Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
[sfrench/cifs-2.6.git] / net / core / sock.c
1 /*
2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
3  *              operating system.  INET is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              Generic socket support routines. Memory allocators, socket lock/release
7  *              handler for protocols to use and generic option handler.
8  *
9  *
10  * Authors:     Ross Biro
11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *              Florian La Roche, <flla@stud.uni-sb.de>
13  *              Alan Cox, <A.Cox@swansea.ac.uk>
14  *
15  * Fixes:
16  *              Alan Cox        :       Numerous verify_area() problems
17  *              Alan Cox        :       Connecting on a connecting socket
18  *                                      now returns an error for tcp.
19  *              Alan Cox        :       sock->protocol is set correctly.
20  *                                      and is not sometimes left as 0.
21  *              Alan Cox        :       connect handles icmp errors on a
22  *                                      connect properly. Unfortunately there
23  *                                      is a restart syscall nasty there. I
24  *                                      can't match BSD without hacking the C
25  *                                      library. Ideas urgently sought!
26  *              Alan Cox        :       Disallow bind() to addresses that are
27  *                                      not ours - especially broadcast ones!!
28  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
29  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
30  *                                      instead they leave that for the DESTROY timer.
31  *              Alan Cox        :       Clean up error flag in accept
32  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
33  *                                      was buggy. Put a remove_sock() in the handler
34  *                                      for memory when we hit 0. Also altered the timer
35  *                                      code. The ACK stuff can wait and needs major
36  *                                      TCP layer surgery.
37  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
38  *                                      and fixed timer/inet_bh race.
39  *              Alan Cox        :       Added zapped flag for TCP
40  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
41  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
43  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
44  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
46  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
47  *      Pauline Middelink       :       identd support
48  *              Alan Cox        :       Fixed connect() taking signals I think.
49  *              Alan Cox        :       SO_LINGER supported
50  *              Alan Cox        :       Error reporting fixes
51  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
52  *              Alan Cox        :       inet sockets don't set sk->type!
53  *              Alan Cox        :       Split socket option code
54  *              Alan Cox        :       Callbacks
55  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
56  *              Alex            :       Removed restriction on inet fioctl
57  *              Alan Cox        :       Splitting INET from NET core
58  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
59  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
60  *              Alan Cox        :       Split IP from generic code
61  *              Alan Cox        :       New kfree_skbmem()
62  *              Alan Cox        :       Make SO_DEBUG superuser only.
63  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
64  *                                      (compatibility fix)
65  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
66  *              Alan Cox        :       Allocator for a socket is settable.
67  *              Alan Cox        :       SO_ERROR includes soft errors.
68  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
69  *              Alan Cox        :       Generic socket allocation to make hooks
70  *                                      easier (suggested by Craig Metz).
71  *              Michael Pall    :       SO_ERROR returns positive errno again
72  *              Steve Whitehouse:       Added default destructor to free
73  *                                      protocol private data.
74  *              Steve Whitehouse:       Added various other default routines
75  *                                      common to several socket families.
76  *              Chris Evans     :       Call suser() check last on F_SETOWN
77  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
79  *              Andi Kleen      :       Fix write_space callback
80  *              Chris Evans     :       Security fixes - signedness again
81  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
82  *
83  * To Fix:
84  *
85  *
86  *              This program is free software; you can redistribute it and/or
87  *              modify it under the terms of the GNU General Public License
88  *              as published by the Free Software Foundation; either version
89  *              2 of the License, or (at your option) any later version.
90  */
91
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
96 #include <linux/in.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113
114 #include <asm/uaccess.h>
115 #include <asm/system.h>
116
117 #include <linux/netdevice.h>
118 #include <net/protocol.h>
119 #include <linux/skbuff.h>
120 #include <net/net_namespace.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <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         memset(&v, 0, sizeof(v));
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         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1141         sk_mem_uncharge(skb->sk, skb->truesize);
1142 }
1143
1144
1145 int sock_i_uid(struct sock *sk)
1146 {
1147         int uid;
1148
1149         read_lock(&sk->sk_callback_lock);
1150         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1151         read_unlock(&sk->sk_callback_lock);
1152         return uid;
1153 }
1154
1155 unsigned long sock_i_ino(struct sock *sk)
1156 {
1157         unsigned long ino;
1158
1159         read_lock(&sk->sk_callback_lock);
1160         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1161         read_unlock(&sk->sk_callback_lock);
1162         return ino;
1163 }
1164
1165 /*
1166  * Allocate a skb from the socket's send buffer.
1167  */
1168 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1169                              gfp_t priority)
1170 {
1171         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1172                 struct sk_buff * skb = alloc_skb(size, priority);
1173                 if (skb) {
1174                         skb_set_owner_w(skb, sk);
1175                         return skb;
1176                 }
1177         }
1178         return NULL;
1179 }
1180
1181 /*
1182  * Allocate a skb from the socket's receive buffer.
1183  */
1184 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1185                              gfp_t priority)
1186 {
1187         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1188                 struct sk_buff *skb = alloc_skb(size, priority);
1189                 if (skb) {
1190                         skb_set_owner_r(skb, sk);
1191                         return skb;
1192                 }
1193         }
1194         return NULL;
1195 }
1196
1197 /*
1198  * Allocate a memory block from the socket's option memory buffer.
1199  */
1200 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1201 {
1202         if ((unsigned)size <= sysctl_optmem_max &&
1203             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1204                 void *mem;
1205                 /* First do the add, to avoid the race if kmalloc
1206                  * might sleep.
1207                  */
1208                 atomic_add(size, &sk->sk_omem_alloc);
1209                 mem = kmalloc(size, priority);
1210                 if (mem)
1211                         return mem;
1212                 atomic_sub(size, &sk->sk_omem_alloc);
1213         }
1214         return NULL;
1215 }
1216
1217 /*
1218  * Free an option memory block.
1219  */
1220 void sock_kfree_s(struct sock *sk, void *mem, int size)
1221 {
1222         kfree(mem);
1223         atomic_sub(size, &sk->sk_omem_alloc);
1224 }
1225
1226 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1227    I think, these locks should be removed for datagram sockets.
1228  */
1229 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1230 {
1231         DEFINE_WAIT(wait);
1232
1233         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1234         for (;;) {
1235                 if (!timeo)
1236                         break;
1237                 if (signal_pending(current))
1238                         break;
1239                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1240                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1241                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1242                         break;
1243                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1244                         break;
1245                 if (sk->sk_err)
1246                         break;
1247                 timeo = schedule_timeout(timeo);
1248         }
1249         finish_wait(sk->sk_sleep, &wait);
1250         return timeo;
1251 }
1252
1253
1254 /*
1255  *      Generic send/receive buffer handlers
1256  */
1257
1258 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1259                                             unsigned long header_len,
1260                                             unsigned long data_len,
1261                                             int noblock, 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
1342 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1343                                     int noblock, int *errcode)
1344 {
1345         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1346 }
1347
1348 static void __lock_sock(struct sock *sk)
1349 {
1350         DEFINE_WAIT(wait);
1351
1352         for (;;) {
1353                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1354                                         TASK_UNINTERRUPTIBLE);
1355                 spin_unlock_bh(&sk->sk_lock.slock);
1356                 schedule();
1357                 spin_lock_bh(&sk->sk_lock.slock);
1358                 if (!sock_owned_by_user(sk))
1359                         break;
1360         }
1361         finish_wait(&sk->sk_lock.wq, &wait);
1362 }
1363
1364 static void __release_sock(struct sock *sk)
1365 {
1366         struct sk_buff *skb = sk->sk_backlog.head;
1367
1368         do {
1369                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1370                 bh_unlock_sock(sk);
1371
1372                 do {
1373                         struct sk_buff *next = skb->next;
1374
1375                         skb->next = NULL;
1376                         sk_backlog_rcv(sk, skb);
1377
1378                         /*
1379                          * We are in process context here with softirqs
1380                          * disabled, use cond_resched_softirq() to preempt.
1381                          * This is safe to do because we've taken the backlog
1382                          * queue private:
1383                          */
1384                         cond_resched_softirq();
1385
1386                         skb = next;
1387                 } while (skb != NULL);
1388
1389                 bh_lock_sock(sk);
1390         } while ((skb = sk->sk_backlog.head) != NULL);
1391 }
1392
1393 /**
1394  * sk_wait_data - wait for data to arrive at sk_receive_queue
1395  * @sk:    sock to wait on
1396  * @timeo: for how long
1397  *
1398  * Now socket state including sk->sk_err is changed only under lock,
1399  * hence we may omit checks after joining wait queue.
1400  * We check receive queue before schedule() only as optimization;
1401  * it is very likely that release_sock() added new data.
1402  */
1403 int sk_wait_data(struct sock *sk, long *timeo)
1404 {
1405         int rc;
1406         DEFINE_WAIT(wait);
1407
1408         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1409         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1410         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1411         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1412         finish_wait(sk->sk_sleep, &wait);
1413         return rc;
1414 }
1415
1416 EXPORT_SYMBOL(sk_wait_data);
1417
1418 /**
1419  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1420  *      @sk: socket
1421  *      @size: memory size to allocate
1422  *      @kind: allocation type
1423  *
1424  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1425  *      rmem allocation. This function assumes that protocols which have
1426  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1427  */
1428 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1429 {
1430         struct proto *prot = sk->sk_prot;
1431         int amt = sk_mem_pages(size);
1432         int allocated;
1433
1434         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1435         allocated = atomic_add_return(amt, prot->memory_allocated);
1436
1437         /* Under limit. */
1438         if (allocated <= prot->sysctl_mem[0]) {
1439                 if (prot->memory_pressure && *prot->memory_pressure)
1440                         *prot->memory_pressure = 0;
1441                 return 1;
1442         }
1443
1444         /* Under pressure. */
1445         if (allocated > prot->sysctl_mem[1])
1446                 if (prot->enter_memory_pressure)
1447                         prot->enter_memory_pressure(sk);
1448
1449         /* Over hard limit. */
1450         if (allocated > prot->sysctl_mem[2])
1451                 goto suppress_allocation;
1452
1453         /* guarantee minimum buffer size under pressure */
1454         if (kind == SK_MEM_RECV) {
1455                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1456                         return 1;
1457         } else { /* SK_MEM_SEND */
1458                 if (sk->sk_type == SOCK_STREAM) {
1459                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1460                                 return 1;
1461                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1462                            prot->sysctl_wmem[0])
1463                                 return 1;
1464         }
1465
1466         if (prot->memory_pressure) {
1467                 int alloc;
1468
1469                 if (!*prot->memory_pressure)
1470                         return 1;
1471                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1472                 if (prot->sysctl_mem[2] > alloc *
1473                     sk_mem_pages(sk->sk_wmem_queued +
1474                                  atomic_read(&sk->sk_rmem_alloc) +
1475                                  sk->sk_forward_alloc))
1476                         return 1;
1477         }
1478
1479 suppress_allocation:
1480
1481         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1482                 sk_stream_moderate_sndbuf(sk);
1483
1484                 /* Fail only if socket is _under_ its sndbuf.
1485                  * In this case we cannot block, so that we have to fail.
1486                  */
1487                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1488                         return 1;
1489         }
1490
1491         /* Alas. Undo changes. */
1492         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1493         atomic_sub(amt, prot->memory_allocated);
1494         return 0;
1495 }
1496
1497 EXPORT_SYMBOL(__sk_mem_schedule);
1498
1499 /**
1500  *      __sk_reclaim - reclaim memory_allocated
1501  *      @sk: socket
1502  */
1503 void __sk_mem_reclaim(struct sock *sk)
1504 {
1505         struct proto *prot = sk->sk_prot;
1506
1507         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1508                    prot->memory_allocated);
1509         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1510
1511         if (prot->memory_pressure && *prot->memory_pressure &&
1512             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1513                 *prot->memory_pressure = 0;
1514 }
1515
1516 EXPORT_SYMBOL(__sk_mem_reclaim);
1517
1518
1519 /*
1520  * Set of default routines for initialising struct proto_ops when
1521  * the protocol does not support a particular function. In certain
1522  * cases where it makes no sense for a protocol to have a "do nothing"
1523  * function, some default processing is provided.
1524  */
1525
1526 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1527 {
1528         return -EOPNOTSUPP;
1529 }
1530
1531 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1532                     int len, int flags)
1533 {
1534         return -EOPNOTSUPP;
1535 }
1536
1537 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1538 {
1539         return -EOPNOTSUPP;
1540 }
1541
1542 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1543 {
1544         return -EOPNOTSUPP;
1545 }
1546
1547 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1548                     int *len, int peer)
1549 {
1550         return -EOPNOTSUPP;
1551 }
1552
1553 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1554 {
1555         return 0;
1556 }
1557
1558 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1559 {
1560         return -EOPNOTSUPP;
1561 }
1562
1563 int sock_no_listen(struct socket *sock, int backlog)
1564 {
1565         return -EOPNOTSUPP;
1566 }
1567
1568 int sock_no_shutdown(struct socket *sock, int how)
1569 {
1570         return -EOPNOTSUPP;
1571 }
1572
1573 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1574                     char __user *optval, int optlen)
1575 {
1576         return -EOPNOTSUPP;
1577 }
1578
1579 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1580                     char __user *optval, int __user *optlen)
1581 {
1582         return -EOPNOTSUPP;
1583 }
1584
1585 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1586                     size_t len)
1587 {
1588         return -EOPNOTSUPP;
1589 }
1590
1591 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1592                     size_t len, int flags)
1593 {
1594         return -EOPNOTSUPP;
1595 }
1596
1597 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1598 {
1599         /* Mirror missing mmap method error code */
1600         return -ENODEV;
1601 }
1602
1603 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1604 {
1605         ssize_t res;
1606         struct msghdr msg = {.msg_flags = flags};
1607         struct kvec iov;
1608         char *kaddr = kmap(page);
1609         iov.iov_base = kaddr + offset;
1610         iov.iov_len = size;
1611         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1612         kunmap(page);
1613         return res;
1614 }
1615
1616 /*
1617  *      Default Socket Callbacks
1618  */
1619
1620 static void sock_def_wakeup(struct sock *sk)
1621 {
1622         read_lock(&sk->sk_callback_lock);
1623         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1624                 wake_up_interruptible_all(sk->sk_sleep);
1625         read_unlock(&sk->sk_callback_lock);
1626 }
1627
1628 static void sock_def_error_report(struct sock *sk)
1629 {
1630         read_lock(&sk->sk_callback_lock);
1631         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1632                 wake_up_interruptible(sk->sk_sleep);
1633         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1634         read_unlock(&sk->sk_callback_lock);
1635 }
1636
1637 static void sock_def_readable(struct sock *sk, int len)
1638 {
1639         read_lock(&sk->sk_callback_lock);
1640         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1641                 wake_up_interruptible_sync(sk->sk_sleep);
1642         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1643         read_unlock(&sk->sk_callback_lock);
1644 }
1645
1646 static void sock_def_write_space(struct sock *sk)
1647 {
1648         read_lock(&sk->sk_callback_lock);
1649
1650         /* Do not wake up a writer until he can make "significant"
1651          * progress.  --DaveM
1652          */
1653         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1654                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1655                         wake_up_interruptible_sync(sk->sk_sleep);
1656
1657                 /* Should agree with poll, otherwise some programs break */
1658                 if (sock_writeable(sk))
1659                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1660         }
1661
1662         read_unlock(&sk->sk_callback_lock);
1663 }
1664
1665 static void sock_def_destruct(struct sock *sk)
1666 {
1667         kfree(sk->sk_protinfo);
1668 }
1669
1670 void sk_send_sigurg(struct sock *sk)
1671 {
1672         if (sk->sk_socket && sk->sk_socket->file)
1673                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1674                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1675 }
1676
1677 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1678                     unsigned long expires)
1679 {
1680         if (!mod_timer(timer, expires))
1681                 sock_hold(sk);
1682 }
1683
1684 EXPORT_SYMBOL(sk_reset_timer);
1685
1686 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1687 {
1688         if (timer_pending(timer) && del_timer(timer))
1689                 __sock_put(sk);
1690 }
1691
1692 EXPORT_SYMBOL(sk_stop_timer);
1693
1694 void sock_init_data(struct socket *sock, struct sock *sk)
1695 {
1696         skb_queue_head_init(&sk->sk_receive_queue);
1697         skb_queue_head_init(&sk->sk_write_queue);
1698         skb_queue_head_init(&sk->sk_error_queue);
1699 #ifdef CONFIG_NET_DMA
1700         skb_queue_head_init(&sk->sk_async_wait_queue);
1701 #endif
1702
1703         sk->sk_send_head        =       NULL;
1704
1705         init_timer(&sk->sk_timer);
1706
1707         sk->sk_allocation       =       GFP_KERNEL;
1708         sk->sk_rcvbuf           =       sysctl_rmem_default;
1709         sk->sk_sndbuf           =       sysctl_wmem_default;
1710         sk->sk_state            =       TCP_CLOSE;
1711         sk_set_socket(sk, sock);
1712
1713         sock_set_flag(sk, SOCK_ZAPPED);
1714
1715         if (sock) {
1716                 sk->sk_type     =       sock->type;
1717                 sk->sk_sleep    =       &sock->wait;
1718                 sock->sk        =       sk;
1719         } else
1720                 sk->sk_sleep    =       NULL;
1721
1722         rwlock_init(&sk->sk_dst_lock);
1723         rwlock_init(&sk->sk_callback_lock);
1724         lockdep_set_class_and_name(&sk->sk_callback_lock,
1725                         af_callback_keys + sk->sk_family,
1726                         af_family_clock_key_strings[sk->sk_family]);
1727
1728         sk->sk_state_change     =       sock_def_wakeup;
1729         sk->sk_data_ready       =       sock_def_readable;
1730         sk->sk_write_space      =       sock_def_write_space;
1731         sk->sk_error_report     =       sock_def_error_report;
1732         sk->sk_destruct         =       sock_def_destruct;
1733
1734         sk->sk_sndmsg_page      =       NULL;
1735         sk->sk_sndmsg_off       =       0;
1736
1737         sk->sk_peercred.pid     =       0;
1738         sk->sk_peercred.uid     =       -1;
1739         sk->sk_peercred.gid     =       -1;
1740         sk->sk_write_pending    =       0;
1741         sk->sk_rcvlowat         =       1;
1742         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1743         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1744
1745         sk->sk_stamp = ktime_set(-1L, 0);
1746
1747         atomic_set(&sk->sk_refcnt, 1);
1748         atomic_set(&sk->sk_drops, 0);
1749 }
1750
1751 void lock_sock_nested(struct sock *sk, int subclass)
1752 {
1753         might_sleep();
1754         spin_lock_bh(&sk->sk_lock.slock);
1755         if (sk->sk_lock.owned)
1756                 __lock_sock(sk);
1757         sk->sk_lock.owned = 1;
1758         spin_unlock(&sk->sk_lock.slock);
1759         /*
1760          * The sk_lock has mutex_lock() semantics here:
1761          */
1762         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1763         local_bh_enable();
1764 }
1765
1766 EXPORT_SYMBOL(lock_sock_nested);
1767
1768 void release_sock(struct sock *sk)
1769 {
1770         /*
1771          * The sk_lock has mutex_unlock() semantics:
1772          */
1773         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1774
1775         spin_lock_bh(&sk->sk_lock.slock);
1776         if (sk->sk_backlog.tail)
1777                 __release_sock(sk);
1778         sk->sk_lock.owned = 0;
1779         if (waitqueue_active(&sk->sk_lock.wq))
1780                 wake_up(&sk->sk_lock.wq);
1781         spin_unlock_bh(&sk->sk_lock.slock);
1782 }
1783 EXPORT_SYMBOL(release_sock);
1784
1785 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1786 {
1787         struct timeval tv;
1788         if (!sock_flag(sk, SOCK_TIMESTAMP))
1789                 sock_enable_timestamp(sk);
1790         tv = ktime_to_timeval(sk->sk_stamp);
1791         if (tv.tv_sec == -1)
1792                 return -ENOENT;
1793         if (tv.tv_sec == 0) {
1794                 sk->sk_stamp = ktime_get_real();
1795                 tv = ktime_to_timeval(sk->sk_stamp);
1796         }
1797         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1798 }
1799 EXPORT_SYMBOL(sock_get_timestamp);
1800
1801 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1802 {
1803         struct timespec ts;
1804         if (!sock_flag(sk, SOCK_TIMESTAMP))
1805                 sock_enable_timestamp(sk);
1806         ts = ktime_to_timespec(sk->sk_stamp);
1807         if (ts.tv_sec == -1)
1808                 return -ENOENT;
1809         if (ts.tv_sec == 0) {
1810                 sk->sk_stamp = ktime_get_real();
1811                 ts = ktime_to_timespec(sk->sk_stamp);
1812         }
1813         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1814 }
1815 EXPORT_SYMBOL(sock_get_timestampns);
1816
1817 void sock_enable_timestamp(struct sock *sk)
1818 {
1819         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1820                 sock_set_flag(sk, SOCK_TIMESTAMP);
1821                 net_enable_timestamp();
1822         }
1823 }
1824
1825 /*
1826  *      Get a socket option on an socket.
1827  *
1828  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1829  *      asynchronous errors should be reported by getsockopt. We assume
1830  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1831  */
1832 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1833                            char __user *optval, int __user *optlen)
1834 {
1835         struct sock *sk = sock->sk;
1836
1837         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1838 }
1839
1840 EXPORT_SYMBOL(sock_common_getsockopt);
1841
1842 #ifdef CONFIG_COMPAT
1843 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1844                                   char __user *optval, int __user *optlen)
1845 {
1846         struct sock *sk = sock->sk;
1847
1848         if (sk->sk_prot->compat_getsockopt != NULL)
1849                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1850                                                       optval, optlen);
1851         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1852 }
1853 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1854 #endif
1855
1856 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1857                         struct msghdr *msg, size_t size, int flags)
1858 {
1859         struct sock *sk = sock->sk;
1860         int addr_len = 0;
1861         int err;
1862
1863         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1864                                    flags & ~MSG_DONTWAIT, &addr_len);
1865         if (err >= 0)
1866                 msg->msg_namelen = addr_len;
1867         return err;
1868 }
1869
1870 EXPORT_SYMBOL(sock_common_recvmsg);
1871
1872 /*
1873  *      Set socket options on an inet socket.
1874  */
1875 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1876                            char __user *optval, int optlen)
1877 {
1878         struct sock *sk = sock->sk;
1879
1880         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1881 }
1882
1883 EXPORT_SYMBOL(sock_common_setsockopt);
1884
1885 #ifdef CONFIG_COMPAT
1886 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1887                                   char __user *optval, int optlen)
1888 {
1889         struct sock *sk = sock->sk;
1890
1891         if (sk->sk_prot->compat_setsockopt != NULL)
1892                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1893                                                       optval, optlen);
1894         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1895 }
1896 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1897 #endif
1898
1899 void sk_common_release(struct sock *sk)
1900 {
1901         if (sk->sk_prot->destroy)
1902                 sk->sk_prot->destroy(sk);
1903
1904         /*
1905          * Observation: when sock_common_release is called, processes have
1906          * no access to socket. But net still has.
1907          * Step one, detach it from networking:
1908          *
1909          * A. Remove from hash tables.
1910          */
1911
1912         sk->sk_prot->unhash(sk);
1913
1914         /*
1915          * In this point socket cannot receive new packets, but it is possible
1916          * that some packets are in flight because some CPU runs receiver and
1917          * did hash table lookup before we unhashed socket. They will achieve
1918          * receive queue and will be purged by socket destructor.
1919          *
1920          * Also we still have packets pending on receive queue and probably,
1921          * our own packets waiting in device queues. sock_destroy will drain
1922          * receive queue, but transmitted packets will delay socket destruction
1923          * until the last reference will be released.
1924          */
1925
1926         sock_orphan(sk);
1927
1928         xfrm_sk_free_policy(sk);
1929
1930         sk_refcnt_debug_release(sk);
1931         sock_put(sk);
1932 }
1933
1934 EXPORT_SYMBOL(sk_common_release);
1935
1936 static DEFINE_RWLOCK(proto_list_lock);
1937 static LIST_HEAD(proto_list);
1938
1939 #ifdef CONFIG_PROC_FS
1940 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
1941 struct prot_inuse {
1942         int val[PROTO_INUSE_NR];
1943 };
1944
1945 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
1946
1947 #ifdef CONFIG_NET_NS
1948 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1949 {
1950         int cpu = smp_processor_id();
1951         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
1952 }
1953 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
1954
1955 int sock_prot_inuse_get(struct net *net, struct proto *prot)
1956 {
1957         int cpu, idx = prot->inuse_idx;
1958         int res = 0;
1959
1960         for_each_possible_cpu(cpu)
1961                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
1962
1963         return res >= 0 ? res : 0;
1964 }
1965 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
1966
1967 static int sock_inuse_init_net(struct net *net)
1968 {
1969         net->core.inuse = alloc_percpu(struct prot_inuse);
1970         return net->core.inuse ? 0 : -ENOMEM;
1971 }
1972
1973 static void sock_inuse_exit_net(struct net *net)
1974 {
1975         free_percpu(net->core.inuse);
1976 }
1977
1978 static struct pernet_operations net_inuse_ops = {
1979         .init = sock_inuse_init_net,
1980         .exit = sock_inuse_exit_net,
1981 };
1982
1983 static __init int net_inuse_init(void)
1984 {
1985         if (register_pernet_subsys(&net_inuse_ops))
1986                 panic("Cannot initialize net inuse counters");
1987
1988         return 0;
1989 }
1990
1991 core_initcall(net_inuse_init);
1992 #else
1993 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
1994
1995 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1996 {
1997         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
1998 }
1999 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2000
2001 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2002 {
2003         int cpu, idx = prot->inuse_idx;
2004         int res = 0;
2005
2006         for_each_possible_cpu(cpu)
2007                 res += per_cpu(prot_inuse, cpu).val[idx];
2008
2009         return res >= 0 ? res : 0;
2010 }
2011 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2012 #endif
2013
2014 static void assign_proto_idx(struct proto *prot)
2015 {
2016         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2017
2018         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2019                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2020                 return;
2021         }
2022
2023         set_bit(prot->inuse_idx, proto_inuse_idx);
2024 }
2025
2026 static void release_proto_idx(struct proto *prot)
2027 {
2028         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2029                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2030 }
2031 #else
2032 static inline void assign_proto_idx(struct proto *prot)
2033 {
2034 }
2035
2036 static inline void release_proto_idx(struct proto *prot)
2037 {
2038 }
2039 #endif
2040
2041 int proto_register(struct proto *prot, int alloc_slab)
2042 {
2043         if (alloc_slab) {
2044                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2045                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2046                                         NULL);
2047
2048                 if (prot->slab == NULL) {
2049                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2050                                prot->name);
2051                         goto out;
2052                 }
2053
2054                 if (prot->rsk_prot != NULL) {
2055                         static const char mask[] = "request_sock_%s";
2056
2057                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2058                         if (prot->rsk_prot->slab_name == NULL)
2059                                 goto out_free_sock_slab;
2060
2061                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2062                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2063                                                                  prot->rsk_prot->obj_size, 0,
2064                                                                  SLAB_HWCACHE_ALIGN, NULL);
2065
2066                         if (prot->rsk_prot->slab == NULL) {
2067                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2068                                        prot->name);
2069                                 goto out_free_request_sock_slab_name;
2070                         }
2071                 }
2072
2073                 if (prot->twsk_prot != NULL) {
2074                         static const char mask[] = "tw_sock_%s";
2075
2076                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2077
2078                         if (prot->twsk_prot->twsk_slab_name == NULL)
2079                                 goto out_free_request_sock_slab;
2080
2081                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2082                         prot->twsk_prot->twsk_slab =
2083                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2084                                                   prot->twsk_prot->twsk_obj_size,
2085                                                   0,
2086                                                   SLAB_HWCACHE_ALIGN |
2087                                                         prot->slab_flags,
2088                                                   NULL);
2089                         if (prot->twsk_prot->twsk_slab == NULL)
2090                                 goto out_free_timewait_sock_slab_name;
2091                 }
2092         }
2093
2094         write_lock(&proto_list_lock);
2095         list_add(&prot->node, &proto_list);
2096         assign_proto_idx(prot);
2097         write_unlock(&proto_list_lock);
2098         return 0;
2099
2100 out_free_timewait_sock_slab_name:
2101         kfree(prot->twsk_prot->twsk_slab_name);
2102 out_free_request_sock_slab:
2103         if (prot->rsk_prot && prot->rsk_prot->slab) {
2104                 kmem_cache_destroy(prot->rsk_prot->slab);
2105                 prot->rsk_prot->slab = NULL;
2106         }
2107 out_free_request_sock_slab_name:
2108         kfree(prot->rsk_prot->slab_name);
2109 out_free_sock_slab:
2110         kmem_cache_destroy(prot->slab);
2111         prot->slab = NULL;
2112 out:
2113         return -ENOBUFS;
2114 }
2115
2116 EXPORT_SYMBOL(proto_register);
2117
2118 void proto_unregister(struct proto *prot)
2119 {
2120         write_lock(&proto_list_lock);
2121         release_proto_idx(prot);
2122         list_del(&prot->node);
2123         write_unlock(&proto_list_lock);
2124
2125         if (prot->slab != NULL) {
2126                 kmem_cache_destroy(prot->slab);
2127                 prot->slab = NULL;
2128         }
2129
2130         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2131                 kmem_cache_destroy(prot->rsk_prot->slab);
2132                 kfree(prot->rsk_prot->slab_name);
2133                 prot->rsk_prot->slab = NULL;
2134         }
2135
2136         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2137                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2138                 kfree(prot->twsk_prot->twsk_slab_name);
2139                 prot->twsk_prot->twsk_slab = NULL;
2140         }
2141 }
2142
2143 EXPORT_SYMBOL(proto_unregister);
2144
2145 #ifdef CONFIG_PROC_FS
2146 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2147         __acquires(proto_list_lock)
2148 {
2149         read_lock(&proto_list_lock);
2150         return seq_list_start_head(&proto_list, *pos);
2151 }
2152
2153 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2154 {
2155         return seq_list_next(v, &proto_list, pos);
2156 }
2157
2158 static void proto_seq_stop(struct seq_file *seq, void *v)
2159         __releases(proto_list_lock)
2160 {
2161         read_unlock(&proto_list_lock);
2162 }
2163
2164 static char proto_method_implemented(const void *method)
2165 {
2166         return method == NULL ? 'n' : 'y';
2167 }
2168
2169 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2170 {
2171         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2172                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2173                    proto->name,
2174                    proto->obj_size,
2175                    sock_prot_inuse_get(seq_file_net(seq), proto),
2176                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2177                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2178                    proto->max_header,
2179                    proto->slab == NULL ? "no" : "yes",
2180                    module_name(proto->owner),
2181                    proto_method_implemented(proto->close),
2182                    proto_method_implemented(proto->connect),
2183                    proto_method_implemented(proto->disconnect),
2184                    proto_method_implemented(proto->accept),
2185                    proto_method_implemented(proto->ioctl),
2186                    proto_method_implemented(proto->init),
2187                    proto_method_implemented(proto->destroy),
2188                    proto_method_implemented(proto->shutdown),
2189                    proto_method_implemented(proto->setsockopt),
2190                    proto_method_implemented(proto->getsockopt),
2191                    proto_method_implemented(proto->sendmsg),
2192                    proto_method_implemented(proto->recvmsg),
2193                    proto_method_implemented(proto->sendpage),
2194                    proto_method_implemented(proto->bind),
2195                    proto_method_implemented(proto->backlog_rcv),
2196                    proto_method_implemented(proto->hash),
2197                    proto_method_implemented(proto->unhash),
2198                    proto_method_implemented(proto->get_port),
2199                    proto_method_implemented(proto->enter_memory_pressure));
2200 }
2201
2202 static int proto_seq_show(struct seq_file *seq, void *v)
2203 {
2204         if (v == &proto_list)
2205                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2206                            "protocol",
2207                            "size",
2208                            "sockets",
2209                            "memory",
2210                            "press",
2211                            "maxhdr",
2212                            "slab",
2213                            "module",
2214                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2215         else
2216                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2217         return 0;
2218 }
2219
2220 static const struct seq_operations proto_seq_ops = {
2221         .start  = proto_seq_start,
2222         .next   = proto_seq_next,
2223         .stop   = proto_seq_stop,
2224         .show   = proto_seq_show,
2225 };
2226
2227 static int proto_seq_open(struct inode *inode, struct file *file)
2228 {
2229         return seq_open_net(inode, file, &proto_seq_ops,
2230                             sizeof(struct seq_net_private));
2231 }
2232
2233 static const struct file_operations proto_seq_fops = {
2234         .owner          = THIS_MODULE,
2235         .open           = proto_seq_open,
2236         .read           = seq_read,
2237         .llseek         = seq_lseek,
2238         .release        = seq_release_net,
2239 };
2240
2241 static __net_init int proto_init_net(struct net *net)
2242 {
2243         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2244                 return -ENOMEM;
2245
2246         return 0;
2247 }
2248
2249 static __net_exit void proto_exit_net(struct net *net)
2250 {
2251         proc_net_remove(net, "protocols");
2252 }
2253
2254
2255 static __net_initdata struct pernet_operations proto_net_ops = {
2256         .init = proto_init_net,
2257         .exit = proto_exit_net,
2258 };
2259
2260 static int __init proto_init(void)
2261 {
2262         return register_pernet_subsys(&proto_net_ops);
2263 }
2264
2265 subsys_initcall(proto_init);
2266
2267 #endif /* PROC_FS */
2268
2269 EXPORT_SYMBOL(sk_alloc);
2270 EXPORT_SYMBOL(sk_free);
2271 EXPORT_SYMBOL(sk_send_sigurg);
2272 EXPORT_SYMBOL(sock_alloc_send_skb);
2273 EXPORT_SYMBOL(sock_init_data);
2274 EXPORT_SYMBOL(sock_kfree_s);
2275 EXPORT_SYMBOL(sock_kmalloc);
2276 EXPORT_SYMBOL(sock_no_accept);
2277 EXPORT_SYMBOL(sock_no_bind);
2278 EXPORT_SYMBOL(sock_no_connect);
2279 EXPORT_SYMBOL(sock_no_getname);
2280 EXPORT_SYMBOL(sock_no_getsockopt);
2281 EXPORT_SYMBOL(sock_no_ioctl);
2282 EXPORT_SYMBOL(sock_no_listen);
2283 EXPORT_SYMBOL(sock_no_mmap);
2284 EXPORT_SYMBOL(sock_no_poll);
2285 EXPORT_SYMBOL(sock_no_recvmsg);
2286 EXPORT_SYMBOL(sock_no_sendmsg);
2287 EXPORT_SYMBOL(sock_no_sendpage);
2288 EXPORT_SYMBOL(sock_no_setsockopt);
2289 EXPORT_SYMBOL(sock_no_shutdown);
2290 EXPORT_SYMBOL(sock_no_socketpair);
2291 EXPORT_SYMBOL(sock_rfree);
2292 EXPORT_SYMBOL(sock_setsockopt);
2293 EXPORT_SYMBOL(sock_wfree);
2294 EXPORT_SYMBOL(sock_wmalloc);
2295 EXPORT_SYMBOL(sock_i_uid);
2296 EXPORT_SYMBOL(sock_i_ino);
2297 EXPORT_SYMBOL(sysctl_optmem_max);