net: add a noref bit on skb dst
[linux-2.6.git] / net / core / sock.c
1 /*
2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
3  *              operating system.  INET is implemented using the  BSD Socket
4  *              interface as the means of communication with the user level.
5  *
6  *              Generic socket support routines. Memory allocators, socket lock/release
7  *              handler for protocols to use and generic option handler.
8  *
9  *
10  * Authors:     Ross Biro
11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *              Florian La Roche, <flla@stud.uni-sb.de>
13  *              Alan Cox, <A.Cox@swansea.ac.uk>
14  *
15  * Fixes:
16  *              Alan Cox        :       Numerous verify_area() problems
17  *              Alan Cox        :       Connecting on a connecting socket
18  *                                      now returns an error for tcp.
19  *              Alan Cox        :       sock->protocol is set correctly.
20  *                                      and is not sometimes left as 0.
21  *              Alan Cox        :       connect handles icmp errors on a
22  *                                      connect properly. Unfortunately there
23  *                                      is a restart syscall nasty there. I
24  *                                      can't match BSD without hacking the C
25  *                                      library. Ideas urgently sought!
26  *              Alan Cox        :       Disallow bind() to addresses that are
27  *                                      not ours - especially broadcast ones!!
28  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
29  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
30  *                                      instead they leave that for the DESTROY timer.
31  *              Alan Cox        :       Clean up error flag in accept
32  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
33  *                                      was buggy. Put a remove_sock() in the handler
34  *                                      for memory when we hit 0. Also altered the timer
35  *                                      code. The ACK stuff can wait and needs major
36  *                                      TCP layer surgery.
37  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
38  *                                      and fixed timer/inet_bh race.
39  *              Alan Cox        :       Added zapped flag for TCP
40  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
41  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
42  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
43  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
44  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
45  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
46  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
47  *      Pauline Middelink       :       identd support
48  *              Alan Cox        :       Fixed connect() taking signals I think.
49  *              Alan Cox        :       SO_LINGER supported
50  *              Alan Cox        :       Error reporting fixes
51  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
52  *              Alan Cox        :       inet sockets don't set sk->type!
53  *              Alan Cox        :       Split socket option code
54  *              Alan Cox        :       Callbacks
55  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
56  *              Alex            :       Removed restriction on inet fioctl
57  *              Alan Cox        :       Splitting INET from NET core
58  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
59  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
60  *              Alan Cox        :       Split IP from generic code
61  *              Alan Cox        :       New kfree_skbmem()
62  *              Alan Cox        :       Make SO_DEBUG superuser only.
63  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
64  *                                      (compatibility fix)
65  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
66  *              Alan Cox        :       Allocator for a socket is settable.
67  *              Alan Cox        :       SO_ERROR includes soft errors.
68  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
69  *              Alan Cox        :       Generic socket allocation to make hooks
70  *                                      easier (suggested by Craig Metz).
71  *              Michael Pall    :       SO_ERROR returns positive errno again
72  *              Steve Whitehouse:       Added default destructor to free
73  *                                      protocol private data.
74  *              Steve Whitehouse:       Added various other default routines
75  *                                      common to several socket families.
76  *              Chris Evans     :       Call suser() check last on F_SETOWN
77  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
78  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
79  *              Andi Kleen      :       Fix write_space callback
80  *              Chris Evans     :       Security fixes - signedness again
81  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
82  *
83  * To Fix:
84  *
85  *
86  *              This program is free software; you can redistribute it and/or
87  *              modify it under the terms of the GNU General Public License
88  *              as published by the Free Software Foundation; either version
89  *              2 of the License, or (at your option) any later version.
90  */
91
92 #include <linux/capability.h>
93 #include <linux/errno.h>
94 #include <linux/types.h>
95 #include <linux/socket.h>
96 #include <linux/in.h>
97 #include <linux/kernel.h>
98 #include <linux/module.h>
99 #include <linux/proc_fs.h>
100 #include <linux/seq_file.h>
101 #include <linux/sched.h>
102 #include <linux/timer.h>
103 #include <linux/string.h>
104 #include <linux/sockios.h>
105 #include <linux/net.h>
106 #include <linux/mm.h>
107 #include <linux/slab.h>
108 #include <linux/interrupt.h>
109 #include <linux/poll.h>
110 #include <linux/tcp.h>
111 #include <linux/init.h>
112 #include <linux/highmem.h>
113
114 #include <asm/uaccess.h>
115 #include <asm/system.h>
116
117 #include <linux/netdevice.h>
118 #include <net/protocol.h>
119 #include <linux/skbuff.h>
120 #include <net/net_namespace.h>
121 #include <net/request_sock.h>
122 #include <net/sock.h>
123 #include <linux/net_tstamp.h>
124 #include <net/xfrm.h>
125 #include <linux/ipsec.h>
126
127 #include <linux/filter.h>
128
129 #ifdef CONFIG_INET
130 #include <net/tcp.h>
131 #endif
132
133 /*
134  * Each address family might have different locking rules, so we have
135  * one slock key per address family:
136  */
137 static struct lock_class_key af_family_keys[AF_MAX];
138 static struct lock_class_key af_family_slock_keys[AF_MAX];
139
140 /*
141  * Make lock validator output more readable. (we pre-construct these
142  * strings build-time, so that runtime initialization of socket
143  * locks is fast):
144  */
145 static const char *const af_family_key_strings[AF_MAX+1] = {
146   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
147   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
148   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
149   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
150   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
151   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
152   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
153   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
154   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
155   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
156   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
157   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
158   "sk_lock-AF_IEEE802154",
159   "sk_lock-AF_MAX"
160 };
161 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
162   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
163   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
164   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
165   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
166   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
167   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
168   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
169   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
170   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
171   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
172   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
173   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
174   "slock-AF_IEEE802154",
175   "slock-AF_MAX"
176 };
177 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
178   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
179   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
180   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
181   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
182   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
183   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
184   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
185   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
186   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
187   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
188   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
189   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
190   "clock-AF_IEEE802154",
191   "clock-AF_MAX"
192 };
193
194 /*
195  * sk_callback_lock locking rules are per-address-family,
196  * so split the lock classes by using a per-AF key:
197  */
198 static struct lock_class_key af_callback_keys[AF_MAX];
199
200 /* Take into consideration the size of the struct sk_buff overhead in the
201  * determination of these values, since that is non-constant across
202  * platforms.  This makes socket queueing behavior and performance
203  * not depend upon such differences.
204  */
205 #define _SK_MEM_PACKETS         256
206 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
207 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
208 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
209
210 /* Run time adjustable parameters. */
211 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
212 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
213 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
214 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
215
216 /* Maximal space eaten by iovec or ancilliary data plus some space */
217 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
218 EXPORT_SYMBOL(sysctl_optmem_max);
219
220 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
221 {
222         struct timeval tv;
223
224         if (optlen < sizeof(tv))
225                 return -EINVAL;
226         if (copy_from_user(&tv, optval, sizeof(tv)))
227                 return -EFAULT;
228         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
229                 return -EDOM;
230
231         if (tv.tv_sec < 0) {
232                 static int warned __read_mostly;
233
234                 *timeo_p = 0;
235                 if (warned < 10 && net_ratelimit()) {
236                         warned++;
237                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
238                                "tries to set negative timeout\n",
239                                 current->comm, task_pid_nr(current));
240                 }
241                 return 0;
242         }
243         *timeo_p = MAX_SCHEDULE_TIMEOUT;
244         if (tv.tv_sec == 0 && tv.tv_usec == 0)
245                 return 0;
246         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
247                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
248         return 0;
249 }
250
251 static void sock_warn_obsolete_bsdism(const char *name)
252 {
253         static int warned;
254         static char warncomm[TASK_COMM_LEN];
255         if (strcmp(warncomm, current->comm) && warned < 5) {
256                 strcpy(warncomm,  current->comm);
257                 printk(KERN_WARNING "process `%s' is using obsolete "
258                        "%s SO_BSDCOMPAT\n", warncomm, name);
259                 warned++;
260         }
261 }
262
263 static void sock_disable_timestamp(struct sock *sk, int flag)
264 {
265         if (sock_flag(sk, flag)) {
266                 sock_reset_flag(sk, flag);
267                 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
268                     !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
269                         net_disable_timestamp();
270                 }
271         }
272 }
273
274
275 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
276 {
277         int err;
278         int skb_len;
279         unsigned long flags;
280         struct sk_buff_head *list = &sk->sk_receive_queue;
281
282         /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
283            number of warnings when compiling with -W --ANK
284          */
285         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
286             (unsigned)sk->sk_rcvbuf) {
287                 atomic_inc(&sk->sk_drops);
288                 return -ENOMEM;
289         }
290
291         err = sk_filter(sk, skb);
292         if (err)
293                 return err;
294
295         if (!sk_rmem_schedule(sk, skb->truesize)) {
296                 atomic_inc(&sk->sk_drops);
297                 return -ENOBUFS;
298         }
299
300         skb->dev = NULL;
301         skb_set_owner_r(skb, sk);
302
303         /* Cache the SKB length before we tack it onto the receive
304          * queue.  Once it is added it no longer belongs to us and
305          * may be freed by other threads of control pulling packets
306          * from the queue.
307          */
308         skb_len = skb->len;
309
310         /* we escape from rcu protected region, make sure we dont leak
311          * a norefcounted dst
312          */
313         skb_dst_force(skb);
314
315         spin_lock_irqsave(&list->lock, flags);
316         skb->dropcount = atomic_read(&sk->sk_drops);
317         __skb_queue_tail(list, skb);
318         spin_unlock_irqrestore(&list->lock, flags);
319
320         if (!sock_flag(sk, SOCK_DEAD))
321                 sk->sk_data_ready(sk, skb_len);
322         return 0;
323 }
324 EXPORT_SYMBOL(sock_queue_rcv_skb);
325
326 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
327 {
328         int rc = NET_RX_SUCCESS;
329
330         if (sk_filter(sk, skb))
331                 goto discard_and_relse;
332
333         skb->dev = NULL;
334
335         if (sk_rcvqueues_full(sk, skb)) {
336                 atomic_inc(&sk->sk_drops);
337                 goto discard_and_relse;
338         }
339         if (nested)
340                 bh_lock_sock_nested(sk);
341         else
342                 bh_lock_sock(sk);
343         if (!sock_owned_by_user(sk)) {
344                 /*
345                  * trylock + unlock semantics:
346                  */
347                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
348
349                 rc = sk_backlog_rcv(sk, skb);
350
351                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
352         } else if (sk_add_backlog(sk, skb)) {
353                 bh_unlock_sock(sk);
354                 atomic_inc(&sk->sk_drops);
355                 goto discard_and_relse;
356         }
357
358         bh_unlock_sock(sk);
359 out:
360         sock_put(sk);
361         return rc;
362 discard_and_relse:
363         kfree_skb(skb);
364         goto out;
365 }
366 EXPORT_SYMBOL(sk_receive_skb);
367
368 void sk_reset_txq(struct sock *sk)
369 {
370         sk_tx_queue_clear(sk);
371 }
372 EXPORT_SYMBOL(sk_reset_txq);
373
374 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
375 {
376         struct dst_entry *dst = __sk_dst_get(sk);
377
378         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
379                 sk_tx_queue_clear(sk);
380                 rcu_assign_pointer(sk->sk_dst_cache, NULL);
381                 dst_release(dst);
382                 return NULL;
383         }
384
385         return dst;
386 }
387 EXPORT_SYMBOL(__sk_dst_check);
388
389 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
390 {
391         struct dst_entry *dst = sk_dst_get(sk);
392
393         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
394                 sk_dst_reset(sk);
395                 dst_release(dst);
396                 return NULL;
397         }
398
399         return dst;
400 }
401 EXPORT_SYMBOL(sk_dst_check);
402
403 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
404 {
405         int ret = -ENOPROTOOPT;
406 #ifdef CONFIG_NETDEVICES
407         struct net *net = sock_net(sk);
408         char devname[IFNAMSIZ];
409         int index;
410
411         /* Sorry... */
412         ret = -EPERM;
413         if (!capable(CAP_NET_RAW))
414                 goto out;
415
416         ret = -EINVAL;
417         if (optlen < 0)
418                 goto out;
419
420         /* Bind this socket to a particular device like "eth0",
421          * as specified in the passed interface name. If the
422          * name is "" or the option length is zero the socket
423          * is not bound.
424          */
425         if (optlen > IFNAMSIZ - 1)
426                 optlen = IFNAMSIZ - 1;
427         memset(devname, 0, sizeof(devname));
428
429         ret = -EFAULT;
430         if (copy_from_user(devname, optval, optlen))
431                 goto out;
432
433         index = 0;
434         if (devname[0] != '\0') {
435                 struct net_device *dev;
436
437                 rcu_read_lock();
438                 dev = dev_get_by_name_rcu(net, devname);
439                 if (dev)
440                         index = dev->ifindex;
441                 rcu_read_unlock();
442                 ret = -ENODEV;
443                 if (!dev)
444                         goto out;
445         }
446
447         lock_sock(sk);
448         sk->sk_bound_dev_if = index;
449         sk_dst_reset(sk);
450         release_sock(sk);
451
452         ret = 0;
453
454 out:
455 #endif
456
457         return ret;
458 }
459
460 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
461 {
462         if (valbool)
463                 sock_set_flag(sk, bit);
464         else
465                 sock_reset_flag(sk, bit);
466 }
467
468 /*
469  *      This is meant for all protocols to use and covers goings on
470  *      at the socket level. Everything here is generic.
471  */
472
473 int sock_setsockopt(struct socket *sock, int level, int optname,
474                     char __user *optval, unsigned int optlen)
475 {
476         struct sock *sk = sock->sk;
477         int val;
478         int valbool;
479         struct linger ling;
480         int ret = 0;
481
482         /*
483          *      Options without arguments
484          */
485
486         if (optname == SO_BINDTODEVICE)
487                 return sock_bindtodevice(sk, optval, optlen);
488
489         if (optlen < sizeof(int))
490                 return -EINVAL;
491
492         if (get_user(val, (int __user *)optval))
493                 return -EFAULT;
494
495         valbool = val ? 1 : 0;
496
497         lock_sock(sk);
498
499         switch (optname) {
500         case SO_DEBUG:
501                 if (val && !capable(CAP_NET_ADMIN))
502                         ret = -EACCES;
503                 else
504                         sock_valbool_flag(sk, SOCK_DBG, valbool);
505                 break;
506         case SO_REUSEADDR:
507                 sk->sk_reuse = valbool;
508                 break;
509         case SO_TYPE:
510         case SO_PROTOCOL:
511         case SO_DOMAIN:
512         case SO_ERROR:
513                 ret = -ENOPROTOOPT;
514                 break;
515         case SO_DONTROUTE:
516                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
517                 break;
518         case SO_BROADCAST:
519                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
520                 break;
521         case SO_SNDBUF:
522                 /* Don't error on this BSD doesn't and if you think
523                    about it this is right. Otherwise apps have to
524                    play 'guess the biggest size' games. RCVBUF/SNDBUF
525                    are treated in BSD as hints */
526
527                 if (val > sysctl_wmem_max)
528                         val = sysctl_wmem_max;
529 set_sndbuf:
530                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
531                 if ((val * 2) < SOCK_MIN_SNDBUF)
532                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
533                 else
534                         sk->sk_sndbuf = val * 2;
535
536                 /*
537                  *      Wake up sending tasks if we
538                  *      upped the value.
539                  */
540                 sk->sk_write_space(sk);
541                 break;
542
543         case SO_SNDBUFFORCE:
544                 if (!capable(CAP_NET_ADMIN)) {
545                         ret = -EPERM;
546                         break;
547                 }
548                 goto set_sndbuf;
549
550         case SO_RCVBUF:
551                 /* Don't error on this BSD doesn't and if you think
552                    about it this is right. Otherwise apps have to
553                    play 'guess the biggest size' games. RCVBUF/SNDBUF
554                    are treated in BSD as hints */
555
556                 if (val > sysctl_rmem_max)
557                         val = sysctl_rmem_max;
558 set_rcvbuf:
559                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
560                 /*
561                  * We double it on the way in to account for
562                  * "struct sk_buff" etc. overhead.   Applications
563                  * assume that the SO_RCVBUF setting they make will
564                  * allow that much actual data to be received on that
565                  * socket.
566                  *
567                  * Applications are unaware that "struct sk_buff" and
568                  * other overheads allocate from the receive buffer
569                  * during socket buffer allocation.
570                  *
571                  * And after considering the possible alternatives,
572                  * returning the value we actually used in getsockopt
573                  * is the most desirable behavior.
574                  */
575                 if ((val * 2) < SOCK_MIN_RCVBUF)
576                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
577                 else
578                         sk->sk_rcvbuf = val * 2;
579                 break;
580
581         case SO_RCVBUFFORCE:
582                 if (!capable(CAP_NET_ADMIN)) {
583                         ret = -EPERM;
584                         break;
585                 }
586                 goto set_rcvbuf;
587
588         case SO_KEEPALIVE:
589 #ifdef CONFIG_INET
590                 if (sk->sk_protocol == IPPROTO_TCP)
591                         tcp_set_keepalive(sk, valbool);
592 #endif
593                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
594                 break;
595
596         case SO_OOBINLINE:
597                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
598                 break;
599
600         case SO_NO_CHECK:
601                 sk->sk_no_check = valbool;
602                 break;
603
604         case SO_PRIORITY:
605                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
606                         sk->sk_priority = val;
607                 else
608                         ret = -EPERM;
609                 break;
610
611         case SO_LINGER:
612                 if (optlen < sizeof(ling)) {
613                         ret = -EINVAL;  /* 1003.1g */
614                         break;
615                 }
616                 if (copy_from_user(&ling, optval, sizeof(ling))) {
617                         ret = -EFAULT;
618                         break;
619                 }
620                 if (!ling.l_onoff)
621                         sock_reset_flag(sk, SOCK_LINGER);
622                 else {
623 #if (BITS_PER_LONG == 32)
624                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
625                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
626                         else
627 #endif
628                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
629                         sock_set_flag(sk, SOCK_LINGER);
630                 }
631                 break;
632
633         case SO_BSDCOMPAT:
634                 sock_warn_obsolete_bsdism("setsockopt");
635                 break;
636
637         case SO_PASSCRED:
638                 if (valbool)
639                         set_bit(SOCK_PASSCRED, &sock->flags);
640                 else
641                         clear_bit(SOCK_PASSCRED, &sock->flags);
642                 break;
643
644         case SO_TIMESTAMP:
645         case SO_TIMESTAMPNS:
646                 if (valbool)  {
647                         if (optname == SO_TIMESTAMP)
648                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
649                         else
650                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
651                         sock_set_flag(sk, SOCK_RCVTSTAMP);
652                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
653                 } else {
654                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
655                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
656                 }
657                 break;
658
659         case SO_TIMESTAMPING:
660                 if (val & ~SOF_TIMESTAMPING_MASK) {
661                         ret = -EINVAL;
662                         break;
663                 }
664                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
665                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
666                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
667                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
668                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
669                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
670                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
671                         sock_enable_timestamp(sk,
672                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
673                 else
674                         sock_disable_timestamp(sk,
675                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
676                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
677                                   val & SOF_TIMESTAMPING_SOFTWARE);
678                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
679                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
680                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
681                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
682                 break;
683
684         case SO_RCVLOWAT:
685                 if (val < 0)
686                         val = INT_MAX;
687                 sk->sk_rcvlowat = val ? : 1;
688                 break;
689
690         case SO_RCVTIMEO:
691                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
692                 break;
693
694         case SO_SNDTIMEO:
695                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
696                 break;
697
698         case SO_ATTACH_FILTER:
699                 ret = -EINVAL;
700                 if (optlen == sizeof(struct sock_fprog)) {
701                         struct sock_fprog fprog;
702
703                         ret = -EFAULT;
704                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
705                                 break;
706
707                         ret = sk_attach_filter(&fprog, sk);
708                 }
709                 break;
710
711         case SO_DETACH_FILTER:
712                 ret = sk_detach_filter(sk);
713                 break;
714
715         case SO_PASSSEC:
716                 if (valbool)
717                         set_bit(SOCK_PASSSEC, &sock->flags);
718                 else
719                         clear_bit(SOCK_PASSSEC, &sock->flags);
720                 break;
721         case SO_MARK:
722                 if (!capable(CAP_NET_ADMIN))
723                         ret = -EPERM;
724                 else
725                         sk->sk_mark = val;
726                 break;
727
728                 /* We implement the SO_SNDLOWAT etc to
729                    not be settable (1003.1g 5.3) */
730         case SO_RXQ_OVFL:
731                 if (valbool)
732                         sock_set_flag(sk, SOCK_RXQ_OVFL);
733                 else
734                         sock_reset_flag(sk, SOCK_RXQ_OVFL);
735                 break;
736         default:
737                 ret = -ENOPROTOOPT;
738                 break;
739         }
740         release_sock(sk);
741         return ret;
742 }
743 EXPORT_SYMBOL(sock_setsockopt);
744
745
746 int sock_getsockopt(struct socket *sock, int level, int optname,
747                     char __user *optval, int __user *optlen)
748 {
749         struct sock *sk = sock->sk;
750
751         union {
752                 int val;
753                 struct linger ling;
754                 struct timeval tm;
755         } v;
756
757         int lv = sizeof(int);
758         int len;
759
760         if (get_user(len, optlen))
761                 return -EFAULT;
762         if (len < 0)
763                 return -EINVAL;
764
765         memset(&v, 0, sizeof(v));
766
767         switch (optname) {
768         case SO_DEBUG:
769                 v.val = sock_flag(sk, SOCK_DBG);
770                 break;
771
772         case SO_DONTROUTE:
773                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
774                 break;
775
776         case SO_BROADCAST:
777                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
778                 break;
779
780         case SO_SNDBUF:
781                 v.val = sk->sk_sndbuf;
782                 break;
783
784         case SO_RCVBUF:
785                 v.val = sk->sk_rcvbuf;
786                 break;
787
788         case SO_REUSEADDR:
789                 v.val = sk->sk_reuse;
790                 break;
791
792         case SO_KEEPALIVE:
793                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
794                 break;
795
796         case SO_TYPE:
797                 v.val = sk->sk_type;
798                 break;
799
800         case SO_PROTOCOL:
801                 v.val = sk->sk_protocol;
802                 break;
803
804         case SO_DOMAIN:
805                 v.val = sk->sk_family;
806                 break;
807
808         case SO_ERROR:
809                 v.val = -sock_error(sk);
810                 if (v.val == 0)
811                         v.val = xchg(&sk->sk_err_soft, 0);
812                 break;
813
814         case SO_OOBINLINE:
815                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
816                 break;
817
818         case SO_NO_CHECK:
819                 v.val = sk->sk_no_check;
820                 break;
821
822         case SO_PRIORITY:
823                 v.val = sk->sk_priority;
824                 break;
825
826         case SO_LINGER:
827                 lv              = sizeof(v.ling);
828                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
829                 v.ling.l_linger = sk->sk_lingertime / HZ;
830                 break;
831
832         case SO_BSDCOMPAT:
833                 sock_warn_obsolete_bsdism("getsockopt");
834                 break;
835
836         case SO_TIMESTAMP:
837                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
838                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
839                 break;
840
841         case SO_TIMESTAMPNS:
842                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
843                 break;
844
845         case SO_TIMESTAMPING:
846                 v.val = 0;
847                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
848                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
849                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
850                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
851                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
852                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
853                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
854                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
855                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
856                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
857                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
858                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
859                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
860                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
861                 break;
862
863         case SO_RCVTIMEO:
864                 lv = sizeof(struct timeval);
865                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
866                         v.tm.tv_sec = 0;
867                         v.tm.tv_usec = 0;
868                 } else {
869                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
870                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
871                 }
872                 break;
873
874         case SO_SNDTIMEO:
875                 lv = sizeof(struct timeval);
876                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
877                         v.tm.tv_sec = 0;
878                         v.tm.tv_usec = 0;
879                 } else {
880                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
881                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
882                 }
883                 break;
884
885         case SO_RCVLOWAT:
886                 v.val = sk->sk_rcvlowat;
887                 break;
888
889         case SO_SNDLOWAT:
890                 v.val = 1;
891                 break;
892
893         case SO_PASSCRED:
894                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
895                 break;
896
897         case SO_PEERCRED:
898                 if (len > sizeof(sk->sk_peercred))
899                         len = sizeof(sk->sk_peercred);
900                 if (copy_to_user(optval, &sk->sk_peercred, len))
901                         return -EFAULT;
902                 goto lenout;
903
904         case SO_PEERNAME:
905         {
906                 char address[128];
907
908                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
909                         return -ENOTCONN;
910                 if (lv < len)
911                         return -EINVAL;
912                 if (copy_to_user(optval, address, len))
913                         return -EFAULT;
914                 goto lenout;
915         }
916
917         /* Dubious BSD thing... Probably nobody even uses it, but
918          * the UNIX standard wants it for whatever reason... -DaveM
919          */
920         case SO_ACCEPTCONN:
921                 v.val = sk->sk_state == TCP_LISTEN;
922                 break;
923
924         case SO_PASSSEC:
925                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
926                 break;
927
928         case SO_PEERSEC:
929                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
930
931         case SO_MARK:
932                 v.val = sk->sk_mark;
933                 break;
934
935         case SO_RXQ_OVFL:
936                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
937                 break;
938
939         default:
940                 return -ENOPROTOOPT;
941         }
942
943         if (len > lv)
944                 len = lv;
945         if (copy_to_user(optval, &v, len))
946                 return -EFAULT;
947 lenout:
948         if (put_user(len, optlen))
949                 return -EFAULT;
950         return 0;
951 }
952
953 /*
954  * Initialize an sk_lock.
955  *
956  * (We also register the sk_lock with the lock validator.)
957  */
958 static inline void sock_lock_init(struct sock *sk)
959 {
960         sock_lock_init_class_and_name(sk,
961                         af_family_slock_key_strings[sk->sk_family],
962                         af_family_slock_keys + sk->sk_family,
963                         af_family_key_strings[sk->sk_family],
964                         af_family_keys + sk->sk_family);
965 }
966
967 /*
968  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
969  * even temporarly, because of RCU lookups. sk_node should also be left as is.
970  */
971 static void sock_copy(struct sock *nsk, const struct sock *osk)
972 {
973 #ifdef CONFIG_SECURITY_NETWORK
974         void *sptr = nsk->sk_security;
975 #endif
976         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
977                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
978                      sizeof(osk->sk_tx_queue_mapping));
979         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
980                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
981 #ifdef CONFIG_SECURITY_NETWORK
982         nsk->sk_security = sptr;
983         security_sk_clone(osk, nsk);
984 #endif
985 }
986
987 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
988                 int family)
989 {
990         struct sock *sk;
991         struct kmem_cache *slab;
992
993         slab = prot->slab;
994         if (slab != NULL) {
995                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
996                 if (!sk)
997                         return sk;
998                 if (priority & __GFP_ZERO) {
999                         /*
1000                          * caches using SLAB_DESTROY_BY_RCU should let
1001                          * sk_node.next un-modified. Special care is taken
1002                          * when initializing object to zero.
1003                          */
1004                         if (offsetof(struct sock, sk_node.next) != 0)
1005                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
1006                         memset(&sk->sk_node.pprev, 0,
1007                                prot->obj_size - offsetof(struct sock,
1008                                                          sk_node.pprev));
1009                 }
1010         }
1011         else
1012                 sk = kmalloc(prot->obj_size, priority);
1013
1014         if (sk != NULL) {
1015                 kmemcheck_annotate_bitfield(sk, flags);
1016
1017                 if (security_sk_alloc(sk, family, priority))
1018                         goto out_free;
1019
1020                 if (!try_module_get(prot->owner))
1021                         goto out_free_sec;
1022                 sk_tx_queue_clear(sk);
1023         }
1024
1025         return sk;
1026
1027 out_free_sec:
1028         security_sk_free(sk);
1029 out_free:
1030         if (slab != NULL)
1031                 kmem_cache_free(slab, sk);
1032         else
1033                 kfree(sk);
1034         return NULL;
1035 }
1036
1037 static void sk_prot_free(struct proto *prot, struct sock *sk)
1038 {
1039         struct kmem_cache *slab;
1040         struct module *owner;
1041
1042         owner = prot->owner;
1043         slab = prot->slab;
1044
1045         security_sk_free(sk);
1046         if (slab != NULL)
1047                 kmem_cache_free(slab, sk);
1048         else
1049                 kfree(sk);
1050         module_put(owner);
1051 }
1052
1053 /**
1054  *      sk_alloc - All socket objects are allocated here
1055  *      @net: the applicable net namespace
1056  *      @family: protocol family
1057  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1058  *      @prot: struct proto associated with this new sock instance
1059  */
1060 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1061                       struct proto *prot)
1062 {
1063         struct sock *sk;
1064
1065         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1066         if (sk) {
1067                 sk->sk_family = family;
1068                 /*
1069                  * See comment in struct sock definition to understand
1070                  * why we need sk_prot_creator -acme
1071                  */
1072                 sk->sk_prot = sk->sk_prot_creator = prot;
1073                 sock_lock_init(sk);
1074                 sock_net_set(sk, get_net(net));
1075                 atomic_set(&sk->sk_wmem_alloc, 1);
1076         }
1077
1078         return sk;
1079 }
1080 EXPORT_SYMBOL(sk_alloc);
1081
1082 static void __sk_free(struct sock *sk)
1083 {
1084         struct sk_filter *filter;
1085
1086         if (sk->sk_destruct)
1087                 sk->sk_destruct(sk);
1088
1089         filter = rcu_dereference_check(sk->sk_filter,
1090                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1091         if (filter) {
1092                 sk_filter_uncharge(sk, filter);
1093                 rcu_assign_pointer(sk->sk_filter, NULL);
1094         }
1095
1096         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1097         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1098
1099         if (atomic_read(&sk->sk_omem_alloc))
1100                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1101                        __func__, atomic_read(&sk->sk_omem_alloc));
1102
1103         put_net(sock_net(sk));
1104         sk_prot_free(sk->sk_prot_creator, sk);
1105 }
1106
1107 void sk_free(struct sock *sk)
1108 {
1109         /*
1110          * We substract one from sk_wmem_alloc and can know if
1111          * some packets are still in some tx queue.
1112          * If not null, sock_wfree() will call __sk_free(sk) later
1113          */
1114         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1115                 __sk_free(sk);
1116 }
1117 EXPORT_SYMBOL(sk_free);
1118
1119 /*
1120  * Last sock_put should drop referrence to sk->sk_net. It has already
1121  * been dropped in sk_change_net. Taking referrence to stopping namespace
1122  * is not an option.
1123  * Take referrence to a socket to remove it from hash _alive_ and after that
1124  * destroy it in the context of init_net.
1125  */
1126 void sk_release_kernel(struct sock *sk)
1127 {
1128         if (sk == NULL || sk->sk_socket == NULL)
1129                 return;
1130
1131         sock_hold(sk);
1132         sock_release(sk->sk_socket);
1133         release_net(sock_net(sk));
1134         sock_net_set(sk, get_net(&init_net));
1135         sock_put(sk);
1136 }
1137 EXPORT_SYMBOL(sk_release_kernel);
1138
1139 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1140 {
1141         struct sock *newsk;
1142
1143         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1144         if (newsk != NULL) {
1145                 struct sk_filter *filter;
1146
1147                 sock_copy(newsk, sk);
1148
1149                 /* SANITY */
1150                 get_net(sock_net(newsk));
1151                 sk_node_init(&newsk->sk_node);
1152                 sock_lock_init(newsk);
1153                 bh_lock_sock(newsk);
1154                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1155                 newsk->sk_backlog.len = 0;
1156
1157                 atomic_set(&newsk->sk_rmem_alloc, 0);
1158                 /*
1159                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1160                  */
1161                 atomic_set(&newsk->sk_wmem_alloc, 1);
1162                 atomic_set(&newsk->sk_omem_alloc, 0);
1163                 skb_queue_head_init(&newsk->sk_receive_queue);
1164                 skb_queue_head_init(&newsk->sk_write_queue);
1165 #ifdef CONFIG_NET_DMA
1166                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1167 #endif
1168
1169                 spin_lock_init(&newsk->sk_dst_lock);
1170                 rwlock_init(&newsk->sk_callback_lock);
1171                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1172                                 af_callback_keys + newsk->sk_family,
1173                                 af_family_clock_key_strings[newsk->sk_family]);
1174
1175                 newsk->sk_dst_cache     = NULL;
1176                 newsk->sk_wmem_queued   = 0;
1177                 newsk->sk_forward_alloc = 0;
1178                 newsk->sk_send_head     = NULL;
1179                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1180
1181                 sock_reset_flag(newsk, SOCK_DONE);
1182                 skb_queue_head_init(&newsk->sk_error_queue);
1183
1184                 filter = newsk->sk_filter;
1185                 if (filter != NULL)
1186                         sk_filter_charge(newsk, filter);
1187
1188                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1189                         /* It is still raw copy of parent, so invalidate
1190                          * destructor and make plain sk_free() */
1191                         newsk->sk_destruct = NULL;
1192                         sk_free(newsk);
1193                         newsk = NULL;
1194                         goto out;
1195                 }
1196
1197                 newsk->sk_err      = 0;
1198                 newsk->sk_priority = 0;
1199                 /*
1200                  * Before updating sk_refcnt, we must commit prior changes to memory
1201                  * (Documentation/RCU/rculist_nulls.txt for details)
1202                  */
1203                 smp_wmb();
1204                 atomic_set(&newsk->sk_refcnt, 2);
1205
1206                 /*
1207                  * Increment the counter in the same struct proto as the master
1208                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1209                  * is the same as sk->sk_prot->socks, as this field was copied
1210                  * with memcpy).
1211                  *
1212                  * This _changes_ the previous behaviour, where
1213                  * tcp_create_openreq_child always was incrementing the
1214                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1215                  * to be taken into account in all callers. -acme
1216                  */
1217                 sk_refcnt_debug_inc(newsk);
1218                 sk_set_socket(newsk, NULL);
1219                 newsk->sk_wq = NULL;
1220
1221                 if (newsk->sk_prot->sockets_allocated)
1222                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1223
1224                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1225                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1226                         net_enable_timestamp();
1227         }
1228 out:
1229         return newsk;
1230 }
1231 EXPORT_SYMBOL_GPL(sk_clone);
1232
1233 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1234 {
1235         __sk_dst_set(sk, dst);
1236         sk->sk_route_caps = dst->dev->features;
1237         if (sk->sk_route_caps & NETIF_F_GSO)
1238                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1239         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1240         if (sk_can_gso(sk)) {
1241                 if (dst->header_len) {
1242                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1243                 } else {
1244                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1245                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1246                 }
1247         }
1248 }
1249 EXPORT_SYMBOL_GPL(sk_setup_caps);
1250
1251 void __init sk_init(void)
1252 {
1253         if (totalram_pages <= 4096) {
1254                 sysctl_wmem_max = 32767;
1255                 sysctl_rmem_max = 32767;
1256                 sysctl_wmem_default = 32767;
1257                 sysctl_rmem_default = 32767;
1258         } else if (totalram_pages >= 131072) {
1259                 sysctl_wmem_max = 131071;
1260                 sysctl_rmem_max = 131071;
1261         }
1262 }
1263
1264 /*
1265  *      Simple resource managers for sockets.
1266  */
1267
1268
1269 /*
1270  * Write buffer destructor automatically called from kfree_skb.
1271  */
1272 void sock_wfree(struct sk_buff *skb)
1273 {
1274         struct sock *sk = skb->sk;
1275         unsigned int len = skb->truesize;
1276
1277         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1278                 /*
1279                  * Keep a reference on sk_wmem_alloc, this will be released
1280                  * after sk_write_space() call
1281                  */
1282                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1283                 sk->sk_write_space(sk);
1284                 len = 1;
1285         }
1286         /*
1287          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1288          * could not do because of in-flight packets
1289          */
1290         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1291                 __sk_free(sk);
1292 }
1293 EXPORT_SYMBOL(sock_wfree);
1294
1295 /*
1296  * Read buffer destructor automatically called from kfree_skb.
1297  */
1298 void sock_rfree(struct sk_buff *skb)
1299 {
1300         struct sock *sk = skb->sk;
1301
1302         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1303         sk_mem_uncharge(skb->sk, skb->truesize);
1304 }
1305 EXPORT_SYMBOL(sock_rfree);
1306
1307
1308 int sock_i_uid(struct sock *sk)
1309 {
1310         int uid;
1311
1312         read_lock(&sk->sk_callback_lock);
1313         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1314         read_unlock(&sk->sk_callback_lock);
1315         return uid;
1316 }
1317 EXPORT_SYMBOL(sock_i_uid);
1318
1319 unsigned long sock_i_ino(struct sock *sk)
1320 {
1321         unsigned long ino;
1322
1323         read_lock(&sk->sk_callback_lock);
1324         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1325         read_unlock(&sk->sk_callback_lock);
1326         return ino;
1327 }
1328 EXPORT_SYMBOL(sock_i_ino);
1329
1330 /*
1331  * Allocate a skb from the socket's send buffer.
1332  */
1333 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1334                              gfp_t priority)
1335 {
1336         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1337                 struct sk_buff *skb = alloc_skb(size, priority);
1338                 if (skb) {
1339                         skb_set_owner_w(skb, sk);
1340                         return skb;
1341                 }
1342         }
1343         return NULL;
1344 }
1345 EXPORT_SYMBOL(sock_wmalloc);
1346
1347 /*
1348  * Allocate a skb from the socket's receive buffer.
1349  */
1350 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1351                              gfp_t priority)
1352 {
1353         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1354                 struct sk_buff *skb = alloc_skb(size, priority);
1355                 if (skb) {
1356                         skb_set_owner_r(skb, sk);
1357                         return skb;
1358                 }
1359         }
1360         return NULL;
1361 }
1362
1363 /*
1364  * Allocate a memory block from the socket's option memory buffer.
1365  */
1366 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1367 {
1368         if ((unsigned)size <= sysctl_optmem_max &&
1369             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1370                 void *mem;
1371                 /* First do the add, to avoid the race if kmalloc
1372                  * might sleep.
1373                  */
1374                 atomic_add(size, &sk->sk_omem_alloc);
1375                 mem = kmalloc(size, priority);
1376                 if (mem)
1377                         return mem;
1378                 atomic_sub(size, &sk->sk_omem_alloc);
1379         }
1380         return NULL;
1381 }
1382 EXPORT_SYMBOL(sock_kmalloc);
1383
1384 /*
1385  * Free an option memory block.
1386  */
1387 void sock_kfree_s(struct sock *sk, void *mem, int size)
1388 {
1389         kfree(mem);
1390         atomic_sub(size, &sk->sk_omem_alloc);
1391 }
1392 EXPORT_SYMBOL(sock_kfree_s);
1393
1394 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1395    I think, these locks should be removed for datagram sockets.
1396  */
1397 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1398 {
1399         DEFINE_WAIT(wait);
1400
1401         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1402         for (;;) {
1403                 if (!timeo)
1404                         break;
1405                 if (signal_pending(current))
1406                         break;
1407                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1408                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1409                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1410                         break;
1411                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1412                         break;
1413                 if (sk->sk_err)
1414                         break;
1415                 timeo = schedule_timeout(timeo);
1416         }
1417         finish_wait(sk_sleep(sk), &wait);
1418         return timeo;
1419 }
1420
1421
1422 /*
1423  *      Generic send/receive buffer handlers
1424  */
1425
1426 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1427                                      unsigned long data_len, int noblock,
1428                                      int *errcode)
1429 {
1430         struct sk_buff *skb;
1431         gfp_t gfp_mask;
1432         long timeo;
1433         int err;
1434
1435         gfp_mask = sk->sk_allocation;
1436         if (gfp_mask & __GFP_WAIT)
1437                 gfp_mask |= __GFP_REPEAT;
1438
1439         timeo = sock_sndtimeo(sk, noblock);
1440         while (1) {
1441                 err = sock_error(sk);
1442                 if (err != 0)
1443                         goto failure;
1444
1445                 err = -EPIPE;
1446                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1447                         goto failure;
1448
1449                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1450                         skb = alloc_skb(header_len, gfp_mask);
1451                         if (skb) {
1452                                 int npages;
1453                                 int i;
1454
1455                                 /* No pages, we're done... */
1456                                 if (!data_len)
1457                                         break;
1458
1459                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1460                                 skb->truesize += data_len;
1461                                 skb_shinfo(skb)->nr_frags = npages;
1462                                 for (i = 0; i < npages; i++) {
1463                                         struct page *page;
1464                                         skb_frag_t *frag;
1465
1466                                         page = alloc_pages(sk->sk_allocation, 0);
1467                                         if (!page) {
1468                                                 err = -ENOBUFS;
1469                                                 skb_shinfo(skb)->nr_frags = i;
1470                                                 kfree_skb(skb);
1471                                                 goto failure;
1472                                         }
1473
1474                                         frag = &skb_shinfo(skb)->frags[i];
1475                                         frag->page = page;
1476                                         frag->page_offset = 0;
1477                                         frag->size = (data_len >= PAGE_SIZE ?
1478                                                       PAGE_SIZE :
1479                                                       data_len);
1480                                         data_len -= PAGE_SIZE;
1481                                 }
1482
1483                                 /* Full success... */
1484                                 break;
1485                         }
1486                         err = -ENOBUFS;
1487                         goto failure;
1488                 }
1489                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1490                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1491                 err = -EAGAIN;
1492                 if (!timeo)
1493                         goto failure;
1494                 if (signal_pending(current))
1495                         goto interrupted;
1496                 timeo = sock_wait_for_wmem(sk, timeo);
1497         }
1498
1499         skb_set_owner_w(skb, sk);
1500         return skb;
1501
1502 interrupted:
1503         err = sock_intr_errno(timeo);
1504 failure:
1505         *errcode = err;
1506         return NULL;
1507 }
1508 EXPORT_SYMBOL(sock_alloc_send_pskb);
1509
1510 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1511                                     int noblock, int *errcode)
1512 {
1513         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1514 }
1515 EXPORT_SYMBOL(sock_alloc_send_skb);
1516
1517 static void __lock_sock(struct sock *sk)
1518 {
1519         DEFINE_WAIT(wait);
1520
1521         for (;;) {
1522                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1523                                         TASK_UNINTERRUPTIBLE);
1524                 spin_unlock_bh(&sk->sk_lock.slock);
1525                 schedule();
1526                 spin_lock_bh(&sk->sk_lock.slock);
1527                 if (!sock_owned_by_user(sk))
1528                         break;
1529         }
1530         finish_wait(&sk->sk_lock.wq, &wait);
1531 }
1532
1533 static void __release_sock(struct sock *sk)
1534 {
1535         struct sk_buff *skb = sk->sk_backlog.head;
1536
1537         do {
1538                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1539                 bh_unlock_sock(sk);
1540
1541                 do {
1542                         struct sk_buff *next = skb->next;
1543
1544                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1545                         skb->next = NULL;
1546                         sk_backlog_rcv(sk, skb);
1547
1548                         /*
1549                          * We are in process context here with softirqs
1550                          * disabled, use cond_resched_softirq() to preempt.
1551                          * This is safe to do because we've taken the backlog
1552                          * queue private:
1553                          */
1554                         cond_resched_softirq();
1555
1556                         skb = next;
1557                 } while (skb != NULL);
1558
1559                 bh_lock_sock(sk);
1560         } while ((skb = sk->sk_backlog.head) != NULL);
1561
1562         /*
1563          * Doing the zeroing here guarantee we can not loop forever
1564          * while a wild producer attempts to flood us.
1565          */
1566         sk->sk_backlog.len = 0;
1567 }
1568
1569 /**
1570  * sk_wait_data - wait for data to arrive at sk_receive_queue
1571  * @sk:    sock to wait on
1572  * @timeo: for how long
1573  *
1574  * Now socket state including sk->sk_err is changed only under lock,
1575  * hence we may omit checks after joining wait queue.
1576  * We check receive queue before schedule() only as optimization;
1577  * it is very likely that release_sock() added new data.
1578  */
1579 int sk_wait_data(struct sock *sk, long *timeo)
1580 {
1581         int rc;
1582         DEFINE_WAIT(wait);
1583
1584         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1585         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1586         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1587         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1588         finish_wait(sk_sleep(sk), &wait);
1589         return rc;
1590 }
1591 EXPORT_SYMBOL(sk_wait_data);
1592
1593 /**
1594  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1595  *      @sk: socket
1596  *      @size: memory size to allocate
1597  *      @kind: allocation type
1598  *
1599  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1600  *      rmem allocation. This function assumes that protocols which have
1601  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1602  */
1603 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1604 {
1605         struct proto *prot = sk->sk_prot;
1606         int amt = sk_mem_pages(size);
1607         int allocated;
1608
1609         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1610         allocated = atomic_add_return(amt, prot->memory_allocated);
1611
1612         /* Under limit. */
1613         if (allocated <= prot->sysctl_mem[0]) {
1614                 if (prot->memory_pressure && *prot->memory_pressure)
1615                         *prot->memory_pressure = 0;
1616                 return 1;
1617         }
1618
1619         /* Under pressure. */
1620         if (allocated > prot->sysctl_mem[1])
1621                 if (prot->enter_memory_pressure)
1622                         prot->enter_memory_pressure(sk);
1623
1624         /* Over hard limit. */
1625         if (allocated > prot->sysctl_mem[2])
1626                 goto suppress_allocation;
1627
1628         /* guarantee minimum buffer size under pressure */
1629         if (kind == SK_MEM_RECV) {
1630                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1631                         return 1;
1632         } else { /* SK_MEM_SEND */
1633                 if (sk->sk_type == SOCK_STREAM) {
1634                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1635                                 return 1;
1636                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1637                            prot->sysctl_wmem[0])
1638                                 return 1;
1639         }
1640
1641         if (prot->memory_pressure) {
1642                 int alloc;
1643
1644                 if (!*prot->memory_pressure)
1645                         return 1;
1646                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1647                 if (prot->sysctl_mem[2] > alloc *
1648                     sk_mem_pages(sk->sk_wmem_queued +
1649                                  atomic_read(&sk->sk_rmem_alloc) +
1650                                  sk->sk_forward_alloc))
1651                         return 1;
1652         }
1653
1654 suppress_allocation:
1655
1656         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1657                 sk_stream_moderate_sndbuf(sk);
1658
1659                 /* Fail only if socket is _under_ its sndbuf.
1660                  * In this case we cannot block, so that we have to fail.
1661                  */
1662                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1663                         return 1;
1664         }
1665
1666         /* Alas. Undo changes. */
1667         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1668         atomic_sub(amt, prot->memory_allocated);
1669         return 0;
1670 }
1671 EXPORT_SYMBOL(__sk_mem_schedule);
1672
1673 /**
1674  *      __sk_reclaim - reclaim memory_allocated
1675  *      @sk: socket
1676  */
1677 void __sk_mem_reclaim(struct sock *sk)
1678 {
1679         struct proto *prot = sk->sk_prot;
1680
1681         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1682                    prot->memory_allocated);
1683         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1684
1685         if (prot->memory_pressure && *prot->memory_pressure &&
1686             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1687                 *prot->memory_pressure = 0;
1688 }
1689 EXPORT_SYMBOL(__sk_mem_reclaim);
1690
1691
1692 /*
1693  * Set of default routines for initialising struct proto_ops when
1694  * the protocol does not support a particular function. In certain
1695  * cases where it makes no sense for a protocol to have a "do nothing"
1696  * function, some default processing is provided.
1697  */
1698
1699 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1700 {
1701         return -EOPNOTSUPP;
1702 }
1703 EXPORT_SYMBOL(sock_no_bind);
1704
1705 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1706                     int len, int flags)
1707 {
1708         return -EOPNOTSUPP;
1709 }
1710 EXPORT_SYMBOL(sock_no_connect);
1711
1712 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1713 {
1714         return -EOPNOTSUPP;
1715 }
1716 EXPORT_SYMBOL(sock_no_socketpair);
1717
1718 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1719 {
1720         return -EOPNOTSUPP;
1721 }
1722 EXPORT_SYMBOL(sock_no_accept);
1723
1724 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1725                     int *len, int peer)
1726 {
1727         return -EOPNOTSUPP;
1728 }
1729 EXPORT_SYMBOL(sock_no_getname);
1730
1731 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1732 {
1733         return 0;
1734 }
1735 EXPORT_SYMBOL(sock_no_poll);
1736
1737 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1738 {
1739         return -EOPNOTSUPP;
1740 }
1741 EXPORT_SYMBOL(sock_no_ioctl);
1742
1743 int sock_no_listen(struct socket *sock, int backlog)
1744 {
1745         return -EOPNOTSUPP;
1746 }
1747 EXPORT_SYMBOL(sock_no_listen);
1748
1749 int sock_no_shutdown(struct socket *sock, int how)
1750 {
1751         return -EOPNOTSUPP;
1752 }
1753 EXPORT_SYMBOL(sock_no_shutdown);
1754
1755 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1756                     char __user *optval, unsigned int optlen)
1757 {
1758         return -EOPNOTSUPP;
1759 }
1760 EXPORT_SYMBOL(sock_no_setsockopt);
1761
1762 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1763                     char __user *optval, int __user *optlen)
1764 {
1765         return -EOPNOTSUPP;
1766 }
1767 EXPORT_SYMBOL(sock_no_getsockopt);
1768
1769 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1770                     size_t len)
1771 {
1772         return -EOPNOTSUPP;
1773 }
1774 EXPORT_SYMBOL(sock_no_sendmsg);
1775
1776 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1777                     size_t len, int flags)
1778 {
1779         return -EOPNOTSUPP;
1780 }
1781 EXPORT_SYMBOL(sock_no_recvmsg);
1782
1783 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1784 {
1785         /* Mirror missing mmap method error code */
1786         return -ENODEV;
1787 }
1788 EXPORT_SYMBOL(sock_no_mmap);
1789
1790 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1791 {
1792         ssize_t res;
1793         struct msghdr msg = {.msg_flags = flags};
1794         struct kvec iov;
1795         char *kaddr = kmap(page);
1796         iov.iov_base = kaddr + offset;
1797         iov.iov_len = size;
1798         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1799         kunmap(page);
1800         return res;
1801 }
1802 EXPORT_SYMBOL(sock_no_sendpage);
1803
1804 /*
1805  *      Default Socket Callbacks
1806  */
1807
1808 static void sock_def_wakeup(struct sock *sk)
1809 {
1810         struct socket_wq *wq;
1811
1812         rcu_read_lock();
1813         wq = rcu_dereference(sk->sk_wq);
1814         if (wq_has_sleeper(wq))
1815                 wake_up_interruptible_all(&wq->wait);
1816         rcu_read_unlock();
1817 }
1818
1819 static void sock_def_error_report(struct sock *sk)
1820 {
1821         struct socket_wq *wq;
1822
1823         rcu_read_lock();
1824         wq = rcu_dereference(sk->sk_wq);
1825         if (wq_has_sleeper(wq))
1826                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1827         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1828         rcu_read_unlock();
1829 }
1830
1831 static void sock_def_readable(struct sock *sk, int len)
1832 {
1833         struct socket_wq *wq;
1834
1835         rcu_read_lock();
1836         wq = rcu_dereference(sk->sk_wq);
1837         if (wq_has_sleeper(wq))
1838                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
1839                                                 POLLRDNORM | POLLRDBAND);
1840         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1841         rcu_read_unlock();
1842 }
1843
1844 static void sock_def_write_space(struct sock *sk)
1845 {
1846         struct socket_wq *wq;
1847
1848         rcu_read_lock();
1849
1850         /* Do not wake up a writer until he can make "significant"
1851          * progress.  --DaveM
1852          */
1853         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1854                 wq = rcu_dereference(sk->sk_wq);
1855                 if (wq_has_sleeper(wq))
1856                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1857                                                 POLLWRNORM | POLLWRBAND);
1858
1859                 /* Should agree with poll, otherwise some programs break */
1860                 if (sock_writeable(sk))
1861                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1862         }
1863
1864         rcu_read_unlock();
1865 }
1866
1867 static void sock_def_destruct(struct sock *sk)
1868 {
1869         kfree(sk->sk_protinfo);
1870 }
1871
1872 void sk_send_sigurg(struct sock *sk)
1873 {
1874         if (sk->sk_socket && sk->sk_socket->file)
1875                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1876                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1877 }
1878 EXPORT_SYMBOL(sk_send_sigurg);
1879
1880 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1881                     unsigned long expires)
1882 {
1883         if (!mod_timer(timer, expires))
1884                 sock_hold(sk);
1885 }
1886 EXPORT_SYMBOL(sk_reset_timer);
1887
1888 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1889 {
1890         if (timer_pending(timer) && del_timer(timer))
1891                 __sock_put(sk);
1892 }
1893 EXPORT_SYMBOL(sk_stop_timer);
1894
1895 void sock_init_data(struct socket *sock, struct sock *sk)
1896 {
1897         skb_queue_head_init(&sk->sk_receive_queue);
1898         skb_queue_head_init(&sk->sk_write_queue);
1899         skb_queue_head_init(&sk->sk_error_queue);
1900 #ifdef CONFIG_NET_DMA
1901         skb_queue_head_init(&sk->sk_async_wait_queue);
1902 #endif
1903
1904         sk->sk_send_head        =       NULL;
1905
1906         init_timer(&sk->sk_timer);
1907
1908         sk->sk_allocation       =       GFP_KERNEL;
1909         sk->sk_rcvbuf           =       sysctl_rmem_default;
1910         sk->sk_sndbuf           =       sysctl_wmem_default;
1911         sk->sk_state            =       TCP_CLOSE;
1912         sk_set_socket(sk, sock);
1913
1914         sock_set_flag(sk, SOCK_ZAPPED);
1915
1916         if (sock) {
1917                 sk->sk_type     =       sock->type;
1918                 sk->sk_wq       =       sock->wq;
1919                 sock->sk        =       sk;
1920         } else
1921                 sk->sk_wq       =       NULL;
1922
1923         spin_lock_init(&sk->sk_dst_lock);
1924         rwlock_init(&sk->sk_callback_lock);
1925         lockdep_set_class_and_name(&sk->sk_callback_lock,
1926                         af_callback_keys + sk->sk_family,
1927                         af_family_clock_key_strings[sk->sk_family]);
1928
1929         sk->sk_state_change     =       sock_def_wakeup;
1930         sk->sk_data_ready       =       sock_def_readable;
1931         sk->sk_write_space      =       sock_def_write_space;
1932         sk->sk_error_report     =       sock_def_error_report;
1933         sk->sk_destruct         =       sock_def_destruct;
1934
1935         sk->sk_sndmsg_page      =       NULL;
1936         sk->sk_sndmsg_off       =       0;
1937
1938         sk->sk_peercred.pid     =       0;
1939         sk->sk_peercred.uid     =       -1;
1940         sk->sk_peercred.gid     =       -1;
1941         sk->sk_write_pending    =       0;
1942         sk->sk_rcvlowat         =       1;
1943         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1944         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1945
1946         sk->sk_stamp = ktime_set(-1L, 0);
1947
1948         /*
1949          * Before updating sk_refcnt, we must commit prior changes to memory
1950          * (Documentation/RCU/rculist_nulls.txt for details)
1951          */
1952         smp_wmb();
1953         atomic_set(&sk->sk_refcnt, 1);
1954         atomic_set(&sk->sk_drops, 0);
1955 }
1956 EXPORT_SYMBOL(sock_init_data);
1957
1958 void lock_sock_nested(struct sock *sk, int subclass)
1959 {
1960         might_sleep();
1961         spin_lock_bh(&sk->sk_lock.slock);
1962         if (sk->sk_lock.owned)
1963                 __lock_sock(sk);
1964         sk->sk_lock.owned = 1;
1965         spin_unlock(&sk->sk_lock.slock);
1966         /*
1967          * The sk_lock has mutex_lock() semantics here:
1968          */
1969         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1970         local_bh_enable();
1971 }
1972 EXPORT_SYMBOL(lock_sock_nested);
1973
1974 void release_sock(struct sock *sk)
1975 {
1976         /*
1977          * The sk_lock has mutex_unlock() semantics:
1978          */
1979         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1980
1981         spin_lock_bh(&sk->sk_lock.slock);
1982         if (sk->sk_backlog.tail)
1983                 __release_sock(sk);
1984         sk->sk_lock.owned = 0;
1985         if (waitqueue_active(&sk->sk_lock.wq))
1986                 wake_up(&sk->sk_lock.wq);
1987         spin_unlock_bh(&sk->sk_lock.slock);
1988 }
1989 EXPORT_SYMBOL(release_sock);
1990
1991 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1992 {
1993         struct timeval tv;
1994         if (!sock_flag(sk, SOCK_TIMESTAMP))
1995                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1996         tv = ktime_to_timeval(sk->sk_stamp);
1997         if (tv.tv_sec == -1)
1998                 return -ENOENT;
1999         if (tv.tv_sec == 0) {
2000                 sk->sk_stamp = ktime_get_real();
2001                 tv = ktime_to_timeval(sk->sk_stamp);
2002         }
2003         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2004 }
2005 EXPORT_SYMBOL(sock_get_timestamp);
2006
2007 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2008 {
2009         struct timespec ts;
2010         if (!sock_flag(sk, SOCK_TIMESTAMP))
2011                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2012         ts = ktime_to_timespec(sk->sk_stamp);
2013         if (ts.tv_sec == -1)
2014                 return -ENOENT;
2015         if (ts.tv_sec == 0) {
2016                 sk->sk_stamp = ktime_get_real();
2017                 ts = ktime_to_timespec(sk->sk_stamp);
2018         }
2019         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2020 }
2021 EXPORT_SYMBOL(sock_get_timestampns);
2022
2023 void sock_enable_timestamp(struct sock *sk, int flag)
2024 {
2025         if (!sock_flag(sk, flag)) {
2026                 sock_set_flag(sk, flag);
2027                 /*
2028                  * we just set one of the two flags which require net
2029                  * time stamping, but time stamping might have been on
2030                  * already because of the other one
2031                  */
2032                 if (!sock_flag(sk,
2033                                 flag == SOCK_TIMESTAMP ?
2034                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2035                                 SOCK_TIMESTAMP))
2036                         net_enable_timestamp();
2037         }
2038 }
2039
2040 /*
2041  *      Get a socket option on an socket.
2042  *
2043  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2044  *      asynchronous errors should be reported by getsockopt. We assume
2045  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2046  */
2047 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2048                            char __user *optval, int __user *optlen)
2049 {
2050         struct sock *sk = sock->sk;
2051
2052         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2053 }
2054 EXPORT_SYMBOL(sock_common_getsockopt);
2055
2056 #ifdef CONFIG_COMPAT
2057 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2058                                   char __user *optval, int __user *optlen)
2059 {
2060         struct sock *sk = sock->sk;
2061
2062         if (sk->sk_prot->compat_getsockopt != NULL)
2063                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2064                                                       optval, optlen);
2065         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2066 }
2067 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2068 #endif
2069
2070 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2071                         struct msghdr *msg, size_t size, int flags)
2072 {
2073         struct sock *sk = sock->sk;
2074         int addr_len = 0;
2075         int err;
2076
2077         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2078                                    flags & ~MSG_DONTWAIT, &addr_len);
2079         if (err >= 0)
2080                 msg->msg_namelen = addr_len;
2081         return err;
2082 }
2083 EXPORT_SYMBOL(sock_common_recvmsg);
2084
2085 /*
2086  *      Set socket options on an inet socket.
2087  */
2088 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2089                            char __user *optval, unsigned int optlen)
2090 {
2091         struct sock *sk = sock->sk;
2092
2093         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2094 }
2095 EXPORT_SYMBOL(sock_common_setsockopt);
2096
2097 #ifdef CONFIG_COMPAT
2098 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2099                                   char __user *optval, unsigned int optlen)
2100 {
2101         struct sock *sk = sock->sk;
2102
2103         if (sk->sk_prot->compat_setsockopt != NULL)
2104                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2105                                                       optval, optlen);
2106         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2107 }
2108 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2109 #endif
2110
2111 void sk_common_release(struct sock *sk)
2112 {
2113         if (sk->sk_prot->destroy)
2114                 sk->sk_prot->destroy(sk);
2115
2116         /*
2117          * Observation: when sock_common_release is called, processes have
2118          * no access to socket. But net still has.
2119          * Step one, detach it from networking:
2120          *
2121          * A. Remove from hash tables.
2122          */
2123
2124         sk->sk_prot->unhash(sk);
2125
2126         /*
2127          * In this point socket cannot receive new packets, but it is possible
2128          * that some packets are in flight because some CPU runs receiver and
2129          * did hash table lookup before we unhashed socket. They will achieve
2130          * receive queue and will be purged by socket destructor.
2131          *
2132          * Also we still have packets pending on receive queue and probably,
2133          * our own packets waiting in device queues. sock_destroy will drain
2134          * receive queue, but transmitted packets will delay socket destruction
2135          * until the last reference will be released.
2136          */
2137
2138         sock_orphan(sk);
2139
2140         xfrm_sk_free_policy(sk);
2141
2142         sk_refcnt_debug_release(sk);
2143         sock_put(sk);
2144 }
2145 EXPORT_SYMBOL(sk_common_release);
2146
2147 static DEFINE_RWLOCK(proto_list_lock);
2148 static LIST_HEAD(proto_list);
2149
2150 #ifdef CONFIG_PROC_FS
2151 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2152 struct prot_inuse {
2153         int val[PROTO_INUSE_NR];
2154 };
2155
2156 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2157
2158 #ifdef CONFIG_NET_NS
2159 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2160 {
2161         int cpu = smp_processor_id();
2162         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2163 }
2164 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2165
2166 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2167 {
2168         int cpu, idx = prot->inuse_idx;
2169         int res = 0;
2170
2171         for_each_possible_cpu(cpu)
2172                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2173
2174         return res >= 0 ? res : 0;
2175 }
2176 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2177
2178 static int __net_init sock_inuse_init_net(struct net *net)
2179 {
2180         net->core.inuse = alloc_percpu(struct prot_inuse);
2181         return net->core.inuse ? 0 : -ENOMEM;
2182 }
2183
2184 static void __net_exit sock_inuse_exit_net(struct net *net)
2185 {
2186         free_percpu(net->core.inuse);
2187 }
2188
2189 static struct pernet_operations net_inuse_ops = {
2190         .init = sock_inuse_init_net,
2191         .exit = sock_inuse_exit_net,
2192 };
2193
2194 static __init int net_inuse_init(void)
2195 {
2196         if (register_pernet_subsys(&net_inuse_ops))
2197                 panic("Cannot initialize net inuse counters");
2198
2199         return 0;
2200 }
2201
2202 core_initcall(net_inuse_init);
2203 #else
2204 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2205
2206 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2207 {
2208         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2209 }
2210 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2211
2212 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2213 {
2214         int cpu, idx = prot->inuse_idx;
2215         int res = 0;
2216
2217         for_each_possible_cpu(cpu)
2218                 res += per_cpu(prot_inuse, cpu).val[idx];
2219
2220         return res >= 0 ? res : 0;
2221 }
2222 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2223 #endif
2224
2225 static void assign_proto_idx(struct proto *prot)
2226 {
2227         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2228
2229         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2230                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2231                 return;
2232         }
2233
2234         set_bit(prot->inuse_idx, proto_inuse_idx);
2235 }
2236
2237 static void release_proto_idx(struct proto *prot)
2238 {
2239         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2240                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2241 }
2242 #else
2243 static inline void assign_proto_idx(struct proto *prot)
2244 {
2245 }
2246
2247 static inline void release_proto_idx(struct proto *prot)
2248 {
2249 }
2250 #endif
2251
2252 int proto_register(struct proto *prot, int alloc_slab)
2253 {
2254         if (alloc_slab) {
2255                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2256                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2257                                         NULL);
2258
2259                 if (prot->slab == NULL) {
2260                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2261                                prot->name);
2262                         goto out;
2263                 }
2264
2265                 if (prot->rsk_prot != NULL) {
2266                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2267                         if (prot->rsk_prot->slab_name == NULL)
2268                                 goto out_free_sock_slab;
2269
2270                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2271                                                                  prot->rsk_prot->obj_size, 0,
2272                                                                  SLAB_HWCACHE_ALIGN, NULL);
2273
2274                         if (prot->rsk_prot->slab == NULL) {
2275                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2276                                        prot->name);
2277                                 goto out_free_request_sock_slab_name;
2278                         }
2279                 }
2280
2281                 if (prot->twsk_prot != NULL) {
2282                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2283
2284                         if (prot->twsk_prot->twsk_slab_name == NULL)
2285                                 goto out_free_request_sock_slab;
2286
2287                         prot->twsk_prot->twsk_slab =
2288                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2289                                                   prot->twsk_prot->twsk_obj_size,
2290                                                   0,
2291                                                   SLAB_HWCACHE_ALIGN |
2292                                                         prot->slab_flags,
2293                                                   NULL);
2294                         if (prot->twsk_prot->twsk_slab == NULL)
2295                                 goto out_free_timewait_sock_slab_name;
2296                 }
2297         }
2298
2299         write_lock(&proto_list_lock);
2300         list_add(&prot->node, &proto_list);
2301         assign_proto_idx(prot);
2302         write_unlock(&proto_list_lock);
2303         return 0;
2304
2305 out_free_timewait_sock_slab_name:
2306         kfree(prot->twsk_prot->twsk_slab_name);
2307 out_free_request_sock_slab:
2308         if (prot->rsk_prot && prot->rsk_prot->slab) {
2309                 kmem_cache_destroy(prot->rsk_prot->slab);
2310                 prot->rsk_prot->slab = NULL;
2311         }
2312 out_free_request_sock_slab_name:
2313         if (prot->rsk_prot)
2314                 kfree(prot->rsk_prot->slab_name);
2315 out_free_sock_slab:
2316         kmem_cache_destroy(prot->slab);
2317         prot->slab = NULL;
2318 out:
2319         return -ENOBUFS;
2320 }
2321 EXPORT_SYMBOL(proto_register);
2322
2323 void proto_unregister(struct proto *prot)
2324 {
2325         write_lock(&proto_list_lock);
2326         release_proto_idx(prot);
2327         list_del(&prot->node);
2328         write_unlock(&proto_list_lock);
2329
2330         if (prot->slab != NULL) {
2331                 kmem_cache_destroy(prot->slab);
2332                 prot->slab = NULL;
2333         }
2334
2335         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2336                 kmem_cache_destroy(prot->rsk_prot->slab);
2337                 kfree(prot->rsk_prot->slab_name);
2338                 prot->rsk_prot->slab = NULL;
2339         }
2340
2341         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2342                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2343                 kfree(prot->twsk_prot->twsk_slab_name);
2344                 prot->twsk_prot->twsk_slab = NULL;
2345         }
2346 }
2347 EXPORT_SYMBOL(proto_unregister);
2348
2349 #ifdef CONFIG_PROC_FS
2350 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2351         __acquires(proto_list_lock)
2352 {
2353         read_lock(&proto_list_lock);
2354         return seq_list_start_head(&proto_list, *pos);
2355 }
2356
2357 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2358 {
2359         return seq_list_next(v, &proto_list, pos);
2360 }
2361
2362 static void proto_seq_stop(struct seq_file *seq, void *v)
2363         __releases(proto_list_lock)
2364 {
2365         read_unlock(&proto_list_lock);
2366 }
2367
2368 static char proto_method_implemented(const void *method)
2369 {
2370         return method == NULL ? 'n' : 'y';
2371 }
2372
2373 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2374 {
2375         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2376                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2377                    proto->name,
2378                    proto->obj_size,
2379                    sock_prot_inuse_get(seq_file_net(seq), proto),
2380                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2381                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2382                    proto->max_header,
2383                    proto->slab == NULL ? "no" : "yes",
2384                    module_name(proto->owner),
2385                    proto_method_implemented(proto->close),
2386                    proto_method_implemented(proto->connect),
2387                    proto_method_implemented(proto->disconnect),
2388                    proto_method_implemented(proto->accept),
2389                    proto_method_implemented(proto->ioctl),
2390                    proto_method_implemented(proto->init),
2391                    proto_method_implemented(proto->destroy),
2392                    proto_method_implemented(proto->shutdown),
2393                    proto_method_implemented(proto->setsockopt),
2394                    proto_method_implemented(proto->getsockopt),
2395                    proto_method_implemented(proto->sendmsg),
2396                    proto_method_implemented(proto->recvmsg),
2397                    proto_method_implemented(proto->sendpage),
2398                    proto_method_implemented(proto->bind),
2399                    proto_method_implemented(proto->backlog_rcv),
2400                    proto_method_implemented(proto->hash),
2401                    proto_method_implemented(proto->unhash),
2402                    proto_method_implemented(proto->get_port),
2403                    proto_method_implemented(proto->enter_memory_pressure));
2404 }
2405
2406 static int proto_seq_show(struct seq_file *seq, void *v)
2407 {
2408         if (v == &proto_list)
2409                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2410                            "protocol",
2411                            "size",
2412                            "sockets",
2413                            "memory",
2414                            "press",
2415                            "maxhdr",
2416                            "slab",
2417                            "module",
2418                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2419         else
2420                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2421         return 0;
2422 }
2423
2424 static const struct seq_operations proto_seq_ops = {
2425         .start  = proto_seq_start,
2426         .next   = proto_seq_next,
2427         .stop   = proto_seq_stop,
2428         .show   = proto_seq_show,
2429 };
2430
2431 static int proto_seq_open(struct inode *inode, struct file *file)
2432 {
2433         return seq_open_net(inode, file, &proto_seq_ops,
2434                             sizeof(struct seq_net_private));
2435 }
2436
2437 static const struct file_operations proto_seq_fops = {
2438         .owner          = THIS_MODULE,
2439         .open           = proto_seq_open,
2440         .read           = seq_read,
2441         .llseek         = seq_lseek,
2442         .release        = seq_release_net,
2443 };
2444
2445 static __net_init int proto_init_net(struct net *net)
2446 {
2447         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2448                 return -ENOMEM;
2449
2450         return 0;
2451 }
2452
2453 static __net_exit void proto_exit_net(struct net *net)
2454 {
2455         proc_net_remove(net, "protocols");
2456 }
2457
2458
2459 static __net_initdata struct pernet_operations proto_net_ops = {
2460         .init = proto_init_net,
2461         .exit = proto_exit_net,
2462 };
2463
2464 static int __init proto_init(void)
2465 {
2466         return register_pernet_subsys(&proto_net_ops);
2467 }
2468
2469 subsys_initcall(proto_init);
2470
2471 #endif /* PROC_FS */