projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tcp: fix return value for partial writes
[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 95#include <linux/errno.h>
cb820f8e 96#include <linux/errqueue.h>
1da177e4
LT		 97#include <linux/types.h>
98#include <linux/socket.h>
99#include <linux/in.h>
100#include <linux/kernel.h>
1da177e4
LT		 101#include <linux/module.h>
102#include <linux/proc_fs.h>
103#include <linux/seq_file.h>
104#include <linux/sched.h>
105#include <linux/timer.h>
106#include <linux/string.h>
107#include <linux/sockios.h>
108#include <linux/net.h>
109#include <linux/mm.h>
110#include <linux/slab.h>
111#include <linux/interrupt.h>
112#include <linux/poll.h>
113#include <linux/tcp.h>
114#include <linux/init.h>
a1f8e7f7 115#include <linux/highmem.h>
3f551f94 116#include <linux/user_namespace.h>
c5905afb 117#include <linux/static_key.h>
3969eb38 118#include <linux/memcontrol.h>
8c1ae10d 119#include <linux/prefetch.h>
1da177e4
LT		 120
121#include <asm/uaccess.h>
1da177e4
LT		 122
123#include <linux/netdevice.h>
124#include <net/protocol.h>
125#include <linux/skbuff.h>
457c4cbc 126#include <net/net_namespace.h>
2e6599cb 127#include <net/request_sock.h>
1da177e4 128#include <net/sock.h>
20d49473 129#include <linux/net_tstamp.h>
1da177e4
LT		 130#include <net/xfrm.h>
131#include <linux/ipsec.h>
f8451725 132#include <net/cls_cgroup.h>
5bc1421e 133#include <net/netprio_cgroup.h>
eb4cb008 134#include <linux/sock_diag.h>
1da177e4
LT		 135
136#include <linux/filter.h>
538950a1 137#include <net/sock_reuseport.h>
1da177e4 138
3847ce32
SM		 139#include <trace/events/sock.h>
140
1da177e4
LT		 141#ifdef CONFIG_INET
142#include <net/tcp.h>
143#endif
144
076bb0c8 145#include <net/busy_poll.h>
06021292 146
36b77a52 147static DEFINE_MUTEX(proto_list_mutex);
d1a4c0b3
GC		 148static LIST_HEAD(proto_list);
149
a3b299da
EB		 150/**
151 * sk_ns_capable - General socket capability test
152 * @sk: Socket to use a capability on or through
153 * @user_ns: The user namespace of the capability to use
154 * @cap: The capability to use
155 *
156 * Test to see if the opener of the socket had when the socket was
157 * created and the current process has the capability @cap in the user
158 * namespace @user_ns.
159 */
160bool sk_ns_capable(const struct sock *sk,
161 struct user_namespace *user_ns, int cap)
162{
163 return file_ns_capable(sk->sk_socket->file, user_ns, cap) &&
164 ns_capable(user_ns, cap);
165}
166EXPORT_SYMBOL(sk_ns_capable);
167
168/**
169 * sk_capable - Socket global capability test
170 * @sk: Socket to use a capability on or through
e793c0f7 171 * @cap: The global capability to use
a3b299da
EB		 172 *
173 * Test to see if the opener of the socket had when the socket was
174 * created and the current process has the capability @cap in all user
175 * namespaces.
176 */
177bool sk_capable(const struct sock *sk, int cap)
178{
179 return sk_ns_capable(sk, &init_user_ns, cap);
180}
181EXPORT_SYMBOL(sk_capable);
182
183/**
184 * sk_net_capable - Network namespace socket capability test
185 * @sk: Socket to use a capability on or through
186 * @cap: The capability to use
187 *
e793c0f7 188 * Test to see if the opener of the socket had when the socket was created
a3b299da
EB		 189 * and the current process has the capability @cap over the network namespace
190 * the socket is a member of.
191 */
192bool sk_net_capable(const struct sock *sk, int cap)
193{
194 return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195}
196EXPORT_SYMBOL(sk_net_capable);
197
da21f24d
IM		 198/*
199 * Each address family might have different locking rules, so we have
200 * one slock key per address family:
201 */
a5b5bb9a
IM		 202static struct lock_class_key af_family_keys[AF_MAX];
203static struct lock_class_key af_family_slock_keys[AF_MAX];
204
a5b5bb9a
IM		 205/*
206 * Make lock validator output more readable. (we pre-construct these
207 * strings build-time, so that runtime initialization of socket
208 * locks is fast):
209 */
36cbd3dc 210static const char *const af_family_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 211 "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX" , "sk_lock-AF_INET" ,
212 "sk_lock-AF_AX25" , "sk_lock-AF_IPX" , "sk_lock-AF_APPLETALK",
213 "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE" , "sk_lock-AF_ATMPVC" ,
214 "sk_lock-AF_X25" , "sk_lock-AF_INET6" , "sk_lock-AF_ROSE" ,
215 "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI" , "sk_lock-AF_SECURITY" ,
216 "sk_lock-AF_KEY" , "sk_lock-AF_NETLINK" , "sk_lock-AF_PACKET" ,
217 "sk_lock-AF_ASH" , "sk_lock-AF_ECONET" , "sk_lock-AF_ATMSVC" ,
cbd151bf 218 "sk_lock-AF_RDS" , "sk_lock-AF_SNA" , "sk_lock-AF_IRDA" ,
a5b5bb9a 219 "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE" , "sk_lock-AF_LLC" ,
cd05acfe 220 "sk_lock-27" , "sk_lock-28" , "sk_lock-AF_CAN" ,
17926a79 221 "sk_lock-AF_TIPC" , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV" ,
bce7b154 222 "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN" , "sk_lock-AF_PHONET" ,
6f107b58 223 "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG" ,
0a1a37b6
DC		 224 "sk_lock-AF_NFC" , "sk_lock-AF_VSOCK" , "sk_lock-AF_KCM" ,
225 "sk_lock-AF_MAX"
a5b5bb9a 226};
36cbd3dc 227static const char *const af_family_slock_key_strings[AF_MAX+1] = {
a5b5bb9a
IM		 228 "slock-AF_UNSPEC", "slock-AF_UNIX" , "slock-AF_INET" ,
229 "slock-AF_AX25" , "slock-AF_IPX" , "slock-AF_APPLETALK",
230 "slock-AF_NETROM", "slock-AF_BRIDGE" , "slock-AF_ATMPVC" ,
231 "slock-AF_X25" , "slock-AF_INET6" , "slock-AF_ROSE" ,
232 "slock-AF_DECnet", "slock-AF_NETBEUI" , "slock-AF_SECURITY" ,
233 "slock-AF_KEY" , "slock-AF_NETLINK" , "slock-AF_PACKET" ,
234 "slock-AF_ASH" , "slock-AF_ECONET" , "slock-AF_ATMSVC" ,
cbd151bf 235 "slock-AF_RDS" , "slock-AF_SNA" , "slock-AF_IRDA" ,
a5b5bb9a 236 "slock-AF_PPPOX" , "slock-AF_WANPIPE" , "slock-AF_LLC" ,
cd05acfe 237 "slock-27" , "slock-28" , "slock-AF_CAN" ,
17926a79 238 "slock-AF_TIPC" , "slock-AF_BLUETOOTH", "slock-AF_IUCV" ,
bce7b154 239 "slock-AF_RXRPC" , "slock-AF_ISDN" , "slock-AF_PHONET" ,
6f107b58 240 "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG" ,
0a1a37b6
DC		 241 "slock-AF_NFC" , "slock-AF_VSOCK" ,"slock-AF_KCM" ,
242 "slock-AF_MAX"
a5b5bb9a 243};
36cbd3dc 244static const char *const af_family_clock_key_strings[AF_MAX+1] = {
443aef0e
PZ		 245 "clock-AF_UNSPEC", "clock-AF_UNIX" , "clock-AF_INET" ,
246 "clock-AF_AX25" , "clock-AF_IPX" , "clock-AF_APPLETALK",
247 "clock-AF_NETROM", "clock-AF_BRIDGE" , "clock-AF_ATMPVC" ,
248 "clock-AF_X25" , "clock-AF_INET6" , "clock-AF_ROSE" ,
249 "clock-AF_DECnet", "clock-AF_NETBEUI" , "clock-AF_SECURITY" ,
250 "clock-AF_KEY" , "clock-AF_NETLINK" , "clock-AF_PACKET" ,
251 "clock-AF_ASH" , "clock-AF_ECONET" , "clock-AF_ATMSVC" ,
cbd151bf 252 "clock-AF_RDS" , "clock-AF_SNA" , "clock-AF_IRDA" ,
443aef0e 253 "clock-AF_PPPOX" , "clock-AF_WANPIPE" , "clock-AF_LLC" ,
b4942af6 254 "clock-27" , "clock-28" , "clock-AF_CAN" ,
e51f802b 255 "clock-AF_TIPC" , "clock-AF_BLUETOOTH", "clock-AF_IUCV" ,
bce7b154 256 "clock-AF_RXRPC" , "clock-AF_ISDN" , "clock-AF_PHONET" ,
6f107b58 257 "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG" ,
0a1a37b6
DC		 258 "clock-AF_NFC" , "clock-AF_VSOCK" , "clock-AF_KCM" ,
259 "clock-AF_MAX"
443aef0e 260};
da21f24d
IM		 261
262/*
263 * sk_callback_lock locking rules are per-address-family,
264 * so split the lock classes by using a per-AF key:
265 */
266static struct lock_class_key af_callback_keys[AF_MAX];
267
1da177e4
LT		 268/* Take into consideration the size of the struct sk_buff overhead in the
269 * determination of these values, since that is non-constant across
270 * platforms. This makes socket queueing behavior and performance
271 * not depend upon such differences.
272 */
273#define _SK_MEM_PACKETS 256
87fb4b7b 274#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
1da177e4
LT		 275#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
276#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
277
278/* Run time adjustable parameters. */
ab32ea5d 279__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
6d8ebc8a 280EXPORT_SYMBOL(sysctl_wmem_max);
ab32ea5d 281__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
6d8ebc8a 282EXPORT_SYMBOL(sysctl_rmem_max);
ab32ea5d
BH		 283__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
284__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
1da177e4 285
25985edc 286/* Maximal space eaten by iovec or ancillary data plus some space */
ab32ea5d 287int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
2a91525c 288EXPORT_SYMBOL(sysctl_optmem_max);
1da177e4 289
b245be1f
WB		 290int sysctl_tstamp_allow_data __read_mostly = 1;
291
c93bdd0e
MG		 292struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
293EXPORT_SYMBOL_GPL(memalloc_socks);
294
7cb02404
MG		 295/**
296 * sk_set_memalloc - sets %SOCK_MEMALLOC
297 * @sk: socket to set it on
298 *
299 * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
300 * It's the responsibility of the admin to adjust min_free_kbytes
301 * to meet the requirements
302 */
303void sk_set_memalloc(struct sock *sk)
304{
305 sock_set_flag(sk, SOCK_MEMALLOC);
306 sk->sk_allocation |= __GFP_MEMALLOC;
c93bdd0e 307 static_key_slow_inc(&memalloc_socks);
7cb02404
MG		 308}
309EXPORT_SYMBOL_GPL(sk_set_memalloc);
310
311void sk_clear_memalloc(struct sock *sk)
312{
313 sock_reset_flag(sk, SOCK_MEMALLOC);
314 sk->sk_allocation &= ~__GFP_MEMALLOC;
c93bdd0e 315 static_key_slow_dec(&memalloc_socks);
c76562b6
MG		 316
317 /*
318 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
5d753610
MG		 319 * progress of swapping. SOCK_MEMALLOC may be cleared while
320 * it has rmem allocations due to the last swapfile being deactivated
321 * but there is a risk that the socket is unusable due to exceeding
322 * the rmem limits. Reclaim the reserves and obey rmem limits again.
c76562b6 323 */
5d753610 324 sk_mem_reclaim(sk);
7cb02404
MG		 325}
326EXPORT_SYMBOL_GPL(sk_clear_memalloc);
327
b4b9e355
MG		 328int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
329{
330 int ret;
331 unsigned long pflags = current->flags;
332
333 /* these should have been dropped before queueing */
334 BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
335
336 current->flags |= PF_MEMALLOC;
337 ret = sk->sk_backlog_rcv(sk, skb);
338 tsk_restore_flags(current, pflags, PF_MEMALLOC);
339
340 return ret;
341}
342EXPORT_SYMBOL(__sk_backlog_rcv);
343
1da177e4
LT		 344static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
345{
346 struct timeval tv;
347
348 if (optlen < sizeof(tv))
349 return -EINVAL;
350 if (copy_from_user(&tv, optval, sizeof(tv)))
351 return -EFAULT;
ba78073e
VA		 352 if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
353 return -EDOM;
1da177e4 354
ba78073e 355 if (tv.tv_sec < 0) {
6f11df83
AM		 356 static int warned __read_mostly;
357
ba78073e 358 *timeo_p = 0;
50aab54f 359 if (warned < 10 && net_ratelimit()) {
ba78073e 360 warned++;
e005d193
JP		 361 pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
362 __func__, current->comm, task_pid_nr(current));
50aab54f 363 }
ba78073e
VA		 364 return 0;
365 }
1da177e4
LT		 366 *timeo_p = MAX_SCHEDULE_TIMEOUT;
367 if (tv.tv_sec == 0 && tv.tv_usec == 0)
368 return 0;
369 if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
370 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
371 return 0;
372}
373
374static void sock_warn_obsolete_bsdism(const char *name)
375{
376 static int warned;
377 static char warncomm[TASK_COMM_LEN];
4ec93edb
YH		 378 if (strcmp(warncomm, current->comm) && warned < 5) {
379 strcpy(warncomm, current->comm);
e005d193
JP		 380 pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
381 warncomm, name);
1da177e4
LT		 382 warned++;
383 }
384}
385
080a270f
HFS		 386static bool sock_needs_netstamp(const struct sock *sk)
387{
388 switch (sk->sk_family) {
389 case AF_UNSPEC:
390 case AF_UNIX:
391 return false;
392 default:
393 return true;
394 }
395}
396
08e29af3 397static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
4ec93edb 398{
08e29af3
ED		 399 if (sk->sk_flags & flags) {
400 sk->sk_flags &= ~flags;
080a270f
HFS		 401 if (sock_needs_netstamp(sk) &&
402 !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
20d49473 403 net_disable_timestamp();
1da177e4
LT		 404 }
405}
406
407
e6afc8ac 408int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
f0088a50 409{
3b885787
NH		 410 unsigned long flags;
411 struct sk_buff_head *list = &sk->sk_receive_queue;
f0088a50 412
0fd7bac6 413 if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
766e9037 414 atomic_inc(&sk->sk_drops);
3847ce32 415 trace_sock_rcvqueue_full(sk, skb);
766e9037 416 return -ENOMEM;
f0088a50
DV		 417 }
418
c76562b6 419 if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
766e9037
ED		 420 atomic_inc(&sk->sk_drops);
421 return -ENOBUFS;
3ab224be
HA		 422 }
423
f0088a50
DV		 424 skb->dev = NULL;
425 skb_set_owner_r(skb, sk);
49ad9599 426
7fee226a
ED		 427 /* we escape from rcu protected region, make sure we dont leak
428 * a norefcounted dst
429 */
430 skb_dst_force(skb);
431
3b885787 432 spin_lock_irqsave(&list->lock, flags);
3bc3b96f 433 sock_skb_set_dropcount(sk, skb);
3b885787
NH		 434 __skb_queue_tail(list, skb);
435 spin_unlock_irqrestore(&list->lock, flags);
f0088a50
DV		 436
437 if (!sock_flag(sk, SOCK_DEAD))
676d2369 438 sk->sk_data_ready(sk);
766e9037 439 return 0;
f0088a50 440}
e6afc8ac 441EXPORT_SYMBOL(__sock_queue_rcv_skb);
442
443int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
444{
445 int err;
446
447 err = sk_filter(sk, skb);
448 if (err)
449 return err;
450
451 return __sock_queue_rcv_skb(sk, skb);
452}
f0088a50
DV		 453EXPORT_SYMBOL(sock_queue_rcv_skb);
454
4f0c40d9
WB		 455int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
456 const int nested, unsigned int trim_cap)
f0088a50
DV		 457{
458 int rc = NET_RX_SUCCESS;
459
4f0c40d9 460 if (sk_filter_trim_cap(sk, skb, trim_cap))
f0088a50
DV		 461 goto discard_and_relse;
462
463 skb->dev = NULL;
464
274f482d 465 if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
c377411f
ED		 466 atomic_inc(&sk->sk_drops);
467 goto discard_and_relse;
468 }
58a5a7b9
ACM		 469 if (nested)
470 bh_lock_sock_nested(sk);
471 else
472 bh_lock_sock(sk);
a5b5bb9a
IM		 473 if (!sock_owned_by_user(sk)) {
474 /*
475 * trylock + unlock semantics:
476 */
477 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
478
c57943a1 479 rc = sk_backlog_rcv(sk, skb);
a5b5bb9a
IM		 480
481 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
f545a38f 482 } else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
8eae939f
ZY		 483 bh_unlock_sock(sk);
484 atomic_inc(&sk->sk_drops);
485 goto discard_and_relse;
486 }
487
f0088a50
DV		 488 bh_unlock_sock(sk);
489out:
490 sock_put(sk);
491 return rc;
492discard_and_relse:
493 kfree_skb(skb);
494 goto out;
495}
4f0c40d9 496EXPORT_SYMBOL(__sk_receive_skb);
f0088a50
DV		 497
498struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
499{
b6c6712a 500 struct dst_entry *dst = __sk_dst_get(sk);
f0088a50
DV		 501
502 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
e022f0b4 503 sk_tx_queue_clear(sk);
a9b3cd7f 504 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
f0088a50
DV		 505 dst_release(dst);
506 return NULL;
507 }
508
509 return dst;
510}
511EXPORT_SYMBOL(__sk_dst_check);
512
513struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
514{
515 struct dst_entry *dst = sk_dst_get(sk);
516
517 if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
518 sk_dst_reset(sk);
519 dst_release(dst);
520 return NULL;
521 }
522
523 return dst;
524}
525EXPORT_SYMBOL(sk_dst_check);
526
c91f6df2
BH		 527static int sock_setbindtodevice(struct sock *sk, char __user *optval,
528 int optlen)
4878809f
DM		 529{
530 int ret = -ENOPROTOOPT;
531#ifdef CONFIG_NETDEVICES
3b1e0a65 532 struct net *net = sock_net(sk);
4878809f
DM		 533 char devname[IFNAMSIZ];
534 int index;
535
536 /* Sorry... */
537 ret = -EPERM;
5e1fccc0 538 if (!ns_capable(net->user_ns, CAP_NET_RAW))
4878809f
DM		 539 goto out;
540
541 ret = -EINVAL;
542 if (optlen < 0)
543 goto out;
544
545 /* Bind this socket to a particular device like "eth0",
546 * as specified in the passed interface name. If the
547 * name is "" or the option length is zero the socket
548 * is not bound.
549 */
550 if (optlen > IFNAMSIZ - 1)
551 optlen = IFNAMSIZ - 1;
552 memset(devname, 0, sizeof(devname));
553
554 ret = -EFAULT;
555 if (copy_from_user(devname, optval, optlen))
556 goto out;
557
000ba2e4
DM		 558 index = 0;
559 if (devname[0] != '\0') {
bf8e56bf 560 struct net_device *dev;
4878809f 561
bf8e56bf
ED		 562 rcu_read_lock();
563 dev = dev_get_by_name_rcu(net, devname);
564 if (dev)
565 index = dev->ifindex;
566 rcu_read_unlock();
4878809f
DM		 567 ret = -ENODEV;
568 if (!dev)
569 goto out;
4878809f
DM		 570 }
571
572 lock_sock(sk);
573 sk->sk_bound_dev_if = index;
574 sk_dst_reset(sk);
575 release_sock(sk);
576
577 ret = 0;
578
579out:
580#endif
581
582 return ret;
583}
584
c91f6df2
BH		 585static int sock_getbindtodevice(struct sock *sk, char __user *optval,
586 int __user *optlen, int len)
587{
588 int ret = -ENOPROTOOPT;
589#ifdef CONFIG_NETDEVICES
590 struct net *net = sock_net(sk);
c91f6df2 591 char devname[IFNAMSIZ];
c91f6df2
BH		 592
593 if (sk->sk_bound_dev_if == 0) {
594 len = 0;
595 goto zero;
596 }
597
598 ret = -EINVAL;
599 if (len < IFNAMSIZ)
600 goto out;
601
5dbe7c17
NS		 602 ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
603 if (ret)
c91f6df2 604 goto out;
c91f6df2
BH		 605
606 len = strlen(devname) + 1;
607
608 ret = -EFAULT;
609 if (copy_to_user(optval, devname, len))
610 goto out;
611
612zero:
613 ret = -EFAULT;
614 if (put_user(len, optlen))
615 goto out;
616
617 ret = 0;
618
619out:
620#endif
621
622 return ret;
623}
624
c0ef877b
PE		 625static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
626{
627 if (valbool)
628 sock_set_flag(sk, bit);
629 else
630 sock_reset_flag(sk, bit);
631}
632
f60e5990 633bool sk_mc_loop(struct sock *sk)
634{
635 if (dev_recursion_level())
636 return false;
637 if (!sk)
638 return true;
639 switch (sk->sk_family) {
640 case AF_INET:
641 return inet_sk(sk)->mc_loop;
642#if IS_ENABLED(CONFIG_IPV6)
643 case AF_INET6:
644 return inet6_sk(sk)->mc_loop;
645#endif
646 }
647 WARN_ON(1);
648 return true;
649}
650EXPORT_SYMBOL(sk_mc_loop);
651
1da177e4
LT		 652/*
653 * This is meant for all protocols to use and covers goings on
654 * at the socket level. Everything here is generic.
655 */
656
657int sock_setsockopt(struct socket *sock, int level, int optname,
b7058842 658 char __user *optval, unsigned int optlen)
1da177e4 659{
2a91525c 660 struct sock *sk = sock->sk;
1da177e4
LT		 661 int val;
662 int valbool;
663 struct linger ling;
664 int ret = 0;
4ec93edb 665
1da177e4
LT		 666 /*
667 * Options without arguments
668 */
669
4878809f 670 if (optname == SO_BINDTODEVICE)
c91f6df2 671 return sock_setbindtodevice(sk, optval, optlen);
4878809f 672
e71a4783
SH		 673 if (optlen < sizeof(int))
674 return -EINVAL;
4ec93edb 675
1da177e4
LT		 676 if (get_user(val, (int __user *)optval))
677 return -EFAULT;
4ec93edb 678
2a91525c 679 valbool = val ? 1 : 0;
1da177e4
LT		 680
681 lock_sock(sk);
682
2a91525c 683 switch (optname) {
e71a4783 684 case SO_DEBUG:
2a91525c 685 if (val && !capable(CAP_NET_ADMIN))
e71a4783 686 ret = -EACCES;
2a91525c 687 else
c0ef877b 688 sock_valbool_flag(sk, SOCK_DBG, valbool);
e71a4783
SH		 689 break;
690 case SO_REUSEADDR:
4a17fd52 691 sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
e71a4783 692 break;
055dc21a
TH		 693 case SO_REUSEPORT:
694 sk->sk_reuseport = valbool;
695 break;
e71a4783 696 case SO_TYPE:
49c794e9 697 case SO_PROTOCOL:
0d6038ee 698 case SO_DOMAIN:
e71a4783
SH		 699 case SO_ERROR:
700 ret = -ENOPROTOOPT;
701 break;
702 case SO_DONTROUTE:
c0ef877b 703 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
e71a4783
SH		 704 break;
705 case SO_BROADCAST:
706 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
707 break;
708 case SO_SNDBUF:
709 /* Don't error on this BSD doesn't and if you think
82981930
ED		 710 * about it this is right. Otherwise apps have to
711 * play 'guess the biggest size' games. RCVBUF/SNDBUF
712 * are treated in BSD as hints
713 */
714 val = min_t(u32, val, sysctl_wmem_max);
b0573dea 715set_sndbuf:
e71a4783 716 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
82981930
ED		 717 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
718 /* Wake up sending tasks if we upped the value. */
e71a4783
SH		 719 sk->sk_write_space(sk);
720 break;
1da177e4 721
e71a4783
SH		 722 case SO_SNDBUFFORCE:
723 if (!capable(CAP_NET_ADMIN)) {
724 ret = -EPERM;
725 break;
726 }
727 goto set_sndbuf;
b0573dea 728
e71a4783
SH		 729 case SO_RCVBUF:
730 /* Don't error on this BSD doesn't and if you think
82981930
ED		 731 * about it this is right. Otherwise apps have to
732 * play 'guess the biggest size' games. RCVBUF/SNDBUF
733 * are treated in BSD as hints
734 */
735 val = min_t(u32, val, sysctl_rmem_max);
b0573dea 736set_rcvbuf:
e71a4783
SH		 737 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
738 /*
739 * We double it on the way in to account for
740 * "struct sk_buff" etc. overhead. Applications
741 * assume that the SO_RCVBUF setting they make will
742 * allow that much actual data to be received on that
743 * socket.
744 *
745 * Applications are unaware that "struct sk_buff" and
746 * other overheads allocate from the receive buffer
747 * during socket buffer allocation.
748 *
749 * And after considering the possible alternatives,
750 * returning the value we actually used in getsockopt
751 * is the most desirable behavior.
752 */
82981930 753 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
e71a4783
SH		 754 break;
755
756 case SO_RCVBUFFORCE:
757 if (!capable(CAP_NET_ADMIN)) {
758 ret = -EPERM;
1da177e4 759 break;
e71a4783
SH		 760 }
761 goto set_rcvbuf;
1da177e4 762
e71a4783 763 case SO_KEEPALIVE:
1da177e4 764#ifdef CONFIG_INET
3e10986d
ED		 765 if (sk->sk_protocol == IPPROTO_TCP &&
766 sk->sk_type == SOCK_STREAM)
e71a4783 767 tcp_set_keepalive(sk, valbool);
1da177e4 768#endif
e71a4783
SH		 769 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
770 break;
771
772 case SO_OOBINLINE:
773 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
774 break;
775
776 case SO_NO_CHECK:
28448b80 777 sk->sk_no_check_tx = valbool;
e71a4783
SH		 778 break;
779
780 case SO_PRIORITY:
5e1fccc0
EB		 781 if ((val >= 0 && val <= 6) ||
782 ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
e71a4783
SH		 783 sk->sk_priority = val;
784 else
785 ret = -EPERM;
786 break;
787
788 case SO_LINGER:
789 if (optlen < sizeof(ling)) {
790 ret = -EINVAL; /* 1003.1g */
1da177e4 791 break;
e71a4783 792 }
2a91525c 793 if (copy_from_user(&ling, optval, sizeof(ling))) {
e71a4783 794 ret = -EFAULT;
1da177e4 795 break;
e71a4783
SH		 796 }
797 if (!ling.l_onoff)
798 sock_reset_flag(sk, SOCK_LINGER);
799 else {
1da177e4 800#if (BITS_PER_LONG == 32)
e71a4783
SH		 801 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
802 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
1da177e4 803 else
e71a4783
SH		 804#endif
805 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
806 sock_set_flag(sk, SOCK_LINGER);
807 }
808 break;
809
810 case SO_BSDCOMPAT:
811 sock_warn_obsolete_bsdism("setsockopt");
812 break;
813
814 case SO_PASSCRED:
815 if (valbool)
816 set_bit(SOCK_PASSCRED, &sock->flags);
817 else
818 clear_bit(SOCK_PASSCRED, &sock->flags);
819 break;
820
821 case SO_TIMESTAMP:
92f37fd2 822 case SO_TIMESTAMPNS:
e71a4783 823 if (valbool) {
92f37fd2
ED		 824 if (optname == SO_TIMESTAMP)
825 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
826 else
827 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783 828 sock_set_flag(sk, SOCK_RCVTSTAMP);
20d49473 829 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
92f37fd2 830 } else {
e71a4783 831 sock_reset_flag(sk, SOCK_RCVTSTAMP);
92f37fd2
ED		 832 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
833 }
e71a4783
SH		 834 break;
835
20d49473
PO		 836 case SO_TIMESTAMPING:
837 if (val & ~SOF_TIMESTAMPING_MASK) {
f249fb78 838 ret = -EINVAL;
20d49473
PO		 839 break;
840 }
b245be1f 841
09c2d251 842 if (val & SOF_TIMESTAMPING_OPT_ID &&
4ed2d765 843 !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
ac5cc977
WC		 844 if (sk->sk_protocol == IPPROTO_TCP &&
845 sk->sk_type == SOCK_STREAM) {
6db8b963
SHY		 846 if ((1 << sk->sk_state) &
847 (TCPF_CLOSE | TCPF_LISTEN)) {
4ed2d765
WB		 848 ret = -EINVAL;
849 break;
850 }
851 sk->sk_tskey = tcp_sk(sk)->snd_una;
852 } else {
853 sk->sk_tskey = 0;
854 }
855 }
b9f40e21 856 sk->sk_tsflags = val;
20d49473
PO		 857 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
858 sock_enable_timestamp(sk,
859 SOCK_TIMESTAMPING_RX_SOFTWARE);
860 else
861 sock_disable_timestamp(sk,
08e29af3 862 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
20d49473
PO		 863 break;
864
e71a4783
SH		 865 case SO_RCVLOWAT:
866 if (val < 0)
867 val = INT_MAX;
868 sk->sk_rcvlowat = val ? : 1;
869 break;
870
871 case SO_RCVTIMEO:
872 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
873 break;
874
875 case SO_SNDTIMEO:
876 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
877 break;
1da177e4 878
e71a4783
SH		 879 case SO_ATTACH_FILTER:
880 ret = -EINVAL;
881 if (optlen == sizeof(struct sock_fprog)) {
882 struct sock_fprog fprog;
1da177e4 883
e71a4783
SH		 884 ret = -EFAULT;
885 if (copy_from_user(&fprog, optval, sizeof(fprog)))
1da177e4 886 break;
e71a4783
SH		 887
888 ret = sk_attach_filter(&fprog, sk);
889 }
890 break;
891
89aa0758
AS		 892 case SO_ATTACH_BPF:
893 ret = -EINVAL;
894 if (optlen == sizeof(u32)) {
895 u32 ufd;
896
897 ret = -EFAULT;
898 if (copy_from_user(&ufd, optval, sizeof(ufd)))
899 break;
900
901 ret = sk_attach_bpf(ufd, sk);
902 }
903 break;
904
538950a1
CG		 905 case SO_ATTACH_REUSEPORT_CBPF:
906 ret = -EINVAL;
907 if (optlen == sizeof(struct sock_fprog)) {
908 struct sock_fprog fprog;
909
910 ret = -EFAULT;
911 if (copy_from_user(&fprog, optval, sizeof(fprog)))
912 break;
913
914 ret = sk_reuseport_attach_filter(&fprog, sk);
915 }
916 break;
917
918 case SO_ATTACH_REUSEPORT_EBPF:
919 ret = -EINVAL;
920 if (optlen == sizeof(u32)) {
921 u32 ufd;
922
923 ret = -EFAULT;
924 if (copy_from_user(&ufd, optval, sizeof(ufd)))
925 break;
926
927 ret = sk_reuseport_attach_bpf(ufd, sk);
928 }
929 break;
930
e71a4783 931 case SO_DETACH_FILTER:
55b33325 932 ret = sk_detach_filter(sk);
e71a4783 933 break;
1da177e4 934
d59577b6
VB		 935 case SO_LOCK_FILTER:
936 if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
937 ret = -EPERM;
938 else
939 sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
940 break;
941
e71a4783
SH		 942 case SO_PASSSEC:
943 if (valbool)
944 set_bit(SOCK_PASSSEC, &sock->flags);
945 else
946 clear_bit(SOCK_PASSSEC, &sock->flags);
947 break;
4a19ec58 948 case SO_MARK:
5e1fccc0 949 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
4a19ec58 950 ret = -EPERM;
2a91525c 951 else
4a19ec58 952 sk->sk_mark = val;
4a19ec58 953 break;
877ce7c1 954
3b885787 955 case SO_RXQ_OVFL:
8083f0fc 956 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
3b885787 957 break;
6e3e939f
JB		 958
959 case SO_WIFI_STATUS:
960 sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
961 break;
962
ef64a54f
PE		 963 case SO_PEEK_OFF:
964 if (sock->ops->set_peek_off)
12663bfc 965 ret = sock->ops->set_peek_off(sk, val);
ef64a54f
PE		 966 else
967 ret = -EOPNOTSUPP;
968 break;
3bdc0eba
BG		 969
970 case SO_NOFCS:
971 sock_valbool_flag(sk, SOCK_NOFCS, valbool);
972 break;
973
7d4c04fc
KJ		 974 case SO_SELECT_ERR_QUEUE:
975 sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
976 break;
977
e0d1095a 978#ifdef CONFIG_NET_RX_BUSY_POLL
64b0dc51 979 case SO_BUSY_POLL:
dafcc438
ET		 980 /* allow unprivileged users to decrease the value */
981 if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
982 ret = -EPERM;
983 else {
984 if (val < 0)
985 ret = -EINVAL;
986 else
987 sk->sk_ll_usec = val;
988 }
989 break;
990#endif
62748f32
ED		 991
992 case SO_MAX_PACING_RATE:
993 sk->sk_max_pacing_rate = val;
994 sk->sk_pacing_rate = min(sk->sk_pacing_rate,
995 sk->sk_max_pacing_rate);
996 break;
997
70da268b
ED		 998 case SO_INCOMING_CPU:
999 sk->sk_incoming_cpu = val;
1000 break;
1001
a87cb3e4
TH		 1002 case SO_CNX_ADVICE:
1003 if (val == 1)
1004 dst_negative_advice(sk);
1005 break;
e71a4783
SH		 1006 default:
1007 ret = -ENOPROTOOPT;
1008 break;
4ec93edb 1009 }
1da177e4
LT		 1010 release_sock(sk);
1011 return ret;
1012}
2a91525c 1013EXPORT_SYMBOL(sock_setsockopt);
1da177e4
LT		 1014
1015
8f09898b 1016static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1017 struct ucred *ucred)
3f551f94
EB		 1018{
1019 ucred->pid = pid_vnr(pid);
1020 ucred->uid = ucred->gid = -1;
1021 if (cred) {
1022 struct user_namespace *current_ns = current_user_ns();
1023
b2e4f544
EB		 1024 ucred->uid = from_kuid_munged(current_ns, cred->euid);
1025 ucred->gid = from_kgid_munged(current_ns, cred->egid);
3f551f94
EB		 1026 }
1027}
1028
1da177e4
LT		 1029int sock_getsockopt(struct socket *sock, int level, int optname,
1030 char __user *optval, int __user *optlen)
1031{
1032 struct sock *sk = sock->sk;
4ec93edb 1033
e71a4783 1034 union {
4ec93edb
YH		 1035 int val;
1036 struct linger ling;
1da177e4
LT		 1037 struct timeval tm;
1038 } v;
4ec93edb 1039
4d0392be 1040 int lv = sizeof(int);
1da177e4 1041 int len;
4ec93edb 1042
e71a4783 1043 if (get_user(len, optlen))
4ec93edb 1044 return -EFAULT;
e71a4783 1045 if (len < 0)
1da177e4 1046 return -EINVAL;
4ec93edb 1047
50fee1de 1048 memset(&v, 0, sizeof(v));
df0bca04 1049
2a91525c 1050 switch (optname) {
e71a4783
SH		 1051 case SO_DEBUG:
1052 v.val = sock_flag(sk, SOCK_DBG);
1053 break;
1054
1055 case SO_DONTROUTE:
1056 v.val = sock_flag(sk, SOCK_LOCALROUTE);
1057 break;
1058
1059 case SO_BROADCAST:
1b23a5df 1060 v.val = sock_flag(sk, SOCK_BROADCAST);
e71a4783
SH		 1061 break;
1062
1063 case SO_SNDBUF:
1064 v.val = sk->sk_sndbuf;
1065 break;
1066
1067 case SO_RCVBUF:
1068 v.val = sk->sk_rcvbuf;
1069 break;
1070
1071 case SO_REUSEADDR:
1072 v.val = sk->sk_reuse;
1073 break;
1074
055dc21a
TH		 1075 case SO_REUSEPORT:
1076 v.val = sk->sk_reuseport;
1077 break;
1078
e71a4783 1079 case SO_KEEPALIVE:
1b23a5df 1080 v.val = sock_flag(sk, SOCK_KEEPOPEN);
e71a4783
SH		 1081 break;
1082
1083 case SO_TYPE:
1084 v.val = sk->sk_type;
1085 break;
1086
49c794e9
JE		 1087 case SO_PROTOCOL:
1088 v.val = sk->sk_protocol;
1089 break;
1090
0d6038ee
JE		 1091 case SO_DOMAIN:
1092 v.val = sk->sk_family;
1093 break;
1094
e71a4783
SH		 1095 case SO_ERROR:
1096 v.val = -sock_error(sk);
2a91525c 1097 if (v.val == 0)
e71a4783
SH		 1098 v.val = xchg(&sk->sk_err_soft, 0);
1099 break;
1100
1101 case SO_OOBINLINE:
1b23a5df 1102 v.val = sock_flag(sk, SOCK_URGINLINE);
e71a4783
SH		 1103 break;
1104
1105 case SO_NO_CHECK:
28448b80 1106 v.val = sk->sk_no_check_tx;
e71a4783
SH		 1107 break;
1108
1109 case SO_PRIORITY:
1110 v.val = sk->sk_priority;
1111 break;
1112
1113 case SO_LINGER:
1114 lv = sizeof(v.ling);
1b23a5df 1115 v.ling.l_onoff = sock_flag(sk, SOCK_LINGER);
e71a4783
SH		 1116 v.ling.l_linger = sk->sk_lingertime / HZ;
1117 break;
1118
1119 case SO_BSDCOMPAT:
1120 sock_warn_obsolete_bsdism("getsockopt");
1121 break;
1122
1123 case SO_TIMESTAMP:
92f37fd2
ED		 1124 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1125 !sock_flag(sk, SOCK_RCVTSTAMPNS);
1126 break;
1127
1128 case SO_TIMESTAMPNS:
1129 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
e71a4783
SH		 1130 break;
1131
20d49473 1132 case SO_TIMESTAMPING:
b9f40e21 1133 v.val = sk->sk_tsflags;
20d49473
PO		 1134 break;
1135
e71a4783 1136 case SO_RCVTIMEO:
2a91525c 1137 lv = sizeof(struct timeval);
e71a4783
SH		 1138 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1139 v.tm.tv_sec = 0;
1140 v.tm.tv_usec = 0;
1141 } else {
1142 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1143 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1144 }
1145 break;
1146
1147 case SO_SNDTIMEO:
2a91525c 1148 lv = sizeof(struct timeval);
e71a4783
SH		 1149 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1150 v.tm.tv_sec = 0;
1151 v.tm.tv_usec = 0;
1152 } else {
1153 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1154 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1155 }
1156 break;
1da177e4 1157
e71a4783
SH		 1158 case SO_RCVLOWAT:
1159 v.val = sk->sk_rcvlowat;
1160 break;
1da177e4 1161
e71a4783 1162 case SO_SNDLOWAT:
2a91525c 1163 v.val = 1;
e71a4783 1164 break;
1da177e4 1165
e71a4783 1166 case SO_PASSCRED:
82981930 1167 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
e71a4783 1168 break;
1da177e4 1169
e71a4783 1170 case SO_PEERCRED:
109f6e39
EB		 1171 {
1172 struct ucred peercred;
1173 if (len > sizeof(peercred))
1174 len = sizeof(peercred);
1175 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1176 if (copy_to_user(optval, &peercred, len))
e71a4783
SH		 1177 return -EFAULT;
1178 goto lenout;
109f6e39 1179 }
1da177e4 1180
e71a4783
SH		 1181 case SO_PEERNAME:
1182 {
1183 char address[128];
1184
1185 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1186 return -ENOTCONN;
1187 if (lv < len)
1188 return -EINVAL;
1189 if (copy_to_user(optval, address, len))
1190 return -EFAULT;
1191 goto lenout;
1192 }
1da177e4 1193
e71a4783
SH		 1194 /* Dubious BSD thing... Probably nobody even uses it, but
1195 * the UNIX standard wants it for whatever reason... -DaveM
1196 */
1197 case SO_ACCEPTCONN:
1198 v.val = sk->sk_state == TCP_LISTEN;
1199 break;
1da177e4 1200
e71a4783 1201 case SO_PASSSEC:
82981930 1202 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
e71a4783 1203 break;
877ce7c1 1204
e71a4783
SH		 1205 case SO_PEERSEC:
1206 return security_socket_getpeersec_stream(sock, optval, optlen, len);
1da177e4 1207
4a19ec58
LAT		 1208 case SO_MARK:
1209 v.val = sk->sk_mark;
1210 break;
1211
3b885787 1212 case SO_RXQ_OVFL:
1b23a5df 1213 v.val = sock_flag(sk, SOCK_RXQ_OVFL);
3b885787
NH		 1214 break;
1215
6e3e939f 1216 case SO_WIFI_STATUS:
1b23a5df 1217 v.val = sock_flag(sk, SOCK_WIFI_STATUS);
6e3e939f
JB		 1218 break;
1219
ef64a54f
PE		 1220 case SO_PEEK_OFF:
1221 if (!sock->ops->set_peek_off)
1222 return -EOPNOTSUPP;
1223
1224 v.val = sk->sk_peek_off;
1225 break;
bc2f7996 1226 case SO_NOFCS:
1b23a5df 1227 v.val = sock_flag(sk, SOCK_NOFCS);
bc2f7996 1228 break;
c91f6df2 1229
f7b86bfe 1230 case SO_BINDTODEVICE:
c91f6df2
BH		 1231 return sock_getbindtodevice(sk, optval, optlen, len);
1232
a8fc9277
PE		 1233 case SO_GET_FILTER:
1234 len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1235 if (len < 0)
1236 return len;
1237
1238 goto lenout;
c91f6df2 1239
d59577b6
VB		 1240 case SO_LOCK_FILTER:
1241 v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1242 break;
1243
ea02f941
MS		 1244 case SO_BPF_EXTENSIONS:
1245 v.val = bpf_tell_extensions();
1246 break;
1247
7d4c04fc
KJ		 1248 case SO_SELECT_ERR_QUEUE:
1249 v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1250 break;
1251
e0d1095a 1252#ifdef CONFIG_NET_RX_BUSY_POLL
64b0dc51 1253 case SO_BUSY_POLL:
dafcc438
ET		 1254 v.val = sk->sk_ll_usec;
1255 break;
1256#endif
1257
62748f32
ED		 1258 case SO_MAX_PACING_RATE:
1259 v.val = sk->sk_max_pacing_rate;
1260 break;
1261
2c8c56e1
ED		 1262 case SO_INCOMING_CPU:
1263 v.val = sk->sk_incoming_cpu;
1264 break;
1265
e71a4783 1266 default:
443b5991
YH		 1267 /* We implement the SO_SNDLOWAT etc to not be settable
1268 * (1003.1g 7).
1269 */
e71a4783 1270 return -ENOPROTOOPT;
1da177e4 1271 }
e71a4783 1272
1da177e4
LT		 1273 if (len > lv)
1274 len = lv;
1275 if (copy_to_user(optval, &v, len))
1276 return -EFAULT;
1277lenout:
4ec93edb
YH		 1278 if (put_user(len, optlen))
1279 return -EFAULT;
1280 return 0;
1da177e4
LT		 1281}
1282
a5b5bb9a
IM		 1283/*
1284 * Initialize an sk_lock.
1285 *
1286 * (We also register the sk_lock with the lock validator.)
1287 */
b6f99a21 1288static inline void sock_lock_init(struct sock *sk)
a5b5bb9a 1289{
ed07536e
PZ		 1290 sock_lock_init_class_and_name(sk,
1291 af_family_slock_key_strings[sk->sk_family],
1292 af_family_slock_keys + sk->sk_family,
1293 af_family_key_strings[sk->sk_family],
1294 af_family_keys + sk->sk_family);
a5b5bb9a
IM		 1295}
1296
4dc6dc71
ED		 1297/*
1298 * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1299 * even temporarly, because of RCU lookups. sk_node should also be left as is.
68835aba 1300 * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
4dc6dc71 1301 */
f1a6c4da
PE		 1302static void sock_copy(struct sock *nsk, const struct sock *osk)
1303{
1304#ifdef CONFIG_SECURITY_NETWORK
1305 void *sptr = nsk->sk_security;
1306#endif
68835aba
ED		 1307 memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1308
1309 memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1310 osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1311
f1a6c4da
PE		 1312#ifdef CONFIG_SECURITY_NETWORK
1313 nsk->sk_security = sptr;
1314 security_sk_clone(osk, nsk);
1315#endif
1316}
1317
2e4afe7b
PE		 1318static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1319 int family)
c308c1b2
PE		 1320{
1321 struct sock *sk;
1322 struct kmem_cache *slab;
1323
1324 slab = prot->slab;
e912b114
ED		 1325 if (slab != NULL) {
1326 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1327 if (!sk)
1328 return sk;
ba2489b0
ED		 1329 if (priority & __GFP_ZERO)
1330 sk_prot_clear_nulls(sk, prot->obj_size);
fcbdf09d 1331 } else
c308c1b2
PE		 1332 sk = kmalloc(prot->obj_size, priority);
1333
2e4afe7b 1334 if (sk != NULL) {
a98b65a3
VN		 1335 kmemcheck_annotate_bitfield(sk, flags);
1336
2e4afe7b
PE		 1337 if (security_sk_alloc(sk, family, priority))
1338 goto out_free;
1339
1340 if (!try_module_get(prot->owner))
1341 goto out_free_sec;
e022f0b4 1342 sk_tx_queue_clear(sk);
2e4afe7b
PE		 1343 }
1344
c308c1b2 1345 return sk;
2e4afe7b
PE		 1346
1347out_free_sec:
1348 security_sk_free(sk);
1349out_free:
1350 if (slab != NULL)
1351 kmem_cache_free(slab, sk);
1352 else
1353 kfree(sk);
1354 return NULL;
c308c1b2
PE		 1355}
1356
1357static void sk_prot_free(struct proto *prot, struct sock *sk)
1358{
1359 struct kmem_cache *slab;
2e4afe7b 1360 struct module *owner;
c308c1b2 1361
2e4afe7b 1362 owner = prot->owner;
c308c1b2 1363 slab = prot->slab;
2e4afe7b 1364
bd1060a1 1365 cgroup_sk_free(&sk->sk_cgrp_data);
2d758073 1366 mem_cgroup_sk_free(sk);
2e4afe7b 1367 security_sk_free(sk);
c308c1b2
PE		 1368 if (slab != NULL)
1369 kmem_cache_free(slab, sk);
1370 else
1371 kfree(sk);
2e4afe7b 1372 module_put(owner);
c308c1b2
PE		 1373}
1374
1da177e4
LT		 1375/**
1376 * sk_alloc - All socket objects are allocated here
c4ea43c5 1377 * @net: the applicable net namespace
4dc3b16b
PP		 1378 * @family: protocol family
1379 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1380 * @prot: struct proto associated with this new sock instance
11aa9c28 1381 * @kern: is this to be a kernel socket?
1da177e4 1382 */
1b8d7ae4 1383struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
11aa9c28 1384 struct proto *prot, int kern)
1da177e4 1385{
c308c1b2 1386 struct sock *sk;
1da177e4 1387
154adbc8 1388 sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1da177e4 1389 if (sk) {
154adbc8
PE		 1390 sk->sk_family = family;
1391 /*
1392 * See comment in struct sock definition to understand
1393 * why we need sk_prot_creator -acme
1394 */
1395 sk->sk_prot = sk->sk_prot_creator = prot;
1396 sock_lock_init(sk);
26abe143
EB		 1397 sk->sk_net_refcnt = kern ? 0 : 1;
1398 if (likely(sk->sk_net_refcnt))
1399 get_net(net);
1400 sock_net_set(sk, net);
d66ee058 1401 atomic_set(&sk->sk_wmem_alloc, 1);
f8451725 1402
2d758073 1403 mem_cgroup_sk_alloc(sk);
d979a39d 1404 cgroup_sk_alloc(&sk->sk_cgrp_data);
2a56a1fe
TH		 1405 sock_update_classid(&sk->sk_cgrp_data);
1406 sock_update_netprioidx(&sk->sk_cgrp_data);
1da177e4 1407 }
a79af59e 1408
2e4afe7b 1409 return sk;
1da177e4 1410}
2a91525c 1411EXPORT_SYMBOL(sk_alloc);
1da177e4 1412
a4298e45
ED		 1413/* Sockets having SOCK_RCU_FREE will call this function after one RCU
1414 * grace period. This is the case for UDP sockets and TCP listeners.
1415 */
1416static void __sk_destruct(struct rcu_head *head)
1da177e4 1417{
a4298e45 1418 struct sock *sk = container_of(head, struct sock, sk_rcu);
1da177e4 1419 struct sk_filter *filter;
1da177e4
LT		 1420
1421 if (sk->sk_destruct)
1422 sk->sk_destruct(sk);
1423
a898def2
PM		 1424 filter = rcu_dereference_check(sk->sk_filter,
1425 atomic_read(&sk->sk_wmem_alloc) == 0);
1da177e4 1426 if (filter) {
309dd5fc 1427 sk_filter_uncharge(sk, filter);
a9b3cd7f 1428 RCU_INIT_POINTER(sk->sk_filter, NULL);
1da177e4 1429 }
538950a1
CG		 1430 if (rcu_access_pointer(sk->sk_reuseport_cb))
1431 reuseport_detach_sock(sk);
1da177e4 1432
08e29af3 1433 sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1da177e4
LT		 1434
1435 if (atomic_read(&sk->sk_omem_alloc))
e005d193
JP		 1436 pr_debug("%s: optmem leakage (%d bytes) detected\n",
1437 __func__, atomic_read(&sk->sk_omem_alloc));
1da177e4 1438
109f6e39
EB		 1439 if (sk->sk_peer_cred)
1440 put_cred(sk->sk_peer_cred);
1441 put_pid(sk->sk_peer_pid);
26abe143
EB		 1442 if (likely(sk->sk_net_refcnt))
1443 put_net(sock_net(sk));
c308c1b2 1444 sk_prot_free(sk->sk_prot_creator, sk);
1da177e4 1445}
2b85a34e 1446
a4298e45
ED		 1447void sk_destruct(struct sock *sk)
1448{
1449 if (sock_flag(sk, SOCK_RCU_FREE))
1450 call_rcu(&sk->sk_rcu, __sk_destruct);
1451 else
1452 __sk_destruct(&sk->sk_rcu);
1453}
1454
eb4cb008
CG		 1455static void __sk_free(struct sock *sk)
1456{
b922622e 1457 if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
eb4cb008
CG		 1458 sock_diag_broadcast_destroy(sk);
1459 else
1460 sk_destruct(sk);
1461}
1462
2b85a34e
ED		 1463void sk_free(struct sock *sk)
1464{
1465 /*
25985edc 1466 * We subtract one from sk_wmem_alloc and can know if
2b85a34e
ED		 1467 * some packets are still in some tx queue.
1468 * If not null, sock_wfree() will call __sk_free(sk) later
1469 */
1470 if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1471 __sk_free(sk);
1472}
2a91525c 1473EXPORT_SYMBOL(sk_free);
1da177e4 1474
e56c57d0
ED		 1475/**
1476 * sk_clone_lock - clone a socket, and lock its clone
1477 * @sk: the socket to clone
1478 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1479 *
1480 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1481 */
1482struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
87d11ceb 1483{
8fd1d178 1484 struct sock *newsk;
278571ba 1485 bool is_charged = true;
87d11ceb 1486
8fd1d178 1487 newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
87d11ceb
ACM		 1488 if (newsk != NULL) {
1489 struct sk_filter *filter;
1490
892c141e 1491 sock_copy(newsk, sk);
87d11ceb
ACM		 1492
1493 /* SANITY */
8a681736
SV		 1494 if (likely(newsk->sk_net_refcnt))
1495 get_net(sock_net(newsk));
87d11ceb
ACM		 1496 sk_node_init(&newsk->sk_node);
1497 sock_lock_init(newsk);
1498 bh_lock_sock(newsk);
fa438ccf 1499 newsk->sk_backlog.head = newsk->sk_backlog.tail = NULL;
8eae939f 1500 newsk->sk_backlog.len = 0;
87d11ceb
ACM		 1501
1502 atomic_set(&newsk->sk_rmem_alloc, 0);
2b85a34e
ED		 1503 /*
1504 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1505 */
1506 atomic_set(&newsk->sk_wmem_alloc, 1);
87d11ceb
ACM		 1507 atomic_set(&newsk->sk_omem_alloc, 0);
1508 skb_queue_head_init(&newsk->sk_receive_queue);
1509 skb_queue_head_init(&newsk->sk_write_queue);
1510
87d11ceb 1511 rwlock_init(&newsk->sk_callback_lock);
443aef0e
PZ		 1512 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1513 af_callback_keys + newsk->sk_family,
1514 af_family_clock_key_strings[newsk->sk_family]);
87d11ceb
ACM		 1515
1516 newsk->sk_dst_cache = NULL;
1517 newsk->sk_wmem_queued = 0;
1518 newsk->sk_forward_alloc = 0;
9caad864 1519 atomic_set(&newsk->sk_drops, 0);
87d11ceb 1520 newsk->sk_send_head = NULL;
87d11ceb
ACM		 1521 newsk->sk_userlocks = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1522
1523 sock_reset_flag(newsk, SOCK_DONE);
1524 skb_queue_head_init(&newsk->sk_error_queue);
1525
0d7da9dd 1526 filter = rcu_dereference_protected(newsk->sk_filter, 1);
87d11ceb 1527 if (filter != NULL)
278571ba
AS		 1528 /* though it's an empty new sock, the charging may fail
1529 * if sysctl_optmem_max was changed between creation of
1530 * original socket and cloning
1531 */
1532 is_charged = sk_filter_charge(newsk, filter);
87d11ceb 1533
d188ba86 1534 if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
87d11ceb
ACM		 1535 /* It is still raw copy of parent, so invalidate
1536 * destructor and make plain sk_free() */
1537 newsk->sk_destruct = NULL;
b0691c8e 1538 bh_unlock_sock(newsk);
87d11ceb
ACM		 1539 sk_free(newsk);
1540 newsk = NULL;
1541 goto out;
1542 }
fa463497 1543 RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
87d11ceb
ACM		 1544
1545 newsk->sk_err = 0;
e551c32d 1546 newsk->sk_err_soft = 0;
87d11ceb 1547 newsk->sk_priority = 0;
2c8c56e1 1548 newsk->sk_incoming_cpu = raw_smp_processor_id();
33cf7c90 1549 atomic64_set(&newsk->sk_cookie, 0);
d979a39d 1550
2d758073 1551 mem_cgroup_sk_alloc(newsk);
d979a39d
JW		 1552 cgroup_sk_alloc(&newsk->sk_cgrp_data);
1553
4dc6dc71
ED		 1554 /*
1555 * Before updating sk_refcnt, we must commit prior changes to memory
1556 * (Documentation/RCU/rculist_nulls.txt for details)
1557 */
1558 smp_wmb();
87d11ceb
ACM		 1559 atomic_set(&newsk->sk_refcnt, 2);
1560
1561 /*
1562 * Increment the counter in the same struct proto as the master
1563 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1564 * is the same as sk->sk_prot->socks, as this field was copied
1565 * with memcpy).
1566 *
1567 * This _changes_ the previous behaviour, where
1568 * tcp_create_openreq_child always was incrementing the
1569 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1570 * to be taken into account in all callers. -acme
1571 */
1572 sk_refcnt_debug_inc(newsk);
972692e0 1573 sk_set_socket(newsk, NULL);
43815482 1574 newsk->sk_wq = NULL;
87d11ceb
ACM		 1575
1576 if (newsk->sk_prot->sockets_allocated)
180d8cd9 1577 sk_sockets_allocated_inc(newsk);
704da560 1578
080a270f
HFS		 1579 if (sock_needs_netstamp(sk) &&
1580 newsk->sk_flags & SK_FLAGS_TIMESTAMP)
704da560 1581 net_enable_timestamp();
87d11ceb
ACM		 1582 }
1583out:
1584 return newsk;
1585}
e56c57d0 1586EXPORT_SYMBOL_GPL(sk_clone_lock);
87d11ceb 1587
9958089a
AK		 1588void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1589{
d6a4e26a
ED		 1590 u32 max_segs = 1;
1591
6bd4f355 1592 sk_dst_set(sk, dst);
9958089a
AK		 1593 sk->sk_route_caps = dst->dev->features;
1594 if (sk->sk_route_caps & NETIF_F_GSO)
4fcd6b99 1595 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
a465419b 1596 sk->sk_route_caps &= ~sk->sk_route_nocaps;
9958089a 1597 if (sk_can_gso(sk)) {
82cc1a7a 1598 if (dst->header_len) {
9958089a 1599 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
82cc1a7a 1600 } else {
9958089a 1601 sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
82cc1a7a 1602 sk->sk_gso_max_size = dst->dev->gso_max_size;
d6a4e26a 1603 max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
82cc1a7a 1604 }
9958089a 1605 }
d6a4e26a 1606 sk->sk_gso_max_segs = max_segs;
9958089a
AK		 1607}
1608EXPORT_SYMBOL_GPL(sk_setup_caps);
1609
1da177e4
LT		 1610/*
1611 * Simple resource managers for sockets.
1612 */
1613
1614
4ec93edb
YH		 1615/*
1616 * Write buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1617 */
1618void sock_wfree(struct sk_buff *skb)
1619{
1620 struct sock *sk = skb->sk;
d99927f4 1621 unsigned int len = skb->truesize;
1da177e4 1622
d99927f4
ED		 1623 if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1624 /*
1625 * Keep a reference on sk_wmem_alloc, this will be released
1626 * after sk_write_space() call
1627 */
1628 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1da177e4 1629 sk->sk_write_space(sk);
d99927f4
ED		 1630 len = 1;
1631 }
2b85a34e 1632 /*
d99927f4
ED		 1633 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1634 * could not do because of in-flight packets
2b85a34e 1635 */
d99927f4 1636 if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
2b85a34e 1637 __sk_free(sk);
1da177e4 1638}
2a91525c 1639EXPORT_SYMBOL(sock_wfree);
1da177e4 1640
1d2077ac
ED		 1641/* This variant of sock_wfree() is used by TCP,
1642 * since it sets SOCK_USE_WRITE_QUEUE.
1643 */
1644void __sock_wfree(struct sk_buff *skb)
1645{
1646 struct sock *sk = skb->sk;
1647
1648 if (atomic_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1649 __sk_free(sk);
1650}
1651
9e17f8a4
ED		 1652void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1653{
1654 skb_orphan(skb);
1655 skb->sk = sk;
1656#ifdef CONFIG_INET
1657 if (unlikely(!sk_fullsock(sk))) {
1658 skb->destructor = sock_edemux;
1659 sock_hold(sk);
1660 return;
1661 }
1662#endif
1663 skb->destructor = sock_wfree;
1664 skb_set_hash_from_sk(skb, sk);
1665 /*
1666 * We used to take a refcount on sk, but following operation
1667 * is enough to guarantee sk_free() wont free this sock until
1668 * all in-flight packets are completed
1669 */
1670 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1671}
1672EXPORT_SYMBOL(skb_set_owner_w);
1673
1d2077ac
ED		 1674/* This helper is used by netem, as it can hold packets in its
1675 * delay queue. We want to allow the owner socket to send more
1676 * packets, as if they were already TX completed by a typical driver.
1677 * But we also want to keep skb->sk set because some packet schedulers
1678 * rely on it (sch_fq for example). So we set skb->truesize to a small
1679 * amount (1) and decrease sk_wmem_alloc accordingly.
1680 */
f2f872f9
ED		 1681void skb_orphan_partial(struct sk_buff *skb)
1682{
1d2077ac
ED		 1683 /* If this skb is a TCP pure ACK or already went here,
1684 * we have nothing to do. 2 is already a very small truesize.
1685 */
1686 if (skb->truesize <= 2)
1687 return;
1688
f2f872f9
ED		 1689 /* TCP stack sets skb->ooo_okay based on sk_wmem_alloc,
1690 * so we do not completely orphan skb, but transfert all
1691 * accounted bytes but one, to avoid unexpected reorders.
1692 */
1693 if (skb->destructor == sock_wfree
1694#ifdef CONFIG_INET
1695 || skb->destructor == tcp_wfree
1696#endif
1697 ) {
1698 atomic_sub(skb->truesize - 1, &skb->sk->sk_wmem_alloc);
1699 skb->truesize = 1;
1700 } else {
1701 skb_orphan(skb);
1702 }
1703}
1704EXPORT_SYMBOL(skb_orphan_partial);
1705
4ec93edb
YH		 1706/*
1707 * Read buffer destructor automatically called from kfree_skb.
1da177e4
LT		 1708 */
1709void sock_rfree(struct sk_buff *skb)
1710{
1711 struct sock *sk = skb->sk;
d361fd59 1712 unsigned int len = skb->truesize;
1da177e4 1713
d361fd59
ED		 1714 atomic_sub(len, &sk->sk_rmem_alloc);
1715 sk_mem_uncharge(sk, len);
1da177e4 1716}
2a91525c 1717EXPORT_SYMBOL(sock_rfree);
1da177e4 1718
7768eed8
OH		 1719/*
1720 * Buffer destructor for skbs that are not used directly in read or write
1721 * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1722 */
62bccb8c
AD		 1723void sock_efree(struct sk_buff *skb)
1724{
1725 sock_put(skb->sk);
1726}
1727EXPORT_SYMBOL(sock_efree);
1728
976d0201 1729kuid_t sock_i_uid(struct sock *sk)
1da177e4 1730{
976d0201 1731 kuid_t uid;
1da177e4 1732
f064af1e 1733 read_lock_bh(&sk->sk_callback_lock);
976d0201 1734 uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
f064af1e 1735 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1736 return uid;
1737}
2a91525c 1738EXPORT_SYMBOL(sock_i_uid);
1da177e4
LT		 1739
1740unsigned long sock_i_ino(struct sock *sk)
1741{
1742 unsigned long ino;
1743
f064af1e 1744 read_lock_bh(&sk->sk_callback_lock);
1da177e4 1745 ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
f064af1e 1746 read_unlock_bh(&sk->sk_callback_lock);
1da177e4
LT		 1747 return ino;
1748}
2a91525c 1749EXPORT_SYMBOL(sock_i_ino);
1da177e4
LT		 1750
1751/*
1752 * Allocate a skb from the socket's send buffer.
1753 */
86a76caf 1754struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
dd0fc66f 1755 gfp_t priority)
1da177e4
LT		 1756{
1757 if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
2a91525c 1758 struct sk_buff *skb = alloc_skb(size, priority);
1da177e4
LT		 1759 if (skb) {
1760 skb_set_owner_w(skb, sk);
1761 return skb;
1762 }
1763 }
1764 return NULL;
1765}
2a91525c 1766EXPORT_SYMBOL(sock_wmalloc);
1da177e4 1767
4ec93edb 1768/*
1da177e4 1769 * Allocate a memory block from the socket's option memory buffer.
4ec93edb 1770 */
dd0fc66f 1771void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1da177e4 1772{
95c96174 1773 if ((unsigned int)size <= sysctl_optmem_max &&
1da177e4
LT		 1774 atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1775 void *mem;
1776 /* First do the add, to avoid the race if kmalloc
4ec93edb 1777 * might sleep.
1da177e4
LT		 1778 */
1779 atomic_add(size, &sk->sk_omem_alloc);
1780 mem = kmalloc(size, priority);
1781 if (mem)
1782 return mem;
1783 atomic_sub(size, &sk->sk_omem_alloc);
1784 }
1785 return NULL;
1786}
2a91525c 1787EXPORT_SYMBOL(sock_kmalloc);
1da177e4 1788
79e88659
DB		 1789/* Free an option memory block. Note, we actually want the inline
1790 * here as this allows gcc to detect the nullify and fold away the
1791 * condition entirely.
1da177e4 1792 */
79e88659
DB		 1793static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
1794 const bool nullify)
1da177e4 1795{
e53da5fb
DM		 1796 if (WARN_ON_ONCE(!mem))
1797 return;
79e88659
DB		 1798 if (nullify)
1799 kzfree(mem);
1800 else
1801 kfree(mem);
1da177e4
LT		 1802 atomic_sub(size, &sk->sk_omem_alloc);
1803}
79e88659
DB		 1804
1805void sock_kfree_s(struct sock *sk, void *mem, int size)
1806{
1807 __sock_kfree_s(sk, mem, size, false);
1808}
2a91525c 1809EXPORT_SYMBOL(sock_kfree_s);
1da177e4 1810
79e88659
DB		 1811void sock_kzfree_s(struct sock *sk, void *mem, int size)
1812{
1813 __sock_kfree_s(sk, mem, size, true);
1814}
1815EXPORT_SYMBOL(sock_kzfree_s);
1816
1da177e4
LT		 1817/* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1818 I think, these locks should be removed for datagram sockets.
1819 */
2a91525c 1820static long sock_wait_for_wmem(struct sock *sk, long timeo)
1da177e4
LT		 1821{
1822 DEFINE_WAIT(wait);
1823
9cd3e072 1824 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1da177e4
LT		 1825 for (;;) {
1826 if (!timeo)
1827 break;
1828 if (signal_pending(current))
1829 break;
1830 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
aa395145 1831 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1da177e4
LT		 1832 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1833 break;
1834 if (sk->sk_shutdown & SEND_SHUTDOWN)
1835 break;
1836 if (sk->sk_err)
1837 break;
1838 timeo = schedule_timeout(timeo);
1839 }
aa395145 1840 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 1841 return timeo;
1842}
1843
1844
1845/*
1846 * Generic send/receive buffer handlers
1847 */
1848
4cc7f68d
HX		 1849struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1850 unsigned long data_len, int noblock,
28d64271 1851 int *errcode, int max_page_order)
1da177e4 1852{
2e4e4410 1853 struct sk_buff *skb;
1da177e4
LT		 1854 long timeo;
1855 int err;
1856
1da177e4 1857 timeo = sock_sndtimeo(sk, noblock);
2e4e4410 1858 for (;;) {
1da177e4
LT		 1859 err = sock_error(sk);
1860 if (err != 0)
1861 goto failure;
1862
1863 err = -EPIPE;
1864 if (sk->sk_shutdown & SEND_SHUTDOWN)
1865 goto failure;
1866
2e4e4410
ED		 1867 if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
1868 break;
28d64271 1869
9cd3e072 1870 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2e4e4410
ED		 1871 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1872 err = -EAGAIN;
1873 if (!timeo)
1da177e4 1874 goto failure;
2e4e4410
ED		 1875 if (signal_pending(current))
1876 goto interrupted;
1877 timeo = sock_wait_for_wmem(sk, timeo);
1da177e4 1878 }
2e4e4410
ED		 1879 skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
1880 errcode, sk->sk_allocation);
1881 if (skb)
1882 skb_set_owner_w(skb, sk);
1da177e4
LT		 1883 return skb;
1884
1885interrupted:
1886 err = sock_intr_errno(timeo);
1887failure:
1888 *errcode = err;
1889 return NULL;
1890}
4cc7f68d 1891EXPORT_SYMBOL(sock_alloc_send_pskb);
1da177e4 1892
4ec93edb 1893struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1da177e4
LT		 1894 int noblock, int *errcode)
1895{
28d64271 1896 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1da177e4 1897}
2a91525c 1898EXPORT_SYMBOL(sock_alloc_send_skb);
1da177e4 1899
39771b12
WB		 1900int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1901 struct sockcm_cookie *sockc)
1902{
3dd17e63
SHY		 1903 u32 tsflags;
1904
39771b12
WB		 1905 switch (cmsg->cmsg_type) {
1906 case SO_MARK:
1907 if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
1908 return -EPERM;
1909 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1910 return -EINVAL;
1911 sockc->mark = *(u32 *)CMSG_DATA(cmsg);
1912 break;
3dd17e63
SHY		 1913 case SO_TIMESTAMPING:
1914 if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
1915 return -EINVAL;
1916
1917 tsflags = *(u32 *)CMSG_DATA(cmsg);
1918 if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
1919 return -EINVAL;
1920
1921 sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
1922 sockc->tsflags |= tsflags;
1923 break;
779f1ede
SHY		 1924 /* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
1925 case SCM_RIGHTS:
1926 case SCM_CREDENTIALS:
1927 break;
39771b12
WB		 1928 default:
1929 return -EINVAL;
1930 }
1931 return 0;
1932}
1933EXPORT_SYMBOL(__sock_cmsg_send);
1934
f28ea365
EJ		 1935int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1936 struct sockcm_cookie *sockc)
1937{
1938 struct cmsghdr *cmsg;
39771b12 1939 int ret;
f28ea365
EJ		 1940
1941 for_each_cmsghdr(cmsg, msg) {
1942 if (!CMSG_OK(msg, cmsg))
1943 return -EINVAL;
1944 if (cmsg->cmsg_level != SOL_SOCKET)
1945 continue;
39771b12
WB		 1946 ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
1947 if (ret)
1948 return ret;
f28ea365
EJ		 1949 }
1950 return 0;
1951}
1952EXPORT_SYMBOL(sock_cmsg_send);
1953
5640f768
ED		 1954/* On 32bit arches, an skb frag is limited to 2^15 */
1955#define SKB_FRAG_PAGE_ORDER get_order(32768)
1956
400dfd3a
ED		 1957/**
1958 * skb_page_frag_refill - check that a page_frag contains enough room
1959 * @sz: minimum size of the fragment we want to get
1960 * @pfrag: pointer to page_frag
82d5e2b8 1961 * @gfp: priority for memory allocation
400dfd3a
ED		 1962 *
1963 * Note: While this allocator tries to use high order pages, there is
1964 * no guarantee that allocations succeed. Therefore, @sz MUST be
1965 * less or equal than PAGE_SIZE.
1966 */
d9b2938a 1967bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
5640f768 1968{
5640f768 1969 if (pfrag->page) {
fe896d18 1970 if (page_ref_count(pfrag->page) == 1) {
5640f768
ED		 1971 pfrag->offset = 0;
1972 return true;
1973 }
400dfd3a 1974 if (pfrag->offset + sz <= pfrag->size)
5640f768
ED		 1975 return true;
1976 put_page(pfrag->page);
1977 }
1978
d9b2938a
ED		 1979 pfrag->offset = 0;
1980 if (SKB_FRAG_PAGE_ORDER) {
d0164adc
MG		 1981 /* Avoid direct reclaim but allow kswapd to wake */
1982 pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
1983 __GFP_COMP | __GFP_NOWARN |
1984 __GFP_NORETRY,
d9b2938a 1985 SKB_FRAG_PAGE_ORDER);
5640f768 1986 if (likely(pfrag->page)) {
d9b2938a 1987 pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
5640f768
ED		 1988 return true;
1989 }
d9b2938a
ED		 1990 }
1991 pfrag->page = alloc_page(gfp);
1992 if (likely(pfrag->page)) {
1993 pfrag->size = PAGE_SIZE;
1994 return true;
1995 }
400dfd3a
ED		 1996 return false;
1997}
1998EXPORT_SYMBOL(skb_page_frag_refill);
1999
2000bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2001{
2002 if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2003 return true;
2004
5640f768
ED		 2005 sk_enter_memory_pressure(sk);
2006 sk_stream_moderate_sndbuf(sk);
2007 return false;
2008}
2009EXPORT_SYMBOL(sk_page_frag_refill);
2010
1da177e4 2011static void __lock_sock(struct sock *sk)
f39234d6
NK		 2012 __releases(&sk->sk_lock.slock)
2013 __acquires(&sk->sk_lock.slock)
1da177e4
LT		 2014{
2015 DEFINE_WAIT(wait);
2016
e71a4783 2017 for (;;) {
1da177e4
LT		 2018 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2019 TASK_UNINTERRUPTIBLE);
2020 spin_unlock_bh(&sk->sk_lock.slock);
2021 schedule();
2022 spin_lock_bh(&sk->sk_lock.slock);
e71a4783 2023 if (!sock_owned_by_user(sk))
1da177e4
LT		 2024 break;
2025 }
2026 finish_wait(&sk->sk_lock.wq, &wait);
2027}
2028
2029static void __release_sock(struct sock *sk)
f39234d6
NK		 2030 __releases(&sk->sk_lock.slock)
2031 __acquires(&sk->sk_lock.slock)
1da177e4 2032{
5413d1ba 2033 struct sk_buff *skb, *next;
1da177e4 2034
5413d1ba 2035 while ((skb = sk->sk_backlog.head) != NULL) {
1da177e4 2036 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1da177e4 2037
5413d1ba 2038 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4 2039
5413d1ba
ED		 2040 do {
2041 next = skb->next;
e4cbb02a 2042 prefetch(next);
7fee226a 2043 WARN_ON_ONCE(skb_dst_is_noref(skb));
1da177e4 2044 skb->next = NULL;
c57943a1 2045 sk_backlog_rcv(sk, skb);
1da177e4 2046
5413d1ba 2047 cond_resched();
1da177e4
LT		 2048
2049 skb = next;
2050 } while (skb != NULL);
2051
5413d1ba
ED		 2052 spin_lock_bh(&sk->sk_lock.slock);
2053 }
8eae939f
ZY		 2054
2055 /*
2056 * Doing the zeroing here guarantee we can not loop forever
2057 * while a wild producer attempts to flood us.
2058 */
2059 sk->sk_backlog.len = 0;
1da177e4
LT		 2060}
2061
d41a69f1
ED		 2062void __sk_flush_backlog(struct sock *sk)
2063{
2064 spin_lock_bh(&sk->sk_lock.slock);
2065 __release_sock(sk);
2066 spin_unlock_bh(&sk->sk_lock.slock);
2067}
2068
1da177e4
LT		 2069/**
2070 * sk_wait_data - wait for data to arrive at sk_receive_queue
4dc3b16b
PP		 2071 * @sk: sock to wait on
2072 * @timeo: for how long
dfbafc99 2073 * @skb: last skb seen on sk_receive_queue
1da177e4
LT		 2074 *
2075 * Now socket state including sk->sk_err is changed only under lock,
2076 * hence we may omit checks after joining wait queue.
2077 * We check receive queue before schedule() only as optimization;
2078 * it is very likely that release_sock() added new data.
2079 */
dfbafc99 2080int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
1da177e4
LT		 2081{
2082 int rc;
2083 DEFINE_WAIT(wait);
2084
aa395145 2085 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
9cd3e072 2086 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
dfbafc99 2087 rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb);
9cd3e072 2088 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
aa395145 2089 finish_wait(sk_sleep(sk), &wait);
1da177e4
LT		 2090 return rc;
2091}
1da177e4
LT		 2092EXPORT_SYMBOL(sk_wait_data);
2093
3ab224be
HA		 2094/**
2095 * __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2096 * @sk: socket
2097 * @size: memory size to allocate
2098 * @kind: allocation type
2099 *
2100 * If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2101 * rmem allocation. This function assumes that protocols which have
2102 * memory_pressure use sk_wmem_queued as write buffer accounting.
2103 */
2104int __sk_mem_schedule(struct sock *sk, int size, int kind)
2105{
2106 struct proto *prot = sk->sk_prot;
2107 int amt = sk_mem_pages(size);
8d987e5c 2108 long allocated;
3ab224be
HA		 2109
2110 sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
180d8cd9 2111
e805605c
JW		 2112 allocated = sk_memory_allocated_add(sk, amt);
2113
baac50bb
JW		 2114 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2115 !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
e805605c 2116 goto suppress_allocation;
3ab224be
HA		 2117
2118 /* Under limit. */
e805605c 2119 if (allocated <= sk_prot_mem_limits(sk, 0)) {
180d8cd9 2120 sk_leave_memory_pressure(sk);
3ab224be
HA		 2121 return 1;
2122 }
2123
e805605c
JW		 2124 /* Under pressure. */
2125 if (allocated > sk_prot_mem_limits(sk, 1))
180d8cd9 2126 sk_enter_memory_pressure(sk);
3ab224be 2127
e805605c
JW		 2128 /* Over hard limit. */
2129 if (allocated > sk_prot_mem_limits(sk, 2))
3ab224be
HA		 2130 goto suppress_allocation;
2131
2132 /* guarantee minimum buffer size under pressure */
2133 if (kind == SK_MEM_RECV) {
2134 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
2135 return 1;
180d8cd9 2136
3ab224be
HA		 2137 } else { /* SK_MEM_SEND */
2138 if (sk->sk_type == SOCK_STREAM) {
2139 if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
2140 return 1;
2141 } else if (atomic_read(&sk->sk_wmem_alloc) <
2142 prot->sysctl_wmem[0])
2143 return 1;
2144 }
2145
180d8cd9 2146 if (sk_has_memory_pressure(sk)) {
1748376b
ED		 2147 int alloc;
2148
180d8cd9 2149 if (!sk_under_memory_pressure(sk))
1748376b 2150 return 1;
180d8cd9
GC		 2151 alloc = sk_sockets_allocated_read_positive(sk);
2152 if (sk_prot_mem_limits(sk, 2) > alloc *
3ab224be
HA		 2153 sk_mem_pages(sk->sk_wmem_queued +
2154 atomic_read(&sk->sk_rmem_alloc) +
2155 sk->sk_forward_alloc))
2156 return 1;
2157 }
2158
2159suppress_allocation:
2160
2161 if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2162 sk_stream_moderate_sndbuf(sk);
2163
2164 /* Fail only if socket is _under_ its sndbuf.
2165 * In this case we cannot block, so that we have to fail.
2166 */
2167 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2168 return 1;
2169 }
2170
3847ce32
SM		 2171 trace_sock_exceed_buf_limit(sk, prot, allocated);
2172
3ab224be
HA		 2173 /* Alas. Undo changes. */
2174 sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
180d8cd9 2175
0e90b31f 2176 sk_memory_allocated_sub(sk, amt);
180d8cd9 2177
baac50bb
JW		 2178 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2179 mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
e805605c 2180
3ab224be
HA		 2181 return 0;
2182}
3ab224be
HA		 2183EXPORT_SYMBOL(__sk_mem_schedule);
2184
2185/**
69dba9bb 2186 * __sk_mem_reclaim - reclaim memory_allocated
3ab224be 2187 * @sk: socket
1a24e04e 2188 * @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
3ab224be 2189 */
1a24e04e 2190void __sk_mem_reclaim(struct sock *sk, int amount)
3ab224be 2191{
1a24e04e
ED		 2192 amount >>= SK_MEM_QUANTUM_SHIFT;
2193 sk_memory_allocated_sub(sk, amount);
2194 sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
3ab224be 2195
baac50bb
JW		 2196 if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2197 mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
e805605c 2198
180d8cd9
GC		 2199 if (sk_under_memory_pressure(sk) &&
2200 (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2201 sk_leave_memory_pressure(sk);
3ab224be 2202}
3ab224be
HA		 2203EXPORT_SYMBOL(__sk_mem_reclaim);
2204
627d2d6b 2205int sk_set_peek_off(struct sock *sk, int val)
2206{
2207 if (val < 0)
2208 return -EINVAL;
2209
2210 sk->sk_peek_off = val;
2211 return 0;
2212}
2213EXPORT_SYMBOL_GPL(sk_set_peek_off);
3ab224be 2214
1da177e4
LT		 2215/*
2216 * Set of default routines for initialising struct proto_ops when
2217 * the protocol does not support a particular function. In certain
2218 * cases where it makes no sense for a protocol to have a "do nothing"
2219 * function, some default processing is provided.
2220 */
2221
2222int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2223{
2224 return -EOPNOTSUPP;
2225}
2a91525c 2226EXPORT_SYMBOL(sock_no_bind);
1da177e4 2227
4ec93edb 2228int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 2229 int len, int flags)
2230{
2231 return -EOPNOTSUPP;
2232}
2a91525c 2233EXPORT_SYMBOL(sock_no_connect);
1da177e4
LT		 2234
2235int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2236{
2237 return -EOPNOTSUPP;
2238}
2a91525c 2239EXPORT_SYMBOL(sock_no_socketpair);
1da177e4
LT		 2240
2241int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2242{
2243 return -EOPNOTSUPP;
2244}
2a91525c 2245EXPORT_SYMBOL(sock_no_accept);
1da177e4 2246
4ec93edb 2247int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1da177e4
LT		 2248 int *len, int peer)
2249{
2250 return -EOPNOTSUPP;
2251}
2a91525c 2252EXPORT_SYMBOL(sock_no_getname);
1da177e4 2253
2a91525c 2254unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1da177e4
LT		 2255{
2256 return 0;
2257}
2a91525c 2258EXPORT_SYMBOL(sock_no_poll);
1da177e4
LT		 2259
2260int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2261{
2262 return -EOPNOTSUPP;
2263}
2a91525c 2264EXPORT_SYMBOL(sock_no_ioctl);
1da177e4
LT		 2265
2266int sock_no_listen(struct socket *sock, int backlog)
2267{
2268 return -EOPNOTSUPP;
2269}
2a91525c 2270EXPORT_SYMBOL(sock_no_listen);
1da177e4
LT		 2271
2272int sock_no_shutdown(struct socket *sock, int how)
2273{
2274 return -EOPNOTSUPP;
2275}
2a91525c 2276EXPORT_SYMBOL(sock_no_shutdown);
1da177e4
LT		 2277
2278int sock_no_setsockopt(struct socket *sock, int level, int optname,
b7058842 2279 char __user *optval, unsigned int optlen)
1da177e4
LT		 2280{
2281 return -EOPNOTSUPP;
2282}
2a91525c 2283EXPORT_SYMBOL(sock_no_setsockopt);
1da177e4
LT		 2284
2285int sock_no_getsockopt(struct socket *sock, int level, int optname,
2286 char __user *optval, int __user *optlen)
2287{
2288 return -EOPNOTSUPP;
2289}
2a91525c 2290EXPORT_SYMBOL(sock_no_getsockopt);
1da177e4 2291
1b784140 2292int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
1da177e4
LT		 2293{
2294 return -EOPNOTSUPP;
2295}
2a91525c 2296EXPORT_SYMBOL(sock_no_sendmsg);
1da177e4 2297
1b784140
YX		 2298int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2299 int flags)
1da177e4
LT		 2300{
2301 return -EOPNOTSUPP;
2302}
2a91525c 2303EXPORT_SYMBOL(sock_no_recvmsg);
1da177e4
LT		 2304
2305int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2306{
2307 /* Mirror missing mmap method error code */
2308 return -ENODEV;
2309}
2a91525c 2310EXPORT_SYMBOL(sock_no_mmap);
1da177e4
LT		 2311
2312ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2313{
2314 ssize_t res;
2315 struct msghdr msg = {.msg_flags = flags};
2316 struct kvec iov;
2317 char *kaddr = kmap(page);
2318 iov.iov_base = kaddr + offset;
2319 iov.iov_len = size;
2320 res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2321 kunmap(page);
2322 return res;
2323}
2a91525c 2324EXPORT_SYMBOL(sock_no_sendpage);
1da177e4
LT		 2325
2326/*
2327 * Default Socket Callbacks
2328 */
2329
2330static void sock_def_wakeup(struct sock *sk)
2331{
43815482
ED		 2332 struct socket_wq *wq;
2333
2334 rcu_read_lock();
2335 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 2336 if (skwq_has_sleeper(wq))
43815482
ED		 2337 wake_up_interruptible_all(&wq->wait);
2338 rcu_read_unlock();
1da177e4
LT		 2339}
2340
2341static void sock_def_error_report(struct sock *sk)
2342{
43815482
ED		 2343 struct socket_wq *wq;
2344
2345 rcu_read_lock();
2346 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 2347 if (skwq_has_sleeper(wq))
43815482 2348 wake_up_interruptible_poll(&wq->wait, POLLERR);
8d8ad9d7 2349 sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
43815482 2350 rcu_read_unlock();
1da177e4
LT		 2351}
2352
676d2369 2353static void sock_def_readable(struct sock *sk)
1da177e4 2354{
43815482
ED		 2355 struct socket_wq *wq;
2356
2357 rcu_read_lock();
2358 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 2359 if (skwq_has_sleeper(wq))
2c6607c6 2360 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
37e5540b 2361 POLLRDNORM | POLLRDBAND);
8d8ad9d7 2362 sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
43815482 2363 rcu_read_unlock();
1da177e4
LT		 2364}
2365
2366static void sock_def_write_space(struct sock *sk)
2367{
43815482
ED		 2368 struct socket_wq *wq;
2369
2370 rcu_read_lock();
1da177e4
LT		 2371
2372 /* Do not wake up a writer until he can make "significant"
2373 * progress. --DaveM
2374 */
e71a4783 2375 if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
43815482 2376 wq = rcu_dereference(sk->sk_wq);
1ce0bf50 2377 if (skwq_has_sleeper(wq))
43815482 2378 wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
37e5540b 2379 POLLWRNORM | POLLWRBAND);
1da177e4
LT		 2380
2381 /* Should agree with poll, otherwise some programs break */
2382 if (sock_writeable(sk))
8d8ad9d7 2383 sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1da177e4
LT		 2384 }
2385
43815482 2386 rcu_read_unlock();
1da177e4
LT		 2387}
2388
2389static void sock_def_destruct(struct sock *sk)
2390{
1da177e4
LT		 2391}
2392
2393void sk_send_sigurg(struct sock *sk)
2394{
2395 if (sk->sk_socket && sk->sk_socket->file)
2396 if (send_sigurg(&sk->sk_socket->file->f_owner))
8d8ad9d7 2397 sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1da177e4 2398}
2a91525c 2399EXPORT_SYMBOL(sk_send_sigurg);
1da177e4
LT		 2400
2401void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2402 unsigned long expires)
2403{
2404 if (!mod_timer(timer, expires))
2405 sock_hold(sk);
2406}
1da177e4
LT		 2407EXPORT_SYMBOL(sk_reset_timer);
2408
2409void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2410{
25cc4ae9 2411 if (del_timer(timer))
1da177e4
LT		 2412 __sock_put(sk);
2413}
1da177e4
LT		 2414EXPORT_SYMBOL(sk_stop_timer);
2415
2416void sock_init_data(struct socket *sock, struct sock *sk)
2417{
2418 skb_queue_head_init(&sk->sk_receive_queue);
2419 skb_queue_head_init(&sk->sk_write_queue);
2420 skb_queue_head_init(&sk->sk_error_queue);
2421
2422 sk->sk_send_head = NULL;
2423
2424 init_timer(&sk->sk_timer);
4ec93edb 2425
1da177e4
LT		 2426 sk->sk_allocation = GFP_KERNEL;
2427 sk->sk_rcvbuf = sysctl_rmem_default;
2428 sk->sk_sndbuf = sysctl_wmem_default;
2429 sk->sk_state = TCP_CLOSE;
972692e0 2430 sk_set_socket(sk, sock);
1da177e4
LT		 2431
2432 sock_set_flag(sk, SOCK_ZAPPED);
2433
e71a4783 2434 if (sock) {
1da177e4 2435 sk->sk_type = sock->type;
43815482 2436 sk->sk_wq = sock->wq;
1da177e4
LT		 2437 sock->sk = sk;
2438 } else
43815482 2439 sk->sk_wq = NULL;
1da177e4 2440
1da177e4 2441 rwlock_init(&sk->sk_callback_lock);
443aef0e
PZ		 2442 lockdep_set_class_and_name(&sk->sk_callback_lock,
2443 af_callback_keys + sk->sk_family,
2444 af_family_clock_key_strings[sk->sk_family]);
1da177e4
LT		 2445
2446 sk->sk_state_change = sock_def_wakeup;
2447 sk->sk_data_ready = sock_def_readable;
2448 sk->sk_write_space = sock_def_write_space;
2449 sk->sk_error_report = sock_def_error_report;
2450 sk->sk_destruct = sock_def_destruct;
2451
5640f768
ED		 2452 sk->sk_frag.page = NULL;
2453 sk->sk_frag.offset = 0;
ef64a54f 2454 sk->sk_peek_off = -1;
1da177e4 2455
109f6e39
EB		 2456 sk->sk_peer_pid = NULL;
2457 sk->sk_peer_cred = NULL;
1da177e4
LT		 2458 sk->sk_write_pending = 0;
2459 sk->sk_rcvlowat = 1;
2460 sk->sk_rcvtimeo = MAX_SCHEDULE_TIMEOUT;
2461 sk->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
2462
f37f0afb 2463 sk->sk_stamp = ktime_set(-1L, 0);
1da177e4 2464
e0d1095a 2465#ifdef CONFIG_NET_RX_BUSY_POLL
06021292 2466 sk->sk_napi_id = 0;
64b0dc51 2467 sk->sk_ll_usec = sysctl_net_busy_read;
06021292
ET		 2468#endif
2469
62748f32 2470 sk->sk_max_pacing_rate = ~0U;
7eec4174 2471 sk->sk_pacing_rate = ~0U;
70da268b 2472 sk->sk_incoming_cpu = -1;
4dc6dc71
ED		 2473 /*
2474 * Before updating sk_refcnt, we must commit prior changes to memory
2475 * (Documentation/RCU/rculist_nulls.txt for details)
2476 */
2477 smp_wmb();
1da177e4 2478 atomic_set(&sk->sk_refcnt, 1);
33c732c3 2479 atomic_set(&sk->sk_drops, 0);
1da177e4 2480}
2a91525c 2481EXPORT_SYMBOL(sock_init_data);
1da177e4 2482
b5606c2d 2483void lock_sock_nested(struct sock *sk, int subclass)
1da177e4
LT		 2484{
2485 might_sleep();
a5b5bb9a 2486 spin_lock_bh(&sk->sk_lock.slock);
d2e9117c 2487 if (sk->sk_lock.owned)
1da177e4 2488 __lock_sock(sk);
d2e9117c 2489 sk->sk_lock.owned = 1;
a5b5bb9a
IM		 2490 spin_unlock(&sk->sk_lock.slock);
2491 /*
2492 * The sk_lock has mutex_lock() semantics here:
2493 */
fcc70d5f 2494 mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
a5b5bb9a 2495 local_bh_enable();
1da177e4 2496}
fcc70d5f 2497EXPORT_SYMBOL(lock_sock_nested);
1da177e4 2498
b5606c2d 2499void release_sock(struct sock *sk)
1da177e4 2500{
a5b5bb9a 2501 spin_lock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2502 if (sk->sk_backlog.tail)
2503 __release_sock(sk);
46d3ceab 2504
c3f9b018
ED		 2505 /* Warning : release_cb() might need to release sk ownership,
2506 * ie call sock_release_ownership(sk) before us.
2507 */
46d3ceab
ED		 2508 if (sk->sk_prot->release_cb)
2509 sk->sk_prot->release_cb(sk);
2510
c3f9b018 2511 sock_release_ownership(sk);
a5b5bb9a
IM		 2512 if (waitqueue_active(&sk->sk_lock.wq))
2513 wake_up(&sk->sk_lock.wq);
2514 spin_unlock_bh(&sk->sk_lock.slock);
1da177e4
LT		 2515}
2516EXPORT_SYMBOL(release_sock);
2517
8a74ad60
ED		 2518/**
2519 * lock_sock_fast - fast version of lock_sock
2520 * @sk: socket
2521 *
2522 * This version should be used for very small section, where process wont block
2523 * return false if fast path is taken
2524 * sk_lock.slock locked, owned = 0, BH disabled
2525 * return true if slow path is taken
2526 * sk_lock.slock unlocked, owned = 1, BH enabled
2527 */
2528bool lock_sock_fast(struct sock *sk)
2529{
2530 might_sleep();
2531 spin_lock_bh(&sk->sk_lock.slock);
2532
2533 if (!sk->sk_lock.owned)
2534 /*
2535 * Note : We must disable BH
2536 */
2537 return false;
2538
2539 __lock_sock(sk);
2540 sk->sk_lock.owned = 1;
2541 spin_unlock(&sk->sk_lock.slock);
2542 /*
2543 * The sk_lock has mutex_lock() semantics here:
2544 */
2545 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2546 local_bh_enable();
2547 return true;
2548}
2549EXPORT_SYMBOL(lock_sock_fast);
2550
1da177e4 2551int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
4ec93edb 2552{
b7aa0bf7 2553 struct timeval tv;
1da177e4 2554 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2555 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
b7aa0bf7
ED		 2556 tv = ktime_to_timeval(sk->sk_stamp);
2557 if (tv.tv_sec == -1)
1da177e4 2558 return -ENOENT;
b7aa0bf7
ED		 2559 if (tv.tv_sec == 0) {
2560 sk->sk_stamp = ktime_get_real();
2561 tv = ktime_to_timeval(sk->sk_stamp);
2562 }
2563 return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
4ec93edb 2564}
1da177e4
LT		 2565EXPORT_SYMBOL(sock_get_timestamp);
2566
ae40eb1e
ED		 2567int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2568{
2569 struct timespec ts;
2570 if (!sock_flag(sk, SOCK_TIMESTAMP))
20d49473 2571 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
ae40eb1e
ED		 2572 ts = ktime_to_timespec(sk->sk_stamp);
2573 if (ts.tv_sec == -1)
2574 return -ENOENT;
2575 if (ts.tv_sec == 0) {
2576 sk->sk_stamp = ktime_get_real();
2577 ts = ktime_to_timespec(sk->sk_stamp);
2578 }
2579 return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2580}
2581EXPORT_SYMBOL(sock_get_timestampns);
2582
20d49473 2583void sock_enable_timestamp(struct sock *sk, int flag)
4ec93edb 2584{
20d49473 2585 if (!sock_flag(sk, flag)) {
08e29af3
ED		 2586 unsigned long previous_flags = sk->sk_flags;
2587
20d49473
PO		 2588 sock_set_flag(sk, flag);
2589 /*
2590 * we just set one of the two flags which require net
2591 * time stamping, but time stamping might have been on
2592 * already because of the other one
2593 */
080a270f
HFS		 2594 if (sock_needs_netstamp(sk) &&
2595 !(previous_flags & SK_FLAGS_TIMESTAMP))
20d49473 2596 net_enable_timestamp();
1da177e4
LT		 2597 }
2598}
1da177e4 2599
cb820f8e
RC		 2600int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2601 int level, int type)
2602{
2603 struct sock_exterr_skb *serr;
364a9e93 2604 struct sk_buff *skb;
cb820f8e
RC		 2605 int copied, err;
2606
2607 err = -EAGAIN;
364a9e93 2608 skb = sock_dequeue_err_skb(sk);
cb820f8e
RC		 2609 if (skb == NULL)
2610 goto out;
2611
2612 copied = skb->len;
2613 if (copied > len) {
2614 msg->msg_flags |= MSG_TRUNC;
2615 copied = len;
2616 }
51f3d02b 2617 err = skb_copy_datagram_msg(skb, 0, msg, copied);
cb820f8e
RC		 2618 if (err)
2619 goto out_free_skb;
2620
2621 sock_recv_timestamp(msg, sk, skb);
2622
2623 serr = SKB_EXT_ERR(skb);
2624 put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2625
2626 msg->msg_flags |= MSG_ERRQUEUE;
2627 err = copied;
2628
cb820f8e
RC		 2629out_free_skb:
2630 kfree_skb(skb);
2631out:
2632 return err;
2633}
2634EXPORT_SYMBOL(sock_recv_errqueue);
2635
1da177e4
LT		 2636/*
2637 * Get a socket option on an socket.
2638 *
2639 * FIX: POSIX 1003.1g is very ambiguous here. It states that
2640 * asynchronous errors should be reported by getsockopt. We assume
2641 * this means if you specify SO_ERROR (otherwise whats the point of it).
2642 */
2643int sock_common_getsockopt(struct socket *sock, int level, int optname,
2644 char __user *optval, int __user *optlen)
2645{
2646 struct sock *sk = sock->sk;
2647
2648 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2649}
1da177e4
LT		 2650EXPORT_SYMBOL(sock_common_getsockopt);
2651
3fdadf7d 2652#ifdef CONFIG_COMPAT
543d9cfe
ACM		 2653int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2654 char __user *optval, int __user *optlen)
3fdadf7d
DM		 2655{
2656 struct sock *sk = sock->sk;
2657
1e51f951 2658 if (sk->sk_prot->compat_getsockopt != NULL)
543d9cfe
ACM		 2659 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2660 optval, optlen);
3fdadf7d
DM		 2661 return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2662}
2663EXPORT_SYMBOL(compat_sock_common_getsockopt);
2664#endif
2665
1b784140
YX		 2666int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2667 int flags)
1da177e4
LT		 2668{
2669 struct sock *sk = sock->sk;
2670 int addr_len = 0;
2671 int err;
2672
1b784140 2673 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
1da177e4
LT		 2674 flags & ~MSG_DONTWAIT, &addr_len);
2675 if (err >= 0)
2676 msg->msg_namelen = addr_len;
2677 return err;
2678}
1da177e4
LT		 2679EXPORT_SYMBOL(sock_common_recvmsg);
2680
2681/*
2682 * Set socket options on an inet socket.
2683 */
2684int sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2685 char __user *optval, unsigned int optlen)
1da177e4
LT		 2686{
2687 struct sock *sk = sock->sk;
2688
2689 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2690}
1da177e4
LT		 2691EXPORT_SYMBOL(sock_common_setsockopt);
2692
3fdadf7d 2693#ifdef CONFIG_COMPAT
543d9cfe 2694int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
b7058842 2695 char __user *optval, unsigned int optlen)
3fdadf7d
DM		 2696{
2697 struct sock *sk = sock->sk;
2698
543d9cfe
ACM		 2699 if (sk->sk_prot->compat_setsockopt != NULL)
2700 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2701 optval, optlen);
3fdadf7d
DM		 2702 return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2703}
2704EXPORT_SYMBOL(compat_sock_common_setsockopt);
2705#endif
2706
1da177e4
LT		 2707void sk_common_release(struct sock *sk)
2708{
2709 if (sk->sk_prot->destroy)
2710 sk->sk_prot->destroy(sk);
2711
2712 /*
2713 * Observation: when sock_common_release is called, processes have
2714 * no access to socket. But net still has.
2715 * Step one, detach it from networking:
2716 *
2717 * A. Remove from hash tables.
2718 */
2719
2720 sk->sk_prot->unhash(sk);
2721
2722 /*
2723 * In this point socket cannot receive new packets, but it is possible
2724 * that some packets are in flight because some CPU runs receiver and
2725 * did hash table lookup before we unhashed socket. They will achieve
2726 * receive queue and will be purged by socket destructor.
2727 *
2728 * Also we still have packets pending on receive queue and probably,
2729 * our own packets waiting in device queues. sock_destroy will drain
2730 * receive queue, but transmitted packets will delay socket destruction
2731 * until the last reference will be released.
2732 */
2733
2734 sock_orphan(sk);
2735
2736 xfrm_sk_free_policy(sk);
2737
e6848976 2738 sk_refcnt_debug_release(sk);
5640f768
ED		 2739
2740 if (sk->sk_frag.page) {
2741 put_page(sk->sk_frag.page);
2742 sk->sk_frag.page = NULL;
2743 }
2744
1da177e4
LT		 2745 sock_put(sk);
2746}
1da177e4
LT		 2747EXPORT_SYMBOL(sk_common_release);
2748
13ff3d6f
PE		 2749#ifdef CONFIG_PROC_FS
2750#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1338d466
PE		 2751struct prot_inuse {
2752 int val[PROTO_INUSE_NR];
2753};
13ff3d6f
PE		 2754
2755static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
70ee1159
PE		 2756
2757#ifdef CONFIG_NET_NS
2758void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2759{
d6d9ca0f 2760 __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
70ee1159
PE		 2761}
2762EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2763
2764int sock_prot_inuse_get(struct net *net, struct proto *prot)
2765{
2766 int cpu, idx = prot->inuse_idx;
2767 int res = 0;
2768
2769 for_each_possible_cpu(cpu)
2770 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2771
2772 return res >= 0 ? res : 0;
2773}
2774EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2775
2c8c1e72 2776static int __net_init sock_inuse_init_net(struct net *net)
70ee1159
PE		 2777{
2778 net->core.inuse = alloc_percpu(struct prot_inuse);
2779 return net->core.inuse ? 0 : -ENOMEM;
2780}
2781
2c8c1e72 2782static void __net_exit sock_inuse_exit_net(struct net *net)
70ee1159
PE		 2783{
2784 free_percpu(net->core.inuse);
2785}
2786
2787static struct pernet_operations net_inuse_ops = {
2788 .init = sock_inuse_init_net,
2789 .exit = sock_inuse_exit_net,
2790};
2791
2792static __init int net_inuse_init(void)
2793{
2794 if (register_pernet_subsys(&net_inuse_ops))
2795 panic("Cannot initialize net inuse counters");
2796
2797 return 0;
2798}
2799
2800core_initcall(net_inuse_init);
2801#else
1338d466
PE		 2802static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2803
c29a0bc4 2804void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
1338d466 2805{
d6d9ca0f 2806 __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
1338d466
PE		 2807}
2808EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2809
c29a0bc4 2810int sock_prot_inuse_get(struct net *net, struct proto *prot)
1338d466
PE		 2811{
2812 int cpu, idx = prot->inuse_idx;
2813 int res = 0;
2814
2815 for_each_possible_cpu(cpu)
2816 res += per_cpu(prot_inuse, cpu).val[idx];
2817
2818 return res >= 0 ? res : 0;
2819}
2820EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
70ee1159 2821#endif
13ff3d6f
PE		 2822
2823static void assign_proto_idx(struct proto *prot)
2824{
2825 prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2826
2827 if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
e005d193 2828 pr_err("PROTO_INUSE_NR exhausted\n");
13ff3d6f
PE		 2829 return;
2830 }
2831
2832 set_bit(prot->inuse_idx, proto_inuse_idx);
2833}
2834
2835static void release_proto_idx(struct proto *prot)
2836{
2837 if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2838 clear_bit(prot->inuse_idx, proto_inuse_idx);
2839}
2840#else
2841static inline void assign_proto_idx(struct proto *prot)
2842{
2843}
2844
2845static inline void release_proto_idx(struct proto *prot)
2846{
2847}
2848#endif
2849
0159dfd3
ED		 2850static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
2851{
2852 if (!rsk_prot)
2853 return;
2854 kfree(rsk_prot->slab_name);
2855 rsk_prot->slab_name = NULL;
adf78eda
JL		 2856 kmem_cache_destroy(rsk_prot->slab);
2857 rsk_prot->slab = NULL;
0159dfd3
ED		 2858}
2859
2860static int req_prot_init(const struct proto *prot)
2861{
2862 struct request_sock_ops *rsk_prot = prot->rsk_prot;
2863
2864 if (!rsk_prot)
2865 return 0;
2866
2867 rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
2868 prot->name);
2869 if (!rsk_prot->slab_name)
2870 return -ENOMEM;
2871
2872 rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
2873 rsk_prot->obj_size, 0,
e96f78ab 2874 prot->slab_flags, NULL);
0159dfd3
ED		 2875
2876 if (!rsk_prot->slab) {
2877 pr_crit("%s: Can't create request sock SLAB cache!\n",
2878 prot->name);
2879 return -ENOMEM;
2880 }
2881 return 0;
2882}
2883
b733c007
PE		 2884int proto_register(struct proto *prot, int alloc_slab)
2885{
1da177e4
LT		 2886 if (alloc_slab) {
2887 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
271b72c7
ED		 2888 SLAB_HWCACHE_ALIGN | prot->slab_flags,
2889 NULL);
1da177e4
LT		 2890
2891 if (prot->slab == NULL) {
e005d193
JP		 2892 pr_crit("%s: Can't create sock SLAB cache!\n",
2893 prot->name);
60e7663d 2894 goto out;
1da177e4 2895 }
2e6599cb 2896
0159dfd3
ED		 2897 if (req_prot_init(prot))
2898 goto out_free_request_sock_slab;
8feaf0c0 2899
6d6ee43e 2900 if (prot->twsk_prot != NULL) {
faf23422 2901 prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
8feaf0c0 2902
7e56b5d6 2903 if (prot->twsk_prot->twsk_slab_name == NULL)
8feaf0c0
ACM		 2904 goto out_free_request_sock_slab;
2905
6d6ee43e 2906 prot->twsk_prot->twsk_slab =
7e56b5d6 2907 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
6d6ee43e 2908 prot->twsk_prot->twsk_obj_size,
3ab5aee7 2909 0,
52db70dc 2910 prot->slab_flags,
20c2df83 2911 NULL);
6d6ee43e 2912 if (prot->twsk_prot->twsk_slab == NULL)
8feaf0c0
ACM		 2913 goto out_free_timewait_sock_slab_name;
2914 }
1da177e4
LT		 2915 }
2916
36b77a52 2917 mutex_lock(&proto_list_mutex);
1da177e4 2918 list_add(&prot->node, &proto_list);
13ff3d6f 2919 assign_proto_idx(prot);
36b77a52 2920 mutex_unlock(&proto_list_mutex);
b733c007
PE		 2921 return 0;
2922
8feaf0c0 2923out_free_timewait_sock_slab_name:
7e56b5d6 2924 kfree(prot->twsk_prot->twsk_slab_name);
8feaf0c0 2925out_free_request_sock_slab:
0159dfd3
ED		 2926 req_prot_cleanup(prot->rsk_prot);
2927
2e6599cb
ACM		 2928 kmem_cache_destroy(prot->slab);
2929 prot->slab = NULL;
b733c007
PE		 2930out:
2931 return -ENOBUFS;
1da177e4 2932}
1da177e4
LT		 2933EXPORT_SYMBOL(proto_register);
2934
2935void proto_unregister(struct proto *prot)
2936{
36b77a52 2937 mutex_lock(&proto_list_mutex);
13ff3d6f 2938 release_proto_idx(prot);
0a3f4358 2939 list_del(&prot->node);
36b77a52 2940 mutex_unlock(&proto_list_mutex);
1da177e4 2941
adf78eda
JL		 2942 kmem_cache_destroy(prot->slab);
2943 prot->slab = NULL;
1da177e4 2944
0159dfd3 2945 req_prot_cleanup(prot->rsk_prot);
2e6599cb 2946
6d6ee43e 2947 if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
6d6ee43e 2948 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
7e56b5d6 2949 kfree(prot->twsk_prot->twsk_slab_name);
6d6ee43e 2950 prot->twsk_prot->twsk_slab = NULL;
8feaf0c0 2951 }
1da177e4 2952}
1da177e4
LT		 2953EXPORT_SYMBOL(proto_unregister);
2954
2955#ifdef CONFIG_PROC_FS
1da177e4 2956static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
36b77a52 2957 __acquires(proto_list_mutex)
1da177e4 2958{
36b77a52 2959 mutex_lock(&proto_list_mutex);
60f0438a 2960 return seq_list_start_head(&proto_list, *pos);
1da177e4
LT		 2961}
2962
2963static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2964{
60f0438a 2965 return seq_list_next(v, &proto_list, pos);
1da177e4
LT		 2966}
2967
2968static void proto_seq_stop(struct seq_file *seq, void *v)
36b77a52 2969 __releases(proto_list_mutex)
1da177e4 2970{
36b77a52 2971 mutex_unlock(&proto_list_mutex);
1da177e4
LT		 2972}
2973
2974static char proto_method_implemented(const void *method)
2975{
2976 return method == NULL ? 'n' : 'y';
2977}
180d8cd9
GC		 2978static long sock_prot_memory_allocated(struct proto *proto)
2979{
cb75a36c 2980 return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
180d8cd9
GC		 2981}
2982
2983static char *sock_prot_memory_pressure(struct proto *proto)
2984{
2985 return proto->memory_pressure != NULL ?
2986 proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2987}
1da177e4
LT		 2988
2989static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2990{
180d8cd9 2991
8d987e5c 2992 seq_printf(seq, "%-9s %4u %6d %6ld %-3s %6u %-3s %-10s "
1da177e4
LT		 2993 "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2994 proto->name,
2995 proto->obj_size,
14e943db 2996 sock_prot_inuse_get(seq_file_net(seq), proto),
180d8cd9
GC		 2997 sock_prot_memory_allocated(proto),
2998 sock_prot_memory_pressure(proto),
1da177e4
LT		 2999 proto->max_header,
3000 proto->slab == NULL ? "no" : "yes",
3001 module_name(proto->owner),
3002 proto_method_implemented(proto->close),
3003 proto_method_implemented(proto->connect),
3004 proto_method_implemented(proto->disconnect),
3005 proto_method_implemented(proto->accept),
3006 proto_method_implemented(proto->ioctl),
3007 proto_method_implemented(proto->init),
3008 proto_method_implemented(proto->destroy),
3009 proto_method_implemented(proto->shutdown),
3010 proto_method_implemented(proto->setsockopt),
3011 proto_method_implemented(proto->getsockopt),
3012 proto_method_implemented(proto->sendmsg),
3013 proto_method_implemented(proto->recvmsg),
3014 proto_method_implemented(proto->sendpage),
3015 proto_method_implemented(proto->bind),
3016 proto_method_implemented(proto->backlog_rcv),
3017 proto_method_implemented(proto->hash),
3018 proto_method_implemented(proto->unhash),
3019 proto_method_implemented(proto->get_port),
3020 proto_method_implemented(proto->enter_memory_pressure));
3021}
3022
3023static int proto_seq_show(struct seq_file *seq, void *v)
3024{
60f0438a 3025 if (v == &proto_list)
1da177e4
LT		 3026 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3027 "protocol",
3028 "size",
3029 "sockets",
3030 "memory",
3031 "press",
3032 "maxhdr",
3033 "slab",
3034 "module",
3035 "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3036 else
60f0438a 3037 proto_seq_printf(seq, list_entry(v, struct proto, node));
1da177e4
LT		 3038 return 0;
3039}
3040
f690808e 3041static const struct seq_operations proto_seq_ops = {
1da177e4
LT		 3042 .start = proto_seq_start,
3043 .next = proto_seq_next,
3044 .stop = proto_seq_stop,
3045 .show = proto_seq_show,
3046};
3047
3048static int proto_seq_open(struct inode *inode, struct file *file)
3049{
14e943db
ED		 3050 return seq_open_net(inode, file, &proto_seq_ops,
3051 sizeof(struct seq_net_private));
1da177e4
LT		 3052}
3053
9a32144e 3054static const struct file_operations proto_seq_fops = {
1da177e4
LT		 3055 .owner = THIS_MODULE,
3056 .open = proto_seq_open,
3057 .read = seq_read,
3058 .llseek = seq_lseek,
14e943db
ED		 3059 .release = seq_release_net,
3060};
3061
3062static __net_init int proto_init_net(struct net *net)
3063{
d4beaa66 3064 if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
14e943db
ED		 3065 return -ENOMEM;
3066
3067 return 0;
3068}
3069
3070static __net_exit void proto_exit_net(struct net *net)
3071{
ece31ffd 3072 remove_proc_entry("protocols", net->proc_net);
14e943db
ED		 3073}
3074
3075
3076static __net_initdata struct pernet_operations proto_net_ops = {
3077 .init = proto_init_net,
3078 .exit = proto_exit_net,
1da177e4
LT		 3079};
3080
3081static int __init proto_init(void)
3082{
14e943db 3083 return register_pernet_subsys(&proto_net_ops);
1da177e4
LT		 3084}
3085
3086subsys_initcall(proto_init);
3087
3088#endif /* PROC_FS */
Linux 6.x block layer and io_uring tree(s)