]> bbs.cooldavid.org Git - net-next-2.6.git/blame - net/core/sock.c
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge branch 'x86-mem-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[net-next-2.6.git] / net / core / sock.c
CommitLineData
1da177e4
LT		 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 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 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
4ec93edb 35 * code. The ACK stuff can wait and needs major
1da177e4
LT		 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
4fc268d2 92#include <linux/capability.h>
1da177e4
LT		 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>
1da177e4
LT		 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>
a1f8e7f7 112#include <linux/highmem.h>
3f551f94 113#include <linux/user_namespace.h>
1da177e4
LT		 114
115#include <asm/uaccess.h>
116#include <asm/system.h>
117
118#include <linux/netdevice.h>
119#include <net/protocol.h>
120#include <linux/skbuff.h>
457c4cbc 121#include <net/net_namespace.h>
2e6599cb 122#include <net/request_sock.h>
1da177e4 123#include <net/sock.h>
20d49473 124#include <linux/net_tstamp.h>
1da177e4
LT		 125#include <net/xfrm.h>
126#include <linux/ipsec.h>
f8451725 127#include <net/cls_cgroup.h>
1da177e4
LT		 128
129#include <linux/filter.h>
130
131#ifdef CONFIG_INET
132#include <net/tcp.h>
133#endif
134
da21f24d
IM		 135/*
136 * Each address family might have different locking rules, so we have
137 * one slock key per address family:
138 */
a5b5bb9a
IM		 139static struct lock_class_key af_family_keys[AF_MAX];
140static struct lock_class_key af_family_slock_keys[AF_MAX];
141
a5b5bb9a
IM		 142/*
143 * Make lock validator output more readable. (we pre-construct these
144 * strings build-time, so that runtime initialization of socket
145 * locks is fast):
146 */
36cbd3dc 147static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 148 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
149 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
150 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
151 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
152 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
153 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
154 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 155 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 156 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 157 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 158 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 159 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
fe33147a 160 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" ,
bce7b154 161 "sk_lock-AF_MAX"
a5b5bb9a 162};
36cbd3dc 163static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 164 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
165 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
166 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
167 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
168 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
169 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
170 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 171 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 172 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 173 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 174 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 175 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
fe33147a 176 "slock-AF_IEEE802154", "slock-AF_CAIF" ,
bce7b154 177 "slock-AF_MAX"
a5b5bb9a 178};
36cbd3dc 179static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 180 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
181 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
182 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
183 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
184 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
185 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
186 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 187 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 188 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 189 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 190 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 191 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
fe33147a 192 "clock-AF_IEEE802154", "clock-AF_CAIF" ,
bce7b154 193 "clock-AF_MAX"
443aef0e 194};
da21f24d
IM		 195
196/*
197 * sk_callback_lock locking rules are per-address-family,
198 * so split the lock classes by using a per-AF key:
199 */
200static struct lock_class_key af_callback_keys[AF_MAX];
201
1da177e4
LT		 202/* Take into consideration the size of the struct sk_buff overhead in the
203 * determination of these values, since that is non-constant across
204 * platforms. This makes socket queueing behavior and performance
205 * not depend upon such differences.
206 */
207#define _SK_MEM_PACKETS 256
208#define _SK_MEM_OVERHEAD (sizeof(struct sk_buff) + 256)
209#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
210#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
211
212/* Run time adjustable parameters. */
ab32ea5d
BH		 213__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
214__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
215__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
216__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4
LT		 217
218/* Maximal space eaten by iovec or ancilliary data plus some space */
ab32ea5d 219int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 220EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 221
f8451725
HX		 222#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
223int net_cls_subsys_id = -1;
224EXPORT_SYMBOL_GPL(net_cls_subsys_id);
225#endif
226
1da177e4
LT		 227static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
228{
229 struct timeval tv;
230
231 if (optlen < sizeof(tv))
232 return -EINVAL;
233 if (copy_from_user(&tv, optval, sizeof(tv)))
234 return -EFAULT;
ba78073e
VA		 235 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
236 return -EDOM;
1da177e4 237
ba78073e 238 if (tv.tv_sec < 0) {
6f11df83
AM		 239 static int warned __read_mostly;
240
ba78073e 241 *timeo_p = 0;
50aab54f 242 if (warned < 10 && net_ratelimit()) {
ba78073e
VA		 243 warned++;
244 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
245 "tries to set negative timeout\n",
ba25f9dc 246 current->comm, task_pid_nr(current));
50aab54f 247 }
ba78073e
VA		 248 return 0;
249 }
1da177e4
LT		 250 *timeo_p = MAX_SCHEDULE_TIMEOUT;
251 if (tv.tv_sec == 0 && tv.tv_usec == 0)
252 return 0;
253 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
254 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
255 return 0;
256}
257
258static void sock_warn_obsolete_bsdism(const char *name)
259{
260 static int warned;
261 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 262 if (strcmp(warncomm, current->comm) && warned < 5) {
263 strcpy(warncomm, current->comm);
1da177e4
LT		 264 printk(KERN_WARNING "process `%s' is using obsolete "
265 "%s SO_BSDCOMPAT\n", warncomm, name);
266 warned++;
267 }
268}
269
20d49473 270static void sock_disable_timestamp(struct sock *sk, int flag)
4ec93edb 271{
20d49473
PO		 272 if (sock_flag(sk, flag)) {
273 sock_reset_flag(sk, flag);
274 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
275 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
276 net_disable_timestamp();
277 }
1da177e4
LT		 278 }
279}
280
281
f0088a50
DV		 282int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
283{
766e9037 284 int err;
f0088a50 285 int skb_len;
3b885787
NH		 286 unsigned long flags;
287 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 288
9ee6b7f1 289 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
f0088a50
DV		 290 number of warnings when compiling with -W --ANK
291 */
292 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
293 (unsigned)sk->sk_rcvbuf) {
766e9037
ED		 294 atomic_inc(&sk->sk_drops);
295 return -ENOMEM;
f0088a50
DV		 296 }
297
fda9ef5d 298 err = sk_filter(sk, skb);
f0088a50 299 if (err)
766e9037 300 return err;
f0088a50 301
3ab224be 302 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 303 atomic_inc(&sk->sk_drops);
304 return -ENOBUFS;
3ab224be
HA		 305 }
306
f0088a50
DV		 307 skb->dev = NULL;
308 skb_set_owner_r(skb, sk);
49ad9599 309
f0088a50
DV		 310 /* Cache the SKB length before we tack it onto the receive
311 * queue. Once it is added it no longer belongs to us and
312 * may be freed by other threads of control pulling packets
313 * from the queue.
314 */
315 skb_len = skb->len;
316
7fee226a
ED		 317 /* we escape from rcu protected region, make sure we dont leak
318 * a norefcounted dst
319 */
320 skb_dst_force(skb);
321
3b885787
NH		 322 spin_lock_irqsave(&list->lock, flags);
323 skb->dropcount = atomic_read(&sk->sk_drops);
324 __skb_queue_tail(list, skb);
325 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 326
327 if (!sock_flag(sk, SOCK_DEAD))
328 sk->sk_data_ready(sk, skb_len);
766e9037 329 return 0;
f0088a50
DV		 330}
331EXPORT_SYMBOL(sock_queue_rcv_skb);
332
58a5a7b9 333int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 334{
335 int rc = NET_RX_SUCCESS;
336
fda9ef5d 337 if (sk_filter(sk, skb))
f0088a50
DV		 338 goto discard_and_relse;
339
340 skb->dev = NULL;
341
c377411f
ED		 342 if (sk_rcvqueues_full(sk, skb)) {
343 atomic_inc(&sk->sk_drops);
344 goto discard_and_relse;
345 }
58a5a7b9
ACM		 346 if (nested)
347 bh_lock_sock_nested(sk);
348 else
349 bh_lock_sock(sk);
a5b5bb9a
IM		 350 if (!sock_owned_by_user(sk)) {
351 /*
352 * trylock + unlock semantics:
353 */
354 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
355
c57943a1 356 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 357
358 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
a3a858ff 359 } else if (sk_add_backlog(sk, skb)) {
8eae939f
ZY		 360 bh_unlock_sock(sk);
361 atomic_inc(&sk->sk_drops);
362 goto discard_and_relse;
363 }
364
f0088a50
DV		 365 bh_unlock_sock(sk);
366out:
367 sock_put(sk);
368 return rc;
369discard_and_relse:
370 kfree_skb(skb);
371 goto out;
372}
373EXPORT_SYMBOL(sk_receive_skb);
374
ea94ff3b
KK		 375void sk_reset_txq(struct sock *sk)
376{
377 sk_tx_queue_clear(sk);
378}
379EXPORT_SYMBOL(sk_reset_txq);
380
f0088a50
DV		 381struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
382{
b6c6712a 383 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 384
385 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 386 sk_tx_queue_clear(sk);
b6c6712a 387 rcu_assign_pointer(sk->sk_dst_cache, NULL);
f0088a50
DV		 388 dst_release(dst);
389 return NULL;
390 }
391
392 return dst;
393}
394EXPORT_SYMBOL(__sk_dst_check);
395
396struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
397{
398 struct dst_entry *dst = sk_dst_get(sk);
399
400 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
401 sk_dst_reset(sk);
402 dst_release(dst);
403 return NULL;
404 }
405
406 return dst;
407}
408EXPORT_SYMBOL(sk_dst_check);
409
4878809f
DM		 410static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
411{
412 int ret = -ENOPROTOOPT;
413#ifdef CONFIG_NETDEVICES
3b1e0a65 414 struct net *net = sock_net(sk);
4878809f
DM		 415 char devname[IFNAMSIZ];
416 int index;
417
418 /* Sorry... */
419 ret = -EPERM;
420 if (!capable(CAP_NET_RAW))
421 goto out;
422
423 ret = -EINVAL;
424 if (optlen < 0)
425 goto out;
426
427 /* Bind this socket to a particular device like "eth0",
428 * as specified in the passed interface name. If the
429 * name is "" or the option length is zero the socket
430 * is not bound.
431 */
432 if (optlen > IFNAMSIZ - 1)
433 optlen = IFNAMSIZ - 1;
434 memset(devname, 0, sizeof(devname));
435
436 ret = -EFAULT;
437 if (copy_from_user(devname, optval, optlen))
438 goto out;
439
000ba2e4
DM		 440 index = 0;
441 if (devname[0] != '\0') {
bf8e56bf 442 struct net_device *dev;
4878809f 443
bf8e56bf
ED		 444 rcu_read_lock();
445 dev = dev_get_by_name_rcu(net, devname);
446 if (dev)
447 index = dev->ifindex;
448 rcu_read_unlock();
4878809f
DM		 449 ret = -ENODEV;
450 if (!dev)
451 goto out;
4878809f
DM		 452 }
453
454 lock_sock(sk);
455 sk->sk_bound_dev_if = index;
456 sk_dst_reset(sk);
457 release_sock(sk);
458
459 ret = 0;
460
461out:
462#endif
463
464 return ret;
465}
466
c0ef877b
PE		 467static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
468{
469 if (valbool)
470 sock_set_flag(sk, bit);
471 else
472 sock_reset_flag(sk, bit);
473}
474
1da177e4
LT		 475/*
476 * This is meant for all protocols to use and covers goings on
477 * at the socket level. Everything here is generic.
478 */
479
480int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 481 char __user *optval, unsigned int optlen)
1da177e4 482{
2a91525c 483 struct sock *sk = sock->sk;
1da177e4
LT		 484 int val;
485 int valbool;
486 struct linger ling;
487 int ret = 0;
4ec93edb 488
1da177e4
LT		 489 /*
490 * Options without arguments
491 */
492
4878809f
DM		 493 if (optname == SO_BINDTODEVICE)
494 return sock_bindtodevice(sk, optval, optlen);
495
e71a4783
SH		 496 if (optlen < sizeof(int))
497 return -EINVAL;
4ec93edb 498
1da177e4
LT		 499 if (get_user(val, (int __user *)optval))
500 return -EFAULT;
4ec93edb 501
2a91525c 502 valbool = val ? 1 : 0;
1da177e4
LT		 503
504 lock_sock(sk);
505
2a91525c 506 switch (optname) {
e71a4783 507 case SO_DEBUG:
2a91525c 508 if (val && !capable(CAP_NET_ADMIN))
e71a4783 509 ret = -EACCES;
2a91525c 510 else
c0ef877b 511 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 512 break;
513 case SO_REUSEADDR:
514 sk->sk_reuse = valbool;
515 break;
516 case SO_TYPE:
49c794e9 517 case SO_PROTOCOL:
0d6038ee 518 case SO_DOMAIN:
e71a4783
SH		 519 case SO_ERROR:
520 ret = -ENOPROTOOPT;
521 break;
522 case SO_DONTROUTE:
c0ef877b 523 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 524 break;
525 case SO_BROADCAST:
526 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
527 break;
528 case SO_SNDBUF:
529 /* Don't error on this BSD doesn't and if you think
530 about it this is right. Otherwise apps have to
531 play 'guess the biggest size' games. RCVBUF/SNDBUF
532 are treated in BSD as hints */
533
534 if (val > sysctl_wmem_max)
535 val = sysctl_wmem_max;
b0573dea 536set_sndbuf:
e71a4783
SH		 537 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
538 if ((val * 2) < SOCK_MIN_SNDBUF)
539 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
540 else
541 sk->sk_sndbuf = val * 2;
1da177e4 542
e71a4783
SH		 543 /*
544 * Wake up sending tasks if we
545 * upped the value.
546 */
547 sk->sk_write_space(sk);
548 break;
1da177e4 549
e71a4783
SH		 550 case SO_SNDBUFFORCE:
551 if (!capable(CAP_NET_ADMIN)) {
552 ret = -EPERM;
553 break;
554 }
555 goto set_sndbuf;
b0573dea 556
e71a4783
SH		 557 case SO_RCVBUF:
558 /* Don't error on this BSD doesn't and if you think
559 about it this is right. Otherwise apps have to
560 play 'guess the biggest size' games. RCVBUF/SNDBUF
561 are treated in BSD as hints */
4ec93edb 562
e71a4783
SH		 563 if (val > sysctl_rmem_max)
564 val = sysctl_rmem_max;
b0573dea 565set_rcvbuf:
e71a4783
SH		 566 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
567 /*
568 * We double it on the way in to account for
569 * "struct sk_buff" etc. overhead. Applications
570 * assume that the SO_RCVBUF setting they make will
571 * allow that much actual data to be received on that
572 * socket.
573 *
574 * Applications are unaware that "struct sk_buff" and
575 * other overheads allocate from the receive buffer
576 * during socket buffer allocation.
577 *
578 * And after considering the possible alternatives,
579 * returning the value we actually used in getsockopt
580 * is the most desirable behavior.
581 */
582 if ((val * 2) < SOCK_MIN_RCVBUF)
583 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
584 else
585 sk->sk_rcvbuf = val * 2;
586 break;
587
588 case SO_RCVBUFFORCE:
589 if (!capable(CAP_NET_ADMIN)) {
590 ret = -EPERM;
1da177e4 591 break;
e71a4783
SH		 592 }
593 goto set_rcvbuf;
1da177e4 594
e71a4783 595 case SO_KEEPALIVE:
1da177e4 596#ifdef CONFIG_INET
e71a4783
SH		 597 if (sk->sk_protocol == IPPROTO_TCP)
598 tcp_set_keepalive(sk, valbool);
1da177e4 599#endif
e71a4783
SH		 600 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
601 break;
602
603 case SO_OOBINLINE:
604 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
605 break;
606
607 case SO_NO_CHECK:
608 sk->sk_no_check = valbool;
609 break;
610
611 case SO_PRIORITY:
612 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
613 sk->sk_priority = val;
614 else
615 ret = -EPERM;
616 break;
617
618 case SO_LINGER:
619 if (optlen < sizeof(ling)) {
620 ret = -EINVAL; /* 1003.1g */
1da177e4 621 break;
e71a4783 622 }
2a91525c 623 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 624 ret = -EFAULT;
1da177e4 625 break;
e71a4783
SH		 626 }
627 if (!ling.l_onoff)
628 sock_reset_flag(sk, SOCK_LINGER);
629 else {
1da177e4 630#if (BITS_PER_LONG == 32)
e71a4783
SH		 631 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
632 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 633 else
e71a4783
SH		 634#endif
635 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
636 sock_set_flag(sk, SOCK_LINGER);
637 }
638 break;
639
640 case SO_BSDCOMPAT:
641 sock_warn_obsolete_bsdism("setsockopt");
642 break;
643
644 case SO_PASSCRED:
645 if (valbool)
646 set_bit(SOCK_PASSCRED, &sock->flags);
647 else
648 clear_bit(SOCK_PASSCRED, &sock->flags);
649 break;
650
651 case SO_TIMESTAMP:
92f37fd2 652 case SO_TIMESTAMPNS:
e71a4783 653 if (valbool) {
92f37fd2
ED		 654 if (optname == SO_TIMESTAMP)
655 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
656 else
657 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 658 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 659 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 660 } else {
e71a4783 661 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 662 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
663 }
e71a4783
SH		 664 break;
665
20d49473
PO		 666 case SO_TIMESTAMPING:
667 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 668 ret = -EINVAL;
20d49473
PO		 669 break;
670 }
671 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
672 val & SOF_TIMESTAMPING_TX_HARDWARE);
673 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
674 val & SOF_TIMESTAMPING_TX_SOFTWARE);
675 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
676 val & SOF_TIMESTAMPING_RX_HARDWARE);
677 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
678 sock_enable_timestamp(sk,
679 SOCK_TIMESTAMPING_RX_SOFTWARE);
680 else
681 sock_disable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
684 val & SOF_TIMESTAMPING_SOFTWARE);
685 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
686 val & SOF_TIMESTAMPING_SYS_HARDWARE);
687 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
688 val & SOF_TIMESTAMPING_RAW_HARDWARE);
689 break;
690
e71a4783
SH		 691 case SO_RCVLOWAT:
692 if (val < 0)
693 val = INT_MAX;
694 sk->sk_rcvlowat = val ? : 1;
695 break;
696
697 case SO_RCVTIMEO:
698 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
699 break;
700
701 case SO_SNDTIMEO:
702 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
703 break;
1da177e4 704
e71a4783
SH		 705 case SO_ATTACH_FILTER:
706 ret = -EINVAL;
707 if (optlen == sizeof(struct sock_fprog)) {
708 struct sock_fprog fprog;
1da177e4 709
e71a4783
SH		 710 ret = -EFAULT;
711 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 712 break;
e71a4783
SH		 713
714 ret = sk_attach_filter(&fprog, sk);
715 }
716 break;
717
718 case SO_DETACH_FILTER:
55b33325 719 ret = sk_detach_filter(sk);
e71a4783 720 break;
1da177e4 721
e71a4783
SH		 722 case SO_PASSSEC:
723 if (valbool)
724 set_bit(SOCK_PASSSEC, &sock->flags);
725 else
726 clear_bit(SOCK_PASSSEC, &sock->flags);
727 break;
4a19ec58
LAT		 728 case SO_MARK:
729 if (!capable(CAP_NET_ADMIN))
730 ret = -EPERM;
2a91525c 731 else
4a19ec58 732 sk->sk_mark = val;
4a19ec58 733 break;
877ce7c1 734
1da177e4
LT		 735 /* We implement the SO_SNDLOWAT etc to
736 not be settable (1003.1g 5.3) */
3b885787
NH		 737 case SO_RXQ_OVFL:
738 if (valbool)
739 sock_set_flag(sk, SOCK_RXQ_OVFL);
740 else
741 sock_reset_flag(sk, SOCK_RXQ_OVFL);
742 break;
e71a4783
SH		 743 default:
744 ret = -ENOPROTOOPT;
745 break;
4ec93edb 746 }
1da177e4
LT		 747 release_sock(sk);
748 return ret;
749}
2a91525c 750EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 751
752
3f551f94
EB		 753void cred_to_ucred(struct pid *pid, const struct cred *cred,
754 struct ucred *ucred)
755{
756 ucred->pid = pid_vnr(pid);
757 ucred->uid = ucred->gid = -1;
758 if (cred) {
759 struct user_namespace *current_ns = current_user_ns();
760
761 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
762 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
763 }
764}
3924773a 765EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 766
1da177e4
LT		 767int sock_getsockopt(struct socket *sock, int level, int optname,
768 char __user *optval, int __user *optlen)
769{
770 struct sock *sk = sock->sk;
4ec93edb 771
e71a4783 772 union {
4ec93edb
YH		 773 int val;
774 struct linger ling;
1da177e4
LT		 775 struct timeval tm;
776 } v;
4ec93edb 777
4d0392be 778 int lv = sizeof(int);
1da177e4 779 int len;
4ec93edb 780
e71a4783 781 if (get_user(len, optlen))
4ec93edb 782 return -EFAULT;
e71a4783 783 if (len < 0)
1da177e4 784 return -EINVAL;
4ec93edb 785
50fee1de 786 memset(&v, 0, sizeof(v));
df0bca04 787
2a91525c 788 switch (optname) {
e71a4783
SH		 789 case SO_DEBUG:
790 v.val = sock_flag(sk, SOCK_DBG);
791 break;
792
793 case SO_DONTROUTE:
794 v.val = sock_flag(sk, SOCK_LOCALROUTE);
795 break;
796
797 case SO_BROADCAST:
798 v.val = !!sock_flag(sk, SOCK_BROADCAST);
799 break;
800
801 case SO_SNDBUF:
802 v.val = sk->sk_sndbuf;
803 break;
804
805 case SO_RCVBUF:
806 v.val = sk->sk_rcvbuf;
807 break;
808
809 case SO_REUSEADDR:
810 v.val = sk->sk_reuse;
811 break;
812
813 case SO_KEEPALIVE:
814 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
815 break;
816
817 case SO_TYPE:
818 v.val = sk->sk_type;
819 break;
820
49c794e9
JE		 821 case SO_PROTOCOL:
822 v.val = sk->sk_protocol;
823 break;
824
0d6038ee
JE		 825 case SO_DOMAIN:
826 v.val = sk->sk_family;
827 break;
828
e71a4783
SH		 829 case SO_ERROR:
830 v.val = -sock_error(sk);
2a91525c 831 if (v.val == 0)
e71a4783
SH		 832 v.val = xchg(&sk->sk_err_soft, 0);
833 break;
834
835 case SO_OOBINLINE:
836 v.val = !!sock_flag(sk, SOCK_URGINLINE);
837 break;
838
839 case SO_NO_CHECK:
840 v.val = sk->sk_no_check;
841 break;
842
843 case SO_PRIORITY:
844 v.val = sk->sk_priority;
845 break;
846
847 case SO_LINGER:
848 lv = sizeof(v.ling);
849 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
850 v.ling.l_linger = sk->sk_lingertime / HZ;
851 break;
852
853 case SO_BSDCOMPAT:
854 sock_warn_obsolete_bsdism("getsockopt");
855 break;
856
857 case SO_TIMESTAMP:
92f37fd2
ED		 858 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
859 !sock_flag(sk, SOCK_RCVTSTAMPNS);
860 break;
861
862 case SO_TIMESTAMPNS:
863 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 864 break;
865
20d49473
PO		 866 case SO_TIMESTAMPING:
867 v.val = 0;
868 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
869 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
870 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
871 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
872 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
873 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
874 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
875 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
876 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
877 v.val |= SOF_TIMESTAMPING_SOFTWARE;
878 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
879 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
880 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
881 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
882 break;
883
e71a4783 884 case SO_RCVTIMEO:
2a91525c 885 lv = sizeof(struct timeval);
e71a4783
SH		 886 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
887 v.tm.tv_sec = 0;
888 v.tm.tv_usec = 0;
889 } else {
890 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
891 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
892 }
893 break;
894
895 case SO_SNDTIMEO:
2a91525c 896 lv = sizeof(struct timeval);
e71a4783
SH		 897 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
898 v.tm.tv_sec = 0;
899 v.tm.tv_usec = 0;
900 } else {
901 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
902 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
903 }
904 break;
1da177e4 905
e71a4783
SH		 906 case SO_RCVLOWAT:
907 v.val = sk->sk_rcvlowat;
908 break;
1da177e4 909
e71a4783 910 case SO_SNDLOWAT:
2a91525c 911 v.val = 1;
e71a4783 912 break;
1da177e4 913
e71a4783
SH		 914 case SO_PASSCRED:
915 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
916 break;
1da177e4 917
e71a4783 918 case SO_PEERCRED:
109f6e39
EB		 919 {
920 struct ucred peercred;
921 if (len > sizeof(peercred))
922 len = sizeof(peercred);
923 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
924 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 925 return -EFAULT;
926 goto lenout;
109f6e39 927 }
1da177e4 928
e71a4783
SH		 929 case SO_PEERNAME:
930 {
931 char address[128];
932
933 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
934 return -ENOTCONN;
935 if (lv < len)
936 return -EINVAL;
937 if (copy_to_user(optval, address, len))
938 return -EFAULT;
939 goto lenout;
940 }
1da177e4 941
e71a4783
SH		 942 /* Dubious BSD thing... Probably nobody even uses it, but
943 * the UNIX standard wants it for whatever reason... -DaveM
944 */
945 case SO_ACCEPTCONN:
946 v.val = sk->sk_state == TCP_LISTEN;
947 break;
1da177e4 948
e71a4783
SH		 949 case SO_PASSSEC:
950 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
951 break;
877ce7c1 952
e71a4783
SH		 953 case SO_PEERSEC:
954 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 955
4a19ec58
LAT		 956 case SO_MARK:
957 v.val = sk->sk_mark;
958 break;
959
3b885787
NH		 960 case SO_RXQ_OVFL:
961 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
962 break;
963
e71a4783
SH		 964 default:
965 return -ENOPROTOOPT;
1da177e4 966 }
e71a4783 967
1da177e4
LT		 968 if (len > lv)
969 len = lv;
970 if (copy_to_user(optval, &v, len))
971 return -EFAULT;
972lenout:
4ec93edb
YH		 973 if (put_user(len, optlen))
974 return -EFAULT;
975 return 0;
1da177e4
LT		 976}
977
a5b5bb9a
IM		 978/*
979 * Initialize an sk_lock.
980 *
981 * (We also register the sk_lock with the lock validator.)
982 */
b6f99a21 983static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 984{
ed07536e
PZ		 985 sock_lock_init_class_and_name(sk,
986 af_family_slock_key_strings[sk->sk_family],
987 af_family_slock_keys + sk->sk_family,
988 af_family_key_strings[sk->sk_family],
989 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 990}
991
4dc6dc71
ED		 992/*
993 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
994 * even temporarly, because of RCU lookups. sk_node should also be left as is.
995 */
f1a6c4da
PE		 996static void sock_copy(struct sock *nsk, const struct sock *osk)
997{
998#ifdef CONFIG_SECURITY_NETWORK
999 void *sptr = nsk->sk_security;
1000#endif
4dc6dc71 1001 BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
e022f0b4
KK		 1002 sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
1003 sizeof(osk->sk_tx_queue_mapping));
4dc6dc71
ED		 1004 memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
1005 osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
f1a6c4da
PE		 1006#ifdef CONFIG_SECURITY_NETWORK
1007 nsk->sk_security = sptr;
1008 security_sk_clone(osk, nsk);
1009#endif
1010}
1011
2e4afe7b
PE		 1012static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1013 int family)
c308c1b2
PE		 1014{
1015 struct sock *sk;
1016 struct kmem_cache *slab;
1017
1018 slab = prot->slab;
e912b114
ED		 1019 if (slab != NULL) {
1020 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1021 if (!sk)
1022 return sk;
1023 if (priority & __GFP_ZERO) {
1024 /*
1025 * caches using SLAB_DESTROY_BY_RCU should let
1026 * sk_node.next un-modified. Special care is taken
1027 * when initializing object to zero.
1028 */
1029 if (offsetof(struct sock, sk_node.next) != 0)
1030 memset(sk, 0, offsetof(struct sock, sk_node.next));
1031 memset(&sk->sk_node.pprev, 0,
1032 prot->obj_size - offsetof(struct sock,
1033 sk_node.pprev));
1034 }
1035 }
c308c1b2
PE		 1036 else
1037 sk = kmalloc(prot->obj_size, priority);
1038
2e4afe7b 1039 if (sk != NULL) {
a98b65a3
VN		 1040 kmemcheck_annotate_bitfield(sk, flags);
1041
2e4afe7b
PE		 1042 if (security_sk_alloc(sk, family, priority))
1043 goto out_free;
1044
1045 if (!try_module_get(prot->owner))
1046 goto out_free_sec;
e022f0b4 1047 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1048 }
1049
c308c1b2 1050 return sk;
2e4afe7b
PE		 1051
1052out_free_sec:
1053 security_sk_free(sk);
1054out_free:
1055 if (slab != NULL)
1056 kmem_cache_free(slab, sk);
1057 else
1058 kfree(sk);
1059 return NULL;
c308c1b2
PE		 1060}
1061
1062static void sk_prot_free(struct proto *prot, struct sock *sk)
1063{
1064 struct kmem_cache *slab;
2e4afe7b 1065 struct module *owner;
c308c1b2 1066
2e4afe7b 1067 owner = prot->owner;
c308c1b2 1068 slab = prot->slab;
2e4afe7b
PE		 1069
1070 security_sk_free(sk);
c308c1b2
PE		 1071 if (slab != NULL)
1072 kmem_cache_free(slab, sk);
1073 else
1074 kfree(sk);
2e4afe7b 1075 module_put(owner);
c308c1b2
PE		 1076}
1077
f8451725
HX		 1078#ifdef CONFIG_CGROUPS
1079void sock_update_classid(struct sock *sk)
1080{
1144182a 1081 u32 classid;
f8451725 1082
1144182a
PM		 1083 rcu_read_lock(); /* doing current task, which cannot vanish. */
1084 classid = task_cls_classid(current);
1085 rcu_read_unlock();
f8451725
HX		 1086 if (classid && classid != sk->sk_classid)
1087 sk->sk_classid = classid;
1088}
82862742 1089EXPORT_SYMBOL(sock_update_classid);
f8451725
HX		 1090#endif
1091
1da177e4
LT		 1092/**
1093 * sk_alloc - All socket objects are allocated here
c4ea43c5 1094 * @net: the applicable net namespace
4dc3b16b
PP		 1095 * @family: protocol family
1096 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1097 * @prot: struct proto associated with this new sock instance
1da177e4 1098 */
1b8d7ae4 1099struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1100 struct proto *prot)
1da177e4 1101{
c308c1b2 1102 struct sock *sk;
1da177e4 1103
154adbc8 1104 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1105 if (sk) {
154adbc8
PE		 1106 sk->sk_family = family;
1107 /*
1108 * See comment in struct sock definition to understand
1109 * why we need sk_prot_creator -acme
1110 */
1111 sk->sk_prot = sk->sk_prot_creator = prot;
1112 sock_lock_init(sk);
3b1e0a65 1113 sock_net_set(sk, get_net(net));
d66ee058 1114 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1115
1116 sock_update_classid(sk);
1da177e4 1117 }
a79af59e 1118
2e4afe7b 1119 return sk;
1da177e4 1120}
2a91525c 1121EXPORT_SYMBOL(sk_alloc);
1da177e4 1122
2b85a34e 1123static void __sk_free(struct sock *sk)
1da177e4
LT		 1124{
1125 struct sk_filter *filter;
1da177e4
LT		 1126
1127 if (sk->sk_destruct)
1128 sk->sk_destruct(sk);
1129
a898def2
PM		 1130 filter = rcu_dereference_check(sk->sk_filter,
1131 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1132 if (filter) {
309dd5fc 1133 sk_filter_uncharge(sk, filter);
fda9ef5d 1134 rcu_assign_pointer(sk->sk_filter, NULL);
1da177e4
LT		 1135 }
1136
20d49473
PO		 1137 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1138 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1da177e4
LT		 1139
1140 if (atomic_read(&sk->sk_omem_alloc))
1141 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1142 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1143
109f6e39
EB		 1144 if (sk->sk_peer_cred)
1145 put_cred(sk->sk_peer_cred);
1146 put_pid(sk->sk_peer_pid);
3b1e0a65 1147 put_net(sock_net(sk));
c308c1b2 1148 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1149}
2b85a34e
ED		 1150
1151void sk_free(struct sock *sk)
1152{
1153 /*
1154 * We substract one from sk_wmem_alloc and can know if
1155 * some packets are still in some tx queue.
1156 * If not null, sock_wfree() will call __sk_free(sk) later
1157 */
1158 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1159 __sk_free(sk);
1160}
2a91525c 1161EXPORT_SYMBOL(sk_free);
1da177e4 1162
edf02087
DL		 1163/*
1164 * Last sock_put should drop referrence to sk->sk_net. It has already
1165 * been dropped in sk_change_net. Taking referrence to stopping namespace
1166 * is not an option.
1167 * Take referrence to a socket to remove it from hash _alive_ and after that
1168 * destroy it in the context of init_net.
1169 */
1170void sk_release_kernel(struct sock *sk)
1171{
1172 if (sk == NULL || sk->sk_socket == NULL)
1173 return;
1174
1175 sock_hold(sk);
1176 sock_release(sk->sk_socket);
65a18ec5 1177 release_net(sock_net(sk));
3b1e0a65 1178 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1179 sock_put(sk);
1180}
45af1754 1181EXPORT_SYMBOL(sk_release_kernel);
edf02087 1182
dd0fc66f 1183struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
87d11ceb 1184{
8fd1d178 1185 struct sock *newsk;
87d11ceb 1186
8fd1d178 1187 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1188 if (newsk != NULL) {
1189 struct sk_filter *filter;
1190
892c141e 1191 sock_copy(newsk, sk);
87d11ceb
ACM		 1192
1193 /* SANITY */
3b1e0a65 1194 get_net(sock_net(newsk));
87d11ceb
ACM		 1195 sk_node_init(&newsk->sk_node);
1196 sock_lock_init(newsk);
1197 bh_lock_sock(newsk);
fa438ccf 1198 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1199 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1200
1201 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1202 /*
1203 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1204 */
1205 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1206 atomic_set(&newsk->sk_omem_alloc, 0);
1207 skb_queue_head_init(&newsk->sk_receive_queue);
1208 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1209#ifdef CONFIG_NET_DMA
1210 skb_queue_head_init(&newsk->sk_async_wait_queue);
1211#endif
87d11ceb 1212
b6c6712a 1213 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1214 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1215 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1216 af_callback_keys + newsk->sk_family,
1217 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1218
1219 newsk->sk_dst_cache = NULL;
1220 newsk->sk_wmem_queued = 0;
1221 newsk->sk_forward_alloc = 0;
1222 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1223 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1224
1225 sock_reset_flag(newsk, SOCK_DONE);
1226 skb_queue_head_init(&newsk->sk_error_queue);
1227
1228 filter = newsk->sk_filter;
1229 if (filter != NULL)
1230 sk_filter_charge(newsk, filter);
1231
1232 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1233 /* It is still raw copy of parent, so invalidate
1234 * destructor and make plain sk_free() */
1235 newsk->sk_destruct = NULL;
1236 sk_free(newsk);
1237 newsk = NULL;
1238 goto out;
1239 }
1240
1241 newsk->sk_err = 0;
1242 newsk->sk_priority = 0;
4dc6dc71
ED		 1243 /*
1244 * Before updating sk_refcnt, we must commit prior changes to memory
1245 * (Documentation/RCU/rculist_nulls.txt for details)
1246 */
1247 smp_wmb();
87d11ceb
ACM		 1248 atomic_set(&newsk->sk_refcnt, 2);
1249
1250 /*
1251 * Increment the counter in the same struct proto as the master
1252 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1253 * is the same as sk->sk_prot->socks, as this field was copied
1254 * with memcpy).
1255 *
1256 * This _changes_ the previous behaviour, where
1257 * tcp_create_openreq_child always was incrementing the
1258 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1259 * to be taken into account in all callers. -acme
1260 */
1261 sk_refcnt_debug_inc(newsk);
972692e0 1262 sk_set_socket(newsk, NULL);
43815482 1263 newsk->sk_wq = NULL;
87d11ceb
ACM		 1264
1265 if (newsk->sk_prot->sockets_allocated)
1748376b 1266 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
704da560
OP		 1267
1268 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1269 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1270 net_enable_timestamp();
87d11ceb
ACM		 1271 }
1272out:
1273 return newsk;
1274}
87d11ceb
ACM		 1275EXPORT_SYMBOL_GPL(sk_clone);
1276
9958089a
AK		 1277void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1278{
1279 __sk_dst_set(sk, dst);
1280 sk->sk_route_caps = dst->dev->features;
1281 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1282 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1283 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1284 if (sk_can_gso(sk)) {
82cc1a7a 1285 if (dst->header_len) {
9958089a 1286 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1287 } else {
9958089a 1288 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1289 sk->sk_gso_max_size = dst->dev->gso_max_size;
1290 }
9958089a
AK		 1291 }
1292}
1293EXPORT_SYMBOL_GPL(sk_setup_caps);
1294
1da177e4
LT		 1295void __init sk_init(void)
1296{
4481374c 1297 if (totalram_pages <= 4096) {
1da177e4
LT		 1298 sysctl_wmem_max = 32767;
1299 sysctl_rmem_max = 32767;
1300 sysctl_wmem_default = 32767;
1301 sysctl_rmem_default = 32767;
4481374c 1302 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1303 sysctl_wmem_max = 131071;
1304 sysctl_rmem_max = 131071;
1305 }
1306}
1307
1308/*
1309 * Simple resource managers for sockets.
1310 */
1311
1312
4ec93edb
YH		 1313/*
1314 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1315 */
1316void sock_wfree(struct sk_buff *skb)
1317{
1318 struct sock *sk = skb->sk;
d99927f4 1319 unsigned int len = skb->truesize;
1da177e4 1320
d99927f4
ED		 1321 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1322 /*
1323 * Keep a reference on sk_wmem_alloc, this will be released
1324 * after sk_write_space() call
1325 */
1326 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1327 sk->sk_write_space(sk);
d99927f4
ED		 1328 len = 1;
1329 }
2b85a34e 1330 /*
d99927f4
ED		 1331 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1332 * could not do because of in-flight packets
2b85a34e 1333 */
d99927f4 1334 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1335 __sk_free(sk);
1da177e4 1336}
2a91525c 1337EXPORT_SYMBOL(sock_wfree);
1da177e4 1338
4ec93edb
YH		 1339/*
1340 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1341 */
1342void sock_rfree(struct sk_buff *skb)
1343{
1344 struct sock *sk = skb->sk;
d361fd59 1345 unsigned int len = skb->truesize;
1da177e4 1346
d361fd59
ED		 1347 atomic_sub(len, &sk->sk_rmem_alloc);
1348 sk_mem_uncharge(sk, len);
1da177e4 1349}
2a91525c 1350EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1351
1352
1353int sock_i_uid(struct sock *sk)
1354{
1355 int uid;
1356
f064af1e 1357 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1358 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1359 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1360 return uid;
1361}
2a91525c 1362EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1363
1364unsigned long sock_i_ino(struct sock *sk)
1365{
1366 unsigned long ino;
1367
f064af1e 1368 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1369 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1370 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1371 return ino;
1372}
2a91525c 1373EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1374
1375/*
1376 * Allocate a skb from the socket's send buffer.
1377 */
86a76caf 1378struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1379 gfp_t priority)
1da177e4
LT		 1380{
1381 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1382 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1383 if (skb) {
1384 skb_set_owner_w(skb, sk);
1385 return skb;
1386 }
1387 }
1388 return NULL;
1389}
2a91525c 1390EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1391
1392/*
1393 * Allocate a skb from the socket's receive buffer.
4ec93edb 1394 */
86a76caf 1395struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1396 gfp_t priority)
1da177e4
LT		 1397{
1398 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1399 struct sk_buff *skb = alloc_skb(size, priority);
1400 if (skb) {
1401 skb_set_owner_r(skb, sk);
1402 return skb;
1403 }
1404 }
1405 return NULL;
1406}
1407
4ec93edb 1408/*
1da177e4 1409 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1410 */
dd0fc66f 1411void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1412{
1413 if ((unsigned)size <= sysctl_optmem_max &&
1414 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1415 void *mem;
1416 /* First do the add, to avoid the race if kmalloc
4ec93edb 1417 * might sleep.
1da177e4
LT		 1418 */
1419 atomic_add(size, &sk->sk_omem_alloc);
1420 mem = kmalloc(size, priority);
1421 if (mem)
1422 return mem;
1423 atomic_sub(size, &sk->sk_omem_alloc);
1424 }
1425 return NULL;
1426}
2a91525c 1427EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1428
1429/*
1430 * Free an option memory block.
1431 */
1432void sock_kfree_s(struct sock *sk, void *mem, int size)
1433{
1434 kfree(mem);
1435 atomic_sub(size, &sk->sk_omem_alloc);
1436}
2a91525c 1437EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1438
1439/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1440 I think, these locks should be removed for datagram sockets.
1441 */
2a91525c 1442static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1443{
1444 DEFINE_WAIT(wait);
1445
1446 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1447 for (;;) {
1448 if (!timeo)
1449 break;
1450 if (signal_pending(current))
1451 break;
1452 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1453 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1454 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1455 break;
1456 if (sk->sk_shutdown & SEND_SHUTDOWN)
1457 break;
1458 if (sk->sk_err)
1459 break;
1460 timeo = schedule_timeout(timeo);
1461 }
aa395145 1462 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1463 return timeo;
1464}
1465
1466
1467/*
1468 * Generic send/receive buffer handlers
1469 */
1470
4cc7f68d
HX		 1471struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1472 unsigned long data_len, int noblock,
1473 int *errcode)
1da177e4
LT		 1474{
1475 struct sk_buff *skb;
7d877f3b 1476 gfp_t gfp_mask;
1da177e4
LT		 1477 long timeo;
1478 int err;
1479
1480 gfp_mask = sk->sk_allocation;
1481 if (gfp_mask & __GFP_WAIT)
1482 gfp_mask |= __GFP_REPEAT;
1483
1484 timeo = sock_sndtimeo(sk, noblock);
1485 while (1) {
1486 err = sock_error(sk);
1487 if (err != 0)
1488 goto failure;
1489
1490 err = -EPIPE;
1491 if (sk->sk_shutdown & SEND_SHUTDOWN)
1492 goto failure;
1493
1494 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1495 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1496 if (skb) {
1497 int npages;
1498 int i;
1499
1500 /* No pages, we're done... */
1501 if (!data_len)
1502 break;
1503
1504 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1505 skb->truesize += data_len;
1506 skb_shinfo(skb)->nr_frags = npages;
1507 for (i = 0; i < npages; i++) {
1508 struct page *page;
1509 skb_frag_t *frag;
1510
1511 page = alloc_pages(sk->sk_allocation, 0);
1512 if (!page) {
1513 err = -ENOBUFS;
1514 skb_shinfo(skb)->nr_frags = i;
1515 kfree_skb(skb);
1516 goto failure;
1517 }
1518
1519 frag = &skb_shinfo(skb)->frags[i];
1520 frag->page = page;
1521 frag->page_offset = 0;
1522 frag->size = (data_len >= PAGE_SIZE ?
1523 PAGE_SIZE :
1524 data_len);
1525 data_len -= PAGE_SIZE;
1526 }
1527
1528 /* Full success... */
1529 break;
1530 }
1531 err = -ENOBUFS;
1532 goto failure;
1533 }
1534 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1535 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1536 err = -EAGAIN;
1537 if (!timeo)
1538 goto failure;
1539 if (signal_pending(current))
1540 goto interrupted;
1541 timeo = sock_wait_for_wmem(sk, timeo);
1542 }
1543
1544 skb_set_owner_w(skb, sk);
1545 return skb;
1546
1547interrupted:
1548 err = sock_intr_errno(timeo);
1549failure:
1550 *errcode = err;
1551 return NULL;
1552}
4cc7f68d 1553EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1554
4ec93edb 1555struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1556 int noblock, int *errcode)
1557{
1558 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1559}
2a91525c 1560EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1561
1562static void __lock_sock(struct sock *sk)
1563{
1564 DEFINE_WAIT(wait);
1565
e71a4783 1566 for (;;) {
1da177e4
LT		 1567 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1568 TASK_UNINTERRUPTIBLE);
1569 spin_unlock_bh(&sk->sk_lock.slock);
1570 schedule();
1571 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1572 if (!sock_owned_by_user(sk))
1da177e4
LT		 1573 break;
1574 }
1575 finish_wait(&sk->sk_lock.wq, &wait);
1576}
1577
1578static void __release_sock(struct sock *sk)
1579{
1580 struct sk_buff *skb = sk->sk_backlog.head;
1581
1582 do {
1583 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1584 bh_unlock_sock(sk);
1585
1586 do {
1587 struct sk_buff *next = skb->next;
1588
7fee226a 1589 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1590 skb->next = NULL;
c57943a1 1591 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1592
1593 /*
1594 * We are in process context here with softirqs
1595 * disabled, use cond_resched_softirq() to preempt.
1596 * This is safe to do because we've taken the backlog
1597 * queue private:
1598 */
1599 cond_resched_softirq();
1600
1601 skb = next;
1602 } while (skb != NULL);
1603
1604 bh_lock_sock(sk);
e71a4783 1605 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1606
1607 /*
1608 * Doing the zeroing here guarantee we can not loop forever
1609 * while a wild producer attempts to flood us.
1610 */
1611 sk->sk_backlog.len = 0;
1da177e4
LT		 1612}
1613
1614/**
1615 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1616 * @sk: sock to wait on
1617 * @timeo: for how long
1da177e4
LT		 1618 *
1619 * Now socket state including sk->sk_err is changed only under lock,
1620 * hence we may omit checks after joining wait queue.
1621 * We check receive queue before schedule() only as optimization;
1622 * it is very likely that release_sock() added new data.
1623 */
1624int sk_wait_data(struct sock *sk, long *timeo)
1625{
1626 int rc;
1627 DEFINE_WAIT(wait);
1628
aa395145 1629 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1630 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1631 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1632 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1633 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1634 return rc;
1635}
1da177e4
LT		 1636EXPORT_SYMBOL(sk_wait_data);
1637
3ab224be
HA		 1638/**
1639 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1640 * @sk: socket
1641 * @size: memory size to allocate
1642 * @kind: allocation type
1643 *
1644 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1645 * rmem allocation. This function assumes that protocols which have
1646 * memory_pressure use sk_wmem_queued as write buffer accounting.
1647 */
1648int __sk_mem_schedule(struct sock *sk, int size, int kind)
1649{
1650 struct proto *prot = sk->sk_prot;
1651 int amt = sk_mem_pages(size);
1652 int allocated;
1653
1654 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1655 allocated = atomic_add_return(amt, prot->memory_allocated);
1656
1657 /* Under limit. */
1658 if (allocated <= prot->sysctl_mem[0]) {
1659 if (prot->memory_pressure && *prot->memory_pressure)
1660 *prot->memory_pressure = 0;
1661 return 1;
1662 }
1663
1664 /* Under pressure. */
1665 if (allocated > prot->sysctl_mem[1])
1666 if (prot->enter_memory_pressure)
5c52ba17 1667 prot->enter_memory_pressure(sk);
3ab224be
HA		 1668
1669 /* Over hard limit. */
1670 if (allocated > prot->sysctl_mem[2])
1671 goto suppress_allocation;
1672
1673 /* guarantee minimum buffer size under pressure */
1674 if (kind == SK_MEM_RECV) {
1675 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1676 return 1;
1677 } else { /* SK_MEM_SEND */
1678 if (sk->sk_type == SOCK_STREAM) {
1679 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1680 return 1;
1681 } else if (atomic_read(&sk->sk_wmem_alloc) <
1682 prot->sysctl_wmem[0])
1683 return 1;
1684 }
1685
1686 if (prot->memory_pressure) {
1748376b
ED		 1687 int alloc;
1688
1689 if (!*prot->memory_pressure)
1690 return 1;
1691 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1692 if (prot->sysctl_mem[2] > alloc *
3ab224be
HA		 1693 sk_mem_pages(sk->sk_wmem_queued +
1694 atomic_read(&sk->sk_rmem_alloc) +
1695 sk->sk_forward_alloc))
1696 return 1;
1697 }
1698
1699suppress_allocation:
1700
1701 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1702 sk_stream_moderate_sndbuf(sk);
1703
1704 /* Fail only if socket is _under_ its sndbuf.
1705 * In this case we cannot block, so that we have to fail.
1706 */
1707 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1708 return 1;
1709 }
1710
1711 /* Alas. Undo changes. */
1712 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1713 atomic_sub(amt, prot->memory_allocated);
1714 return 0;
1715}
3ab224be
HA		 1716EXPORT_SYMBOL(__sk_mem_schedule);
1717
1718/**
1719 * __sk_reclaim - reclaim memory_allocated
1720 * @sk: socket
1721 */
1722void __sk_mem_reclaim(struct sock *sk)
1723{
1724 struct proto *prot = sk->sk_prot;
1725
680a5a50 1726 atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1727 prot->memory_allocated);
1728 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1729
1730 if (prot->memory_pressure && *prot->memory_pressure &&
1731 (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1732 *prot->memory_pressure = 0;
1733}
3ab224be
HA		 1734EXPORT_SYMBOL(__sk_mem_reclaim);
1735
1736
1da177e4
LT		 1737/*
1738 * Set of default routines for initialising struct proto_ops when
1739 * the protocol does not support a particular function. In certain
1740 * cases where it makes no sense for a protocol to have a "do nothing"
1741 * function, some default processing is provided.
1742 */
1743
1744int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1745{
1746 return -EOPNOTSUPP;
1747}
2a91525c 1748EXPORT_SYMBOL(sock_no_bind);
1da177e4 1749
4ec93edb 1750int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1751 int len, int flags)
1752{
1753 return -EOPNOTSUPP;
1754}
2a91525c 1755EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1756
1757int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1758{
1759 return -EOPNOTSUPP;
1760}
2a91525c 1761EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1762
1763int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1764{
1765 return -EOPNOTSUPP;
1766}
2a91525c 1767EXPORT_SYMBOL(sock_no_accept);
1da177e4 1768
4ec93edb 1769int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1770 int *len, int peer)
1771{
1772 return -EOPNOTSUPP;
1773}
2a91525c 1774EXPORT_SYMBOL(sock_no_getname);
1da177e4 1775
2a91525c 1776unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1777{
1778 return 0;
1779}
2a91525c 1780EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1781
1782int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1783{
1784 return -EOPNOTSUPP;
1785}
2a91525c 1786EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1787
1788int sock_no_listen(struct socket *sock, int backlog)
1789{
1790 return -EOPNOTSUPP;
1791}
2a91525c 1792EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1793
1794int sock_no_shutdown(struct socket *sock, int how)
1795{
1796 return -EOPNOTSUPP;
1797}
2a91525c 1798EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1799
1800int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1801 char __user *optval, unsigned int optlen)
1da177e4
LT		 1802{
1803 return -EOPNOTSUPP;
1804}
2a91525c 1805EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1806
1807int sock_no_getsockopt(struct socket *sock, int level, int optname,
1808 char __user *optval, int __user *optlen)
1809{
1810 return -EOPNOTSUPP;
1811}
2a91525c 1812EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1813
1814int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1815 size_t len)
1816{
1817 return -EOPNOTSUPP;
1818}
2a91525c 1819EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1820
1821int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1822 size_t len, int flags)
1823{
1824 return -EOPNOTSUPP;
1825}
2a91525c 1826EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1827
1828int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1829{
1830 /* Mirror missing mmap method error code */
1831 return -ENODEV;
1832}
2a91525c 1833EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1834
1835ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1836{
1837 ssize_t res;
1838 struct msghdr msg = {.msg_flags = flags};
1839 struct kvec iov;
1840 char *kaddr = kmap(page);
1841 iov.iov_base = kaddr + offset;
1842 iov.iov_len = size;
1843 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1844 kunmap(page);
1845 return res;
1846}
2a91525c 1847EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1848
1849/*
1850 * Default Socket Callbacks
1851 */
1852
1853static void sock_def_wakeup(struct sock *sk)
1854{
43815482
ED		 1855 struct socket_wq *wq;
1856
1857 rcu_read_lock();
1858 wq = rcu_dereference(sk->sk_wq);
1859 if (wq_has_sleeper(wq))
1860 wake_up_interruptible_all(&wq->wait);
1861 rcu_read_unlock();
1da177e4
LT		 1862}
1863
1864static void sock_def_error_report(struct sock *sk)
1865{
43815482
ED		 1866 struct socket_wq *wq;
1867
1868 rcu_read_lock();
1869 wq = rcu_dereference(sk->sk_wq);
1870 if (wq_has_sleeper(wq))
1871 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1872 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1873 rcu_read_unlock();
1da177e4
LT		 1874}
1875
1876static void sock_def_readable(struct sock *sk, int len)
1877{
43815482
ED		 1878 struct socket_wq *wq;
1879
1880 rcu_read_lock();
1881 wq = rcu_dereference(sk->sk_wq);
1882 if (wq_has_sleeper(wq))
1883 wake_up_interruptible_sync_poll(&wq->wait, POLLIN |
37e5540b 1884 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1885 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1886 rcu_read_unlock();
1da177e4
LT		 1887}
1888
1889static void sock_def_write_space(struct sock *sk)
1890{
43815482
ED		 1891 struct socket_wq *wq;
1892
1893 rcu_read_lock();
1da177e4
LT		 1894
1895 /* Do not wake up a writer until he can make "significant"
1896 * progress. --DaveM
1897 */
e71a4783 1898 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 1899 wq = rcu_dereference(sk->sk_wq);
1900 if (wq_has_sleeper(wq))
1901 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 1902 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 1903
1904 /* Should agree with poll, otherwise some programs break */
1905 if (sock_writeable(sk))
8d8ad9d7 1906 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1907 }
1908
43815482 1909 rcu_read_unlock();
1da177e4
LT		 1910}
1911
1912static void sock_def_destruct(struct sock *sk)
1913{
a51482bd 1914 kfree(sk->sk_protinfo);
1da177e4
LT		 1915}
1916
1917void sk_send_sigurg(struct sock *sk)
1918{
1919 if (sk->sk_socket && sk->sk_socket->file)
1920 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 1921 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 1922}
2a91525c 1923EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 1924
1925void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1926 unsigned long expires)
1927{
1928 if (!mod_timer(timer, expires))
1929 sock_hold(sk);
1930}
1da177e4
LT		 1931EXPORT_SYMBOL(sk_reset_timer);
1932
1933void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1934{
1935 if (timer_pending(timer) && del_timer(timer))
1936 __sock_put(sk);
1937}
1da177e4
LT		 1938EXPORT_SYMBOL(sk_stop_timer);
1939
1940void sock_init_data(struct socket *sock, struct sock *sk)
1941{
1942 skb_queue_head_init(&sk->sk_receive_queue);
1943 skb_queue_head_init(&sk->sk_write_queue);
1944 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 1945#ifdef CONFIG_NET_DMA
1946 skb_queue_head_init(&sk->sk_async_wait_queue);
1947#endif
1da177e4
LT		 1948
1949 sk->sk_send_head = NULL;
1950
1951 init_timer(&sk->sk_timer);
4ec93edb 1952
1da177e4
LT		 1953 sk->sk_allocation = GFP_KERNEL;
1954 sk->sk_rcvbuf = sysctl_rmem_default;
1955 sk->sk_sndbuf = sysctl_wmem_default;
1956 sk->sk_state = TCP_CLOSE;
972692e0 1957 sk_set_socket(sk, sock);
1da177e4
LT		 1958
1959 sock_set_flag(sk, SOCK_ZAPPED);
1960
e71a4783 1961 if (sock) {
1da177e4 1962 sk->sk_type = sock->type;
43815482 1963 sk->sk_wq = sock->wq;
1da177e4
LT		 1964 sock->sk = sk;
1965 } else
43815482 1966 sk->sk_wq = NULL;
1da177e4 1967
b6c6712a 1968 spin_lock_init(&sk->sk_dst_lock);
1da177e4 1969 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 1970 lockdep_set_class_and_name(&sk->sk_callback_lock,
1971 af_callback_keys + sk->sk_family,
1972 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 1973
1974 sk->sk_state_change = sock_def_wakeup;
1975 sk->sk_data_ready = sock_def_readable;
1976 sk->sk_write_space = sock_def_write_space;
1977 sk->sk_error_report = sock_def_error_report;
1978 sk->sk_destruct = sock_def_destruct;
1979
1980 sk->sk_sndmsg_page = NULL;
1981 sk->sk_sndmsg_off = 0;
1982
109f6e39
EB		 1983 sk->sk_peer_pid = NULL;
1984 sk->sk_peer_cred = NULL;
1da177e4
LT		 1985 sk->sk_write_pending = 0;
1986 sk->sk_rcvlowat = 1;
1987 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
1988 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
1989
f37f0afb 1990 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 1991
4dc6dc71
ED		 1992 /*
1993 * Before updating sk_refcnt, we must commit prior changes to memory
1994 * (Documentation/RCU/rculist_nulls.txt for details)
1995 */
1996 smp_wmb();
1da177e4 1997 atomic_set(&sk->sk_refcnt, 1);
33c732c3 1998 atomic_set(&sk->sk_drops, 0);
1da177e4 1999}
2a91525c 2000EXPORT_SYMBOL(sock_init_data);
1da177e4 2001
b5606c2d 2002void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2003{
2004 might_sleep();
a5b5bb9a 2005 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2006 if (sk->sk_lock.owned)
1da177e4 2007 __lock_sock(sk);
d2e9117c 2008 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2009 spin_unlock(&sk->sk_lock.slock);
2010 /*
2011 * The sk_lock has mutex_lock() semantics here:
2012 */
fcc70d5f 2013 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2014 local_bh_enable();
1da177e4 2015}
fcc70d5f 2016EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2017
b5606c2d 2018void release_sock(struct sock *sk)
1da177e4 2019{
a5b5bb9a
IM		 2020 /*
2021 * The sk_lock has mutex_unlock() semantics:
2022 */
2023 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2024
2025 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2026 if (sk->sk_backlog.tail)
2027 __release_sock(sk);
d2e9117c 2028 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2029 if (waitqueue_active(&sk->sk_lock.wq))
2030 wake_up(&sk->sk_lock.wq);
2031 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2032}
2033EXPORT_SYMBOL(release_sock);
2034
8a74ad60
ED		 2035/**
2036 * lock_sock_fast - fast version of lock_sock
2037 * @sk: socket
2038 *
2039 * This version should be used for very small section, where process wont block
2040 * return false if fast path is taken
2041 * sk_lock.slock locked, owned = 0, BH disabled
2042 * return true if slow path is taken
2043 * sk_lock.slock unlocked, owned = 1, BH enabled
2044 */
2045bool lock_sock_fast(struct sock *sk)
2046{
2047 might_sleep();
2048 spin_lock_bh(&sk->sk_lock.slock);
2049
2050 if (!sk->sk_lock.owned)
2051 /*
2052 * Note : We must disable BH
2053 */
2054 return false;
2055
2056 __lock_sock(sk);
2057 sk->sk_lock.owned = 1;
2058 spin_unlock(&sk->sk_lock.slock);
2059 /*
2060 * The sk_lock has mutex_lock() semantics here:
2061 */
2062 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2063 local_bh_enable();
2064 return true;
2065}
2066EXPORT_SYMBOL(lock_sock_fast);
2067
1da177e4 2068int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2069{
b7aa0bf7 2070 struct timeval tv;
1da177e4 2071 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2072 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2073 tv = ktime_to_timeval(sk->sk_stamp);
2074 if (tv.tv_sec == -1)
1da177e4 2075 return -ENOENT;
b7aa0bf7
ED		 2076 if (tv.tv_sec == 0) {
2077 sk->sk_stamp = ktime_get_real();
2078 tv = ktime_to_timeval(sk->sk_stamp);
2079 }
2080 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2081}
1da177e4
LT		 2082EXPORT_SYMBOL(sock_get_timestamp);
2083
ae40eb1e
ED		 2084int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2085{
2086 struct timespec ts;
2087 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2088 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2089 ts = ktime_to_timespec(sk->sk_stamp);
2090 if (ts.tv_sec == -1)
2091 return -ENOENT;
2092 if (ts.tv_sec == 0) {
2093 sk->sk_stamp = ktime_get_real();
2094 ts = ktime_to_timespec(sk->sk_stamp);
2095 }
2096 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2097}
2098EXPORT_SYMBOL(sock_get_timestampns);
2099
20d49473 2100void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2101{
20d49473
PO		 2102 if (!sock_flag(sk, flag)) {
2103 sock_set_flag(sk, flag);
2104 /*
2105 * we just set one of the two flags which require net
2106 * time stamping, but time stamping might have been on
2107 * already because of the other one
2108 */
2109 if (!sock_flag(sk,
2110 flag == SOCK_TIMESTAMP ?
2111 SOCK_TIMESTAMPING_RX_SOFTWARE :
2112 SOCK_TIMESTAMP))
2113 net_enable_timestamp();
1da177e4
LT		 2114 }
2115}
1da177e4
LT		 2116
2117/*
2118 * Get a socket option on an socket.
2119 *
2120 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2121 * asynchronous errors should be reported by getsockopt. We assume
2122 * this means if you specify SO_ERROR (otherwise whats the point of it).
2123 */
2124int sock_common_getsockopt(struct socket *sock, int level, int optname,
2125 char __user *optval, int __user *optlen)
2126{
2127 struct sock *sk = sock->sk;
2128
2129 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2130}
1da177e4
LT		 2131EXPORT_SYMBOL(sock_common_getsockopt);
2132
3fdadf7d 2133#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2134int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2135 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2136{
2137 struct sock *sk = sock->sk;
2138
1e51f951 2139 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2140 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2141 optval, optlen);
3fdadf7d
DM		 2142 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2143}
2144EXPORT_SYMBOL(compat_sock_common_getsockopt);
2145#endif
2146
1da177e4
LT		 2147int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2148 struct msghdr *msg, size_t size, int flags)
2149{
2150 struct sock *sk = sock->sk;
2151 int addr_len = 0;
2152 int err;
2153
2154 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2155 flags & ~MSG_DONTWAIT, &addr_len);
2156 if (err >= 0)
2157 msg->msg_namelen = addr_len;
2158 return err;
2159}
1da177e4
LT		 2160EXPORT_SYMBOL(sock_common_recvmsg);
2161
2162/*
2163 * Set socket options on an inet socket.
2164 */
2165int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2166 char __user *optval, unsigned int optlen)
1da177e4
LT		 2167{
2168 struct sock *sk = sock->sk;
2169
2170 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2171}
1da177e4
LT		 2172EXPORT_SYMBOL(sock_common_setsockopt);
2173
3fdadf7d 2174#ifdef CONFIG_COMPAT
543d9cfe 2175int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2176 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2177{
2178 struct sock *sk = sock->sk;
2179
543d9cfe
ACM		 2180 if (sk->sk_prot->compat_setsockopt != NULL)
2181 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2182 optval, optlen);
3fdadf7d
DM		 2183 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2184}
2185EXPORT_SYMBOL(compat_sock_common_setsockopt);
2186#endif
2187
1da177e4
LT		 2188void sk_common_release(struct sock *sk)
2189{
2190 if (sk->sk_prot->destroy)
2191 sk->sk_prot->destroy(sk);
2192
2193 /*
2194 * Observation: when sock_common_release is called, processes have
2195 * no access to socket. But net still has.
2196 * Step one, detach it from networking:
2197 *
2198 * A. Remove from hash tables.
2199 */
2200
2201 sk->sk_prot->unhash(sk);
2202
2203 /*
2204 * In this point socket cannot receive new packets, but it is possible
2205 * that some packets are in flight because some CPU runs receiver and
2206 * did hash table lookup before we unhashed socket. They will achieve
2207 * receive queue and will be purged by socket destructor.
2208 *
2209 * Also we still have packets pending on receive queue and probably,
2210 * our own packets waiting in device queues. sock_destroy will drain
2211 * receive queue, but transmitted packets will delay socket destruction
2212 * until the last reference will be released.
2213 */
2214
2215 sock_orphan(sk);
2216
2217 xfrm_sk_free_policy(sk);
2218
e6848976 2219 sk_refcnt_debug_release(sk);
1da177e4
LT		 2220 sock_put(sk);
2221}
1da177e4
LT		 2222EXPORT_SYMBOL(sk_common_release);
2223
2224static DEFINE_RWLOCK(proto_list_lock);
2225static LIST_HEAD(proto_list);
2226
13ff3d6f
PE		 2227#ifdef CONFIG_PROC_FS
2228#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2229struct prot_inuse {
2230 int val[PROTO_INUSE_NR];
2231};
13ff3d6f
PE		 2232
2233static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2234
2235#ifdef CONFIG_NET_NS
2236void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2237{
d6d9ca0f 2238 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2239}
2240EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2241
2242int sock_prot_inuse_get(struct net *net, struct proto *prot)
2243{
2244 int cpu, idx = prot->inuse_idx;
2245 int res = 0;
2246
2247 for_each_possible_cpu(cpu)
2248 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2249
2250 return res >= 0 ? res : 0;
2251}
2252EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2253
2c8c1e72 2254static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2255{
2256 net->core.inuse = alloc_percpu(struct prot_inuse);
2257 return net->core.inuse ? 0 : -ENOMEM;
2258}
2259
2c8c1e72 2260static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2261{
2262 free_percpu(net->core.inuse);
2263}
2264
2265static struct pernet_operations net_inuse_ops = {
2266 .init = sock_inuse_init_net,
2267 .exit = sock_inuse_exit_net,
2268};
2269
2270static __init int net_inuse_init(void)
2271{
2272 if (register_pernet_subsys(&net_inuse_ops))
2273 panic("Cannot initialize net inuse counters");
2274
2275 return 0;
2276}
2277
2278core_initcall(net_inuse_init);
2279#else
1338d466
PE		 2280static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2281
c29a0bc4 2282void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2283{
d6d9ca0f 2284 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2285}
2286EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2287
c29a0bc4 2288int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2289{
2290 int cpu, idx = prot->inuse_idx;
2291 int res = 0;
2292
2293 for_each_possible_cpu(cpu)
2294 res += per_cpu(prot_inuse, cpu).val[idx];
2295
2296 return res >= 0 ? res : 0;
2297}
2298EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2299#endif
13ff3d6f
PE		 2300
2301static void assign_proto_idx(struct proto *prot)
2302{
2303 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2304
2305 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2306 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2307 return;
2308 }
2309
2310 set_bit(prot->inuse_idx, proto_inuse_idx);
2311}
2312
2313static void release_proto_idx(struct proto *prot)
2314{
2315 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2316 clear_bit(prot->inuse_idx, proto_inuse_idx);
2317}
2318#else
2319static inline void assign_proto_idx(struct proto *prot)
2320{
2321}
2322
2323static inline void release_proto_idx(struct proto *prot)
2324{
2325}
2326#endif
2327
b733c007
PE		 2328int proto_register(struct proto *prot, int alloc_slab)
2329{
1da177e4
LT		 2330 if (alloc_slab) {
2331 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2332 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2333 NULL);
1da177e4
LT		 2334
2335 if (prot->slab == NULL) {
2336 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2337 prot->name);
60e7663d 2338 goto out;
1da177e4 2339 }
2e6599cb
ACM		 2340
2341 if (prot->rsk_prot != NULL) {
faf23422 2342 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2343 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2344 goto out_free_sock_slab;
2345
7e56b5d6 2346 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2347 prot->rsk_prot->obj_size, 0,
20c2df83 2348 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2349
2350 if (prot->rsk_prot->slab == NULL) {
2351 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2352 prot->name);
2353 goto out_free_request_sock_slab_name;
2354 }
2355 }
8feaf0c0 2356
6d6ee43e 2357 if (prot->twsk_prot != NULL) {
faf23422 2358 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2359
7e56b5d6 2360 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2361 goto out_free_request_sock_slab;
2362
6d6ee43e 2363 prot->twsk_prot->twsk_slab =
7e56b5d6 2364 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2365 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2366 0,
2367 SLAB_HWCACHE_ALIGN |
2368 prot->slab_flags,
20c2df83 2369 NULL);
6d6ee43e 2370 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2371 goto out_free_timewait_sock_slab_name;
2372 }
1da177e4
LT		 2373 }
2374
2a278051 2375 write_lock(&proto_list_lock);
1da177e4 2376 list_add(&prot->node, &proto_list);
13ff3d6f 2377 assign_proto_idx(prot);
1da177e4 2378 write_unlock(&proto_list_lock);
b733c007
PE		 2379 return 0;
2380
8feaf0c0 2381out_free_timewait_sock_slab_name:
7e56b5d6 2382 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2383out_free_request_sock_slab:
2384 if (prot->rsk_prot && prot->rsk_prot->slab) {
2385 kmem_cache_destroy(prot->rsk_prot->slab);
2386 prot->rsk_prot->slab = NULL;
2387 }
2e6599cb 2388out_free_request_sock_slab_name:
72150e9b
DC		 2389 if (prot->rsk_prot)
2390 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2391out_free_sock_slab:
2392 kmem_cache_destroy(prot->slab);
2393 prot->slab = NULL;
b733c007
PE		 2394out:
2395 return -ENOBUFS;
1da177e4 2396}
1da177e4
LT		 2397EXPORT_SYMBOL(proto_register);
2398
2399void proto_unregister(struct proto *prot)
2400{
2401 write_lock(&proto_list_lock);
13ff3d6f 2402 release_proto_idx(prot);
0a3f4358
PM		 2403 list_del(&prot->node);
2404 write_unlock(&proto_list_lock);
1da177e4
LT		 2405
2406 if (prot->slab != NULL) {
2407 kmem_cache_destroy(prot->slab);
2408 prot->slab = NULL;
2409 }
2410
2e6599cb 2411 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2412 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2413 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2414 prot->rsk_prot->slab = NULL;
2415 }
2416
6d6ee43e 2417 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2418 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2419 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2420 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2421 }
1da177e4 2422}
1da177e4
LT		 2423EXPORT_SYMBOL(proto_unregister);
2424
2425#ifdef CONFIG_PROC_FS
1da177e4 2426static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2427 __acquires(proto_list_lock)
1da177e4
LT		 2428{
2429 read_lock(&proto_list_lock);
60f0438a 2430 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2431}
2432
2433static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2434{
60f0438a 2435 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2436}
2437
2438static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2439 __releases(proto_list_lock)
1da177e4
LT		 2440{
2441 read_unlock(&proto_list_lock);
2442}
2443
2444static char proto_method_implemented(const void *method)
2445{
2446 return method == NULL ? 'n' : 'y';
2447}
2448
2449static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2450{
2451 seq_printf(seq, "%-9s %4u %6d %6d %-3s %6u %-3s %-10s "
2452 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2453 proto->name,
2454 proto->obj_size,
14e943db 2455 sock_prot_inuse_get(seq_file_net(seq), proto),
1da177e4
LT		 2456 proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2457 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2458 proto->max_header,
2459 proto->slab == NULL ? "no" : "yes",
2460 module_name(proto->owner),
2461 proto_method_implemented(proto->close),
2462 proto_method_implemented(proto->connect),
2463 proto_method_implemented(proto->disconnect),
2464 proto_method_implemented(proto->accept),
2465 proto_method_implemented(proto->ioctl),
2466 proto_method_implemented(proto->init),
2467 proto_method_implemented(proto->destroy),
2468 proto_method_implemented(proto->shutdown),
2469 proto_method_implemented(proto->setsockopt),
2470 proto_method_implemented(proto->getsockopt),
2471 proto_method_implemented(proto->sendmsg),
2472 proto_method_implemented(proto->recvmsg),
2473 proto_method_implemented(proto->sendpage),
2474 proto_method_implemented(proto->bind),
2475 proto_method_implemented(proto->backlog_rcv),
2476 proto_method_implemented(proto->hash),
2477 proto_method_implemented(proto->unhash),
2478 proto_method_implemented(proto->get_port),
2479 proto_method_implemented(proto->enter_memory_pressure));
2480}
2481
2482static int proto_seq_show(struct seq_file *seq, void *v)
2483{
60f0438a 2484 if (v == &proto_list)
1da177e4
LT		 2485 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2486 "protocol",
2487 "size",
2488 "sockets",
2489 "memory",
2490 "press",
2491 "maxhdr",
2492 "slab",
2493 "module",
2494 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2495 else
60f0438a 2496 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2497 return 0;
2498}
2499
f690808e 2500static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2501 .start = proto_seq_start,
2502 .next = proto_seq_next,
2503 .stop = proto_seq_stop,
2504 .show = proto_seq_show,
2505};
2506
2507static int proto_seq_open(struct inode *inode, struct file *file)
2508{
14e943db
ED		 2509 return seq_open_net(inode, file, &proto_seq_ops,
2510 sizeof(struct seq_net_private));
1da177e4
LT		 2511}
2512
9a32144e 2513static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2514 .owner = THIS_MODULE,
2515 .open = proto_seq_open,
2516 .read = seq_read,
2517 .llseek = seq_lseek,
14e943db
ED		 2518 .release = seq_release_net,
2519};
2520
2521static __net_init int proto_init_net(struct net *net)
2522{
2523 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2524 return -ENOMEM;
2525
2526 return 0;
2527}
2528
2529static __net_exit void proto_exit_net(struct net *net)
2530{
2531 proc_net_remove(net, "protocols");
2532}
2533
2534
2535static __net_initdata struct pernet_operations proto_net_ops = {
2536 .init = proto_init_net,
2537 .exit = proto_exit_net,
1da177e4
LT		 2538};
2539
2540static int __init proto_init(void)
2541{
14e943db 2542 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2543}
2544
2545subsys_initcall(proto_init);
2546
2547#endif /* PROC_FS */
Clone of davem's net-next-2.6 tree