Merge branch 'master' of /home/davem/src/GIT/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 <net/xfrm.h>
124 #include <linux/ipsec.h>
125
126 #include <linux/filter.h>
127
128 #ifdef CONFIG_INET
129 #include <net/tcp.h>
130 #endif
131
132 /*
133  * Each address family might have different locking rules, so we have
134  * one slock key per address family:
135  */
136 static struct lock_class_key af_family_keys[AF_MAX];
137 static struct lock_class_key af_family_slock_keys[AF_MAX];
138
139 /*
140  * Make lock validator output more readable. (we pre-construct these
141  * strings build-time, so that runtime initialization of socket
142  * locks is fast):
143  */
144 static const char *af_family_key_strings[AF_MAX+1] = {
145   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
146   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
147   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
148   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
149   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
150   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
151   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
152   "sk_lock-21"       , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
153   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
154   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
155   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
156   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
157   "sk_lock-AF_MAX"
158 };
159 static const char *af_family_slock_key_strings[AF_MAX+1] = {
160   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
161   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
162   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
163   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
164   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
165   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
166   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
167   "slock-21"       , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
168   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
169   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
170   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
171   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
172   "slock-AF_MAX"
173 };
174 static const char *af_family_clock_key_strings[AF_MAX+1] = {
175   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
176   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
177   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
178   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
179   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
180   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
181   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
182   "clock-21"       , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
183   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
184   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
185   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
186   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
187   "clock-AF_MAX"
188 };
189
190 /*
191  * sk_callback_lock locking rules are per-address-family,
192  * so split the lock classes by using a per-AF key:
193  */
194 static struct lock_class_key af_callback_keys[AF_MAX];
195
196 /* Take into consideration the size of the struct sk_buff overhead in the
197  * determination of these values, since that is non-constant across
198  * platforms.  This makes socket queueing behavior and performance
199  * not depend upon such differences.
200  */
201 #define _SK_MEM_PACKETS         256
202 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
203 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
204 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
205
206 /* Run time adjustable parameters. */
207 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
208 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
209 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
210 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
211
212 /* Maximal space eaten by iovec or ancilliary data plus some space */
213 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
214
215 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
216 {
217         struct timeval tv;
218
219         if (optlen < sizeof(tv))
220                 return -EINVAL;
221         if (copy_from_user(&tv, optval, sizeof(tv)))
222                 return -EFAULT;
223         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
224                 return -EDOM;
225
226         if (tv.tv_sec < 0) {
227                 static int warned __read_mostly;
228
229                 *timeo_p = 0;
230                 if (warned < 10 && net_ratelimit()) {
231                         warned++;
232                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
233                                "tries to set negative timeout\n",
234                                 current->comm, task_pid_nr(current));
235                 }
236                 return 0;
237         }
238         *timeo_p = MAX_SCHEDULE_TIMEOUT;
239         if (tv.tv_sec == 0 && tv.tv_usec == 0)
240                 return 0;
241         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
242                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
243         return 0;
244 }
245
246 static void sock_warn_obsolete_bsdism(const char *name)
247 {
248         static int warned;
249         static char warncomm[TASK_COMM_LEN];
250         if (strcmp(warncomm, current->comm) && warned < 5) {
251                 strcpy(warncomm,  current->comm);
252                 printk(KERN_WARNING "process `%s' is using obsolete "
253                        "%s SO_BSDCOMPAT\n", warncomm, name);
254                 warned++;
255         }
256 }
257
258 static void sock_disable_timestamp(struct sock *sk)
259 {
260         if (sock_flag(sk, SOCK_TIMESTAMP)) {
261                 sock_reset_flag(sk, SOCK_TIMESTAMP);
262                 net_disable_timestamp();
263         }
264 }
265
266
267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
268 {
269         int err = 0;
270         int skb_len;
271
272         /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
273            number of warnings when compiling with -W --ANK
274          */
275         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
276             (unsigned)sk->sk_rcvbuf) {
277                 err = -ENOMEM;
278                 goto out;
279         }
280
281         err = sk_filter(sk, skb);
282         if (err)
283                 goto out;
284
285         if (!sk_rmem_schedule(sk, skb->truesize)) {
286                 err = -ENOBUFS;
287                 goto out;
288         }
289
290         skb->dev = NULL;
291         skb_set_owner_r(skb, sk);
292
293         /* Cache the SKB length before we tack it onto the receive
294          * queue.  Once it is added it no longer belongs to us and
295          * may be freed by other threads of control pulling packets
296          * from the queue.
297          */
298         skb_len = skb->len;
299
300         skb_queue_tail(&sk->sk_receive_queue, skb);
301
302         if (!sock_flag(sk, SOCK_DEAD))
303                 sk->sk_data_ready(sk, skb_len);
304 out:
305         return err;
306 }
307 EXPORT_SYMBOL(sock_queue_rcv_skb);
308
309 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
310 {
311         int rc = NET_RX_SUCCESS;
312
313         if (sk_filter(sk, skb))
314                 goto discard_and_relse;
315
316         skb->dev = NULL;
317
318         if (nested)
319                 bh_lock_sock_nested(sk);
320         else
321                 bh_lock_sock(sk);
322         if (!sock_owned_by_user(sk)) {
323                 /*
324                  * trylock + unlock semantics:
325                  */
326                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
327
328                 rc = sk_backlog_rcv(sk, skb);
329
330                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
331         } else
332                 sk_add_backlog(sk, skb);
333         bh_unlock_sock(sk);
334 out:
335         sock_put(sk);
336         return rc;
337 discard_and_relse:
338         kfree_skb(skb);
339         goto out;
340 }
341 EXPORT_SYMBOL(sk_receive_skb);
342
343 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
344 {
345         struct dst_entry *dst = sk->sk_dst_cache;
346
347         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
348                 sk->sk_dst_cache = NULL;
349                 dst_release(dst);
350                 return NULL;
351         }
352
353         return dst;
354 }
355 EXPORT_SYMBOL(__sk_dst_check);
356
357 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
358 {
359         struct dst_entry *dst = sk_dst_get(sk);
360
361         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
362                 sk_dst_reset(sk);
363                 dst_release(dst);
364                 return NULL;
365         }
366
367         return dst;
368 }
369 EXPORT_SYMBOL(sk_dst_check);
370
371 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
372 {
373         int ret = -ENOPROTOOPT;
374 #ifdef CONFIG_NETDEVICES
375         struct net *net = sock_net(sk);
376         char devname[IFNAMSIZ];
377         int index;
378
379         /* Sorry... */
380         ret = -EPERM;
381         if (!capable(CAP_NET_RAW))
382                 goto out;
383
384         ret = -EINVAL;
385         if (optlen < 0)
386                 goto out;
387
388         /* Bind this socket to a particular device like "eth0",
389          * as specified in the passed interface name. If the
390          * name is "" or the option length is zero the socket
391          * is not bound.
392          */
393         if (optlen > IFNAMSIZ - 1)
394                 optlen = IFNAMSIZ - 1;
395         memset(devname, 0, sizeof(devname));
396
397         ret = -EFAULT;
398         if (copy_from_user(devname, optval, optlen))
399                 goto out;
400
401         if (devname[0] == '\0') {
402                 index = 0;
403         } else {
404                 struct net_device *dev = dev_get_by_name(net, devname);
405
406                 ret = -ENODEV;
407                 if (!dev)
408                         goto out;
409
410                 index = dev->ifindex;
411                 dev_put(dev);
412         }
413
414         lock_sock(sk);
415         sk->sk_bound_dev_if = index;
416         sk_dst_reset(sk);
417         release_sock(sk);
418
419         ret = 0;
420
421 out:
422 #endif
423
424         return ret;
425 }
426
427 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
428 {
429         if (valbool)
430                 sock_set_flag(sk, bit);
431         else
432                 sock_reset_flag(sk, bit);
433 }
434
435 /*
436  *      This is meant for all protocols to use and covers goings on
437  *      at the socket level. Everything here is generic.
438  */
439
440 int sock_setsockopt(struct socket *sock, int level, int optname,
441                     char __user *optval, int optlen)
442 {
443         struct sock *sk=sock->sk;
444         int val;
445         int valbool;
446         struct linger ling;
447         int ret = 0;
448
449         /*
450          *      Options without arguments
451          */
452
453         if (optname == SO_BINDTODEVICE)
454                 return sock_bindtodevice(sk, optval, optlen);
455
456         if (optlen < sizeof(int))
457                 return -EINVAL;
458
459         if (get_user(val, (int __user *)optval))
460                 return -EFAULT;
461
462         valbool = val?1:0;
463
464         lock_sock(sk);
465
466         switch(optname) {
467         case SO_DEBUG:
468                 if (val && !capable(CAP_NET_ADMIN)) {
469                         ret = -EACCES;
470                 } else
471                         sock_valbool_flag(sk, SOCK_DBG, valbool);
472                 break;
473         case SO_REUSEADDR:
474                 sk->sk_reuse = valbool;
475                 break;
476         case SO_TYPE:
477         case SO_ERROR:
478                 ret = -ENOPROTOOPT;
479                 break;
480         case SO_DONTROUTE:
481                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
482                 break;
483         case SO_BROADCAST:
484                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
485                 break;
486         case SO_SNDBUF:
487                 /* Don't error on this BSD doesn't and if you think
488                    about it this is right. Otherwise apps have to
489                    play 'guess the biggest size' games. RCVBUF/SNDBUF
490                    are treated in BSD as hints */
491
492                 if (val > sysctl_wmem_max)
493                         val = sysctl_wmem_max;
494 set_sndbuf:
495                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
496                 if ((val * 2) < SOCK_MIN_SNDBUF)
497                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
498                 else
499                         sk->sk_sndbuf = val * 2;
500
501                 /*
502                  *      Wake up sending tasks if we
503                  *      upped the value.
504                  */
505                 sk->sk_write_space(sk);
506                 break;
507
508         case SO_SNDBUFFORCE:
509                 if (!capable(CAP_NET_ADMIN)) {
510                         ret = -EPERM;
511                         break;
512                 }
513                 goto set_sndbuf;
514
515         case SO_RCVBUF:
516                 /* Don't error on this BSD doesn't and if you think
517                    about it this is right. Otherwise apps have to
518                    play 'guess the biggest size' games. RCVBUF/SNDBUF
519                    are treated in BSD as hints */
520
521                 if (val > sysctl_rmem_max)
522                         val = sysctl_rmem_max;
523 set_rcvbuf:
524                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
525                 /*
526                  * We double it on the way in to account for
527                  * "struct sk_buff" etc. overhead.   Applications
528                  * assume that the SO_RCVBUF setting they make will
529                  * allow that much actual data to be received on that
530                  * socket.
531                  *
532                  * Applications are unaware that "struct sk_buff" and
533                  * other overheads allocate from the receive buffer
534                  * during socket buffer allocation.
535                  *
536                  * And after considering the possible alternatives,
537                  * returning the value we actually used in getsockopt
538                  * is the most desirable behavior.
539                  */
540                 if ((val * 2) < SOCK_MIN_RCVBUF)
541                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
542                 else
543                         sk->sk_rcvbuf = val * 2;
544                 break;
545
546         case SO_RCVBUFFORCE:
547                 if (!capable(CAP_NET_ADMIN)) {
548                         ret = -EPERM;
549                         break;
550                 }
551                 goto set_rcvbuf;
552
553         case SO_KEEPALIVE:
554 #ifdef CONFIG_INET
555                 if (sk->sk_protocol == IPPROTO_TCP)
556                         tcp_set_keepalive(sk, valbool);
557 #endif
558                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
559                 break;
560
561         case SO_OOBINLINE:
562                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
563                 break;
564
565         case SO_NO_CHECK:
566                 sk->sk_no_check = valbool;
567                 break;
568
569         case SO_PRIORITY:
570                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
571                         sk->sk_priority = val;
572                 else
573                         ret = -EPERM;
574                 break;
575
576         case SO_LINGER:
577                 if (optlen < sizeof(ling)) {
578                         ret = -EINVAL;  /* 1003.1g */
579                         break;
580                 }
581                 if (copy_from_user(&ling,optval,sizeof(ling))) {
582                         ret = -EFAULT;
583                         break;
584                 }
585                 if (!ling.l_onoff)
586                         sock_reset_flag(sk, SOCK_LINGER);
587                 else {
588 #if (BITS_PER_LONG == 32)
589                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
590                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
591                         else
592 #endif
593                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
594                         sock_set_flag(sk, SOCK_LINGER);
595                 }
596                 break;
597
598         case SO_BSDCOMPAT:
599                 sock_warn_obsolete_bsdism("setsockopt");
600                 break;
601
602         case SO_PASSCRED:
603                 if (valbool)
604                         set_bit(SOCK_PASSCRED, &sock->flags);
605                 else
606                         clear_bit(SOCK_PASSCRED, &sock->flags);
607                 break;
608
609         case SO_TIMESTAMP:
610         case SO_TIMESTAMPNS:
611                 if (valbool)  {
612                         if (optname == SO_TIMESTAMP)
613                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
614                         else
615                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
616                         sock_set_flag(sk, SOCK_RCVTSTAMP);
617                         sock_enable_timestamp(sk);
618                 } else {
619                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
620                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
621                 }
622                 break;
623
624         case SO_RCVLOWAT:
625                 if (val < 0)
626                         val = INT_MAX;
627                 sk->sk_rcvlowat = val ? : 1;
628                 break;
629
630         case SO_RCVTIMEO:
631                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
632                 break;
633
634         case SO_SNDTIMEO:
635                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
636                 break;
637
638         case SO_ATTACH_FILTER:
639                 ret = -EINVAL;
640                 if (optlen == sizeof(struct sock_fprog)) {
641                         struct sock_fprog fprog;
642
643                         ret = -EFAULT;
644                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
645                                 break;
646
647                         ret = sk_attach_filter(&fprog, sk);
648                 }
649                 break;
650
651         case SO_DETACH_FILTER:
652                 ret = sk_detach_filter(sk);
653                 break;
654
655         case SO_PASSSEC:
656                 if (valbool)
657                         set_bit(SOCK_PASSSEC, &sock->flags);
658                 else
659                         clear_bit(SOCK_PASSSEC, &sock->flags);
660                 break;
661         case SO_MARK:
662                 if (!capable(CAP_NET_ADMIN))
663                         ret = -EPERM;
664                 else {
665                         sk->sk_mark = val;
666                 }
667                 break;
668
669                 /* We implement the SO_SNDLOWAT etc to
670                    not be settable (1003.1g 5.3) */
671         default:
672                 ret = -ENOPROTOOPT;
673                 break;
674         }
675         release_sock(sk);
676         return ret;
677 }
678
679
680 int sock_getsockopt(struct socket *sock, int level, int optname,
681                     char __user *optval, int __user *optlen)
682 {
683         struct sock *sk = sock->sk;
684
685         union {
686                 int val;
687                 struct linger ling;
688                 struct timeval tm;
689         } v;
690
691         unsigned int lv = sizeof(int);
692         int len;
693
694         if (get_user(len, optlen))
695                 return -EFAULT;
696         if (len < 0)
697                 return -EINVAL;
698
699         v.val = 0;
700
701         switch(optname) {
702         case SO_DEBUG:
703                 v.val = sock_flag(sk, SOCK_DBG);
704                 break;
705
706         case SO_DONTROUTE:
707                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
708                 break;
709
710         case SO_BROADCAST:
711                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
712                 break;
713
714         case SO_SNDBUF:
715                 v.val = sk->sk_sndbuf;
716                 break;
717
718         case SO_RCVBUF:
719                 v.val = sk->sk_rcvbuf;
720                 break;
721
722         case SO_REUSEADDR:
723                 v.val = sk->sk_reuse;
724                 break;
725
726         case SO_KEEPALIVE:
727                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
728                 break;
729
730         case SO_TYPE:
731                 v.val = sk->sk_type;
732                 break;
733
734         case SO_ERROR:
735                 v.val = -sock_error(sk);
736                 if (v.val==0)
737                         v.val = xchg(&sk->sk_err_soft, 0);
738                 break;
739
740         case SO_OOBINLINE:
741                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
742                 break;
743
744         case SO_NO_CHECK:
745                 v.val = sk->sk_no_check;
746                 break;
747
748         case SO_PRIORITY:
749                 v.val = sk->sk_priority;
750                 break;
751
752         case SO_LINGER:
753                 lv              = sizeof(v.ling);
754                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
755                 v.ling.l_linger = sk->sk_lingertime / HZ;
756                 break;
757
758         case SO_BSDCOMPAT:
759                 sock_warn_obsolete_bsdism("getsockopt");
760                 break;
761
762         case SO_TIMESTAMP:
763                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
764                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
765                 break;
766
767         case SO_TIMESTAMPNS:
768                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
769                 break;
770
771         case SO_RCVTIMEO:
772                 lv=sizeof(struct timeval);
773                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
774                         v.tm.tv_sec = 0;
775                         v.tm.tv_usec = 0;
776                 } else {
777                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
778                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
779                 }
780                 break;
781
782         case SO_SNDTIMEO:
783                 lv=sizeof(struct timeval);
784                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
785                         v.tm.tv_sec = 0;
786                         v.tm.tv_usec = 0;
787                 } else {
788                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
789                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
790                 }
791                 break;
792
793         case SO_RCVLOWAT:
794                 v.val = sk->sk_rcvlowat;
795                 break;
796
797         case SO_SNDLOWAT:
798                 v.val=1;
799                 break;
800
801         case SO_PASSCRED:
802                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
803                 break;
804
805         case SO_PEERCRED:
806                 if (len > sizeof(sk->sk_peercred))
807                         len = sizeof(sk->sk_peercred);
808                 if (copy_to_user(optval, &sk->sk_peercred, len))
809                         return -EFAULT;
810                 goto lenout;
811
812         case SO_PEERNAME:
813         {
814                 char address[128];
815
816                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
817                         return -ENOTCONN;
818                 if (lv < len)
819                         return -EINVAL;
820                 if (copy_to_user(optval, address, len))
821                         return -EFAULT;
822                 goto lenout;
823         }
824
825         /* Dubious BSD thing... Probably nobody even uses it, but
826          * the UNIX standard wants it for whatever reason... -DaveM
827          */
828         case SO_ACCEPTCONN:
829                 v.val = sk->sk_state == TCP_LISTEN;
830                 break;
831
832         case SO_PASSSEC:
833                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
834                 break;
835
836         case SO_PEERSEC:
837                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
838
839         case SO_MARK:
840                 v.val = sk->sk_mark;
841                 break;
842
843         default:
844                 return -ENOPROTOOPT;
845         }
846
847         if (len > lv)
848                 len = lv;
849         if (copy_to_user(optval, &v, len))
850                 return -EFAULT;
851 lenout:
852         if (put_user(len, optlen))
853                 return -EFAULT;
854         return 0;
855 }
856
857 /*
858  * Initialize an sk_lock.
859  *
860  * (We also register the sk_lock with the lock validator.)
861  */
862 static inline void sock_lock_init(struct sock *sk)
863 {
864         sock_lock_init_class_and_name(sk,
865                         af_family_slock_key_strings[sk->sk_family],
866                         af_family_slock_keys + sk->sk_family,
867                         af_family_key_strings[sk->sk_family],
868                         af_family_keys + sk->sk_family);
869 }
870
871 static void sock_copy(struct sock *nsk, const struct sock *osk)
872 {
873 #ifdef CONFIG_SECURITY_NETWORK
874         void *sptr = nsk->sk_security;
875 #endif
876
877         memcpy(nsk, osk, osk->sk_prot->obj_size);
878 #ifdef CONFIG_SECURITY_NETWORK
879         nsk->sk_security = sptr;
880         security_sk_clone(osk, nsk);
881 #endif
882 }
883
884 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
885                 int family)
886 {
887         struct sock *sk;
888         struct kmem_cache *slab;
889
890         slab = prot->slab;
891         if (slab != NULL)
892                 sk = kmem_cache_alloc(slab, priority);
893         else
894                 sk = kmalloc(prot->obj_size, priority);
895
896         if (sk != NULL) {
897                 if (security_sk_alloc(sk, family, priority))
898                         goto out_free;
899
900                 if (!try_module_get(prot->owner))
901                         goto out_free_sec;
902         }
903
904         return sk;
905
906 out_free_sec:
907         security_sk_free(sk);
908 out_free:
909         if (slab != NULL)
910                 kmem_cache_free(slab, sk);
911         else
912                 kfree(sk);
913         return NULL;
914 }
915
916 static void sk_prot_free(struct proto *prot, struct sock *sk)
917 {
918         struct kmem_cache *slab;
919         struct module *owner;
920
921         owner = prot->owner;
922         slab = prot->slab;
923
924         security_sk_free(sk);
925         if (slab != NULL)
926                 kmem_cache_free(slab, sk);
927         else
928                 kfree(sk);
929         module_put(owner);
930 }
931
932 /**
933  *      sk_alloc - All socket objects are allocated here
934  *      @net: the applicable net namespace
935  *      @family: protocol family
936  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
937  *      @prot: struct proto associated with this new sock instance
938  */
939 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
940                       struct proto *prot)
941 {
942         struct sock *sk;
943
944         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
945         if (sk) {
946                 sk->sk_family = family;
947                 /*
948                  * See comment in struct sock definition to understand
949                  * why we need sk_prot_creator -acme
950                  */
951                 sk->sk_prot = sk->sk_prot_creator = prot;
952                 sock_lock_init(sk);
953                 sock_net_set(sk, get_net(net));
954         }
955
956         return sk;
957 }
958
959 void sk_free(struct sock *sk)
960 {
961         struct sk_filter *filter;
962
963         if (sk->sk_destruct)
964                 sk->sk_destruct(sk);
965
966         filter = rcu_dereference(sk->sk_filter);
967         if (filter) {
968                 sk_filter_uncharge(sk, filter);
969                 rcu_assign_pointer(sk->sk_filter, NULL);
970         }
971
972         sock_disable_timestamp(sk);
973
974         if (atomic_read(&sk->sk_omem_alloc))
975                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
976                        __func__, atomic_read(&sk->sk_omem_alloc));
977
978         put_net(sock_net(sk));
979         sk_prot_free(sk->sk_prot_creator, sk);
980 }
981
982 /*
983  * Last sock_put should drop referrence to sk->sk_net. It has already
984  * been dropped in sk_change_net. Taking referrence to stopping namespace
985  * is not an option.
986  * Take referrence to a socket to remove it from hash _alive_ and after that
987  * destroy it in the context of init_net.
988  */
989 void sk_release_kernel(struct sock *sk)
990 {
991         if (sk == NULL || sk->sk_socket == NULL)
992                 return;
993
994         sock_hold(sk);
995         sock_release(sk->sk_socket);
996         release_net(sock_net(sk));
997         sock_net_set(sk, get_net(&init_net));
998         sock_put(sk);
999 }
1000 EXPORT_SYMBOL(sk_release_kernel);
1001
1002 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1003 {
1004         struct sock *newsk;
1005
1006         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1007         if (newsk != NULL) {
1008                 struct sk_filter *filter;
1009
1010                 sock_copy(newsk, sk);
1011
1012                 /* SANITY */
1013                 get_net(sock_net(newsk));
1014                 sk_node_init(&newsk->sk_node);
1015                 sock_lock_init(newsk);
1016                 bh_lock_sock(newsk);
1017                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1018
1019                 atomic_set(&newsk->sk_rmem_alloc, 0);
1020                 atomic_set(&newsk->sk_wmem_alloc, 0);
1021                 atomic_set(&newsk->sk_omem_alloc, 0);
1022                 skb_queue_head_init(&newsk->sk_receive_queue);
1023                 skb_queue_head_init(&newsk->sk_write_queue);
1024 #ifdef CONFIG_NET_DMA
1025                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1026 #endif
1027
1028                 rwlock_init(&newsk->sk_dst_lock);
1029                 rwlock_init(&newsk->sk_callback_lock);
1030                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1031                                 af_callback_keys + newsk->sk_family,
1032                                 af_family_clock_key_strings[newsk->sk_family]);
1033
1034                 newsk->sk_dst_cache     = NULL;
1035                 newsk->sk_wmem_queued   = 0;
1036                 newsk->sk_forward_alloc = 0;
1037                 newsk->sk_send_head     = NULL;
1038                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1039
1040                 sock_reset_flag(newsk, SOCK_DONE);
1041                 skb_queue_head_init(&newsk->sk_error_queue);
1042
1043                 filter = newsk->sk_filter;
1044                 if (filter != NULL)
1045                         sk_filter_charge(newsk, filter);
1046
1047                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1048                         /* It is still raw copy of parent, so invalidate
1049                          * destructor and make plain sk_free() */
1050                         newsk->sk_destruct = NULL;
1051                         sk_free(newsk);
1052                         newsk = NULL;
1053                         goto out;
1054                 }
1055
1056                 newsk->sk_err      = 0;
1057                 newsk->sk_priority = 0;
1058                 atomic_set(&newsk->sk_refcnt, 2);
1059
1060                 /*
1061                  * Increment the counter in the same struct proto as the master
1062                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1063                  * is the same as sk->sk_prot->socks, as this field was copied
1064                  * with memcpy).
1065                  *
1066                  * This _changes_ the previous behaviour, where
1067                  * tcp_create_openreq_child always was incrementing the
1068                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1069                  * to be taken into account in all callers. -acme
1070                  */
1071                 sk_refcnt_debug_inc(newsk);
1072                 sk_set_socket(newsk, NULL);
1073                 newsk->sk_sleep  = NULL;
1074
1075                 if (newsk->sk_prot->sockets_allocated)
1076                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1077         }
1078 out:
1079         return newsk;
1080 }
1081
1082 EXPORT_SYMBOL_GPL(sk_clone);
1083
1084 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1085 {
1086         __sk_dst_set(sk, dst);
1087         sk->sk_route_caps = dst->dev->features;
1088         if (sk->sk_route_caps & NETIF_F_GSO)
1089                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1090         if (sk_can_gso(sk)) {
1091                 if (dst->header_len) {
1092                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1093                 } else {
1094                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1095                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1096                 }
1097         }
1098 }
1099 EXPORT_SYMBOL_GPL(sk_setup_caps);
1100
1101 void __init sk_init(void)
1102 {
1103         if (num_physpages <= 4096) {
1104                 sysctl_wmem_max = 32767;
1105                 sysctl_rmem_max = 32767;
1106                 sysctl_wmem_default = 32767;
1107                 sysctl_rmem_default = 32767;
1108         } else if (num_physpages >= 131072) {
1109                 sysctl_wmem_max = 131071;
1110                 sysctl_rmem_max = 131071;
1111         }
1112 }
1113
1114 /*
1115  *      Simple resource managers for sockets.
1116  */
1117
1118
1119 /*
1120  * Write buffer destructor automatically called from kfree_skb.
1121  */
1122 void sock_wfree(struct sk_buff *skb)
1123 {
1124         struct sock *sk = skb->sk;
1125
1126         /* In case it might be waiting for more memory. */
1127         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1128         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1129                 sk->sk_write_space(sk);
1130         sock_put(sk);
1131 }
1132
1133 /*
1134  * Read buffer destructor automatically called from kfree_skb.
1135  */
1136 void sock_rfree(struct sk_buff *skb)
1137 {
1138         struct sock *sk = skb->sk;
1139
1140         skb_truesize_check(skb);
1141         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1142         sk_mem_uncharge(skb->sk, skb->truesize);
1143 }
1144
1145
1146 int sock_i_uid(struct sock *sk)
1147 {
1148         int uid;
1149
1150         read_lock(&sk->sk_callback_lock);
1151         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1152         read_unlock(&sk->sk_callback_lock);
1153         return uid;
1154 }
1155
1156 unsigned long sock_i_ino(struct sock *sk)
1157 {
1158         unsigned long ino;
1159
1160         read_lock(&sk->sk_callback_lock);
1161         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1162         read_unlock(&sk->sk_callback_lock);
1163         return ino;
1164 }
1165
1166 /*
1167  * Allocate a skb from the socket's send buffer.
1168  */
1169 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1170                              gfp_t priority)
1171 {
1172         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1173                 struct sk_buff * skb = alloc_skb(size, priority);
1174                 if (skb) {
1175                         skb_set_owner_w(skb, sk);
1176                         return skb;
1177                 }
1178         }
1179         return NULL;
1180 }
1181
1182 /*
1183  * Allocate a skb from the socket's receive buffer.
1184  */
1185 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1186                              gfp_t priority)
1187 {
1188         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1189                 struct sk_buff *skb = alloc_skb(size, priority);
1190                 if (skb) {
1191                         skb_set_owner_r(skb, sk);
1192                         return skb;
1193                 }
1194         }
1195         return NULL;
1196 }
1197
1198 /*
1199  * Allocate a memory block from the socket's option memory buffer.
1200  */
1201 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1202 {
1203         if ((unsigned)size <= sysctl_optmem_max &&
1204             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1205                 void *mem;
1206                 /* First do the add, to avoid the race if kmalloc
1207                  * might sleep.
1208                  */
1209                 atomic_add(size, &sk->sk_omem_alloc);
1210                 mem = kmalloc(size, priority);
1211                 if (mem)
1212                         return mem;
1213                 atomic_sub(size, &sk->sk_omem_alloc);
1214         }
1215         return NULL;
1216 }
1217
1218 /*
1219  * Free an option memory block.
1220  */
1221 void sock_kfree_s(struct sock *sk, void *mem, int size)
1222 {
1223         kfree(mem);
1224         atomic_sub(size, &sk->sk_omem_alloc);
1225 }
1226
1227 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1228    I think, these locks should be removed for datagram sockets.
1229  */
1230 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1231 {
1232         DEFINE_WAIT(wait);
1233
1234         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1235         for (;;) {
1236                 if (!timeo)
1237                         break;
1238                 if (signal_pending(current))
1239                         break;
1240                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1241                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1242                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1243                         break;
1244                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1245                         break;
1246                 if (sk->sk_err)
1247                         break;
1248                 timeo = schedule_timeout(timeo);
1249         }
1250         finish_wait(sk->sk_sleep, &wait);
1251         return timeo;
1252 }
1253
1254
1255 /*
1256  *      Generic send/receive buffer handlers
1257  */
1258
1259 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1260                                      unsigned long data_len, int noblock,
1261                                      int *errcode)
1262 {
1263         struct sk_buff *skb;
1264         gfp_t gfp_mask;
1265         long timeo;
1266         int err;
1267
1268         gfp_mask = sk->sk_allocation;
1269         if (gfp_mask & __GFP_WAIT)
1270                 gfp_mask |= __GFP_REPEAT;
1271
1272         timeo = sock_sndtimeo(sk, noblock);
1273         while (1) {
1274                 err = sock_error(sk);
1275                 if (err != 0)
1276                         goto failure;
1277
1278                 err = -EPIPE;
1279                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1280                         goto failure;
1281
1282                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1283                         skb = alloc_skb(header_len, gfp_mask);
1284                         if (skb) {
1285                                 int npages;
1286                                 int i;
1287
1288                                 /* No pages, we're done... */
1289                                 if (!data_len)
1290                                         break;
1291
1292                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1293                                 skb->truesize += data_len;
1294                                 skb_shinfo(skb)->nr_frags = npages;
1295                                 for (i = 0; i < npages; i++) {
1296                                         struct page *page;
1297                                         skb_frag_t *frag;
1298
1299                                         page = alloc_pages(sk->sk_allocation, 0);
1300                                         if (!page) {
1301                                                 err = -ENOBUFS;
1302                                                 skb_shinfo(skb)->nr_frags = i;
1303                                                 kfree_skb(skb);
1304                                                 goto failure;
1305                                         }
1306
1307                                         frag = &skb_shinfo(skb)->frags[i];
1308                                         frag->page = page;
1309                                         frag->page_offset = 0;
1310                                         frag->size = (data_len >= PAGE_SIZE ?
1311                                                       PAGE_SIZE :
1312                                                       data_len);
1313                                         data_len -= PAGE_SIZE;
1314                                 }
1315
1316                                 /* Full success... */
1317                                 break;
1318                         }
1319                         err = -ENOBUFS;
1320                         goto failure;
1321                 }
1322                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1323                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1324                 err = -EAGAIN;
1325                 if (!timeo)
1326                         goto failure;
1327                 if (signal_pending(current))
1328                         goto interrupted;
1329                 timeo = sock_wait_for_wmem(sk, timeo);
1330         }
1331
1332         skb_set_owner_w(skb, sk);
1333         return skb;
1334
1335 interrupted:
1336         err = sock_intr_errno(timeo);
1337 failure:
1338         *errcode = err;
1339         return NULL;
1340 }
1341 EXPORT_SYMBOL(sock_alloc_send_pskb);
1342
1343 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1344                                     int noblock, int *errcode)
1345 {
1346         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1347 }
1348
1349 static void __lock_sock(struct sock *sk)
1350 {
1351         DEFINE_WAIT(wait);
1352
1353         for (;;) {
1354                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1355                                         TASK_UNINTERRUPTIBLE);
1356                 spin_unlock_bh(&sk->sk_lock.slock);
1357                 schedule();
1358                 spin_lock_bh(&sk->sk_lock.slock);
1359                 if (!sock_owned_by_user(sk))
1360                         break;
1361         }
1362         finish_wait(&sk->sk_lock.wq, &wait);
1363 }
1364
1365 static void __release_sock(struct sock *sk)
1366 {
1367         struct sk_buff *skb = sk->sk_backlog.head;
1368
1369         do {
1370                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1371                 bh_unlock_sock(sk);
1372
1373                 do {
1374                         struct sk_buff *next = skb->next;
1375
1376                         skb->next = NULL;
1377                         sk_backlog_rcv(sk, skb);
1378
1379                         /*
1380                          * We are in process context here with softirqs
1381                          * disabled, use cond_resched_softirq() to preempt.
1382                          * This is safe to do because we've taken the backlog
1383                          * queue private:
1384                          */
1385                         cond_resched_softirq();
1386
1387                         skb = next;
1388                 } while (skb != NULL);
1389
1390                 bh_lock_sock(sk);
1391         } while ((skb = sk->sk_backlog.head) != NULL);
1392 }
1393
1394 /**
1395  * sk_wait_data - wait for data to arrive at sk_receive_queue
1396  * @sk:    sock to wait on
1397  * @timeo: for how long
1398  *
1399  * Now socket state including sk->sk_err is changed only under lock,
1400  * hence we may omit checks after joining wait queue.
1401  * We check receive queue before schedule() only as optimization;
1402  * it is very likely that release_sock() added new data.
1403  */
1404 int sk_wait_data(struct sock *sk, long *timeo)
1405 {
1406         int rc;
1407         DEFINE_WAIT(wait);
1408
1409         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1410         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1411         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1412         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1413         finish_wait(sk->sk_sleep, &wait);
1414         return rc;
1415 }
1416
1417 EXPORT_SYMBOL(sk_wait_data);
1418
1419 /**
1420  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1421  *      @sk: socket
1422  *      @size: memory size to allocate
1423  *      @kind: allocation type
1424  *
1425  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1426  *      rmem allocation. This function assumes that protocols which have
1427  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1428  */
1429 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1430 {
1431         struct proto *prot = sk->sk_prot;
1432         int amt = sk_mem_pages(size);
1433         int allocated;
1434
1435         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1436         allocated = atomic_add_return(amt, prot->memory_allocated);
1437
1438         /* Under limit. */
1439         if (allocated <= prot->sysctl_mem[0]) {
1440                 if (prot->memory_pressure && *prot->memory_pressure)
1441                         *prot->memory_pressure = 0;
1442                 return 1;
1443         }
1444
1445         /* Under pressure. */
1446         if (allocated > prot->sysctl_mem[1])
1447                 if (prot->enter_memory_pressure)
1448                         prot->enter_memory_pressure(sk);
1449
1450         /* Over hard limit. */
1451         if (allocated > prot->sysctl_mem[2])
1452                 goto suppress_allocation;
1453
1454         /* guarantee minimum buffer size under pressure */
1455         if (kind == SK_MEM_RECV) {
1456                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1457                         return 1;
1458         } else { /* SK_MEM_SEND */
1459                 if (sk->sk_type == SOCK_STREAM) {
1460                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1461                                 return 1;
1462                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1463                            prot->sysctl_wmem[0])
1464                                 return 1;
1465         }
1466
1467         if (prot->memory_pressure) {
1468                 int alloc;
1469
1470                 if (!*prot->memory_pressure)
1471                         return 1;
1472                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1473                 if (prot->sysctl_mem[2] > alloc *
1474                     sk_mem_pages(sk->sk_wmem_queued +
1475                                  atomic_read(&sk->sk_rmem_alloc) +
1476                                  sk->sk_forward_alloc))
1477                         return 1;
1478         }
1479
1480 suppress_allocation:
1481
1482         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1483                 sk_stream_moderate_sndbuf(sk);
1484
1485                 /* Fail only if socket is _under_ its sndbuf.
1486                  * In this case we cannot block, so that we have to fail.
1487                  */
1488                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1489                         return 1;
1490         }
1491
1492         /* Alas. Undo changes. */
1493         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1494         atomic_sub(amt, prot->memory_allocated);
1495         return 0;
1496 }
1497
1498 EXPORT_SYMBOL(__sk_mem_schedule);
1499
1500 /**
1501  *      __sk_reclaim - reclaim memory_allocated
1502  *      @sk: socket
1503  */
1504 void __sk_mem_reclaim(struct sock *sk)
1505 {
1506         struct proto *prot = sk->sk_prot;
1507
1508         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1509                    prot->memory_allocated);
1510         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1511
1512         if (prot->memory_pressure && *prot->memory_pressure &&
1513             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1514                 *prot->memory_pressure = 0;
1515 }
1516
1517 EXPORT_SYMBOL(__sk_mem_reclaim);
1518
1519
1520 /*
1521  * Set of default routines for initialising struct proto_ops when
1522  * the protocol does not support a particular function. In certain
1523  * cases where it makes no sense for a protocol to have a "do nothing"
1524  * function, some default processing is provided.
1525  */
1526
1527 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1528 {
1529         return -EOPNOTSUPP;
1530 }
1531
1532 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1533                     int len, int flags)
1534 {
1535         return -EOPNOTSUPP;
1536 }
1537
1538 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1539 {
1540         return -EOPNOTSUPP;
1541 }
1542
1543 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1544 {
1545         return -EOPNOTSUPP;
1546 }
1547
1548 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1549                     int *len, int peer)
1550 {
1551         return -EOPNOTSUPP;
1552 }
1553
1554 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1555 {
1556         return 0;
1557 }
1558
1559 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1560 {
1561         return -EOPNOTSUPP;
1562 }
1563
1564 int sock_no_listen(struct socket *sock, int backlog)
1565 {
1566         return -EOPNOTSUPP;
1567 }
1568
1569 int sock_no_shutdown(struct socket *sock, int how)
1570 {
1571         return -EOPNOTSUPP;
1572 }
1573
1574 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1575                     char __user *optval, int optlen)
1576 {
1577         return -EOPNOTSUPP;
1578 }
1579
1580 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1581                     char __user *optval, int __user *optlen)
1582 {
1583         return -EOPNOTSUPP;
1584 }
1585
1586 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1587                     size_t len)
1588 {
1589         return -EOPNOTSUPP;
1590 }
1591
1592 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1593                     size_t len, int flags)
1594 {
1595         return -EOPNOTSUPP;
1596 }
1597
1598 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1599 {
1600         /* Mirror missing mmap method error code */
1601         return -ENODEV;
1602 }
1603
1604 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1605 {
1606         ssize_t res;
1607         struct msghdr msg = {.msg_flags = flags};
1608         struct kvec iov;
1609         char *kaddr = kmap(page);
1610         iov.iov_base = kaddr + offset;
1611         iov.iov_len = size;
1612         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1613         kunmap(page);
1614         return res;
1615 }
1616
1617 /*
1618  *      Default Socket Callbacks
1619  */
1620
1621 static void sock_def_wakeup(struct sock *sk)
1622 {
1623         read_lock(&sk->sk_callback_lock);
1624         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1625                 wake_up_interruptible_all(sk->sk_sleep);
1626         read_unlock(&sk->sk_callback_lock);
1627 }
1628
1629 static void sock_def_error_report(struct sock *sk)
1630 {
1631         read_lock(&sk->sk_callback_lock);
1632         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1633                 wake_up_interruptible(sk->sk_sleep);
1634         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1635         read_unlock(&sk->sk_callback_lock);
1636 }
1637
1638 static void sock_def_readable(struct sock *sk, int len)
1639 {
1640         read_lock(&sk->sk_callback_lock);
1641         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1642                 wake_up_interruptible_sync(sk->sk_sleep);
1643         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1644         read_unlock(&sk->sk_callback_lock);
1645 }
1646
1647 static void sock_def_write_space(struct sock *sk)
1648 {
1649         read_lock(&sk->sk_callback_lock);
1650
1651         /* Do not wake up a writer until he can make "significant"
1652          * progress.  --DaveM
1653          */
1654         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1655                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1656                         wake_up_interruptible_sync(sk->sk_sleep);
1657
1658                 /* Should agree with poll, otherwise some programs break */
1659                 if (sock_writeable(sk))
1660                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1661         }
1662
1663         read_unlock(&sk->sk_callback_lock);
1664 }
1665
1666 static void sock_def_destruct(struct sock *sk)
1667 {
1668         kfree(sk->sk_protinfo);
1669 }
1670
1671 void sk_send_sigurg(struct sock *sk)
1672 {
1673         if (sk->sk_socket && sk->sk_socket->file)
1674                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1675                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1676 }
1677
1678 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1679                     unsigned long expires)
1680 {
1681         if (!mod_timer(timer, expires))
1682                 sock_hold(sk);
1683 }
1684
1685 EXPORT_SYMBOL(sk_reset_timer);
1686
1687 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1688 {
1689         if (timer_pending(timer) && del_timer(timer))
1690                 __sock_put(sk);
1691 }
1692
1693 EXPORT_SYMBOL(sk_stop_timer);
1694
1695 void sock_init_data(struct socket *sock, struct sock *sk)
1696 {
1697         skb_queue_head_init(&sk->sk_receive_queue);
1698         skb_queue_head_init(&sk->sk_write_queue);
1699         skb_queue_head_init(&sk->sk_error_queue);
1700 #ifdef CONFIG_NET_DMA
1701         skb_queue_head_init(&sk->sk_async_wait_queue);
1702 #endif
1703
1704         sk->sk_send_head        =       NULL;
1705
1706         init_timer(&sk->sk_timer);
1707
1708         sk->sk_allocation       =       GFP_KERNEL;
1709         sk->sk_rcvbuf           =       sysctl_rmem_default;
1710         sk->sk_sndbuf           =       sysctl_wmem_default;
1711         sk->sk_state            =       TCP_CLOSE;
1712         sk_set_socket(sk, sock);
1713
1714         sock_set_flag(sk, SOCK_ZAPPED);
1715
1716         if (sock) {
1717                 sk->sk_type     =       sock->type;
1718                 sk->sk_sleep    =       &sock->wait;
1719                 sock->sk        =       sk;
1720         } else
1721                 sk->sk_sleep    =       NULL;
1722
1723         rwlock_init(&sk->sk_dst_lock);
1724         rwlock_init(&sk->sk_callback_lock);
1725         lockdep_set_class_and_name(&sk->sk_callback_lock,
1726                         af_callback_keys + sk->sk_family,
1727                         af_family_clock_key_strings[sk->sk_family]);
1728
1729         sk->sk_state_change     =       sock_def_wakeup;
1730         sk->sk_data_ready       =       sock_def_readable;
1731         sk->sk_write_space      =       sock_def_write_space;
1732         sk->sk_error_report     =       sock_def_error_report;
1733         sk->sk_destruct         =       sock_def_destruct;
1734
1735         sk->sk_sndmsg_page      =       NULL;
1736         sk->sk_sndmsg_off       =       0;
1737
1738         sk->sk_peercred.pid     =       0;
1739         sk->sk_peercred.uid     =       -1;
1740         sk->sk_peercred.gid     =       -1;
1741         sk->sk_write_pending    =       0;
1742         sk->sk_rcvlowat         =       1;
1743         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1744         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1745
1746         sk->sk_stamp = ktime_set(-1L, 0);
1747
1748         atomic_set(&sk->sk_refcnt, 1);
1749         atomic_set(&sk->sk_drops, 0);
1750 }
1751
1752 void lock_sock_nested(struct sock *sk, int subclass)
1753 {
1754         might_sleep();
1755         spin_lock_bh(&sk->sk_lock.slock);
1756         if (sk->sk_lock.owned)
1757                 __lock_sock(sk);
1758         sk->sk_lock.owned = 1;
1759         spin_unlock(&sk->sk_lock.slock);
1760         /*
1761          * The sk_lock has mutex_lock() semantics here:
1762          */
1763         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1764         local_bh_enable();
1765 }
1766
1767 EXPORT_SYMBOL(lock_sock_nested);
1768
1769 void release_sock(struct sock *sk)
1770 {
1771         /*
1772          * The sk_lock has mutex_unlock() semantics:
1773          */
1774         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1775
1776         spin_lock_bh(&sk->sk_lock.slock);
1777         if (sk->sk_backlog.tail)
1778                 __release_sock(sk);
1779         sk->sk_lock.owned = 0;
1780         if (waitqueue_active(&sk->sk_lock.wq))
1781                 wake_up(&sk->sk_lock.wq);
1782         spin_unlock_bh(&sk->sk_lock.slock);
1783 }
1784 EXPORT_SYMBOL(release_sock);
1785
1786 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1787 {
1788         struct timeval tv;
1789         if (!sock_flag(sk, SOCK_TIMESTAMP))
1790                 sock_enable_timestamp(sk);
1791         tv = ktime_to_timeval(sk->sk_stamp);
1792         if (tv.tv_sec == -1)
1793                 return -ENOENT;
1794         if (tv.tv_sec == 0) {
1795                 sk->sk_stamp = ktime_get_real();
1796                 tv = ktime_to_timeval(sk->sk_stamp);
1797         }
1798         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1799 }
1800 EXPORT_SYMBOL(sock_get_timestamp);
1801
1802 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1803 {
1804         struct timespec ts;
1805         if (!sock_flag(sk, SOCK_TIMESTAMP))
1806                 sock_enable_timestamp(sk);
1807         ts = ktime_to_timespec(sk->sk_stamp);
1808         if (ts.tv_sec == -1)
1809                 return -ENOENT;
1810         if (ts.tv_sec == 0) {
1811                 sk->sk_stamp = ktime_get_real();
1812                 ts = ktime_to_timespec(sk->sk_stamp);
1813         }
1814         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1815 }
1816 EXPORT_SYMBOL(sock_get_timestampns);
1817
1818 void sock_enable_timestamp(struct sock *sk)
1819 {
1820         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1821                 sock_set_flag(sk, SOCK_TIMESTAMP);
1822                 net_enable_timestamp();
1823         }
1824 }
1825
1826 /*
1827  *      Get a socket option on an socket.
1828  *
1829  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1830  *      asynchronous errors should be reported by getsockopt. We assume
1831  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1832  */
1833 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1834                            char __user *optval, int __user *optlen)
1835 {
1836         struct sock *sk = sock->sk;
1837
1838         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1839 }
1840
1841 EXPORT_SYMBOL(sock_common_getsockopt);
1842
1843 #ifdef CONFIG_COMPAT
1844 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1845                                   char __user *optval, int __user *optlen)
1846 {
1847         struct sock *sk = sock->sk;
1848
1849         if (sk->sk_prot->compat_getsockopt != NULL)
1850                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1851                                                       optval, optlen);
1852         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1853 }
1854 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1855 #endif
1856
1857 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1858                         struct msghdr *msg, size_t size, int flags)
1859 {
1860         struct sock *sk = sock->sk;
1861         int addr_len = 0;
1862         int err;
1863
1864         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1865                                    flags & ~MSG_DONTWAIT, &addr_len);
1866         if (err >= 0)
1867                 msg->msg_namelen = addr_len;
1868         return err;
1869 }
1870
1871 EXPORT_SYMBOL(sock_common_recvmsg);
1872
1873 /*
1874  *      Set socket options on an inet socket.
1875  */
1876 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1877                            char __user *optval, int optlen)
1878 {
1879         struct sock *sk = sock->sk;
1880
1881         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1882 }
1883
1884 EXPORT_SYMBOL(sock_common_setsockopt);
1885
1886 #ifdef CONFIG_COMPAT
1887 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1888                                   char __user *optval, int optlen)
1889 {
1890         struct sock *sk = sock->sk;
1891
1892         if (sk->sk_prot->compat_setsockopt != NULL)
1893                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1894                                                       optval, optlen);
1895         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1896 }
1897 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1898 #endif
1899
1900 void sk_common_release(struct sock *sk)
1901 {
1902         if (sk->sk_prot->destroy)
1903                 sk->sk_prot->destroy(sk);
1904
1905         /*
1906          * Observation: when sock_common_release is called, processes have
1907          * no access to socket. But net still has.
1908          * Step one, detach it from networking:
1909          *
1910          * A. Remove from hash tables.
1911          */
1912
1913         sk->sk_prot->unhash(sk);
1914
1915         /*
1916          * In this point socket cannot receive new packets, but it is possible
1917          * that some packets are in flight because some CPU runs receiver and
1918          * did hash table lookup before we unhashed socket. They will achieve
1919          * receive queue and will be purged by socket destructor.
1920          *
1921          * Also we still have packets pending on receive queue and probably,
1922          * our own packets waiting in device queues. sock_destroy will drain
1923          * receive queue, but transmitted packets will delay socket destruction
1924          * until the last reference will be released.
1925          */
1926
1927         sock_orphan(sk);
1928
1929         xfrm_sk_free_policy(sk);
1930
1931         sk_refcnt_debug_release(sk);
1932         sock_put(sk);
1933 }
1934
1935 EXPORT_SYMBOL(sk_common_release);
1936
1937 static DEFINE_RWLOCK(proto_list_lock);
1938 static LIST_HEAD(proto_list);
1939
1940 #ifdef CONFIG_PROC_FS
1941 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
1942 struct prot_inuse {
1943         int val[PROTO_INUSE_NR];
1944 };
1945
1946 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1947
1948 #ifdef CONFIG_NET_NS
1949 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1950 {
1951         int cpu = smp_processor_id();
1952         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1953 }
1954 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1955
1956 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1957 {
1958         int cpu, idx = prot->inuse_idx;
1959         int res = 0;
1960
1961         for_each_possible_cpu(cpu)
1962                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1963
1964         return res >= 0 ? res : 0;
1965 }
1966 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1967
1968 static int sock_inuse_init_net(struct net *net)
1969 {
1970         net->core.inuse = alloc_percpu(struct prot_inuse);
1971         return net->core.inuse ? 0 : -ENOMEM;
1972 }
1973
1974 static void sock_inuse_exit_net(struct net *net)
1975 {
1976         free_percpu(net->core.inuse);
1977 }
1978
1979 static struct pernet_operations net_inuse_ops = {
1980         .init = sock_inuse_init_net,
1981         .exit = sock_inuse_exit_net,
1982 };
1983
1984 static __init int net_inuse_init(void)
1985 {
1986         if (register_pernet_subsys(&net_inuse_ops))
1987                 panic("Cannot initialize net inuse counters");
1988
1989         return 0;
1990 }
1991
1992 core_initcall(net_inuse_init);
1993 #else
1994 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1995
1996 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1997 {
1998         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1999 }
2000 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2001
2002 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2003 {
2004         int cpu, idx = prot->inuse_idx;
2005         int res = 0;
2006
2007         for_each_possible_cpu(cpu)
2008                 res += per_cpu(prot_inuse, cpu).val[idx];
2009
2010         return res >= 0 ? res : 0;
2011 }
2012 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2013 #endif
2014
2015 static void assign_proto_idx(struct proto *prot)
2016 {
2017         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2018
2019         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2020                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2021                 return;
2022         }
2023
2024         set_bit(prot->inuse_idx, proto_inuse_idx);
2025 }
2026
2027 static void release_proto_idx(struct proto *prot)
2028 {
2029         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2030                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2031 }
2032 #else
2033 static inline void assign_proto_idx(struct proto *prot)
2034 {
2035 }
2036
2037 static inline void release_proto_idx(struct proto *prot)
2038 {
2039 }
2040 #endif
2041
2042 int proto_register(struct proto *prot, int alloc_slab)
2043 {
2044         if (alloc_slab) {
2045                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2046                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2047                                         NULL);
2048
2049                 if (prot->slab == NULL) {
2050                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2051                                prot->name);
2052                         goto out;
2053                 }
2054
2055                 if (prot->rsk_prot != NULL) {
2056                         static const char mask[] = "request_sock_%s";
2057
2058                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2059                         if (prot->rsk_prot->slab_name == NULL)
2060                                 goto out_free_sock_slab;
2061
2062                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2063                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2064                                                                  prot->rsk_prot->obj_size, 0,
2065                                                                  SLAB_HWCACHE_ALIGN, NULL);
2066
2067                         if (prot->rsk_prot->slab == NULL) {
2068                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2069                                        prot->name);
2070                                 goto out_free_request_sock_slab_name;
2071                         }
2072                 }
2073
2074                 if (prot->twsk_prot != NULL) {
2075                         static const char mask[] = "tw_sock_%s";
2076
2077                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2078
2079                         if (prot->twsk_prot->twsk_slab_name == NULL)
2080                                 goto out_free_request_sock_slab;
2081
2082                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2083                         prot->twsk_prot->twsk_slab =
2084                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2085                                                   prot->twsk_prot->twsk_obj_size,
2086                                                   0,
2087                                                   SLAB_HWCACHE_ALIGN |
2088                                                         prot->slab_flags,
2089                                                   NULL);
2090                         if (prot->twsk_prot->twsk_slab == NULL)
2091                                 goto out_free_timewait_sock_slab_name;
2092                 }
2093         }
2094
2095         write_lock(&proto_list_lock);
2096         list_add(&prot->node, &proto_list);
2097         assign_proto_idx(prot);
2098         write_unlock(&proto_list_lock);
2099         return 0;
2100
2101 out_free_timewait_sock_slab_name:
2102         kfree(prot->twsk_prot->twsk_slab_name);
2103 out_free_request_sock_slab:
2104         if (prot->rsk_prot && prot->rsk_prot->slab) {
2105                 kmem_cache_destroy(prot->rsk_prot->slab);
2106                 prot->rsk_prot->slab = NULL;
2107         }
2108 out_free_request_sock_slab_name:
2109         kfree(prot->rsk_prot->slab_name);
2110 out_free_sock_slab:
2111         kmem_cache_destroy(prot->slab);
2112         prot->slab = NULL;
2113 out:
2114         return -ENOBUFS;
2115 }
2116
2117 EXPORT_SYMBOL(proto_register);
2118
2119 void proto_unregister(struct proto *prot)
2120 {
2121         write_lock(&proto_list_lock);
2122         release_proto_idx(prot);
2123         list_del(&prot->node);
2124         write_unlock(&proto_list_lock);
2125
2126         if (prot->slab != NULL) {
2127                 kmem_cache_destroy(prot->slab);
2128                 prot->slab = NULL;
2129         }
2130
2131         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2132                 kmem_cache_destroy(prot->rsk_prot->slab);
2133                 kfree(prot->rsk_prot->slab_name);
2134                 prot->rsk_prot->slab = NULL;
2135         }
2136
2137         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2138                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2139                 kfree(prot->twsk_prot->twsk_slab_name);
2140                 prot->twsk_prot->twsk_slab = NULL;
2141         }
2142 }
2143
2144 EXPORT_SYMBOL(proto_unregister);
2145
2146 #ifdef CONFIG_PROC_FS
2147 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2148         __acquires(proto_list_lock)
2149 {
2150         read_lock(&proto_list_lock);
2151         return seq_list_start_head(&proto_list, *pos);
2152 }
2153
2154 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2155 {
2156         return seq_list_next(v, &proto_list, pos);
2157 }
2158
2159 static void proto_seq_stop(struct seq_file *seq, void *v)
2160         __releases(proto_list_lock)
2161 {
2162         read_unlock(&proto_list_lock);
2163 }
2164
2165 static char proto_method_implemented(const void *method)
2166 {
2167         return method == NULL ? 'n' : 'y';
2168 }
2169
2170 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2171 {
2172         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2173                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2174                    proto->name,
2175                    proto->obj_size,
2176                    sock_prot_inuse_get(seq_file_net(seq), proto),
2177                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2178                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2179                    proto->max_header,
2180                    proto->slab == NULL ? "no" : "yes",
2181                    module_name(proto->owner),
2182                    proto_method_implemented(proto->close),
2183                    proto_method_implemented(proto->connect),
2184                    proto_method_implemented(proto->disconnect),
2185                    proto_method_implemented(proto->accept),
2186                    proto_method_implemented(proto->ioctl),
2187                    proto_method_implemented(proto->init),
2188                    proto_method_implemented(proto->destroy),
2189                    proto_method_implemented(proto->shutdown),
2190                    proto_method_implemented(proto->setsockopt),
2191                    proto_method_implemented(proto->getsockopt),
2192                    proto_method_implemented(proto->sendmsg),
2193                    proto_method_implemented(proto->recvmsg),
2194                    proto_method_implemented(proto->sendpage),
2195                    proto_method_implemented(proto->bind),
2196                    proto_method_implemented(proto->backlog_rcv),
2197                    proto_method_implemented(proto->hash),
2198                    proto_method_implemented(proto->unhash),
2199                    proto_method_implemented(proto->get_port),
2200                    proto_method_implemented(proto->enter_memory_pressure));
2201 }
2202
2203 static int proto_seq_show(struct seq_file *seq, void *v)
2204 {
2205         if (v == &proto_list)
2206                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2207                            "protocol",
2208                            "size",
2209                            "sockets",
2210                            "memory",
2211                            "press",
2212                            "maxhdr",
2213                            "slab",
2214                            "module",
2215                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2216         else
2217                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2218         return 0;
2219 }
2220
2221 static const struct seq_operations proto_seq_ops = {
2222         .start  = proto_seq_start,
2223         .next   = proto_seq_next,
2224         .stop   = proto_seq_stop,
2225         .show   = proto_seq_show,
2226 };
2227
2228 static int proto_seq_open(struct inode *inode, struct file *file)
2229 {
2230         return seq_open_net(inode, file, &proto_seq_ops,
2231                             sizeof(struct seq_net_private));
2232 }
2233
2234 static const struct file_operations proto_seq_fops = {
2235         .owner          = THIS_MODULE,
2236         .open           = proto_seq_open,
2237         .read           = seq_read,
2238         .llseek         = seq_lseek,
2239         .release        = seq_release_net,
2240 };
2241
2242 static __net_init int proto_init_net(struct net *net)
2243 {
2244         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2245                 return -ENOMEM;
2246
2247         return 0;
2248 }
2249
2250 static __net_exit void proto_exit_net(struct net *net)
2251 {
2252         proc_net_remove(net, "protocols");
2253 }
2254
2255
2256 static __net_initdata struct pernet_operations proto_net_ops = {
2257         .init = proto_init_net,
2258         .exit = proto_exit_net,
2259 };
2260
2261 static int __init proto_init(void)
2262 {
2263         return register_pernet_subsys(&proto_net_ops);
2264 }
2265
2266 subsys_initcall(proto_init);
2267
2268 #endif /* PROC_FS */
2269
2270 EXPORT_SYMBOL(sk_alloc);
2271 EXPORT_SYMBOL(sk_free);
2272 EXPORT_SYMBOL(sk_send_sigurg);
2273 EXPORT_SYMBOL(sock_alloc_send_skb);
2274 EXPORT_SYMBOL(sock_init_data);
2275 EXPORT_SYMBOL(sock_kfree_s);
2276 EXPORT_SYMBOL(sock_kmalloc);
2277 EXPORT_SYMBOL(sock_no_accept);
2278 EXPORT_SYMBOL(sock_no_bind);
2279 EXPORT_SYMBOL(sock_no_connect);
2280 EXPORT_SYMBOL(sock_no_getname);
2281 EXPORT_SYMBOL(sock_no_getsockopt);
2282 EXPORT_SYMBOL(sock_no_ioctl);
2283 EXPORT_SYMBOL(sock_no_listen);
2284 EXPORT_SYMBOL(sock_no_mmap);
2285 EXPORT_SYMBOL(sock_no_poll);
2286 EXPORT_SYMBOL(sock_no_recvmsg);
2287 EXPORT_SYMBOL(sock_no_sendmsg);
2288 EXPORT_SYMBOL(sock_no_sendpage);
2289 EXPORT_SYMBOL(sock_no_setsockopt);
2290 EXPORT_SYMBOL(sock_no_shutdown);
2291 EXPORT_SYMBOL(sock_no_socketpair);
2292 EXPORT_SYMBOL(sock_rfree);
2293 EXPORT_SYMBOL(sock_setsockopt);
2294 EXPORT_SYMBOL(sock_wfree);
2295 EXPORT_SYMBOL(sock_wmalloc);
2296 EXPORT_SYMBOL(sock_i_uid);
2297 EXPORT_SYMBOL(sock_i_ino);
2298 EXPORT_SYMBOL(sysctl_optmem_max);