) {
2108                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2109                                prot->name);
2110                         goto out;
2111                 }
2112
2113                 if (prot->rsk_prot != NULL) {
2114                         static const char mask[] = "request_sock_%s";
2115
2116                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2117                         if (prot->rsk_prot->slab_name == NULL)
2118                                 goto out_free_sock_slab;
2119
2120                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2121                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2122                                                                  prot->rsk_prot->obj_size, 0,
2123                                                                  SLAB_HWCACHE_ALIGN, NULL);
2124
2125                         if (prot->rsk_prot->slab == NULL) {
2126                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2127                                        prot->name);
2128                                 goto out_free_request_sock_slab_name;
2129                         }
2130                 }
2131
2132                 if (prot->twsk_prot != NULL) {
2133                         static const char mask[] = "tw_sock_%s";
2134
2135                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2136
2137                         if (prot->twsk_prot->twsk_slab_name == NULL)
2138                                 goto out_free_request_sock_slab;
2139
2140                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2141                         prot->twsk_prot->twsk_slab =
2142                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2143                                                   prot->twsk_prot->twsk_obj_size,
2144                                                   0,
2145                                                   SLAB_HWCACHE_ALIGN |
2146                                                         prot->slab_flags,
2147                                                   NULL);
2148                         if (prot->twsk_prot->twsk_slab == NULL)
2149                                 goto out_free_timewait_sock_slab_name;
2150                 }
2151         }
2152
2153         write_lock(&proto_list_lock);
2154         list_add(&prot->node, &proto_list);
2155         assign_proto_idx(prot);
2156         write_unlock(&proto_list_lock);
2157         return 0;
2158
2159 out_free_timewait_sock_slab_name:
2160         kfree(prot->twsk_prot->twsk_slab_name);
2161 out_free_request_sock_slab:
2162         if (prot->rsk_prot && prot->rsk_prot->slab) {
2163                 kmem_cache_destroy(prot->rsk_prot->slab);
2164                 prot->rsk_prot->slab = NULL;
2165         }
2166 out_free_request_sock_slab_name:
2167         kfree(prot->rsk_prot->slab_name);
2168 out_free_sock_slab:
2169         kmem_cache_destroy(prot->slab);
2170         prot->slab = NULL;
2171 out:
2172         return -ENOBUFS;
2173 }
2174
2175 EXPORT_SYMBOL(proto_register);
2176
2177 void proto_unregister(struct proto *prot)
2178 {
2179         write_lock(&proto_list_lock);
2180         release_proto_idx(prot);
2181         list_del(&prot->node);
2182         write_unlock(&proto_list_lock);
2183
2184         if (prot->slab != NULL) {
2185                 kmem_cache_destroy(prot->slab);
2186                 prot->slab = NULL;
2187         }
2188
2189         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2190                 kmem_cache_destroy(prot->rsk_prot->slab);
2191                 kfree(prot->rsk_prot->slab_name);
2192                 prot->rsk_prot->slab = NULL;
2193         }
2194
2195         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2196                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2197                 kfree(prot->twsk_prot->twsk_slab_name);
2198                 prot->twsk_prot->twsk_slab = NULL;
2199         }
2200 }
2201
2202 EXPORT_SYMBOL(proto_unregister);
2203
2204 #ifdef CONFIG_PROC_FS
2205 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2206         __acquires(proto_list_lock)
2207 {
2208         read_lock(&proto_list_lock);
2209         return seq_list_start_head(&proto_list, *pos);
2210 }
2211
2212 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2213 {
2214         return seq_list_next(v, &proto_list, pos);
2215 }
2216
2217 static void proto_seq_stop(struct seq_file *seq, void *v)
2218         __releases(proto_list_lock)
2219 {
2220         read_unlock(&proto_list_lock);
2221 }
2222
2223 static char proto_method_implemented(const void *method)
2224 {
2225         return method == NULL ? 'n' : 'y';
2226 }
2227
2228 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2229 {
2230         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2231                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2232                    proto->name,
2233                    proto->obj_size,
2234                    sock_prot_inuse_get(seq_file_net(seq), proto),
2235                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2236                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2237                    proto->max_header,
2238                    proto->slab == NULL ? "no" : "yes",
2239                    module_name(proto->owner),
2240                    proto_method_implemented(proto->close),
2241                    proto_method_implemented(proto->connect),
2242                    proto_method_implemented(proto->disconnect),
2243                    proto_method_implemented(proto->accept),
2244                    proto_method_implemented(proto->ioctl),
2245                    proto_method_implemented(proto->init),
2246                    proto_method_implemented(proto->destroy),
2247                    proto_method_implemented(proto->shutdown),
2248                    proto_method_implemented(proto->setsockopt),
2249                    proto_method_implemented(proto->getsockopt),
2250                    proto_method_implemented(proto->sendmsg),
2251                    proto_method_implemented(proto->recvmsg),
2252                    proto_method_implemented(proto->sendpage),
2253                    proto_method_implemented(proto->bind),
2254                    proto_method_implemented(proto->backlog_rcv),
2255                    proto_method_implemented(proto->hash),
2256                    proto_method_implemented(proto->unhash),
2257                    proto_method_implemented(proto->get_port),
2258                    proto_method_implemented(proto->enter_memory_pressure));
2259 }
2260
2261 static int proto_seq_show(struct seq_file *seq, void *v)
2262 {
2263         if (v == &proto_list)
2264                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2265                            "protocol",
2266                            "size",
2267                            "sockets",
2268                            "memory",
2269                            "press",
2270                            "maxhdr",
2271                            "slab",
2272                            "module",
2273                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2274         else
2275                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2276         return 0;
2277 }
2278
2279 static const struct seq_operations proto_seq_ops = {
2280         .start  = proto_seq_start,
2281         .next   = proto_seq_next,
2282         .stop   = proto_seq_stop,
2283         .show   = proto_seq_show,
2284 };
2285
2286 static int proto_seq_open(struct inode *inode, struct file *file)
2287 {
2288         return seq_open_net(inode, file, &proto_seq_ops,
2289                             sizeof(struct seq_net_private));
2290 }
2291
2292 static const struct file_operations proto_seq_fops = {
2293         .owner          = THIS_MODULE,
2294         .open           = proto_seq_open,
2295         .read           = seq_read,
2296         .llseek         = seq_lseek,
2297         .release        = seq_release_net,
2298 };
2299
2300 static __net_init int proto_init_net(struct net *net)
2301 {
2302         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2303                 return -ENOMEM;
2304
2305         return 0;
2306 }
2307
2308 static __net_exit void proto_exit_net(struct net *net)
2309 {
2310         proc_net_remove(net, "protocols");
2311 }
2312
2313
2314 static __net_initdata struct pernet_operations proto_net_ops = {
2315         .init = proto_init_net,
2316         .exit = proto_exit_net,
2317 };
2318
2319 static int __init proto_init(void)
2320 {
2321         return register_pernet_subsys(&proto_net_ops);
2322 }
2323
2324 subsys_initcall(proto_init);
2325
2326 #endif /* PROC_FS */
2327
2328 EXPORT_SYMBOL(sk_alloc);
2329 EXPORT_SYMBOL(sk_free);
2330 EXPORT_SYMBOL(sk_send_sigurg);
2331 EXPORT_SYMBOL(sock_alloc_send_skb);
2332 EXPORT_SYMBOL(sock_init_data);
2333 EXPORT_SYMBOL(sock_kfree_s);
2334 EXPORT_SYMBOL(sock_kmalloc);
2335 EXPORT_SYMBOL(sock_no_accept);
2336 EXPORT_SYMBOL(sock_no_bind);
2337 EXPORT_SYMBOL(sock_no_connect);
2338 EXPORT_SYMBOL(sock_no_getname);
2339 EXPORT_SYMBOL(sock_no_getsockopt);
2340 EXPORT_SYMBOL(sock_no_ioctl);
2341 EXPORT_SYMBOL(sock_no_listen);
2342 EXPORT_SYMBOL(sock_no_mmap);
2343 EXPORT_SYMBOL(sock_no_poll);
2344 EXPORT_SYMBOL(sock_no_recvmsg);
2345 EXPORT_SYMBOL(sock_no_sendmsg);
2346 EXPORT_SYMBOL(sock_no_sendpage);
2347 EXPORT_SYMBOL(sock_no_setsockopt);
2348 EXPORT_SYMBOL(sock_no_shutdown);
2349 EXPORT_SYMBOL(sock_no_socketpair);
2350 EXPORT_SYMBOL(sock_rfree);
2351 EXPORT_SYMBOL(sock_setsockopt);
2352 EXPORT_SYMBOL(sock_wfree);
2353 EXPORT_SYMBOL(sock_wmalloc);
2354 EXPORT_SYMBOL(sock_i_uid);
2355 EXPORT_SYMBOL(sock_i_ino);
2356 EXPORT_SYMBOL(sysctl_optmem_max);