projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tg3: Fix advertisement handling
[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>
1da177e4
LT		 128
129#include <linux/filter.h>
130
3847ce32
SM		 131#include <trace/events/sock.h>
132
1da177e4
LT		 133#ifdef CONFIG_INET
134#include <net/tcp.h>
135#endif
136
da21f24d
IM		 137/*
138 * Each address family might have different locking rules, so we have
139 * one slock key per address family:
140 */
a5b5bb9a
IM		 141static struct lock_class_key af_family_keys[AF_MAX];
142static struct lock_class_key af_family_slock_keys[AF_MAX];
143
a5b5bb9a
IM		 144/*
145 * Make lock validator output more readable. (we pre-construct these
146 * strings build-time, so that runtime initialization of socket
147 * locks is fast):
148 */
36cbd3dc 149static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 150 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
151 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
152 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
153 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
154 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
155 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
156 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 157 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 158 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 159 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 160 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 161 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 162 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
c7fe3b52 163 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
a5b5bb9a 164};
36cbd3dc 165static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 166 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
167 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
168 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
169 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
170 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
171 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
172 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 173 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 174 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 175 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 176 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 177 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 178 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
c7fe3b52 179 "slock-AF_NFC" , "slock-AF_MAX"
a5b5bb9a 180};
36cbd3dc 181static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 182 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
183 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
184 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
185 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
186 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
187 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
188 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 189 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 190 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 191 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 192 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 193 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 194 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
c7fe3b52 195 "clock-AF_NFC" , "clock-AF_MAX"
443aef0e 196};
da21f24d
IM		 197
198/*
199 * sk_callback_lock locking rules are per-address-family,
200 * so split the lock classes by using a per-AF key:
201 */
202static struct lock_class_key af_callback_keys[AF_MAX];
203
1da177e4
LT		 204/* Take into consideration the size of the struct sk_buff overhead in the
205 * determination of these values, since that is non-constant across
206 * platforms. This makes socket queueing behavior and performance
207 * not depend upon such differences.
208 */
209#define _SK_MEM_PACKETS 256
87fb4b7b 210#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 211#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
212#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
213
214/* Run time adjustable parameters. */
ab32ea5d
BH		 215__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
216__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
217__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
218__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 219
25985edc 220/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 221int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 222EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 223
f8451725
HX		 224#if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
225int net_cls_subsys_id = -1;
226EXPORT_SYMBOL_GPL(net_cls_subsys_id);
227#endif
228
1da177e4
LT		 229static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
230{
231 struct timeval tv;
232
233 if (optlen < sizeof(tv))
234 return -EINVAL;
235 if (copy_from_user(&tv, optval, sizeof(tv)))
236 return -EFAULT;
ba78073e
VA		 237 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
238 return -EDOM;
1da177e4 239
ba78073e 240 if (tv.tv_sec < 0) {
6f11df83
AM		 241 static int warned __read_mostly;
242
ba78073e 243 *timeo_p = 0;
50aab54f 244 if (warned < 10 && net_ratelimit()) {
ba78073e
VA		 245 warned++;
246 printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
247 "tries to set negative timeout\n",
ba25f9dc 248 current->comm, task_pid_nr(current));
50aab54f 249 }
ba78073e
VA		 250 return 0;
251 }
1da177e4
LT		 252 *timeo_p = MAX_SCHEDULE_TIMEOUT;
253 if (tv.tv_sec == 0 && tv.tv_usec == 0)
254 return 0;
255 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
256 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
257 return 0;
258}
259
260static void sock_warn_obsolete_bsdism(const char *name)
261{
262 static int warned;
263 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 264 if (strcmp(warncomm, current->comm) && warned < 5) {
265 strcpy(warncomm, current->comm);
1da177e4
LT		 266 printk(KERN_WARNING "process `%s' is using obsolete "
267 "%s SO_BSDCOMPAT\n", warncomm, name);
268 warned++;
269 }
270}
271
20d49473 272static void sock_disable_timestamp(struct sock *sk, int flag)
4ec93edb 273{
20d49473
PO		 274 if (sock_flag(sk, flag)) {
275 sock_reset_flag(sk, flag);
276 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
277 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
278 net_disable_timestamp();
279 }
1da177e4
LT		 280 }
281}
282
283
f0088a50
DV		 284int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
285{
766e9037 286 int err;
f0088a50 287 int skb_len;
3b885787
NH		 288 unsigned long flags;
289 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 290
9ee6b7f1 291 /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
f0088a50
DV		 292 number of warnings when compiling with -W --ANK
293 */
294 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
295 (unsigned)sk->sk_rcvbuf) {
766e9037 296 atomic_inc(&sk->sk_drops);
3847ce32 297 trace_sock_rcvqueue_full(sk, skb);
766e9037 298 return -ENOMEM;
f0088a50
DV		 299 }
300
fda9ef5d 301 err = sk_filter(sk, skb);
f0088a50 302 if (err)
766e9037 303 return err;
f0088a50 304
3ab224be 305 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 306 atomic_inc(&sk->sk_drops);
307 return -ENOBUFS;
3ab224be
HA		 308 }
309
f0088a50
DV		 310 skb->dev = NULL;
311 skb_set_owner_r(skb, sk);
49ad9599 312
f0088a50
DV		 313 /* Cache the SKB length before we tack it onto the receive
314 * queue. Once it is added it no longer belongs to us and
315 * may be freed by other threads of control pulling packets
316 * from the queue.
317 */
318 skb_len = skb->len;
319
7fee226a
ED		 320 /* we escape from rcu protected region, make sure we dont leak
321 * a norefcounted dst
322 */
323 skb_dst_force(skb);
324
3b885787
NH		 325 spin_lock_irqsave(&list->lock, flags);
326 skb->dropcount = atomic_read(&sk->sk_drops);
327 __skb_queue_tail(list, skb);
328 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 329
330 if (!sock_flag(sk, SOCK_DEAD))
331 sk->sk_data_ready(sk, skb_len);
766e9037 332 return 0;
f0088a50
DV		 333}
334EXPORT_SYMBOL(sock_queue_rcv_skb);
335
58a5a7b9 336int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 337{
338 int rc = NET_RX_SUCCESS;
339
fda9ef5d 340 if (sk_filter(sk, skb))
f0088a50
DV		 341 goto discard_and_relse;
342
343 skb->dev = NULL;
344
c377411f
ED		 345 if (sk_rcvqueues_full(sk, skb)) {
346 atomic_inc(&sk->sk_drops);
347 goto discard_and_relse;
348 }
58a5a7b9
ACM		 349 if (nested)
350 bh_lock_sock_nested(sk);
351 else
352 bh_lock_sock(sk);
a5b5bb9a
IM		 353 if (!sock_owned_by_user(sk)) {
354 /*
355 * trylock + unlock semantics:
356 */
357 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
358
c57943a1 359 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 360
361 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
a3a858ff 362 } else if (sk_add_backlog(sk, skb)) {
8eae939f
ZY		 363 bh_unlock_sock(sk);
364 atomic_inc(&sk->sk_drops);
365 goto discard_and_relse;
366 }
367
f0088a50
DV		 368 bh_unlock_sock(sk);
369out:
370 sock_put(sk);
371 return rc;
372discard_and_relse:
373 kfree_skb(skb);
374 goto out;
375}
376EXPORT_SYMBOL(sk_receive_skb);
377
ea94ff3b
KK		 378void sk_reset_txq(struct sock *sk)
379{
380 sk_tx_queue_clear(sk);
381}
382EXPORT_SYMBOL(sk_reset_txq);
383
f0088a50
DV		 384struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
385{
b6c6712a 386 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 387
388 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 389 sk_tx_queue_clear(sk);
a9b3cd7f 390 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 391 dst_release(dst);
392 return NULL;
393 }
394
395 return dst;
396}
397EXPORT_SYMBOL(__sk_dst_check);
398
399struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
400{
401 struct dst_entry *dst = sk_dst_get(sk);
402
403 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
404 sk_dst_reset(sk);
405 dst_release(dst);
406 return NULL;
407 }
408
409 return dst;
410}
411EXPORT_SYMBOL(sk_dst_check);
412
4878809f
DM		 413static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
414{
415 int ret = -ENOPROTOOPT;
416#ifdef CONFIG_NETDEVICES
3b1e0a65 417 struct net *net = sock_net(sk);
4878809f
DM		 418 char devname[IFNAMSIZ];
419 int index;
420
421 /* Sorry... */
422 ret = -EPERM;
423 if (!capable(CAP_NET_RAW))
424 goto out;
425
426 ret = -EINVAL;
427 if (optlen < 0)
428 goto out;
429
430 /* Bind this socket to a particular device like "eth0",
431 * as specified in the passed interface name. If the
432 * name is "" or the option length is zero the socket
433 * is not bound.
434 */
435 if (optlen > IFNAMSIZ - 1)
436 optlen = IFNAMSIZ - 1;
437 memset(devname, 0, sizeof(devname));
438
439 ret = -EFAULT;
440 if (copy_from_user(devname, optval, optlen))
441 goto out;
442
000ba2e4
DM		 443 index = 0;
444 if (devname[0] != '\0') {
bf8e56bf 445 struct net_device *dev;
4878809f 446
bf8e56bf
ED		 447 rcu_read_lock();
448 dev = dev_get_by_name_rcu(net, devname);
449 if (dev)
450 index = dev->ifindex;
451 rcu_read_unlock();
4878809f
DM		 452 ret = -ENODEV;
453 if (!dev)
454 goto out;
4878809f
DM		 455 }
456
457 lock_sock(sk);
458 sk->sk_bound_dev_if = index;
459 sk_dst_reset(sk);
460 release_sock(sk);
461
462 ret = 0;
463
464out:
465#endif
466
467 return ret;
468}
469
c0ef877b
PE		 470static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
471{
472 if (valbool)
473 sock_set_flag(sk, bit);
474 else
475 sock_reset_flag(sk, bit);
476}
477
1da177e4
LT		 478/*
479 * This is meant for all protocols to use and covers goings on
480 * at the socket level. Everything here is generic.
481 */
482
483int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 484 char __user *optval, unsigned int optlen)
1da177e4 485{
2a91525c 486 struct sock *sk = sock->sk;
1da177e4
LT		 487 int val;
488 int valbool;
489 struct linger ling;
490 int ret = 0;
4ec93edb 491
1da177e4
LT		 492 /*
493 * Options without arguments
494 */
495
4878809f
DM		 496 if (optname == SO_BINDTODEVICE)
497 return sock_bindtodevice(sk, optval, optlen);
498
e71a4783
SH		 499 if (optlen < sizeof(int))
500 return -EINVAL;
4ec93edb 501
1da177e4
LT		 502 if (get_user(val, (int __user *)optval))
503 return -EFAULT;
4ec93edb 504
2a91525c 505 valbool = val ? 1 : 0;
1da177e4
LT		 506
507 lock_sock(sk);
508
2a91525c 509 switch (optname) {
e71a4783 510 case SO_DEBUG:
2a91525c 511 if (val && !capable(CAP_NET_ADMIN))
e71a4783 512 ret = -EACCES;
2a91525c 513 else
c0ef877b 514 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 515 break;
516 case SO_REUSEADDR:
517 sk->sk_reuse = valbool;
518 break;
519 case SO_TYPE:
49c794e9 520 case SO_PROTOCOL:
0d6038ee 521 case SO_DOMAIN:
e71a4783
SH		 522 case SO_ERROR:
523 ret = -ENOPROTOOPT;
524 break;
525 case SO_DONTROUTE:
c0ef877b 526 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 527 break;
528 case SO_BROADCAST:
529 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
530 break;
531 case SO_SNDBUF:
532 /* Don't error on this BSD doesn't and if you think
533 about it this is right. Otherwise apps have to
534 play 'guess the biggest size' games. RCVBUF/SNDBUF
535 are treated in BSD as hints */
536
537 if (val > sysctl_wmem_max)
538 val = sysctl_wmem_max;
b0573dea 539set_sndbuf:
e71a4783
SH		 540 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
541 if ((val * 2) < SOCK_MIN_SNDBUF)
542 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
543 else
544 sk->sk_sndbuf = val * 2;
1da177e4 545
e71a4783
SH		 546 /*
547 * Wake up sending tasks if we
548 * upped the value.
549 */
550 sk->sk_write_space(sk);
551 break;
1da177e4 552
e71a4783
SH		 553 case SO_SNDBUFFORCE:
554 if (!capable(CAP_NET_ADMIN)) {
555 ret = -EPERM;
556 break;
557 }
558 goto set_sndbuf;
b0573dea 559
e71a4783
SH		 560 case SO_RCVBUF:
561 /* Don't error on this BSD doesn't and if you think
562 about it this is right. Otherwise apps have to
563 play 'guess the biggest size' games. RCVBUF/SNDBUF
564 are treated in BSD as hints */
4ec93edb 565
e71a4783
SH		 566 if (val > sysctl_rmem_max)
567 val = sysctl_rmem_max;
b0573dea 568set_rcvbuf:
e71a4783
SH		 569 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
570 /*
571 * We double it on the way in to account for
572 * "struct sk_buff" etc. overhead. Applications
573 * assume that the SO_RCVBUF setting they make will
574 * allow that much actual data to be received on that
575 * socket.
576 *
577 * Applications are unaware that "struct sk_buff" and
578 * other overheads allocate from the receive buffer
579 * during socket buffer allocation.
580 *
581 * And after considering the possible alternatives,
582 * returning the value we actually used in getsockopt
583 * is the most desirable behavior.
584 */
585 if ((val * 2) < SOCK_MIN_RCVBUF)
586 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
587 else
588 sk->sk_rcvbuf = val * 2;
589 break;
590
591 case SO_RCVBUFFORCE:
592 if (!capable(CAP_NET_ADMIN)) {
593 ret = -EPERM;
1da177e4 594 break;
e71a4783
SH		 595 }
596 goto set_rcvbuf;
1da177e4 597
e71a4783 598 case SO_KEEPALIVE:
1da177e4 599#ifdef CONFIG_INET
e71a4783
SH		 600 if (sk->sk_protocol == IPPROTO_TCP)
601 tcp_set_keepalive(sk, valbool);
1da177e4 602#endif
e71a4783
SH		 603 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
604 break;
605
606 case SO_OOBINLINE:
607 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
608 break;
609
610 case SO_NO_CHECK:
611 sk->sk_no_check = valbool;
612 break;
613
614 case SO_PRIORITY:
615 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
616 sk->sk_priority = val;
617 else
618 ret = -EPERM;
619 break;
620
621 case SO_LINGER:
622 if (optlen < sizeof(ling)) {
623 ret = -EINVAL; /* 1003.1g */
1da177e4 624 break;
e71a4783 625 }
2a91525c 626 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 627 ret = -EFAULT;
1da177e4 628 break;
e71a4783
SH		 629 }
630 if (!ling.l_onoff)
631 sock_reset_flag(sk, SOCK_LINGER);
632 else {
1da177e4 633#if (BITS_PER_LONG == 32)
e71a4783
SH		 634 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
635 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 636 else
e71a4783
SH		 637#endif
638 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
639 sock_set_flag(sk, SOCK_LINGER);
640 }
641 break;
642
643 case SO_BSDCOMPAT:
644 sock_warn_obsolete_bsdism("setsockopt");
645 break;
646
647 case SO_PASSCRED:
648 if (valbool)
649 set_bit(SOCK_PASSCRED, &sock->flags);
650 else
651 clear_bit(SOCK_PASSCRED, &sock->flags);
652 break;
653
654 case SO_TIMESTAMP:
92f37fd2 655 case SO_TIMESTAMPNS:
e71a4783 656 if (valbool) {
92f37fd2
ED		 657 if (optname == SO_TIMESTAMP)
658 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
659 else
660 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 661 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 662 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 663 } else {
e71a4783 664 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 665 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
666 }
e71a4783
SH		 667 break;
668
20d49473
PO		 669 case SO_TIMESTAMPING:
670 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 671 ret = -EINVAL;
20d49473
PO		 672 break;
673 }
674 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
675 val & SOF_TIMESTAMPING_TX_HARDWARE);
676 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
677 val & SOF_TIMESTAMPING_TX_SOFTWARE);
678 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
679 val & SOF_TIMESTAMPING_RX_HARDWARE);
680 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
681 sock_enable_timestamp(sk,
682 SOCK_TIMESTAMPING_RX_SOFTWARE);
683 else
684 sock_disable_timestamp(sk,
685 SOCK_TIMESTAMPING_RX_SOFTWARE);
686 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
687 val & SOF_TIMESTAMPING_SOFTWARE);
688 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
689 val & SOF_TIMESTAMPING_SYS_HARDWARE);
690 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
691 val & SOF_TIMESTAMPING_RAW_HARDWARE);
692 break;
693
e71a4783
SH		 694 case SO_RCVLOWAT:
695 if (val < 0)
696 val = INT_MAX;
697 sk->sk_rcvlowat = val ? : 1;
698 break;
699
700 case SO_RCVTIMEO:
701 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
702 break;
703
704 case SO_SNDTIMEO:
705 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
706 break;
1da177e4 707
e71a4783
SH		 708 case SO_ATTACH_FILTER:
709 ret = -EINVAL;
710 if (optlen == sizeof(struct sock_fprog)) {
711 struct sock_fprog fprog;
1da177e4 712
e71a4783
SH		 713 ret = -EFAULT;
714 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 715 break;
e71a4783
SH		 716
717 ret = sk_attach_filter(&fprog, sk);
718 }
719 break;
720
721 case SO_DETACH_FILTER:
55b33325 722 ret = sk_detach_filter(sk);
e71a4783 723 break;
1da177e4 724
e71a4783
SH		 725 case SO_PASSSEC:
726 if (valbool)
727 set_bit(SOCK_PASSSEC, &sock->flags);
728 else
729 clear_bit(SOCK_PASSSEC, &sock->flags);
730 break;
4a19ec58
LAT		 731 case SO_MARK:
732 if (!capable(CAP_NET_ADMIN))
733 ret = -EPERM;
2a91525c 734 else
4a19ec58 735 sk->sk_mark = val;
4a19ec58 736 break;
877ce7c1 737
1da177e4
LT		 738 /* We implement the SO_SNDLOWAT etc to
739 not be settable (1003.1g 5.3) */
3b885787 740 case SO_RXQ_OVFL:
8083f0fc 741 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 742 break;
6e3e939f
JB		 743
744 case SO_WIFI_STATUS:
745 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
746 break;
747
e71a4783
SH		 748 default:
749 ret = -ENOPROTOOPT;
750 break;
4ec93edb 751 }
1da177e4
LT		 752 release_sock(sk);
753 return ret;
754}
2a91525c 755EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 756
757
3f551f94
EB		 758void cred_to_ucred(struct pid *pid, const struct cred *cred,
759 struct ucred *ucred)
760{
761 ucred->pid = pid_vnr(pid);
762 ucred->uid = ucred->gid = -1;
763 if (cred) {
764 struct user_namespace *current_ns = current_user_ns();
765
766 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
767 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
768 }
769}
3924773a 770EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 771
1da177e4
LT		 772int sock_getsockopt(struct socket *sock, int level, int optname,
773 char __user *optval, int __user *optlen)
774{
775 struct sock *sk = sock->sk;
4ec93edb 776
e71a4783 777 union {
4ec93edb
YH		 778 int val;
779 struct linger ling;
1da177e4
LT		 780 struct timeval tm;
781 } v;
4ec93edb 782
4d0392be 783 int lv = sizeof(int);
1da177e4 784 int len;
4ec93edb 785
e71a4783 786 if (get_user(len, optlen))
4ec93edb 787 return -EFAULT;
e71a4783 788 if (len < 0)
1da177e4 789 return -EINVAL;
4ec93edb 790
50fee1de 791 memset(&v, 0, sizeof(v));
df0bca04 792
2a91525c 793 switch (optname) {
e71a4783
SH		 794 case SO_DEBUG:
795 v.val = sock_flag(sk, SOCK_DBG);
796 break;
797
798 case SO_DONTROUTE:
799 v.val = sock_flag(sk, SOCK_LOCALROUTE);
800 break;
801
802 case SO_BROADCAST:
803 v.val = !!sock_flag(sk, SOCK_BROADCAST);
804 break;
805
806 case SO_SNDBUF:
807 v.val = sk->sk_sndbuf;
808 break;
809
810 case SO_RCVBUF:
811 v.val = sk->sk_rcvbuf;
812 break;
813
814 case SO_REUSEADDR:
815 v.val = sk->sk_reuse;
816 break;
817
818 case SO_KEEPALIVE:
819 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
820 break;
821
822 case SO_TYPE:
823 v.val = sk->sk_type;
824 break;
825
49c794e9
JE		 826 case SO_PROTOCOL:
827 v.val = sk->sk_protocol;
828 break;
829
0d6038ee
JE		 830 case SO_DOMAIN:
831 v.val = sk->sk_family;
832 break;
833
e71a4783
SH		 834 case SO_ERROR:
835 v.val = -sock_error(sk);
2a91525c 836 if (v.val == 0)
e71a4783
SH		 837 v.val = xchg(&sk->sk_err_soft, 0);
838 break;
839
840 case SO_OOBINLINE:
841 v.val = !!sock_flag(sk, SOCK_URGINLINE);
842 break;
843
844 case SO_NO_CHECK:
845 v.val = sk->sk_no_check;
846 break;
847
848 case SO_PRIORITY:
849 v.val = sk->sk_priority;
850 break;
851
852 case SO_LINGER:
853 lv = sizeof(v.ling);
854 v.ling.l_onoff = !!sock_flag(sk, SOCK_LINGER);
855 v.ling.l_linger = sk->sk_lingertime / HZ;
856 break;
857
858 case SO_BSDCOMPAT:
859 sock_warn_obsolete_bsdism("getsockopt");
860 break;
861
862 case SO_TIMESTAMP:
92f37fd2
ED		 863 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
864 !sock_flag(sk, SOCK_RCVTSTAMPNS);
865 break;
866
867 case SO_TIMESTAMPNS:
868 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 869 break;
870
20d49473
PO		 871 case SO_TIMESTAMPING:
872 v.val = 0;
873 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
874 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
875 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
876 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
877 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
878 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
879 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
880 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
881 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
882 v.val |= SOF_TIMESTAMPING_SOFTWARE;
883 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
884 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
885 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
886 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
887 break;
888
e71a4783 889 case SO_RCVTIMEO:
2a91525c 890 lv = sizeof(struct timeval);
e71a4783
SH		 891 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
892 v.tm.tv_sec = 0;
893 v.tm.tv_usec = 0;
894 } else {
895 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
896 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
897 }
898 break;
899
900 case SO_SNDTIMEO:
2a91525c 901 lv = sizeof(struct timeval);
e71a4783
SH		 902 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
903 v.tm.tv_sec = 0;
904 v.tm.tv_usec = 0;
905 } else {
906 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
907 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
908 }
909 break;
1da177e4 910
e71a4783
SH		 911 case SO_RCVLOWAT:
912 v.val = sk->sk_rcvlowat;
913 break;
1da177e4 914
e71a4783 915 case SO_SNDLOWAT:
2a91525c 916 v.val = 1;
e71a4783 917 break;
1da177e4 918
e71a4783
SH		 919 case SO_PASSCRED:
920 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
921 break;
1da177e4 922
e71a4783 923 case SO_PEERCRED:
109f6e39
EB		 924 {
925 struct ucred peercred;
926 if (len > sizeof(peercred))
927 len = sizeof(peercred);
928 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
929 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 930 return -EFAULT;
931 goto lenout;
109f6e39 932 }
1da177e4 933
e71a4783
SH		 934 case SO_PEERNAME:
935 {
936 char address[128];
937
938 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
939 return -ENOTCONN;
940 if (lv < len)
941 return -EINVAL;
942 if (copy_to_user(optval, address, len))
943 return -EFAULT;
944 goto lenout;
945 }
1da177e4 946
e71a4783
SH		 947 /* Dubious BSD thing... Probably nobody even uses it, but
948 * the UNIX standard wants it for whatever reason... -DaveM
949 */
950 case SO_ACCEPTCONN:
951 v.val = sk->sk_state == TCP_LISTEN;
952 break;
1da177e4 953
e71a4783
SH		 954 case SO_PASSSEC:
955 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
956 break;
877ce7c1 957
e71a4783
SH		 958 case SO_PEERSEC:
959 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 960
4a19ec58
LAT		 961 case SO_MARK:
962 v.val = sk->sk_mark;
963 break;
964
3b885787
NH		 965 case SO_RXQ_OVFL:
966 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
967 break;
968
6e3e939f
JB		 969 case SO_WIFI_STATUS:
970 v.val = !!sock_flag(sk, SOCK_WIFI_STATUS);
971 break;
972
e71a4783
SH		 973 default:
974 return -ENOPROTOOPT;
1da177e4 975 }
e71a4783 976
1da177e4
LT		 977 if (len > lv)
978 len = lv;
979 if (copy_to_user(optval, &v, len))
980 return -EFAULT;
981lenout:
4ec93edb
YH		 982 if (put_user(len, optlen))
983 return -EFAULT;
984 return 0;
1da177e4
LT		 985}
986
a5b5bb9a
IM		 987/*
988 * Initialize an sk_lock.
989 *
990 * (We also register the sk_lock with the lock validator.)
991 */
b6f99a21 992static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 993{
ed07536e
PZ		 994 sock_lock_init_class_and_name(sk,
995 af_family_slock_key_strings[sk->sk_family],
996 af_family_slock_keys + sk->sk_family,
997 af_family_key_strings[sk->sk_family],
998 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 999}
1000
4dc6dc71
ED		 1001/*
1002 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1003 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1004 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1005 */
f1a6c4da
PE		 1006static void sock_copy(struct sock *nsk, const struct sock *osk)
1007{
1008#ifdef CONFIG_SECURITY_NETWORK
1009 void *sptr = nsk->sk_security;
1010#endif
68835aba
ED		 1011 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1012
1013 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1014 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1015
f1a6c4da
PE		 1016#ifdef CONFIG_SECURITY_NETWORK
1017 nsk->sk_security = sptr;
1018 security_sk_clone(osk, nsk);
1019#endif
1020}
1021
fcbdf09d
OP		 1022/*
1023 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1024 * un-modified. Special care is taken when initializing object to zero.
1025 */
1026static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1027{
1028 if (offsetof(struct sock, sk_node.next) != 0)
1029 memset(sk, 0, offsetof(struct sock, sk_node.next));
1030 memset(&sk->sk_node.pprev, 0,
1031 size - offsetof(struct sock, sk_node.pprev));
1032}
1033
1034void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1035{
1036 unsigned long nulls1, nulls2;
1037
1038 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1039 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1040 if (nulls1 > nulls2)
1041 swap(nulls1, nulls2);
1042
1043 if (nulls1 != 0)
1044 memset((char *)sk, 0, nulls1);
1045 memset((char *)sk + nulls1 + sizeof(void *), 0,
1046 nulls2 - nulls1 - sizeof(void *));
1047 memset((char *)sk + nulls2 + sizeof(void *), 0,
1048 size - nulls2 - sizeof(void *));
1049}
1050EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1051
2e4afe7b
PE		 1052static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1053 int family)
c308c1b2
PE		 1054{
1055 struct sock *sk;
1056 struct kmem_cache *slab;
1057
1058 slab = prot->slab;
e912b114
ED		 1059 if (slab != NULL) {
1060 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1061 if (!sk)
1062 return sk;
1063 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1064 if (prot->clear_sk)
1065 prot->clear_sk(sk, prot->obj_size);
1066 else
1067 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1068 }
fcbdf09d 1069 } else
c308c1b2
PE		 1070 sk = kmalloc(prot->obj_size, priority);
1071
2e4afe7b 1072 if (sk != NULL) {
a98b65a3
VN		 1073 kmemcheck_annotate_bitfield(sk, flags);
1074
2e4afe7b
PE		 1075 if (security_sk_alloc(sk, family, priority))
1076 goto out_free;
1077
1078 if (!try_module_get(prot->owner))
1079 goto out_free_sec;
e022f0b4 1080 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1081 }
1082
c308c1b2 1083 return sk;
2e4afe7b
PE		 1084
1085out_free_sec:
1086 security_sk_free(sk);
1087out_free:
1088 if (slab != NULL)
1089 kmem_cache_free(slab, sk);
1090 else
1091 kfree(sk);
1092 return NULL;
c308c1b2
PE		 1093}
1094
1095static void sk_prot_free(struct proto *prot, struct sock *sk)
1096{
1097 struct kmem_cache *slab;
2e4afe7b 1098 struct module *owner;
c308c1b2 1099
2e4afe7b 1100 owner = prot->owner;
c308c1b2 1101 slab = prot->slab;
2e4afe7b
PE		 1102
1103 security_sk_free(sk);
c308c1b2
PE		 1104 if (slab != NULL)
1105 kmem_cache_free(slab, sk);
1106 else
1107 kfree(sk);
2e4afe7b 1108 module_put(owner);
c308c1b2
PE		 1109}
1110
f8451725
HX		 1111#ifdef CONFIG_CGROUPS
1112void sock_update_classid(struct sock *sk)
1113{
1144182a 1114 u32 classid;
f8451725 1115
1144182a
PM		 1116 rcu_read_lock(); /* doing current task, which cannot vanish. */
1117 classid = task_cls_classid(current);
1118 rcu_read_unlock();
f8451725
HX		 1119 if (classid && classid != sk->sk_classid)
1120 sk->sk_classid = classid;
1121}
82862742 1122EXPORT_SYMBOL(sock_update_classid);
f8451725
HX		 1123#endif
1124
1da177e4
LT		 1125/**
1126 * sk_alloc - All socket objects are allocated here
c4ea43c5 1127 * @net: the applicable net namespace
4dc3b16b
PP		 1128 * @family: protocol family
1129 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1130 * @prot: struct proto associated with this new sock instance
1da177e4 1131 */
1b8d7ae4 1132struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1133 struct proto *prot)
1da177e4 1134{
c308c1b2 1135 struct sock *sk;
1da177e4 1136
154adbc8 1137 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1138 if (sk) {
154adbc8
PE		 1139 sk->sk_family = family;
1140 /*
1141 * See comment in struct sock definition to understand
1142 * why we need sk_prot_creator -acme
1143 */
1144 sk->sk_prot = sk->sk_prot_creator = prot;
1145 sock_lock_init(sk);
3b1e0a65 1146 sock_net_set(sk, get_net(net));
d66ee058 1147 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1148
1149 sock_update_classid(sk);
1da177e4 1150 }
a79af59e 1151
2e4afe7b 1152 return sk;
1da177e4 1153}
2a91525c 1154EXPORT_SYMBOL(sk_alloc);
1da177e4 1155
2b85a34e 1156static void __sk_free(struct sock *sk)
1da177e4
LT		 1157{
1158 struct sk_filter *filter;
1da177e4
LT		 1159
1160 if (sk->sk_destruct)
1161 sk->sk_destruct(sk);
1162
a898def2
PM		 1163 filter = rcu_dereference_check(sk->sk_filter,
1164 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1165 if (filter) {
309dd5fc 1166 sk_filter_uncharge(sk, filter);
a9b3cd7f 1167 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1168 }
1169
20d49473
PO		 1170 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1171 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1da177e4
LT		 1172
1173 if (atomic_read(&sk->sk_omem_alloc))
1174 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1175 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1176
109f6e39
EB		 1177 if (sk->sk_peer_cred)
1178 put_cred(sk->sk_peer_cred);
1179 put_pid(sk->sk_peer_pid);
3b1e0a65 1180 put_net(sock_net(sk));
c308c1b2 1181 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1182}
2b85a34e
ED		 1183
1184void sk_free(struct sock *sk)
1185{
1186 /*
25985edc 1187 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1188 * some packets are still in some tx queue.
1189 * If not null, sock_wfree() will call __sk_free(sk) later
1190 */
1191 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1192 __sk_free(sk);
1193}
2a91525c 1194EXPORT_SYMBOL(sk_free);
1da177e4 1195
edf02087 1196/*
25985edc
LDM		 1197 * Last sock_put should drop reference to sk->sk_net. It has already
1198 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1199 * is not an option.
25985edc 1200 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1201 * destroy it in the context of init_net.
1202 */
1203void sk_release_kernel(struct sock *sk)
1204{
1205 if (sk == NULL || sk->sk_socket == NULL)
1206 return;
1207
1208 sock_hold(sk);
1209 sock_release(sk->sk_socket);
65a18ec5 1210 release_net(sock_net(sk));
3b1e0a65 1211 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1212 sock_put(sk);
1213}
45af1754 1214EXPORT_SYMBOL(sk_release_kernel);
edf02087 1215
e56c57d0
ED		 1216/**
1217 * sk_clone_lock - clone a socket, and lock its clone
1218 * @sk: the socket to clone
1219 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1220 *
1221 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1222 */
1223struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1224{
8fd1d178 1225 struct sock *newsk;
87d11ceb 1226
8fd1d178 1227 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1228 if (newsk != NULL) {
1229 struct sk_filter *filter;
1230
892c141e 1231 sock_copy(newsk, sk);
87d11ceb
ACM		 1232
1233 /* SANITY */
3b1e0a65 1234 get_net(sock_net(newsk));
87d11ceb
ACM		 1235 sk_node_init(&newsk->sk_node);
1236 sock_lock_init(newsk);
1237 bh_lock_sock(newsk);
fa438ccf 1238 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1239 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1240
1241 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1242 /*
1243 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1244 */
1245 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1246 atomic_set(&newsk->sk_omem_alloc, 0);
1247 skb_queue_head_init(&newsk->sk_receive_queue);
1248 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1249#ifdef CONFIG_NET_DMA
1250 skb_queue_head_init(&newsk->sk_async_wait_queue);
1251#endif
87d11ceb 1252
b6c6712a 1253 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1254 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1255 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1256 af_callback_keys + newsk->sk_family,
1257 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1258
1259 newsk->sk_dst_cache = NULL;
1260 newsk->sk_wmem_queued = 0;
1261 newsk->sk_forward_alloc = 0;
1262 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1263 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1264
1265 sock_reset_flag(newsk, SOCK_DONE);
1266 skb_queue_head_init(&newsk->sk_error_queue);
1267
0d7da9dd 1268 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1269 if (filter != NULL)
1270 sk_filter_charge(newsk, filter);
1271
1272 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1273 /* It is still raw copy of parent, so invalidate
1274 * destructor and make plain sk_free() */
1275 newsk->sk_destruct = NULL;
b0691c8e 1276 bh_unlock_sock(newsk);
87d11ceb
ACM		 1277 sk_free(newsk);
1278 newsk = NULL;
1279 goto out;
1280 }
1281
1282 newsk->sk_err = 0;
1283 newsk->sk_priority = 0;
4dc6dc71
ED		 1284 /*
1285 * Before updating sk_refcnt, we must commit prior changes to memory
1286 * (Documentation/RCU/rculist_nulls.txt for details)
1287 */
1288 smp_wmb();
87d11ceb
ACM		 1289 atomic_set(&newsk->sk_refcnt, 2);
1290
1291 /*
1292 * Increment the counter in the same struct proto as the master
1293 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1294 * is the same as sk->sk_prot->socks, as this field was copied
1295 * with memcpy).
1296 *
1297 * This _changes_ the previous behaviour, where
1298 * tcp_create_openreq_child always was incrementing the
1299 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1300 * to be taken into account in all callers. -acme
1301 */
1302 sk_refcnt_debug_inc(newsk);
972692e0 1303 sk_set_socket(newsk, NULL);
43815482 1304 newsk->sk_wq = NULL;
87d11ceb
ACM		 1305
1306 if (newsk->sk_prot->sockets_allocated)
1748376b 1307 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
704da560
OP		 1308
1309 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1310 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1311 net_enable_timestamp();
87d11ceb
ACM		 1312 }
1313out:
1314 return newsk;
1315}
e56c57d0 1316EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1317
9958089a
AK		 1318void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1319{
1320 __sk_dst_set(sk, dst);
1321 sk->sk_route_caps = dst->dev->features;
1322 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1323 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1324 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1325 if (sk_can_gso(sk)) {
82cc1a7a 1326 if (dst->header_len) {
9958089a 1327 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1328 } else {
9958089a 1329 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1330 sk->sk_gso_max_size = dst->dev->gso_max_size;
1331 }
9958089a
AK		 1332 }
1333}
1334EXPORT_SYMBOL_GPL(sk_setup_caps);
1335
1da177e4
LT		 1336void __init sk_init(void)
1337{
4481374c 1338 if (totalram_pages <= 4096) {
1da177e4
LT		 1339 sysctl_wmem_max = 32767;
1340 sysctl_rmem_max = 32767;
1341 sysctl_wmem_default = 32767;
1342 sysctl_rmem_default = 32767;
4481374c 1343 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1344 sysctl_wmem_max = 131071;
1345 sysctl_rmem_max = 131071;
1346 }
1347}
1348
1349/*
1350 * Simple resource managers for sockets.
1351 */
1352
1353
4ec93edb
YH		 1354/*
1355 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1356 */
1357void sock_wfree(struct sk_buff *skb)
1358{
1359 struct sock *sk = skb->sk;
d99927f4 1360 unsigned int len = skb->truesize;
1da177e4 1361
d99927f4
ED		 1362 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1363 /*
1364 * Keep a reference on sk_wmem_alloc, this will be released
1365 * after sk_write_space() call
1366 */
1367 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1368 sk->sk_write_space(sk);
d99927f4
ED		 1369 len = 1;
1370 }
2b85a34e 1371 /*
d99927f4
ED		 1372 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1373 * could not do because of in-flight packets
2b85a34e 1374 */
d99927f4 1375 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1376 __sk_free(sk);
1da177e4 1377}
2a91525c 1378EXPORT_SYMBOL(sock_wfree);
1da177e4 1379
4ec93edb
YH		 1380/*
1381 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1382 */
1383void sock_rfree(struct sk_buff *skb)
1384{
1385 struct sock *sk = skb->sk;
d361fd59 1386 unsigned int len = skb->truesize;
1da177e4 1387
d361fd59
ED		 1388 atomic_sub(len, &sk->sk_rmem_alloc);
1389 sk_mem_uncharge(sk, len);
1da177e4 1390}
2a91525c 1391EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1392
1393
1394int sock_i_uid(struct sock *sk)
1395{
1396 int uid;
1397
f064af1e 1398 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1399 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1400 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1401 return uid;
1402}
2a91525c 1403EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1404
1405unsigned long sock_i_ino(struct sock *sk)
1406{
1407 unsigned long ino;
1408
f064af1e 1409 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1410 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1411 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1412 return ino;
1413}
2a91525c 1414EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1415
1416/*
1417 * Allocate a skb from the socket's send buffer.
1418 */
86a76caf 1419struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1420 gfp_t priority)
1da177e4
LT		 1421{
1422 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1423 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1424 if (skb) {
1425 skb_set_owner_w(skb, sk);
1426 return skb;
1427 }
1428 }
1429 return NULL;
1430}
2a91525c 1431EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1432
1433/*
1434 * Allocate a skb from the socket's receive buffer.
4ec93edb 1435 */
86a76caf 1436struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1437 gfp_t priority)
1da177e4
LT		 1438{
1439 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1440 struct sk_buff *skb = alloc_skb(size, priority);
1441 if (skb) {
1442 skb_set_owner_r(skb, sk);
1443 return skb;
1444 }
1445 }
1446 return NULL;
1447}
1448
4ec93edb 1449/*
1da177e4 1450 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1451 */
dd0fc66f 1452void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1453{
1454 if ((unsigned)size <= sysctl_optmem_max &&
1455 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1456 void *mem;
1457 /* First do the add, to avoid the race if kmalloc
4ec93edb 1458 * might sleep.
1da177e4
LT		 1459 */
1460 atomic_add(size, &sk->sk_omem_alloc);
1461 mem = kmalloc(size, priority);
1462 if (mem)
1463 return mem;
1464 atomic_sub(size, &sk->sk_omem_alloc);
1465 }
1466 return NULL;
1467}
2a91525c 1468EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1469
1470/*
1471 * Free an option memory block.
1472 */
1473void sock_kfree_s(struct sock *sk, void *mem, int size)
1474{
1475 kfree(mem);
1476 atomic_sub(size, &sk->sk_omem_alloc);
1477}
2a91525c 1478EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1479
1480/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1481 I think, these locks should be removed for datagram sockets.
1482 */
2a91525c 1483static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1484{
1485 DEFINE_WAIT(wait);
1486
1487 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1488 for (;;) {
1489 if (!timeo)
1490 break;
1491 if (signal_pending(current))
1492 break;
1493 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1494 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1495 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1496 break;
1497 if (sk->sk_shutdown & SEND_SHUTDOWN)
1498 break;
1499 if (sk->sk_err)
1500 break;
1501 timeo = schedule_timeout(timeo);
1502 }
aa395145 1503 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1504 return timeo;
1505}
1506
1507
1508/*
1509 * Generic send/receive buffer handlers
1510 */
1511
4cc7f68d
HX		 1512struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1513 unsigned long data_len, int noblock,
1514 int *errcode)
1da177e4
LT		 1515{
1516 struct sk_buff *skb;
7d877f3b 1517 gfp_t gfp_mask;
1da177e4
LT		 1518 long timeo;
1519 int err;
1520
1521 gfp_mask = sk->sk_allocation;
1522 if (gfp_mask & __GFP_WAIT)
1523 gfp_mask |= __GFP_REPEAT;
1524
1525 timeo = sock_sndtimeo(sk, noblock);
1526 while (1) {
1527 err = sock_error(sk);
1528 if (err != 0)
1529 goto failure;
1530
1531 err = -EPIPE;
1532 if (sk->sk_shutdown & SEND_SHUTDOWN)
1533 goto failure;
1534
1535 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1536 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1537 if (skb) {
1538 int npages;
1539 int i;
1540
1541 /* No pages, we're done... */
1542 if (!data_len)
1543 break;
1544
1545 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1546 skb->truesize += data_len;
1547 skb_shinfo(skb)->nr_frags = npages;
1548 for (i = 0; i < npages; i++) {
1549 struct page *page;
1da177e4
LT		 1550
1551 page = alloc_pages(sk->sk_allocation, 0);
1552 if (!page) {
1553 err = -ENOBUFS;
1554 skb_shinfo(skb)->nr_frags = i;
1555 kfree_skb(skb);
1556 goto failure;
1557 }
1558
ea2ab693
IC		 1559 __skb_fill_page_desc(skb, i,
1560 page, 0,
1561 (data_len >= PAGE_SIZE ?
1562 PAGE_SIZE :
1563 data_len));
1da177e4
LT		 1564 data_len -= PAGE_SIZE;
1565 }
1566
1567 /* Full success... */
1568 break;
1569 }
1570 err = -ENOBUFS;
1571 goto failure;
1572 }
1573 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1574 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1575 err = -EAGAIN;
1576 if (!timeo)
1577 goto failure;
1578 if (signal_pending(current))
1579 goto interrupted;
1580 timeo = sock_wait_for_wmem(sk, timeo);
1581 }
1582
1583 skb_set_owner_w(skb, sk);
1584 return skb;
1585
1586interrupted:
1587 err = sock_intr_errno(timeo);
1588failure:
1589 *errcode = err;
1590 return NULL;
1591}
4cc7f68d 1592EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1593
4ec93edb 1594struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1595 int noblock, int *errcode)
1596{
1597 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1598}
2a91525c 1599EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1600
1601static void __lock_sock(struct sock *sk)
f39234d6
NK		 1602 __releases(&sk->sk_lock.slock)
1603 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1604{
1605 DEFINE_WAIT(wait);
1606
e71a4783 1607 for (;;) {
1da177e4
LT		 1608 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1609 TASK_UNINTERRUPTIBLE);
1610 spin_unlock_bh(&sk->sk_lock.slock);
1611 schedule();
1612 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1613 if (!sock_owned_by_user(sk))
1da177e4
LT		 1614 break;
1615 }
1616 finish_wait(&sk->sk_lock.wq, &wait);
1617}
1618
1619static void __release_sock(struct sock *sk)
f39234d6
NK		 1620 __releases(&sk->sk_lock.slock)
1621 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1622{
1623 struct sk_buff *skb = sk->sk_backlog.head;
1624
1625 do {
1626 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1627 bh_unlock_sock(sk);
1628
1629 do {
1630 struct sk_buff *next = skb->next;
1631
7fee226a 1632 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1633 skb->next = NULL;
c57943a1 1634 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1635
1636 /*
1637 * We are in process context here with softirqs
1638 * disabled, use cond_resched_softirq() to preempt.
1639 * This is safe to do because we've taken the backlog
1640 * queue private:
1641 */
1642 cond_resched_softirq();
1643
1644 skb = next;
1645 } while (skb != NULL);
1646
1647 bh_lock_sock(sk);
e71a4783 1648 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1649
1650 /*
1651 * Doing the zeroing here guarantee we can not loop forever
1652 * while a wild producer attempts to flood us.
1653 */
1654 sk->sk_backlog.len = 0;
1da177e4
LT		 1655}
1656
1657/**
1658 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1659 * @sk: sock to wait on
1660 * @timeo: for how long
1da177e4
LT		 1661 *
1662 * Now socket state including sk->sk_err is changed only under lock,
1663 * hence we may omit checks after joining wait queue.
1664 * We check receive queue before schedule() only as optimization;
1665 * it is very likely that release_sock() added new data.
1666 */
1667int sk_wait_data(struct sock *sk, long *timeo)
1668{
1669 int rc;
1670 DEFINE_WAIT(wait);
1671
aa395145 1672 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1673 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1674 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1675 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1676 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1677 return rc;
1678}
1da177e4
LT		 1679EXPORT_SYMBOL(sk_wait_data);
1680
3ab224be
HA		 1681/**
1682 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1683 * @sk: socket
1684 * @size: memory size to allocate
1685 * @kind: allocation type
1686 *
1687 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1688 * rmem allocation. This function assumes that protocols which have
1689 * memory_pressure use sk_wmem_queued as write buffer accounting.
1690 */
1691int __sk_mem_schedule(struct sock *sk, int size, int kind)
1692{
1693 struct proto *prot = sk->sk_prot;
1694 int amt = sk_mem_pages(size);
8d987e5c 1695 long allocated;
3ab224be
HA		 1696
1697 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
8d987e5c 1698 allocated = atomic_long_add_return(amt, prot->memory_allocated);
3ab224be
HA		 1699
1700 /* Under limit. */
1701 if (allocated <= prot->sysctl_mem[0]) {
1702 if (prot->memory_pressure && *prot->memory_pressure)
1703 *prot->memory_pressure = 0;
1704 return 1;
1705 }
1706
1707 /* Under pressure. */
1708 if (allocated > prot->sysctl_mem[1])
1709 if (prot->enter_memory_pressure)
5c52ba17 1710 prot->enter_memory_pressure(sk);
3ab224be
HA		 1711
1712 /* Over hard limit. */
1713 if (allocated > prot->sysctl_mem[2])
1714 goto suppress_allocation;
1715
1716 /* guarantee minimum buffer size under pressure */
1717 if (kind == SK_MEM_RECV) {
1718 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1719 return 1;
1720 } else { /* SK_MEM_SEND */
1721 if (sk->sk_type == SOCK_STREAM) {
1722 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1723 return 1;
1724 } else if (atomic_read(&sk->sk_wmem_alloc) <
1725 prot->sysctl_wmem[0])
1726 return 1;
1727 }
1728
1729 if (prot->memory_pressure) {
1748376b
ED		 1730 int alloc;
1731
1732 if (!*prot->memory_pressure)
1733 return 1;
1734 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1735 if (prot->sysctl_mem[2] > alloc *
3ab224be
HA		 1736 sk_mem_pages(sk->sk_wmem_queued +
1737 atomic_read(&sk->sk_rmem_alloc) +
1738 sk->sk_forward_alloc))
1739 return 1;
1740 }
1741
1742suppress_allocation:
1743
1744 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1745 sk_stream_moderate_sndbuf(sk);
1746
1747 /* Fail only if socket is _under_ its sndbuf.
1748 * In this case we cannot block, so that we have to fail.
1749 */
1750 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1751 return 1;
1752 }
1753
3847ce32
SM		 1754 trace_sock_exceed_buf_limit(sk, prot, allocated);
1755
3ab224be
HA		 1756 /* Alas. Undo changes. */
1757 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
8d987e5c 1758 atomic_long_sub(amt, prot->memory_allocated);
3ab224be
HA		 1759 return 0;
1760}
3ab224be
HA		 1761EXPORT_SYMBOL(__sk_mem_schedule);
1762
1763/**
1764 * __sk_reclaim - reclaim memory_allocated
1765 * @sk: socket
1766 */
1767void __sk_mem_reclaim(struct sock *sk)
1768{
1769 struct proto *prot = sk->sk_prot;
1770
8d987e5c 1771 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1772 prot->memory_allocated);
1773 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1774
1775 if (prot->memory_pressure && *prot->memory_pressure &&
8d987e5c 1776 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
3ab224be
HA		 1777 *prot->memory_pressure = 0;
1778}
3ab224be
HA		 1779EXPORT_SYMBOL(__sk_mem_reclaim);
1780
1781
1da177e4
LT		 1782/*
1783 * Set of default routines for initialising struct proto_ops when
1784 * the protocol does not support a particular function. In certain
1785 * cases where it makes no sense for a protocol to have a "do nothing"
1786 * function, some default processing is provided.
1787 */
1788
1789int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1790{
1791 return -EOPNOTSUPP;
1792}
2a91525c 1793EXPORT_SYMBOL(sock_no_bind);
1da177e4 1794
4ec93edb 1795int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1796 int len, int flags)
1797{
1798 return -EOPNOTSUPP;
1799}
2a91525c 1800EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1801
1802int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1803{
1804 return -EOPNOTSUPP;
1805}
2a91525c 1806EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1807
1808int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1809{
1810 return -EOPNOTSUPP;
1811}
2a91525c 1812EXPORT_SYMBOL(sock_no_accept);
1da177e4 1813
4ec93edb 1814int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1815 int *len, int peer)
1816{
1817 return -EOPNOTSUPP;
1818}
2a91525c 1819EXPORT_SYMBOL(sock_no_getname);
1da177e4 1820
2a91525c 1821unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1822{
1823 return 0;
1824}
2a91525c 1825EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1826
1827int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1828{
1829 return -EOPNOTSUPP;
1830}
2a91525c 1831EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1832
1833int sock_no_listen(struct socket *sock, int backlog)
1834{
1835 return -EOPNOTSUPP;
1836}
2a91525c 1837EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1838
1839int sock_no_shutdown(struct socket *sock, int how)
1840{
1841 return -EOPNOTSUPP;
1842}
2a91525c 1843EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1844
1845int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1846 char __user *optval, unsigned int optlen)
1da177e4
LT		 1847{
1848 return -EOPNOTSUPP;
1849}
2a91525c 1850EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1851
1852int sock_no_getsockopt(struct socket *sock, int level, int optname,
1853 char __user *optval, int __user *optlen)
1854{
1855 return -EOPNOTSUPP;
1856}
2a91525c 1857EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1858
1859int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1860 size_t len)
1861{
1862 return -EOPNOTSUPP;
1863}
2a91525c 1864EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1865
1866int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1867 size_t len, int flags)
1868{
1869 return -EOPNOTSUPP;
1870}
2a91525c 1871EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1872
1873int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1874{
1875 /* Mirror missing mmap method error code */
1876 return -ENODEV;
1877}
2a91525c 1878EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1879
1880ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1881{
1882 ssize_t res;
1883 struct msghdr msg = {.msg_flags = flags};
1884 struct kvec iov;
1885 char *kaddr = kmap(page);
1886 iov.iov_base = kaddr + offset;
1887 iov.iov_len = size;
1888 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1889 kunmap(page);
1890 return res;
1891}
2a91525c 1892EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1893
1894/*
1895 * Default Socket Callbacks
1896 */
1897
1898static void sock_def_wakeup(struct sock *sk)
1899{
43815482
ED		 1900 struct socket_wq *wq;
1901
1902 rcu_read_lock();
1903 wq = rcu_dereference(sk->sk_wq);
1904 if (wq_has_sleeper(wq))
1905 wake_up_interruptible_all(&wq->wait);
1906 rcu_read_unlock();
1da177e4
LT		 1907}
1908
1909static void sock_def_error_report(struct sock *sk)
1910{
43815482
ED		 1911 struct socket_wq *wq;
1912
1913 rcu_read_lock();
1914 wq = rcu_dereference(sk->sk_wq);
1915 if (wq_has_sleeper(wq))
1916 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1917 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1918 rcu_read_unlock();
1da177e4
LT		 1919}
1920
1921static void sock_def_readable(struct sock *sk, int len)
1922{
43815482
ED		 1923 struct socket_wq *wq;
1924
1925 rcu_read_lock();
1926 wq = rcu_dereference(sk->sk_wq);
1927 if (wq_has_sleeper(wq))
2c6607c6 1928 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1929 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1930 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1931 rcu_read_unlock();
1da177e4
LT		 1932}
1933
1934static void sock_def_write_space(struct sock *sk)
1935{
43815482
ED		 1936 struct socket_wq *wq;
1937
1938 rcu_read_lock();
1da177e4
LT		 1939
1940 /* Do not wake up a writer until he can make "significant"
1941 * progress. --DaveM
1942 */
e71a4783 1943 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 1944 wq = rcu_dereference(sk->sk_wq);
1945 if (wq_has_sleeper(wq))
1946 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 1947 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 1948
1949 /* Should agree with poll, otherwise some programs break */
1950 if (sock_writeable(sk))
8d8ad9d7 1951 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1952 }
1953
43815482 1954 rcu_read_unlock();
1da177e4
LT		 1955}
1956
1957static void sock_def_destruct(struct sock *sk)
1958{
a51482bd 1959 kfree(sk->sk_protinfo);
1da177e4
LT		 1960}
1961
1962void sk_send_sigurg(struct sock *sk)
1963{
1964 if (sk->sk_socket && sk->sk_socket->file)
1965 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 1966 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 1967}
2a91525c 1968EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 1969
1970void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1971 unsigned long expires)
1972{
1973 if (!mod_timer(timer, expires))
1974 sock_hold(sk);
1975}
1da177e4
LT		 1976EXPORT_SYMBOL(sk_reset_timer);
1977
1978void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1979{
1980 if (timer_pending(timer) && del_timer(timer))
1981 __sock_put(sk);
1982}
1da177e4
LT		 1983EXPORT_SYMBOL(sk_stop_timer);
1984
1985void sock_init_data(struct socket *sock, struct sock *sk)
1986{
1987 skb_queue_head_init(&sk->sk_receive_queue);
1988 skb_queue_head_init(&sk->sk_write_queue);
1989 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 1990#ifdef CONFIG_NET_DMA
1991 skb_queue_head_init(&sk->sk_async_wait_queue);
1992#endif
1da177e4
LT		 1993
1994 sk->sk_send_head = NULL;
1995
1996 init_timer(&sk->sk_timer);
4ec93edb 1997
1da177e4
LT		 1998 sk->sk_allocation = GFP_KERNEL;
1999 sk->sk_rcvbuf = sysctl_rmem_default;
2000 sk->sk_sndbuf = sysctl_wmem_default;
2001 sk->sk_state = TCP_CLOSE;
972692e0 2002 sk_set_socket(sk, sock);
1da177e4
LT		 2003
2004 sock_set_flag(sk, SOCK_ZAPPED);
2005
e71a4783 2006 if (sock) {
1da177e4 2007 sk->sk_type = sock->type;
43815482 2008 sk->sk_wq = sock->wq;
1da177e4
LT		 2009 sock->sk = sk;
2010 } else
43815482 2011 sk->sk_wq = NULL;
1da177e4 2012
b6c6712a 2013 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2014 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2015 lockdep_set_class_and_name(&sk->sk_callback_lock,
2016 af_callback_keys + sk->sk_family,
2017 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2018
2019 sk->sk_state_change = sock_def_wakeup;
2020 sk->sk_data_ready = sock_def_readable;
2021 sk->sk_write_space = sock_def_write_space;
2022 sk->sk_error_report = sock_def_error_report;
2023 sk->sk_destruct = sock_def_destruct;
2024
2025 sk->sk_sndmsg_page = NULL;
2026 sk->sk_sndmsg_off = 0;
2027
109f6e39
EB		 2028 sk->sk_peer_pid = NULL;
2029 sk->sk_peer_cred = NULL;
1da177e4
LT		 2030 sk->sk_write_pending = 0;
2031 sk->sk_rcvlowat = 1;
2032 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2033 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2034
f37f0afb 2035 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2036
4dc6dc71
ED		 2037 /*
2038 * Before updating sk_refcnt, we must commit prior changes to memory
2039 * (Documentation/RCU/rculist_nulls.txt for details)
2040 */
2041 smp_wmb();
1da177e4 2042 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2043 atomic_set(&sk->sk_drops, 0);
1da177e4 2044}
2a91525c 2045EXPORT_SYMBOL(sock_init_data);
1da177e4 2046
b5606c2d 2047void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2048{
2049 might_sleep();
a5b5bb9a 2050 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2051 if (sk->sk_lock.owned)
1da177e4 2052 __lock_sock(sk);
d2e9117c 2053 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2054 spin_unlock(&sk->sk_lock.slock);
2055 /*
2056 * The sk_lock has mutex_lock() semantics here:
2057 */
fcc70d5f 2058 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2059 local_bh_enable();
1da177e4 2060}
fcc70d5f 2061EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2062
b5606c2d 2063void release_sock(struct sock *sk)
1da177e4 2064{
a5b5bb9a
IM		 2065 /*
2066 * The sk_lock has mutex_unlock() semantics:
2067 */
2068 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2069
2070 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2071 if (sk->sk_backlog.tail)
2072 __release_sock(sk);
d2e9117c 2073 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2074 if (waitqueue_active(&sk->sk_lock.wq))
2075 wake_up(&sk->sk_lock.wq);
2076 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2077}
2078EXPORT_SYMBOL(release_sock);
2079
8a74ad60
ED		 2080/**
2081 * lock_sock_fast - fast version of lock_sock
2082 * @sk: socket
2083 *
2084 * This version should be used for very small section, where process wont block
2085 * return false if fast path is taken
2086 * sk_lock.slock locked, owned = 0, BH disabled
2087 * return true if slow path is taken
2088 * sk_lock.slock unlocked, owned = 1, BH enabled
2089 */
2090bool lock_sock_fast(struct sock *sk)
2091{
2092 might_sleep();
2093 spin_lock_bh(&sk->sk_lock.slock);
2094
2095 if (!sk->sk_lock.owned)
2096 /*
2097 * Note : We must disable BH
2098 */
2099 return false;
2100
2101 __lock_sock(sk);
2102 sk->sk_lock.owned = 1;
2103 spin_unlock(&sk->sk_lock.slock);
2104 /*
2105 * The sk_lock has mutex_lock() semantics here:
2106 */
2107 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2108 local_bh_enable();
2109 return true;
2110}
2111EXPORT_SYMBOL(lock_sock_fast);
2112
1da177e4 2113int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2114{
b7aa0bf7 2115 struct timeval tv;
1da177e4 2116 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2117 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2118 tv = ktime_to_timeval(sk->sk_stamp);
2119 if (tv.tv_sec == -1)
1da177e4 2120 return -ENOENT;
b7aa0bf7
ED		 2121 if (tv.tv_sec == 0) {
2122 sk->sk_stamp = ktime_get_real();
2123 tv = ktime_to_timeval(sk->sk_stamp);
2124 }
2125 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2126}
1da177e4
LT		 2127EXPORT_SYMBOL(sock_get_timestamp);
2128
ae40eb1e
ED		 2129int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2130{
2131 struct timespec ts;
2132 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2133 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2134 ts = ktime_to_timespec(sk->sk_stamp);
2135 if (ts.tv_sec == -1)
2136 return -ENOENT;
2137 if (ts.tv_sec == 0) {
2138 sk->sk_stamp = ktime_get_real();
2139 ts = ktime_to_timespec(sk->sk_stamp);
2140 }
2141 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2142}
2143EXPORT_SYMBOL(sock_get_timestampns);
2144
20d49473 2145void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2146{
20d49473
PO		 2147 if (!sock_flag(sk, flag)) {
2148 sock_set_flag(sk, flag);
2149 /*
2150 * we just set one of the two flags which require net
2151 * time stamping, but time stamping might have been on
2152 * already because of the other one
2153 */
2154 if (!sock_flag(sk,
2155 flag == SOCK_TIMESTAMP ?
2156 SOCK_TIMESTAMPING_RX_SOFTWARE :
2157 SOCK_TIMESTAMP))
2158 net_enable_timestamp();
1da177e4
LT		 2159 }
2160}
1da177e4
LT		 2161
2162/*
2163 * Get a socket option on an socket.
2164 *
2165 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2166 * asynchronous errors should be reported by getsockopt. We assume
2167 * this means if you specify SO_ERROR (otherwise whats the point of it).
2168 */
2169int sock_common_getsockopt(struct socket *sock, int level, int optname,
2170 char __user *optval, int __user *optlen)
2171{
2172 struct sock *sk = sock->sk;
2173
2174 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2175}
1da177e4
LT		 2176EXPORT_SYMBOL(sock_common_getsockopt);
2177
3fdadf7d 2178#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2179int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2180 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2181{
2182 struct sock *sk = sock->sk;
2183
1e51f951 2184 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2185 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2186 optval, optlen);
3fdadf7d
DM		 2187 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2188}
2189EXPORT_SYMBOL(compat_sock_common_getsockopt);
2190#endif
2191
1da177e4
LT		 2192int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2193 struct msghdr *msg, size_t size, int flags)
2194{
2195 struct sock *sk = sock->sk;
2196 int addr_len = 0;
2197 int err;
2198
2199 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2200 flags & ~MSG_DONTWAIT, &addr_len);
2201 if (err >= 0)
2202 msg->msg_namelen = addr_len;
2203 return err;
2204}
1da177e4
LT		 2205EXPORT_SYMBOL(sock_common_recvmsg);
2206
2207/*
2208 * Set socket options on an inet socket.
2209 */
2210int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2211 char __user *optval, unsigned int optlen)
1da177e4
LT		 2212{
2213 struct sock *sk = sock->sk;
2214
2215 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2216}
1da177e4
LT		 2217EXPORT_SYMBOL(sock_common_setsockopt);
2218
3fdadf7d 2219#ifdef CONFIG_COMPAT
543d9cfe 2220int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2221 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2222{
2223 struct sock *sk = sock->sk;
2224
543d9cfe
ACM		 2225 if (sk->sk_prot->compat_setsockopt != NULL)
2226 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2227 optval, optlen);
3fdadf7d
DM		 2228 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2229}
2230EXPORT_SYMBOL(compat_sock_common_setsockopt);
2231#endif
2232
1da177e4
LT		 2233void sk_common_release(struct sock *sk)
2234{
2235 if (sk->sk_prot->destroy)
2236 sk->sk_prot->destroy(sk);
2237
2238 /*
2239 * Observation: when sock_common_release is called, processes have
2240 * no access to socket. But net still has.
2241 * Step one, detach it from networking:
2242 *
2243 * A. Remove from hash tables.
2244 */
2245
2246 sk->sk_prot->unhash(sk);
2247
2248 /*
2249 * In this point socket cannot receive new packets, but it is possible
2250 * that some packets are in flight because some CPU runs receiver and
2251 * did hash table lookup before we unhashed socket. They will achieve
2252 * receive queue and will be purged by socket destructor.
2253 *
2254 * Also we still have packets pending on receive queue and probably,
2255 * our own packets waiting in device queues. sock_destroy will drain
2256 * receive queue, but transmitted packets will delay socket destruction
2257 * until the last reference will be released.
2258 */
2259
2260 sock_orphan(sk);
2261
2262 xfrm_sk_free_policy(sk);
2263
e6848976 2264 sk_refcnt_debug_release(sk);
1da177e4
LT		 2265 sock_put(sk);
2266}
1da177e4
LT		 2267EXPORT_SYMBOL(sk_common_release);
2268
2269static DEFINE_RWLOCK(proto_list_lock);
2270static LIST_HEAD(proto_list);
2271
13ff3d6f
PE		 2272#ifdef CONFIG_PROC_FS
2273#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2274struct prot_inuse {
2275 int val[PROTO_INUSE_NR];
2276};
13ff3d6f
PE		 2277
2278static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2279
2280#ifdef CONFIG_NET_NS
2281void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2282{
d6d9ca0f 2283 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2284}
2285EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2286
2287int sock_prot_inuse_get(struct net *net, struct proto *prot)
2288{
2289 int cpu, idx = prot->inuse_idx;
2290 int res = 0;
2291
2292 for_each_possible_cpu(cpu)
2293 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2294
2295 return res >= 0 ? res : 0;
2296}
2297EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2298
2c8c1e72 2299static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2300{
2301 net->core.inuse = alloc_percpu(struct prot_inuse);
2302 return net->core.inuse ? 0 : -ENOMEM;
2303}
2304
2c8c1e72 2305static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2306{
2307 free_percpu(net->core.inuse);
2308}
2309
2310static struct pernet_operations net_inuse_ops = {
2311 .init = sock_inuse_init_net,
2312 .exit = sock_inuse_exit_net,
2313};
2314
2315static __init int net_inuse_init(void)
2316{
2317 if (register_pernet_subsys(&net_inuse_ops))
2318 panic("Cannot initialize net inuse counters");
2319
2320 return 0;
2321}
2322
2323core_initcall(net_inuse_init);
2324#else
1338d466
PE		 2325static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2326
c29a0bc4 2327void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2328{
d6d9ca0f 2329 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2330}
2331EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2332
c29a0bc4 2333int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2334{
2335 int cpu, idx = prot->inuse_idx;
2336 int res = 0;
2337
2338 for_each_possible_cpu(cpu)
2339 res += per_cpu(prot_inuse, cpu).val[idx];
2340
2341 return res >= 0 ? res : 0;
2342}
2343EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2344#endif
13ff3d6f
PE		 2345
2346static void assign_proto_idx(struct proto *prot)
2347{
2348 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2349
2350 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2351 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2352 return;
2353 }
2354
2355 set_bit(prot->inuse_idx, proto_inuse_idx);
2356}
2357
2358static void release_proto_idx(struct proto *prot)
2359{
2360 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2361 clear_bit(prot->inuse_idx, proto_inuse_idx);
2362}
2363#else
2364static inline void assign_proto_idx(struct proto *prot)
2365{
2366}
2367
2368static inline void release_proto_idx(struct proto *prot)
2369{
2370}
2371#endif
2372
b733c007
PE		 2373int proto_register(struct proto *prot, int alloc_slab)
2374{
1da177e4
LT		 2375 if (alloc_slab) {
2376 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2377 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2378 NULL);
1da177e4
LT		 2379
2380 if (prot->slab == NULL) {
2381 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2382 prot->name);
60e7663d 2383 goto out;
1da177e4 2384 }
2e6599cb
ACM		 2385
2386 if (prot->rsk_prot != NULL) {
faf23422 2387 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2388 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2389 goto out_free_sock_slab;
2390
7e56b5d6 2391 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2392 prot->rsk_prot->obj_size, 0,
20c2df83 2393 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2394
2395 if (prot->rsk_prot->slab == NULL) {
2396 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2397 prot->name);
2398 goto out_free_request_sock_slab_name;
2399 }
2400 }
8feaf0c0 2401
6d6ee43e 2402 if (prot->twsk_prot != NULL) {
faf23422 2403 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2404
7e56b5d6 2405 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2406 goto out_free_request_sock_slab;
2407
6d6ee43e 2408 prot->twsk_prot->twsk_slab =
7e56b5d6 2409 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2410 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2411 0,
2412 SLAB_HWCACHE_ALIGN |
2413 prot->slab_flags,
20c2df83 2414 NULL);
6d6ee43e 2415 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2416 goto out_free_timewait_sock_slab_name;
2417 }
1da177e4
LT		 2418 }
2419
2a278051 2420 write_lock(&proto_list_lock);
1da177e4 2421 list_add(&prot->node, &proto_list);
13ff3d6f 2422 assign_proto_idx(prot);
1da177e4 2423 write_unlock(&proto_list_lock);
b733c007
PE		 2424 return 0;
2425
8feaf0c0 2426out_free_timewait_sock_slab_name:
7e56b5d6 2427 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2428out_free_request_sock_slab:
2429 if (prot->rsk_prot && prot->rsk_prot->slab) {
2430 kmem_cache_destroy(prot->rsk_prot->slab);
2431 prot->rsk_prot->slab = NULL;
2432 }
2e6599cb 2433out_free_request_sock_slab_name:
72150e9b
DC		 2434 if (prot->rsk_prot)
2435 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2436out_free_sock_slab:
2437 kmem_cache_destroy(prot->slab);
2438 prot->slab = NULL;
b733c007
PE		 2439out:
2440 return -ENOBUFS;
1da177e4 2441}
1da177e4
LT		 2442EXPORT_SYMBOL(proto_register);
2443
2444void proto_unregister(struct proto *prot)
2445{
2446 write_lock(&proto_list_lock);
13ff3d6f 2447 release_proto_idx(prot);
0a3f4358
PM		 2448 list_del(&prot->node);
2449 write_unlock(&proto_list_lock);
1da177e4
LT		 2450
2451 if (prot->slab != NULL) {
2452 kmem_cache_destroy(prot->slab);
2453 prot->slab = NULL;
2454 }
2455
2e6599cb 2456 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2457 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2458 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2459 prot->rsk_prot->slab = NULL;
2460 }
2461
6d6ee43e 2462 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2463 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2464 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2465 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2466 }
1da177e4 2467}
1da177e4
LT		 2468EXPORT_SYMBOL(proto_unregister);
2469
2470#ifdef CONFIG_PROC_FS
1da177e4 2471static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2472 __acquires(proto_list_lock)
1da177e4
LT		 2473{
2474 read_lock(&proto_list_lock);
60f0438a 2475 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2476}
2477
2478static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2479{
60f0438a 2480 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2481}
2482
2483static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2484 __releases(proto_list_lock)
1da177e4
LT		 2485{
2486 read_unlock(&proto_list_lock);
2487}
2488
2489static char proto_method_implemented(const void *method)
2490{
2491 return method == NULL ? 'n' : 'y';
2492}
2493
2494static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2495{
8d987e5c 2496 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2497 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2498 proto->name,
2499 proto->obj_size,
14e943db 2500 sock_prot_inuse_get(seq_file_net(seq), proto),
8d987e5c 2501 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
1da177e4
LT		 2502 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2503 proto->max_header,
2504 proto->slab == NULL ? "no" : "yes",
2505 module_name(proto->owner),
2506 proto_method_implemented(proto->close),
2507 proto_method_implemented(proto->connect),
2508 proto_method_implemented(proto->disconnect),
2509 proto_method_implemented(proto->accept),
2510 proto_method_implemented(proto->ioctl),
2511 proto_method_implemented(proto->init),
2512 proto_method_implemented(proto->destroy),
2513 proto_method_implemented(proto->shutdown),
2514 proto_method_implemented(proto->setsockopt),
2515 proto_method_implemented(proto->getsockopt),
2516 proto_method_implemented(proto->sendmsg),
2517 proto_method_implemented(proto->recvmsg),
2518 proto_method_implemented(proto->sendpage),
2519 proto_method_implemented(proto->bind),
2520 proto_method_implemented(proto->backlog_rcv),
2521 proto_method_implemented(proto->hash),
2522 proto_method_implemented(proto->unhash),
2523 proto_method_implemented(proto->get_port),
2524 proto_method_implemented(proto->enter_memory_pressure));
2525}
2526
2527static int proto_seq_show(struct seq_file *seq, void *v)
2528{
60f0438a 2529 if (v == &proto_list)
1da177e4
LT		 2530 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2531 "protocol",
2532 "size",
2533 "sockets",
2534 "memory",
2535 "press",
2536 "maxhdr",
2537 "slab",
2538 "module",
2539 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2540 else
60f0438a 2541 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2542 return 0;
2543}
2544
f690808e 2545static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2546 .start = proto_seq_start,
2547 .next = proto_seq_next,
2548 .stop = proto_seq_stop,
2549 .show = proto_seq_show,
2550};
2551
2552static int proto_seq_open(struct inode *inode, struct file *file)
2553{
14e943db
ED		 2554 return seq_open_net(inode, file, &proto_seq_ops,
2555 sizeof(struct seq_net_private));
1da177e4
LT		 2556}
2557
9a32144e 2558static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2559 .owner = THIS_MODULE,
2560 .open = proto_seq_open,
2561 .read = seq_read,
2562 .llseek = seq_lseek,
14e943db
ED		 2563 .release = seq_release_net,
2564};
2565
2566static __net_init int proto_init_net(struct net *net)
2567{
2568 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2569 return -ENOMEM;
2570
2571 return 0;
2572}
2573
2574static __net_exit void proto_exit_net(struct net *net)
2575{
2576 proc_net_remove(net, "protocols");
2577}
2578
2579
2580static __net_initdata struct pernet_operations proto_net_ops = {
2581 .init = proto_init_net,
2582 .exit = proto_exit_net,
1da177e4
LT		 2583};
2584
2585static int __init proto_init(void)
2586{
14e943db 2587 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2588}
2589
2590subsys_initcall(proto_init);
2591
2592#endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.776998 seconds and 5 git commands to generate.