projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Basic kernel memory functionality for the Memory Controller
[deliverable/linux.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>
5bc1421e 128#include <net/netprio_cgroup.h>
1da177e4
LT		 129
130#include <linux/filter.h>
131
3847ce32
SM		 132#include <trace/events/sock.h>
133
1da177e4
LT		 134#ifdef CONFIG_INET
135#include <net/tcp.h>
136#endif
137
da21f24d
IM		 138/*
139 * Each address family might have different locking rules, so we have
140 * one slock key per address family:
141 */
a5b5bb9a
IM		 142static struct lock_class_key af_family_keys[AF_MAX];
143static struct lock_class_key af_family_slock_keys[AF_MAX];
144
a5b5bb9a
IM		 145/*
146 * Make lock validator output more readable. (we pre-construct these
147 * strings build-time, so that runtime initialization of socket
148 * locks is fast):
149 */
36cbd3dc 150static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 151 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
152 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
153 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
154 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
155 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
156 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
157 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 158 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 159 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 160 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 161 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 162 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 163 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
c7fe3b52 164 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
a5b5bb9a 165};
36cbd3dc 166static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 167 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
168 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
169 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
170 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
171 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
172 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
173 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 174 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 175 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 176 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 177 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 178 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 179 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
c7fe3b52 180 "slock-AF_NFC" , "slock-AF_MAX"
a5b5bb9a 181};
36cbd3dc 182static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 183 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
184 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
185 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
186 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
187 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
188 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
189 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 190 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 191 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 192 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 193 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 194 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 195 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
c7fe3b52 196 "clock-AF_NFC" , "clock-AF_MAX"
443aef0e 197};
da21f24d
IM		 198
199/*
200 * sk_callback_lock locking rules are per-address-family,
201 * so split the lock classes by using a per-AF key:
202 */
203static struct lock_class_key af_callback_keys[AF_MAX];
204
1da177e4
LT		 205/* Take into consideration the size of the struct sk_buff overhead in the
206 * determination of these values, since that is non-constant across
207 * platforms. This makes socket queueing behavior and performance
208 * not depend upon such differences.
209 */
210#define _SK_MEM_PACKETS 256
87fb4b7b 211#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 212#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
213#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
214
215/* Run time adjustable parameters. */
ab32ea5d
BH		 216__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
217__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
218__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
219__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 220
25985edc 221/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 222int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 223EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 224
5bc1421e
NH		 225#if defined(CONFIG_CGROUPS)
226#if !defined(CONFIG_NET_CLS_CGROUP)
f8451725
HX		 227int net_cls_subsys_id = -1;
228EXPORT_SYMBOL_GPL(net_cls_subsys_id);
229#endif
5bc1421e
NH		 230#if !defined(CONFIG_NETPRIO_CGROUP)
231int net_prio_subsys_id = -1;
232EXPORT_SYMBOL_GPL(net_prio_subsys_id);
233#endif
234#endif
f8451725 235
1da177e4
LT		 236static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
237{
238 struct timeval tv;
239
240 if (optlen < sizeof(tv))
241 return -EINVAL;
242 if (copy_from_user(&tv, optval, sizeof(tv)))
243 return -EFAULT;
ba78073e
VA		 244 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
245 return -EDOM;
1da177e4 246
ba78073e 247 if (tv.tv_sec < 0) {
6f11df83
AM		 248 static int warned __read_mostly;
249
ba78073e 250 *timeo_p = 0;
50aab54f 251 if (warned < 10 && net_ratelimit()) {
ba78073e
VA		 252 warned++;
253 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
254 "tries to set negative timeout\n",
ba25f9dc 255 current->comm, task_pid_nr(current));
50aab54f 256 }
ba78073e
VA		 257 return 0;
258 }
1da177e4
LT		 259 *timeo_p = MAX_SCHEDULE_TIMEOUT;
260 if (tv.tv_sec == 0 && tv.tv_usec == 0)
261 return 0;
262 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
263 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
264 return 0;
265}
266
267static void sock_warn_obsolete_bsdism(const char *name)
268{
269 static int warned;
270 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 271 if (strcmp(warncomm, current->comm) && warned < 5) {
272 strcpy(warncomm, current->comm);
1da177e4
LT		 273 printk(KERN_WARNING "process `%s' is using obsolete "
274 "%s SO_BSDCOMPAT\n", warncomm, name);
275 warned++;
276 }
277}
278
08e29af3
ED		 279#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
280
281static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 282{
08e29af3
ED		 283 if (sk->sk_flags & flags) {
284 sk->sk_flags &= ~flags;
285 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 286 net_disable_timestamp();
1da177e4
LT		 287 }
288}
289
290
f0088a50
DV		 291int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
292{
766e9037 293 int err;
f0088a50 294 int skb_len;
3b885787
NH		 295 unsigned long flags;
296 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 297
9ee6b7f1 298 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
f0088a50
DV		 299 number of warnings when compiling with -W --ANK
300 */
301 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
302 (unsigned)sk->sk_rcvbuf) {
766e9037 303 atomic_inc(&sk->sk_drops);
3847ce32 304 trace_sock_rcvqueue_full(sk, skb);
766e9037 305 return -ENOMEM;
f0088a50
DV		 306 }
307
fda9ef5d 308 err = sk_filter(sk, skb);
f0088a50 309 if (err)
766e9037 310 return err;
f0088a50 311
3ab224be 312 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 313 atomic_inc(&sk->sk_drops);
314 return -ENOBUFS;
3ab224be
HA		 315 }
316
f0088a50
DV		 317 skb->dev = NULL;
318 skb_set_owner_r(skb, sk);
49ad9599 319
f0088a50
DV		 320 /* Cache the SKB length before we tack it onto the receive
321 * queue. Once it is added it no longer belongs to us and
322 * may be freed by other threads of control pulling packets
323 * from the queue.
324 */
325 skb_len = skb->len;
326
7fee226a
ED		 327 /* we escape from rcu protected region, make sure we dont leak
328 * a norefcounted dst
329 */
330 skb_dst_force(skb);
331
3b885787
NH		 332 spin_lock_irqsave(&list->lock, flags);
333 skb->dropcount = atomic_read(&sk->sk_drops);
334 __skb_queue_tail(list, skb);
335 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 336
337 if (!sock_flag(sk, SOCK_DEAD))
338 sk->sk_data_ready(sk, skb_len);
766e9037 339 return 0;
f0088a50
DV		 340}
341EXPORT_SYMBOL(sock_queue_rcv_skb);
342
58a5a7b9 343int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 344{
345 int rc = NET_RX_SUCCESS;
346
fda9ef5d 347 if (sk_filter(sk, skb))
f0088a50
DV		 348 goto discard_and_relse;
349
350 skb->dev = NULL;
351
c377411f
ED		 352 if (sk_rcvqueues_full(sk, skb)) {
353 atomic_inc(&sk->sk_drops);
354 goto discard_and_relse;
355 }
58a5a7b9
ACM		 356 if (nested)
357 bh_lock_sock_nested(sk);
358 else
359 bh_lock_sock(sk);
a5b5bb9a
IM		 360 if (!sock_owned_by_user(sk)) {
361 /*
362 * trylock + unlock semantics:
363 */
364 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
365
c57943a1 366 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 367
368 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
a3a858ff 369 } else if (sk_add_backlog(sk, skb)) {
8eae939f
ZY		 370 bh_unlock_sock(sk);
371 atomic_inc(&sk->sk_drops);
372 goto discard_and_relse;
373 }
374
f0088a50
DV		 375 bh_unlock_sock(sk);
376out:
377 sock_put(sk);
378 return rc;
379discard_and_relse:
380 kfree_skb(skb);
381 goto out;
382}
383EXPORT_SYMBOL(sk_receive_skb);
384
ea94ff3b
KK		 385void sk_reset_txq(struct sock *sk)
386{
387 sk_tx_queue_clear(sk);
388}
389EXPORT_SYMBOL(sk_reset_txq);
390
f0088a50
DV		 391struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
392{
b6c6712a 393 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 394
395 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 396 sk_tx_queue_clear(sk);
a9b3cd7f 397 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 398 dst_release(dst);
399 return NULL;
400 }
401
402 return dst;
403}
404EXPORT_SYMBOL(__sk_dst_check);
405
406struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
407{
408 struct dst_entry *dst = sk_dst_get(sk);
409
410 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
411 sk_dst_reset(sk);
412 dst_release(dst);
413 return NULL;
414 }
415
416 return dst;
417}
418EXPORT_SYMBOL(sk_dst_check);
419
4878809f
DM		 420static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
421{
422 int ret = -ENOPROTOOPT;
423#ifdef CONFIG_NETDEVICES
3b1e0a65 424 struct net *net = sock_net(sk);
4878809f
DM		 425 char devname[IFNAMSIZ];
426 int index;
427
428 /* Sorry... */
429 ret = -EPERM;
430 if (!capable(CAP_NET_RAW))
431 goto out;
432
433 ret = -EINVAL;
434 if (optlen < 0)
435 goto out;
436
437 /* Bind this socket to a particular device like "eth0",
438 * as specified in the passed interface name. If the
439 * name is "" or the option length is zero the socket
440 * is not bound.
441 */
442 if (optlen > IFNAMSIZ - 1)
443 optlen = IFNAMSIZ - 1;
444 memset(devname, 0, sizeof(devname));
445
446 ret = -EFAULT;
447 if (copy_from_user(devname, optval, optlen))
448 goto out;
449
000ba2e4
DM		 450 index = 0;
451 if (devname[0] != '\0') {
bf8e56bf 452 struct net_device *dev;
4878809f 453
bf8e56bf
ED		 454 rcu_read_lock();
455 dev = dev_get_by_name_rcu(net, devname);
456 if (dev)
457 index = dev->ifindex;
458 rcu_read_unlock();
4878809f
DM		 459 ret = -ENODEV;
460 if (!dev)
461 goto out;
4878809f
DM		 462 }
463
464 lock_sock(sk);
465 sk->sk_bound_dev_if = index;
466 sk_dst_reset(sk);
467 release_sock(sk);
468
469 ret = 0;
470
471out:
472#endif
473
474 return ret;
475}
476
c0ef877b
PE		 477static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
478{
479 if (valbool)
480 sock_set_flag(sk, bit);
481 else
482 sock_reset_flag(sk, bit);
483}
484
1da177e4
LT		 485/*
486 * This is meant for all protocols to use and covers goings on
487 * at the socket level. Everything here is generic.
488 */
489
490int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 491 char __user *optval, unsigned int optlen)
1da177e4 492{
2a91525c 493 struct sock *sk = sock->sk;
1da177e4
LT		 494 int val;
495 int valbool;
496 struct linger ling;
497 int ret = 0;
4ec93edb 498
1da177e4
LT		 499 /*
500 * Options without arguments
501 */
502
4878809f
DM		 503 if (optname == SO_BINDTODEVICE)
504 return sock_bindtodevice(sk, optval, optlen);
505
e71a4783
SH		 506 if (optlen < sizeof(int))
507 return -EINVAL;
4ec93edb 508
1da177e4
LT		 509 if (get_user(val, (int __user *)optval))
510 return -EFAULT;
4ec93edb 511
2a91525c 512 valbool = val ? 1 : 0;
1da177e4
LT		 513
514 lock_sock(sk);
515
2a91525c 516 switch (optname) {
e71a4783 517 case SO_DEBUG:
2a91525c 518 if (val && !capable(CAP_NET_ADMIN))
e71a4783 519 ret = -EACCES;
2a91525c 520 else
c0ef877b 521 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 522 break;
523 case SO_REUSEADDR:
524 sk->sk_reuse = valbool;
525 break;
526 case SO_TYPE:
49c794e9 527 case SO_PROTOCOL:
0d6038ee 528 case SO_DOMAIN:
e71a4783
SH		 529 case SO_ERROR:
530 ret = -ENOPROTOOPT;
531 break;
532 case SO_DONTROUTE:
c0ef877b 533 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 534 break;
535 case SO_BROADCAST:
536 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
537 break;
538 case SO_SNDBUF:
539 /* Don't error on this BSD doesn't and if you think
540 about it this is right. Otherwise apps have to
541 play 'guess the biggest size' games. RCVBUF/SNDBUF
542 are treated in BSD as hints */
543
544 if (val > sysctl_wmem_max)
545 val = sysctl_wmem_max;
b0573dea 546set_sndbuf:
e71a4783
SH		 547 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
548 if ((val * 2) < SOCK_MIN_SNDBUF)
549 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
550 else
551 sk->sk_sndbuf = val * 2;
1da177e4 552
e71a4783
SH		 553 /*
554 * Wake up sending tasks if we
555 * upped the value.
556 */
557 sk->sk_write_space(sk);
558 break;
1da177e4 559
e71a4783
SH		 560 case SO_SNDBUFFORCE:
561 if (!capable(CAP_NET_ADMIN)) {
562 ret = -EPERM;
563 break;
564 }
565 goto set_sndbuf;
b0573dea 566
e71a4783
SH		 567 case SO_RCVBUF:
568 /* Don't error on this BSD doesn't and if you think
569 about it this is right. Otherwise apps have to
570 play 'guess the biggest size' games. RCVBUF/SNDBUF
571 are treated in BSD as hints */
4ec93edb 572
e71a4783
SH		 573 if (val > sysctl_rmem_max)
574 val = sysctl_rmem_max;
b0573dea 575set_rcvbuf:
e71a4783
SH		 576 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
577 /*
578 * We double it on the way in to account for
579 * "struct sk_buff" etc. overhead. Applications
580 * assume that the SO_RCVBUF setting they make will
581 * allow that much actual data to be received on that
582 * socket.
583 *
584 * Applications are unaware that "struct sk_buff" and
585 * other overheads allocate from the receive buffer
586 * during socket buffer allocation.
587 *
588 * And after considering the possible alternatives,
589 * returning the value we actually used in getsockopt
590 * is the most desirable behavior.
591 */
592 if ((val * 2) < SOCK_MIN_RCVBUF)
593 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
594 else
595 sk->sk_rcvbuf = val * 2;
596 break;
597
598 case SO_RCVBUFFORCE:
599 if (!capable(CAP_NET_ADMIN)) {
600 ret = -EPERM;
1da177e4 601 break;
e71a4783
SH		 602 }
603 goto set_rcvbuf;
1da177e4 604
e71a4783 605 case SO_KEEPALIVE:
1da177e4 606#ifdef CONFIG_INET
e71a4783
SH		 607 if (sk->sk_protocol == IPPROTO_TCP)
608 tcp_set_keepalive(sk, valbool);
1da177e4 609#endif
e71a4783
SH		 610 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
611 break;
612
613 case SO_OOBINLINE:
614 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
615 break;
616
617 case SO_NO_CHECK:
618 sk->sk_no_check = valbool;
619 break;
620
621 case SO_PRIORITY:
622 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
623 sk->sk_priority = val;
624 else
625 ret = -EPERM;
626 break;
627
628 case SO_LINGER:
629 if (optlen < sizeof(ling)) {
630 ret = -EINVAL; /* 1003.1g */
1da177e4 631 break;
e71a4783 632 }
2a91525c 633 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 634 ret = -EFAULT;
1da177e4 635 break;
e71a4783
SH		 636 }
637 if (!ling.l_onoff)
638 sock_reset_flag(sk, SOCK_LINGER);
639 else {
1da177e4 640#if (BITS_PER_LONG == 32)
e71a4783
SH		 641 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
642 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 643 else
e71a4783
SH		 644#endif
645 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
646 sock_set_flag(sk, SOCK_LINGER);
647 }
648 break;
649
650 case SO_BSDCOMPAT:
651 sock_warn_obsolete_bsdism("setsockopt");
652 break;
653
654 case SO_PASSCRED:
655 if (valbool)
656 set_bit(SOCK_PASSCRED, &sock->flags);
657 else
658 clear_bit(SOCK_PASSCRED, &sock->flags);
659 break;
660
661 case SO_TIMESTAMP:
92f37fd2 662 case SO_TIMESTAMPNS:
e71a4783 663 if (valbool) {
92f37fd2
ED		 664 if (optname == SO_TIMESTAMP)
665 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
666 else
667 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 668 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 669 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 670 } else {
e71a4783 671 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 672 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
673 }
e71a4783
SH		 674 break;
675
20d49473
PO		 676 case SO_TIMESTAMPING:
677 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 678 ret = -EINVAL;
20d49473
PO		 679 break;
680 }
681 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
682 val & SOF_TIMESTAMPING_TX_HARDWARE);
683 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
684 val & SOF_TIMESTAMPING_TX_SOFTWARE);
685 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
686 val & SOF_TIMESTAMPING_RX_HARDWARE);
687 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
688 sock_enable_timestamp(sk,
689 SOCK_TIMESTAMPING_RX_SOFTWARE);
690 else
691 sock_disable_timestamp(sk,
08e29af3 692 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
20d49473
PO		 693 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
694 val & SOF_TIMESTAMPING_SOFTWARE);
695 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
696 val & SOF_TIMESTAMPING_SYS_HARDWARE);
697 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
698 val & SOF_TIMESTAMPING_RAW_HARDWARE);
699 break;
700
e71a4783
SH		 701 case SO_RCVLOWAT:
702 if (val < 0)
703 val = INT_MAX;
704 sk->sk_rcvlowat = val ? : 1;
705 break;
706
707 case SO_RCVTIMEO:
708 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
709 break;
710
711 case SO_SNDTIMEO:
712 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
713 break;
1da177e4 714
e71a4783
SH		 715 case SO_ATTACH_FILTER:
716 ret = -EINVAL;
717 if (optlen == sizeof(struct sock_fprog)) {
718 struct sock_fprog fprog;
1da177e4 719
e71a4783
SH		 720 ret = -EFAULT;
721 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 722 break;
e71a4783
SH		 723
724 ret = sk_attach_filter(&fprog, sk);
725 }
726 break;
727
728 case SO_DETACH_FILTER:
55b33325 729 ret = sk_detach_filter(sk);
e71a4783 730 break;
1da177e4 731
e71a4783
SH		 732 case SO_PASSSEC:
733 if (valbool)
734 set_bit(SOCK_PASSSEC, &sock->flags);
735 else
736 clear_bit(SOCK_PASSSEC, &sock->flags);
737 break;
4a19ec58
LAT		 738 case SO_MARK:
739 if (!capable(CAP_NET_ADMIN))
740 ret = -EPERM;
2a91525c 741 else
4a19ec58 742 sk->sk_mark = val;
4a19ec58 743 break;
877ce7c1 744
1da177e4
LT		 745 /* We implement the SO_SNDLOWAT etc to
746 not be settable (1003.1g 5.3) */
3b885787 747 case SO_RXQ_OVFL:
8083f0fc 748 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 749 break;
6e3e939f
JB		 750
751 case SO_WIFI_STATUS:
752 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
753 break;
754
e71a4783
SH		 755 default:
756 ret = -ENOPROTOOPT;
757 break;
4ec93edb 758 }
1da177e4
LT		 759 release_sock(sk);
760 return ret;
761}
2a91525c 762EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 763
764
3f551f94
EB		 765void cred_to_ucred(struct pid *pid, const struct cred *cred,
766 struct ucred *ucred)
767{
768 ucred->pid = pid_vnr(pid);
769 ucred->uid = ucred->gid = -1;
770 if (cred) {
771 struct user_namespace *current_ns = current_user_ns();
772
773 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
774 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
775 }
776}
3924773a 777EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 778
1da177e4
LT		 779int sock_getsockopt(struct socket *sock, int level, int optname,
780 char __user *optval, int __user *optlen)
781{
782 struct sock *sk = sock->sk;
4ec93edb 783
e71a4783 784 union {
4ec93edb
YH		 785 int val;
786 struct linger ling;
1da177e4
LT		 787 struct timeval tm;
788 } v;
4ec93edb 789
4d0392be 790 int lv = sizeof(int);
1da177e4 791 int len;
4ec93edb 792
e71a4783 793 if (get_user(len, optlen))
4ec93edb 794 return -EFAULT;
e71a4783 795 if (len < 0)
1da177e4 796 return -EINVAL;
4ec93edb 797
50fee1de 798 memset(&v, 0, sizeof(v));
df0bca04 799
2a91525c 800 switch (optname) {
e71a4783
SH		 801 case SO_DEBUG:
802 v.val = sock_flag(sk, SOCK_DBG);
803 break;
804
805 case SO_DONTROUTE:
806 v.val = sock_flag(sk, SOCK_LOCALROUTE);
807 break;
808
809 case SO_BROADCAST:
810 v.val = !!sock_flag(sk, SOCK_BROADCAST);
811 break;
812
813 case SO_SNDBUF:
814 v.val = sk->sk_sndbuf;
815 break;
816
817 case SO_RCVBUF:
818 v.val = sk->sk_rcvbuf;
819 break;
820
821 case SO_REUSEADDR:
822 v.val = sk->sk_reuse;
823 break;
824
825 case SO_KEEPALIVE:
826 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
827 break;
828
829 case SO_TYPE:
830 v.val = sk->sk_type;
831 break;
832
49c794e9
JE		 833 case SO_PROTOCOL:
834 v.val = sk->sk_protocol;
835 break;
836
0d6038ee
JE		 837 case SO_DOMAIN:
838 v.val = sk->sk_family;
839 break;
840
e71a4783
SH		 841 case SO_ERROR:
842 v.val = -sock_error(sk);
2a91525c 843 if (v.val == 0)
e71a4783
SH		 844 v.val = xchg(&sk->sk_err_soft, 0);
845 break;
846
847 case SO_OOBINLINE:
848 v.val = !!sock_flag(sk, SOCK_URGINLINE);
849 break;
850
851 case SO_NO_CHECK:
852 v.val = sk->sk_no_check;
853 break;
854
855 case SO_PRIORITY:
856 v.val = sk->sk_priority;
857 break;
858
859 case SO_LINGER:
860 lv = sizeof(v.ling);
861 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
862 v.ling.l_linger = sk->sk_lingertime / HZ;
863 break;
864
865 case SO_BSDCOMPAT:
866 sock_warn_obsolete_bsdism("getsockopt");
867 break;
868
869 case SO_TIMESTAMP:
92f37fd2
ED		 870 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
871 !sock_flag(sk, SOCK_RCVTSTAMPNS);
872 break;
873
874 case SO_TIMESTAMPNS:
875 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 876 break;
877
20d49473
PO		 878 case SO_TIMESTAMPING:
879 v.val = 0;
880 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
881 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
882 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
883 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
884 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
885 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
886 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
887 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
888 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
889 v.val |= SOF_TIMESTAMPING_SOFTWARE;
890 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
891 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
892 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
893 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
894 break;
895
e71a4783 896 case SO_RCVTIMEO:
2a91525c 897 lv = sizeof(struct timeval);
e71a4783
SH		 898 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
899 v.tm.tv_sec = 0;
900 v.tm.tv_usec = 0;
901 } else {
902 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
903 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
904 }
905 break;
906
907 case SO_SNDTIMEO:
2a91525c 908 lv = sizeof(struct timeval);
e71a4783
SH		 909 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
910 v.tm.tv_sec = 0;
911 v.tm.tv_usec = 0;
912 } else {
913 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
914 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
915 }
916 break;
1da177e4 917
e71a4783
SH		 918 case SO_RCVLOWAT:
919 v.val = sk->sk_rcvlowat;
920 break;
1da177e4 921
e71a4783 922 case SO_SNDLOWAT:
2a91525c 923 v.val = 1;
e71a4783 924 break;
1da177e4 925
e71a4783
SH		 926 case SO_PASSCRED:
927 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
928 break;
1da177e4 929
e71a4783 930 case SO_PEERCRED:
109f6e39
EB		 931 {
932 struct ucred peercred;
933 if (len > sizeof(peercred))
934 len = sizeof(peercred);
935 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
936 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 937 return -EFAULT;
938 goto lenout;
109f6e39 939 }
1da177e4 940
e71a4783
SH		 941 case SO_PEERNAME:
942 {
943 char address[128];
944
945 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
946 return -ENOTCONN;
947 if (lv < len)
948 return -EINVAL;
949 if (copy_to_user(optval, address, len))
950 return -EFAULT;
951 goto lenout;
952 }
1da177e4 953
e71a4783
SH		 954 /* Dubious BSD thing... Probably nobody even uses it, but
955 * the UNIX standard wants it for whatever reason... -DaveM
956 */
957 case SO_ACCEPTCONN:
958 v.val = sk->sk_state == TCP_LISTEN;
959 break;
1da177e4 960
e71a4783
SH		 961 case SO_PASSSEC:
962 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
963 break;
877ce7c1 964
e71a4783
SH		 965 case SO_PEERSEC:
966 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 967
4a19ec58
LAT		 968 case SO_MARK:
969 v.val = sk->sk_mark;
970 break;
971
3b885787
NH		 972 case SO_RXQ_OVFL:
973 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
974 break;
975
6e3e939f
JB		 976 case SO_WIFI_STATUS:
977 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
978 break;
979
e71a4783
SH		 980 default:
981 return -ENOPROTOOPT;
1da177e4 982 }
e71a4783 983
1da177e4
LT		 984 if (len > lv)
985 len = lv;
986 if (copy_to_user(optval, &v, len))
987 return -EFAULT;
988lenout:
4ec93edb
YH		 989 if (put_user(len, optlen))
990 return -EFAULT;
991 return 0;
1da177e4
LT		 992}
993
a5b5bb9a
IM		 994/*
995 * Initialize an sk_lock.
996 *
997 * (We also register the sk_lock with the lock validator.)
998 */
b6f99a21 999static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 1000{
ed07536e
PZ		 1001 sock_lock_init_class_and_name(sk,
1002 af_family_slock_key_strings[sk->sk_family],
1003 af_family_slock_keys + sk->sk_family,
1004 af_family_key_strings[sk->sk_family],
1005 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 1006}
1007
4dc6dc71
ED		 1008/*
1009 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1010 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1011 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1012 */
f1a6c4da
PE		 1013static void sock_copy(struct sock *nsk, const struct sock *osk)
1014{
1015#ifdef CONFIG_SECURITY_NETWORK
1016 void *sptr = nsk->sk_security;
1017#endif
68835aba
ED		 1018 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1019
1020 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1021 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1022
f1a6c4da
PE		 1023#ifdef CONFIG_SECURITY_NETWORK
1024 nsk->sk_security = sptr;
1025 security_sk_clone(osk, nsk);
1026#endif
1027}
1028
fcbdf09d
OP		 1029/*
1030 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1031 * un-modified. Special care is taken when initializing object to zero.
1032 */
1033static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1034{
1035 if (offsetof(struct sock, sk_node.next) != 0)
1036 memset(sk, 0, offsetof(struct sock, sk_node.next));
1037 memset(&sk->sk_node.pprev, 0,
1038 size - offsetof(struct sock, sk_node.pprev));
1039}
1040
1041void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1042{
1043 unsigned long nulls1, nulls2;
1044
1045 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1046 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1047 if (nulls1 > nulls2)
1048 swap(nulls1, nulls2);
1049
1050 if (nulls1 != 0)
1051 memset((char *)sk, 0, nulls1);
1052 memset((char *)sk + nulls1 + sizeof(void *), 0,
1053 nulls2 - nulls1 - sizeof(void *));
1054 memset((char *)sk + nulls2 + sizeof(void *), 0,
1055 size - nulls2 - sizeof(void *));
1056}
1057EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1058
2e4afe7b
PE		 1059static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1060 int family)
c308c1b2
PE		 1061{
1062 struct sock *sk;
1063 struct kmem_cache *slab;
1064
1065 slab = prot->slab;
e912b114
ED		 1066 if (slab != NULL) {
1067 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1068 if (!sk)
1069 return sk;
1070 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1071 if (prot->clear_sk)
1072 prot->clear_sk(sk, prot->obj_size);
1073 else
1074 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1075 }
fcbdf09d 1076 } else
c308c1b2
PE		 1077 sk = kmalloc(prot->obj_size, priority);
1078
2e4afe7b 1079 if (sk != NULL) {
a98b65a3
VN		 1080 kmemcheck_annotate_bitfield(sk, flags);
1081
2e4afe7b
PE		 1082 if (security_sk_alloc(sk, family, priority))
1083 goto out_free;
1084
1085 if (!try_module_get(prot->owner))
1086 goto out_free_sec;
e022f0b4 1087 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1088 }
1089
c308c1b2 1090 return sk;
2e4afe7b
PE		 1091
1092out_free_sec:
1093 security_sk_free(sk);
1094out_free:
1095 if (slab != NULL)
1096 kmem_cache_free(slab, sk);
1097 else
1098 kfree(sk);
1099 return NULL;
c308c1b2
PE		 1100}
1101
1102static void sk_prot_free(struct proto *prot, struct sock *sk)
1103{
1104 struct kmem_cache *slab;
2e4afe7b 1105 struct module *owner;
c308c1b2 1106
2e4afe7b 1107 owner = prot->owner;
c308c1b2 1108 slab = prot->slab;
2e4afe7b
PE		 1109
1110 security_sk_free(sk);
c308c1b2
PE		 1111 if (slab != NULL)
1112 kmem_cache_free(slab, sk);
1113 else
1114 kfree(sk);
2e4afe7b 1115 module_put(owner);
c308c1b2
PE		 1116}
1117
f8451725
HX		 1118#ifdef CONFIG_CGROUPS
1119void sock_update_classid(struct sock *sk)
1120{
1144182a 1121 u32 classid;
f8451725 1122
1144182a
PM		 1123 rcu_read_lock(); /* doing current task, which cannot vanish. */
1124 classid = task_cls_classid(current);
1125 rcu_read_unlock();
f8451725
HX		 1126 if (classid && classid != sk->sk_classid)
1127 sk->sk_classid = classid;
1128}
82862742 1129EXPORT_SYMBOL(sock_update_classid);
5bc1421e
NH		 1130
1131void sock_update_netprioidx(struct sock *sk)
1132{
1133 struct cgroup_netprio_state *state;
1134 if (in_interrupt())
1135 return;
1136 rcu_read_lock();
1137 state = task_netprio_state(current);
1138 sk->sk_cgrp_prioidx = state ? state->prioidx : 0;
1139 rcu_read_unlock();
1140}
1141EXPORT_SYMBOL_GPL(sock_update_netprioidx);
f8451725
HX		 1142#endif
1143
1da177e4
LT		 1144/**
1145 * sk_alloc - All socket objects are allocated here
c4ea43c5 1146 * @net: the applicable net namespace
4dc3b16b
PP		 1147 * @family: protocol family
1148 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1149 * @prot: struct proto associated with this new sock instance
1da177e4 1150 */
1b8d7ae4 1151struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1152 struct proto *prot)
1da177e4 1153{
c308c1b2 1154 struct sock *sk;
1da177e4 1155
154adbc8 1156 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1157 if (sk) {
154adbc8
PE		 1158 sk->sk_family = family;
1159 /*
1160 * See comment in struct sock definition to understand
1161 * why we need sk_prot_creator -acme
1162 */
1163 sk->sk_prot = sk->sk_prot_creator = prot;
1164 sock_lock_init(sk);
3b1e0a65 1165 sock_net_set(sk, get_net(net));
d66ee058 1166 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1167
1168 sock_update_classid(sk);
5bc1421e 1169 sock_update_netprioidx(sk);
1da177e4 1170 }
a79af59e 1171
2e4afe7b 1172 return sk;
1da177e4 1173}
2a91525c 1174EXPORT_SYMBOL(sk_alloc);
1da177e4 1175
2b85a34e 1176static void __sk_free(struct sock *sk)
1da177e4
LT		 1177{
1178 struct sk_filter *filter;
1da177e4
LT		 1179
1180 if (sk->sk_destruct)
1181 sk->sk_destruct(sk);
1182
a898def2
PM		 1183 filter = rcu_dereference_check(sk->sk_filter,
1184 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1185 if (filter) {
309dd5fc 1186 sk_filter_uncharge(sk, filter);
a9b3cd7f 1187 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1188 }
1189
08e29af3 1190 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4
LT		 1191
1192 if (atomic_read(&sk->sk_omem_alloc))
1193 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1194 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1195
109f6e39
EB		 1196 if (sk->sk_peer_cred)
1197 put_cred(sk->sk_peer_cred);
1198 put_pid(sk->sk_peer_pid);
3b1e0a65 1199 put_net(sock_net(sk));
c308c1b2 1200 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1201}
2b85a34e
ED		 1202
1203void sk_free(struct sock *sk)
1204{
1205 /*
25985edc 1206 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1207 * some packets are still in some tx queue.
1208 * If not null, sock_wfree() will call __sk_free(sk) later
1209 */
1210 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1211 __sk_free(sk);
1212}
2a91525c 1213EXPORT_SYMBOL(sk_free);
1da177e4 1214
edf02087 1215/*
25985edc
LDM		 1216 * Last sock_put should drop reference to sk->sk_net. It has already
1217 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1218 * is not an option.
25985edc 1219 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1220 * destroy it in the context of init_net.
1221 */
1222void sk_release_kernel(struct sock *sk)
1223{
1224 if (sk == NULL || sk->sk_socket == NULL)
1225 return;
1226
1227 sock_hold(sk);
1228 sock_release(sk->sk_socket);
65a18ec5 1229 release_net(sock_net(sk));
3b1e0a65 1230 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1231 sock_put(sk);
1232}
45af1754 1233EXPORT_SYMBOL(sk_release_kernel);
edf02087 1234
e56c57d0
ED		 1235/**
1236 * sk_clone_lock - clone a socket, and lock its clone
1237 * @sk: the socket to clone
1238 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1239 *
1240 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1241 */
1242struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1243{
8fd1d178 1244 struct sock *newsk;
87d11ceb 1245
8fd1d178 1246 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1247 if (newsk != NULL) {
1248 struct sk_filter *filter;
1249
892c141e 1250 sock_copy(newsk, sk);
87d11ceb
ACM		 1251
1252 /* SANITY */
3b1e0a65 1253 get_net(sock_net(newsk));
87d11ceb
ACM		 1254 sk_node_init(&newsk->sk_node);
1255 sock_lock_init(newsk);
1256 bh_lock_sock(newsk);
fa438ccf 1257 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1258 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1259
1260 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1261 /*
1262 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1263 */
1264 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1265 atomic_set(&newsk->sk_omem_alloc, 0);
1266 skb_queue_head_init(&newsk->sk_receive_queue);
1267 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1268#ifdef CONFIG_NET_DMA
1269 skb_queue_head_init(&newsk->sk_async_wait_queue);
1270#endif
87d11ceb 1271
b6c6712a 1272 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1273 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1274 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1275 af_callback_keys + newsk->sk_family,
1276 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1277
1278 newsk->sk_dst_cache = NULL;
1279 newsk->sk_wmem_queued = 0;
1280 newsk->sk_forward_alloc = 0;
1281 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1282 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1283
1284 sock_reset_flag(newsk, SOCK_DONE);
1285 skb_queue_head_init(&newsk->sk_error_queue);
1286
0d7da9dd 1287 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1288 if (filter != NULL)
1289 sk_filter_charge(newsk, filter);
1290
1291 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1292 /* It is still raw copy of parent, so invalidate
1293 * destructor and make plain sk_free() */
1294 newsk->sk_destruct = NULL;
b0691c8e 1295 bh_unlock_sock(newsk);
87d11ceb
ACM		 1296 sk_free(newsk);
1297 newsk = NULL;
1298 goto out;
1299 }
1300
1301 newsk->sk_err = 0;
1302 newsk->sk_priority = 0;
4dc6dc71
ED		 1303 /*
1304 * Before updating sk_refcnt, we must commit prior changes to memory
1305 * (Documentation/RCU/rculist_nulls.txt for details)
1306 */
1307 smp_wmb();
87d11ceb
ACM		 1308 atomic_set(&newsk->sk_refcnt, 2);
1309
1310 /*
1311 * Increment the counter in the same struct proto as the master
1312 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1313 * is the same as sk->sk_prot->socks, as this field was copied
1314 * with memcpy).
1315 *
1316 * This _changes_ the previous behaviour, where
1317 * tcp_create_openreq_child always was incrementing the
1318 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1319 * to be taken into account in all callers. -acme
1320 */
1321 sk_refcnt_debug_inc(newsk);
972692e0 1322 sk_set_socket(newsk, NULL);
43815482 1323 newsk->sk_wq = NULL;
87d11ceb
ACM		 1324
1325 if (newsk->sk_prot->sockets_allocated)
1748376b 1326 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
704da560 1327
08e29af3 1328 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1329 net_enable_timestamp();
87d11ceb
ACM		 1330 }
1331out:
1332 return newsk;
1333}
e56c57d0 1334EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1335
9958089a
AK		 1336void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1337{
1338 __sk_dst_set(sk, dst);
1339 sk->sk_route_caps = dst->dev->features;
1340 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1341 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1342 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1343 if (sk_can_gso(sk)) {
82cc1a7a 1344 if (dst->header_len) {
9958089a 1345 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1346 } else {
9958089a 1347 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1348 sk->sk_gso_max_size = dst->dev->gso_max_size;
1349 }
9958089a
AK		 1350 }
1351}
1352EXPORT_SYMBOL_GPL(sk_setup_caps);
1353
1da177e4
LT		 1354void __init sk_init(void)
1355{
4481374c 1356 if (totalram_pages <= 4096) {
1da177e4
LT		 1357 sysctl_wmem_max = 32767;
1358 sysctl_rmem_max = 32767;
1359 sysctl_wmem_default = 32767;
1360 sysctl_rmem_default = 32767;
4481374c 1361 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1362 sysctl_wmem_max = 131071;
1363 sysctl_rmem_max = 131071;
1364 }
1365}
1366
1367/*
1368 * Simple resource managers for sockets.
1369 */
1370
1371
4ec93edb
YH		 1372/*
1373 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1374 */
1375void sock_wfree(struct sk_buff *skb)
1376{
1377 struct sock *sk = skb->sk;
d99927f4 1378 unsigned int len = skb->truesize;
1da177e4 1379
d99927f4
ED		 1380 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1381 /*
1382 * Keep a reference on sk_wmem_alloc, this will be released
1383 * after sk_write_space() call
1384 */
1385 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1386 sk->sk_write_space(sk);
d99927f4
ED		 1387 len = 1;
1388 }
2b85a34e 1389 /*
d99927f4
ED		 1390 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1391 * could not do because of in-flight packets
2b85a34e 1392 */
d99927f4 1393 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1394 __sk_free(sk);
1da177e4 1395}
2a91525c 1396EXPORT_SYMBOL(sock_wfree);
1da177e4 1397
4ec93edb
YH		 1398/*
1399 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1400 */
1401void sock_rfree(struct sk_buff *skb)
1402{
1403 struct sock *sk = skb->sk;
d361fd59 1404 unsigned int len = skb->truesize;
1da177e4 1405
d361fd59
ED		 1406 atomic_sub(len, &sk->sk_rmem_alloc);
1407 sk_mem_uncharge(sk, len);
1da177e4 1408}
2a91525c 1409EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1410
1411
1412int sock_i_uid(struct sock *sk)
1413{
1414 int uid;
1415
f064af1e 1416 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1417 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1418 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1419 return uid;
1420}
2a91525c 1421EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1422
1423unsigned long sock_i_ino(struct sock *sk)
1424{
1425 unsigned long ino;
1426
f064af1e 1427 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1428 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1429 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1430 return ino;
1431}
2a91525c 1432EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1433
1434/*
1435 * Allocate a skb from the socket's send buffer.
1436 */
86a76caf 1437struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1438 gfp_t priority)
1da177e4
LT		 1439{
1440 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1441 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1442 if (skb) {
1443 skb_set_owner_w(skb, sk);
1444 return skb;
1445 }
1446 }
1447 return NULL;
1448}
2a91525c 1449EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1450
1451/*
1452 * Allocate a skb from the socket's receive buffer.
4ec93edb 1453 */
86a76caf 1454struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1455 gfp_t priority)
1da177e4
LT		 1456{
1457 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1458 struct sk_buff *skb = alloc_skb(size, priority);
1459 if (skb) {
1460 skb_set_owner_r(skb, sk);
1461 return skb;
1462 }
1463 }
1464 return NULL;
1465}
1466
4ec93edb 1467/*
1da177e4 1468 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1469 */
dd0fc66f 1470void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1471{
1472 if ((unsigned)size <= sysctl_optmem_max &&
1473 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1474 void *mem;
1475 /* First do the add, to avoid the race if kmalloc
4ec93edb 1476 * might sleep.
1da177e4
LT		 1477 */
1478 atomic_add(size, &sk->sk_omem_alloc);
1479 mem = kmalloc(size, priority);
1480 if (mem)
1481 return mem;
1482 atomic_sub(size, &sk->sk_omem_alloc);
1483 }
1484 return NULL;
1485}
2a91525c 1486EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1487
1488/*
1489 * Free an option memory block.
1490 */
1491void sock_kfree_s(struct sock *sk, void *mem, int size)
1492{
1493 kfree(mem);
1494 atomic_sub(size, &sk->sk_omem_alloc);
1495}
2a91525c 1496EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1497
1498/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1499 I think, these locks should be removed for datagram sockets.
1500 */
2a91525c 1501static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1502{
1503 DEFINE_WAIT(wait);
1504
1505 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1506 for (;;) {
1507 if (!timeo)
1508 break;
1509 if (signal_pending(current))
1510 break;
1511 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1512 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1513 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1514 break;
1515 if (sk->sk_shutdown & SEND_SHUTDOWN)
1516 break;
1517 if (sk->sk_err)
1518 break;
1519 timeo = schedule_timeout(timeo);
1520 }
aa395145 1521 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1522 return timeo;
1523}
1524
1525
1526/*
1527 * Generic send/receive buffer handlers
1528 */
1529
4cc7f68d
HX		 1530struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1531 unsigned long data_len, int noblock,
1532 int *errcode)
1da177e4
LT		 1533{
1534 struct sk_buff *skb;
7d877f3b 1535 gfp_t gfp_mask;
1da177e4
LT		 1536 long timeo;
1537 int err;
1538
1539 gfp_mask = sk->sk_allocation;
1540 if (gfp_mask & __GFP_WAIT)
1541 gfp_mask |= __GFP_REPEAT;
1542
1543 timeo = sock_sndtimeo(sk, noblock);
1544 while (1) {
1545 err = sock_error(sk);
1546 if (err != 0)
1547 goto failure;
1548
1549 err = -EPIPE;
1550 if (sk->sk_shutdown & SEND_SHUTDOWN)
1551 goto failure;
1552
1553 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1554 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1555 if (skb) {
1556 int npages;
1557 int i;
1558
1559 /* No pages, we're done... */
1560 if (!data_len)
1561 break;
1562
1563 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1564 skb->truesize += data_len;
1565 skb_shinfo(skb)->nr_frags = npages;
1566 for (i = 0; i < npages; i++) {
1567 struct page *page;
1da177e4
LT		 1568
1569 page = alloc_pages(sk->sk_allocation, 0);
1570 if (!page) {
1571 err = -ENOBUFS;
1572 skb_shinfo(skb)->nr_frags = i;
1573 kfree_skb(skb);
1574 goto failure;
1575 }
1576
ea2ab693
IC		 1577 __skb_fill_page_desc(skb, i,
1578 page, 0,
1579 (data_len >= PAGE_SIZE ?
1580 PAGE_SIZE :
1581 data_len));
1da177e4
LT		 1582 data_len -= PAGE_SIZE;
1583 }
1584
1585 /* Full success... */
1586 break;
1587 }
1588 err = -ENOBUFS;
1589 goto failure;
1590 }
1591 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1592 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1593 err = -EAGAIN;
1594 if (!timeo)
1595 goto failure;
1596 if (signal_pending(current))
1597 goto interrupted;
1598 timeo = sock_wait_for_wmem(sk, timeo);
1599 }
1600
1601 skb_set_owner_w(skb, sk);
1602 return skb;
1603
1604interrupted:
1605 err = sock_intr_errno(timeo);
1606failure:
1607 *errcode = err;
1608 return NULL;
1609}
4cc7f68d 1610EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1611
4ec93edb 1612struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1613 int noblock, int *errcode)
1614{
1615 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1616}
2a91525c 1617EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1618
1619static void __lock_sock(struct sock *sk)
f39234d6
NK		 1620 __releases(&sk->sk_lock.slock)
1621 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1622{
1623 DEFINE_WAIT(wait);
1624
e71a4783 1625 for (;;) {
1da177e4
LT		 1626 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1627 TASK_UNINTERRUPTIBLE);
1628 spin_unlock_bh(&sk->sk_lock.slock);
1629 schedule();
1630 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1631 if (!sock_owned_by_user(sk))
1da177e4
LT		 1632 break;
1633 }
1634 finish_wait(&sk->sk_lock.wq, &wait);
1635}
1636
1637static void __release_sock(struct sock *sk)
f39234d6
NK		 1638 __releases(&sk->sk_lock.slock)
1639 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1640{
1641 struct sk_buff *skb = sk->sk_backlog.head;
1642
1643 do {
1644 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1645 bh_unlock_sock(sk);
1646
1647 do {
1648 struct sk_buff *next = skb->next;
1649
7fee226a 1650 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1651 skb->next = NULL;
c57943a1 1652 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1653
1654 /*
1655 * We are in process context here with softirqs
1656 * disabled, use cond_resched_softirq() to preempt.
1657 * This is safe to do because we've taken the backlog
1658 * queue private:
1659 */
1660 cond_resched_softirq();
1661
1662 skb = next;
1663 } while (skb != NULL);
1664
1665 bh_lock_sock(sk);
e71a4783 1666 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1667
1668 /*
1669 * Doing the zeroing here guarantee we can not loop forever
1670 * while a wild producer attempts to flood us.
1671 */
1672 sk->sk_backlog.len = 0;
1da177e4
LT		 1673}
1674
1675/**
1676 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1677 * @sk: sock to wait on
1678 * @timeo: for how long
1da177e4
LT		 1679 *
1680 * Now socket state including sk->sk_err is changed only under lock,
1681 * hence we may omit checks after joining wait queue.
1682 * We check receive queue before schedule() only as optimization;
1683 * it is very likely that release_sock() added new data.
1684 */
1685int sk_wait_data(struct sock *sk, long *timeo)
1686{
1687 int rc;
1688 DEFINE_WAIT(wait);
1689
aa395145 1690 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1691 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1692 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1693 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1694 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1695 return rc;
1696}
1da177e4
LT		 1697EXPORT_SYMBOL(sk_wait_data);
1698
3ab224be
HA		 1699/**
1700 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1701 * @sk: socket
1702 * @size: memory size to allocate
1703 * @kind: allocation type
1704 *
1705 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1706 * rmem allocation. This function assumes that protocols which have
1707 * memory_pressure use sk_wmem_queued as write buffer accounting.
1708 */
1709int __sk_mem_schedule(struct sock *sk, int size, int kind)
1710{
1711 struct proto *prot = sk->sk_prot;
1712 int amt = sk_mem_pages(size);
8d987e5c 1713 long allocated;
3ab224be
HA		 1714
1715 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
8d987e5c 1716 allocated = atomic_long_add_return(amt, prot->memory_allocated);
3ab224be
HA		 1717
1718 /* Under limit. */
1719 if (allocated <= prot->sysctl_mem[0]) {
1720 if (prot->memory_pressure && *prot->memory_pressure)
1721 *prot->memory_pressure = 0;
1722 return 1;
1723 }
1724
1725 /* Under pressure. */
1726 if (allocated > prot->sysctl_mem[1])
1727 if (prot->enter_memory_pressure)
5c52ba17 1728 prot->enter_memory_pressure(sk);
3ab224be
HA		 1729
1730 /* Over hard limit. */
1731 if (allocated > prot->sysctl_mem[2])
1732 goto suppress_allocation;
1733
1734 /* guarantee minimum buffer size under pressure */
1735 if (kind == SK_MEM_RECV) {
1736 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1737 return 1;
1738 } else { /* SK_MEM_SEND */
1739 if (sk->sk_type == SOCK_STREAM) {
1740 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1741 return 1;
1742 } else if (atomic_read(&sk->sk_wmem_alloc) <
1743 prot->sysctl_wmem[0])
1744 return 1;
1745 }
1746
1747 if (prot->memory_pressure) {
1748376b
ED		 1748 int alloc;
1749
1750 if (!*prot->memory_pressure)
1751 return 1;
1752 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1753 if (prot->sysctl_mem[2] > alloc *
3ab224be
HA		 1754 sk_mem_pages(sk->sk_wmem_queued +
1755 atomic_read(&sk->sk_rmem_alloc) +
1756 sk->sk_forward_alloc))
1757 return 1;
1758 }
1759
1760suppress_allocation:
1761
1762 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1763 sk_stream_moderate_sndbuf(sk);
1764
1765 /* Fail only if socket is _under_ its sndbuf.
1766 * In this case we cannot block, so that we have to fail.
1767 */
1768 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1769 return 1;
1770 }
1771
3847ce32
SM		 1772 trace_sock_exceed_buf_limit(sk, prot, allocated);
1773
3ab224be
HA		 1774 /* Alas. Undo changes. */
1775 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
8d987e5c 1776 atomic_long_sub(amt, prot->memory_allocated);
3ab224be
HA		 1777 return 0;
1778}
3ab224be
HA		 1779EXPORT_SYMBOL(__sk_mem_schedule);
1780
1781/**
1782 * __sk_reclaim - reclaim memory_allocated
1783 * @sk: socket
1784 */
1785void __sk_mem_reclaim(struct sock *sk)
1786{
1787 struct proto *prot = sk->sk_prot;
1788
8d987e5c 1789 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1790 prot->memory_allocated);
1791 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1792
1793 if (prot->memory_pressure && *prot->memory_pressure &&
8d987e5c 1794 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
3ab224be
HA		 1795 *prot->memory_pressure = 0;
1796}
3ab224be
HA		 1797EXPORT_SYMBOL(__sk_mem_reclaim);
1798
1799
1da177e4
LT		 1800/*
1801 * Set of default routines for initialising struct proto_ops when
1802 * the protocol does not support a particular function. In certain
1803 * cases where it makes no sense for a protocol to have a "do nothing"
1804 * function, some default processing is provided.
1805 */
1806
1807int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1808{
1809 return -EOPNOTSUPP;
1810}
2a91525c 1811EXPORT_SYMBOL(sock_no_bind);
1da177e4 1812
4ec93edb 1813int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1814 int len, int flags)
1815{
1816 return -EOPNOTSUPP;
1817}
2a91525c 1818EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1819
1820int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1821{
1822 return -EOPNOTSUPP;
1823}
2a91525c 1824EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1825
1826int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1827{
1828 return -EOPNOTSUPP;
1829}
2a91525c 1830EXPORT_SYMBOL(sock_no_accept);
1da177e4 1831
4ec93edb 1832int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1833 int *len, int peer)
1834{
1835 return -EOPNOTSUPP;
1836}
2a91525c 1837EXPORT_SYMBOL(sock_no_getname);
1da177e4 1838
2a91525c 1839unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1840{
1841 return 0;
1842}
2a91525c 1843EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1844
1845int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1846{
1847 return -EOPNOTSUPP;
1848}
2a91525c 1849EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1850
1851int sock_no_listen(struct socket *sock, int backlog)
1852{
1853 return -EOPNOTSUPP;
1854}
2a91525c 1855EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1856
1857int sock_no_shutdown(struct socket *sock, int how)
1858{
1859 return -EOPNOTSUPP;
1860}
2a91525c 1861EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1862
1863int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1864 char __user *optval, unsigned int optlen)
1da177e4
LT		 1865{
1866 return -EOPNOTSUPP;
1867}
2a91525c 1868EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1869
1870int sock_no_getsockopt(struct socket *sock, int level, int optname,
1871 char __user *optval, int __user *optlen)
1872{
1873 return -EOPNOTSUPP;
1874}
2a91525c 1875EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1876
1877int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1878 size_t len)
1879{
1880 return -EOPNOTSUPP;
1881}
2a91525c 1882EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1883
1884int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1885 size_t len, int flags)
1886{
1887 return -EOPNOTSUPP;
1888}
2a91525c 1889EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1890
1891int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1892{
1893 /* Mirror missing mmap method error code */
1894 return -ENODEV;
1895}
2a91525c 1896EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1897
1898ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1899{
1900 ssize_t res;
1901 struct msghdr msg = {.msg_flags = flags};
1902 struct kvec iov;
1903 char *kaddr = kmap(page);
1904 iov.iov_base = kaddr + offset;
1905 iov.iov_len = size;
1906 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1907 kunmap(page);
1908 return res;
1909}
2a91525c 1910EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1911
1912/*
1913 * Default Socket Callbacks
1914 */
1915
1916static void sock_def_wakeup(struct sock *sk)
1917{
43815482
ED		 1918 struct socket_wq *wq;
1919
1920 rcu_read_lock();
1921 wq = rcu_dereference(sk->sk_wq);
1922 if (wq_has_sleeper(wq))
1923 wake_up_interruptible_all(&wq->wait);
1924 rcu_read_unlock();
1da177e4
LT		 1925}
1926
1927static void sock_def_error_report(struct sock *sk)
1928{
43815482
ED		 1929 struct socket_wq *wq;
1930
1931 rcu_read_lock();
1932 wq = rcu_dereference(sk->sk_wq);
1933 if (wq_has_sleeper(wq))
1934 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1935 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1936 rcu_read_unlock();
1da177e4
LT		 1937}
1938
1939static void sock_def_readable(struct sock *sk, int len)
1940{
43815482
ED		 1941 struct socket_wq *wq;
1942
1943 rcu_read_lock();
1944 wq = rcu_dereference(sk->sk_wq);
1945 if (wq_has_sleeper(wq))
2c6607c6 1946 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1947 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1948 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1949 rcu_read_unlock();
1da177e4
LT		 1950}
1951
1952static void sock_def_write_space(struct sock *sk)
1953{
43815482
ED		 1954 struct socket_wq *wq;
1955
1956 rcu_read_lock();
1da177e4
LT		 1957
1958 /* Do not wake up a writer until he can make "significant"
1959 * progress. --DaveM
1960 */
e71a4783 1961 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 1962 wq = rcu_dereference(sk->sk_wq);
1963 if (wq_has_sleeper(wq))
1964 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 1965 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 1966
1967 /* Should agree with poll, otherwise some programs break */
1968 if (sock_writeable(sk))
8d8ad9d7 1969 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1970 }
1971
43815482 1972 rcu_read_unlock();
1da177e4
LT		 1973}
1974
1975static void sock_def_destruct(struct sock *sk)
1976{
a51482bd 1977 kfree(sk->sk_protinfo);
1da177e4
LT		 1978}
1979
1980void sk_send_sigurg(struct sock *sk)
1981{
1982 if (sk->sk_socket && sk->sk_socket->file)
1983 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 1984 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 1985}
2a91525c 1986EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 1987
1988void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1989 unsigned long expires)
1990{
1991 if (!mod_timer(timer, expires))
1992 sock_hold(sk);
1993}
1da177e4
LT		 1994EXPORT_SYMBOL(sk_reset_timer);
1995
1996void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1997{
1998 if (timer_pending(timer) && del_timer(timer))
1999 __sock_put(sk);
2000}
1da177e4
LT		 2001EXPORT_SYMBOL(sk_stop_timer);
2002
2003void sock_init_data(struct socket *sock, struct sock *sk)
2004{
2005 skb_queue_head_init(&sk->sk_receive_queue);
2006 skb_queue_head_init(&sk->sk_write_queue);
2007 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2008#ifdef CONFIG_NET_DMA
2009 skb_queue_head_init(&sk->sk_async_wait_queue);
2010#endif
1da177e4
LT		 2011
2012 sk->sk_send_head = NULL;
2013
2014 init_timer(&sk->sk_timer);
4ec93edb 2015
1da177e4
LT		 2016 sk->sk_allocation = GFP_KERNEL;
2017 sk->sk_rcvbuf = sysctl_rmem_default;
2018 sk->sk_sndbuf = sysctl_wmem_default;
2019 sk->sk_state = TCP_CLOSE;
972692e0 2020 sk_set_socket(sk, sock);
1da177e4
LT		 2021
2022 sock_set_flag(sk, SOCK_ZAPPED);
2023
e71a4783 2024 if (sock) {
1da177e4 2025 sk->sk_type = sock->type;
43815482 2026 sk->sk_wq = sock->wq;
1da177e4
LT		 2027 sock->sk = sk;
2028 } else
43815482 2029 sk->sk_wq = NULL;
1da177e4 2030
b6c6712a 2031 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2032 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2033 lockdep_set_class_and_name(&sk->sk_callback_lock,
2034 af_callback_keys + sk->sk_family,
2035 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2036
2037 sk->sk_state_change = sock_def_wakeup;
2038 sk->sk_data_ready = sock_def_readable;
2039 sk->sk_write_space = sock_def_write_space;
2040 sk->sk_error_report = sock_def_error_report;
2041 sk->sk_destruct = sock_def_destruct;
2042
2043 sk->sk_sndmsg_page = NULL;
2044 sk->sk_sndmsg_off = 0;
2045
109f6e39
EB		 2046 sk->sk_peer_pid = NULL;
2047 sk->sk_peer_cred = NULL;
1da177e4
LT		 2048 sk->sk_write_pending = 0;
2049 sk->sk_rcvlowat = 1;
2050 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2051 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2052
f37f0afb 2053 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2054
4dc6dc71
ED		 2055 /*
2056 * Before updating sk_refcnt, we must commit prior changes to memory
2057 * (Documentation/RCU/rculist_nulls.txt for details)
2058 */
2059 smp_wmb();
1da177e4 2060 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2061 atomic_set(&sk->sk_drops, 0);
1da177e4 2062}
2a91525c 2063EXPORT_SYMBOL(sock_init_data);
1da177e4 2064
b5606c2d 2065void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2066{
2067 might_sleep();
a5b5bb9a 2068 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2069 if (sk->sk_lock.owned)
1da177e4 2070 __lock_sock(sk);
d2e9117c 2071 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2072 spin_unlock(&sk->sk_lock.slock);
2073 /*
2074 * The sk_lock has mutex_lock() semantics here:
2075 */
fcc70d5f 2076 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2077 local_bh_enable();
1da177e4 2078}
fcc70d5f 2079EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2080
b5606c2d 2081void release_sock(struct sock *sk)
1da177e4 2082{
a5b5bb9a
IM		 2083 /*
2084 * The sk_lock has mutex_unlock() semantics:
2085 */
2086 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2087
2088 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2089 if (sk->sk_backlog.tail)
2090 __release_sock(sk);
d2e9117c 2091 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2092 if (waitqueue_active(&sk->sk_lock.wq))
2093 wake_up(&sk->sk_lock.wq);
2094 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2095}
2096EXPORT_SYMBOL(release_sock);
2097
8a74ad60
ED		 2098/**
2099 * lock_sock_fast - fast version of lock_sock
2100 * @sk: socket
2101 *
2102 * This version should be used for very small section, where process wont block
2103 * return false if fast path is taken
2104 * sk_lock.slock locked, owned = 0, BH disabled
2105 * return true if slow path is taken
2106 * sk_lock.slock unlocked, owned = 1, BH enabled
2107 */
2108bool lock_sock_fast(struct sock *sk)
2109{
2110 might_sleep();
2111 spin_lock_bh(&sk->sk_lock.slock);
2112
2113 if (!sk->sk_lock.owned)
2114 /*
2115 * Note : We must disable BH
2116 */
2117 return false;
2118
2119 __lock_sock(sk);
2120 sk->sk_lock.owned = 1;
2121 spin_unlock(&sk->sk_lock.slock);
2122 /*
2123 * The sk_lock has mutex_lock() semantics here:
2124 */
2125 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2126 local_bh_enable();
2127 return true;
2128}
2129EXPORT_SYMBOL(lock_sock_fast);
2130
1da177e4 2131int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2132{
b7aa0bf7 2133 struct timeval tv;
1da177e4 2134 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2135 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2136 tv = ktime_to_timeval(sk->sk_stamp);
2137 if (tv.tv_sec == -1)
1da177e4 2138 return -ENOENT;
b7aa0bf7
ED		 2139 if (tv.tv_sec == 0) {
2140 sk->sk_stamp = ktime_get_real();
2141 tv = ktime_to_timeval(sk->sk_stamp);
2142 }
2143 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2144}
1da177e4
LT		 2145EXPORT_SYMBOL(sock_get_timestamp);
2146
ae40eb1e
ED		 2147int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2148{
2149 struct timespec ts;
2150 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2151 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2152 ts = ktime_to_timespec(sk->sk_stamp);
2153 if (ts.tv_sec == -1)
2154 return -ENOENT;
2155 if (ts.tv_sec == 0) {
2156 sk->sk_stamp = ktime_get_real();
2157 ts = ktime_to_timespec(sk->sk_stamp);
2158 }
2159 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2160}
2161EXPORT_SYMBOL(sock_get_timestampns);
2162
20d49473 2163void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2164{
20d49473 2165 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2166 unsigned long previous_flags = sk->sk_flags;
2167
20d49473
PO		 2168 sock_set_flag(sk, flag);
2169 /*
2170 * we just set one of the two flags which require net
2171 * time stamping, but time stamping might have been on
2172 * already because of the other one
2173 */
08e29af3 2174 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2175 net_enable_timestamp();
1da177e4
LT		 2176 }
2177}
1da177e4
LT		 2178
2179/*
2180 * Get a socket option on an socket.
2181 *
2182 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2183 * asynchronous errors should be reported by getsockopt. We assume
2184 * this means if you specify SO_ERROR (otherwise whats the point of it).
2185 */
2186int sock_common_getsockopt(struct socket *sock, int level, int optname,
2187 char __user *optval, int __user *optlen)
2188{
2189 struct sock *sk = sock->sk;
2190
2191 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2192}
1da177e4
LT		 2193EXPORT_SYMBOL(sock_common_getsockopt);
2194
3fdadf7d 2195#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2196int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2197 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2198{
2199 struct sock *sk = sock->sk;
2200
1e51f951 2201 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2202 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2203 optval, optlen);
3fdadf7d
DM		 2204 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2205}
2206EXPORT_SYMBOL(compat_sock_common_getsockopt);
2207#endif
2208
1da177e4
LT		 2209int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2210 struct msghdr *msg, size_t size, int flags)
2211{
2212 struct sock *sk = sock->sk;
2213 int addr_len = 0;
2214 int err;
2215
2216 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2217 flags & ~MSG_DONTWAIT, &addr_len);
2218 if (err >= 0)
2219 msg->msg_namelen = addr_len;
2220 return err;
2221}
1da177e4
LT		 2222EXPORT_SYMBOL(sock_common_recvmsg);
2223
2224/*
2225 * Set socket options on an inet socket.
2226 */
2227int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2228 char __user *optval, unsigned int optlen)
1da177e4
LT		 2229{
2230 struct sock *sk = sock->sk;
2231
2232 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2233}
1da177e4
LT		 2234EXPORT_SYMBOL(sock_common_setsockopt);
2235
3fdadf7d 2236#ifdef CONFIG_COMPAT
543d9cfe 2237int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2238 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2239{
2240 struct sock *sk = sock->sk;
2241
543d9cfe
ACM		 2242 if (sk->sk_prot->compat_setsockopt != NULL)
2243 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2244 optval, optlen);
3fdadf7d
DM		 2245 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2246}
2247EXPORT_SYMBOL(compat_sock_common_setsockopt);
2248#endif
2249
1da177e4
LT		 2250void sk_common_release(struct sock *sk)
2251{
2252 if (sk->sk_prot->destroy)
2253 sk->sk_prot->destroy(sk);
2254
2255 /*
2256 * Observation: when sock_common_release is called, processes have
2257 * no access to socket. But net still has.
2258 * Step one, detach it from networking:
2259 *
2260 * A. Remove from hash tables.
2261 */
2262
2263 sk->sk_prot->unhash(sk);
2264
2265 /*
2266 * In this point socket cannot receive new packets, but it is possible
2267 * that some packets are in flight because some CPU runs receiver and
2268 * did hash table lookup before we unhashed socket. They will achieve
2269 * receive queue and will be purged by socket destructor.
2270 *
2271 * Also we still have packets pending on receive queue and probably,
2272 * our own packets waiting in device queues. sock_destroy will drain
2273 * receive queue, but transmitted packets will delay socket destruction
2274 * until the last reference will be released.
2275 */
2276
2277 sock_orphan(sk);
2278
2279 xfrm_sk_free_policy(sk);
2280
e6848976 2281 sk_refcnt_debug_release(sk);
1da177e4
LT		 2282 sock_put(sk);
2283}
1da177e4
LT		 2284EXPORT_SYMBOL(sk_common_release);
2285
2286static DEFINE_RWLOCK(proto_list_lock);
2287static LIST_HEAD(proto_list);
2288
13ff3d6f
PE		 2289#ifdef CONFIG_PROC_FS
2290#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2291struct prot_inuse {
2292 int val[PROTO_INUSE_NR];
2293};
13ff3d6f
PE		 2294
2295static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2296
2297#ifdef CONFIG_NET_NS
2298void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2299{
d6d9ca0f 2300 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2301}
2302EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2303
2304int sock_prot_inuse_get(struct net *net, struct proto *prot)
2305{
2306 int cpu, idx = prot->inuse_idx;
2307 int res = 0;
2308
2309 for_each_possible_cpu(cpu)
2310 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2311
2312 return res >= 0 ? res : 0;
2313}
2314EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2315
2c8c1e72 2316static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2317{
2318 net->core.inuse = alloc_percpu(struct prot_inuse);
2319 return net->core.inuse ? 0 : -ENOMEM;
2320}
2321
2c8c1e72 2322static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2323{
2324 free_percpu(net->core.inuse);
2325}
2326
2327static struct pernet_operations net_inuse_ops = {
2328 .init = sock_inuse_init_net,
2329 .exit = sock_inuse_exit_net,
2330};
2331
2332static __init int net_inuse_init(void)
2333{
2334 if (register_pernet_subsys(&net_inuse_ops))
2335 panic("Cannot initialize net inuse counters");
2336
2337 return 0;
2338}
2339
2340core_initcall(net_inuse_init);
2341#else
1338d466
PE		 2342static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2343
c29a0bc4 2344void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2345{
d6d9ca0f 2346 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2347}
2348EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2349
c29a0bc4 2350int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2351{
2352 int cpu, idx = prot->inuse_idx;
2353 int res = 0;
2354
2355 for_each_possible_cpu(cpu)
2356 res += per_cpu(prot_inuse, cpu).val[idx];
2357
2358 return res >= 0 ? res : 0;
2359}
2360EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2361#endif
13ff3d6f
PE		 2362
2363static void assign_proto_idx(struct proto *prot)
2364{
2365 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2366
2367 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2368 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2369 return;
2370 }
2371
2372 set_bit(prot->inuse_idx, proto_inuse_idx);
2373}
2374
2375static void release_proto_idx(struct proto *prot)
2376{
2377 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2378 clear_bit(prot->inuse_idx, proto_inuse_idx);
2379}
2380#else
2381static inline void assign_proto_idx(struct proto *prot)
2382{
2383}
2384
2385static inline void release_proto_idx(struct proto *prot)
2386{
2387}
2388#endif
2389
b733c007
PE		 2390int proto_register(struct proto *prot, int alloc_slab)
2391{
1da177e4
LT		 2392 if (alloc_slab) {
2393 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2394 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2395 NULL);
1da177e4
LT		 2396
2397 if (prot->slab == NULL) {
2398 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2399 prot->name);
60e7663d 2400 goto out;
1da177e4 2401 }
2e6599cb
ACM		 2402
2403 if (prot->rsk_prot != NULL) {
faf23422 2404 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2405 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2406 goto out_free_sock_slab;
2407
7e56b5d6 2408 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2409 prot->rsk_prot->obj_size, 0,
20c2df83 2410 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2411
2412 if (prot->rsk_prot->slab == NULL) {
2413 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2414 prot->name);
2415 goto out_free_request_sock_slab_name;
2416 }
2417 }
8feaf0c0 2418
6d6ee43e 2419 if (prot->twsk_prot != NULL) {
faf23422 2420 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2421
7e56b5d6 2422 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2423 goto out_free_request_sock_slab;
2424
6d6ee43e 2425 prot->twsk_prot->twsk_slab =
7e56b5d6 2426 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2427 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2428 0,
2429 SLAB_HWCACHE_ALIGN |
2430 prot->slab_flags,
20c2df83 2431 NULL);
6d6ee43e 2432 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2433 goto out_free_timewait_sock_slab_name;
2434 }
1da177e4
LT		 2435 }
2436
2a278051 2437 write_lock(&proto_list_lock);
1da177e4 2438 list_add(&prot->node, &proto_list);
13ff3d6f 2439 assign_proto_idx(prot);
1da177e4 2440 write_unlock(&proto_list_lock);
b733c007
PE		 2441 return 0;
2442
8feaf0c0 2443out_free_timewait_sock_slab_name:
7e56b5d6 2444 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2445out_free_request_sock_slab:
2446 if (prot->rsk_prot && prot->rsk_prot->slab) {
2447 kmem_cache_destroy(prot->rsk_prot->slab);
2448 prot->rsk_prot->slab = NULL;
2449 }
2e6599cb 2450out_free_request_sock_slab_name:
72150e9b
DC		 2451 if (prot->rsk_prot)
2452 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2453out_free_sock_slab:
2454 kmem_cache_destroy(prot->slab);
2455 prot->slab = NULL;
b733c007
PE		 2456out:
2457 return -ENOBUFS;
1da177e4 2458}
1da177e4
LT		 2459EXPORT_SYMBOL(proto_register);
2460
2461void proto_unregister(struct proto *prot)
2462{
2463 write_lock(&proto_list_lock);
13ff3d6f 2464 release_proto_idx(prot);
0a3f4358
PM		 2465 list_del(&prot->node);
2466 write_unlock(&proto_list_lock);
1da177e4
LT		 2467
2468 if (prot->slab != NULL) {
2469 kmem_cache_destroy(prot->slab);
2470 prot->slab = NULL;
2471 }
2472
2e6599cb 2473 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2474 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2475 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2476 prot->rsk_prot->slab = NULL;
2477 }
2478
6d6ee43e 2479 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2480 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2481 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2482 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2483 }
1da177e4 2484}
1da177e4
LT		 2485EXPORT_SYMBOL(proto_unregister);
2486
2487#ifdef CONFIG_PROC_FS
1da177e4 2488static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2489 __acquires(proto_list_lock)
1da177e4
LT		 2490{
2491 read_lock(&proto_list_lock);
60f0438a 2492 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2493}
2494
2495static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2496{
60f0438a 2497 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2498}
2499
2500static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2501 __releases(proto_list_lock)
1da177e4
LT		 2502{
2503 read_unlock(&proto_list_lock);
2504}
2505
2506static char proto_method_implemented(const void *method)
2507{
2508 return method == NULL ? 'n' : 'y';
2509}
2510
2511static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2512{
8d987e5c 2513 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2514 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2515 proto->name,
2516 proto->obj_size,
14e943db 2517 sock_prot_inuse_get(seq_file_net(seq), proto),
8d987e5c 2518 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
1da177e4
LT		 2519 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2520 proto->max_header,
2521 proto->slab == NULL ? "no" : "yes",
2522 module_name(proto->owner),
2523 proto_method_implemented(proto->close),
2524 proto_method_implemented(proto->connect),
2525 proto_method_implemented(proto->disconnect),
2526 proto_method_implemented(proto->accept),
2527 proto_method_implemented(proto->ioctl),
2528 proto_method_implemented(proto->init),
2529 proto_method_implemented(proto->destroy),
2530 proto_method_implemented(proto->shutdown),
2531 proto_method_implemented(proto->setsockopt),
2532 proto_method_implemented(proto->getsockopt),
2533 proto_method_implemented(proto->sendmsg),
2534 proto_method_implemented(proto->recvmsg),
2535 proto_method_implemented(proto->sendpage),
2536 proto_method_implemented(proto->bind),
2537 proto_method_implemented(proto->backlog_rcv),
2538 proto_method_implemented(proto->hash),
2539 proto_method_implemented(proto->unhash),
2540 proto_method_implemented(proto->get_port),
2541 proto_method_implemented(proto->enter_memory_pressure));
2542}
2543
2544static int proto_seq_show(struct seq_file *seq, void *v)
2545{
60f0438a 2546 if (v == &proto_list)
1da177e4
LT		 2547 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2548 "protocol",
2549 "size",
2550 "sockets",
2551 "memory",
2552 "press",
2553 "maxhdr",
2554 "slab",
2555 "module",
2556 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2557 else
60f0438a 2558 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2559 return 0;
2560}
2561
f690808e 2562static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2563 .start = proto_seq_start,
2564 .next = proto_seq_next,
2565 .stop = proto_seq_stop,
2566 .show = proto_seq_show,
2567};
2568
2569static int proto_seq_open(struct inode *inode, struct file *file)
2570{
14e943db
ED		 2571 return seq_open_net(inode, file, &proto_seq_ops,
2572 sizeof(struct seq_net_private));
1da177e4
LT		 2573}
2574
9a32144e 2575static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2576 .owner = THIS_MODULE,
2577 .open = proto_seq_open,
2578 .read = seq_read,
2579 .llseek = seq_lseek,
14e943db
ED		 2580 .release = seq_release_net,
2581};
2582
2583static __net_init int proto_init_net(struct net *net)
2584{
2585 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2586 return -ENOMEM;
2587
2588 return 0;
2589}
2590
2591static __net_exit void proto_exit_net(struct net *net)
2592{
2593 proc_net_remove(net, "protocols");
2594}
2595
2596
2597static __net_initdata struct pernet_operations proto_net_ops = {
2598 .init = proto_init_net,
2599 .exit = proto_exit_net,
1da177e4
LT		 2600};
2601
2602static int __init proto_init(void)
2603{
14e943db 2604 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2605}
2606
2607subsys_initcall(proto_init);
2608
2609#endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.776405 seconds and 5 git commands to generate.