projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
net: dont call jump_label_dec from irq context
[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
20d49473 279static void sock_disable_timestamp(struct sock *sk, int flag)
4ec93edb 280{
20d49473
PO		 281 if (sock_flag(sk, flag)) {
282 sock_reset_flag(sk, flag);
283 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
284 !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
285 net_disable_timestamp();
286 }
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,
692 SOCK_TIMESTAMPING_RX_SOFTWARE);
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
20d49473
PO		 1190 sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1191 sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1da177e4
LT		 1192
1193 if (atomic_read(&sk->sk_omem_alloc))
1194 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
0dc47877 1195 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1196
109f6e39
EB		 1197 if (sk->sk_peer_cred)
1198 put_cred(sk->sk_peer_cred);
1199 put_pid(sk->sk_peer_pid);
3b1e0a65 1200 put_net(sock_net(sk));
c308c1b2 1201 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1202}
2b85a34e
ED		 1203
1204void sk_free(struct sock *sk)
1205{
1206 /*
25985edc 1207 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1208 * some packets are still in some tx queue.
1209 * If not null, sock_wfree() will call __sk_free(sk) later
1210 */
1211 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1212 __sk_free(sk);
1213}
2a91525c 1214EXPORT_SYMBOL(sk_free);
1da177e4 1215
edf02087 1216/*
25985edc
LDM		 1217 * Last sock_put should drop reference to sk->sk_net. It has already
1218 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1219 * is not an option.
25985edc 1220 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1221 * destroy it in the context of init_net.
1222 */
1223void sk_release_kernel(struct sock *sk)
1224{
1225 if (sk == NULL || sk->sk_socket == NULL)
1226 return;
1227
1228 sock_hold(sk);
1229 sock_release(sk->sk_socket);
65a18ec5 1230 release_net(sock_net(sk));
3b1e0a65 1231 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1232 sock_put(sk);
1233}
45af1754 1234EXPORT_SYMBOL(sk_release_kernel);
edf02087 1235
e56c57d0
ED		 1236/**
1237 * sk_clone_lock - clone a socket, and lock its clone
1238 * @sk: the socket to clone
1239 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1240 *
1241 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1242 */
1243struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1244{
8fd1d178 1245 struct sock *newsk;
87d11ceb 1246
8fd1d178 1247 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1248 if (newsk != NULL) {
1249 struct sk_filter *filter;
1250
892c141e 1251 sock_copy(newsk, sk);
87d11ceb
ACM		 1252
1253 /* SANITY */
3b1e0a65 1254 get_net(sock_net(newsk));
87d11ceb
ACM		 1255 sk_node_init(&newsk->sk_node);
1256 sock_lock_init(newsk);
1257 bh_lock_sock(newsk);
fa438ccf 1258 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1259 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1260
1261 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1262 /*
1263 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1264 */
1265 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1266 atomic_set(&newsk->sk_omem_alloc, 0);
1267 skb_queue_head_init(&newsk->sk_receive_queue);
1268 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1269#ifdef CONFIG_NET_DMA
1270 skb_queue_head_init(&newsk->sk_async_wait_queue);
1271#endif
87d11ceb 1272
b6c6712a 1273 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1274 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1275 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1276 af_callback_keys + newsk->sk_family,
1277 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1278
1279 newsk->sk_dst_cache = NULL;
1280 newsk->sk_wmem_queued = 0;
1281 newsk->sk_forward_alloc = 0;
1282 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1283 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1284
1285 sock_reset_flag(newsk, SOCK_DONE);
1286 skb_queue_head_init(&newsk->sk_error_queue);
1287
0d7da9dd 1288 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1289 if (filter != NULL)
1290 sk_filter_charge(newsk, filter);
1291
1292 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1293 /* It is still raw copy of parent, so invalidate
1294 * destructor and make plain sk_free() */
1295 newsk->sk_destruct = NULL;
b0691c8e 1296 bh_unlock_sock(newsk);
87d11ceb
ACM		 1297 sk_free(newsk);
1298 newsk = NULL;
1299 goto out;
1300 }
1301
1302 newsk->sk_err = 0;
1303 newsk->sk_priority = 0;
4dc6dc71
ED		 1304 /*
1305 * Before updating sk_refcnt, we must commit prior changes to memory
1306 * (Documentation/RCU/rculist_nulls.txt for details)
1307 */
1308 smp_wmb();
87d11ceb
ACM		 1309 atomic_set(&newsk->sk_refcnt, 2);
1310
1311 /*
1312 * Increment the counter in the same struct proto as the master
1313 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1314 * is the same as sk->sk_prot->socks, as this field was copied
1315 * with memcpy).
1316 *
1317 * This _changes_ the previous behaviour, where
1318 * tcp_create_openreq_child always was incrementing the
1319 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1320 * to be taken into account in all callers. -acme
1321 */
1322 sk_refcnt_debug_inc(newsk);
972692e0 1323 sk_set_socket(newsk, NULL);
43815482 1324 newsk->sk_wq = NULL;
87d11ceb
ACM		 1325
1326 if (newsk->sk_prot->sockets_allocated)
1748376b 1327 percpu_counter_inc(newsk->sk_prot->sockets_allocated);
704da560
OP		 1328
1329 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1330 sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1331 net_enable_timestamp();
87d11ceb
ACM		 1332 }
1333out:
1334 return newsk;
1335}
e56c57d0 1336EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1337
9958089a
AK		 1338void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1339{
1340 __sk_dst_set(sk, dst);
1341 sk->sk_route_caps = dst->dev->features;
1342 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1343 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1344 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1345 if (sk_can_gso(sk)) {
82cc1a7a 1346 if (dst->header_len) {
9958089a 1347 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1348 } else {
9958089a 1349 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1350 sk->sk_gso_max_size = dst->dev->gso_max_size;
1351 }
9958089a
AK		 1352 }
1353}
1354EXPORT_SYMBOL_GPL(sk_setup_caps);
1355
1da177e4
LT		 1356void __init sk_init(void)
1357{
4481374c 1358 if (totalram_pages <= 4096) {
1da177e4
LT		 1359 sysctl_wmem_max = 32767;
1360 sysctl_rmem_max = 32767;
1361 sysctl_wmem_default = 32767;
1362 sysctl_rmem_default = 32767;
4481374c 1363 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1364 sysctl_wmem_max = 131071;
1365 sysctl_rmem_max = 131071;
1366 }
1367}
1368
1369/*
1370 * Simple resource managers for sockets.
1371 */
1372
1373
4ec93edb
YH		 1374/*
1375 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1376 */
1377void sock_wfree(struct sk_buff *skb)
1378{
1379 struct sock *sk = skb->sk;
d99927f4 1380 unsigned int len = skb->truesize;
1da177e4 1381
d99927f4
ED		 1382 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1383 /*
1384 * Keep a reference on sk_wmem_alloc, this will be released
1385 * after sk_write_space() call
1386 */
1387 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1388 sk->sk_write_space(sk);
d99927f4
ED		 1389 len = 1;
1390 }
2b85a34e 1391 /*
d99927f4
ED		 1392 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1393 * could not do because of in-flight packets
2b85a34e 1394 */
d99927f4 1395 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1396 __sk_free(sk);
1da177e4 1397}
2a91525c 1398EXPORT_SYMBOL(sock_wfree);
1da177e4 1399
4ec93edb
YH		 1400/*
1401 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1402 */
1403void sock_rfree(struct sk_buff *skb)
1404{
1405 struct sock *sk = skb->sk;
d361fd59 1406 unsigned int len = skb->truesize;
1da177e4 1407
d361fd59
ED		 1408 atomic_sub(len, &sk->sk_rmem_alloc);
1409 sk_mem_uncharge(sk, len);
1da177e4 1410}
2a91525c 1411EXPORT_SYMBOL(sock_rfree);
1da177e4
LT		 1412
1413
1414int sock_i_uid(struct sock *sk)
1415{
1416 int uid;
1417
f064af1e 1418 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1419 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1420 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1421 return uid;
1422}
2a91525c 1423EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1424
1425unsigned long sock_i_ino(struct sock *sk)
1426{
1427 unsigned long ino;
1428
f064af1e 1429 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1430 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1431 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1432 return ino;
1433}
2a91525c 1434EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1435
1436/*
1437 * Allocate a skb from the socket's send buffer.
1438 */
86a76caf 1439struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1440 gfp_t priority)
1da177e4
LT		 1441{
1442 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1443 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1444 if (skb) {
1445 skb_set_owner_w(skb, sk);
1446 return skb;
1447 }
1448 }
1449 return NULL;
1450}
2a91525c 1451EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1452
1453/*
1454 * Allocate a skb from the socket's receive buffer.
4ec93edb 1455 */
86a76caf 1456struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1457 gfp_t priority)
1da177e4
LT		 1458{
1459 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1460 struct sk_buff *skb = alloc_skb(size, priority);
1461 if (skb) {
1462 skb_set_owner_r(skb, sk);
1463 return skb;
1464 }
1465 }
1466 return NULL;
1467}
1468
4ec93edb 1469/*
1da177e4 1470 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1471 */
dd0fc66f 1472void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4
LT		 1473{
1474 if ((unsigned)size <= sysctl_optmem_max &&
1475 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1476 void *mem;
1477 /* First do the add, to avoid the race if kmalloc
4ec93edb 1478 * might sleep.
1da177e4
LT		 1479 */
1480 atomic_add(size, &sk->sk_omem_alloc);
1481 mem = kmalloc(size, priority);
1482 if (mem)
1483 return mem;
1484 atomic_sub(size, &sk->sk_omem_alloc);
1485 }
1486 return NULL;
1487}
2a91525c 1488EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1489
1490/*
1491 * Free an option memory block.
1492 */
1493void sock_kfree_s(struct sock *sk, void *mem, int size)
1494{
1495 kfree(mem);
1496 atomic_sub(size, &sk->sk_omem_alloc);
1497}
2a91525c 1498EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1499
1500/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1501 I think, these locks should be removed for datagram sockets.
1502 */
2a91525c 1503static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1504{
1505 DEFINE_WAIT(wait);
1506
1507 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1508 for (;;) {
1509 if (!timeo)
1510 break;
1511 if (signal_pending(current))
1512 break;
1513 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1514 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1515 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1516 break;
1517 if (sk->sk_shutdown & SEND_SHUTDOWN)
1518 break;
1519 if (sk->sk_err)
1520 break;
1521 timeo = schedule_timeout(timeo);
1522 }
aa395145 1523 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1524 return timeo;
1525}
1526
1527
1528/*
1529 * Generic send/receive buffer handlers
1530 */
1531
4cc7f68d
HX		 1532struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1533 unsigned long data_len, int noblock,
1534 int *errcode)
1da177e4
LT		 1535{
1536 struct sk_buff *skb;
7d877f3b 1537 gfp_t gfp_mask;
1da177e4
LT		 1538 long timeo;
1539 int err;
1540
1541 gfp_mask = sk->sk_allocation;
1542 if (gfp_mask & __GFP_WAIT)
1543 gfp_mask |= __GFP_REPEAT;
1544
1545 timeo = sock_sndtimeo(sk, noblock);
1546 while (1) {
1547 err = sock_error(sk);
1548 if (err != 0)
1549 goto failure;
1550
1551 err = -EPIPE;
1552 if (sk->sk_shutdown & SEND_SHUTDOWN)
1553 goto failure;
1554
1555 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1556 skb = alloc_skb(header_len, gfp_mask);
1da177e4
LT		 1557 if (skb) {
1558 int npages;
1559 int i;
1560
1561 /* No pages, we're done... */
1562 if (!data_len)
1563 break;
1564
1565 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1566 skb->truesize += data_len;
1567 skb_shinfo(skb)->nr_frags = npages;
1568 for (i = 0; i < npages; i++) {
1569 struct page *page;
1da177e4
LT		 1570
1571 page = alloc_pages(sk->sk_allocation, 0);
1572 if (!page) {
1573 err = -ENOBUFS;
1574 skb_shinfo(skb)->nr_frags = i;
1575 kfree_skb(skb);
1576 goto failure;
1577 }
1578
ea2ab693
IC		 1579 __skb_fill_page_desc(skb, i,
1580 page, 0,
1581 (data_len >= PAGE_SIZE ?
1582 PAGE_SIZE :
1583 data_len));
1da177e4
LT		 1584 data_len -= PAGE_SIZE;
1585 }
1586
1587 /* Full success... */
1588 break;
1589 }
1590 err = -ENOBUFS;
1591 goto failure;
1592 }
1593 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1594 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1595 err = -EAGAIN;
1596 if (!timeo)
1597 goto failure;
1598 if (signal_pending(current))
1599 goto interrupted;
1600 timeo = sock_wait_for_wmem(sk, timeo);
1601 }
1602
1603 skb_set_owner_w(skb, sk);
1604 return skb;
1605
1606interrupted:
1607 err = sock_intr_errno(timeo);
1608failure:
1609 *errcode = err;
1610 return NULL;
1611}
4cc7f68d 1612EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1613
4ec93edb 1614struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1615 int noblock, int *errcode)
1616{
1617 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1618}
2a91525c 1619EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1620
1621static void __lock_sock(struct sock *sk)
f39234d6
NK		 1622 __releases(&sk->sk_lock.slock)
1623 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1624{
1625 DEFINE_WAIT(wait);
1626
e71a4783 1627 for (;;) {
1da177e4
LT		 1628 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1629 TASK_UNINTERRUPTIBLE);
1630 spin_unlock_bh(&sk->sk_lock.slock);
1631 schedule();
1632 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1633 if (!sock_owned_by_user(sk))
1da177e4
LT		 1634 break;
1635 }
1636 finish_wait(&sk->sk_lock.wq, &wait);
1637}
1638
1639static void __release_sock(struct sock *sk)
f39234d6
NK		 1640 __releases(&sk->sk_lock.slock)
1641 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1642{
1643 struct sk_buff *skb = sk->sk_backlog.head;
1644
1645 do {
1646 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1647 bh_unlock_sock(sk);
1648
1649 do {
1650 struct sk_buff *next = skb->next;
1651
7fee226a 1652 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1653 skb->next = NULL;
c57943a1 1654 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1655
1656 /*
1657 * We are in process context here with softirqs
1658 * disabled, use cond_resched_softirq() to preempt.
1659 * This is safe to do because we've taken the backlog
1660 * queue private:
1661 */
1662 cond_resched_softirq();
1663
1664 skb = next;
1665 } while (skb != NULL);
1666
1667 bh_lock_sock(sk);
e71a4783 1668 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1669
1670 /*
1671 * Doing the zeroing here guarantee we can not loop forever
1672 * while a wild producer attempts to flood us.
1673 */
1674 sk->sk_backlog.len = 0;
1da177e4
LT		 1675}
1676
1677/**
1678 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1679 * @sk: sock to wait on
1680 * @timeo: for how long
1da177e4
LT		 1681 *
1682 * Now socket state including sk->sk_err is changed only under lock,
1683 * hence we may omit checks after joining wait queue.
1684 * We check receive queue before schedule() only as optimization;
1685 * it is very likely that release_sock() added new data.
1686 */
1687int sk_wait_data(struct sock *sk, long *timeo)
1688{
1689 int rc;
1690 DEFINE_WAIT(wait);
1691
aa395145 1692 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1693 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1694 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1695 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1696 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1697 return rc;
1698}
1da177e4
LT		 1699EXPORT_SYMBOL(sk_wait_data);
1700
3ab224be
HA		 1701/**
1702 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1703 * @sk: socket
1704 * @size: memory size to allocate
1705 * @kind: allocation type
1706 *
1707 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1708 * rmem allocation. This function assumes that protocols which have
1709 * memory_pressure use sk_wmem_queued as write buffer accounting.
1710 */
1711int __sk_mem_schedule(struct sock *sk, int size, int kind)
1712{
1713 struct proto *prot = sk->sk_prot;
1714 int amt = sk_mem_pages(size);
8d987e5c 1715 long allocated;
3ab224be
HA		 1716
1717 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
8d987e5c 1718 allocated = atomic_long_add_return(amt, prot->memory_allocated);
3ab224be
HA		 1719
1720 /* Under limit. */
1721 if (allocated <= prot->sysctl_mem[0]) {
1722 if (prot->memory_pressure && *prot->memory_pressure)
1723 *prot->memory_pressure = 0;
1724 return 1;
1725 }
1726
1727 /* Under pressure. */
1728 if (allocated > prot->sysctl_mem[1])
1729 if (prot->enter_memory_pressure)
5c52ba17 1730 prot->enter_memory_pressure(sk);
3ab224be
HA		 1731
1732 /* Over hard limit. */
1733 if (allocated > prot->sysctl_mem[2])
1734 goto suppress_allocation;
1735
1736 /* guarantee minimum buffer size under pressure */
1737 if (kind == SK_MEM_RECV) {
1738 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1739 return 1;
1740 } else { /* SK_MEM_SEND */
1741 if (sk->sk_type == SOCK_STREAM) {
1742 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1743 return 1;
1744 } else if (atomic_read(&sk->sk_wmem_alloc) <
1745 prot->sysctl_wmem[0])
1746 return 1;
1747 }
1748
1749 if (prot->memory_pressure) {
1748376b
ED		 1750 int alloc;
1751
1752 if (!*prot->memory_pressure)
1753 return 1;
1754 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1755 if (prot->sysctl_mem[2] > alloc *
3ab224be
HA		 1756 sk_mem_pages(sk->sk_wmem_queued +
1757 atomic_read(&sk->sk_rmem_alloc) +
1758 sk->sk_forward_alloc))
1759 return 1;
1760 }
1761
1762suppress_allocation:
1763
1764 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1765 sk_stream_moderate_sndbuf(sk);
1766
1767 /* Fail only if socket is _under_ its sndbuf.
1768 * In this case we cannot block, so that we have to fail.
1769 */
1770 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1771 return 1;
1772 }
1773
3847ce32
SM		 1774 trace_sock_exceed_buf_limit(sk, prot, allocated);
1775
3ab224be
HA		 1776 /* Alas. Undo changes. */
1777 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
8d987e5c 1778 atomic_long_sub(amt, prot->memory_allocated);
3ab224be
HA		 1779 return 0;
1780}
3ab224be
HA		 1781EXPORT_SYMBOL(__sk_mem_schedule);
1782
1783/**
1784 * __sk_reclaim - reclaim memory_allocated
1785 * @sk: socket
1786 */
1787void __sk_mem_reclaim(struct sock *sk)
1788{
1789 struct proto *prot = sk->sk_prot;
1790
8d987e5c 1791 atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
3ab224be
HA		 1792 prot->memory_allocated);
1793 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1794
1795 if (prot->memory_pressure && *prot->memory_pressure &&
8d987e5c 1796 (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
3ab224be
HA		 1797 *prot->memory_pressure = 0;
1798}
3ab224be
HA		 1799EXPORT_SYMBOL(__sk_mem_reclaim);
1800
1801
1da177e4
LT		 1802/*
1803 * Set of default routines for initialising struct proto_ops when
1804 * the protocol does not support a particular function. In certain
1805 * cases where it makes no sense for a protocol to have a "do nothing"
1806 * function, some default processing is provided.
1807 */
1808
1809int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1810{
1811 return -EOPNOTSUPP;
1812}
2a91525c 1813EXPORT_SYMBOL(sock_no_bind);
1da177e4 1814
4ec93edb 1815int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1816 int len, int flags)
1817{
1818 return -EOPNOTSUPP;
1819}
2a91525c 1820EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1821
1822int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1823{
1824 return -EOPNOTSUPP;
1825}
2a91525c 1826EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1827
1828int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1829{
1830 return -EOPNOTSUPP;
1831}
2a91525c 1832EXPORT_SYMBOL(sock_no_accept);
1da177e4 1833
4ec93edb 1834int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1835 int *len, int peer)
1836{
1837 return -EOPNOTSUPP;
1838}
2a91525c 1839EXPORT_SYMBOL(sock_no_getname);
1da177e4 1840
2a91525c 1841unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1842{
1843 return 0;
1844}
2a91525c 1845EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1846
1847int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1848{
1849 return -EOPNOTSUPP;
1850}
2a91525c 1851EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1852
1853int sock_no_listen(struct socket *sock, int backlog)
1854{
1855 return -EOPNOTSUPP;
1856}
2a91525c 1857EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1858
1859int sock_no_shutdown(struct socket *sock, int how)
1860{
1861 return -EOPNOTSUPP;
1862}
2a91525c 1863EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1864
1865int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1866 char __user *optval, unsigned int optlen)
1da177e4
LT		 1867{
1868 return -EOPNOTSUPP;
1869}
2a91525c 1870EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1871
1872int sock_no_getsockopt(struct socket *sock, int level, int optname,
1873 char __user *optval, int __user *optlen)
1874{
1875 return -EOPNOTSUPP;
1876}
2a91525c 1877EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1878
1879int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1880 size_t len)
1881{
1882 return -EOPNOTSUPP;
1883}
2a91525c 1884EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1885
1886int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1887 size_t len, int flags)
1888{
1889 return -EOPNOTSUPP;
1890}
2a91525c 1891EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 1892
1893int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1894{
1895 /* Mirror missing mmap method error code */
1896 return -ENODEV;
1897}
2a91525c 1898EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 1899
1900ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1901{
1902 ssize_t res;
1903 struct msghdr msg = {.msg_flags = flags};
1904 struct kvec iov;
1905 char *kaddr = kmap(page);
1906 iov.iov_base = kaddr + offset;
1907 iov.iov_len = size;
1908 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1909 kunmap(page);
1910 return res;
1911}
2a91525c 1912EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 1913
1914/*
1915 * Default Socket Callbacks
1916 */
1917
1918static void sock_def_wakeup(struct sock *sk)
1919{
43815482
ED		 1920 struct socket_wq *wq;
1921
1922 rcu_read_lock();
1923 wq = rcu_dereference(sk->sk_wq);
1924 if (wq_has_sleeper(wq))
1925 wake_up_interruptible_all(&wq->wait);
1926 rcu_read_unlock();
1da177e4
LT		 1927}
1928
1929static void sock_def_error_report(struct sock *sk)
1930{
43815482
ED		 1931 struct socket_wq *wq;
1932
1933 rcu_read_lock();
1934 wq = rcu_dereference(sk->sk_wq);
1935 if (wq_has_sleeper(wq))
1936 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 1937 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 1938 rcu_read_unlock();
1da177e4
LT		 1939}
1940
1941static void sock_def_readable(struct sock *sk, int len)
1942{
43815482
ED		 1943 struct socket_wq *wq;
1944
1945 rcu_read_lock();
1946 wq = rcu_dereference(sk->sk_wq);
1947 if (wq_has_sleeper(wq))
2c6607c6 1948 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 1949 POLLRDNORM | POLLRDBAND);
8d8ad9d7 1950 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 1951 rcu_read_unlock();
1da177e4
LT		 1952}
1953
1954static void sock_def_write_space(struct sock *sk)
1955{
43815482
ED		 1956 struct socket_wq *wq;
1957
1958 rcu_read_lock();
1da177e4
LT		 1959
1960 /* Do not wake up a writer until he can make "significant"
1961 * progress. --DaveM
1962 */
e71a4783 1963 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 1964 wq = rcu_dereference(sk->sk_wq);
1965 if (wq_has_sleeper(wq))
1966 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 1967 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 1968
1969 /* Should agree with poll, otherwise some programs break */
1970 if (sock_writeable(sk))
8d8ad9d7 1971 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 1972 }
1973
43815482 1974 rcu_read_unlock();
1da177e4
LT		 1975}
1976
1977static void sock_def_destruct(struct sock *sk)
1978{
a51482bd 1979 kfree(sk->sk_protinfo);
1da177e4
LT		 1980}
1981
1982void sk_send_sigurg(struct sock *sk)
1983{
1984 if (sk->sk_socket && sk->sk_socket->file)
1985 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 1986 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 1987}
2a91525c 1988EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 1989
1990void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1991 unsigned long expires)
1992{
1993 if (!mod_timer(timer, expires))
1994 sock_hold(sk);
1995}
1da177e4
LT		 1996EXPORT_SYMBOL(sk_reset_timer);
1997
1998void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1999{
2000 if (timer_pending(timer) && del_timer(timer))
2001 __sock_put(sk);
2002}
1da177e4
LT		 2003EXPORT_SYMBOL(sk_stop_timer);
2004
2005void sock_init_data(struct socket *sock, struct sock *sk)
2006{
2007 skb_queue_head_init(&sk->sk_receive_queue);
2008 skb_queue_head_init(&sk->sk_write_queue);
2009 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2010#ifdef CONFIG_NET_DMA
2011 skb_queue_head_init(&sk->sk_async_wait_queue);
2012#endif
1da177e4
LT		 2013
2014 sk->sk_send_head = NULL;
2015
2016 init_timer(&sk->sk_timer);
4ec93edb 2017
1da177e4
LT		 2018 sk->sk_allocation = GFP_KERNEL;
2019 sk->sk_rcvbuf = sysctl_rmem_default;
2020 sk->sk_sndbuf = sysctl_wmem_default;
2021 sk->sk_state = TCP_CLOSE;
972692e0 2022 sk_set_socket(sk, sock);
1da177e4
LT		 2023
2024 sock_set_flag(sk, SOCK_ZAPPED);
2025
e71a4783 2026 if (sock) {
1da177e4 2027 sk->sk_type = sock->type;
43815482 2028 sk->sk_wq = sock->wq;
1da177e4
LT		 2029 sock->sk = sk;
2030 } else
43815482 2031 sk->sk_wq = NULL;
1da177e4 2032
b6c6712a 2033 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2034 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2035 lockdep_set_class_and_name(&sk->sk_callback_lock,
2036 af_callback_keys + sk->sk_family,
2037 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2038
2039 sk->sk_state_change = sock_def_wakeup;
2040 sk->sk_data_ready = sock_def_readable;
2041 sk->sk_write_space = sock_def_write_space;
2042 sk->sk_error_report = sock_def_error_report;
2043 sk->sk_destruct = sock_def_destruct;
2044
2045 sk->sk_sndmsg_page = NULL;
2046 sk->sk_sndmsg_off = 0;
2047
109f6e39
EB		 2048 sk->sk_peer_pid = NULL;
2049 sk->sk_peer_cred = NULL;
1da177e4
LT		 2050 sk->sk_write_pending = 0;
2051 sk->sk_rcvlowat = 1;
2052 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2053 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2054
f37f0afb 2055 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2056
4dc6dc71
ED		 2057 /*
2058 * Before updating sk_refcnt, we must commit prior changes to memory
2059 * (Documentation/RCU/rculist_nulls.txt for details)
2060 */
2061 smp_wmb();
1da177e4 2062 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2063 atomic_set(&sk->sk_drops, 0);
1da177e4 2064}
2a91525c 2065EXPORT_SYMBOL(sock_init_data);
1da177e4 2066
b5606c2d 2067void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2068{
2069 might_sleep();
a5b5bb9a 2070 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2071 if (sk->sk_lock.owned)
1da177e4 2072 __lock_sock(sk);
d2e9117c 2073 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2074 spin_unlock(&sk->sk_lock.slock);
2075 /*
2076 * The sk_lock has mutex_lock() semantics here:
2077 */
fcc70d5f 2078 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2079 local_bh_enable();
1da177e4 2080}
fcc70d5f 2081EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2082
b5606c2d 2083void release_sock(struct sock *sk)
1da177e4 2084{
a5b5bb9a
IM		 2085 /*
2086 * The sk_lock has mutex_unlock() semantics:
2087 */
2088 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2089
2090 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2091 if (sk->sk_backlog.tail)
2092 __release_sock(sk);
d2e9117c 2093 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2094 if (waitqueue_active(&sk->sk_lock.wq))
2095 wake_up(&sk->sk_lock.wq);
2096 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2097}
2098EXPORT_SYMBOL(release_sock);
2099
8a74ad60
ED		 2100/**
2101 * lock_sock_fast - fast version of lock_sock
2102 * @sk: socket
2103 *
2104 * This version should be used for very small section, where process wont block
2105 * return false if fast path is taken
2106 * sk_lock.slock locked, owned = 0, BH disabled
2107 * return true if slow path is taken
2108 * sk_lock.slock unlocked, owned = 1, BH enabled
2109 */
2110bool lock_sock_fast(struct sock *sk)
2111{
2112 might_sleep();
2113 spin_lock_bh(&sk->sk_lock.slock);
2114
2115 if (!sk->sk_lock.owned)
2116 /*
2117 * Note : We must disable BH
2118 */
2119 return false;
2120
2121 __lock_sock(sk);
2122 sk->sk_lock.owned = 1;
2123 spin_unlock(&sk->sk_lock.slock);
2124 /*
2125 * The sk_lock has mutex_lock() semantics here:
2126 */
2127 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2128 local_bh_enable();
2129 return true;
2130}
2131EXPORT_SYMBOL(lock_sock_fast);
2132
1da177e4 2133int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2134{
b7aa0bf7 2135 struct timeval tv;
1da177e4 2136 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2137 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2138 tv = ktime_to_timeval(sk->sk_stamp);
2139 if (tv.tv_sec == -1)
1da177e4 2140 return -ENOENT;
b7aa0bf7
ED		 2141 if (tv.tv_sec == 0) {
2142 sk->sk_stamp = ktime_get_real();
2143 tv = ktime_to_timeval(sk->sk_stamp);
2144 }
2145 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2146}
1da177e4
LT		 2147EXPORT_SYMBOL(sock_get_timestamp);
2148
ae40eb1e
ED		 2149int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2150{
2151 struct timespec ts;
2152 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2153 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2154 ts = ktime_to_timespec(sk->sk_stamp);
2155 if (ts.tv_sec == -1)
2156 return -ENOENT;
2157 if (ts.tv_sec == 0) {
2158 sk->sk_stamp = ktime_get_real();
2159 ts = ktime_to_timespec(sk->sk_stamp);
2160 }
2161 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2162}
2163EXPORT_SYMBOL(sock_get_timestampns);
2164
20d49473 2165void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2166{
20d49473
PO		 2167 if (!sock_flag(sk, flag)) {
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 */
2174 if (!sock_flag(sk,
2175 flag == SOCK_TIMESTAMP ?
2176 SOCK_TIMESTAMPING_RX_SOFTWARE :
2177 SOCK_TIMESTAMP))
2178 net_enable_timestamp();
1da177e4
LT		 2179 }
2180}
1da177e4
LT		 2181
2182/*
2183 * Get a socket option on an socket.
2184 *
2185 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2186 * asynchronous errors should be reported by getsockopt. We assume
2187 * this means if you specify SO_ERROR (otherwise whats the point of it).
2188 */
2189int sock_common_getsockopt(struct socket *sock, int level, int optname,
2190 char __user *optval, int __user *optlen)
2191{
2192 struct sock *sk = sock->sk;
2193
2194 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2195}
1da177e4
LT		 2196EXPORT_SYMBOL(sock_common_getsockopt);
2197
3fdadf7d 2198#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2199int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2200 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2201{
2202 struct sock *sk = sock->sk;
2203
1e51f951 2204 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2205 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2206 optval, optlen);
3fdadf7d
DM		 2207 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2208}
2209EXPORT_SYMBOL(compat_sock_common_getsockopt);
2210#endif
2211
1da177e4
LT		 2212int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2213 struct msghdr *msg, size_t size, int flags)
2214{
2215 struct sock *sk = sock->sk;
2216 int addr_len = 0;
2217 int err;
2218
2219 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2220 flags & ~MSG_DONTWAIT, &addr_len);
2221 if (err >= 0)
2222 msg->msg_namelen = addr_len;
2223 return err;
2224}
1da177e4
LT		 2225EXPORT_SYMBOL(sock_common_recvmsg);
2226
2227/*
2228 * Set socket options on an inet socket.
2229 */
2230int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2231 char __user *optval, unsigned int optlen)
1da177e4
LT		 2232{
2233 struct sock *sk = sock->sk;
2234
2235 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2236}
1da177e4
LT		 2237EXPORT_SYMBOL(sock_common_setsockopt);
2238
3fdadf7d 2239#ifdef CONFIG_COMPAT
543d9cfe 2240int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2241 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2242{
2243 struct sock *sk = sock->sk;
2244
543d9cfe
ACM		 2245 if (sk->sk_prot->compat_setsockopt != NULL)
2246 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2247 optval, optlen);
3fdadf7d
DM		 2248 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2249}
2250EXPORT_SYMBOL(compat_sock_common_setsockopt);
2251#endif
2252
1da177e4
LT		 2253void sk_common_release(struct sock *sk)
2254{
2255 if (sk->sk_prot->destroy)
2256 sk->sk_prot->destroy(sk);
2257
2258 /*
2259 * Observation: when sock_common_release is called, processes have
2260 * no access to socket. But net still has.
2261 * Step one, detach it from networking:
2262 *
2263 * A. Remove from hash tables.
2264 */
2265
2266 sk->sk_prot->unhash(sk);
2267
2268 /*
2269 * In this point socket cannot receive new packets, but it is possible
2270 * that some packets are in flight because some CPU runs receiver and
2271 * did hash table lookup before we unhashed socket. They will achieve
2272 * receive queue and will be purged by socket destructor.
2273 *
2274 * Also we still have packets pending on receive queue and probably,
2275 * our own packets waiting in device queues. sock_destroy will drain
2276 * receive queue, but transmitted packets will delay socket destruction
2277 * until the last reference will be released.
2278 */
2279
2280 sock_orphan(sk);
2281
2282 xfrm_sk_free_policy(sk);
2283
e6848976 2284 sk_refcnt_debug_release(sk);
1da177e4
LT		 2285 sock_put(sk);
2286}
1da177e4
LT		 2287EXPORT_SYMBOL(sk_common_release);
2288
2289static DEFINE_RWLOCK(proto_list_lock);
2290static LIST_HEAD(proto_list);
2291
13ff3d6f
PE		 2292#ifdef CONFIG_PROC_FS
2293#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2294struct prot_inuse {
2295 int val[PROTO_INUSE_NR];
2296};
13ff3d6f
PE		 2297
2298static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2299
2300#ifdef CONFIG_NET_NS
2301void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2302{
d6d9ca0f 2303 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2304}
2305EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2306
2307int sock_prot_inuse_get(struct net *net, struct proto *prot)
2308{
2309 int cpu, idx = prot->inuse_idx;
2310 int res = 0;
2311
2312 for_each_possible_cpu(cpu)
2313 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2314
2315 return res >= 0 ? res : 0;
2316}
2317EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2318
2c8c1e72 2319static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2320{
2321 net->core.inuse = alloc_percpu(struct prot_inuse);
2322 return net->core.inuse ? 0 : -ENOMEM;
2323}
2324
2c8c1e72 2325static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2326{
2327 free_percpu(net->core.inuse);
2328}
2329
2330static struct pernet_operations net_inuse_ops = {
2331 .init = sock_inuse_init_net,
2332 .exit = sock_inuse_exit_net,
2333};
2334
2335static __init int net_inuse_init(void)
2336{
2337 if (register_pernet_subsys(&net_inuse_ops))
2338 panic("Cannot initialize net inuse counters");
2339
2340 return 0;
2341}
2342
2343core_initcall(net_inuse_init);
2344#else
1338d466
PE		 2345static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2346
c29a0bc4 2347void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2348{
d6d9ca0f 2349 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2350}
2351EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2352
c29a0bc4 2353int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2354{
2355 int cpu, idx = prot->inuse_idx;
2356 int res = 0;
2357
2358 for_each_possible_cpu(cpu)
2359 res += per_cpu(prot_inuse, cpu).val[idx];
2360
2361 return res >= 0 ? res : 0;
2362}
2363EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2364#endif
13ff3d6f
PE		 2365
2366static void assign_proto_idx(struct proto *prot)
2367{
2368 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2369
2370 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2371 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2372 return;
2373 }
2374
2375 set_bit(prot->inuse_idx, proto_inuse_idx);
2376}
2377
2378static void release_proto_idx(struct proto *prot)
2379{
2380 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2381 clear_bit(prot->inuse_idx, proto_inuse_idx);
2382}
2383#else
2384static inline void assign_proto_idx(struct proto *prot)
2385{
2386}
2387
2388static inline void release_proto_idx(struct proto *prot)
2389{
2390}
2391#endif
2392
b733c007
PE		 2393int proto_register(struct proto *prot, int alloc_slab)
2394{
1da177e4
LT		 2395 if (alloc_slab) {
2396 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2397 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2398 NULL);
1da177e4
LT		 2399
2400 if (prot->slab == NULL) {
2401 printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2402 prot->name);
60e7663d 2403 goto out;
1da177e4 2404 }
2e6599cb
ACM		 2405
2406 if (prot->rsk_prot != NULL) {
faf23422 2407 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2408 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2409 goto out_free_sock_slab;
2410
7e56b5d6 2411 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2412 prot->rsk_prot->obj_size, 0,
20c2df83 2413 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2414
2415 if (prot->rsk_prot->slab == NULL) {
2416 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2417 prot->name);
2418 goto out_free_request_sock_slab_name;
2419 }
2420 }
8feaf0c0 2421
6d6ee43e 2422 if (prot->twsk_prot != NULL) {
faf23422 2423 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2424
7e56b5d6 2425 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2426 goto out_free_request_sock_slab;
2427
6d6ee43e 2428 prot->twsk_prot->twsk_slab =
7e56b5d6 2429 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2430 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2431 0,
2432 SLAB_HWCACHE_ALIGN |
2433 prot->slab_flags,
20c2df83 2434 NULL);
6d6ee43e 2435 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2436 goto out_free_timewait_sock_slab_name;
2437 }
1da177e4
LT		 2438 }
2439
2a278051 2440 write_lock(&proto_list_lock);
1da177e4 2441 list_add(&prot->node, &proto_list);
13ff3d6f 2442 assign_proto_idx(prot);
1da177e4 2443 write_unlock(&proto_list_lock);
b733c007
PE		 2444 return 0;
2445
8feaf0c0 2446out_free_timewait_sock_slab_name:
7e56b5d6 2447 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2448out_free_request_sock_slab:
2449 if (prot->rsk_prot && prot->rsk_prot->slab) {
2450 kmem_cache_destroy(prot->rsk_prot->slab);
2451 prot->rsk_prot->slab = NULL;
2452 }
2e6599cb 2453out_free_request_sock_slab_name:
72150e9b
DC		 2454 if (prot->rsk_prot)
2455 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2456out_free_sock_slab:
2457 kmem_cache_destroy(prot->slab);
2458 prot->slab = NULL;
b733c007
PE		 2459out:
2460 return -ENOBUFS;
1da177e4 2461}
1da177e4
LT		 2462EXPORT_SYMBOL(proto_register);
2463
2464void proto_unregister(struct proto *prot)
2465{
2466 write_lock(&proto_list_lock);
13ff3d6f 2467 release_proto_idx(prot);
0a3f4358
PM		 2468 list_del(&prot->node);
2469 write_unlock(&proto_list_lock);
1da177e4
LT		 2470
2471 if (prot->slab != NULL) {
2472 kmem_cache_destroy(prot->slab);
2473 prot->slab = NULL;
2474 }
2475
2e6599cb 2476 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2477 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2478 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2479 prot->rsk_prot->slab = NULL;
2480 }
2481
6d6ee43e 2482 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2483 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2484 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2485 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2486 }
1da177e4 2487}
1da177e4
LT		 2488EXPORT_SYMBOL(proto_unregister);
2489
2490#ifdef CONFIG_PROC_FS
1da177e4 2491static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
9a429c49 2492 __acquires(proto_list_lock)
1da177e4
LT		 2493{
2494 read_lock(&proto_list_lock);
60f0438a 2495 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2496}
2497
2498static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2499{
60f0438a 2500 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2501}
2502
2503static void proto_seq_stop(struct seq_file *seq, void *v)
9a429c49 2504 __releases(proto_list_lock)
1da177e4
LT		 2505{
2506 read_unlock(&proto_list_lock);
2507}
2508
2509static char proto_method_implemented(const void *method)
2510{
2511 return method == NULL ? 'n' : 'y';
2512}
2513
2514static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2515{
8d987e5c 2516 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2517 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2518 proto->name,
2519 proto->obj_size,
14e943db 2520 sock_prot_inuse_get(seq_file_net(seq), proto),
8d987e5c 2521 proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
1da177e4
LT		 2522 proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2523 proto->max_header,
2524 proto->slab == NULL ? "no" : "yes",
2525 module_name(proto->owner),
2526 proto_method_implemented(proto->close),
2527 proto_method_implemented(proto->connect),
2528 proto_method_implemented(proto->disconnect),
2529 proto_method_implemented(proto->accept),
2530 proto_method_implemented(proto->ioctl),
2531 proto_method_implemented(proto->init),
2532 proto_method_implemented(proto->destroy),
2533 proto_method_implemented(proto->shutdown),
2534 proto_method_implemented(proto->setsockopt),
2535 proto_method_implemented(proto->getsockopt),
2536 proto_method_implemented(proto->sendmsg),
2537 proto_method_implemented(proto->recvmsg),
2538 proto_method_implemented(proto->sendpage),
2539 proto_method_implemented(proto->bind),
2540 proto_method_implemented(proto->backlog_rcv),
2541 proto_method_implemented(proto->hash),
2542 proto_method_implemented(proto->unhash),
2543 proto_method_implemented(proto->get_port),
2544 proto_method_implemented(proto->enter_memory_pressure));
2545}
2546
2547static int proto_seq_show(struct seq_file *seq, void *v)
2548{
60f0438a 2549 if (v == &proto_list)
1da177e4
LT		 2550 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2551 "protocol",
2552 "size",
2553 "sockets",
2554 "memory",
2555 "press",
2556 "maxhdr",
2557 "slab",
2558 "module",
2559 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2560 else
60f0438a 2561 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2562 return 0;
2563}
2564
f690808e 2565static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2566 .start = proto_seq_start,
2567 .next = proto_seq_next,
2568 .stop = proto_seq_stop,
2569 .show = proto_seq_show,
2570};
2571
2572static int proto_seq_open(struct inode *inode, struct file *file)
2573{
14e943db
ED		 2574 return seq_open_net(inode, file, &proto_seq_ops,
2575 sizeof(struct seq_net_private));
1da177e4
LT		 2576}
2577
9a32144e 2578static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2579 .owner = THIS_MODULE,
2580 .open = proto_seq_open,
2581 .read = seq_read,
2582 .llseek = seq_lseek,
14e943db
ED		 2583 .release = seq_release_net,
2584};
2585
2586static __net_init int proto_init_net(struct net *net)
2587{
2588 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2589 return -ENOMEM;
2590
2591 return 0;
2592}
2593
2594static __net_exit void proto_exit_net(struct net *net)
2595{
2596 proc_net_remove(net, "protocols");
2597}
2598
2599
2600static __net_initdata struct pernet_operations proto_net_ops = {
2601 .init = proto_init_net,
2602 .exit = proto_exit_net,
1da177e4
LT		 2603};
2604
2605static int __init proto_init(void)
2606{
14e943db 2607 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2608}
2609
2610subsys_initcall(proto_init);
2611
2612#endif /* PROC_FS */
Linux kernel - Deliverables
This page took 0.798654 seconds and 5 git commands to generate.