projects / deliverable / linux.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mm: account for the number of times direct reclaimers get throttled
[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
e005d193
JP		 92#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93
4fc268d2 94#include <linux/capability.h>
1da177e4
LT		 95#include <linux/errno.h>
96#include <linux/types.h>
97#include <linux/socket.h>
98#include <linux/in.h>
99#include <linux/kernel.h>
1da177e4
LT		 100#include <linux/module.h>
101#include <linux/proc_fs.h>
102#include <linux/seq_file.h>
103#include <linux/sched.h>
104#include <linux/timer.h>
105#include <linux/string.h>
106#include <linux/sockios.h>
107#include <linux/net.h>
108#include <linux/mm.h>
109#include <linux/slab.h>
110#include <linux/interrupt.h>
111#include <linux/poll.h>
112#include <linux/tcp.h>
113#include <linux/init.h>
a1f8e7f7 114#include <linux/highmem.h>
3f551f94 115#include <linux/user_namespace.h>
c5905afb 116#include <linux/static_key.h>
3969eb38 117#include <linux/memcontrol.h>
8c1ae10d 118#include <linux/prefetch.h>
1da177e4
LT		 119
120#include <asm/uaccess.h>
1da177e4
LT		 121
122#include <linux/netdevice.h>
123#include <net/protocol.h>
124#include <linux/skbuff.h>
457c4cbc 125#include <net/net_namespace.h>
2e6599cb 126#include <net/request_sock.h>
1da177e4 127#include <net/sock.h>
20d49473 128#include <linux/net_tstamp.h>
1da177e4
LT		 129#include <net/xfrm.h>
130#include <linux/ipsec.h>
f8451725 131#include <net/cls_cgroup.h>
5bc1421e 132#include <net/netprio_cgroup.h>
1da177e4
LT		 133
134#include <linux/filter.h>
135
3847ce32
SM		 136#include <trace/events/sock.h>
137
1da177e4
LT		 138#ifdef CONFIG_INET
139#include <net/tcp.h>
140#endif
141
36b77a52 142static DEFINE_MUTEX(proto_list_mutex);
d1a4c0b3
GC		 143static LIST_HEAD(proto_list);
144
c255a458 145#ifdef CONFIG_MEMCG_KMEM
1d62e436 146int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
d1a4c0b3
GC		 147{
148 struct proto *proto;
149 int ret = 0;
150
36b77a52 151 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 152 list_for_each_entry(proto, &proto_list, node) {
153 if (proto->init_cgroup) {
1d62e436 154 ret = proto->init_cgroup(memcg, ss);
d1a4c0b3
GC		 155 if (ret)
156 goto out;
157 }
158 }
159
36b77a52 160 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 161 return ret;
162out:
163 list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 if (proto->destroy_cgroup)
1d62e436 165 proto->destroy_cgroup(memcg);
36b77a52 166 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 167 return ret;
168}
169
1d62e436 170void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
d1a4c0b3
GC		 171{
172 struct proto *proto;
173
36b77a52 174 mutex_lock(&proto_list_mutex);
d1a4c0b3
GC		 175 list_for_each_entry_reverse(proto, &proto_list, node)
176 if (proto->destroy_cgroup)
1d62e436 177 proto->destroy_cgroup(memcg);
36b77a52 178 mutex_unlock(&proto_list_mutex);
d1a4c0b3
GC		 179}
180#endif
181
da21f24d
IM		 182/*
183 * Each address family might have different locking rules, so we have
184 * one slock key per address family:
185 */
a5b5bb9a
IM		 186static struct lock_class_key af_family_keys[AF_MAX];
187static struct lock_class_key af_family_slock_keys[AF_MAX];
188
c5905afb 189struct static_key memcg_socket_limit_enabled;
e1aab161
GC		 190EXPORT_SYMBOL(memcg_socket_limit_enabled);
191
a5b5bb9a
IM		 192/*
193 * Make lock validator output more readable. (we pre-construct these
194 * strings build-time, so that runtime initialization of socket
195 * locks is fast):
196 */
36cbd3dc 197static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 198 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
199 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
200 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
201 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
202 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
203 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
204 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 205 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 206 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 207 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 208 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 209 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 210 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
c7fe3b52 211 "sk_lock-AF_NFC" , "sk_lock-AF_MAX"
a5b5bb9a 212};
36cbd3dc 213static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 214 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
215 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
216 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
217 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
218 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
219 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
220 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 221 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 222 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 223 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 224 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 225 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 226 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
c7fe3b52 227 "slock-AF_NFC" , "slock-AF_MAX"
a5b5bb9a 228};
36cbd3dc 229static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 230 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
231 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
232 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
233 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
234 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
235 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
236 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 237 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 238 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 239 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 240 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 241 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 242 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
c7fe3b52 243 "clock-AF_NFC" , "clock-AF_MAX"
443aef0e 244};
da21f24d
IM		 245
246/*
247 * sk_callback_lock locking rules are per-address-family,
248 * so split the lock classes by using a per-AF key:
249 */
250static struct lock_class_key af_callback_keys[AF_MAX];
251
1da177e4
LT		 252/* Take into consideration the size of the struct sk_buff overhead in the
253 * determination of these values, since that is non-constant across
254 * platforms. This makes socket queueing behavior and performance
255 * not depend upon such differences.
256 */
257#define _SK_MEM_PACKETS 256
87fb4b7b 258#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 259#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
261
262/* Run time adjustable parameters. */
ab32ea5d 263__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
6d8ebc8a 264EXPORT_SYMBOL(sysctl_wmem_max);
ab32ea5d 265__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
6d8ebc8a 266EXPORT_SYMBOL(sysctl_rmem_max);
ab32ea5d
BH		 267__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 269
25985edc 270/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 271int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 272EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 273
c93bdd0e
MG		 274struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
275EXPORT_SYMBOL_GPL(memalloc_socks);
276
7cb02404
MG		 277/**
278 * sk_set_memalloc - sets %SOCK_MEMALLOC
279 * @sk: socket to set it on
280 *
281 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
282 * It's the responsibility of the admin to adjust min_free_kbytes
283 * to meet the requirements
284 */
285void sk_set_memalloc(struct sock *sk)
286{
287 sock_set_flag(sk, SOCK_MEMALLOC);
288 sk->sk_allocation |= __GFP_MEMALLOC;
c93bdd0e 289 static_key_slow_inc(&memalloc_socks);
7cb02404
MG		 290}
291EXPORT_SYMBOL_GPL(sk_set_memalloc);
292
293void sk_clear_memalloc(struct sock *sk)
294{
295 sock_reset_flag(sk, SOCK_MEMALLOC);
296 sk->sk_allocation &= ~__GFP_MEMALLOC;
c93bdd0e 297 static_key_slow_dec(&memalloc_socks);
7cb02404
MG		 298}
299EXPORT_SYMBOL_GPL(sk_clear_memalloc);
300
b4b9e355
MG		 301int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
302{
303 int ret;
304 unsigned long pflags = current->flags;
305
306 /* these should have been dropped before queueing */
307 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
308
309 current->flags |= PF_MEMALLOC;
310 ret = sk->sk_backlog_rcv(sk, skb);
311 tsk_restore_flags(current, pflags, PF_MEMALLOC);
312
313 return ret;
314}
315EXPORT_SYMBOL(__sk_backlog_rcv);
316
5bc1421e
NH		 317#if defined(CONFIG_CGROUPS)
318#if !defined(CONFIG_NET_CLS_CGROUP)
f8451725
HX		 319int net_cls_subsys_id = -1;
320EXPORT_SYMBOL_GPL(net_cls_subsys_id);
321#endif
5bc1421e
NH		 322#if !defined(CONFIG_NETPRIO_CGROUP)
323int net_prio_subsys_id = -1;
324EXPORT_SYMBOL_GPL(net_prio_subsys_id);
325#endif
326#endif
f8451725 327
1da177e4
LT		 328static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
329{
330 struct timeval tv;
331
332 if (optlen < sizeof(tv))
333 return -EINVAL;
334 if (copy_from_user(&tv, optval, sizeof(tv)))
335 return -EFAULT;
ba78073e
VA		 336 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
337 return -EDOM;
1da177e4 338
ba78073e 339 if (tv.tv_sec < 0) {
6f11df83
AM		 340 static int warned __read_mostly;
341
ba78073e 342 *timeo_p = 0;
50aab54f 343 if (warned < 10 && net_ratelimit()) {
ba78073e 344 warned++;
e005d193
JP		 345 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
346 __func__, current->comm, task_pid_nr(current));
50aab54f 347 }
ba78073e
VA		 348 return 0;
349 }
1da177e4
LT		 350 *timeo_p = MAX_SCHEDULE_TIMEOUT;
351 if (tv.tv_sec == 0 && tv.tv_usec == 0)
352 return 0;
353 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
354 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
355 return 0;
356}
357
358static void sock_warn_obsolete_bsdism(const char *name)
359{
360 static int warned;
361 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 362 if (strcmp(warncomm, current->comm) && warned < 5) {
363 strcpy(warncomm, current->comm);
e005d193
JP		 364 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
365 warncomm, name);
1da177e4
LT		 366 warned++;
367 }
368}
369
08e29af3
ED		 370#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
371
372static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 373{
08e29af3
ED		 374 if (sk->sk_flags & flags) {
375 sk->sk_flags &= ~flags;
376 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 377 net_disable_timestamp();
1da177e4
LT		 378 }
379}
380
381
f0088a50
DV		 382int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
383{
766e9037 384 int err;
f0088a50 385 int skb_len;
3b885787
NH		 386 unsigned long flags;
387 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 388
0fd7bac6 389 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
766e9037 390 atomic_inc(&sk->sk_drops);
3847ce32 391 trace_sock_rcvqueue_full(sk, skb);
766e9037 392 return -ENOMEM;
f0088a50
DV		 393 }
394
fda9ef5d 395 err = sk_filter(sk, skb);
f0088a50 396 if (err)
766e9037 397 return err;
f0088a50 398
3ab224be 399 if (!sk_rmem_schedule(sk, skb->truesize)) {
766e9037
ED		 400 atomic_inc(&sk->sk_drops);
401 return -ENOBUFS;
3ab224be
HA		 402 }
403
f0088a50
DV		 404 skb->dev = NULL;
405 skb_set_owner_r(skb, sk);
49ad9599 406
f0088a50
DV		 407 /* Cache the SKB length before we tack it onto the receive
408 * queue. Once it is added it no longer belongs to us and
409 * may be freed by other threads of control pulling packets
410 * from the queue.
411 */
412 skb_len = skb->len;
413
7fee226a
ED		 414 /* we escape from rcu protected region, make sure we dont leak
415 * a norefcounted dst
416 */
417 skb_dst_force(skb);
418
3b885787
NH		 419 spin_lock_irqsave(&list->lock, flags);
420 skb->dropcount = atomic_read(&sk->sk_drops);
421 __skb_queue_tail(list, skb);
422 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 423
424 if (!sock_flag(sk, SOCK_DEAD))
425 sk->sk_data_ready(sk, skb_len);
766e9037 426 return 0;
f0088a50
DV		 427}
428EXPORT_SYMBOL(sock_queue_rcv_skb);
429
58a5a7b9 430int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 431{
432 int rc = NET_RX_SUCCESS;
433
fda9ef5d 434 if (sk_filter(sk, skb))
f0088a50
DV		 435 goto discard_and_relse;
436
437 skb->dev = NULL;
438
f545a38f 439 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
c377411f
ED		 440 atomic_inc(&sk->sk_drops);
441 goto discard_and_relse;
442 }
58a5a7b9
ACM		 443 if (nested)
444 bh_lock_sock_nested(sk);
445 else
446 bh_lock_sock(sk);
a5b5bb9a
IM		 447 if (!sock_owned_by_user(sk)) {
448 /*
449 * trylock + unlock semantics:
450 */
451 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
452
c57943a1 453 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 454
455 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
f545a38f 456 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
8eae939f
ZY		 457 bh_unlock_sock(sk);
458 atomic_inc(&sk->sk_drops);
459 goto discard_and_relse;
460 }
461
f0088a50
DV		 462 bh_unlock_sock(sk);
463out:
464 sock_put(sk);
465 return rc;
466discard_and_relse:
467 kfree_skb(skb);
468 goto out;
469}
470EXPORT_SYMBOL(sk_receive_skb);
471
ea94ff3b
KK		 472void sk_reset_txq(struct sock *sk)
473{
474 sk_tx_queue_clear(sk);
475}
476EXPORT_SYMBOL(sk_reset_txq);
477
f0088a50
DV		 478struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
479{
b6c6712a 480 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 481
482 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 483 sk_tx_queue_clear(sk);
a9b3cd7f 484 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 485 dst_release(dst);
486 return NULL;
487 }
488
489 return dst;
490}
491EXPORT_SYMBOL(__sk_dst_check);
492
493struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
494{
495 struct dst_entry *dst = sk_dst_get(sk);
496
497 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
498 sk_dst_reset(sk);
499 dst_release(dst);
500 return NULL;
501 }
502
503 return dst;
504}
505EXPORT_SYMBOL(sk_dst_check);
506
4878809f
DM		 507static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
508{
509 int ret = -ENOPROTOOPT;
510#ifdef CONFIG_NETDEVICES
3b1e0a65 511 struct net *net = sock_net(sk);
4878809f
DM		 512 char devname[IFNAMSIZ];
513 int index;
514
515 /* Sorry... */
516 ret = -EPERM;
517 if (!capable(CAP_NET_RAW))
518 goto out;
519
520 ret = -EINVAL;
521 if (optlen < 0)
522 goto out;
523
524 /* Bind this socket to a particular device like "eth0",
525 * as specified in the passed interface name. If the
526 * name is "" or the option length is zero the socket
527 * is not bound.
528 */
529 if (optlen > IFNAMSIZ - 1)
530 optlen = IFNAMSIZ - 1;
531 memset(devname, 0, sizeof(devname));
532
533 ret = -EFAULT;
534 if (copy_from_user(devname, optval, optlen))
535 goto out;
536
000ba2e4
DM		 537 index = 0;
538 if (devname[0] != '\0') {
bf8e56bf 539 struct net_device *dev;
4878809f 540
bf8e56bf
ED		 541 rcu_read_lock();
542 dev = dev_get_by_name_rcu(net, devname);
543 if (dev)
544 index = dev->ifindex;
545 rcu_read_unlock();
4878809f
DM		 546 ret = -ENODEV;
547 if (!dev)
548 goto out;
4878809f
DM		 549 }
550
551 lock_sock(sk);
552 sk->sk_bound_dev_if = index;
553 sk_dst_reset(sk);
554 release_sock(sk);
555
556 ret = 0;
557
558out:
559#endif
560
561 return ret;
562}
563
c0ef877b
PE		 564static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
565{
566 if (valbool)
567 sock_set_flag(sk, bit);
568 else
569 sock_reset_flag(sk, bit);
570}
571
1da177e4
LT		 572/*
573 * This is meant for all protocols to use and covers goings on
574 * at the socket level. Everything here is generic.
575 */
576
577int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 578 char __user *optval, unsigned int optlen)
1da177e4 579{
2a91525c 580 struct sock *sk = sock->sk;
1da177e4
LT		 581 int val;
582 int valbool;
583 struct linger ling;
584 int ret = 0;
4ec93edb 585
1da177e4
LT		 586 /*
587 * Options without arguments
588 */
589
4878809f
DM		 590 if (optname == SO_BINDTODEVICE)
591 return sock_bindtodevice(sk, optval, optlen);
592
e71a4783
SH		 593 if (optlen < sizeof(int))
594 return -EINVAL;
4ec93edb 595
1da177e4
LT		 596 if (get_user(val, (int __user *)optval))
597 return -EFAULT;
4ec93edb 598
2a91525c 599 valbool = val ? 1 : 0;
1da177e4
LT		 600
601 lock_sock(sk);
602
2a91525c 603 switch (optname) {
e71a4783 604 case SO_DEBUG:
2a91525c 605 if (val && !capable(CAP_NET_ADMIN))
e71a4783 606 ret = -EACCES;
2a91525c 607 else
c0ef877b 608 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 609 break;
610 case SO_REUSEADDR:
4a17fd52 611 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
e71a4783
SH		 612 break;
613 case SO_TYPE:
49c794e9 614 case SO_PROTOCOL:
0d6038ee 615 case SO_DOMAIN:
e71a4783
SH		 616 case SO_ERROR:
617 ret = -ENOPROTOOPT;
618 break;
619 case SO_DONTROUTE:
c0ef877b 620 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 621 break;
622 case SO_BROADCAST:
623 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
624 break;
625 case SO_SNDBUF:
626 /* Don't error on this BSD doesn't and if you think
82981930
ED		 627 * about it this is right. Otherwise apps have to
628 * play 'guess the biggest size' games. RCVBUF/SNDBUF
629 * are treated in BSD as hints
630 */
631 val = min_t(u32, val, sysctl_wmem_max);
b0573dea 632set_sndbuf:
e71a4783 633 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
82981930
ED		 634 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
635 /* Wake up sending tasks if we upped the value. */
e71a4783
SH		 636 sk->sk_write_space(sk);
637 break;
1da177e4 638
e71a4783
SH		 639 case SO_SNDBUFFORCE:
640 if (!capable(CAP_NET_ADMIN)) {
641 ret = -EPERM;
642 break;
643 }
644 goto set_sndbuf;
b0573dea 645
e71a4783
SH		 646 case SO_RCVBUF:
647 /* Don't error on this BSD doesn't and if you think
82981930
ED		 648 * about it this is right. Otherwise apps have to
649 * play 'guess the biggest size' games. RCVBUF/SNDBUF
650 * are treated in BSD as hints
651 */
652 val = min_t(u32, val, sysctl_rmem_max);
b0573dea 653set_rcvbuf:
e71a4783
SH		 654 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
655 /*
656 * We double it on the way in to account for
657 * "struct sk_buff" etc. overhead. Applications
658 * assume that the SO_RCVBUF setting they make will
659 * allow that much actual data to be received on that
660 * socket.
661 *
662 * Applications are unaware that "struct sk_buff" and
663 * other overheads allocate from the receive buffer
664 * during socket buffer allocation.
665 *
666 * And after considering the possible alternatives,
667 * returning the value we actually used in getsockopt
668 * is the most desirable behavior.
669 */
82981930 670 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
e71a4783
SH		 671 break;
672
673 case SO_RCVBUFFORCE:
674 if (!capable(CAP_NET_ADMIN)) {
675 ret = -EPERM;
1da177e4 676 break;
e71a4783
SH		 677 }
678 goto set_rcvbuf;
1da177e4 679
e71a4783 680 case SO_KEEPALIVE:
1da177e4 681#ifdef CONFIG_INET
e71a4783
SH		 682 if (sk->sk_protocol == IPPROTO_TCP)
683 tcp_set_keepalive(sk, valbool);
1da177e4 684#endif
e71a4783
SH		 685 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
686 break;
687
688 case SO_OOBINLINE:
689 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
690 break;
691
692 case SO_NO_CHECK:
693 sk->sk_no_check = valbool;
694 break;
695
696 case SO_PRIORITY:
697 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
698 sk->sk_priority = val;
699 else
700 ret = -EPERM;
701 break;
702
703 case SO_LINGER:
704 if (optlen < sizeof(ling)) {
705 ret = -EINVAL; /* 1003.1g */
1da177e4 706 break;
e71a4783 707 }
2a91525c 708 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 709 ret = -EFAULT;
1da177e4 710 break;
e71a4783
SH		 711 }
712 if (!ling.l_onoff)
713 sock_reset_flag(sk, SOCK_LINGER);
714 else {
1da177e4 715#if (BITS_PER_LONG == 32)
e71a4783
SH		 716 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
717 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 718 else
e71a4783
SH		 719#endif
720 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
721 sock_set_flag(sk, SOCK_LINGER);
722 }
723 break;
724
725 case SO_BSDCOMPAT:
726 sock_warn_obsolete_bsdism("setsockopt");
727 break;
728
729 case SO_PASSCRED:
730 if (valbool)
731 set_bit(SOCK_PASSCRED, &sock->flags);
732 else
733 clear_bit(SOCK_PASSCRED, &sock->flags);
734 break;
735
736 case SO_TIMESTAMP:
92f37fd2 737 case SO_TIMESTAMPNS:
e71a4783 738 if (valbool) {
92f37fd2
ED		 739 if (optname == SO_TIMESTAMP)
740 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
741 else
742 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 743 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 744 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 745 } else {
e71a4783 746 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 747 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
748 }
e71a4783
SH		 749 break;
750
20d49473
PO		 751 case SO_TIMESTAMPING:
752 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 753 ret = -EINVAL;
20d49473
PO		 754 break;
755 }
756 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
757 val & SOF_TIMESTAMPING_TX_HARDWARE);
758 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
759 val & SOF_TIMESTAMPING_TX_SOFTWARE);
760 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
761 val & SOF_TIMESTAMPING_RX_HARDWARE);
762 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
763 sock_enable_timestamp(sk,
764 SOCK_TIMESTAMPING_RX_SOFTWARE);
765 else
766 sock_disable_timestamp(sk,
08e29af3 767 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
20d49473
PO		 768 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
769 val & SOF_TIMESTAMPING_SOFTWARE);
770 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
771 val & SOF_TIMESTAMPING_SYS_HARDWARE);
772 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
773 val & SOF_TIMESTAMPING_RAW_HARDWARE);
774 break;
775
e71a4783
SH		 776 case SO_RCVLOWAT:
777 if (val < 0)
778 val = INT_MAX;
779 sk->sk_rcvlowat = val ? : 1;
780 break;
781
782 case SO_RCVTIMEO:
783 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
784 break;
785
786 case SO_SNDTIMEO:
787 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
788 break;
1da177e4 789
e71a4783
SH		 790 case SO_ATTACH_FILTER:
791 ret = -EINVAL;
792 if (optlen == sizeof(struct sock_fprog)) {
793 struct sock_fprog fprog;
1da177e4 794
e71a4783
SH		 795 ret = -EFAULT;
796 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 797 break;
e71a4783
SH		 798
799 ret = sk_attach_filter(&fprog, sk);
800 }
801 break;
802
803 case SO_DETACH_FILTER:
55b33325 804 ret = sk_detach_filter(sk);
e71a4783 805 break;
1da177e4 806
e71a4783
SH		 807 case SO_PASSSEC:
808 if (valbool)
809 set_bit(SOCK_PASSSEC, &sock->flags);
810 else
811 clear_bit(SOCK_PASSSEC, &sock->flags);
812 break;
4a19ec58
LAT		 813 case SO_MARK:
814 if (!capable(CAP_NET_ADMIN))
815 ret = -EPERM;
2a91525c 816 else
4a19ec58 817 sk->sk_mark = val;
4a19ec58 818 break;
877ce7c1 819
1da177e4
LT		 820 /* We implement the SO_SNDLOWAT etc to
821 not be settable (1003.1g 5.3) */
3b885787 822 case SO_RXQ_OVFL:
8083f0fc 823 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 824 break;
6e3e939f
JB		 825
826 case SO_WIFI_STATUS:
827 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
828 break;
829
ef64a54f
PE		 830 case SO_PEEK_OFF:
831 if (sock->ops->set_peek_off)
832 sock->ops->set_peek_off(sk, val);
833 else
834 ret = -EOPNOTSUPP;
835 break;
3bdc0eba
BG		 836
837 case SO_NOFCS:
838 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
839 break;
840
e71a4783
SH		 841 default:
842 ret = -ENOPROTOOPT;
843 break;
4ec93edb 844 }
1da177e4
LT		 845 release_sock(sk);
846 return ret;
847}
2a91525c 848EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 849
850
3f551f94
EB		 851void cred_to_ucred(struct pid *pid, const struct cred *cred,
852 struct ucred *ucred)
853{
854 ucred->pid = pid_vnr(pid);
855 ucred->uid = ucred->gid = -1;
856 if (cred) {
857 struct user_namespace *current_ns = current_user_ns();
858
76b6db01
EB		 859 ucred->uid = from_kuid(current_ns, cred->euid);
860 ucred->gid = from_kgid(current_ns, cred->egid);
3f551f94
EB		 861 }
862}
3924773a 863EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 864
1da177e4
LT		 865int sock_getsockopt(struct socket *sock, int level, int optname,
866 char __user *optval, int __user *optlen)
867{
868 struct sock *sk = sock->sk;
4ec93edb 869
e71a4783 870 union {
4ec93edb
YH		 871 int val;
872 struct linger ling;
1da177e4
LT		 873 struct timeval tm;
874 } v;
4ec93edb 875
4d0392be 876 int lv = sizeof(int);
1da177e4 877 int len;
4ec93edb 878
e71a4783 879 if (get_user(len, optlen))
4ec93edb 880 return -EFAULT;
e71a4783 881 if (len < 0)
1da177e4 882 return -EINVAL;
4ec93edb 883
50fee1de 884 memset(&v, 0, sizeof(v));
df0bca04 885
2a91525c 886 switch (optname) {
e71a4783
SH		 887 case SO_DEBUG:
888 v.val = sock_flag(sk, SOCK_DBG);
889 break;
890
891 case SO_DONTROUTE:
892 v.val = sock_flag(sk, SOCK_LOCALROUTE);
893 break;
894
895 case SO_BROADCAST:
1b23a5df 896 v.val = sock_flag(sk, SOCK_BROADCAST);
e71a4783
SH		 897 break;
898
899 case SO_SNDBUF:
900 v.val = sk->sk_sndbuf;
901 break;
902
903 case SO_RCVBUF:
904 v.val = sk->sk_rcvbuf;
905 break;
906
907 case SO_REUSEADDR:
908 v.val = sk->sk_reuse;
909 break;
910
911 case SO_KEEPALIVE:
1b23a5df 912 v.val = sock_flag(sk, SOCK_KEEPOPEN);
e71a4783
SH		 913 break;
914
915 case SO_TYPE:
916 v.val = sk->sk_type;
917 break;
918
49c794e9
JE		 919 case SO_PROTOCOL:
920 v.val = sk->sk_protocol;
921 break;
922
0d6038ee
JE		 923 case SO_DOMAIN:
924 v.val = sk->sk_family;
925 break;
926
e71a4783
SH		 927 case SO_ERROR:
928 v.val = -sock_error(sk);
2a91525c 929 if (v.val == 0)
e71a4783
SH		 930 v.val = xchg(&sk->sk_err_soft, 0);
931 break;
932
933 case SO_OOBINLINE:
1b23a5df 934 v.val = sock_flag(sk, SOCK_URGINLINE);
e71a4783
SH		 935 break;
936
937 case SO_NO_CHECK:
938 v.val = sk->sk_no_check;
939 break;
940
941 case SO_PRIORITY:
942 v.val = sk->sk_priority;
943 break;
944
945 case SO_LINGER:
946 lv = sizeof(v.ling);
1b23a5df 947 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
e71a4783
SH		 948 v.ling.l_linger = sk->sk_lingertime / HZ;
949 break;
950
951 case SO_BSDCOMPAT:
952 sock_warn_obsolete_bsdism("getsockopt");
953 break;
954
955 case SO_TIMESTAMP:
92f37fd2
ED		 956 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
957 !sock_flag(sk, SOCK_RCVTSTAMPNS);
958 break;
959
960 case SO_TIMESTAMPNS:
961 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 962 break;
963
20d49473
PO		 964 case SO_TIMESTAMPING:
965 v.val = 0;
966 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
967 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
968 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
969 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
970 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
971 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
972 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
973 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
974 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
975 v.val |= SOF_TIMESTAMPING_SOFTWARE;
976 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
977 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
978 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
979 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
980 break;
981
e71a4783 982 case SO_RCVTIMEO:
2a91525c 983 lv = sizeof(struct timeval);
e71a4783
SH		 984 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
985 v.tm.tv_sec = 0;
986 v.tm.tv_usec = 0;
987 } else {
988 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
989 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
990 }
991 break;
992
993 case SO_SNDTIMEO:
2a91525c 994 lv = sizeof(struct timeval);
e71a4783
SH		 995 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
996 v.tm.tv_sec = 0;
997 v.tm.tv_usec = 0;
998 } else {
999 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1000 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1001 }
1002 break;
1da177e4 1003
e71a4783
SH		 1004 case SO_RCVLOWAT:
1005 v.val = sk->sk_rcvlowat;
1006 break;
1da177e4 1007
e71a4783 1008 case SO_SNDLOWAT:
2a91525c 1009 v.val = 1;
e71a4783 1010 break;
1da177e4 1011
e71a4783 1012 case SO_PASSCRED:
82981930 1013 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
e71a4783 1014 break;
1da177e4 1015
e71a4783 1016 case SO_PEERCRED:
109f6e39
EB		 1017 {
1018 struct ucred peercred;
1019 if (len > sizeof(peercred))
1020 len = sizeof(peercred);
1021 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1022 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 1023 return -EFAULT;
1024 goto lenout;
109f6e39 1025 }
1da177e4 1026
e71a4783
SH		 1027 case SO_PEERNAME:
1028 {
1029 char address[128];
1030
1031 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1032 return -ENOTCONN;
1033 if (lv < len)
1034 return -EINVAL;
1035 if (copy_to_user(optval, address, len))
1036 return -EFAULT;
1037 goto lenout;
1038 }
1da177e4 1039
e71a4783
SH		 1040 /* Dubious BSD thing... Probably nobody even uses it, but
1041 * the UNIX standard wants it for whatever reason... -DaveM
1042 */
1043 case SO_ACCEPTCONN:
1044 v.val = sk->sk_state == TCP_LISTEN;
1045 break;
1da177e4 1046
e71a4783 1047 case SO_PASSSEC:
82981930 1048 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
e71a4783 1049 break;
877ce7c1 1050
e71a4783
SH		 1051 case SO_PEERSEC:
1052 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 1053
4a19ec58
LAT		 1054 case SO_MARK:
1055 v.val = sk->sk_mark;
1056 break;
1057
3b885787 1058 case SO_RXQ_OVFL:
1b23a5df 1059 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
3b885787
NH		 1060 break;
1061
6e3e939f 1062 case SO_WIFI_STATUS:
1b23a5df 1063 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
6e3e939f
JB		 1064 break;
1065
ef64a54f
PE		 1066 case SO_PEEK_OFF:
1067 if (!sock->ops->set_peek_off)
1068 return -EOPNOTSUPP;
1069
1070 v.val = sk->sk_peek_off;
1071 break;
bc2f7996 1072 case SO_NOFCS:
1b23a5df 1073 v.val = sock_flag(sk, SOCK_NOFCS);
bc2f7996 1074 break;
e71a4783
SH		 1075 default:
1076 return -ENOPROTOOPT;
1da177e4 1077 }
e71a4783 1078
1da177e4
LT		 1079 if (len > lv)
1080 len = lv;
1081 if (copy_to_user(optval, &v, len))
1082 return -EFAULT;
1083lenout:
4ec93edb
YH		 1084 if (put_user(len, optlen))
1085 return -EFAULT;
1086 return 0;
1da177e4
LT		 1087}
1088
a5b5bb9a
IM		 1089/*
1090 * Initialize an sk_lock.
1091 *
1092 * (We also register the sk_lock with the lock validator.)
1093 */
b6f99a21 1094static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 1095{
ed07536e
PZ		 1096 sock_lock_init_class_and_name(sk,
1097 af_family_slock_key_strings[sk->sk_family],
1098 af_family_slock_keys + sk->sk_family,
1099 af_family_key_strings[sk->sk_family],
1100 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 1101}
1102
4dc6dc71
ED		 1103/*
1104 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1105 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1106 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1107 */
f1a6c4da
PE		 1108static void sock_copy(struct sock *nsk, const struct sock *osk)
1109{
1110#ifdef CONFIG_SECURITY_NETWORK
1111 void *sptr = nsk->sk_security;
1112#endif
68835aba
ED		 1113 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1114
1115 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1116 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1117
f1a6c4da
PE		 1118#ifdef CONFIG_SECURITY_NETWORK
1119 nsk->sk_security = sptr;
1120 security_sk_clone(osk, nsk);
1121#endif
1122}
1123
fcbdf09d
OP		 1124/*
1125 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1126 * un-modified. Special care is taken when initializing object to zero.
1127 */
1128static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1129{
1130 if (offsetof(struct sock, sk_node.next) != 0)
1131 memset(sk, 0, offsetof(struct sock, sk_node.next));
1132 memset(&sk->sk_node.pprev, 0,
1133 size - offsetof(struct sock, sk_node.pprev));
1134}
1135
1136void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1137{
1138 unsigned long nulls1, nulls2;
1139
1140 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1141 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1142 if (nulls1 > nulls2)
1143 swap(nulls1, nulls2);
1144
1145 if (nulls1 != 0)
1146 memset((char *)sk, 0, nulls1);
1147 memset((char *)sk + nulls1 + sizeof(void *), 0,
1148 nulls2 - nulls1 - sizeof(void *));
1149 memset((char *)sk + nulls2 + sizeof(void *), 0,
1150 size - nulls2 - sizeof(void *));
1151}
1152EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1153
2e4afe7b
PE		 1154static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1155 int family)
c308c1b2
PE		 1156{
1157 struct sock *sk;
1158 struct kmem_cache *slab;
1159
1160 slab = prot->slab;
e912b114
ED		 1161 if (slab != NULL) {
1162 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1163 if (!sk)
1164 return sk;
1165 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1166 if (prot->clear_sk)
1167 prot->clear_sk(sk, prot->obj_size);
1168 else
1169 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1170 }
fcbdf09d 1171 } else
c308c1b2
PE		 1172 sk = kmalloc(prot->obj_size, priority);
1173
2e4afe7b 1174 if (sk != NULL) {
a98b65a3
VN		 1175 kmemcheck_annotate_bitfield(sk, flags);
1176
2e4afe7b
PE		 1177 if (security_sk_alloc(sk, family, priority))
1178 goto out_free;
1179
1180 if (!try_module_get(prot->owner))
1181 goto out_free_sec;
e022f0b4 1182 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1183 }
1184
c308c1b2 1185 return sk;
2e4afe7b
PE		 1186
1187out_free_sec:
1188 security_sk_free(sk);
1189out_free:
1190 if (slab != NULL)
1191 kmem_cache_free(slab, sk);
1192 else
1193 kfree(sk);
1194 return NULL;
c308c1b2
PE		 1195}
1196
1197static void sk_prot_free(struct proto *prot, struct sock *sk)
1198{
1199 struct kmem_cache *slab;
2e4afe7b 1200 struct module *owner;
c308c1b2 1201
2e4afe7b 1202 owner = prot->owner;
c308c1b2 1203 slab = prot->slab;
2e4afe7b
PE		 1204
1205 security_sk_free(sk);
c308c1b2
PE		 1206 if (slab != NULL)
1207 kmem_cache_free(slab, sk);
1208 else
1209 kfree(sk);
2e4afe7b 1210 module_put(owner);
c308c1b2
PE		 1211}
1212
f8451725
HX		 1213#ifdef CONFIG_CGROUPS
1214void sock_update_classid(struct sock *sk)
1215{
1144182a 1216 u32 classid;
f8451725 1217
1144182a
PM		 1218 rcu_read_lock(); /* doing current task, which cannot vanish. */
1219 classid = task_cls_classid(current);
1220 rcu_read_unlock();
f8451725
HX		 1221 if (classid && classid != sk->sk_classid)
1222 sk->sk_classid = classid;
1223}
82862742 1224EXPORT_SYMBOL(sock_update_classid);
5bc1421e 1225
406a3c63 1226void sock_update_netprioidx(struct sock *sk, struct task_struct *task)
5bc1421e 1227{
5bc1421e
NH		 1228 if (in_interrupt())
1229 return;
2b73bc65 1230
406a3c63 1231 sk->sk_cgrp_prioidx = task_netprioidx(task);
5bc1421e
NH		 1232}
1233EXPORT_SYMBOL_GPL(sock_update_netprioidx);
f8451725
HX		 1234#endif
1235
1da177e4
LT		 1236/**
1237 * sk_alloc - All socket objects are allocated here
c4ea43c5 1238 * @net: the applicable net namespace
4dc3b16b
PP		 1239 * @family: protocol family
1240 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1241 * @prot: struct proto associated with this new sock instance
1da177e4 1242 */
1b8d7ae4 1243struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1244 struct proto *prot)
1da177e4 1245{
c308c1b2 1246 struct sock *sk;
1da177e4 1247
154adbc8 1248 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1249 if (sk) {
154adbc8
PE		 1250 sk->sk_family = family;
1251 /*
1252 * See comment in struct sock definition to understand
1253 * why we need sk_prot_creator -acme
1254 */
1255 sk->sk_prot = sk->sk_prot_creator = prot;
1256 sock_lock_init(sk);
3b1e0a65 1257 sock_net_set(sk, get_net(net));
d66ee058 1258 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1259
1260 sock_update_classid(sk);
406a3c63 1261 sock_update_netprioidx(sk, current);
1da177e4 1262 }
a79af59e 1263
2e4afe7b 1264 return sk;
1da177e4 1265}
2a91525c 1266EXPORT_SYMBOL(sk_alloc);
1da177e4 1267
2b85a34e 1268static void __sk_free(struct sock *sk)
1da177e4
LT		 1269{
1270 struct sk_filter *filter;
1da177e4
LT		 1271
1272 if (sk->sk_destruct)
1273 sk->sk_destruct(sk);
1274
a898def2
PM		 1275 filter = rcu_dereference_check(sk->sk_filter,
1276 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1277 if (filter) {
309dd5fc 1278 sk_filter_uncharge(sk, filter);
a9b3cd7f 1279 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1280 }
1281
08e29af3 1282 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4
LT		 1283
1284 if (atomic_read(&sk->sk_omem_alloc))
e005d193
JP		 1285 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1286 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1287
109f6e39
EB		 1288 if (sk->sk_peer_cred)
1289 put_cred(sk->sk_peer_cred);
1290 put_pid(sk->sk_peer_pid);
3b1e0a65 1291 put_net(sock_net(sk));
c308c1b2 1292 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1293}
2b85a34e
ED		 1294
1295void sk_free(struct sock *sk)
1296{
1297 /*
25985edc 1298 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1299 * some packets are still in some tx queue.
1300 * If not null, sock_wfree() will call __sk_free(sk) later
1301 */
1302 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1303 __sk_free(sk);
1304}
2a91525c 1305EXPORT_SYMBOL(sk_free);
1da177e4 1306
edf02087 1307/*
25985edc
LDM		 1308 * Last sock_put should drop reference to sk->sk_net. It has already
1309 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1310 * is not an option.
25985edc 1311 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1312 * destroy it in the context of init_net.
1313 */
1314void sk_release_kernel(struct sock *sk)
1315{
1316 if (sk == NULL || sk->sk_socket == NULL)
1317 return;
1318
1319 sock_hold(sk);
1320 sock_release(sk->sk_socket);
65a18ec5 1321 release_net(sock_net(sk));
3b1e0a65 1322 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1323 sock_put(sk);
1324}
45af1754 1325EXPORT_SYMBOL(sk_release_kernel);
edf02087 1326
475f1b52
SR		 1327static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1328{
1329 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1330 sock_update_memcg(newsk);
1331}
1332
e56c57d0
ED		 1333/**
1334 * sk_clone_lock - clone a socket, and lock its clone
1335 * @sk: the socket to clone
1336 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1337 *
1338 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1339 */
1340struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1341{
8fd1d178 1342 struct sock *newsk;
87d11ceb 1343
8fd1d178 1344 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1345 if (newsk != NULL) {
1346 struct sk_filter *filter;
1347
892c141e 1348 sock_copy(newsk, sk);
87d11ceb
ACM		 1349
1350 /* SANITY */
3b1e0a65 1351 get_net(sock_net(newsk));
87d11ceb
ACM		 1352 sk_node_init(&newsk->sk_node);
1353 sock_lock_init(newsk);
1354 bh_lock_sock(newsk);
fa438ccf 1355 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1356 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1357
1358 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1359 /*
1360 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1361 */
1362 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1363 atomic_set(&newsk->sk_omem_alloc, 0);
1364 skb_queue_head_init(&newsk->sk_receive_queue);
1365 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1366#ifdef CONFIG_NET_DMA
1367 skb_queue_head_init(&newsk->sk_async_wait_queue);
1368#endif
87d11ceb 1369
b6c6712a 1370 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1371 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1372 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1373 af_callback_keys + newsk->sk_family,
1374 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1375
1376 newsk->sk_dst_cache = NULL;
1377 newsk->sk_wmem_queued = 0;
1378 newsk->sk_forward_alloc = 0;
1379 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1380 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1381
1382 sock_reset_flag(newsk, SOCK_DONE);
1383 skb_queue_head_init(&newsk->sk_error_queue);
1384
0d7da9dd 1385 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1386 if (filter != NULL)
1387 sk_filter_charge(newsk, filter);
1388
1389 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1390 /* It is still raw copy of parent, so invalidate
1391 * destructor and make plain sk_free() */
1392 newsk->sk_destruct = NULL;
b0691c8e 1393 bh_unlock_sock(newsk);
87d11ceb
ACM		 1394 sk_free(newsk);
1395 newsk = NULL;
1396 goto out;
1397 }
1398
1399 newsk->sk_err = 0;
1400 newsk->sk_priority = 0;
4dc6dc71
ED		 1401 /*
1402 * Before updating sk_refcnt, we must commit prior changes to memory
1403 * (Documentation/RCU/rculist_nulls.txt for details)
1404 */
1405 smp_wmb();
87d11ceb
ACM		 1406 atomic_set(&newsk->sk_refcnt, 2);
1407
1408 /*
1409 * Increment the counter in the same struct proto as the master
1410 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1411 * is the same as sk->sk_prot->socks, as this field was copied
1412 * with memcpy).
1413 *
1414 * This _changes_ the previous behaviour, where
1415 * tcp_create_openreq_child always was incrementing the
1416 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1417 * to be taken into account in all callers. -acme
1418 */
1419 sk_refcnt_debug_inc(newsk);
972692e0 1420 sk_set_socket(newsk, NULL);
43815482 1421 newsk->sk_wq = NULL;
87d11ceb 1422
f3f511e1
GC		 1423 sk_update_clone(sk, newsk);
1424
87d11ceb 1425 if (newsk->sk_prot->sockets_allocated)
180d8cd9 1426 sk_sockets_allocated_inc(newsk);
704da560 1427
08e29af3 1428 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1429 net_enable_timestamp();
87d11ceb
ACM		 1430 }
1431out:
1432 return newsk;
1433}
e56c57d0 1434EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1435
9958089a
AK		 1436void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1437{
1438 __sk_dst_set(sk, dst);
1439 sk->sk_route_caps = dst->dev->features;
1440 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1441 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1442 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1443 if (sk_can_gso(sk)) {
82cc1a7a 1444 if (dst->header_len) {
9958089a 1445 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1446 } else {
9958089a 1447 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a
PWJ		 1448 sk->sk_gso_max_size = dst->dev->gso_max_size;
1449 }
9958089a
AK		 1450 }
1451}
1452EXPORT_SYMBOL_GPL(sk_setup_caps);
1453
1da177e4
LT		 1454void __init sk_init(void)
1455{
4481374c 1456 if (totalram_pages <= 4096) {
1da177e4
LT		 1457 sysctl_wmem_max = 32767;
1458 sysctl_rmem_max = 32767;
1459 sysctl_wmem_default = 32767;
1460 sysctl_rmem_default = 32767;
4481374c 1461 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1462 sysctl_wmem_max = 131071;
1463 sysctl_rmem_max = 131071;
1464 }
1465}
1466
1467/*
1468 * Simple resource managers for sockets.
1469 */
1470
1471
4ec93edb
YH		 1472/*
1473 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1474 */
1475void sock_wfree(struct sk_buff *skb)
1476{
1477 struct sock *sk = skb->sk;
d99927f4 1478 unsigned int len = skb->truesize;
1da177e4 1479
d99927f4
ED		 1480 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1481 /*
1482 * Keep a reference on sk_wmem_alloc, this will be released
1483 * after sk_write_space() call
1484 */
1485 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1486 sk->sk_write_space(sk);
d99927f4
ED		 1487 len = 1;
1488 }
2b85a34e 1489 /*
d99927f4
ED		 1490 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1491 * could not do because of in-flight packets
2b85a34e 1492 */
d99927f4 1493 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1494 __sk_free(sk);
1da177e4 1495}
2a91525c 1496EXPORT_SYMBOL(sock_wfree);
1da177e4 1497
4ec93edb
YH		 1498/*
1499 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1500 */
1501void sock_rfree(struct sk_buff *skb)
1502{
1503 struct sock *sk = skb->sk;
d361fd59 1504 unsigned int len = skb->truesize;
1da177e4 1505
d361fd59
ED		 1506 atomic_sub(len, &sk->sk_rmem_alloc);
1507 sk_mem_uncharge(sk, len);
1da177e4 1508}
2a91525c 1509EXPORT_SYMBOL(sock_rfree);
1da177e4 1510
41063e9d
DM		 1511void sock_edemux(struct sk_buff *skb)
1512{
1513 sock_put(skb->sk);
1514}
1515EXPORT_SYMBOL(sock_edemux);
1da177e4
LT		 1516
1517int sock_i_uid(struct sock *sk)
1518{
1519 int uid;
1520
f064af1e 1521 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1522 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1523 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1524 return uid;
1525}
2a91525c 1526EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1527
1528unsigned long sock_i_ino(struct sock *sk)
1529{
1530 unsigned long ino;
1531
f064af1e 1532 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1533 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1534 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1535 return ino;
1536}
2a91525c 1537EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1538
1539/*
1540 * Allocate a skb from the socket's send buffer.
1541 */
86a76caf 1542struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1543 gfp_t priority)
1da177e4
LT		 1544{
1545 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1546 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1547 if (skb) {
1548 skb_set_owner_w(skb, sk);
1549 return skb;
1550 }
1551 }
1552 return NULL;
1553}
2a91525c 1554EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1555
1556/*
1557 * Allocate a skb from the socket's receive buffer.
4ec93edb 1558 */
86a76caf 1559struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1560 gfp_t priority)
1da177e4
LT		 1561{
1562 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1563 struct sk_buff *skb = alloc_skb(size, priority);
1564 if (skb) {
1565 skb_set_owner_r(skb, sk);
1566 return skb;
1567 }
1568 }
1569 return NULL;
1570}
1571
4ec93edb 1572/*
1da177e4 1573 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1574 */
dd0fc66f 1575void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4 1576{
95c96174 1577 if ((unsigned int)size <= sysctl_optmem_max &&
1da177e4
LT		 1578 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1579 void *mem;
1580 /* First do the add, to avoid the race if kmalloc
4ec93edb 1581 * might sleep.
1da177e4
LT		 1582 */
1583 atomic_add(size, &sk->sk_omem_alloc);
1584 mem = kmalloc(size, priority);
1585 if (mem)
1586 return mem;
1587 atomic_sub(size, &sk->sk_omem_alloc);
1588 }
1589 return NULL;
1590}
2a91525c 1591EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1592
1593/*
1594 * Free an option memory block.
1595 */
1596void sock_kfree_s(struct sock *sk, void *mem, int size)
1597{
1598 kfree(mem);
1599 atomic_sub(size, &sk->sk_omem_alloc);
1600}
2a91525c 1601EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1602
1603/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1604 I think, these locks should be removed for datagram sockets.
1605 */
2a91525c 1606static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1607{
1608 DEFINE_WAIT(wait);
1609
1610 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1611 for (;;) {
1612 if (!timeo)
1613 break;
1614 if (signal_pending(current))
1615 break;
1616 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1617 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1618 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1619 break;
1620 if (sk->sk_shutdown & SEND_SHUTDOWN)
1621 break;
1622 if (sk->sk_err)
1623 break;
1624 timeo = schedule_timeout(timeo);
1625 }
aa395145 1626 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1627 return timeo;
1628}
1629
1630
1631/*
1632 * Generic send/receive buffer handlers
1633 */
1634
4cc7f68d
HX		 1635struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1636 unsigned long data_len, int noblock,
1637 int *errcode)
1da177e4
LT		 1638{
1639 struct sk_buff *skb;
7d877f3b 1640 gfp_t gfp_mask;
1da177e4
LT		 1641 long timeo;
1642 int err;
cc9b17ad
JW		 1643 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1644
1645 err = -EMSGSIZE;
1646 if (npages > MAX_SKB_FRAGS)
1647 goto failure;
1da177e4
LT		 1648
1649 gfp_mask = sk->sk_allocation;
1650 if (gfp_mask & __GFP_WAIT)
1651 gfp_mask |= __GFP_REPEAT;
1652
1653 timeo = sock_sndtimeo(sk, noblock);
1654 while (1) {
1655 err = sock_error(sk);
1656 if (err != 0)
1657 goto failure;
1658
1659 err = -EPIPE;
1660 if (sk->sk_shutdown & SEND_SHUTDOWN)
1661 goto failure;
1662
1663 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1664 skb = alloc_skb(header_len, gfp_mask);
1da177e4 1665 if (skb) {
1da177e4
LT		 1666 int i;
1667
1668 /* No pages, we're done... */
1669 if (!data_len)
1670 break;
1671
1da177e4
LT		 1672 skb->truesize += data_len;
1673 skb_shinfo(skb)->nr_frags = npages;
1674 for (i = 0; i < npages; i++) {
1675 struct page *page;
1da177e4
LT		 1676
1677 page = alloc_pages(sk->sk_allocation, 0);
1678 if (!page) {
1679 err = -ENOBUFS;
1680 skb_shinfo(skb)->nr_frags = i;
1681 kfree_skb(skb);
1682 goto failure;
1683 }
1684
ea2ab693
IC		 1685 __skb_fill_page_desc(skb, i,
1686 page, 0,
1687 (data_len >= PAGE_SIZE ?
1688 PAGE_SIZE :
1689 data_len));
1da177e4
LT		 1690 data_len -= PAGE_SIZE;
1691 }
1692
1693 /* Full success... */
1694 break;
1695 }
1696 err = -ENOBUFS;
1697 goto failure;
1698 }
1699 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1700 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1701 err = -EAGAIN;
1702 if (!timeo)
1703 goto failure;
1704 if (signal_pending(current))
1705 goto interrupted;
1706 timeo = sock_wait_for_wmem(sk, timeo);
1707 }
1708
1709 skb_set_owner_w(skb, sk);
1710 return skb;
1711
1712interrupted:
1713 err = sock_intr_errno(timeo);
1714failure:
1715 *errcode = err;
1716 return NULL;
1717}
4cc7f68d 1718EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1719
4ec93edb 1720struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1721 int noblock, int *errcode)
1722{
1723 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1724}
2a91525c 1725EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1726
1727static void __lock_sock(struct sock *sk)
f39234d6
NK		 1728 __releases(&sk->sk_lock.slock)
1729 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1730{
1731 DEFINE_WAIT(wait);
1732
e71a4783 1733 for (;;) {
1da177e4
LT		 1734 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1735 TASK_UNINTERRUPTIBLE);
1736 spin_unlock_bh(&sk->sk_lock.slock);
1737 schedule();
1738 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1739 if (!sock_owned_by_user(sk))
1da177e4
LT		 1740 break;
1741 }
1742 finish_wait(&sk->sk_lock.wq, &wait);
1743}
1744
1745static void __release_sock(struct sock *sk)
f39234d6
NK		 1746 __releases(&sk->sk_lock.slock)
1747 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1748{
1749 struct sk_buff *skb = sk->sk_backlog.head;
1750
1751 do {
1752 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1753 bh_unlock_sock(sk);
1754
1755 do {
1756 struct sk_buff *next = skb->next;
1757
e4cbb02a 1758 prefetch(next);
7fee226a 1759 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1760 skb->next = NULL;
c57943a1 1761 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1762
1763 /*
1764 * We are in process context here with softirqs
1765 * disabled, use cond_resched_softirq() to preempt.
1766 * This is safe to do because we've taken the backlog
1767 * queue private:
1768 */
1769 cond_resched_softirq();
1770
1771 skb = next;
1772 } while (skb != NULL);
1773
1774 bh_lock_sock(sk);
e71a4783 1775 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1776
1777 /*
1778 * Doing the zeroing here guarantee we can not loop forever
1779 * while a wild producer attempts to flood us.
1780 */
1781 sk->sk_backlog.len = 0;
1da177e4
LT		 1782}
1783
1784/**
1785 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1786 * @sk: sock to wait on
1787 * @timeo: for how long
1da177e4
LT		 1788 *
1789 * Now socket state including sk->sk_err is changed only under lock,
1790 * hence we may omit checks after joining wait queue.
1791 * We check receive queue before schedule() only as optimization;
1792 * it is very likely that release_sock() added new data.
1793 */
1794int sk_wait_data(struct sock *sk, long *timeo)
1795{
1796 int rc;
1797 DEFINE_WAIT(wait);
1798
aa395145 1799 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1800 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1801 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1802 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1803 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1804 return rc;
1805}
1da177e4
LT		 1806EXPORT_SYMBOL(sk_wait_data);
1807
3ab224be
HA		 1808/**
1809 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1810 * @sk: socket
1811 * @size: memory size to allocate
1812 * @kind: allocation type
1813 *
1814 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1815 * rmem allocation. This function assumes that protocols which have
1816 * memory_pressure use sk_wmem_queued as write buffer accounting.
1817 */
1818int __sk_mem_schedule(struct sock *sk, int size, int kind)
1819{
1820 struct proto *prot = sk->sk_prot;
1821 int amt = sk_mem_pages(size);
8d987e5c 1822 long allocated;
e1aab161 1823 int parent_status = UNDER_LIMIT;
3ab224be
HA		 1824
1825 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
180d8cd9 1826
e1aab161 1827 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
3ab224be
HA		 1828
1829 /* Under limit. */
e1aab161
GC		 1830 if (parent_status == UNDER_LIMIT &&
1831 allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 1832 sk_leave_memory_pressure(sk);
3ab224be
HA		 1833 return 1;
1834 }
1835
e1aab161
GC		 1836 /* Under pressure. (we or our parents) */
1837 if ((parent_status > SOFT_LIMIT) ||
1838 allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 1839 sk_enter_memory_pressure(sk);
3ab224be 1840
e1aab161
GC		 1841 /* Over hard limit (we or our parents) */
1842 if ((parent_status == OVER_LIMIT) ||
1843 (allocated > sk_prot_mem_limits(sk, 2)))
3ab224be
HA		 1844 goto suppress_allocation;
1845
1846 /* guarantee minimum buffer size under pressure */
1847 if (kind == SK_MEM_RECV) {
1848 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1849 return 1;
180d8cd9 1850
3ab224be
HA		 1851 } else { /* SK_MEM_SEND */
1852 if (sk->sk_type == SOCK_STREAM) {
1853 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1854 return 1;
1855 } else if (atomic_read(&sk->sk_wmem_alloc) <
1856 prot->sysctl_wmem[0])
1857 return 1;
1858 }
1859
180d8cd9 1860 if (sk_has_memory_pressure(sk)) {
1748376b
ED		 1861 int alloc;
1862
180d8cd9 1863 if (!sk_under_memory_pressure(sk))
1748376b 1864 return 1;
180d8cd9
GC		 1865 alloc = sk_sockets_allocated_read_positive(sk);
1866 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 1867 sk_mem_pages(sk->sk_wmem_queued +
1868 atomic_read(&sk->sk_rmem_alloc) +
1869 sk->sk_forward_alloc))
1870 return 1;
1871 }
1872
1873suppress_allocation:
1874
1875 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1876 sk_stream_moderate_sndbuf(sk);
1877
1878 /* Fail only if socket is _under_ its sndbuf.
1879 * In this case we cannot block, so that we have to fail.
1880 */
1881 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1882 return 1;
1883 }
1884
3847ce32
SM		 1885 trace_sock_exceed_buf_limit(sk, prot, allocated);
1886
3ab224be
HA		 1887 /* Alas. Undo changes. */
1888 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
180d8cd9 1889
0e90b31f 1890 sk_memory_allocated_sub(sk, amt);
180d8cd9 1891
3ab224be
HA		 1892 return 0;
1893}
3ab224be
HA		 1894EXPORT_SYMBOL(__sk_mem_schedule);
1895
1896/**
1897 * __sk_reclaim - reclaim memory_allocated
1898 * @sk: socket
1899 */
1900void __sk_mem_reclaim(struct sock *sk)
1901{
180d8cd9 1902 sk_memory_allocated_sub(sk,
0e90b31f 1903 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
3ab224be
HA		 1904 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1905
180d8cd9
GC		 1906 if (sk_under_memory_pressure(sk) &&
1907 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1908 sk_leave_memory_pressure(sk);
3ab224be 1909}
3ab224be
HA		 1910EXPORT_SYMBOL(__sk_mem_reclaim);
1911
1912
1da177e4
LT		 1913/*
1914 * Set of default routines for initialising struct proto_ops when
1915 * the protocol does not support a particular function. In certain
1916 * cases where it makes no sense for a protocol to have a "do nothing"
1917 * function, some default processing is provided.
1918 */
1919
1920int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1921{
1922 return -EOPNOTSUPP;
1923}
2a91525c 1924EXPORT_SYMBOL(sock_no_bind);
1da177e4 1925
4ec93edb 1926int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1927 int len, int flags)
1928{
1929 return -EOPNOTSUPP;
1930}
2a91525c 1931EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1932
1933int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1934{
1935 return -EOPNOTSUPP;
1936}
2a91525c 1937EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1938
1939int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1940{
1941 return -EOPNOTSUPP;
1942}
2a91525c 1943EXPORT_SYMBOL(sock_no_accept);
1da177e4 1944
4ec93edb 1945int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1946 int *len, int peer)
1947{
1948 return -EOPNOTSUPP;
1949}
2a91525c 1950EXPORT_SYMBOL(sock_no_getname);
1da177e4 1951
2a91525c 1952unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1953{
1954 return 0;
1955}
2a91525c 1956EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1957
1958int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1959{
1960 return -EOPNOTSUPP;
1961}
2a91525c 1962EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1963
1964int sock_no_listen(struct socket *sock, int backlog)
1965{
1966 return -EOPNOTSUPP;
1967}
2a91525c 1968EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1969
1970int sock_no_shutdown(struct socket *sock, int how)
1971{
1972 return -EOPNOTSUPP;
1973}
2a91525c 1974EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1975
1976int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1977 char __user *optval, unsigned int optlen)
1da177e4
LT		 1978{
1979 return -EOPNOTSUPP;
1980}
2a91525c 1981EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1982
1983int sock_no_getsockopt(struct socket *sock, int level, int optname,
1984 char __user *optval, int __user *optlen)
1985{
1986 return -EOPNOTSUPP;
1987}
2a91525c 1988EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 1989
1990int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1991 size_t len)
1992{
1993 return -EOPNOTSUPP;
1994}
2a91525c 1995EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 1996
1997int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1998 size_t len, int flags)
1999{
2000 return -EOPNOTSUPP;
2001}
2a91525c 2002EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 2003
2004int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2005{
2006 /* Mirror missing mmap method error code */
2007 return -ENODEV;
2008}
2a91525c 2009EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 2010
2011ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2012{
2013 ssize_t res;
2014 struct msghdr msg = {.msg_flags = flags};
2015 struct kvec iov;
2016 char *kaddr = kmap(page);
2017 iov.iov_base = kaddr + offset;
2018 iov.iov_len = size;
2019 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2020 kunmap(page);
2021 return res;
2022}
2a91525c 2023EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 2024
2025/*
2026 * Default Socket Callbacks
2027 */
2028
2029static void sock_def_wakeup(struct sock *sk)
2030{
43815482
ED		 2031 struct socket_wq *wq;
2032
2033 rcu_read_lock();
2034 wq = rcu_dereference(sk->sk_wq);
2035 if (wq_has_sleeper(wq))
2036 wake_up_interruptible_all(&wq->wait);
2037 rcu_read_unlock();
1da177e4
LT		 2038}
2039
2040static void sock_def_error_report(struct sock *sk)
2041{
43815482
ED		 2042 struct socket_wq *wq;
2043
2044 rcu_read_lock();
2045 wq = rcu_dereference(sk->sk_wq);
2046 if (wq_has_sleeper(wq))
2047 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 2048 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 2049 rcu_read_unlock();
1da177e4
LT		 2050}
2051
2052static void sock_def_readable(struct sock *sk, int len)
2053{
43815482
ED		 2054 struct socket_wq *wq;
2055
2056 rcu_read_lock();
2057 wq = rcu_dereference(sk->sk_wq);
2058 if (wq_has_sleeper(wq))
2c6607c6 2059 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 2060 POLLRDNORM | POLLRDBAND);
8d8ad9d7 2061 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 2062 rcu_read_unlock();
1da177e4
LT		 2063}
2064
2065static void sock_def_write_space(struct sock *sk)
2066{
43815482
ED		 2067 struct socket_wq *wq;
2068
2069 rcu_read_lock();
1da177e4
LT		 2070
2071 /* Do not wake up a writer until he can make "significant"
2072 * progress. --DaveM
2073 */
e71a4783 2074 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 2075 wq = rcu_dereference(sk->sk_wq);
2076 if (wq_has_sleeper(wq))
2077 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 2078 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 2079
2080 /* Should agree with poll, otherwise some programs break */
2081 if (sock_writeable(sk))
8d8ad9d7 2082 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 2083 }
2084
43815482 2085 rcu_read_unlock();
1da177e4
LT		 2086}
2087
2088static void sock_def_destruct(struct sock *sk)
2089{
a51482bd 2090 kfree(sk->sk_protinfo);
1da177e4
LT		 2091}
2092
2093void sk_send_sigurg(struct sock *sk)
2094{
2095 if (sk->sk_socket && sk->sk_socket->file)
2096 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 2097 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 2098}
2a91525c 2099EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 2100
2101void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2102 unsigned long expires)
2103{
2104 if (!mod_timer(timer, expires))
2105 sock_hold(sk);
2106}
1da177e4
LT		 2107EXPORT_SYMBOL(sk_reset_timer);
2108
2109void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2110{
2111 if (timer_pending(timer) && del_timer(timer))
2112 __sock_put(sk);
2113}
1da177e4
LT		 2114EXPORT_SYMBOL(sk_stop_timer);
2115
2116void sock_init_data(struct socket *sock, struct sock *sk)
2117{
2118 skb_queue_head_init(&sk->sk_receive_queue);
2119 skb_queue_head_init(&sk->sk_write_queue);
2120 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2121#ifdef CONFIG_NET_DMA
2122 skb_queue_head_init(&sk->sk_async_wait_queue);
2123#endif
1da177e4
LT		 2124
2125 sk->sk_send_head = NULL;
2126
2127 init_timer(&sk->sk_timer);
4ec93edb 2128
1da177e4
LT		 2129 sk->sk_allocation = GFP_KERNEL;
2130 sk->sk_rcvbuf = sysctl_rmem_default;
2131 sk->sk_sndbuf = sysctl_wmem_default;
2132 sk->sk_state = TCP_CLOSE;
972692e0 2133 sk_set_socket(sk, sock);
1da177e4
LT		 2134
2135 sock_set_flag(sk, SOCK_ZAPPED);
2136
e71a4783 2137 if (sock) {
1da177e4 2138 sk->sk_type = sock->type;
43815482 2139 sk->sk_wq = sock->wq;
1da177e4
LT		 2140 sock->sk = sk;
2141 } else
43815482 2142 sk->sk_wq = NULL;
1da177e4 2143
b6c6712a 2144 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2145 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2146 lockdep_set_class_and_name(&sk->sk_callback_lock,
2147 af_callback_keys + sk->sk_family,
2148 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2149
2150 sk->sk_state_change = sock_def_wakeup;
2151 sk->sk_data_ready = sock_def_readable;
2152 sk->sk_write_space = sock_def_write_space;
2153 sk->sk_error_report = sock_def_error_report;
2154 sk->sk_destruct = sock_def_destruct;
2155
2156 sk->sk_sndmsg_page = NULL;
2157 sk->sk_sndmsg_off = 0;
ef64a54f 2158 sk->sk_peek_off = -1;
1da177e4 2159
109f6e39
EB		 2160 sk->sk_peer_pid = NULL;
2161 sk->sk_peer_cred = NULL;
1da177e4
LT		 2162 sk->sk_write_pending = 0;
2163 sk->sk_rcvlowat = 1;
2164 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2165 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2166
f37f0afb 2167 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2168
4dc6dc71
ED		 2169 /*
2170 * Before updating sk_refcnt, we must commit prior changes to memory
2171 * (Documentation/RCU/rculist_nulls.txt for details)
2172 */
2173 smp_wmb();
1da177e4 2174 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2175 atomic_set(&sk->sk_drops, 0);
1da177e4 2176}
2a91525c 2177EXPORT_SYMBOL(sock_init_data);
1da177e4 2178
b5606c2d 2179void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2180{
2181 might_sleep();
a5b5bb9a 2182 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2183 if (sk->sk_lock.owned)
1da177e4 2184 __lock_sock(sk);
d2e9117c 2185 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2186 spin_unlock(&sk->sk_lock.slock);
2187 /*
2188 * The sk_lock has mutex_lock() semantics here:
2189 */
fcc70d5f 2190 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2191 local_bh_enable();
1da177e4 2192}
fcc70d5f 2193EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2194
b5606c2d 2195void release_sock(struct sock *sk)
1da177e4 2196{
a5b5bb9a
IM		 2197 /*
2198 * The sk_lock has mutex_unlock() semantics:
2199 */
2200 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2201
2202 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2203 if (sk->sk_backlog.tail)
2204 __release_sock(sk);
46d3ceab
ED		 2205
2206 if (sk->sk_prot->release_cb)
2207 sk->sk_prot->release_cb(sk);
2208
d2e9117c 2209 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2210 if (waitqueue_active(&sk->sk_lock.wq))
2211 wake_up(&sk->sk_lock.wq);
2212 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2213}
2214EXPORT_SYMBOL(release_sock);
2215
8a74ad60
ED		 2216/**
2217 * lock_sock_fast - fast version of lock_sock
2218 * @sk: socket
2219 *
2220 * This version should be used for very small section, where process wont block
2221 * return false if fast path is taken
2222 * sk_lock.slock locked, owned = 0, BH disabled
2223 * return true if slow path is taken
2224 * sk_lock.slock unlocked, owned = 1, BH enabled
2225 */
2226bool lock_sock_fast(struct sock *sk)
2227{
2228 might_sleep();
2229 spin_lock_bh(&sk->sk_lock.slock);
2230
2231 if (!sk->sk_lock.owned)
2232 /*
2233 * Note : We must disable BH
2234 */
2235 return false;
2236
2237 __lock_sock(sk);
2238 sk->sk_lock.owned = 1;
2239 spin_unlock(&sk->sk_lock.slock);
2240 /*
2241 * The sk_lock has mutex_lock() semantics here:
2242 */
2243 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2244 local_bh_enable();
2245 return true;
2246}
2247EXPORT_SYMBOL(lock_sock_fast);
2248
1da177e4 2249int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2250{
b7aa0bf7 2251 struct timeval tv;
1da177e4 2252 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2253 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2254 tv = ktime_to_timeval(sk->sk_stamp);
2255 if (tv.tv_sec == -1)
1da177e4 2256 return -ENOENT;
b7aa0bf7
ED		 2257 if (tv.tv_sec == 0) {
2258 sk->sk_stamp = ktime_get_real();
2259 tv = ktime_to_timeval(sk->sk_stamp);
2260 }
2261 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2262}
1da177e4
LT		 2263EXPORT_SYMBOL(sock_get_timestamp);
2264
ae40eb1e
ED		 2265int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2266{
2267 struct timespec ts;
2268 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2269 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2270 ts = ktime_to_timespec(sk->sk_stamp);
2271 if (ts.tv_sec == -1)
2272 return -ENOENT;
2273 if (ts.tv_sec == 0) {
2274 sk->sk_stamp = ktime_get_real();
2275 ts = ktime_to_timespec(sk->sk_stamp);
2276 }
2277 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2278}
2279EXPORT_SYMBOL(sock_get_timestampns);
2280
20d49473 2281void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2282{
20d49473 2283 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2284 unsigned long previous_flags = sk->sk_flags;
2285
20d49473
PO		 2286 sock_set_flag(sk, flag);
2287 /*
2288 * we just set one of the two flags which require net
2289 * time stamping, but time stamping might have been on
2290 * already because of the other one
2291 */
08e29af3 2292 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2293 net_enable_timestamp();
1da177e4
LT		 2294 }
2295}
1da177e4
LT		 2296
2297/*
2298 * Get a socket option on an socket.
2299 *
2300 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2301 * asynchronous errors should be reported by getsockopt. We assume
2302 * this means if you specify SO_ERROR (otherwise whats the point of it).
2303 */
2304int sock_common_getsockopt(struct socket *sock, int level, int optname,
2305 char __user *optval, int __user *optlen)
2306{
2307 struct sock *sk = sock->sk;
2308
2309 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2310}
1da177e4
LT		 2311EXPORT_SYMBOL(sock_common_getsockopt);
2312
3fdadf7d 2313#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2314int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2315 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2316{
2317 struct sock *sk = sock->sk;
2318
1e51f951 2319 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2320 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2321 optval, optlen);
3fdadf7d
DM		 2322 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2323}
2324EXPORT_SYMBOL(compat_sock_common_getsockopt);
2325#endif
2326
1da177e4
LT		 2327int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2328 struct msghdr *msg, size_t size, int flags)
2329{
2330 struct sock *sk = sock->sk;
2331 int addr_len = 0;
2332 int err;
2333
2334 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2335 flags & ~MSG_DONTWAIT, &addr_len);
2336 if (err >= 0)
2337 msg->msg_namelen = addr_len;
2338 return err;
2339}
1da177e4
LT		 2340EXPORT_SYMBOL(sock_common_recvmsg);
2341
2342/*
2343 * Set socket options on an inet socket.
2344 */
2345int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2346 char __user *optval, unsigned int optlen)
1da177e4
LT		 2347{
2348 struct sock *sk = sock->sk;
2349
2350 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2351}
1da177e4
LT		 2352EXPORT_SYMBOL(sock_common_setsockopt);
2353
3fdadf7d 2354#ifdef CONFIG_COMPAT
543d9cfe 2355int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2356 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2357{
2358 struct sock *sk = sock->sk;
2359
543d9cfe
ACM		 2360 if (sk->sk_prot->compat_setsockopt != NULL)
2361 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2362 optval, optlen);
3fdadf7d
DM		 2363 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2364}
2365EXPORT_SYMBOL(compat_sock_common_setsockopt);
2366#endif
2367
1da177e4
LT		 2368void sk_common_release(struct sock *sk)
2369{
2370 if (sk->sk_prot->destroy)
2371 sk->sk_prot->destroy(sk);
2372
2373 /*
2374 * Observation: when sock_common_release is called, processes have
2375 * no access to socket. But net still has.
2376 * Step one, detach it from networking:
2377 *
2378 * A. Remove from hash tables.
2379 */
2380
2381 sk->sk_prot->unhash(sk);
2382
2383 /*
2384 * In this point socket cannot receive new packets, but it is possible
2385 * that some packets are in flight because some CPU runs receiver and
2386 * did hash table lookup before we unhashed socket. They will achieve
2387 * receive queue and will be purged by socket destructor.
2388 *
2389 * Also we still have packets pending on receive queue and probably,
2390 * our own packets waiting in device queues. sock_destroy will drain
2391 * receive queue, but transmitted packets will delay socket destruction
2392 * until the last reference will be released.
2393 */
2394
2395 sock_orphan(sk);
2396
2397 xfrm_sk_free_policy(sk);
2398
e6848976 2399 sk_refcnt_debug_release(sk);
1da177e4
LT		 2400 sock_put(sk);
2401}
1da177e4
LT		 2402EXPORT_SYMBOL(sk_common_release);
2403
13ff3d6f
PE		 2404#ifdef CONFIG_PROC_FS
2405#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2406struct prot_inuse {
2407 int val[PROTO_INUSE_NR];
2408};
13ff3d6f
PE		 2409
2410static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2411
2412#ifdef CONFIG_NET_NS
2413void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2414{
d6d9ca0f 2415 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2416}
2417EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2418
2419int sock_prot_inuse_get(struct net *net, struct proto *prot)
2420{
2421 int cpu, idx = prot->inuse_idx;
2422 int res = 0;
2423
2424 for_each_possible_cpu(cpu)
2425 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2426
2427 return res >= 0 ? res : 0;
2428}
2429EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2430
2c8c1e72 2431static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2432{
2433 net->core.inuse = alloc_percpu(struct prot_inuse);
2434 return net->core.inuse ? 0 : -ENOMEM;
2435}
2436
2c8c1e72 2437static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2438{
2439 free_percpu(net->core.inuse);
2440}
2441
2442static struct pernet_operations net_inuse_ops = {
2443 .init = sock_inuse_init_net,
2444 .exit = sock_inuse_exit_net,
2445};
2446
2447static __init int net_inuse_init(void)
2448{
2449 if (register_pernet_subsys(&net_inuse_ops))
2450 panic("Cannot initialize net inuse counters");
2451
2452 return 0;
2453}
2454
2455core_initcall(net_inuse_init);
2456#else
1338d466
PE		 2457static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2458
c29a0bc4 2459void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2460{
d6d9ca0f 2461 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2462}
2463EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2464
c29a0bc4 2465int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2466{
2467 int cpu, idx = prot->inuse_idx;
2468 int res = 0;
2469
2470 for_each_possible_cpu(cpu)
2471 res += per_cpu(prot_inuse, cpu).val[idx];
2472
2473 return res >= 0 ? res : 0;
2474}
2475EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2476#endif
13ff3d6f
PE		 2477
2478static void assign_proto_idx(struct proto *prot)
2479{
2480 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2481
2482 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
e005d193 2483 pr_err("PROTO_INUSE_NR exhausted\n");
13ff3d6f
PE		 2484 return;
2485 }
2486
2487 set_bit(prot->inuse_idx, proto_inuse_idx);
2488}
2489
2490static void release_proto_idx(struct proto *prot)
2491{
2492 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2493 clear_bit(prot->inuse_idx, proto_inuse_idx);
2494}
2495#else
2496static inline void assign_proto_idx(struct proto *prot)
2497{
2498}
2499
2500static inline void release_proto_idx(struct proto *prot)
2501{
2502}
2503#endif
2504
b733c007
PE		 2505int proto_register(struct proto *prot, int alloc_slab)
2506{
1da177e4
LT		 2507 if (alloc_slab) {
2508 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2509 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2510 NULL);
1da177e4
LT		 2511
2512 if (prot->slab == NULL) {
e005d193
JP		 2513 pr_crit("%s: Can't create sock SLAB cache!\n",
2514 prot->name);
60e7663d 2515 goto out;
1da177e4 2516 }
2e6599cb
ACM		 2517
2518 if (prot->rsk_prot != NULL) {
faf23422 2519 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2520 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2521 goto out_free_sock_slab;
2522
7e56b5d6 2523 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2524 prot->rsk_prot->obj_size, 0,
20c2df83 2525 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2526
2527 if (prot->rsk_prot->slab == NULL) {
e005d193
JP		 2528 pr_crit("%s: Can't create request sock SLAB cache!\n",
2529 prot->name);
2e6599cb
ACM		 2530 goto out_free_request_sock_slab_name;
2531 }
2532 }
8feaf0c0 2533
6d6ee43e 2534 if (prot->twsk_prot != NULL) {
faf23422 2535 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2536
7e56b5d6 2537 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2538 goto out_free_request_sock_slab;
2539
6d6ee43e 2540 prot->twsk_prot->twsk_slab =
7e56b5d6 2541 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2542 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2543 0,
2544 SLAB_HWCACHE_ALIGN |
2545 prot->slab_flags,
20c2df83 2546 NULL);
6d6ee43e 2547 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2548 goto out_free_timewait_sock_slab_name;
2549 }
1da177e4
LT		 2550 }
2551
36b77a52 2552 mutex_lock(&proto_list_mutex);
1da177e4 2553 list_add(&prot->node, &proto_list);
13ff3d6f 2554 assign_proto_idx(prot);
36b77a52 2555 mutex_unlock(&proto_list_mutex);
b733c007
PE		 2556 return 0;
2557
8feaf0c0 2558out_free_timewait_sock_slab_name:
7e56b5d6 2559 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2560out_free_request_sock_slab:
2561 if (prot->rsk_prot && prot->rsk_prot->slab) {
2562 kmem_cache_destroy(prot->rsk_prot->slab);
2563 prot->rsk_prot->slab = NULL;
2564 }
2e6599cb 2565out_free_request_sock_slab_name:
72150e9b
DC		 2566 if (prot->rsk_prot)
2567 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2568out_free_sock_slab:
2569 kmem_cache_destroy(prot->slab);
2570 prot->slab = NULL;
b733c007
PE		 2571out:
2572 return -ENOBUFS;
1da177e4 2573}
1da177e4
LT		 2574EXPORT_SYMBOL(proto_register);
2575
2576void proto_unregister(struct proto *prot)
2577{
36b77a52 2578 mutex_lock(&proto_list_mutex);
13ff3d6f 2579 release_proto_idx(prot);
0a3f4358 2580 list_del(&prot->node);
36b77a52 2581 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2582
2583 if (prot->slab != NULL) {
2584 kmem_cache_destroy(prot->slab);
2585 prot->slab = NULL;
2586 }
2587
2e6599cb 2588 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2589 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2590 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2591 prot->rsk_prot->slab = NULL;
2592 }
2593
6d6ee43e 2594 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2595 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2596 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2597 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2598 }
1da177e4 2599}
1da177e4
LT		 2600EXPORT_SYMBOL(proto_unregister);
2601
2602#ifdef CONFIG_PROC_FS
1da177e4 2603static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 2604 __acquires(proto_list_mutex)
1da177e4 2605{
36b77a52 2606 mutex_lock(&proto_list_mutex);
60f0438a 2607 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2608}
2609
2610static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2611{
60f0438a 2612 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2613}
2614
2615static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 2616 __releases(proto_list_mutex)
1da177e4 2617{
36b77a52 2618 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2619}
2620
2621static char proto_method_implemented(const void *method)
2622{
2623 return method == NULL ? 'n' : 'y';
2624}
180d8cd9
GC		 2625static long sock_prot_memory_allocated(struct proto *proto)
2626{
cb75a36c 2627 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
180d8cd9
GC		 2628}
2629
2630static char *sock_prot_memory_pressure(struct proto *proto)
2631{
2632 return proto->memory_pressure != NULL ?
2633 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2634}
1da177e4
LT		 2635
2636static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2637{
180d8cd9 2638
8d987e5c 2639 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2640 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2641 proto->name,
2642 proto->obj_size,
14e943db 2643 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 2644 sock_prot_memory_allocated(proto),
2645 sock_prot_memory_pressure(proto),
1da177e4
LT		 2646 proto->max_header,
2647 proto->slab == NULL ? "no" : "yes",
2648 module_name(proto->owner),
2649 proto_method_implemented(proto->close),
2650 proto_method_implemented(proto->connect),
2651 proto_method_implemented(proto->disconnect),
2652 proto_method_implemented(proto->accept),
2653 proto_method_implemented(proto->ioctl),
2654 proto_method_implemented(proto->init),
2655 proto_method_implemented(proto->destroy),
2656 proto_method_implemented(proto->shutdown),
2657 proto_method_implemented(proto->setsockopt),
2658 proto_method_implemented(proto->getsockopt),
2659 proto_method_implemented(proto->sendmsg),
2660 proto_method_implemented(proto->recvmsg),
2661 proto_method_implemented(proto->sendpage),
2662 proto_method_implemented(proto->bind),
2663 proto_method_implemented(proto->backlog_rcv),
2664 proto_method_implemented(proto->hash),
2665 proto_method_implemented(proto->unhash),
2666 proto_method_implemented(proto->get_port),
2667 proto_method_implemented(proto->enter_memory_pressure));
2668}
2669
2670static int proto_seq_show(struct seq_file *seq, void *v)
2671{
60f0438a 2672 if (v == &proto_list)
1da177e4
LT		 2673 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2674 "protocol",
2675 "size",
2676 "sockets",
2677 "memory",
2678 "press",
2679 "maxhdr",
2680 "slab",
2681 "module",
2682 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2683 else
60f0438a 2684 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2685 return 0;
2686}
2687
f690808e 2688static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2689 .start = proto_seq_start,
2690 .next = proto_seq_next,
2691 .stop = proto_seq_stop,
2692 .show = proto_seq_show,
2693};
2694
2695static int proto_seq_open(struct inode *inode, struct file *file)
2696{
14e943db
ED		 2697 return seq_open_net(inode, file, &proto_seq_ops,
2698 sizeof(struct seq_net_private));
1da177e4
LT		 2699}
2700
9a32144e 2701static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2702 .owner = THIS_MODULE,
2703 .open = proto_seq_open,
2704 .read = seq_read,
2705 .llseek = seq_lseek,
14e943db
ED		 2706 .release = seq_release_net,
2707};
2708
2709static __net_init int proto_init_net(struct net *net)
2710{
2711 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2712 return -ENOMEM;
2713
2714 return 0;
2715}
2716
2717static __net_exit void proto_exit_net(struct net *net)
2718{
2719 proc_net_remove(net, "protocols");
2720}
2721
2722
2723static __net_initdata struct pernet_operations proto_net_ops = {
2724 .init = proto_init_net,
2725 .exit = proto_exit_net,
1da177e4
LT		 2726};
2727
2728static int __init proto_init(void)
2729{
14e943db 2730 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2731}
2732
2733subsys_initcall(proto_init);
2734
2735#endif /* PROC_FS */
Linux kernel - Deliverables
This page took 1.440762 seconds and 5 git commands to generate.