projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
sctp: Make the address lists per network namespace
[linux-2.6-block.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);
c76562b6
MG		 298
299 /*
300 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
301 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
302 * it has rmem allocations there is a risk that the user of the
303 * socket cannot make forward progress due to exceeding the rmem
304 * limits. By rights, sk_clear_memalloc() should only be called
305 * on sockets being torn down but warn and reset the accounting if
306 * that assumption breaks.
307 */
308 if (WARN_ON(sk->sk_forward_alloc))
309 sk_mem_reclaim(sk);
7cb02404
MG		 310}
311EXPORT_SYMBOL_GPL(sk_clear_memalloc);
312
b4b9e355
MG		 313int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
314{
315 int ret;
316 unsigned long pflags = current->flags;
317
318 /* these should have been dropped before queueing */
319 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
320
321 current->flags |= PF_MEMALLOC;
322 ret = sk->sk_backlog_rcv(sk, skb);
323 tsk_restore_flags(current, pflags, PF_MEMALLOC);
324
325 return ret;
326}
327EXPORT_SYMBOL(__sk_backlog_rcv);
328
5bc1421e
NH		 329#if defined(CONFIG_CGROUPS)
330#if !defined(CONFIG_NET_CLS_CGROUP)
f8451725
HX		 331int net_cls_subsys_id = -1;
332EXPORT_SYMBOL_GPL(net_cls_subsys_id);
333#endif
5bc1421e
NH		 334#if !defined(CONFIG_NETPRIO_CGROUP)
335int net_prio_subsys_id = -1;
336EXPORT_SYMBOL_GPL(net_prio_subsys_id);
337#endif
338#endif
f8451725 339
1da177e4
LT		 340static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
341{
342 struct timeval tv;
343
344 if (optlen < sizeof(tv))
345 return -EINVAL;
346 if (copy_from_user(&tv, optval, sizeof(tv)))
347 return -EFAULT;
ba78073e
VA		 348 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
349 return -EDOM;
1da177e4 350
ba78073e 351 if (tv.tv_sec < 0) {
6f11df83
AM		 352 static int warned __read_mostly;
353
ba78073e 354 *timeo_p = 0;
50aab54f 355 if (warned < 10 && net_ratelimit()) {
ba78073e 356 warned++;
e005d193
JP		 357 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
358 __func__, current->comm, task_pid_nr(current));
50aab54f 359 }
ba78073e
VA		 360 return 0;
361 }
1da177e4
LT		 362 *timeo_p = MAX_SCHEDULE_TIMEOUT;
363 if (tv.tv_sec == 0 && tv.tv_usec == 0)
364 return 0;
365 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
366 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
367 return 0;
368}
369
370static void sock_warn_obsolete_bsdism(const char *name)
371{
372 static int warned;
373 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 374 if (strcmp(warncomm, current->comm) && warned < 5) {
375 strcpy(warncomm, current->comm);
e005d193
JP		 376 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
377 warncomm, name);
1da177e4
LT		 378 warned++;
379 }
380}
381
08e29af3
ED		 382#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
383
384static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 385{
08e29af3
ED		 386 if (sk->sk_flags & flags) {
387 sk->sk_flags &= ~flags;
388 if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 389 net_disable_timestamp();
1da177e4
LT		 390 }
391}
392
393
f0088a50
DV		 394int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
395{
766e9037 396 int err;
f0088a50 397 int skb_len;
3b885787
NH		 398 unsigned long flags;
399 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 400
0fd7bac6 401 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
766e9037 402 atomic_inc(&sk->sk_drops);
3847ce32 403 trace_sock_rcvqueue_full(sk, skb);
766e9037 404 return -ENOMEM;
f0088a50
DV		 405 }
406
fda9ef5d 407 err = sk_filter(sk, skb);
f0088a50 408 if (err)
766e9037 409 return err;
f0088a50 410
c76562b6 411 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
766e9037
ED		 412 atomic_inc(&sk->sk_drops);
413 return -ENOBUFS;
3ab224be
HA		 414 }
415
f0088a50
DV		 416 skb->dev = NULL;
417 skb_set_owner_r(skb, sk);
49ad9599 418
f0088a50
DV		 419 /* Cache the SKB length before we tack it onto the receive
420 * queue. Once it is added it no longer belongs to us and
421 * may be freed by other threads of control pulling packets
422 * from the queue.
423 */
424 skb_len = skb->len;
425
7fee226a
ED		 426 /* we escape from rcu protected region, make sure we dont leak
427 * a norefcounted dst
428 */
429 skb_dst_force(skb);
430
3b885787
NH		 431 spin_lock_irqsave(&list->lock, flags);
432 skb->dropcount = atomic_read(&sk->sk_drops);
433 __skb_queue_tail(list, skb);
434 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 435
436 if (!sock_flag(sk, SOCK_DEAD))
437 sk->sk_data_ready(sk, skb_len);
766e9037 438 return 0;
f0088a50
DV		 439}
440EXPORT_SYMBOL(sock_queue_rcv_skb);
441
58a5a7b9 442int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
f0088a50
DV		 443{
444 int rc = NET_RX_SUCCESS;
445
fda9ef5d 446 if (sk_filter(sk, skb))
f0088a50
DV		 447 goto discard_and_relse;
448
449 skb->dev = NULL;
450
f545a38f 451 if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
c377411f
ED		 452 atomic_inc(&sk->sk_drops);
453 goto discard_and_relse;
454 }
58a5a7b9
ACM		 455 if (nested)
456 bh_lock_sock_nested(sk);
457 else
458 bh_lock_sock(sk);
a5b5bb9a
IM		 459 if (!sock_owned_by_user(sk)) {
460 /*
461 * trylock + unlock semantics:
462 */
463 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
464
c57943a1 465 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 466
467 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
f545a38f 468 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
8eae939f
ZY		 469 bh_unlock_sock(sk);
470 atomic_inc(&sk->sk_drops);
471 goto discard_and_relse;
472 }
473
f0088a50
DV		 474 bh_unlock_sock(sk);
475out:
476 sock_put(sk);
477 return rc;
478discard_and_relse:
479 kfree_skb(skb);
480 goto out;
481}
482EXPORT_SYMBOL(sk_receive_skb);
483
ea94ff3b
KK		 484void sk_reset_txq(struct sock *sk)
485{
486 sk_tx_queue_clear(sk);
487}
488EXPORT_SYMBOL(sk_reset_txq);
489
f0088a50
DV		 490struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
491{
b6c6712a 492 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 493
494 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 495 sk_tx_queue_clear(sk);
a9b3cd7f 496 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 497 dst_release(dst);
498 return NULL;
499 }
500
501 return dst;
502}
503EXPORT_SYMBOL(__sk_dst_check);
504
505struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
506{
507 struct dst_entry *dst = sk_dst_get(sk);
508
509 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
510 sk_dst_reset(sk);
511 dst_release(dst);
512 return NULL;
513 }
514
515 return dst;
516}
517EXPORT_SYMBOL(sk_dst_check);
518
4878809f
DM		 519static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
520{
521 int ret = -ENOPROTOOPT;
522#ifdef CONFIG_NETDEVICES
3b1e0a65 523 struct net *net = sock_net(sk);
4878809f
DM		 524 char devname[IFNAMSIZ];
525 int index;
526
527 /* Sorry... */
528 ret = -EPERM;
529 if (!capable(CAP_NET_RAW))
530 goto out;
531
532 ret = -EINVAL;
533 if (optlen < 0)
534 goto out;
535
536 /* Bind this socket to a particular device like "eth0",
537 * as specified in the passed interface name. If the
538 * name is "" or the option length is zero the socket
539 * is not bound.
540 */
541 if (optlen > IFNAMSIZ - 1)
542 optlen = IFNAMSIZ - 1;
543 memset(devname, 0, sizeof(devname));
544
545 ret = -EFAULT;
546 if (copy_from_user(devname, optval, optlen))
547 goto out;
548
000ba2e4
DM		 549 index = 0;
550 if (devname[0] != '\0') {
bf8e56bf 551 struct net_device *dev;
4878809f 552
bf8e56bf
ED		 553 rcu_read_lock();
554 dev = dev_get_by_name_rcu(net, devname);
555 if (dev)
556 index = dev->ifindex;
557 rcu_read_unlock();
4878809f
DM		 558 ret = -ENODEV;
559 if (!dev)
560 goto out;
4878809f
DM		 561 }
562
563 lock_sock(sk);
564 sk->sk_bound_dev_if = index;
565 sk_dst_reset(sk);
566 release_sock(sk);
567
568 ret = 0;
569
570out:
571#endif
572
573 return ret;
574}
575
c0ef877b
PE		 576static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
577{
578 if (valbool)
579 sock_set_flag(sk, bit);
580 else
581 sock_reset_flag(sk, bit);
582}
583
1da177e4
LT		 584/*
585 * This is meant for all protocols to use and covers goings on
586 * at the socket level. Everything here is generic.
587 */
588
589int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 590 char __user *optval, unsigned int optlen)
1da177e4 591{
2a91525c 592 struct sock *sk = sock->sk;
1da177e4
LT		 593 int val;
594 int valbool;
595 struct linger ling;
596 int ret = 0;
4ec93edb 597
1da177e4
LT		 598 /*
599 * Options without arguments
600 */
601
4878809f
DM		 602 if (optname == SO_BINDTODEVICE)
603 return sock_bindtodevice(sk, optval, optlen);
604
e71a4783
SH		 605 if (optlen < sizeof(int))
606 return -EINVAL;
4ec93edb 607
1da177e4
LT		 608 if (get_user(val, (int __user *)optval))
609 return -EFAULT;
4ec93edb 610
2a91525c 611 valbool = val ? 1 : 0;
1da177e4
LT		 612
613 lock_sock(sk);
614
2a91525c 615 switch (optname) {
e71a4783 616 case SO_DEBUG:
2a91525c 617 if (val && !capable(CAP_NET_ADMIN))
e71a4783 618 ret = -EACCES;
2a91525c 619 else
c0ef877b 620 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 621 break;
622 case SO_REUSEADDR:
4a17fd52 623 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
e71a4783
SH		 624 break;
625 case SO_TYPE:
49c794e9 626 case SO_PROTOCOL:
0d6038ee 627 case SO_DOMAIN:
e71a4783
SH		 628 case SO_ERROR:
629 ret = -ENOPROTOOPT;
630 break;
631 case SO_DONTROUTE:
c0ef877b 632 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 633 break;
634 case SO_BROADCAST:
635 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
636 break;
637 case SO_SNDBUF:
638 /* Don't error on this BSD doesn't and if you think
82981930
ED		 639 * about it this is right. Otherwise apps have to
640 * play 'guess the biggest size' games. RCVBUF/SNDBUF
641 * are treated in BSD as hints
642 */
643 val = min_t(u32, val, sysctl_wmem_max);
b0573dea 644set_sndbuf:
e71a4783 645 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
82981930
ED		 646 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
647 /* Wake up sending tasks if we upped the value. */
e71a4783
SH		 648 sk->sk_write_space(sk);
649 break;
1da177e4 650
e71a4783
SH		 651 case SO_SNDBUFFORCE:
652 if (!capable(CAP_NET_ADMIN)) {
653 ret = -EPERM;
654 break;
655 }
656 goto set_sndbuf;
b0573dea 657
e71a4783
SH		 658 case SO_RCVBUF:
659 /* Don't error on this BSD doesn't and if you think
82981930
ED		 660 * about it this is right. Otherwise apps have to
661 * play 'guess the biggest size' games. RCVBUF/SNDBUF
662 * are treated in BSD as hints
663 */
664 val = min_t(u32, val, sysctl_rmem_max);
b0573dea 665set_rcvbuf:
e71a4783
SH		 666 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
667 /*
668 * We double it on the way in to account for
669 * "struct sk_buff" etc. overhead. Applications
670 * assume that the SO_RCVBUF setting they make will
671 * allow that much actual data to be received on that
672 * socket.
673 *
674 * Applications are unaware that "struct sk_buff" and
675 * other overheads allocate from the receive buffer
676 * during socket buffer allocation.
677 *
678 * And after considering the possible alternatives,
679 * returning the value we actually used in getsockopt
680 * is the most desirable behavior.
681 */
82981930 682 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
e71a4783
SH		 683 break;
684
685 case SO_RCVBUFFORCE:
686 if (!capable(CAP_NET_ADMIN)) {
687 ret = -EPERM;
1da177e4 688 break;
e71a4783
SH		 689 }
690 goto set_rcvbuf;
1da177e4 691
e71a4783 692 case SO_KEEPALIVE:
1da177e4 693#ifdef CONFIG_INET
e71a4783
SH		 694 if (sk->sk_protocol == IPPROTO_TCP)
695 tcp_set_keepalive(sk, valbool);
1da177e4 696#endif
e71a4783
SH		 697 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
698 break;
699
700 case SO_OOBINLINE:
701 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
702 break;
703
704 case SO_NO_CHECK:
705 sk->sk_no_check = valbool;
706 break;
707
708 case SO_PRIORITY:
709 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
710 sk->sk_priority = val;
711 else
712 ret = -EPERM;
713 break;
714
715 case SO_LINGER:
716 if (optlen < sizeof(ling)) {
717 ret = -EINVAL; /* 1003.1g */
1da177e4 718 break;
e71a4783 719 }
2a91525c 720 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 721 ret = -EFAULT;
1da177e4 722 break;
e71a4783
SH		 723 }
724 if (!ling.l_onoff)
725 sock_reset_flag(sk, SOCK_LINGER);
726 else {
1da177e4 727#if (BITS_PER_LONG == 32)
e71a4783
SH		 728 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
729 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 730 else
e71a4783
SH		 731#endif
732 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
733 sock_set_flag(sk, SOCK_LINGER);
734 }
735 break;
736
737 case SO_BSDCOMPAT:
738 sock_warn_obsolete_bsdism("setsockopt");
739 break;
740
741 case SO_PASSCRED:
742 if (valbool)
743 set_bit(SOCK_PASSCRED, &sock->flags);
744 else
745 clear_bit(SOCK_PASSCRED, &sock->flags);
746 break;
747
748 case SO_TIMESTAMP:
92f37fd2 749 case SO_TIMESTAMPNS:
e71a4783 750 if (valbool) {
92f37fd2
ED		 751 if (optname == SO_TIMESTAMP)
752 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
753 else
754 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 755 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 756 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 757 } else {
e71a4783 758 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 759 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
760 }
e71a4783
SH		 761 break;
762
20d49473
PO		 763 case SO_TIMESTAMPING:
764 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 765 ret = -EINVAL;
20d49473
PO		 766 break;
767 }
768 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
769 val & SOF_TIMESTAMPING_TX_HARDWARE);
770 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
771 val & SOF_TIMESTAMPING_TX_SOFTWARE);
772 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
773 val & SOF_TIMESTAMPING_RX_HARDWARE);
774 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
775 sock_enable_timestamp(sk,
776 SOCK_TIMESTAMPING_RX_SOFTWARE);
777 else
778 sock_disable_timestamp(sk,
08e29af3 779 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
20d49473
PO		 780 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
781 val & SOF_TIMESTAMPING_SOFTWARE);
782 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
783 val & SOF_TIMESTAMPING_SYS_HARDWARE);
784 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
785 val & SOF_TIMESTAMPING_RAW_HARDWARE);
786 break;
787
e71a4783
SH		 788 case SO_RCVLOWAT:
789 if (val < 0)
790 val = INT_MAX;
791 sk->sk_rcvlowat = val ? : 1;
792 break;
793
794 case SO_RCVTIMEO:
795 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
796 break;
797
798 case SO_SNDTIMEO:
799 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
800 break;
1da177e4 801
e71a4783
SH		 802 case SO_ATTACH_FILTER:
803 ret = -EINVAL;
804 if (optlen == sizeof(struct sock_fprog)) {
805 struct sock_fprog fprog;
1da177e4 806
e71a4783
SH		 807 ret = -EFAULT;
808 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 809 break;
e71a4783
SH		 810
811 ret = sk_attach_filter(&fprog, sk);
812 }
813 break;
814
815 case SO_DETACH_FILTER:
55b33325 816 ret = sk_detach_filter(sk);
e71a4783 817 break;
1da177e4 818
e71a4783
SH		 819 case SO_PASSSEC:
820 if (valbool)
821 set_bit(SOCK_PASSSEC, &sock->flags);
822 else
823 clear_bit(SOCK_PASSSEC, &sock->flags);
824 break;
4a19ec58
LAT		 825 case SO_MARK:
826 if (!capable(CAP_NET_ADMIN))
827 ret = -EPERM;
2a91525c 828 else
4a19ec58 829 sk->sk_mark = val;
4a19ec58 830 break;
877ce7c1 831
1da177e4
LT		 832 /* We implement the SO_SNDLOWAT etc to
833 not be settable (1003.1g 5.3) */
3b885787 834 case SO_RXQ_OVFL:
8083f0fc 835 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 836 break;
6e3e939f
JB		 837
838 case SO_WIFI_STATUS:
839 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
840 break;
841
ef64a54f
PE		 842 case SO_PEEK_OFF:
843 if (sock->ops->set_peek_off)
844 sock->ops->set_peek_off(sk, val);
845 else
846 ret = -EOPNOTSUPP;
847 break;
3bdc0eba
BG		 848
849 case SO_NOFCS:
850 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
851 break;
852
e71a4783
SH		 853 default:
854 ret = -ENOPROTOOPT;
855 break;
4ec93edb 856 }
1da177e4
LT		 857 release_sock(sk);
858 return ret;
859}
2a91525c 860EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 861
862
3f551f94
EB		 863void cred_to_ucred(struct pid *pid, const struct cred *cred,
864 struct ucred *ucred)
865{
866 ucred->pid = pid_vnr(pid);
867 ucred->uid = ucred->gid = -1;
868 if (cred) {
869 struct user_namespace *current_ns = current_user_ns();
870
76b6db01
EB		 871 ucred->uid = from_kuid(current_ns, cred->euid);
872 ucred->gid = from_kgid(current_ns, cred->egid);
3f551f94
EB		 873 }
874}
3924773a 875EXPORT_SYMBOL_GPL(cred_to_ucred);
3f551f94 876
1da177e4
LT		 877int sock_getsockopt(struct socket *sock, int level, int optname,
878 char __user *optval, int __user *optlen)
879{
880 struct sock *sk = sock->sk;
4ec93edb 881
e71a4783 882 union {
4ec93edb
YH		 883 int val;
884 struct linger ling;
1da177e4
LT		 885 struct timeval tm;
886 } v;
4ec93edb 887
4d0392be 888 int lv = sizeof(int);
1da177e4 889 int len;
4ec93edb 890
e71a4783 891 if (get_user(len, optlen))
4ec93edb 892 return -EFAULT;
e71a4783 893 if (len < 0)
1da177e4 894 return -EINVAL;
4ec93edb 895
50fee1de 896 memset(&v, 0, sizeof(v));
df0bca04 897
2a91525c 898 switch (optname) {
e71a4783
SH		 899 case SO_DEBUG:
900 v.val = sock_flag(sk, SOCK_DBG);
901 break;
902
903 case SO_DONTROUTE:
904 v.val = sock_flag(sk, SOCK_LOCALROUTE);
905 break;
906
907 case SO_BROADCAST:
1b23a5df 908 v.val = sock_flag(sk, SOCK_BROADCAST);
e71a4783
SH		 909 break;
910
911 case SO_SNDBUF:
912 v.val = sk->sk_sndbuf;
913 break;
914
915 case SO_RCVBUF:
916 v.val = sk->sk_rcvbuf;
917 break;
918
919 case SO_REUSEADDR:
920 v.val = sk->sk_reuse;
921 break;
922
923 case SO_KEEPALIVE:
1b23a5df 924 v.val = sock_flag(sk, SOCK_KEEPOPEN);
e71a4783
SH		 925 break;
926
927 case SO_TYPE:
928 v.val = sk->sk_type;
929 break;
930
49c794e9
JE		 931 case SO_PROTOCOL:
932 v.val = sk->sk_protocol;
933 break;
934
0d6038ee
JE		 935 case SO_DOMAIN:
936 v.val = sk->sk_family;
937 break;
938
e71a4783
SH		 939 case SO_ERROR:
940 v.val = -sock_error(sk);
2a91525c 941 if (v.val == 0)
e71a4783
SH		 942 v.val = xchg(&sk->sk_err_soft, 0);
943 break;
944
945 case SO_OOBINLINE:
1b23a5df 946 v.val = sock_flag(sk, SOCK_URGINLINE);
e71a4783
SH		 947 break;
948
949 case SO_NO_CHECK:
950 v.val = sk->sk_no_check;
951 break;
952
953 case SO_PRIORITY:
954 v.val = sk->sk_priority;
955 break;
956
957 case SO_LINGER:
958 lv = sizeof(v.ling);
1b23a5df 959 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
e71a4783
SH		 960 v.ling.l_linger = sk->sk_lingertime / HZ;
961 break;
962
963 case SO_BSDCOMPAT:
964 sock_warn_obsolete_bsdism("getsockopt");
965 break;
966
967 case SO_TIMESTAMP:
92f37fd2
ED		 968 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
969 !sock_flag(sk, SOCK_RCVTSTAMPNS);
970 break;
971
972 case SO_TIMESTAMPNS:
973 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 974 break;
975
20d49473
PO		 976 case SO_TIMESTAMPING:
977 v.val = 0;
978 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
979 v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
980 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
981 v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
982 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
983 v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
984 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
985 v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
986 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
987 v.val |= SOF_TIMESTAMPING_SOFTWARE;
988 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
989 v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
990 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
991 v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
992 break;
993
e71a4783 994 case SO_RCVTIMEO:
2a91525c 995 lv = sizeof(struct timeval);
e71a4783
SH		 996 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
997 v.tm.tv_sec = 0;
998 v.tm.tv_usec = 0;
999 } else {
1000 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1001 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1002 }
1003 break;
1004
1005 case SO_SNDTIMEO:
2a91525c 1006 lv = sizeof(struct timeval);
e71a4783
SH		 1007 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1008 v.tm.tv_sec = 0;
1009 v.tm.tv_usec = 0;
1010 } else {
1011 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1012 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1013 }
1014 break;
1da177e4 1015
e71a4783
SH		 1016 case SO_RCVLOWAT:
1017 v.val = sk->sk_rcvlowat;
1018 break;
1da177e4 1019
e71a4783 1020 case SO_SNDLOWAT:
2a91525c 1021 v.val = 1;
e71a4783 1022 break;
1da177e4 1023
e71a4783 1024 case SO_PASSCRED:
82981930 1025 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
e71a4783 1026 break;
1da177e4 1027
e71a4783 1028 case SO_PEERCRED:
109f6e39
EB		 1029 {
1030 struct ucred peercred;
1031 if (len > sizeof(peercred))
1032 len = sizeof(peercred);
1033 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1034 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 1035 return -EFAULT;
1036 goto lenout;
109f6e39 1037 }
1da177e4 1038
e71a4783
SH		 1039 case SO_PEERNAME:
1040 {
1041 char address[128];
1042
1043 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1044 return -ENOTCONN;
1045 if (lv < len)
1046 return -EINVAL;
1047 if (copy_to_user(optval, address, len))
1048 return -EFAULT;
1049 goto lenout;
1050 }
1da177e4 1051
e71a4783
SH		 1052 /* Dubious BSD thing... Probably nobody even uses it, but
1053 * the UNIX standard wants it for whatever reason... -DaveM
1054 */
1055 case SO_ACCEPTCONN:
1056 v.val = sk->sk_state == TCP_LISTEN;
1057 break;
1da177e4 1058
e71a4783 1059 case SO_PASSSEC:
82981930 1060 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
e71a4783 1061 break;
877ce7c1 1062
e71a4783
SH		 1063 case SO_PEERSEC:
1064 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 1065
4a19ec58
LAT		 1066 case SO_MARK:
1067 v.val = sk->sk_mark;
1068 break;
1069
3b885787 1070 case SO_RXQ_OVFL:
1b23a5df 1071 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
3b885787
NH		 1072 break;
1073
6e3e939f 1074 case SO_WIFI_STATUS:
1b23a5df 1075 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
6e3e939f
JB		 1076 break;
1077
ef64a54f
PE		 1078 case SO_PEEK_OFF:
1079 if (!sock->ops->set_peek_off)
1080 return -EOPNOTSUPP;
1081
1082 v.val = sk->sk_peek_off;
1083 break;
bc2f7996 1084 case SO_NOFCS:
1b23a5df 1085 v.val = sock_flag(sk, SOCK_NOFCS);
bc2f7996 1086 break;
e71a4783
SH		 1087 default:
1088 return -ENOPROTOOPT;
1da177e4 1089 }
e71a4783 1090
1da177e4
LT		 1091 if (len > lv)
1092 len = lv;
1093 if (copy_to_user(optval, &v, len))
1094 return -EFAULT;
1095lenout:
4ec93edb
YH		 1096 if (put_user(len, optlen))
1097 return -EFAULT;
1098 return 0;
1da177e4
LT		 1099}
1100
a5b5bb9a
IM		 1101/*
1102 * Initialize an sk_lock.
1103 *
1104 * (We also register the sk_lock with the lock validator.)
1105 */
b6f99a21 1106static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 1107{
ed07536e
PZ		 1108 sock_lock_init_class_and_name(sk,
1109 af_family_slock_key_strings[sk->sk_family],
1110 af_family_slock_keys + sk->sk_family,
1111 af_family_key_strings[sk->sk_family],
1112 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 1113}
1114
4dc6dc71
ED		 1115/*
1116 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1117 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1118 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1119 */
f1a6c4da
PE		 1120static void sock_copy(struct sock *nsk, const struct sock *osk)
1121{
1122#ifdef CONFIG_SECURITY_NETWORK
1123 void *sptr = nsk->sk_security;
1124#endif
68835aba
ED		 1125 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1126
1127 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1128 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1129
f1a6c4da
PE		 1130#ifdef CONFIG_SECURITY_NETWORK
1131 nsk->sk_security = sptr;
1132 security_sk_clone(osk, nsk);
1133#endif
1134}
1135
fcbdf09d
OP		 1136/*
1137 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1138 * un-modified. Special care is taken when initializing object to zero.
1139 */
1140static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1141{
1142 if (offsetof(struct sock, sk_node.next) != 0)
1143 memset(sk, 0, offsetof(struct sock, sk_node.next));
1144 memset(&sk->sk_node.pprev, 0,
1145 size - offsetof(struct sock, sk_node.pprev));
1146}
1147
1148void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1149{
1150 unsigned long nulls1, nulls2;
1151
1152 nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1153 nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1154 if (nulls1 > nulls2)
1155 swap(nulls1, nulls2);
1156
1157 if (nulls1 != 0)
1158 memset((char *)sk, 0, nulls1);
1159 memset((char *)sk + nulls1 + sizeof(void *), 0,
1160 nulls2 - nulls1 - sizeof(void *));
1161 memset((char *)sk + nulls2 + sizeof(void *), 0,
1162 size - nulls2 - sizeof(void *));
1163}
1164EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1165
2e4afe7b
PE		 1166static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1167 int family)
c308c1b2
PE		 1168{
1169 struct sock *sk;
1170 struct kmem_cache *slab;
1171
1172 slab = prot->slab;
e912b114
ED		 1173 if (slab != NULL) {
1174 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1175 if (!sk)
1176 return sk;
1177 if (priority & __GFP_ZERO) {
fcbdf09d
OP		 1178 if (prot->clear_sk)
1179 prot->clear_sk(sk, prot->obj_size);
1180 else
1181 sk_prot_clear_nulls(sk, prot->obj_size);
e912b114 1182 }
fcbdf09d 1183 } else
c308c1b2
PE		 1184 sk = kmalloc(prot->obj_size, priority);
1185
2e4afe7b 1186 if (sk != NULL) {
a98b65a3
VN		 1187 kmemcheck_annotate_bitfield(sk, flags);
1188
2e4afe7b
PE		 1189 if (security_sk_alloc(sk, family, priority))
1190 goto out_free;
1191
1192 if (!try_module_get(prot->owner))
1193 goto out_free_sec;
e022f0b4 1194 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1195 }
1196
c308c1b2 1197 return sk;
2e4afe7b
PE		 1198
1199out_free_sec:
1200 security_sk_free(sk);
1201out_free:
1202 if (slab != NULL)
1203 kmem_cache_free(slab, sk);
1204 else
1205 kfree(sk);
1206 return NULL;
c308c1b2
PE		 1207}
1208
1209static void sk_prot_free(struct proto *prot, struct sock *sk)
1210{
1211 struct kmem_cache *slab;
2e4afe7b 1212 struct module *owner;
c308c1b2 1213
2e4afe7b 1214 owner = prot->owner;
c308c1b2 1215 slab = prot->slab;
2e4afe7b
PE		 1216
1217 security_sk_free(sk);
c308c1b2
PE		 1218 if (slab != NULL)
1219 kmem_cache_free(slab, sk);
1220 else
1221 kfree(sk);
2e4afe7b 1222 module_put(owner);
c308c1b2
PE		 1223}
1224
f8451725
HX		 1225#ifdef CONFIG_CGROUPS
1226void sock_update_classid(struct sock *sk)
1227{
1144182a 1228 u32 classid;
f8451725 1229
1144182a
PM		 1230 rcu_read_lock(); /* doing current task, which cannot vanish. */
1231 classid = task_cls_classid(current);
1232 rcu_read_unlock();
f8451725
HX		 1233 if (classid && classid != sk->sk_classid)
1234 sk->sk_classid = classid;
1235}
82862742 1236EXPORT_SYMBOL(sock_update_classid);
5bc1421e 1237
406a3c63 1238void sock_update_netprioidx(struct sock *sk, struct task_struct *task)
5bc1421e 1239{
5bc1421e
NH		 1240 if (in_interrupt())
1241 return;
2b73bc65 1242
406a3c63 1243 sk->sk_cgrp_prioidx = task_netprioidx(task);
5bc1421e
NH		 1244}
1245EXPORT_SYMBOL_GPL(sock_update_netprioidx);
f8451725
HX		 1246#endif
1247
1da177e4
LT		 1248/**
1249 * sk_alloc - All socket objects are allocated here
c4ea43c5 1250 * @net: the applicable net namespace
4dc3b16b
PP		 1251 * @family: protocol family
1252 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1253 * @prot: struct proto associated with this new sock instance
1da177e4 1254 */
1b8d7ae4 1255struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
6257ff21 1256 struct proto *prot)
1da177e4 1257{
c308c1b2 1258 struct sock *sk;
1da177e4 1259
154adbc8 1260 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1261 if (sk) {
154adbc8
PE		 1262 sk->sk_family = family;
1263 /*
1264 * See comment in struct sock definition to understand
1265 * why we need sk_prot_creator -acme
1266 */
1267 sk->sk_prot = sk->sk_prot_creator = prot;
1268 sock_lock_init(sk);
3b1e0a65 1269 sock_net_set(sk, get_net(net));
d66ee058 1270 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725
HX		 1271
1272 sock_update_classid(sk);
406a3c63 1273 sock_update_netprioidx(sk, current);
1da177e4 1274 }
a79af59e 1275
2e4afe7b 1276 return sk;
1da177e4 1277}
2a91525c 1278EXPORT_SYMBOL(sk_alloc);
1da177e4 1279
2b85a34e 1280static void __sk_free(struct sock *sk)
1da177e4
LT		 1281{
1282 struct sk_filter *filter;
1da177e4
LT		 1283
1284 if (sk->sk_destruct)
1285 sk->sk_destruct(sk);
1286
a898def2
PM		 1287 filter = rcu_dereference_check(sk->sk_filter,
1288 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1289 if (filter) {
309dd5fc 1290 sk_filter_uncharge(sk, filter);
a9b3cd7f 1291 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4
LT		 1292 }
1293
08e29af3 1294 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4
LT		 1295
1296 if (atomic_read(&sk->sk_omem_alloc))
e005d193
JP		 1297 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1298 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1299
109f6e39
EB		 1300 if (sk->sk_peer_cred)
1301 put_cred(sk->sk_peer_cred);
1302 put_pid(sk->sk_peer_pid);
3b1e0a65 1303 put_net(sock_net(sk));
c308c1b2 1304 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1305}
2b85a34e
ED		 1306
1307void sk_free(struct sock *sk)
1308{
1309 /*
25985edc 1310 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1311 * some packets are still in some tx queue.
1312 * If not null, sock_wfree() will call __sk_free(sk) later
1313 */
1314 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1315 __sk_free(sk);
1316}
2a91525c 1317EXPORT_SYMBOL(sk_free);
1da177e4 1318
edf02087 1319/*
25985edc
LDM		 1320 * Last sock_put should drop reference to sk->sk_net. It has already
1321 * been dropped in sk_change_net. Taking reference to stopping namespace
edf02087 1322 * is not an option.
25985edc 1323 * Take reference to a socket to remove it from hash _alive_ and after that
edf02087
DL		 1324 * destroy it in the context of init_net.
1325 */
1326void sk_release_kernel(struct sock *sk)
1327{
1328 if (sk == NULL || sk->sk_socket == NULL)
1329 return;
1330
1331 sock_hold(sk);
1332 sock_release(sk->sk_socket);
65a18ec5 1333 release_net(sock_net(sk));
3b1e0a65 1334 sock_net_set(sk, get_net(&init_net));
edf02087
DL		 1335 sock_put(sk);
1336}
45af1754 1337EXPORT_SYMBOL(sk_release_kernel);
edf02087 1338
475f1b52
SR		 1339static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1340{
1341 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1342 sock_update_memcg(newsk);
1343}
1344
e56c57d0
ED		 1345/**
1346 * sk_clone_lock - clone a socket, and lock its clone
1347 * @sk: the socket to clone
1348 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1349 *
1350 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1351 */
1352struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1353{
8fd1d178 1354 struct sock *newsk;
87d11ceb 1355
8fd1d178 1356 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1357 if (newsk != NULL) {
1358 struct sk_filter *filter;
1359
892c141e 1360 sock_copy(newsk, sk);
87d11ceb
ACM		 1361
1362 /* SANITY */
3b1e0a65 1363 get_net(sock_net(newsk));
87d11ceb
ACM		 1364 sk_node_init(&newsk->sk_node);
1365 sock_lock_init(newsk);
1366 bh_lock_sock(newsk);
fa438ccf 1367 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1368 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1369
1370 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1371 /*
1372 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1373 */
1374 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1375 atomic_set(&newsk->sk_omem_alloc, 0);
1376 skb_queue_head_init(&newsk->sk_receive_queue);
1377 skb_queue_head_init(&newsk->sk_write_queue);
97fc2f08
CL		 1378#ifdef CONFIG_NET_DMA
1379 skb_queue_head_init(&newsk->sk_async_wait_queue);
1380#endif
87d11ceb 1381
b6c6712a 1382 spin_lock_init(&newsk->sk_dst_lock);
87d11ceb 1383 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1384 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1385 af_callback_keys + newsk->sk_family,
1386 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1387
1388 newsk->sk_dst_cache = NULL;
1389 newsk->sk_wmem_queued = 0;
1390 newsk->sk_forward_alloc = 0;
1391 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1392 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1393
1394 sock_reset_flag(newsk, SOCK_DONE);
1395 skb_queue_head_init(&newsk->sk_error_queue);
1396
0d7da9dd 1397 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb
ACM		 1398 if (filter != NULL)
1399 sk_filter_charge(newsk, filter);
1400
1401 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1402 /* It is still raw copy of parent, so invalidate
1403 * destructor and make plain sk_free() */
1404 newsk->sk_destruct = NULL;
b0691c8e 1405 bh_unlock_sock(newsk);
87d11ceb
ACM		 1406 sk_free(newsk);
1407 newsk = NULL;
1408 goto out;
1409 }
1410
1411 newsk->sk_err = 0;
1412 newsk->sk_priority = 0;
4dc6dc71
ED		 1413 /*
1414 * Before updating sk_refcnt, we must commit prior changes to memory
1415 * (Documentation/RCU/rculist_nulls.txt for details)
1416 */
1417 smp_wmb();
87d11ceb
ACM		 1418 atomic_set(&newsk->sk_refcnt, 2);
1419
1420 /*
1421 * Increment the counter in the same struct proto as the master
1422 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1423 * is the same as sk->sk_prot->socks, as this field was copied
1424 * with memcpy).
1425 *
1426 * This _changes_ the previous behaviour, where
1427 * tcp_create_openreq_child always was incrementing the
1428 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1429 * to be taken into account in all callers. -acme
1430 */
1431 sk_refcnt_debug_inc(newsk);
972692e0 1432 sk_set_socket(newsk, NULL);
43815482 1433 newsk->sk_wq = NULL;
87d11ceb 1434
f3f511e1
GC		 1435 sk_update_clone(sk, newsk);
1436
87d11ceb 1437 if (newsk->sk_prot->sockets_allocated)
180d8cd9 1438 sk_sockets_allocated_inc(newsk);
704da560 1439
08e29af3 1440 if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1441 net_enable_timestamp();
87d11ceb
ACM		 1442 }
1443out:
1444 return newsk;
1445}
e56c57d0 1446EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1447
9958089a
AK		 1448void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1449{
1450 __sk_dst_set(sk, dst);
1451 sk->sk_route_caps = dst->dev->features;
1452 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1453 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1454 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1455 if (sk_can_gso(sk)) {
82cc1a7a 1456 if (dst->header_len) {
9958089a 1457 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1458 } else {
9958089a 1459 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a 1460 sk->sk_gso_max_size = dst->dev->gso_max_size;
1485348d 1461 sk->sk_gso_max_segs = dst->dev->gso_max_segs;
82cc1a7a 1462 }
9958089a
AK		 1463 }
1464}
1465EXPORT_SYMBOL_GPL(sk_setup_caps);
1466
1da177e4
LT		 1467void __init sk_init(void)
1468{
4481374c 1469 if (totalram_pages <= 4096) {
1da177e4
LT		 1470 sysctl_wmem_max = 32767;
1471 sysctl_rmem_max = 32767;
1472 sysctl_wmem_default = 32767;
1473 sysctl_rmem_default = 32767;
4481374c 1474 } else if (totalram_pages >= 131072) {
1da177e4
LT		 1475 sysctl_wmem_max = 131071;
1476 sysctl_rmem_max = 131071;
1477 }
1478}
1479
1480/*
1481 * Simple resource managers for sockets.
1482 */
1483
1484
4ec93edb
YH		 1485/*
1486 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1487 */
1488void sock_wfree(struct sk_buff *skb)
1489{
1490 struct sock *sk = skb->sk;
d99927f4 1491 unsigned int len = skb->truesize;
1da177e4 1492
d99927f4
ED		 1493 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1494 /*
1495 * Keep a reference on sk_wmem_alloc, this will be released
1496 * after sk_write_space() call
1497 */
1498 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1499 sk->sk_write_space(sk);
d99927f4
ED		 1500 len = 1;
1501 }
2b85a34e 1502 /*
d99927f4
ED		 1503 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1504 * could not do because of in-flight packets
2b85a34e 1505 */
d99927f4 1506 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1507 __sk_free(sk);
1da177e4 1508}
2a91525c 1509EXPORT_SYMBOL(sock_wfree);
1da177e4 1510
4ec93edb
YH		 1511/*
1512 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1513 */
1514void sock_rfree(struct sk_buff *skb)
1515{
1516 struct sock *sk = skb->sk;
d361fd59 1517 unsigned int len = skb->truesize;
1da177e4 1518
d361fd59
ED		 1519 atomic_sub(len, &sk->sk_rmem_alloc);
1520 sk_mem_uncharge(sk, len);
1da177e4 1521}
2a91525c 1522EXPORT_SYMBOL(sock_rfree);
1da177e4 1523
41063e9d
DM		 1524void sock_edemux(struct sk_buff *skb)
1525{
1526 sock_put(skb->sk);
1527}
1528EXPORT_SYMBOL(sock_edemux);
1da177e4
LT		 1529
1530int sock_i_uid(struct sock *sk)
1531{
1532 int uid;
1533
f064af1e 1534 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1535 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
f064af1e 1536 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1537 return uid;
1538}
2a91525c 1539EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1540
1541unsigned long sock_i_ino(struct sock *sk)
1542{
1543 unsigned long ino;
1544
f064af1e 1545 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1546 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1547 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1548 return ino;
1549}
2a91525c 1550EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1551
1552/*
1553 * Allocate a skb from the socket's send buffer.
1554 */
86a76caf 1555struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1556 gfp_t priority)
1da177e4
LT		 1557{
1558 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1559 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1560 if (skb) {
1561 skb_set_owner_w(skb, sk);
1562 return skb;
1563 }
1564 }
1565 return NULL;
1566}
2a91525c 1567EXPORT_SYMBOL(sock_wmalloc);
1da177e4
LT		 1568
1569/*
1570 * Allocate a skb from the socket's receive buffer.
4ec93edb 1571 */
86a76caf 1572struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1573 gfp_t priority)
1da177e4
LT		 1574{
1575 if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1576 struct sk_buff *skb = alloc_skb(size, priority);
1577 if (skb) {
1578 skb_set_owner_r(skb, sk);
1579 return skb;
1580 }
1581 }
1582 return NULL;
1583}
1584
4ec93edb 1585/*
1da177e4 1586 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1587 */
dd0fc66f 1588void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4 1589{
95c96174 1590 if ((unsigned int)size <= sysctl_optmem_max &&
1da177e4
LT		 1591 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1592 void *mem;
1593 /* First do the add, to avoid the race if kmalloc
4ec93edb 1594 * might sleep.
1da177e4
LT		 1595 */
1596 atomic_add(size, &sk->sk_omem_alloc);
1597 mem = kmalloc(size, priority);
1598 if (mem)
1599 return mem;
1600 atomic_sub(size, &sk->sk_omem_alloc);
1601 }
1602 return NULL;
1603}
2a91525c 1604EXPORT_SYMBOL(sock_kmalloc);
1da177e4
LT		 1605
1606/*
1607 * Free an option memory block.
1608 */
1609void sock_kfree_s(struct sock *sk, void *mem, int size)
1610{
1611 kfree(mem);
1612 atomic_sub(size, &sk->sk_omem_alloc);
1613}
2a91525c 1614EXPORT_SYMBOL(sock_kfree_s);
1da177e4
LT		 1615
1616/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1617 I think, these locks should be removed for datagram sockets.
1618 */
2a91525c 1619static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1620{
1621 DEFINE_WAIT(wait);
1622
1623 clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1624 for (;;) {
1625 if (!timeo)
1626 break;
1627 if (signal_pending(current))
1628 break;
1629 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1630 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1631 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1632 break;
1633 if (sk->sk_shutdown & SEND_SHUTDOWN)
1634 break;
1635 if (sk->sk_err)
1636 break;
1637 timeo = schedule_timeout(timeo);
1638 }
aa395145 1639 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1640 return timeo;
1641}
1642
1643
1644/*
1645 * Generic send/receive buffer handlers
1646 */
1647
4cc7f68d
HX		 1648struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1649 unsigned long data_len, int noblock,
1650 int *errcode)
1da177e4
LT		 1651{
1652 struct sk_buff *skb;
7d877f3b 1653 gfp_t gfp_mask;
1da177e4
LT		 1654 long timeo;
1655 int err;
cc9b17ad
JW		 1656 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1657
1658 err = -EMSGSIZE;
1659 if (npages > MAX_SKB_FRAGS)
1660 goto failure;
1da177e4
LT		 1661
1662 gfp_mask = sk->sk_allocation;
1663 if (gfp_mask & __GFP_WAIT)
1664 gfp_mask |= __GFP_REPEAT;
1665
1666 timeo = sock_sndtimeo(sk, noblock);
1667 while (1) {
1668 err = sock_error(sk);
1669 if (err != 0)
1670 goto failure;
1671
1672 err = -EPIPE;
1673 if (sk->sk_shutdown & SEND_SHUTDOWN)
1674 goto failure;
1675
1676 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
db38c179 1677 skb = alloc_skb(header_len, gfp_mask);
1da177e4 1678 if (skb) {
1da177e4
LT		 1679 int i;
1680
1681 /* No pages, we're done... */
1682 if (!data_len)
1683 break;
1684
1da177e4
LT		 1685 skb->truesize += data_len;
1686 skb_shinfo(skb)->nr_frags = npages;
1687 for (i = 0; i < npages; i++) {
1688 struct page *page;
1da177e4
LT		 1689
1690 page = alloc_pages(sk->sk_allocation, 0);
1691 if (!page) {
1692 err = -ENOBUFS;
1693 skb_shinfo(skb)->nr_frags = i;
1694 kfree_skb(skb);
1695 goto failure;
1696 }
1697
ea2ab693
IC		 1698 __skb_fill_page_desc(skb, i,
1699 page, 0,
1700 (data_len >= PAGE_SIZE ?
1701 PAGE_SIZE :
1702 data_len));
1da177e4
LT		 1703 data_len -= PAGE_SIZE;
1704 }
1705
1706 /* Full success... */
1707 break;
1708 }
1709 err = -ENOBUFS;
1710 goto failure;
1711 }
1712 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1713 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1714 err = -EAGAIN;
1715 if (!timeo)
1716 goto failure;
1717 if (signal_pending(current))
1718 goto interrupted;
1719 timeo = sock_wait_for_wmem(sk, timeo);
1720 }
1721
1722 skb_set_owner_w(skb, sk);
1723 return skb;
1724
1725interrupted:
1726 err = sock_intr_errno(timeo);
1727failure:
1728 *errcode = err;
1729 return NULL;
1730}
4cc7f68d 1731EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1732
4ec93edb 1733struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1734 int noblock, int *errcode)
1735{
1736 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1737}
2a91525c 1738EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4
LT		 1739
1740static void __lock_sock(struct sock *sk)
f39234d6
NK		 1741 __releases(&sk->sk_lock.slock)
1742 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1743{
1744 DEFINE_WAIT(wait);
1745
e71a4783 1746 for (;;) {
1da177e4
LT		 1747 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1748 TASK_UNINTERRUPTIBLE);
1749 spin_unlock_bh(&sk->sk_lock.slock);
1750 schedule();
1751 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 1752 if (!sock_owned_by_user(sk))
1da177e4
LT		 1753 break;
1754 }
1755 finish_wait(&sk->sk_lock.wq, &wait);
1756}
1757
1758static void __release_sock(struct sock *sk)
f39234d6
NK		 1759 __releases(&sk->sk_lock.slock)
1760 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 1761{
1762 struct sk_buff *skb = sk->sk_backlog.head;
1763
1764 do {
1765 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1766 bh_unlock_sock(sk);
1767
1768 do {
1769 struct sk_buff *next = skb->next;
1770
e4cbb02a 1771 prefetch(next);
7fee226a 1772 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 1773 skb->next = NULL;
c57943a1 1774 sk_backlog_rcv(sk, skb);
1da177e4
LT		 1775
1776 /*
1777 * We are in process context here with softirqs
1778 * disabled, use cond_resched_softirq() to preempt.
1779 * This is safe to do because we've taken the backlog
1780 * queue private:
1781 */
1782 cond_resched_softirq();
1783
1784 skb = next;
1785 } while (skb != NULL);
1786
1787 bh_lock_sock(sk);
e71a4783 1788 } while ((skb = sk->sk_backlog.head) != NULL);
8eae939f
ZY		 1789
1790 /*
1791 * Doing the zeroing here guarantee we can not loop forever
1792 * while a wild producer attempts to flood us.
1793 */
1794 sk->sk_backlog.len = 0;
1da177e4
LT		 1795}
1796
1797/**
1798 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 1799 * @sk: sock to wait on
1800 * @timeo: for how long
1da177e4
LT		 1801 *
1802 * Now socket state including sk->sk_err is changed only under lock,
1803 * hence we may omit checks after joining wait queue.
1804 * We check receive queue before schedule() only as optimization;
1805 * it is very likely that release_sock() added new data.
1806 */
1807int sk_wait_data(struct sock *sk, long *timeo)
1808{
1809 int rc;
1810 DEFINE_WAIT(wait);
1811
aa395145 1812 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1813 set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1814 rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1815 clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
aa395145 1816 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1817 return rc;
1818}
1da177e4
LT		 1819EXPORT_SYMBOL(sk_wait_data);
1820
3ab224be
HA		 1821/**
1822 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1823 * @sk: socket
1824 * @size: memory size to allocate
1825 * @kind: allocation type
1826 *
1827 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1828 * rmem allocation. This function assumes that protocols which have
1829 * memory_pressure use sk_wmem_queued as write buffer accounting.
1830 */
1831int __sk_mem_schedule(struct sock *sk, int size, int kind)
1832{
1833 struct proto *prot = sk->sk_prot;
1834 int amt = sk_mem_pages(size);
8d987e5c 1835 long allocated;
e1aab161 1836 int parent_status = UNDER_LIMIT;
3ab224be
HA		 1837
1838 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
180d8cd9 1839
e1aab161 1840 allocated = sk_memory_allocated_add(sk, amt, &parent_status);
3ab224be
HA		 1841
1842 /* Under limit. */
e1aab161
GC		 1843 if (parent_status == UNDER_LIMIT &&
1844 allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 1845 sk_leave_memory_pressure(sk);
3ab224be
HA		 1846 return 1;
1847 }
1848
e1aab161
GC		 1849 /* Under pressure. (we or our parents) */
1850 if ((parent_status > SOFT_LIMIT) ||
1851 allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 1852 sk_enter_memory_pressure(sk);
3ab224be 1853
e1aab161
GC		 1854 /* Over hard limit (we or our parents) */
1855 if ((parent_status == OVER_LIMIT) ||
1856 (allocated > sk_prot_mem_limits(sk, 2)))
3ab224be
HA		 1857 goto suppress_allocation;
1858
1859 /* guarantee minimum buffer size under pressure */
1860 if (kind == SK_MEM_RECV) {
1861 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1862 return 1;
180d8cd9 1863
3ab224be
HA		 1864 } else { /* SK_MEM_SEND */
1865 if (sk->sk_type == SOCK_STREAM) {
1866 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1867 return 1;
1868 } else if (atomic_read(&sk->sk_wmem_alloc) <
1869 prot->sysctl_wmem[0])
1870 return 1;
1871 }
1872
180d8cd9 1873 if (sk_has_memory_pressure(sk)) {
1748376b
ED		 1874 int alloc;
1875
180d8cd9 1876 if (!sk_under_memory_pressure(sk))
1748376b 1877 return 1;
180d8cd9
GC		 1878 alloc = sk_sockets_allocated_read_positive(sk);
1879 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 1880 sk_mem_pages(sk->sk_wmem_queued +
1881 atomic_read(&sk->sk_rmem_alloc) +
1882 sk->sk_forward_alloc))
1883 return 1;
1884 }
1885
1886suppress_allocation:
1887
1888 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1889 sk_stream_moderate_sndbuf(sk);
1890
1891 /* Fail only if socket is _under_ its sndbuf.
1892 * In this case we cannot block, so that we have to fail.
1893 */
1894 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1895 return 1;
1896 }
1897
3847ce32
SM		 1898 trace_sock_exceed_buf_limit(sk, prot, allocated);
1899
3ab224be
HA		 1900 /* Alas. Undo changes. */
1901 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
180d8cd9 1902
0e90b31f 1903 sk_memory_allocated_sub(sk, amt);
180d8cd9 1904
3ab224be
HA		 1905 return 0;
1906}
3ab224be
HA		 1907EXPORT_SYMBOL(__sk_mem_schedule);
1908
1909/**
1910 * __sk_reclaim - reclaim memory_allocated
1911 * @sk: socket
1912 */
1913void __sk_mem_reclaim(struct sock *sk)
1914{
180d8cd9 1915 sk_memory_allocated_sub(sk,
0e90b31f 1916 sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
3ab224be
HA		 1917 sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1918
180d8cd9
GC		 1919 if (sk_under_memory_pressure(sk) &&
1920 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1921 sk_leave_memory_pressure(sk);
3ab224be 1922}
3ab224be
HA		 1923EXPORT_SYMBOL(__sk_mem_reclaim);
1924
1925
1da177e4
LT		 1926/*
1927 * Set of default routines for initialising struct proto_ops when
1928 * the protocol does not support a particular function. In certain
1929 * cases where it makes no sense for a protocol to have a "do nothing"
1930 * function, some default processing is provided.
1931 */
1932
1933int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1934{
1935 return -EOPNOTSUPP;
1936}
2a91525c 1937EXPORT_SYMBOL(sock_no_bind);
1da177e4 1938
4ec93edb 1939int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1940 int len, int flags)
1941{
1942 return -EOPNOTSUPP;
1943}
2a91525c 1944EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 1945
1946int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1947{
1948 return -EOPNOTSUPP;
1949}
2a91525c 1950EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 1951
1952int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1953{
1954 return -EOPNOTSUPP;
1955}
2a91525c 1956EXPORT_SYMBOL(sock_no_accept);
1da177e4 1957
4ec93edb 1958int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 1959 int *len, int peer)
1960{
1961 return -EOPNOTSUPP;
1962}
2a91525c 1963EXPORT_SYMBOL(sock_no_getname);
1da177e4 1964
2a91525c 1965unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 1966{
1967 return 0;
1968}
2a91525c 1969EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 1970
1971int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1972{
1973 return -EOPNOTSUPP;
1974}
2a91525c 1975EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 1976
1977int sock_no_listen(struct socket *sock, int backlog)
1978{
1979 return -EOPNOTSUPP;
1980}
2a91525c 1981EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 1982
1983int sock_no_shutdown(struct socket *sock, int how)
1984{
1985 return -EOPNOTSUPP;
1986}
2a91525c 1987EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 1988
1989int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 1990 char __user *optval, unsigned int optlen)
1da177e4
LT		 1991{
1992 return -EOPNOTSUPP;
1993}
2a91525c 1994EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 1995
1996int sock_no_getsockopt(struct socket *sock, int level, int optname,
1997 char __user *optval, int __user *optlen)
1998{
1999 return -EOPNOTSUPP;
2000}
2a91525c 2001EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4
LT		 2002
2003int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2004 size_t len)
2005{
2006 return -EOPNOTSUPP;
2007}
2a91525c 2008EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4
LT		 2009
2010int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2011 size_t len, int flags)
2012{
2013 return -EOPNOTSUPP;
2014}
2a91525c 2015EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 2016
2017int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2018{
2019 /* Mirror missing mmap method error code */
2020 return -ENODEV;
2021}
2a91525c 2022EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 2023
2024ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2025{
2026 ssize_t res;
2027 struct msghdr msg = {.msg_flags = flags};
2028 struct kvec iov;
2029 char *kaddr = kmap(page);
2030 iov.iov_base = kaddr + offset;
2031 iov.iov_len = size;
2032 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2033 kunmap(page);
2034 return res;
2035}
2a91525c 2036EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 2037
2038/*
2039 * Default Socket Callbacks
2040 */
2041
2042static void sock_def_wakeup(struct sock *sk)
2043{
43815482
ED		 2044 struct socket_wq *wq;
2045
2046 rcu_read_lock();
2047 wq = rcu_dereference(sk->sk_wq);
2048 if (wq_has_sleeper(wq))
2049 wake_up_interruptible_all(&wq->wait);
2050 rcu_read_unlock();
1da177e4
LT		 2051}
2052
2053static void sock_def_error_report(struct sock *sk)
2054{
43815482
ED		 2055 struct socket_wq *wq;
2056
2057 rcu_read_lock();
2058 wq = rcu_dereference(sk->sk_wq);
2059 if (wq_has_sleeper(wq))
2060 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 2061 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 2062 rcu_read_unlock();
1da177e4
LT		 2063}
2064
2065static void sock_def_readable(struct sock *sk, int len)
2066{
43815482
ED		 2067 struct socket_wq *wq;
2068
2069 rcu_read_lock();
2070 wq = rcu_dereference(sk->sk_wq);
2071 if (wq_has_sleeper(wq))
2c6607c6 2072 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 2073 POLLRDNORM | POLLRDBAND);
8d8ad9d7 2074 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 2075 rcu_read_unlock();
1da177e4
LT		 2076}
2077
2078static void sock_def_write_space(struct sock *sk)
2079{
43815482
ED		 2080 struct socket_wq *wq;
2081
2082 rcu_read_lock();
1da177e4
LT		 2083
2084 /* Do not wake up a writer until he can make "significant"
2085 * progress. --DaveM
2086 */
e71a4783 2087 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482
ED		 2088 wq = rcu_dereference(sk->sk_wq);
2089 if (wq_has_sleeper(wq))
2090 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 2091 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 2092
2093 /* Should agree with poll, otherwise some programs break */
2094 if (sock_writeable(sk))
8d8ad9d7 2095 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 2096 }
2097
43815482 2098 rcu_read_unlock();
1da177e4
LT		 2099}
2100
2101static void sock_def_destruct(struct sock *sk)
2102{
a51482bd 2103 kfree(sk->sk_protinfo);
1da177e4
LT		 2104}
2105
2106void sk_send_sigurg(struct sock *sk)
2107{
2108 if (sk->sk_socket && sk->sk_socket->file)
2109 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 2110 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 2111}
2a91525c 2112EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 2113
2114void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2115 unsigned long expires)
2116{
2117 if (!mod_timer(timer, expires))
2118 sock_hold(sk);
2119}
1da177e4
LT		 2120EXPORT_SYMBOL(sk_reset_timer);
2121
2122void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2123{
2124 if (timer_pending(timer) && del_timer(timer))
2125 __sock_put(sk);
2126}
1da177e4
LT		 2127EXPORT_SYMBOL(sk_stop_timer);
2128
2129void sock_init_data(struct socket *sock, struct sock *sk)
2130{
2131 skb_queue_head_init(&sk->sk_receive_queue);
2132 skb_queue_head_init(&sk->sk_write_queue);
2133 skb_queue_head_init(&sk->sk_error_queue);
97fc2f08
CL		 2134#ifdef CONFIG_NET_DMA
2135 skb_queue_head_init(&sk->sk_async_wait_queue);
2136#endif
1da177e4
LT		 2137
2138 sk->sk_send_head = NULL;
2139
2140 init_timer(&sk->sk_timer);
4ec93edb 2141
1da177e4
LT		 2142 sk->sk_allocation = GFP_KERNEL;
2143 sk->sk_rcvbuf = sysctl_rmem_default;
2144 sk->sk_sndbuf = sysctl_wmem_default;
2145 sk->sk_state = TCP_CLOSE;
972692e0 2146 sk_set_socket(sk, sock);
1da177e4
LT		 2147
2148 sock_set_flag(sk, SOCK_ZAPPED);
2149
e71a4783 2150 if (sock) {
1da177e4 2151 sk->sk_type = sock->type;
43815482 2152 sk->sk_wq = sock->wq;
1da177e4
LT		 2153 sock->sk = sk;
2154 } else
43815482 2155 sk->sk_wq = NULL;
1da177e4 2156
b6c6712a 2157 spin_lock_init(&sk->sk_dst_lock);
1da177e4 2158 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2159 lockdep_set_class_and_name(&sk->sk_callback_lock,
2160 af_callback_keys + sk->sk_family,
2161 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2162
2163 sk->sk_state_change = sock_def_wakeup;
2164 sk->sk_data_ready = sock_def_readable;
2165 sk->sk_write_space = sock_def_write_space;
2166 sk->sk_error_report = sock_def_error_report;
2167 sk->sk_destruct = sock_def_destruct;
2168
2169 sk->sk_sndmsg_page = NULL;
2170 sk->sk_sndmsg_off = 0;
ef64a54f 2171 sk->sk_peek_off = -1;
1da177e4 2172
109f6e39
EB		 2173 sk->sk_peer_pid = NULL;
2174 sk->sk_peer_cred = NULL;
1da177e4
LT		 2175 sk->sk_write_pending = 0;
2176 sk->sk_rcvlowat = 1;
2177 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2178 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2179
f37f0afb 2180 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2181
4dc6dc71
ED		 2182 /*
2183 * Before updating sk_refcnt, we must commit prior changes to memory
2184 * (Documentation/RCU/rculist_nulls.txt for details)
2185 */
2186 smp_wmb();
1da177e4 2187 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2188 atomic_set(&sk->sk_drops, 0);
1da177e4 2189}
2a91525c 2190EXPORT_SYMBOL(sock_init_data);
1da177e4 2191
b5606c2d 2192void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2193{
2194 might_sleep();
a5b5bb9a 2195 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2196 if (sk->sk_lock.owned)
1da177e4 2197 __lock_sock(sk);
d2e9117c 2198 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2199 spin_unlock(&sk->sk_lock.slock);
2200 /*
2201 * The sk_lock has mutex_lock() semantics here:
2202 */
fcc70d5f 2203 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2204 local_bh_enable();
1da177e4 2205}
fcc70d5f 2206EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2207
b5606c2d 2208void release_sock(struct sock *sk)
1da177e4 2209{
a5b5bb9a
IM		 2210 /*
2211 * The sk_lock has mutex_unlock() semantics:
2212 */
2213 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2214
2215 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2216 if (sk->sk_backlog.tail)
2217 __release_sock(sk);
46d3ceab
ED		 2218
2219 if (sk->sk_prot->release_cb)
2220 sk->sk_prot->release_cb(sk);
2221
d2e9117c 2222 sk->sk_lock.owned = 0;
a5b5bb9a
IM		 2223 if (waitqueue_active(&sk->sk_lock.wq))
2224 wake_up(&sk->sk_lock.wq);
2225 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2226}
2227EXPORT_SYMBOL(release_sock);
2228
8a74ad60
ED		 2229/**
2230 * lock_sock_fast - fast version of lock_sock
2231 * @sk: socket
2232 *
2233 * This version should be used for very small section, where process wont block
2234 * return false if fast path is taken
2235 * sk_lock.slock locked, owned = 0, BH disabled
2236 * return true if slow path is taken
2237 * sk_lock.slock unlocked, owned = 1, BH enabled
2238 */
2239bool lock_sock_fast(struct sock *sk)
2240{
2241 might_sleep();
2242 spin_lock_bh(&sk->sk_lock.slock);
2243
2244 if (!sk->sk_lock.owned)
2245 /*
2246 * Note : We must disable BH
2247 */
2248 return false;
2249
2250 __lock_sock(sk);
2251 sk->sk_lock.owned = 1;
2252 spin_unlock(&sk->sk_lock.slock);
2253 /*
2254 * The sk_lock has mutex_lock() semantics here:
2255 */
2256 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2257 local_bh_enable();
2258 return true;
2259}
2260EXPORT_SYMBOL(lock_sock_fast);
2261
1da177e4 2262int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2263{
b7aa0bf7 2264 struct timeval tv;
1da177e4 2265 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2266 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2267 tv = ktime_to_timeval(sk->sk_stamp);
2268 if (tv.tv_sec == -1)
1da177e4 2269 return -ENOENT;
b7aa0bf7
ED		 2270 if (tv.tv_sec == 0) {
2271 sk->sk_stamp = ktime_get_real();
2272 tv = ktime_to_timeval(sk->sk_stamp);
2273 }
2274 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2275}
1da177e4
LT		 2276EXPORT_SYMBOL(sock_get_timestamp);
2277
ae40eb1e
ED		 2278int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2279{
2280 struct timespec ts;
2281 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2282 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2283 ts = ktime_to_timespec(sk->sk_stamp);
2284 if (ts.tv_sec == -1)
2285 return -ENOENT;
2286 if (ts.tv_sec == 0) {
2287 sk->sk_stamp = ktime_get_real();
2288 ts = ktime_to_timespec(sk->sk_stamp);
2289 }
2290 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2291}
2292EXPORT_SYMBOL(sock_get_timestampns);
2293
20d49473 2294void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2295{
20d49473 2296 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2297 unsigned long previous_flags = sk->sk_flags;
2298
20d49473
PO		 2299 sock_set_flag(sk, flag);
2300 /*
2301 * we just set one of the two flags which require net
2302 * time stamping, but time stamping might have been on
2303 * already because of the other one
2304 */
08e29af3 2305 if (!(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2306 net_enable_timestamp();
1da177e4
LT		 2307 }
2308}
1da177e4
LT		 2309
2310/*
2311 * Get a socket option on an socket.
2312 *
2313 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2314 * asynchronous errors should be reported by getsockopt. We assume
2315 * this means if you specify SO_ERROR (otherwise whats the point of it).
2316 */
2317int sock_common_getsockopt(struct socket *sock, int level, int optname,
2318 char __user *optval, int __user *optlen)
2319{
2320 struct sock *sk = sock->sk;
2321
2322 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2323}
1da177e4
LT		 2324EXPORT_SYMBOL(sock_common_getsockopt);
2325
3fdadf7d 2326#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2327int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2328 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2329{
2330 struct sock *sk = sock->sk;
2331
1e51f951 2332 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2333 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2334 optval, optlen);
3fdadf7d
DM		 2335 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2336}
2337EXPORT_SYMBOL(compat_sock_common_getsockopt);
2338#endif
2339
1da177e4
LT		 2340int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2341 struct msghdr *msg, size_t size, int flags)
2342{
2343 struct sock *sk = sock->sk;
2344 int addr_len = 0;
2345 int err;
2346
2347 err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2348 flags & ~MSG_DONTWAIT, &addr_len);
2349 if (err >= 0)
2350 msg->msg_namelen = addr_len;
2351 return err;
2352}
1da177e4
LT		 2353EXPORT_SYMBOL(sock_common_recvmsg);
2354
2355/*
2356 * Set socket options on an inet socket.
2357 */
2358int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2359 char __user *optval, unsigned int optlen)
1da177e4
LT		 2360{
2361 struct sock *sk = sock->sk;
2362
2363 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2364}
1da177e4
LT		 2365EXPORT_SYMBOL(sock_common_setsockopt);
2366
3fdadf7d 2367#ifdef CONFIG_COMPAT
543d9cfe 2368int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2369 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2370{
2371 struct sock *sk = sock->sk;
2372
543d9cfe
ACM		 2373 if (sk->sk_prot->compat_setsockopt != NULL)
2374 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2375 optval, optlen);
3fdadf7d
DM		 2376 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2377}
2378EXPORT_SYMBOL(compat_sock_common_setsockopt);
2379#endif
2380
1da177e4
LT		 2381void sk_common_release(struct sock *sk)
2382{
2383 if (sk->sk_prot->destroy)
2384 sk->sk_prot->destroy(sk);
2385
2386 /*
2387 * Observation: when sock_common_release is called, processes have
2388 * no access to socket. But net still has.
2389 * Step one, detach it from networking:
2390 *
2391 * A. Remove from hash tables.
2392 */
2393
2394 sk->sk_prot->unhash(sk);
2395
2396 /*
2397 * In this point socket cannot receive new packets, but it is possible
2398 * that some packets are in flight because some CPU runs receiver and
2399 * did hash table lookup before we unhashed socket. They will achieve
2400 * receive queue and will be purged by socket destructor.
2401 *
2402 * Also we still have packets pending on receive queue and probably,
2403 * our own packets waiting in device queues. sock_destroy will drain
2404 * receive queue, but transmitted packets will delay socket destruction
2405 * until the last reference will be released.
2406 */
2407
2408 sock_orphan(sk);
2409
2410 xfrm_sk_free_policy(sk);
2411
e6848976 2412 sk_refcnt_debug_release(sk);
1da177e4
LT		 2413 sock_put(sk);
2414}
1da177e4
LT		 2415EXPORT_SYMBOL(sk_common_release);
2416
13ff3d6f
PE		 2417#ifdef CONFIG_PROC_FS
2418#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2419struct prot_inuse {
2420 int val[PROTO_INUSE_NR];
2421};
13ff3d6f
PE		 2422
2423static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2424
2425#ifdef CONFIG_NET_NS
2426void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2427{
d6d9ca0f 2428 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2429}
2430EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2431
2432int sock_prot_inuse_get(struct net *net, struct proto *prot)
2433{
2434 int cpu, idx = prot->inuse_idx;
2435 int res = 0;
2436
2437 for_each_possible_cpu(cpu)
2438 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2439
2440 return res >= 0 ? res : 0;
2441}
2442EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2443
2c8c1e72 2444static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2445{
2446 net->core.inuse = alloc_percpu(struct prot_inuse);
2447 return net->core.inuse ? 0 : -ENOMEM;
2448}
2449
2c8c1e72 2450static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2451{
2452 free_percpu(net->core.inuse);
2453}
2454
2455static struct pernet_operations net_inuse_ops = {
2456 .init = sock_inuse_init_net,
2457 .exit = sock_inuse_exit_net,
2458};
2459
2460static __init int net_inuse_init(void)
2461{
2462 if (register_pernet_subsys(&net_inuse_ops))
2463 panic("Cannot initialize net inuse counters");
2464
2465 return 0;
2466}
2467
2468core_initcall(net_inuse_init);
2469#else
1338d466
PE		 2470static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2471
c29a0bc4 2472void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2473{
d6d9ca0f 2474 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2475}
2476EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2477
c29a0bc4 2478int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2479{
2480 int cpu, idx = prot->inuse_idx;
2481 int res = 0;
2482
2483 for_each_possible_cpu(cpu)
2484 res += per_cpu(prot_inuse, cpu).val[idx];
2485
2486 return res >= 0 ? res : 0;
2487}
2488EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2489#endif
13ff3d6f
PE		 2490
2491static void assign_proto_idx(struct proto *prot)
2492{
2493 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2494
2495 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
e005d193 2496 pr_err("PROTO_INUSE_NR exhausted\n");
13ff3d6f
PE		 2497 return;
2498 }
2499
2500 set_bit(prot->inuse_idx, proto_inuse_idx);
2501}
2502
2503static void release_proto_idx(struct proto *prot)
2504{
2505 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2506 clear_bit(prot->inuse_idx, proto_inuse_idx);
2507}
2508#else
2509static inline void assign_proto_idx(struct proto *prot)
2510{
2511}
2512
2513static inline void release_proto_idx(struct proto *prot)
2514{
2515}
2516#endif
2517
b733c007
PE		 2518int proto_register(struct proto *prot, int alloc_slab)
2519{
1da177e4
LT		 2520 if (alloc_slab) {
2521 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2522 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2523 NULL);
1da177e4
LT		 2524
2525 if (prot->slab == NULL) {
e005d193
JP		 2526 pr_crit("%s: Can't create sock SLAB cache!\n",
2527 prot->name);
60e7663d 2528 goto out;
1da177e4 2529 }
2e6599cb
ACM		 2530
2531 if (prot->rsk_prot != NULL) {
faf23422 2532 prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
7e56b5d6 2533 if (prot->rsk_prot->slab_name == NULL)
2e6599cb
ACM		 2534 goto out_free_sock_slab;
2535
7e56b5d6 2536 prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2e6599cb 2537 prot->rsk_prot->obj_size, 0,
20c2df83 2538 SLAB_HWCACHE_ALIGN, NULL);
2e6599cb
ACM		 2539
2540 if (prot->rsk_prot->slab == NULL) {
e005d193
JP		 2541 pr_crit("%s: Can't create request sock SLAB cache!\n",
2542 prot->name);
2e6599cb
ACM		 2543 goto out_free_request_sock_slab_name;
2544 }
2545 }
8feaf0c0 2546
6d6ee43e 2547 if (prot->twsk_prot != NULL) {
faf23422 2548 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2549
7e56b5d6 2550 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2551 goto out_free_request_sock_slab;
2552
6d6ee43e 2553 prot->twsk_prot->twsk_slab =
7e56b5d6 2554 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2555 prot->twsk_prot->twsk_obj_size,
3ab5aee7
ED		 2556 0,
2557 SLAB_HWCACHE_ALIGN |
2558 prot->slab_flags,
20c2df83 2559 NULL);
6d6ee43e 2560 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2561 goto out_free_timewait_sock_slab_name;
2562 }
1da177e4
LT		 2563 }
2564
36b77a52 2565 mutex_lock(&proto_list_mutex);
1da177e4 2566 list_add(&prot->node, &proto_list);
13ff3d6f 2567 assign_proto_idx(prot);
36b77a52 2568 mutex_unlock(&proto_list_mutex);
b733c007
PE		 2569 return 0;
2570
8feaf0c0 2571out_free_timewait_sock_slab_name:
7e56b5d6 2572 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0
ACM		 2573out_free_request_sock_slab:
2574 if (prot->rsk_prot && prot->rsk_prot->slab) {
2575 kmem_cache_destroy(prot->rsk_prot->slab);
2576 prot->rsk_prot->slab = NULL;
2577 }
2e6599cb 2578out_free_request_sock_slab_name:
72150e9b
DC		 2579 if (prot->rsk_prot)
2580 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2581out_free_sock_slab:
2582 kmem_cache_destroy(prot->slab);
2583 prot->slab = NULL;
b733c007
PE		 2584out:
2585 return -ENOBUFS;
1da177e4 2586}
1da177e4
LT		 2587EXPORT_SYMBOL(proto_register);
2588
2589void proto_unregister(struct proto *prot)
2590{
36b77a52 2591 mutex_lock(&proto_list_mutex);
13ff3d6f 2592 release_proto_idx(prot);
0a3f4358 2593 list_del(&prot->node);
36b77a52 2594 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2595
2596 if (prot->slab != NULL) {
2597 kmem_cache_destroy(prot->slab);
2598 prot->slab = NULL;
2599 }
2600
2e6599cb 2601 if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2e6599cb 2602 kmem_cache_destroy(prot->rsk_prot->slab);
7e56b5d6 2603 kfree(prot->rsk_prot->slab_name);
2e6599cb
ACM		 2604 prot->rsk_prot->slab = NULL;
2605 }
2606
6d6ee43e 2607 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2608 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2609 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2610 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2611 }
1da177e4 2612}
1da177e4
LT		 2613EXPORT_SYMBOL(proto_unregister);
2614
2615#ifdef CONFIG_PROC_FS
1da177e4 2616static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 2617 __acquires(proto_list_mutex)
1da177e4 2618{
36b77a52 2619 mutex_lock(&proto_list_mutex);
60f0438a 2620 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2621}
2622
2623static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2624{
60f0438a 2625 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2626}
2627
2628static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 2629 __releases(proto_list_mutex)
1da177e4 2630{
36b77a52 2631 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2632}
2633
2634static char proto_method_implemented(const void *method)
2635{
2636 return method == NULL ? 'n' : 'y';
2637}
180d8cd9
GC		 2638static long sock_prot_memory_allocated(struct proto *proto)
2639{
cb75a36c 2640 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
180d8cd9
GC		 2641}
2642
2643static char *sock_prot_memory_pressure(struct proto *proto)
2644{
2645 return proto->memory_pressure != NULL ?
2646 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2647}
1da177e4
LT		 2648
2649static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2650{
180d8cd9 2651
8d987e5c 2652 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2653 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2654 proto->name,
2655 proto->obj_size,
14e943db 2656 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 2657 sock_prot_memory_allocated(proto),
2658 sock_prot_memory_pressure(proto),
1da177e4
LT		 2659 proto->max_header,
2660 proto->slab == NULL ? "no" : "yes",
2661 module_name(proto->owner),
2662 proto_method_implemented(proto->close),
2663 proto_method_implemented(proto->connect),
2664 proto_method_implemented(proto->disconnect),
2665 proto_method_implemented(proto->accept),
2666 proto_method_implemented(proto->ioctl),
2667 proto_method_implemented(proto->init),
2668 proto_method_implemented(proto->destroy),
2669 proto_method_implemented(proto->shutdown),
2670 proto_method_implemented(proto->setsockopt),
2671 proto_method_implemented(proto->getsockopt),
2672 proto_method_implemented(proto->sendmsg),
2673 proto_method_implemented(proto->recvmsg),
2674 proto_method_implemented(proto->sendpage),
2675 proto_method_implemented(proto->bind),
2676 proto_method_implemented(proto->backlog_rcv),
2677 proto_method_implemented(proto->hash),
2678 proto_method_implemented(proto->unhash),
2679 proto_method_implemented(proto->get_port),
2680 proto_method_implemented(proto->enter_memory_pressure));
2681}
2682
2683static int proto_seq_show(struct seq_file *seq, void *v)
2684{
60f0438a 2685 if (v == &proto_list)
1da177e4
LT		 2686 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2687 "protocol",
2688 "size",
2689 "sockets",
2690 "memory",
2691 "press",
2692 "maxhdr",
2693 "slab",
2694 "module",
2695 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2696 else
60f0438a 2697 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 2698 return 0;
2699}
2700
f690808e 2701static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 2702 .start = proto_seq_start,
2703 .next = proto_seq_next,
2704 .stop = proto_seq_stop,
2705 .show = proto_seq_show,
2706};
2707
2708static int proto_seq_open(struct inode *inode, struct file *file)
2709{
14e943db
ED		 2710 return seq_open_net(inode, file, &proto_seq_ops,
2711 sizeof(struct seq_net_private));
1da177e4
LT		 2712}
2713
9a32144e 2714static const struct file_operations proto_seq_fops = {
1da177e4
LT		 2715 .owner = THIS_MODULE,
2716 .open = proto_seq_open,
2717 .read = seq_read,
2718 .llseek = seq_lseek,
14e943db
ED		 2719 .release = seq_release_net,
2720};
2721
2722static __net_init int proto_init_net(struct net *net)
2723{
2724 if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2725 return -ENOMEM;
2726
2727 return 0;
2728}
2729
2730static __net_exit void proto_exit_net(struct net *net)
2731{
2732 proc_net_remove(net, "protocols");
2733}
2734
2735
2736static __net_initdata struct pernet_operations proto_net_ops = {
2737 .init = proto_init_net,
2738 .exit = proto_exit_net,
1da177e4
LT		 2739};
2740
2741static int __init proto_init(void)
2742{
14e943db 2743 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 2744}
2745
2746subsys_initcall(proto_init);
2747
2748#endif /* PROC_FS */
Linux 6.x block layer and io_uring tree(s)