xref: /openbmc/linux/net/core/sock.c (revision 9dae47aba0a055f761176d9297371d5bb24289ec)
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  *
10  * Authors:	Ross Biro
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
35  *					code. The ACK stuff can wait and needs major
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 
92 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
93 
94 #include <linux/capability.h>
95 #include <linux/errno.h>
96 #include <linux/errqueue.h>
97 #include <linux/types.h>
98 #include <linux/socket.h>
99 #include <linux/in.h>
100 #include <linux/kernel.h>
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/sched/mm.h>
106 #include <linux/timer.h>
107 #include <linux/string.h>
108 #include <linux/sockios.h>
109 #include <linux/net.h>
110 #include <linux/mm.h>
111 #include <linux/slab.h>
112 #include <linux/interrupt.h>
113 #include <linux/poll.h>
114 #include <linux/tcp.h>
115 #include <linux/init.h>
116 #include <linux/highmem.h>
117 #include <linux/user_namespace.h>
118 #include <linux/static_key.h>
119 #include <linux/memcontrol.h>
120 #include <linux/prefetch.h>
121 
122 #include <linux/uaccess.h>
123 
124 #include <linux/netdevice.h>
125 #include <net/protocol.h>
126 #include <linux/skbuff.h>
127 #include <net/net_namespace.h>
128 #include <net/request_sock.h>
129 #include <net/sock.h>
130 #include <linux/net_tstamp.h>
131 #include <net/xfrm.h>
132 #include <linux/ipsec.h>
133 #include <net/cls_cgroup.h>
134 #include <net/netprio_cgroup.h>
135 #include <linux/sock_diag.h>
136 
137 #include <linux/filter.h>
138 #include <net/sock_reuseport.h>
139 
140 #include <trace/events/sock.h>
141 
142 #include <net/tcp.h>
143 #include <net/busy_poll.h>
144 
145 static DEFINE_MUTEX(proto_list_mutex);
146 static LIST_HEAD(proto_list);
147 
148 static void sock_inuse_add(struct net *net, int val);
149 
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  */
160 bool 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 }
166 EXPORT_SYMBOL(sk_ns_capable);
167 
168 /**
169  * sk_capable - Socket global capability test
170  * @sk: Socket to use a capability on or through
171  * @cap: The global capability to use
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  */
177 bool sk_capable(const struct sock *sk, int cap)
178 {
179 	return sk_ns_capable(sk, &init_user_ns, cap);
180 }
181 EXPORT_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  *
188  * Test to see if the opener of the socket had when the socket was created
189  * and the current process has the capability @cap over the network namespace
190  * the socket is a member of.
191  */
192 bool sk_net_capable(const struct sock *sk, int cap)
193 {
194 	return sk_ns_capable(sk, sock_net(sk)->user_ns, cap);
195 }
196 EXPORT_SYMBOL(sk_net_capable);
197 
198 /*
199  * Each address family might have different locking rules, so we have
200  * one slock key per address family and separate keys for internal and
201  * userspace sockets.
202  */
203 static struct lock_class_key af_family_keys[AF_MAX];
204 static struct lock_class_key af_family_kern_keys[AF_MAX];
205 static struct lock_class_key af_family_slock_keys[AF_MAX];
206 static struct lock_class_key af_family_kern_slock_keys[AF_MAX];
207 
208 /*
209  * Make lock validator output more readable. (we pre-construct these
210  * strings build-time, so that runtime initialization of socket
211  * locks is fast):
212  */
213 
214 #define _sock_locks(x)						  \
215   x "AF_UNSPEC",	x "AF_UNIX"     ,	x "AF_INET"     , \
216   x "AF_AX25"  ,	x "AF_IPX"      ,	x "AF_APPLETALK", \
217   x "AF_NETROM",	x "AF_BRIDGE"   ,	x "AF_ATMPVC"   , \
218   x "AF_X25"   ,	x "AF_INET6"    ,	x "AF_ROSE"     , \
219   x "AF_DECnet",	x "AF_NETBEUI"  ,	x "AF_SECURITY" , \
220   x "AF_KEY"   ,	x "AF_NETLINK"  ,	x "AF_PACKET"   , \
221   x "AF_ASH"   ,	x "AF_ECONET"   ,	x "AF_ATMSVC"   , \
222   x "AF_RDS"   ,	x "AF_SNA"      ,	x "AF_IRDA"     , \
223   x "AF_PPPOX" ,	x "AF_WANPIPE"  ,	x "AF_LLC"      , \
224   x "27"       ,	x "28"          ,	x "AF_CAN"      , \
225   x "AF_TIPC"  ,	x "AF_BLUETOOTH",	x "IUCV"        , \
226   x "AF_RXRPC" ,	x "AF_ISDN"     ,	x "AF_PHONET"   , \
227   x "AF_IEEE802154",	x "AF_CAIF"	,	x "AF_ALG"      , \
228   x "AF_NFC"   ,	x "AF_VSOCK"    ,	x "AF_KCM"      , \
229   x "AF_QIPCRTR",	x "AF_SMC"	,	x "AF_MAX"
230 
231 static const char *const af_family_key_strings[AF_MAX+1] = {
232 	_sock_locks("sk_lock-")
233 };
234 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
235 	_sock_locks("slock-")
236 };
237 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
238 	_sock_locks("clock-")
239 };
240 
241 static const char *const af_family_kern_key_strings[AF_MAX+1] = {
242 	_sock_locks("k-sk_lock-")
243 };
244 static const char *const af_family_kern_slock_key_strings[AF_MAX+1] = {
245 	_sock_locks("k-slock-")
246 };
247 static const char *const af_family_kern_clock_key_strings[AF_MAX+1] = {
248 	_sock_locks("k-clock-")
249 };
250 static const char *const af_family_rlock_key_strings[AF_MAX+1] = {
251   "rlock-AF_UNSPEC", "rlock-AF_UNIX"     , "rlock-AF_INET"     ,
252   "rlock-AF_AX25"  , "rlock-AF_IPX"      , "rlock-AF_APPLETALK",
253   "rlock-AF_NETROM", "rlock-AF_BRIDGE"   , "rlock-AF_ATMPVC"   ,
254   "rlock-AF_X25"   , "rlock-AF_INET6"    , "rlock-AF_ROSE"     ,
255   "rlock-AF_DECnet", "rlock-AF_NETBEUI"  , "rlock-AF_SECURITY" ,
256   "rlock-AF_KEY"   , "rlock-AF_NETLINK"  , "rlock-AF_PACKET"   ,
257   "rlock-AF_ASH"   , "rlock-AF_ECONET"   , "rlock-AF_ATMSVC"   ,
258   "rlock-AF_RDS"   , "rlock-AF_SNA"      , "rlock-AF_IRDA"     ,
259   "rlock-AF_PPPOX" , "rlock-AF_WANPIPE"  , "rlock-AF_LLC"      ,
260   "rlock-27"       , "rlock-28"          , "rlock-AF_CAN"      ,
261   "rlock-AF_TIPC"  , "rlock-AF_BLUETOOTH", "rlock-AF_IUCV"     ,
262   "rlock-AF_RXRPC" , "rlock-AF_ISDN"     , "rlock-AF_PHONET"   ,
263   "rlock-AF_IEEE802154", "rlock-AF_CAIF" , "rlock-AF_ALG"      ,
264   "rlock-AF_NFC"   , "rlock-AF_VSOCK"    , "rlock-AF_KCM"      ,
265   "rlock-AF_QIPCRTR", "rlock-AF_SMC"     , "rlock-AF_MAX"
266 };
267 static const char *const af_family_wlock_key_strings[AF_MAX+1] = {
268   "wlock-AF_UNSPEC", "wlock-AF_UNIX"     , "wlock-AF_INET"     ,
269   "wlock-AF_AX25"  , "wlock-AF_IPX"      , "wlock-AF_APPLETALK",
270   "wlock-AF_NETROM", "wlock-AF_BRIDGE"   , "wlock-AF_ATMPVC"   ,
271   "wlock-AF_X25"   , "wlock-AF_INET6"    , "wlock-AF_ROSE"     ,
272   "wlock-AF_DECnet", "wlock-AF_NETBEUI"  , "wlock-AF_SECURITY" ,
273   "wlock-AF_KEY"   , "wlock-AF_NETLINK"  , "wlock-AF_PACKET"   ,
274   "wlock-AF_ASH"   , "wlock-AF_ECONET"   , "wlock-AF_ATMSVC"   ,
275   "wlock-AF_RDS"   , "wlock-AF_SNA"      , "wlock-AF_IRDA"     ,
276   "wlock-AF_PPPOX" , "wlock-AF_WANPIPE"  , "wlock-AF_LLC"      ,
277   "wlock-27"       , "wlock-28"          , "wlock-AF_CAN"      ,
278   "wlock-AF_TIPC"  , "wlock-AF_BLUETOOTH", "wlock-AF_IUCV"     ,
279   "wlock-AF_RXRPC" , "wlock-AF_ISDN"     , "wlock-AF_PHONET"   ,
280   "wlock-AF_IEEE802154", "wlock-AF_CAIF" , "wlock-AF_ALG"      ,
281   "wlock-AF_NFC"   , "wlock-AF_VSOCK"    , "wlock-AF_KCM"      ,
282   "wlock-AF_QIPCRTR", "wlock-AF_SMC"     , "wlock-AF_MAX"
283 };
284 static const char *const af_family_elock_key_strings[AF_MAX+1] = {
285   "elock-AF_UNSPEC", "elock-AF_UNIX"     , "elock-AF_INET"     ,
286   "elock-AF_AX25"  , "elock-AF_IPX"      , "elock-AF_APPLETALK",
287   "elock-AF_NETROM", "elock-AF_BRIDGE"   , "elock-AF_ATMPVC"   ,
288   "elock-AF_X25"   , "elock-AF_INET6"    , "elock-AF_ROSE"     ,
289   "elock-AF_DECnet", "elock-AF_NETBEUI"  , "elock-AF_SECURITY" ,
290   "elock-AF_KEY"   , "elock-AF_NETLINK"  , "elock-AF_PACKET"   ,
291   "elock-AF_ASH"   , "elock-AF_ECONET"   , "elock-AF_ATMSVC"   ,
292   "elock-AF_RDS"   , "elock-AF_SNA"      , "elock-AF_IRDA"     ,
293   "elock-AF_PPPOX" , "elock-AF_WANPIPE"  , "elock-AF_LLC"      ,
294   "elock-27"       , "elock-28"          , "elock-AF_CAN"      ,
295   "elock-AF_TIPC"  , "elock-AF_BLUETOOTH", "elock-AF_IUCV"     ,
296   "elock-AF_RXRPC" , "elock-AF_ISDN"     , "elock-AF_PHONET"   ,
297   "elock-AF_IEEE802154", "elock-AF_CAIF" , "elock-AF_ALG"      ,
298   "elock-AF_NFC"   , "elock-AF_VSOCK"    , "elock-AF_KCM"      ,
299   "elock-AF_QIPCRTR", "elock-AF_SMC"     , "elock-AF_MAX"
300 };
301 
302 /*
303  * sk_callback_lock and sk queues locking rules are per-address-family,
304  * so split the lock classes by using a per-AF key:
305  */
306 static struct lock_class_key af_callback_keys[AF_MAX];
307 static struct lock_class_key af_rlock_keys[AF_MAX];
308 static struct lock_class_key af_wlock_keys[AF_MAX];
309 static struct lock_class_key af_elock_keys[AF_MAX];
310 static struct lock_class_key af_kern_callback_keys[AF_MAX];
311 
312 /* Run time adjustable parameters. */
313 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
314 EXPORT_SYMBOL(sysctl_wmem_max);
315 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
316 EXPORT_SYMBOL(sysctl_rmem_max);
317 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
318 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
319 
320 /* Maximal space eaten by iovec or ancillary data plus some space */
321 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
322 EXPORT_SYMBOL(sysctl_optmem_max);
323 
324 int sysctl_tstamp_allow_data __read_mostly = 1;
325 
326 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
327 EXPORT_SYMBOL_GPL(memalloc_socks);
328 
329 /**
330  * sk_set_memalloc - sets %SOCK_MEMALLOC
331  * @sk: socket to set it on
332  *
333  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
334  * It's the responsibility of the admin to adjust min_free_kbytes
335  * to meet the requirements
336  */
337 void sk_set_memalloc(struct sock *sk)
338 {
339 	sock_set_flag(sk, SOCK_MEMALLOC);
340 	sk->sk_allocation |= __GFP_MEMALLOC;
341 	static_key_slow_inc(&memalloc_socks);
342 }
343 EXPORT_SYMBOL_GPL(sk_set_memalloc);
344 
345 void sk_clear_memalloc(struct sock *sk)
346 {
347 	sock_reset_flag(sk, SOCK_MEMALLOC);
348 	sk->sk_allocation &= ~__GFP_MEMALLOC;
349 	static_key_slow_dec(&memalloc_socks);
350 
351 	/*
352 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
353 	 * progress of swapping. SOCK_MEMALLOC may be cleared while
354 	 * it has rmem allocations due to the last swapfile being deactivated
355 	 * but there is a risk that the socket is unusable due to exceeding
356 	 * the rmem limits. Reclaim the reserves and obey rmem limits again.
357 	 */
358 	sk_mem_reclaim(sk);
359 }
360 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
361 
362 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
363 {
364 	int ret;
365 	unsigned int noreclaim_flag;
366 
367 	/* these should have been dropped before queueing */
368 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
369 
370 	noreclaim_flag = memalloc_noreclaim_save();
371 	ret = sk->sk_backlog_rcv(sk, skb);
372 	memalloc_noreclaim_restore(noreclaim_flag);
373 
374 	return ret;
375 }
376 EXPORT_SYMBOL(__sk_backlog_rcv);
377 
378 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
379 {
380 	struct timeval tv;
381 
382 	if (optlen < sizeof(tv))
383 		return -EINVAL;
384 	if (copy_from_user(&tv, optval, sizeof(tv)))
385 		return -EFAULT;
386 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
387 		return -EDOM;
388 
389 	if (tv.tv_sec < 0) {
390 		static int warned __read_mostly;
391 
392 		*timeo_p = 0;
393 		if (warned < 10 && net_ratelimit()) {
394 			warned++;
395 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
396 				__func__, current->comm, task_pid_nr(current));
397 		}
398 		return 0;
399 	}
400 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
401 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
402 		return 0;
403 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
404 		*timeo_p = tv.tv_sec * HZ + DIV_ROUND_UP(tv.tv_usec, USEC_PER_SEC / HZ);
405 	return 0;
406 }
407 
408 static void sock_warn_obsolete_bsdism(const char *name)
409 {
410 	static int warned;
411 	static char warncomm[TASK_COMM_LEN];
412 	if (strcmp(warncomm, current->comm) && warned < 5) {
413 		strcpy(warncomm,  current->comm);
414 		pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
415 			warncomm, name);
416 		warned++;
417 	}
418 }
419 
420 static bool sock_needs_netstamp(const struct sock *sk)
421 {
422 	switch (sk->sk_family) {
423 	case AF_UNSPEC:
424 	case AF_UNIX:
425 		return false;
426 	default:
427 		return true;
428 	}
429 }
430 
431 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
432 {
433 	if (sk->sk_flags & flags) {
434 		sk->sk_flags &= ~flags;
435 		if (sock_needs_netstamp(sk) &&
436 		    !(sk->sk_flags & SK_FLAGS_TIMESTAMP))
437 			net_disable_timestamp();
438 	}
439 }
440 
441 
442 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
443 {
444 	unsigned long flags;
445 	struct sk_buff_head *list = &sk->sk_receive_queue;
446 
447 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
448 		atomic_inc(&sk->sk_drops);
449 		trace_sock_rcvqueue_full(sk, skb);
450 		return -ENOMEM;
451 	}
452 
453 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
454 		atomic_inc(&sk->sk_drops);
455 		return -ENOBUFS;
456 	}
457 
458 	skb->dev = NULL;
459 	skb_set_owner_r(skb, sk);
460 
461 	/* we escape from rcu protected region, make sure we dont leak
462 	 * a norefcounted dst
463 	 */
464 	skb_dst_force(skb);
465 
466 	spin_lock_irqsave(&list->lock, flags);
467 	sock_skb_set_dropcount(sk, skb);
468 	__skb_queue_tail(list, skb);
469 	spin_unlock_irqrestore(&list->lock, flags);
470 
471 	if (!sock_flag(sk, SOCK_DEAD))
472 		sk->sk_data_ready(sk);
473 	return 0;
474 }
475 EXPORT_SYMBOL(__sock_queue_rcv_skb);
476 
477 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
478 {
479 	int err;
480 
481 	err = sk_filter(sk, skb);
482 	if (err)
483 		return err;
484 
485 	return __sock_queue_rcv_skb(sk, skb);
486 }
487 EXPORT_SYMBOL(sock_queue_rcv_skb);
488 
489 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
490 		     const int nested, unsigned int trim_cap, bool refcounted)
491 {
492 	int rc = NET_RX_SUCCESS;
493 
494 	if (sk_filter_trim_cap(sk, skb, trim_cap))
495 		goto discard_and_relse;
496 
497 	skb->dev = NULL;
498 
499 	if (sk_rcvqueues_full(sk, sk->sk_rcvbuf)) {
500 		atomic_inc(&sk->sk_drops);
501 		goto discard_and_relse;
502 	}
503 	if (nested)
504 		bh_lock_sock_nested(sk);
505 	else
506 		bh_lock_sock(sk);
507 	if (!sock_owned_by_user(sk)) {
508 		/*
509 		 * trylock + unlock semantics:
510 		 */
511 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
512 
513 		rc = sk_backlog_rcv(sk, skb);
514 
515 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
516 	} else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
517 		bh_unlock_sock(sk);
518 		atomic_inc(&sk->sk_drops);
519 		goto discard_and_relse;
520 	}
521 
522 	bh_unlock_sock(sk);
523 out:
524 	if (refcounted)
525 		sock_put(sk);
526 	return rc;
527 discard_and_relse:
528 	kfree_skb(skb);
529 	goto out;
530 }
531 EXPORT_SYMBOL(__sk_receive_skb);
532 
533 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
534 {
535 	struct dst_entry *dst = __sk_dst_get(sk);
536 
537 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
538 		sk_tx_queue_clear(sk);
539 		sk->sk_dst_pending_confirm = 0;
540 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
541 		dst_release(dst);
542 		return NULL;
543 	}
544 
545 	return dst;
546 }
547 EXPORT_SYMBOL(__sk_dst_check);
548 
549 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
550 {
551 	struct dst_entry *dst = sk_dst_get(sk);
552 
553 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
554 		sk_dst_reset(sk);
555 		dst_release(dst);
556 		return NULL;
557 	}
558 
559 	return dst;
560 }
561 EXPORT_SYMBOL(sk_dst_check);
562 
563 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
564 				int optlen)
565 {
566 	int ret = -ENOPROTOOPT;
567 #ifdef CONFIG_NETDEVICES
568 	struct net *net = sock_net(sk);
569 	char devname[IFNAMSIZ];
570 	int index;
571 
572 	/* Sorry... */
573 	ret = -EPERM;
574 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
575 		goto out;
576 
577 	ret = -EINVAL;
578 	if (optlen < 0)
579 		goto out;
580 
581 	/* Bind this socket to a particular device like "eth0",
582 	 * as specified in the passed interface name. If the
583 	 * name is "" or the option length is zero the socket
584 	 * is not bound.
585 	 */
586 	if (optlen > IFNAMSIZ - 1)
587 		optlen = IFNAMSIZ - 1;
588 	memset(devname, 0, sizeof(devname));
589 
590 	ret = -EFAULT;
591 	if (copy_from_user(devname, optval, optlen))
592 		goto out;
593 
594 	index = 0;
595 	if (devname[0] != '\0') {
596 		struct net_device *dev;
597 
598 		rcu_read_lock();
599 		dev = dev_get_by_name_rcu(net, devname);
600 		if (dev)
601 			index = dev->ifindex;
602 		rcu_read_unlock();
603 		ret = -ENODEV;
604 		if (!dev)
605 			goto out;
606 	}
607 
608 	lock_sock(sk);
609 	sk->sk_bound_dev_if = index;
610 	sk_dst_reset(sk);
611 	release_sock(sk);
612 
613 	ret = 0;
614 
615 out:
616 #endif
617 
618 	return ret;
619 }
620 
621 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
622 				int __user *optlen, int len)
623 {
624 	int ret = -ENOPROTOOPT;
625 #ifdef CONFIG_NETDEVICES
626 	struct net *net = sock_net(sk);
627 	char devname[IFNAMSIZ];
628 
629 	if (sk->sk_bound_dev_if == 0) {
630 		len = 0;
631 		goto zero;
632 	}
633 
634 	ret = -EINVAL;
635 	if (len < IFNAMSIZ)
636 		goto out;
637 
638 	ret = netdev_get_name(net, devname, sk->sk_bound_dev_if);
639 	if (ret)
640 		goto out;
641 
642 	len = strlen(devname) + 1;
643 
644 	ret = -EFAULT;
645 	if (copy_to_user(optval, devname, len))
646 		goto out;
647 
648 zero:
649 	ret = -EFAULT;
650 	if (put_user(len, optlen))
651 		goto out;
652 
653 	ret = 0;
654 
655 out:
656 #endif
657 
658 	return ret;
659 }
660 
661 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
662 {
663 	if (valbool)
664 		sock_set_flag(sk, bit);
665 	else
666 		sock_reset_flag(sk, bit);
667 }
668 
669 bool sk_mc_loop(struct sock *sk)
670 {
671 	if (dev_recursion_level())
672 		return false;
673 	if (!sk)
674 		return true;
675 	switch (sk->sk_family) {
676 	case AF_INET:
677 		return inet_sk(sk)->mc_loop;
678 #if IS_ENABLED(CONFIG_IPV6)
679 	case AF_INET6:
680 		return inet6_sk(sk)->mc_loop;
681 #endif
682 	}
683 	WARN_ON(1);
684 	return true;
685 }
686 EXPORT_SYMBOL(sk_mc_loop);
687 
688 /*
689  *	This is meant for all protocols to use and covers goings on
690  *	at the socket level. Everything here is generic.
691  */
692 
693 int sock_setsockopt(struct socket *sock, int level, int optname,
694 		    char __user *optval, unsigned int optlen)
695 {
696 	struct sock *sk = sock->sk;
697 	int val;
698 	int valbool;
699 	struct linger ling;
700 	int ret = 0;
701 
702 	/*
703 	 *	Options without arguments
704 	 */
705 
706 	if (optname == SO_BINDTODEVICE)
707 		return sock_setbindtodevice(sk, optval, optlen);
708 
709 	if (optlen < sizeof(int))
710 		return -EINVAL;
711 
712 	if (get_user(val, (int __user *)optval))
713 		return -EFAULT;
714 
715 	valbool = val ? 1 : 0;
716 
717 	lock_sock(sk);
718 
719 	switch (optname) {
720 	case SO_DEBUG:
721 		if (val && !capable(CAP_NET_ADMIN))
722 			ret = -EACCES;
723 		else
724 			sock_valbool_flag(sk, SOCK_DBG, valbool);
725 		break;
726 	case SO_REUSEADDR:
727 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
728 		break;
729 	case SO_REUSEPORT:
730 		sk->sk_reuseport = valbool;
731 		break;
732 	case SO_TYPE:
733 	case SO_PROTOCOL:
734 	case SO_DOMAIN:
735 	case SO_ERROR:
736 		ret = -ENOPROTOOPT;
737 		break;
738 	case SO_DONTROUTE:
739 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
740 		break;
741 	case SO_BROADCAST:
742 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
743 		break;
744 	case SO_SNDBUF:
745 		/* Don't error on this BSD doesn't and if you think
746 		 * about it this is right. Otherwise apps have to
747 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
748 		 * are treated in BSD as hints
749 		 */
750 		val = min_t(u32, val, sysctl_wmem_max);
751 set_sndbuf:
752 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
753 		sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
754 		/* Wake up sending tasks if we upped the value. */
755 		sk->sk_write_space(sk);
756 		break;
757 
758 	case SO_SNDBUFFORCE:
759 		if (!capable(CAP_NET_ADMIN)) {
760 			ret = -EPERM;
761 			break;
762 		}
763 		goto set_sndbuf;
764 
765 	case SO_RCVBUF:
766 		/* Don't error on this BSD doesn't and if you think
767 		 * about it this is right. Otherwise apps have to
768 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
769 		 * are treated in BSD as hints
770 		 */
771 		val = min_t(u32, val, sysctl_rmem_max);
772 set_rcvbuf:
773 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
774 		/*
775 		 * We double it on the way in to account for
776 		 * "struct sk_buff" etc. overhead.   Applications
777 		 * assume that the SO_RCVBUF setting they make will
778 		 * allow that much actual data to be received on that
779 		 * socket.
780 		 *
781 		 * Applications are unaware that "struct sk_buff" and
782 		 * other overheads allocate from the receive buffer
783 		 * during socket buffer allocation.
784 		 *
785 		 * And after considering the possible alternatives,
786 		 * returning the value we actually used in getsockopt
787 		 * is the most desirable behavior.
788 		 */
789 		sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
790 		break;
791 
792 	case SO_RCVBUFFORCE:
793 		if (!capable(CAP_NET_ADMIN)) {
794 			ret = -EPERM;
795 			break;
796 		}
797 		goto set_rcvbuf;
798 
799 	case SO_KEEPALIVE:
800 		if (sk->sk_prot->keepalive)
801 			sk->sk_prot->keepalive(sk, valbool);
802 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
803 		break;
804 
805 	case SO_OOBINLINE:
806 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
807 		break;
808 
809 	case SO_NO_CHECK:
810 		sk->sk_no_check_tx = valbool;
811 		break;
812 
813 	case SO_PRIORITY:
814 		if ((val >= 0 && val <= 6) ||
815 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
816 			sk->sk_priority = val;
817 		else
818 			ret = -EPERM;
819 		break;
820 
821 	case SO_LINGER:
822 		if (optlen < sizeof(ling)) {
823 			ret = -EINVAL;	/* 1003.1g */
824 			break;
825 		}
826 		if (copy_from_user(&ling, optval, sizeof(ling))) {
827 			ret = -EFAULT;
828 			break;
829 		}
830 		if (!ling.l_onoff)
831 			sock_reset_flag(sk, SOCK_LINGER);
832 		else {
833 #if (BITS_PER_LONG == 32)
834 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
835 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
836 			else
837 #endif
838 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
839 			sock_set_flag(sk, SOCK_LINGER);
840 		}
841 		break;
842 
843 	case SO_BSDCOMPAT:
844 		sock_warn_obsolete_bsdism("setsockopt");
845 		break;
846 
847 	case SO_PASSCRED:
848 		if (valbool)
849 			set_bit(SOCK_PASSCRED, &sock->flags);
850 		else
851 			clear_bit(SOCK_PASSCRED, &sock->flags);
852 		break;
853 
854 	case SO_TIMESTAMP:
855 	case SO_TIMESTAMPNS:
856 		if (valbool)  {
857 			if (optname == SO_TIMESTAMP)
858 				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
859 			else
860 				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
861 			sock_set_flag(sk, SOCK_RCVTSTAMP);
862 			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
863 		} else {
864 			sock_reset_flag(sk, SOCK_RCVTSTAMP);
865 			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
866 		}
867 		break;
868 
869 	case SO_TIMESTAMPING:
870 		if (val & ~SOF_TIMESTAMPING_MASK) {
871 			ret = -EINVAL;
872 			break;
873 		}
874 
875 		if (val & SOF_TIMESTAMPING_OPT_ID &&
876 		    !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID)) {
877 			if (sk->sk_protocol == IPPROTO_TCP &&
878 			    sk->sk_type == SOCK_STREAM) {
879 				if ((1 << sk->sk_state) &
880 				    (TCPF_CLOSE | TCPF_LISTEN)) {
881 					ret = -EINVAL;
882 					break;
883 				}
884 				sk->sk_tskey = tcp_sk(sk)->snd_una;
885 			} else {
886 				sk->sk_tskey = 0;
887 			}
888 		}
889 
890 		if (val & SOF_TIMESTAMPING_OPT_STATS &&
891 		    !(val & SOF_TIMESTAMPING_OPT_TSONLY)) {
892 			ret = -EINVAL;
893 			break;
894 		}
895 
896 		sk->sk_tsflags = val;
897 		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
898 			sock_enable_timestamp(sk,
899 					      SOCK_TIMESTAMPING_RX_SOFTWARE);
900 		else
901 			sock_disable_timestamp(sk,
902 					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
903 		break;
904 
905 	case SO_RCVLOWAT:
906 		if (val < 0)
907 			val = INT_MAX;
908 		sk->sk_rcvlowat = val ? : 1;
909 		break;
910 
911 	case SO_RCVTIMEO:
912 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
913 		break;
914 
915 	case SO_SNDTIMEO:
916 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
917 		break;
918 
919 	case SO_ATTACH_FILTER:
920 		ret = -EINVAL;
921 		if (optlen == sizeof(struct sock_fprog)) {
922 			struct sock_fprog fprog;
923 
924 			ret = -EFAULT;
925 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
926 				break;
927 
928 			ret = sk_attach_filter(&fprog, sk);
929 		}
930 		break;
931 
932 	case SO_ATTACH_BPF:
933 		ret = -EINVAL;
934 		if (optlen == sizeof(u32)) {
935 			u32 ufd;
936 
937 			ret = -EFAULT;
938 			if (copy_from_user(&ufd, optval, sizeof(ufd)))
939 				break;
940 
941 			ret = sk_attach_bpf(ufd, sk);
942 		}
943 		break;
944 
945 	case SO_ATTACH_REUSEPORT_CBPF:
946 		ret = -EINVAL;
947 		if (optlen == sizeof(struct sock_fprog)) {
948 			struct sock_fprog fprog;
949 
950 			ret = -EFAULT;
951 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
952 				break;
953 
954 			ret = sk_reuseport_attach_filter(&fprog, sk);
955 		}
956 		break;
957 
958 	case SO_ATTACH_REUSEPORT_EBPF:
959 		ret = -EINVAL;
960 		if (optlen == sizeof(u32)) {
961 			u32 ufd;
962 
963 			ret = -EFAULT;
964 			if (copy_from_user(&ufd, optval, sizeof(ufd)))
965 				break;
966 
967 			ret = sk_reuseport_attach_bpf(ufd, sk);
968 		}
969 		break;
970 
971 	case SO_DETACH_FILTER:
972 		ret = sk_detach_filter(sk);
973 		break;
974 
975 	case SO_LOCK_FILTER:
976 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
977 			ret = -EPERM;
978 		else
979 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
980 		break;
981 
982 	case SO_PASSSEC:
983 		if (valbool)
984 			set_bit(SOCK_PASSSEC, &sock->flags);
985 		else
986 			clear_bit(SOCK_PASSSEC, &sock->flags);
987 		break;
988 	case SO_MARK:
989 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
990 			ret = -EPERM;
991 		else
992 			sk->sk_mark = val;
993 		break;
994 
995 	case SO_RXQ_OVFL:
996 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
997 		break;
998 
999 	case SO_WIFI_STATUS:
1000 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
1001 		break;
1002 
1003 	case SO_PEEK_OFF:
1004 		if (sock->ops->set_peek_off)
1005 			ret = sock->ops->set_peek_off(sk, val);
1006 		else
1007 			ret = -EOPNOTSUPP;
1008 		break;
1009 
1010 	case SO_NOFCS:
1011 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
1012 		break;
1013 
1014 	case SO_SELECT_ERR_QUEUE:
1015 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
1016 		break;
1017 
1018 #ifdef CONFIG_NET_RX_BUSY_POLL
1019 	case SO_BUSY_POLL:
1020 		/* allow unprivileged users to decrease the value */
1021 		if ((val > sk->sk_ll_usec) && !capable(CAP_NET_ADMIN))
1022 			ret = -EPERM;
1023 		else {
1024 			if (val < 0)
1025 				ret = -EINVAL;
1026 			else
1027 				sk->sk_ll_usec = val;
1028 		}
1029 		break;
1030 #endif
1031 
1032 	case SO_MAX_PACING_RATE:
1033 		if (val != ~0U)
1034 			cmpxchg(&sk->sk_pacing_status,
1035 				SK_PACING_NONE,
1036 				SK_PACING_NEEDED);
1037 		sk->sk_max_pacing_rate = val;
1038 		sk->sk_pacing_rate = min(sk->sk_pacing_rate,
1039 					 sk->sk_max_pacing_rate);
1040 		break;
1041 
1042 	case SO_INCOMING_CPU:
1043 		sk->sk_incoming_cpu = val;
1044 		break;
1045 
1046 	case SO_CNX_ADVICE:
1047 		if (val == 1)
1048 			dst_negative_advice(sk);
1049 		break;
1050 
1051 	case SO_ZEROCOPY:
1052 		if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1053 			ret = -ENOTSUPP;
1054 		else if (sk->sk_protocol != IPPROTO_TCP)
1055 			ret = -ENOTSUPP;
1056 		else if (sk->sk_state != TCP_CLOSE)
1057 			ret = -EBUSY;
1058 		else if (val < 0 || val > 1)
1059 			ret = -EINVAL;
1060 		else
1061 			sock_valbool_flag(sk, SOCK_ZEROCOPY, valbool);
1062 		break;
1063 
1064 	default:
1065 		ret = -ENOPROTOOPT;
1066 		break;
1067 	}
1068 	release_sock(sk);
1069 	return ret;
1070 }
1071 EXPORT_SYMBOL(sock_setsockopt);
1072 
1073 
1074 static void cred_to_ucred(struct pid *pid, const struct cred *cred,
1075 			  struct ucred *ucred)
1076 {
1077 	ucred->pid = pid_vnr(pid);
1078 	ucred->uid = ucred->gid = -1;
1079 	if (cred) {
1080 		struct user_namespace *current_ns = current_user_ns();
1081 
1082 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
1083 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
1084 	}
1085 }
1086 
1087 static int groups_to_user(gid_t __user *dst, const struct group_info *src)
1088 {
1089 	struct user_namespace *user_ns = current_user_ns();
1090 	int i;
1091 
1092 	for (i = 0; i < src->ngroups; i++)
1093 		if (put_user(from_kgid_munged(user_ns, src->gid[i]), dst + i))
1094 			return -EFAULT;
1095 
1096 	return 0;
1097 }
1098 
1099 int sock_getsockopt(struct socket *sock, int level, int optname,
1100 		    char __user *optval, int __user *optlen)
1101 {
1102 	struct sock *sk = sock->sk;
1103 
1104 	union {
1105 		int val;
1106 		u64 val64;
1107 		struct linger ling;
1108 		struct timeval tm;
1109 	} v;
1110 
1111 	int lv = sizeof(int);
1112 	int len;
1113 
1114 	if (get_user(len, optlen))
1115 		return -EFAULT;
1116 	if (len < 0)
1117 		return -EINVAL;
1118 
1119 	memset(&v, 0, sizeof(v));
1120 
1121 	switch (optname) {
1122 	case SO_DEBUG:
1123 		v.val = sock_flag(sk, SOCK_DBG);
1124 		break;
1125 
1126 	case SO_DONTROUTE:
1127 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
1128 		break;
1129 
1130 	case SO_BROADCAST:
1131 		v.val = sock_flag(sk, SOCK_BROADCAST);
1132 		break;
1133 
1134 	case SO_SNDBUF:
1135 		v.val = sk->sk_sndbuf;
1136 		break;
1137 
1138 	case SO_RCVBUF:
1139 		v.val = sk->sk_rcvbuf;
1140 		break;
1141 
1142 	case SO_REUSEADDR:
1143 		v.val = sk->sk_reuse;
1144 		break;
1145 
1146 	case SO_REUSEPORT:
1147 		v.val = sk->sk_reuseport;
1148 		break;
1149 
1150 	case SO_KEEPALIVE:
1151 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
1152 		break;
1153 
1154 	case SO_TYPE:
1155 		v.val = sk->sk_type;
1156 		break;
1157 
1158 	case SO_PROTOCOL:
1159 		v.val = sk->sk_protocol;
1160 		break;
1161 
1162 	case SO_DOMAIN:
1163 		v.val = sk->sk_family;
1164 		break;
1165 
1166 	case SO_ERROR:
1167 		v.val = -sock_error(sk);
1168 		if (v.val == 0)
1169 			v.val = xchg(&sk->sk_err_soft, 0);
1170 		break;
1171 
1172 	case SO_OOBINLINE:
1173 		v.val = sock_flag(sk, SOCK_URGINLINE);
1174 		break;
1175 
1176 	case SO_NO_CHECK:
1177 		v.val = sk->sk_no_check_tx;
1178 		break;
1179 
1180 	case SO_PRIORITY:
1181 		v.val = sk->sk_priority;
1182 		break;
1183 
1184 	case SO_LINGER:
1185 		lv		= sizeof(v.ling);
1186 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1187 		v.ling.l_linger	= sk->sk_lingertime / HZ;
1188 		break;
1189 
1190 	case SO_BSDCOMPAT:
1191 		sock_warn_obsolete_bsdism("getsockopt");
1192 		break;
1193 
1194 	case SO_TIMESTAMP:
1195 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1196 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1197 		break;
1198 
1199 	case SO_TIMESTAMPNS:
1200 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1201 		break;
1202 
1203 	case SO_TIMESTAMPING:
1204 		v.val = sk->sk_tsflags;
1205 		break;
1206 
1207 	case SO_RCVTIMEO:
1208 		lv = sizeof(struct timeval);
1209 		if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1210 			v.tm.tv_sec = 0;
1211 			v.tm.tv_usec = 0;
1212 		} else {
1213 			v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1214 			v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * USEC_PER_SEC) / HZ;
1215 		}
1216 		break;
1217 
1218 	case SO_SNDTIMEO:
1219 		lv = sizeof(struct timeval);
1220 		if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1221 			v.tm.tv_sec = 0;
1222 			v.tm.tv_usec = 0;
1223 		} else {
1224 			v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1225 			v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * USEC_PER_SEC) / HZ;
1226 		}
1227 		break;
1228 
1229 	case SO_RCVLOWAT:
1230 		v.val = sk->sk_rcvlowat;
1231 		break;
1232 
1233 	case SO_SNDLOWAT:
1234 		v.val = 1;
1235 		break;
1236 
1237 	case SO_PASSCRED:
1238 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1239 		break;
1240 
1241 	case SO_PEERCRED:
1242 	{
1243 		struct ucred peercred;
1244 		if (len > sizeof(peercred))
1245 			len = sizeof(peercred);
1246 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1247 		if (copy_to_user(optval, &peercred, len))
1248 			return -EFAULT;
1249 		goto lenout;
1250 	}
1251 
1252 	case SO_PEERGROUPS:
1253 	{
1254 		int ret, n;
1255 
1256 		if (!sk->sk_peer_cred)
1257 			return -ENODATA;
1258 
1259 		n = sk->sk_peer_cred->group_info->ngroups;
1260 		if (len < n * sizeof(gid_t)) {
1261 			len = n * sizeof(gid_t);
1262 			return put_user(len, optlen) ? -EFAULT : -ERANGE;
1263 		}
1264 		len = n * sizeof(gid_t);
1265 
1266 		ret = groups_to_user((gid_t __user *)optval,
1267 				     sk->sk_peer_cred->group_info);
1268 		if (ret)
1269 			return ret;
1270 		goto lenout;
1271 	}
1272 
1273 	case SO_PEERNAME:
1274 	{
1275 		char address[128];
1276 
1277 		if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1278 			return -ENOTCONN;
1279 		if (lv < len)
1280 			return -EINVAL;
1281 		if (copy_to_user(optval, address, len))
1282 			return -EFAULT;
1283 		goto lenout;
1284 	}
1285 
1286 	/* Dubious BSD thing... Probably nobody even uses it, but
1287 	 * the UNIX standard wants it for whatever reason... -DaveM
1288 	 */
1289 	case SO_ACCEPTCONN:
1290 		v.val = sk->sk_state == TCP_LISTEN;
1291 		break;
1292 
1293 	case SO_PASSSEC:
1294 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1295 		break;
1296 
1297 	case SO_PEERSEC:
1298 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1299 
1300 	case SO_MARK:
1301 		v.val = sk->sk_mark;
1302 		break;
1303 
1304 	case SO_RXQ_OVFL:
1305 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1306 		break;
1307 
1308 	case SO_WIFI_STATUS:
1309 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1310 		break;
1311 
1312 	case SO_PEEK_OFF:
1313 		if (!sock->ops->set_peek_off)
1314 			return -EOPNOTSUPP;
1315 
1316 		v.val = sk->sk_peek_off;
1317 		break;
1318 	case SO_NOFCS:
1319 		v.val = sock_flag(sk, SOCK_NOFCS);
1320 		break;
1321 
1322 	case SO_BINDTODEVICE:
1323 		return sock_getbindtodevice(sk, optval, optlen, len);
1324 
1325 	case SO_GET_FILTER:
1326 		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1327 		if (len < 0)
1328 			return len;
1329 
1330 		goto lenout;
1331 
1332 	case SO_LOCK_FILTER:
1333 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1334 		break;
1335 
1336 	case SO_BPF_EXTENSIONS:
1337 		v.val = bpf_tell_extensions();
1338 		break;
1339 
1340 	case SO_SELECT_ERR_QUEUE:
1341 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1342 		break;
1343 
1344 #ifdef CONFIG_NET_RX_BUSY_POLL
1345 	case SO_BUSY_POLL:
1346 		v.val = sk->sk_ll_usec;
1347 		break;
1348 #endif
1349 
1350 	case SO_MAX_PACING_RATE:
1351 		v.val = sk->sk_max_pacing_rate;
1352 		break;
1353 
1354 	case SO_INCOMING_CPU:
1355 		v.val = sk->sk_incoming_cpu;
1356 		break;
1357 
1358 	case SO_MEMINFO:
1359 	{
1360 		u32 meminfo[SK_MEMINFO_VARS];
1361 
1362 		if (get_user(len, optlen))
1363 			return -EFAULT;
1364 
1365 		sk_get_meminfo(sk, meminfo);
1366 
1367 		len = min_t(unsigned int, len, sizeof(meminfo));
1368 		if (copy_to_user(optval, &meminfo, len))
1369 			return -EFAULT;
1370 
1371 		goto lenout;
1372 	}
1373 
1374 #ifdef CONFIG_NET_RX_BUSY_POLL
1375 	case SO_INCOMING_NAPI_ID:
1376 		v.val = READ_ONCE(sk->sk_napi_id);
1377 
1378 		/* aggregate non-NAPI IDs down to 0 */
1379 		if (v.val < MIN_NAPI_ID)
1380 			v.val = 0;
1381 
1382 		break;
1383 #endif
1384 
1385 	case SO_COOKIE:
1386 		lv = sizeof(u64);
1387 		if (len < lv)
1388 			return -EINVAL;
1389 		v.val64 = sock_gen_cookie(sk);
1390 		break;
1391 
1392 	case SO_ZEROCOPY:
1393 		v.val = sock_flag(sk, SOCK_ZEROCOPY);
1394 		break;
1395 
1396 	default:
1397 		/* We implement the SO_SNDLOWAT etc to not be settable
1398 		 * (1003.1g 7).
1399 		 */
1400 		return -ENOPROTOOPT;
1401 	}
1402 
1403 	if (len > lv)
1404 		len = lv;
1405 	if (copy_to_user(optval, &v, len))
1406 		return -EFAULT;
1407 lenout:
1408 	if (put_user(len, optlen))
1409 		return -EFAULT;
1410 	return 0;
1411 }
1412 
1413 /*
1414  * Initialize an sk_lock.
1415  *
1416  * (We also register the sk_lock with the lock validator.)
1417  */
1418 static inline void sock_lock_init(struct sock *sk)
1419 {
1420 	if (sk->sk_kern_sock)
1421 		sock_lock_init_class_and_name(
1422 			sk,
1423 			af_family_kern_slock_key_strings[sk->sk_family],
1424 			af_family_kern_slock_keys + sk->sk_family,
1425 			af_family_kern_key_strings[sk->sk_family],
1426 			af_family_kern_keys + sk->sk_family);
1427 	else
1428 		sock_lock_init_class_and_name(
1429 			sk,
1430 			af_family_slock_key_strings[sk->sk_family],
1431 			af_family_slock_keys + sk->sk_family,
1432 			af_family_key_strings[sk->sk_family],
1433 			af_family_keys + sk->sk_family);
1434 }
1435 
1436 /*
1437  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1438  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1439  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1440  */
1441 static void sock_copy(struct sock *nsk, const struct sock *osk)
1442 {
1443 #ifdef CONFIG_SECURITY_NETWORK
1444 	void *sptr = nsk->sk_security;
1445 #endif
1446 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1447 
1448 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1449 	       osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1450 
1451 #ifdef CONFIG_SECURITY_NETWORK
1452 	nsk->sk_security = sptr;
1453 	security_sk_clone(osk, nsk);
1454 #endif
1455 }
1456 
1457 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1458 		int family)
1459 {
1460 	struct sock *sk;
1461 	struct kmem_cache *slab;
1462 
1463 	slab = prot->slab;
1464 	if (slab != NULL) {
1465 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1466 		if (!sk)
1467 			return sk;
1468 		if (priority & __GFP_ZERO)
1469 			sk_prot_clear_nulls(sk, prot->obj_size);
1470 	} else
1471 		sk = kmalloc(prot->obj_size, priority);
1472 
1473 	if (sk != NULL) {
1474 		if (security_sk_alloc(sk, family, priority))
1475 			goto out_free;
1476 
1477 		if (!try_module_get(prot->owner))
1478 			goto out_free_sec;
1479 		sk_tx_queue_clear(sk);
1480 	}
1481 
1482 	return sk;
1483 
1484 out_free_sec:
1485 	security_sk_free(sk);
1486 out_free:
1487 	if (slab != NULL)
1488 		kmem_cache_free(slab, sk);
1489 	else
1490 		kfree(sk);
1491 	return NULL;
1492 }
1493 
1494 static void sk_prot_free(struct proto *prot, struct sock *sk)
1495 {
1496 	struct kmem_cache *slab;
1497 	struct module *owner;
1498 
1499 	owner = prot->owner;
1500 	slab = prot->slab;
1501 
1502 	cgroup_sk_free(&sk->sk_cgrp_data);
1503 	mem_cgroup_sk_free(sk);
1504 	security_sk_free(sk);
1505 	if (slab != NULL)
1506 		kmem_cache_free(slab, sk);
1507 	else
1508 		kfree(sk);
1509 	module_put(owner);
1510 }
1511 
1512 /**
1513  *	sk_alloc - All socket objects are allocated here
1514  *	@net: the applicable net namespace
1515  *	@family: protocol family
1516  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1517  *	@prot: struct proto associated with this new sock instance
1518  *	@kern: is this to be a kernel socket?
1519  */
1520 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1521 		      struct proto *prot, int kern)
1522 {
1523 	struct sock *sk;
1524 
1525 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1526 	if (sk) {
1527 		sk->sk_family = family;
1528 		/*
1529 		 * See comment in struct sock definition to understand
1530 		 * why we need sk_prot_creator -acme
1531 		 */
1532 		sk->sk_prot = sk->sk_prot_creator = prot;
1533 		sk->sk_kern_sock = kern;
1534 		sock_lock_init(sk);
1535 		sk->sk_net_refcnt = kern ? 0 : 1;
1536 		if (likely(sk->sk_net_refcnt)) {
1537 			get_net(net);
1538 			sock_inuse_add(net, 1);
1539 		}
1540 
1541 		sock_net_set(sk, net);
1542 		refcount_set(&sk->sk_wmem_alloc, 1);
1543 
1544 		mem_cgroup_sk_alloc(sk);
1545 		cgroup_sk_alloc(&sk->sk_cgrp_data);
1546 		sock_update_classid(&sk->sk_cgrp_data);
1547 		sock_update_netprioidx(&sk->sk_cgrp_data);
1548 	}
1549 
1550 	return sk;
1551 }
1552 EXPORT_SYMBOL(sk_alloc);
1553 
1554 /* Sockets having SOCK_RCU_FREE will call this function after one RCU
1555  * grace period. This is the case for UDP sockets and TCP listeners.
1556  */
1557 static void __sk_destruct(struct rcu_head *head)
1558 {
1559 	struct sock *sk = container_of(head, struct sock, sk_rcu);
1560 	struct sk_filter *filter;
1561 
1562 	if (sk->sk_destruct)
1563 		sk->sk_destruct(sk);
1564 
1565 	filter = rcu_dereference_check(sk->sk_filter,
1566 				       refcount_read(&sk->sk_wmem_alloc) == 0);
1567 	if (filter) {
1568 		sk_filter_uncharge(sk, filter);
1569 		RCU_INIT_POINTER(sk->sk_filter, NULL);
1570 	}
1571 	if (rcu_access_pointer(sk->sk_reuseport_cb))
1572 		reuseport_detach_sock(sk);
1573 
1574 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1575 
1576 	if (atomic_read(&sk->sk_omem_alloc))
1577 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1578 			 __func__, atomic_read(&sk->sk_omem_alloc));
1579 
1580 	if (sk->sk_frag.page) {
1581 		put_page(sk->sk_frag.page);
1582 		sk->sk_frag.page = NULL;
1583 	}
1584 
1585 	if (sk->sk_peer_cred)
1586 		put_cred(sk->sk_peer_cred);
1587 	put_pid(sk->sk_peer_pid);
1588 	if (likely(sk->sk_net_refcnt))
1589 		put_net(sock_net(sk));
1590 	sk_prot_free(sk->sk_prot_creator, sk);
1591 }
1592 
1593 void sk_destruct(struct sock *sk)
1594 {
1595 	if (sock_flag(sk, SOCK_RCU_FREE))
1596 		call_rcu(&sk->sk_rcu, __sk_destruct);
1597 	else
1598 		__sk_destruct(&sk->sk_rcu);
1599 }
1600 
1601 static void __sk_free(struct sock *sk)
1602 {
1603 	if (likely(sk->sk_net_refcnt))
1604 		sock_inuse_add(sock_net(sk), -1);
1605 
1606 	if (unlikely(sock_diag_has_destroy_listeners(sk) && sk->sk_net_refcnt))
1607 		sock_diag_broadcast_destroy(sk);
1608 	else
1609 		sk_destruct(sk);
1610 }
1611 
1612 void sk_free(struct sock *sk)
1613 {
1614 	/*
1615 	 * We subtract one from sk_wmem_alloc and can know if
1616 	 * some packets are still in some tx queue.
1617 	 * If not null, sock_wfree() will call __sk_free(sk) later
1618 	 */
1619 	if (refcount_dec_and_test(&sk->sk_wmem_alloc))
1620 		__sk_free(sk);
1621 }
1622 EXPORT_SYMBOL(sk_free);
1623 
1624 static void sk_init_common(struct sock *sk)
1625 {
1626 	skb_queue_head_init(&sk->sk_receive_queue);
1627 	skb_queue_head_init(&sk->sk_write_queue);
1628 	skb_queue_head_init(&sk->sk_error_queue);
1629 
1630 	rwlock_init(&sk->sk_callback_lock);
1631 	lockdep_set_class_and_name(&sk->sk_receive_queue.lock,
1632 			af_rlock_keys + sk->sk_family,
1633 			af_family_rlock_key_strings[sk->sk_family]);
1634 	lockdep_set_class_and_name(&sk->sk_write_queue.lock,
1635 			af_wlock_keys + sk->sk_family,
1636 			af_family_wlock_key_strings[sk->sk_family]);
1637 	lockdep_set_class_and_name(&sk->sk_error_queue.lock,
1638 			af_elock_keys + sk->sk_family,
1639 			af_family_elock_key_strings[sk->sk_family]);
1640 	lockdep_set_class_and_name(&sk->sk_callback_lock,
1641 			af_callback_keys + sk->sk_family,
1642 			af_family_clock_key_strings[sk->sk_family]);
1643 }
1644 
1645 /**
1646  *	sk_clone_lock - clone a socket, and lock its clone
1647  *	@sk: the socket to clone
1648  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1649  *
1650  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1651  */
1652 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1653 {
1654 	struct sock *newsk;
1655 	bool is_charged = true;
1656 
1657 	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1658 	if (newsk != NULL) {
1659 		struct sk_filter *filter;
1660 
1661 		sock_copy(newsk, sk);
1662 
1663 		newsk->sk_prot_creator = sk->sk_prot;
1664 
1665 		/* SANITY */
1666 		if (likely(newsk->sk_net_refcnt))
1667 			get_net(sock_net(newsk));
1668 		sk_node_init(&newsk->sk_node);
1669 		sock_lock_init(newsk);
1670 		bh_lock_sock(newsk);
1671 		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1672 		newsk->sk_backlog.len = 0;
1673 
1674 		atomic_set(&newsk->sk_rmem_alloc, 0);
1675 		/*
1676 		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1677 		 */
1678 		refcount_set(&newsk->sk_wmem_alloc, 1);
1679 		atomic_set(&newsk->sk_omem_alloc, 0);
1680 		sk_init_common(newsk);
1681 
1682 		newsk->sk_dst_cache	= NULL;
1683 		newsk->sk_dst_pending_confirm = 0;
1684 		newsk->sk_wmem_queued	= 0;
1685 		newsk->sk_forward_alloc = 0;
1686 
1687 		/* sk->sk_memcg will be populated at accept() time */
1688 		newsk->sk_memcg = NULL;
1689 
1690 		atomic_set(&newsk->sk_drops, 0);
1691 		newsk->sk_send_head	= NULL;
1692 		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1693 		atomic_set(&newsk->sk_zckey, 0);
1694 
1695 		sock_reset_flag(newsk, SOCK_DONE);
1696 		cgroup_sk_alloc(&newsk->sk_cgrp_data);
1697 
1698 		rcu_read_lock();
1699 		filter = rcu_dereference(sk->sk_filter);
1700 		if (filter != NULL)
1701 			/* though it's an empty new sock, the charging may fail
1702 			 * if sysctl_optmem_max was changed between creation of
1703 			 * original socket and cloning
1704 			 */
1705 			is_charged = sk_filter_charge(newsk, filter);
1706 		RCU_INIT_POINTER(newsk->sk_filter, filter);
1707 		rcu_read_unlock();
1708 
1709 		if (unlikely(!is_charged || xfrm_sk_clone_policy(newsk, sk))) {
1710 			/* We need to make sure that we don't uncharge the new
1711 			 * socket if we couldn't charge it in the first place
1712 			 * as otherwise we uncharge the parent's filter.
1713 			 */
1714 			if (!is_charged)
1715 				RCU_INIT_POINTER(newsk->sk_filter, NULL);
1716 			sk_free_unlock_clone(newsk);
1717 			newsk = NULL;
1718 			goto out;
1719 		}
1720 		RCU_INIT_POINTER(newsk->sk_reuseport_cb, NULL);
1721 
1722 		newsk->sk_err	   = 0;
1723 		newsk->sk_err_soft = 0;
1724 		newsk->sk_priority = 0;
1725 		newsk->sk_incoming_cpu = raw_smp_processor_id();
1726 		atomic64_set(&newsk->sk_cookie, 0);
1727 		if (likely(newsk->sk_net_refcnt))
1728 			sock_inuse_add(sock_net(newsk), 1);
1729 
1730 		/*
1731 		 * Before updating sk_refcnt, we must commit prior changes to memory
1732 		 * (Documentation/RCU/rculist_nulls.txt for details)
1733 		 */
1734 		smp_wmb();
1735 		refcount_set(&newsk->sk_refcnt, 2);
1736 
1737 		/*
1738 		 * Increment the counter in the same struct proto as the master
1739 		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1740 		 * is the same as sk->sk_prot->socks, as this field was copied
1741 		 * with memcpy).
1742 		 *
1743 		 * This _changes_ the previous behaviour, where
1744 		 * tcp_create_openreq_child always was incrementing the
1745 		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1746 		 * to be taken into account in all callers. -acme
1747 		 */
1748 		sk_refcnt_debug_inc(newsk);
1749 		sk_set_socket(newsk, NULL);
1750 		newsk->sk_wq = NULL;
1751 
1752 		if (newsk->sk_prot->sockets_allocated)
1753 			sk_sockets_allocated_inc(newsk);
1754 
1755 		if (sock_needs_netstamp(sk) &&
1756 		    newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1757 			net_enable_timestamp();
1758 	}
1759 out:
1760 	return newsk;
1761 }
1762 EXPORT_SYMBOL_GPL(sk_clone_lock);
1763 
1764 void sk_free_unlock_clone(struct sock *sk)
1765 {
1766 	/* It is still raw copy of parent, so invalidate
1767 	 * destructor and make plain sk_free() */
1768 	sk->sk_destruct = NULL;
1769 	bh_unlock_sock(sk);
1770 	sk_free(sk);
1771 }
1772 EXPORT_SYMBOL_GPL(sk_free_unlock_clone);
1773 
1774 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1775 {
1776 	u32 max_segs = 1;
1777 
1778 	sk_dst_set(sk, dst);
1779 	sk->sk_route_caps = dst->dev->features;
1780 	if (sk->sk_route_caps & NETIF_F_GSO)
1781 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1782 	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1783 	if (sk_can_gso(sk)) {
1784 		if (dst->header_len && !xfrm_dst_offload_ok(dst)) {
1785 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1786 		} else {
1787 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1788 			sk->sk_gso_max_size = dst->dev->gso_max_size;
1789 			max_segs = max_t(u32, dst->dev->gso_max_segs, 1);
1790 		}
1791 	}
1792 	sk->sk_gso_max_segs = max_segs;
1793 }
1794 EXPORT_SYMBOL_GPL(sk_setup_caps);
1795 
1796 /*
1797  *	Simple resource managers for sockets.
1798  */
1799 
1800 
1801 /*
1802  * Write buffer destructor automatically called from kfree_skb.
1803  */
1804 void sock_wfree(struct sk_buff *skb)
1805 {
1806 	struct sock *sk = skb->sk;
1807 	unsigned int len = skb->truesize;
1808 
1809 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1810 		/*
1811 		 * Keep a reference on sk_wmem_alloc, this will be released
1812 		 * after sk_write_space() call
1813 		 */
1814 		WARN_ON(refcount_sub_and_test(len - 1, &sk->sk_wmem_alloc));
1815 		sk->sk_write_space(sk);
1816 		len = 1;
1817 	}
1818 	/*
1819 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1820 	 * could not do because of in-flight packets
1821 	 */
1822 	if (refcount_sub_and_test(len, &sk->sk_wmem_alloc))
1823 		__sk_free(sk);
1824 }
1825 EXPORT_SYMBOL(sock_wfree);
1826 
1827 /* This variant of sock_wfree() is used by TCP,
1828  * since it sets SOCK_USE_WRITE_QUEUE.
1829  */
1830 void __sock_wfree(struct sk_buff *skb)
1831 {
1832 	struct sock *sk = skb->sk;
1833 
1834 	if (refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc))
1835 		__sk_free(sk);
1836 }
1837 
1838 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1839 {
1840 	skb_orphan(skb);
1841 	skb->sk = sk;
1842 #ifdef CONFIG_INET
1843 	if (unlikely(!sk_fullsock(sk))) {
1844 		skb->destructor = sock_edemux;
1845 		sock_hold(sk);
1846 		return;
1847 	}
1848 #endif
1849 	skb->destructor = sock_wfree;
1850 	skb_set_hash_from_sk(skb, sk);
1851 	/*
1852 	 * We used to take a refcount on sk, but following operation
1853 	 * is enough to guarantee sk_free() wont free this sock until
1854 	 * all in-flight packets are completed
1855 	 */
1856 	refcount_add(skb->truesize, &sk->sk_wmem_alloc);
1857 }
1858 EXPORT_SYMBOL(skb_set_owner_w);
1859 
1860 /* This helper is used by netem, as it can hold packets in its
1861  * delay queue. We want to allow the owner socket to send more
1862  * packets, as if they were already TX completed by a typical driver.
1863  * But we also want to keep skb->sk set because some packet schedulers
1864  * rely on it (sch_fq for example).
1865  */
1866 void skb_orphan_partial(struct sk_buff *skb)
1867 {
1868 	if (skb_is_tcp_pure_ack(skb))
1869 		return;
1870 
1871 	if (skb->destructor == sock_wfree
1872 #ifdef CONFIG_INET
1873 	    || skb->destructor == tcp_wfree
1874 #endif
1875 		) {
1876 		struct sock *sk = skb->sk;
1877 
1878 		if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1879 			WARN_ON(refcount_sub_and_test(skb->truesize, &sk->sk_wmem_alloc));
1880 			skb->destructor = sock_efree;
1881 		}
1882 	} else {
1883 		skb_orphan(skb);
1884 	}
1885 }
1886 EXPORT_SYMBOL(skb_orphan_partial);
1887 
1888 /*
1889  * Read buffer destructor automatically called from kfree_skb.
1890  */
1891 void sock_rfree(struct sk_buff *skb)
1892 {
1893 	struct sock *sk = skb->sk;
1894 	unsigned int len = skb->truesize;
1895 
1896 	atomic_sub(len, &sk->sk_rmem_alloc);
1897 	sk_mem_uncharge(sk, len);
1898 }
1899 EXPORT_SYMBOL(sock_rfree);
1900 
1901 /*
1902  * Buffer destructor for skbs that are not used directly in read or write
1903  * path, e.g. for error handler skbs. Automatically called from kfree_skb.
1904  */
1905 void sock_efree(struct sk_buff *skb)
1906 {
1907 	sock_put(skb->sk);
1908 }
1909 EXPORT_SYMBOL(sock_efree);
1910 
1911 kuid_t sock_i_uid(struct sock *sk)
1912 {
1913 	kuid_t uid;
1914 
1915 	read_lock_bh(&sk->sk_callback_lock);
1916 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1917 	read_unlock_bh(&sk->sk_callback_lock);
1918 	return uid;
1919 }
1920 EXPORT_SYMBOL(sock_i_uid);
1921 
1922 unsigned long sock_i_ino(struct sock *sk)
1923 {
1924 	unsigned long ino;
1925 
1926 	read_lock_bh(&sk->sk_callback_lock);
1927 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1928 	read_unlock_bh(&sk->sk_callback_lock);
1929 	return ino;
1930 }
1931 EXPORT_SYMBOL(sock_i_ino);
1932 
1933 /*
1934  * Allocate a skb from the socket's send buffer.
1935  */
1936 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1937 			     gfp_t priority)
1938 {
1939 	if (force || refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1940 		struct sk_buff *skb = alloc_skb(size, priority);
1941 		if (skb) {
1942 			skb_set_owner_w(skb, sk);
1943 			return skb;
1944 		}
1945 	}
1946 	return NULL;
1947 }
1948 EXPORT_SYMBOL(sock_wmalloc);
1949 
1950 static void sock_ofree(struct sk_buff *skb)
1951 {
1952 	struct sock *sk = skb->sk;
1953 
1954 	atomic_sub(skb->truesize, &sk->sk_omem_alloc);
1955 }
1956 
1957 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1958 			     gfp_t priority)
1959 {
1960 	struct sk_buff *skb;
1961 
1962 	/* small safe race: SKB_TRUESIZE may differ from final skb->truesize */
1963 	if (atomic_read(&sk->sk_omem_alloc) + SKB_TRUESIZE(size) >
1964 	    sysctl_optmem_max)
1965 		return NULL;
1966 
1967 	skb = alloc_skb(size, priority);
1968 	if (!skb)
1969 		return NULL;
1970 
1971 	atomic_add(skb->truesize, &sk->sk_omem_alloc);
1972 	skb->sk = sk;
1973 	skb->destructor = sock_ofree;
1974 	return skb;
1975 }
1976 
1977 /*
1978  * Allocate a memory block from the socket's option memory buffer.
1979  */
1980 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1981 {
1982 	if ((unsigned int)size <= sysctl_optmem_max &&
1983 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1984 		void *mem;
1985 		/* First do the add, to avoid the race if kmalloc
1986 		 * might sleep.
1987 		 */
1988 		atomic_add(size, &sk->sk_omem_alloc);
1989 		mem = kmalloc(size, priority);
1990 		if (mem)
1991 			return mem;
1992 		atomic_sub(size, &sk->sk_omem_alloc);
1993 	}
1994 	return NULL;
1995 }
1996 EXPORT_SYMBOL(sock_kmalloc);
1997 
1998 /* Free an option memory block. Note, we actually want the inline
1999  * here as this allows gcc to detect the nullify and fold away the
2000  * condition entirely.
2001  */
2002 static inline void __sock_kfree_s(struct sock *sk, void *mem, int size,
2003 				  const bool nullify)
2004 {
2005 	if (WARN_ON_ONCE(!mem))
2006 		return;
2007 	if (nullify)
2008 		kzfree(mem);
2009 	else
2010 		kfree(mem);
2011 	atomic_sub(size, &sk->sk_omem_alloc);
2012 }
2013 
2014 void sock_kfree_s(struct sock *sk, void *mem, int size)
2015 {
2016 	__sock_kfree_s(sk, mem, size, false);
2017 }
2018 EXPORT_SYMBOL(sock_kfree_s);
2019 
2020 void sock_kzfree_s(struct sock *sk, void *mem, int size)
2021 {
2022 	__sock_kfree_s(sk, mem, size, true);
2023 }
2024 EXPORT_SYMBOL(sock_kzfree_s);
2025 
2026 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
2027    I think, these locks should be removed for datagram sockets.
2028  */
2029 static long sock_wait_for_wmem(struct sock *sk, long timeo)
2030 {
2031 	DEFINE_WAIT(wait);
2032 
2033 	sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2034 	for (;;) {
2035 		if (!timeo)
2036 			break;
2037 		if (signal_pending(current))
2038 			break;
2039 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2040 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
2041 		if (refcount_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
2042 			break;
2043 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2044 			break;
2045 		if (sk->sk_err)
2046 			break;
2047 		timeo = schedule_timeout(timeo);
2048 	}
2049 	finish_wait(sk_sleep(sk), &wait);
2050 	return timeo;
2051 }
2052 
2053 
2054 /*
2055  *	Generic send/receive buffer handlers
2056  */
2057 
2058 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
2059 				     unsigned long data_len, int noblock,
2060 				     int *errcode, int max_page_order)
2061 {
2062 	struct sk_buff *skb;
2063 	long timeo;
2064 	int err;
2065 
2066 	timeo = sock_sndtimeo(sk, noblock);
2067 	for (;;) {
2068 		err = sock_error(sk);
2069 		if (err != 0)
2070 			goto failure;
2071 
2072 		err = -EPIPE;
2073 		if (sk->sk_shutdown & SEND_SHUTDOWN)
2074 			goto failure;
2075 
2076 		if (sk_wmem_alloc_get(sk) < sk->sk_sndbuf)
2077 			break;
2078 
2079 		sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2080 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2081 		err = -EAGAIN;
2082 		if (!timeo)
2083 			goto failure;
2084 		if (signal_pending(current))
2085 			goto interrupted;
2086 		timeo = sock_wait_for_wmem(sk, timeo);
2087 	}
2088 	skb = alloc_skb_with_frags(header_len, data_len, max_page_order,
2089 				   errcode, sk->sk_allocation);
2090 	if (skb)
2091 		skb_set_owner_w(skb, sk);
2092 	return skb;
2093 
2094 interrupted:
2095 	err = sock_intr_errno(timeo);
2096 failure:
2097 	*errcode = err;
2098 	return NULL;
2099 }
2100 EXPORT_SYMBOL(sock_alloc_send_pskb);
2101 
2102 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
2103 				    int noblock, int *errcode)
2104 {
2105 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
2106 }
2107 EXPORT_SYMBOL(sock_alloc_send_skb);
2108 
2109 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
2110 		     struct sockcm_cookie *sockc)
2111 {
2112 	u32 tsflags;
2113 
2114 	switch (cmsg->cmsg_type) {
2115 	case SO_MARK:
2116 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
2117 			return -EPERM;
2118 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2119 			return -EINVAL;
2120 		sockc->mark = *(u32 *)CMSG_DATA(cmsg);
2121 		break;
2122 	case SO_TIMESTAMPING:
2123 		if (cmsg->cmsg_len != CMSG_LEN(sizeof(u32)))
2124 			return -EINVAL;
2125 
2126 		tsflags = *(u32 *)CMSG_DATA(cmsg);
2127 		if (tsflags & ~SOF_TIMESTAMPING_TX_RECORD_MASK)
2128 			return -EINVAL;
2129 
2130 		sockc->tsflags &= ~SOF_TIMESTAMPING_TX_RECORD_MASK;
2131 		sockc->tsflags |= tsflags;
2132 		break;
2133 	/* SCM_RIGHTS and SCM_CREDENTIALS are semantically in SOL_UNIX. */
2134 	case SCM_RIGHTS:
2135 	case SCM_CREDENTIALS:
2136 		break;
2137 	default:
2138 		return -EINVAL;
2139 	}
2140 	return 0;
2141 }
2142 EXPORT_SYMBOL(__sock_cmsg_send);
2143 
2144 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
2145 		   struct sockcm_cookie *sockc)
2146 {
2147 	struct cmsghdr *cmsg;
2148 	int ret;
2149 
2150 	for_each_cmsghdr(cmsg, msg) {
2151 		if (!CMSG_OK(msg, cmsg))
2152 			return -EINVAL;
2153 		if (cmsg->cmsg_level != SOL_SOCKET)
2154 			continue;
2155 		ret = __sock_cmsg_send(sk, msg, cmsg, sockc);
2156 		if (ret)
2157 			return ret;
2158 	}
2159 	return 0;
2160 }
2161 EXPORT_SYMBOL(sock_cmsg_send);
2162 
2163 static void sk_enter_memory_pressure(struct sock *sk)
2164 {
2165 	if (!sk->sk_prot->enter_memory_pressure)
2166 		return;
2167 
2168 	sk->sk_prot->enter_memory_pressure(sk);
2169 }
2170 
2171 static void sk_leave_memory_pressure(struct sock *sk)
2172 {
2173 	if (sk->sk_prot->leave_memory_pressure) {
2174 		sk->sk_prot->leave_memory_pressure(sk);
2175 	} else {
2176 		unsigned long *memory_pressure = sk->sk_prot->memory_pressure;
2177 
2178 		if (memory_pressure && *memory_pressure)
2179 			*memory_pressure = 0;
2180 	}
2181 }
2182 
2183 /* On 32bit arches, an skb frag is limited to 2^15 */
2184 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
2185 
2186 /**
2187  * skb_page_frag_refill - check that a page_frag contains enough room
2188  * @sz: minimum size of the fragment we want to get
2189  * @pfrag: pointer to page_frag
2190  * @gfp: priority for memory allocation
2191  *
2192  * Note: While this allocator tries to use high order pages, there is
2193  * no guarantee that allocations succeed. Therefore, @sz MUST be
2194  * less or equal than PAGE_SIZE.
2195  */
2196 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t gfp)
2197 {
2198 	if (pfrag->page) {
2199 		if (page_ref_count(pfrag->page) == 1) {
2200 			pfrag->offset = 0;
2201 			return true;
2202 		}
2203 		if (pfrag->offset + sz <= pfrag->size)
2204 			return true;
2205 		put_page(pfrag->page);
2206 	}
2207 
2208 	pfrag->offset = 0;
2209 	if (SKB_FRAG_PAGE_ORDER) {
2210 		/* Avoid direct reclaim but allow kswapd to wake */
2211 		pfrag->page = alloc_pages((gfp & ~__GFP_DIRECT_RECLAIM) |
2212 					  __GFP_COMP | __GFP_NOWARN |
2213 					  __GFP_NORETRY,
2214 					  SKB_FRAG_PAGE_ORDER);
2215 		if (likely(pfrag->page)) {
2216 			pfrag->size = PAGE_SIZE << SKB_FRAG_PAGE_ORDER;
2217 			return true;
2218 		}
2219 	}
2220 	pfrag->page = alloc_page(gfp);
2221 	if (likely(pfrag->page)) {
2222 		pfrag->size = PAGE_SIZE;
2223 		return true;
2224 	}
2225 	return false;
2226 }
2227 EXPORT_SYMBOL(skb_page_frag_refill);
2228 
2229 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
2230 {
2231 	if (likely(skb_page_frag_refill(32U, pfrag, sk->sk_allocation)))
2232 		return true;
2233 
2234 	sk_enter_memory_pressure(sk);
2235 	sk_stream_moderate_sndbuf(sk);
2236 	return false;
2237 }
2238 EXPORT_SYMBOL(sk_page_frag_refill);
2239 
2240 static void __lock_sock(struct sock *sk)
2241 	__releases(&sk->sk_lock.slock)
2242 	__acquires(&sk->sk_lock.slock)
2243 {
2244 	DEFINE_WAIT(wait);
2245 
2246 	for (;;) {
2247 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
2248 					TASK_UNINTERRUPTIBLE);
2249 		spin_unlock_bh(&sk->sk_lock.slock);
2250 		schedule();
2251 		spin_lock_bh(&sk->sk_lock.slock);
2252 		if (!sock_owned_by_user(sk))
2253 			break;
2254 	}
2255 	finish_wait(&sk->sk_lock.wq, &wait);
2256 }
2257 
2258 static void __release_sock(struct sock *sk)
2259 	__releases(&sk->sk_lock.slock)
2260 	__acquires(&sk->sk_lock.slock)
2261 {
2262 	struct sk_buff *skb, *next;
2263 
2264 	while ((skb = sk->sk_backlog.head) != NULL) {
2265 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
2266 
2267 		spin_unlock_bh(&sk->sk_lock.slock);
2268 
2269 		do {
2270 			next = skb->next;
2271 			prefetch(next);
2272 			WARN_ON_ONCE(skb_dst_is_noref(skb));
2273 			skb->next = NULL;
2274 			sk_backlog_rcv(sk, skb);
2275 
2276 			cond_resched();
2277 
2278 			skb = next;
2279 		} while (skb != NULL);
2280 
2281 		spin_lock_bh(&sk->sk_lock.slock);
2282 	}
2283 
2284 	/*
2285 	 * Doing the zeroing here guarantee we can not loop forever
2286 	 * while a wild producer attempts to flood us.
2287 	 */
2288 	sk->sk_backlog.len = 0;
2289 }
2290 
2291 void __sk_flush_backlog(struct sock *sk)
2292 {
2293 	spin_lock_bh(&sk->sk_lock.slock);
2294 	__release_sock(sk);
2295 	spin_unlock_bh(&sk->sk_lock.slock);
2296 }
2297 
2298 /**
2299  * sk_wait_data - wait for data to arrive at sk_receive_queue
2300  * @sk:    sock to wait on
2301  * @timeo: for how long
2302  * @skb:   last skb seen on sk_receive_queue
2303  *
2304  * Now socket state including sk->sk_err is changed only under lock,
2305  * hence we may omit checks after joining wait queue.
2306  * We check receive queue before schedule() only as optimization;
2307  * it is very likely that release_sock() added new data.
2308  */
2309 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb)
2310 {
2311 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
2312 	int rc;
2313 
2314 	add_wait_queue(sk_sleep(sk), &wait);
2315 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2316 	rc = sk_wait_event(sk, timeo, skb_peek_tail(&sk->sk_receive_queue) != skb, &wait);
2317 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
2318 	remove_wait_queue(sk_sleep(sk), &wait);
2319 	return rc;
2320 }
2321 EXPORT_SYMBOL(sk_wait_data);
2322 
2323 /**
2324  *	__sk_mem_raise_allocated - increase memory_allocated
2325  *	@sk: socket
2326  *	@size: memory size to allocate
2327  *	@amt: pages to allocate
2328  *	@kind: allocation type
2329  *
2330  *	Similar to __sk_mem_schedule(), but does not update sk_forward_alloc
2331  */
2332 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind)
2333 {
2334 	struct proto *prot = sk->sk_prot;
2335 	long allocated = sk_memory_allocated_add(sk, amt);
2336 
2337 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
2338 	    !mem_cgroup_charge_skmem(sk->sk_memcg, amt))
2339 		goto suppress_allocation;
2340 
2341 	/* Under limit. */
2342 	if (allocated <= sk_prot_mem_limits(sk, 0)) {
2343 		sk_leave_memory_pressure(sk);
2344 		return 1;
2345 	}
2346 
2347 	/* Under pressure. */
2348 	if (allocated > sk_prot_mem_limits(sk, 1))
2349 		sk_enter_memory_pressure(sk);
2350 
2351 	/* Over hard limit. */
2352 	if (allocated > sk_prot_mem_limits(sk, 2))
2353 		goto suppress_allocation;
2354 
2355 	/* guarantee minimum buffer size under pressure */
2356 	if (kind == SK_MEM_RECV) {
2357 		if (atomic_read(&sk->sk_rmem_alloc) < sk_get_rmem0(sk, prot))
2358 			return 1;
2359 
2360 	} else { /* SK_MEM_SEND */
2361 		int wmem0 = sk_get_wmem0(sk, prot);
2362 
2363 		if (sk->sk_type == SOCK_STREAM) {
2364 			if (sk->sk_wmem_queued < wmem0)
2365 				return 1;
2366 		} else if (refcount_read(&sk->sk_wmem_alloc) < wmem0) {
2367 				return 1;
2368 		}
2369 	}
2370 
2371 	if (sk_has_memory_pressure(sk)) {
2372 		int alloc;
2373 
2374 		if (!sk_under_memory_pressure(sk))
2375 			return 1;
2376 		alloc = sk_sockets_allocated_read_positive(sk);
2377 		if (sk_prot_mem_limits(sk, 2) > alloc *
2378 		    sk_mem_pages(sk->sk_wmem_queued +
2379 				 atomic_read(&sk->sk_rmem_alloc) +
2380 				 sk->sk_forward_alloc))
2381 			return 1;
2382 	}
2383 
2384 suppress_allocation:
2385 
2386 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
2387 		sk_stream_moderate_sndbuf(sk);
2388 
2389 		/* Fail only if socket is _under_ its sndbuf.
2390 		 * In this case we cannot block, so that we have to fail.
2391 		 */
2392 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2393 			return 1;
2394 	}
2395 
2396 	trace_sock_exceed_buf_limit(sk, prot, allocated);
2397 
2398 	sk_memory_allocated_sub(sk, amt);
2399 
2400 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2401 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amt);
2402 
2403 	return 0;
2404 }
2405 EXPORT_SYMBOL(__sk_mem_raise_allocated);
2406 
2407 /**
2408  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
2409  *	@sk: socket
2410  *	@size: memory size to allocate
2411  *	@kind: allocation type
2412  *
2413  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
2414  *	rmem allocation. This function assumes that protocols which have
2415  *	memory_pressure use sk_wmem_queued as write buffer accounting.
2416  */
2417 int __sk_mem_schedule(struct sock *sk, int size, int kind)
2418 {
2419 	int ret, amt = sk_mem_pages(size);
2420 
2421 	sk->sk_forward_alloc += amt << SK_MEM_QUANTUM_SHIFT;
2422 	ret = __sk_mem_raise_allocated(sk, size, amt, kind);
2423 	if (!ret)
2424 		sk->sk_forward_alloc -= amt << SK_MEM_QUANTUM_SHIFT;
2425 	return ret;
2426 }
2427 EXPORT_SYMBOL(__sk_mem_schedule);
2428 
2429 /**
2430  *	__sk_mem_reduce_allocated - reclaim memory_allocated
2431  *	@sk: socket
2432  *	@amount: number of quanta
2433  *
2434  *	Similar to __sk_mem_reclaim(), but does not update sk_forward_alloc
2435  */
2436 void __sk_mem_reduce_allocated(struct sock *sk, int amount)
2437 {
2438 	sk_memory_allocated_sub(sk, amount);
2439 
2440 	if (mem_cgroup_sockets_enabled && sk->sk_memcg)
2441 		mem_cgroup_uncharge_skmem(sk->sk_memcg, amount);
2442 
2443 	if (sk_under_memory_pressure(sk) &&
2444 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2445 		sk_leave_memory_pressure(sk);
2446 }
2447 EXPORT_SYMBOL(__sk_mem_reduce_allocated);
2448 
2449 /**
2450  *	__sk_mem_reclaim - reclaim sk_forward_alloc and memory_allocated
2451  *	@sk: socket
2452  *	@amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
2453  */
2454 void __sk_mem_reclaim(struct sock *sk, int amount)
2455 {
2456 	amount >>= SK_MEM_QUANTUM_SHIFT;
2457 	sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
2458 	__sk_mem_reduce_allocated(sk, amount);
2459 }
2460 EXPORT_SYMBOL(__sk_mem_reclaim);
2461 
2462 int sk_set_peek_off(struct sock *sk, int val)
2463 {
2464 	sk->sk_peek_off = val;
2465 	return 0;
2466 }
2467 EXPORT_SYMBOL_GPL(sk_set_peek_off);
2468 
2469 /*
2470  * Set of default routines for initialising struct proto_ops when
2471  * the protocol does not support a particular function. In certain
2472  * cases where it makes no sense for a protocol to have a "do nothing"
2473  * function, some default processing is provided.
2474  */
2475 
2476 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2477 {
2478 	return -EOPNOTSUPP;
2479 }
2480 EXPORT_SYMBOL(sock_no_bind);
2481 
2482 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2483 		    int len, int flags)
2484 {
2485 	return -EOPNOTSUPP;
2486 }
2487 EXPORT_SYMBOL(sock_no_connect);
2488 
2489 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2490 {
2491 	return -EOPNOTSUPP;
2492 }
2493 EXPORT_SYMBOL(sock_no_socketpair);
2494 
2495 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags,
2496 		   bool kern)
2497 {
2498 	return -EOPNOTSUPP;
2499 }
2500 EXPORT_SYMBOL(sock_no_accept);
2501 
2502 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2503 		    int *len, int peer)
2504 {
2505 	return -EOPNOTSUPP;
2506 }
2507 EXPORT_SYMBOL(sock_no_getname);
2508 
2509 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2510 {
2511 	return 0;
2512 }
2513 EXPORT_SYMBOL(sock_no_poll);
2514 
2515 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2516 {
2517 	return -EOPNOTSUPP;
2518 }
2519 EXPORT_SYMBOL(sock_no_ioctl);
2520 
2521 int sock_no_listen(struct socket *sock, int backlog)
2522 {
2523 	return -EOPNOTSUPP;
2524 }
2525 EXPORT_SYMBOL(sock_no_listen);
2526 
2527 int sock_no_shutdown(struct socket *sock, int how)
2528 {
2529 	return -EOPNOTSUPP;
2530 }
2531 EXPORT_SYMBOL(sock_no_shutdown);
2532 
2533 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2534 		    char __user *optval, unsigned int optlen)
2535 {
2536 	return -EOPNOTSUPP;
2537 }
2538 EXPORT_SYMBOL(sock_no_setsockopt);
2539 
2540 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2541 		    char __user *optval, int __user *optlen)
2542 {
2543 	return -EOPNOTSUPP;
2544 }
2545 EXPORT_SYMBOL(sock_no_getsockopt);
2546 
2547 int sock_no_sendmsg(struct socket *sock, struct msghdr *m, size_t len)
2548 {
2549 	return -EOPNOTSUPP;
2550 }
2551 EXPORT_SYMBOL(sock_no_sendmsg);
2552 
2553 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *m, size_t len)
2554 {
2555 	return -EOPNOTSUPP;
2556 }
2557 EXPORT_SYMBOL(sock_no_sendmsg_locked);
2558 
2559 int sock_no_recvmsg(struct socket *sock, struct msghdr *m, size_t len,
2560 		    int flags)
2561 {
2562 	return -EOPNOTSUPP;
2563 }
2564 EXPORT_SYMBOL(sock_no_recvmsg);
2565 
2566 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2567 {
2568 	/* Mirror missing mmap method error code */
2569 	return -ENODEV;
2570 }
2571 EXPORT_SYMBOL(sock_no_mmap);
2572 
2573 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2574 {
2575 	ssize_t res;
2576 	struct msghdr msg = {.msg_flags = flags};
2577 	struct kvec iov;
2578 	char *kaddr = kmap(page);
2579 	iov.iov_base = kaddr + offset;
2580 	iov.iov_len = size;
2581 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2582 	kunmap(page);
2583 	return res;
2584 }
2585 EXPORT_SYMBOL(sock_no_sendpage);
2586 
2587 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
2588 				int offset, size_t size, int flags)
2589 {
2590 	ssize_t res;
2591 	struct msghdr msg = {.msg_flags = flags};
2592 	struct kvec iov;
2593 	char *kaddr = kmap(page);
2594 
2595 	iov.iov_base = kaddr + offset;
2596 	iov.iov_len = size;
2597 	res = kernel_sendmsg_locked(sk, &msg, &iov, 1, size);
2598 	kunmap(page);
2599 	return res;
2600 }
2601 EXPORT_SYMBOL(sock_no_sendpage_locked);
2602 
2603 /*
2604  *	Default Socket Callbacks
2605  */
2606 
2607 static void sock_def_wakeup(struct sock *sk)
2608 {
2609 	struct socket_wq *wq;
2610 
2611 	rcu_read_lock();
2612 	wq = rcu_dereference(sk->sk_wq);
2613 	if (skwq_has_sleeper(wq))
2614 		wake_up_interruptible_all(&wq->wait);
2615 	rcu_read_unlock();
2616 }
2617 
2618 static void sock_def_error_report(struct sock *sk)
2619 {
2620 	struct socket_wq *wq;
2621 
2622 	rcu_read_lock();
2623 	wq = rcu_dereference(sk->sk_wq);
2624 	if (skwq_has_sleeper(wq))
2625 		wake_up_interruptible_poll(&wq->wait, POLLERR);
2626 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2627 	rcu_read_unlock();
2628 }
2629 
2630 static void sock_def_readable(struct sock *sk)
2631 {
2632 	struct socket_wq *wq;
2633 
2634 	rcu_read_lock();
2635 	wq = rcu_dereference(sk->sk_wq);
2636 	if (skwq_has_sleeper(wq))
2637 		wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2638 						POLLRDNORM | POLLRDBAND);
2639 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2640 	rcu_read_unlock();
2641 }
2642 
2643 static void sock_def_write_space(struct sock *sk)
2644 {
2645 	struct socket_wq *wq;
2646 
2647 	rcu_read_lock();
2648 
2649 	/* Do not wake up a writer until he can make "significant"
2650 	 * progress.  --DaveM
2651 	 */
2652 	if ((refcount_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2653 		wq = rcu_dereference(sk->sk_wq);
2654 		if (skwq_has_sleeper(wq))
2655 			wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2656 						POLLWRNORM | POLLWRBAND);
2657 
2658 		/* Should agree with poll, otherwise some programs break */
2659 		if (sock_writeable(sk))
2660 			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2661 	}
2662 
2663 	rcu_read_unlock();
2664 }
2665 
2666 static void sock_def_destruct(struct sock *sk)
2667 {
2668 }
2669 
2670 void sk_send_sigurg(struct sock *sk)
2671 {
2672 	if (sk->sk_socket && sk->sk_socket->file)
2673 		if (send_sigurg(&sk->sk_socket->file->f_owner))
2674 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2675 }
2676 EXPORT_SYMBOL(sk_send_sigurg);
2677 
2678 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2679 		    unsigned long expires)
2680 {
2681 	if (!mod_timer(timer, expires))
2682 		sock_hold(sk);
2683 }
2684 EXPORT_SYMBOL(sk_reset_timer);
2685 
2686 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2687 {
2688 	if (del_timer(timer))
2689 		__sock_put(sk);
2690 }
2691 EXPORT_SYMBOL(sk_stop_timer);
2692 
2693 void sock_init_data(struct socket *sock, struct sock *sk)
2694 {
2695 	sk_init_common(sk);
2696 	sk->sk_send_head	=	NULL;
2697 
2698 	timer_setup(&sk->sk_timer, NULL, 0);
2699 
2700 	sk->sk_allocation	=	GFP_KERNEL;
2701 	sk->sk_rcvbuf		=	sysctl_rmem_default;
2702 	sk->sk_sndbuf		=	sysctl_wmem_default;
2703 	sk->sk_state		=	TCP_CLOSE;
2704 	sk_set_socket(sk, sock);
2705 
2706 	sock_set_flag(sk, SOCK_ZAPPED);
2707 
2708 	if (sock) {
2709 		sk->sk_type	=	sock->type;
2710 		sk->sk_wq	=	sock->wq;
2711 		sock->sk	=	sk;
2712 		sk->sk_uid	=	SOCK_INODE(sock)->i_uid;
2713 	} else {
2714 		sk->sk_wq	=	NULL;
2715 		sk->sk_uid	=	make_kuid(sock_net(sk)->user_ns, 0);
2716 	}
2717 
2718 	rwlock_init(&sk->sk_callback_lock);
2719 	if (sk->sk_kern_sock)
2720 		lockdep_set_class_and_name(
2721 			&sk->sk_callback_lock,
2722 			af_kern_callback_keys + sk->sk_family,
2723 			af_family_kern_clock_key_strings[sk->sk_family]);
2724 	else
2725 		lockdep_set_class_and_name(
2726 			&sk->sk_callback_lock,
2727 			af_callback_keys + sk->sk_family,
2728 			af_family_clock_key_strings[sk->sk_family]);
2729 
2730 	sk->sk_state_change	=	sock_def_wakeup;
2731 	sk->sk_data_ready	=	sock_def_readable;
2732 	sk->sk_write_space	=	sock_def_write_space;
2733 	sk->sk_error_report	=	sock_def_error_report;
2734 	sk->sk_destruct		=	sock_def_destruct;
2735 
2736 	sk->sk_frag.page	=	NULL;
2737 	sk->sk_frag.offset	=	0;
2738 	sk->sk_peek_off		=	-1;
2739 
2740 	sk->sk_peer_pid 	=	NULL;
2741 	sk->sk_peer_cred	=	NULL;
2742 	sk->sk_write_pending	=	0;
2743 	sk->sk_rcvlowat		=	1;
2744 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
2745 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
2746 
2747 	sk->sk_stamp = SK_DEFAULT_STAMP;
2748 	atomic_set(&sk->sk_zckey, 0);
2749 
2750 #ifdef CONFIG_NET_RX_BUSY_POLL
2751 	sk->sk_napi_id		=	0;
2752 	sk->sk_ll_usec		=	sysctl_net_busy_read;
2753 #endif
2754 
2755 	sk->sk_max_pacing_rate = ~0U;
2756 	sk->sk_pacing_rate = ~0U;
2757 	sk->sk_pacing_shift = 10;
2758 	sk->sk_incoming_cpu = -1;
2759 	/*
2760 	 * Before updating sk_refcnt, we must commit prior changes to memory
2761 	 * (Documentation/RCU/rculist_nulls.txt for details)
2762 	 */
2763 	smp_wmb();
2764 	refcount_set(&sk->sk_refcnt, 1);
2765 	atomic_set(&sk->sk_drops, 0);
2766 }
2767 EXPORT_SYMBOL(sock_init_data);
2768 
2769 void lock_sock_nested(struct sock *sk, int subclass)
2770 {
2771 	might_sleep();
2772 	spin_lock_bh(&sk->sk_lock.slock);
2773 	if (sk->sk_lock.owned)
2774 		__lock_sock(sk);
2775 	sk->sk_lock.owned = 1;
2776 	spin_unlock(&sk->sk_lock.slock);
2777 	/*
2778 	 * The sk_lock has mutex_lock() semantics here:
2779 	 */
2780 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2781 	local_bh_enable();
2782 }
2783 EXPORT_SYMBOL(lock_sock_nested);
2784 
2785 void release_sock(struct sock *sk)
2786 {
2787 	spin_lock_bh(&sk->sk_lock.slock);
2788 	if (sk->sk_backlog.tail)
2789 		__release_sock(sk);
2790 
2791 	/* Warning : release_cb() might need to release sk ownership,
2792 	 * ie call sock_release_ownership(sk) before us.
2793 	 */
2794 	if (sk->sk_prot->release_cb)
2795 		sk->sk_prot->release_cb(sk);
2796 
2797 	sock_release_ownership(sk);
2798 	if (waitqueue_active(&sk->sk_lock.wq))
2799 		wake_up(&sk->sk_lock.wq);
2800 	spin_unlock_bh(&sk->sk_lock.slock);
2801 }
2802 EXPORT_SYMBOL(release_sock);
2803 
2804 /**
2805  * lock_sock_fast - fast version of lock_sock
2806  * @sk: socket
2807  *
2808  * This version should be used for very small section, where process wont block
2809  * return false if fast path is taken:
2810  *
2811  *   sk_lock.slock locked, owned = 0, BH disabled
2812  *
2813  * return true if slow path is taken:
2814  *
2815  *   sk_lock.slock unlocked, owned = 1, BH enabled
2816  */
2817 bool lock_sock_fast(struct sock *sk)
2818 {
2819 	might_sleep();
2820 	spin_lock_bh(&sk->sk_lock.slock);
2821 
2822 	if (!sk->sk_lock.owned)
2823 		/*
2824 		 * Note : We must disable BH
2825 		 */
2826 		return false;
2827 
2828 	__lock_sock(sk);
2829 	sk->sk_lock.owned = 1;
2830 	spin_unlock(&sk->sk_lock.slock);
2831 	/*
2832 	 * The sk_lock has mutex_lock() semantics here:
2833 	 */
2834 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2835 	local_bh_enable();
2836 	return true;
2837 }
2838 EXPORT_SYMBOL(lock_sock_fast);
2839 
2840 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2841 {
2842 	struct timeval tv;
2843 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2844 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2845 	tv = ktime_to_timeval(sk->sk_stamp);
2846 	if (tv.tv_sec == -1)
2847 		return -ENOENT;
2848 	if (tv.tv_sec == 0) {
2849 		sk->sk_stamp = ktime_get_real();
2850 		tv = ktime_to_timeval(sk->sk_stamp);
2851 	}
2852 	return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2853 }
2854 EXPORT_SYMBOL(sock_get_timestamp);
2855 
2856 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2857 {
2858 	struct timespec ts;
2859 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2860 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2861 	ts = ktime_to_timespec(sk->sk_stamp);
2862 	if (ts.tv_sec == -1)
2863 		return -ENOENT;
2864 	if (ts.tv_sec == 0) {
2865 		sk->sk_stamp = ktime_get_real();
2866 		ts = ktime_to_timespec(sk->sk_stamp);
2867 	}
2868 	return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2869 }
2870 EXPORT_SYMBOL(sock_get_timestampns);
2871 
2872 void sock_enable_timestamp(struct sock *sk, int flag)
2873 {
2874 	if (!sock_flag(sk, flag)) {
2875 		unsigned long previous_flags = sk->sk_flags;
2876 
2877 		sock_set_flag(sk, flag);
2878 		/*
2879 		 * we just set one of the two flags which require net
2880 		 * time stamping, but time stamping might have been on
2881 		 * already because of the other one
2882 		 */
2883 		if (sock_needs_netstamp(sk) &&
2884 		    !(previous_flags & SK_FLAGS_TIMESTAMP))
2885 			net_enable_timestamp();
2886 	}
2887 }
2888 
2889 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len,
2890 		       int level, int type)
2891 {
2892 	struct sock_exterr_skb *serr;
2893 	struct sk_buff *skb;
2894 	int copied, err;
2895 
2896 	err = -EAGAIN;
2897 	skb = sock_dequeue_err_skb(sk);
2898 	if (skb == NULL)
2899 		goto out;
2900 
2901 	copied = skb->len;
2902 	if (copied > len) {
2903 		msg->msg_flags |= MSG_TRUNC;
2904 		copied = len;
2905 	}
2906 	err = skb_copy_datagram_msg(skb, 0, msg, copied);
2907 	if (err)
2908 		goto out_free_skb;
2909 
2910 	sock_recv_timestamp(msg, sk, skb);
2911 
2912 	serr = SKB_EXT_ERR(skb);
2913 	put_cmsg(msg, level, type, sizeof(serr->ee), &serr->ee);
2914 
2915 	msg->msg_flags |= MSG_ERRQUEUE;
2916 	err = copied;
2917 
2918 out_free_skb:
2919 	kfree_skb(skb);
2920 out:
2921 	return err;
2922 }
2923 EXPORT_SYMBOL(sock_recv_errqueue);
2924 
2925 /*
2926  *	Get a socket option on an socket.
2927  *
2928  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
2929  *	asynchronous errors should be reported by getsockopt. We assume
2930  *	this means if you specify SO_ERROR (otherwise whats the point of it).
2931  */
2932 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2933 			   char __user *optval, int __user *optlen)
2934 {
2935 	struct sock *sk = sock->sk;
2936 
2937 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2938 }
2939 EXPORT_SYMBOL(sock_common_getsockopt);
2940 
2941 #ifdef CONFIG_COMPAT
2942 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2943 				  char __user *optval, int __user *optlen)
2944 {
2945 	struct sock *sk = sock->sk;
2946 
2947 	if (sk->sk_prot->compat_getsockopt != NULL)
2948 		return sk->sk_prot->compat_getsockopt(sk, level, optname,
2949 						      optval, optlen);
2950 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2951 }
2952 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2953 #endif
2954 
2955 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
2956 			int flags)
2957 {
2958 	struct sock *sk = sock->sk;
2959 	int addr_len = 0;
2960 	int err;
2961 
2962 	err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT,
2963 				   flags & ~MSG_DONTWAIT, &addr_len);
2964 	if (err >= 0)
2965 		msg->msg_namelen = addr_len;
2966 	return err;
2967 }
2968 EXPORT_SYMBOL(sock_common_recvmsg);
2969 
2970 /*
2971  *	Set socket options on an inet socket.
2972  */
2973 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2974 			   char __user *optval, unsigned int optlen)
2975 {
2976 	struct sock *sk = sock->sk;
2977 
2978 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2979 }
2980 EXPORT_SYMBOL(sock_common_setsockopt);
2981 
2982 #ifdef CONFIG_COMPAT
2983 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2984 				  char __user *optval, unsigned int optlen)
2985 {
2986 	struct sock *sk = sock->sk;
2987 
2988 	if (sk->sk_prot->compat_setsockopt != NULL)
2989 		return sk->sk_prot->compat_setsockopt(sk, level, optname,
2990 						      optval, optlen);
2991 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2992 }
2993 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2994 #endif
2995 
2996 void sk_common_release(struct sock *sk)
2997 {
2998 	if (sk->sk_prot->destroy)
2999 		sk->sk_prot->destroy(sk);
3000 
3001 	/*
3002 	 * Observation: when sock_common_release is called, processes have
3003 	 * no access to socket. But net still has.
3004 	 * Step one, detach it from networking:
3005 	 *
3006 	 * A. Remove from hash tables.
3007 	 */
3008 
3009 	sk->sk_prot->unhash(sk);
3010 
3011 	/*
3012 	 * In this point socket cannot receive new packets, but it is possible
3013 	 * that some packets are in flight because some CPU runs receiver and
3014 	 * did hash table lookup before we unhashed socket. They will achieve
3015 	 * receive queue and will be purged by socket destructor.
3016 	 *
3017 	 * Also we still have packets pending on receive queue and probably,
3018 	 * our own packets waiting in device queues. sock_destroy will drain
3019 	 * receive queue, but transmitted packets will delay socket destruction
3020 	 * until the last reference will be released.
3021 	 */
3022 
3023 	sock_orphan(sk);
3024 
3025 	xfrm_sk_free_policy(sk);
3026 
3027 	sk_refcnt_debug_release(sk);
3028 
3029 	sock_put(sk);
3030 }
3031 EXPORT_SYMBOL(sk_common_release);
3032 
3033 void sk_get_meminfo(const struct sock *sk, u32 *mem)
3034 {
3035 	memset(mem, 0, sizeof(*mem) * SK_MEMINFO_VARS);
3036 
3037 	mem[SK_MEMINFO_RMEM_ALLOC] = sk_rmem_alloc_get(sk);
3038 	mem[SK_MEMINFO_RCVBUF] = sk->sk_rcvbuf;
3039 	mem[SK_MEMINFO_WMEM_ALLOC] = sk_wmem_alloc_get(sk);
3040 	mem[SK_MEMINFO_SNDBUF] = sk->sk_sndbuf;
3041 	mem[SK_MEMINFO_FWD_ALLOC] = sk->sk_forward_alloc;
3042 	mem[SK_MEMINFO_WMEM_QUEUED] = sk->sk_wmem_queued;
3043 	mem[SK_MEMINFO_OPTMEM] = atomic_read(&sk->sk_omem_alloc);
3044 	mem[SK_MEMINFO_BACKLOG] = sk->sk_backlog.len;
3045 	mem[SK_MEMINFO_DROPS] = atomic_read(&sk->sk_drops);
3046 }
3047 
3048 #ifdef CONFIG_PROC_FS
3049 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
3050 struct prot_inuse {
3051 	int val[PROTO_INUSE_NR];
3052 };
3053 
3054 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
3055 
3056 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
3057 {
3058 	__this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
3059 }
3060 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
3061 
3062 int sock_prot_inuse_get(struct net *net, struct proto *prot)
3063 {
3064 	int cpu, idx = prot->inuse_idx;
3065 	int res = 0;
3066 
3067 	for_each_possible_cpu(cpu)
3068 		res += per_cpu_ptr(net->core.prot_inuse, cpu)->val[idx];
3069 
3070 	return res >= 0 ? res : 0;
3071 }
3072 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
3073 
3074 static void sock_inuse_add(struct net *net, int val)
3075 {
3076 	this_cpu_add(*net->core.sock_inuse, val);
3077 }
3078 
3079 int sock_inuse_get(struct net *net)
3080 {
3081 	int cpu, res = 0;
3082 
3083 	for_each_possible_cpu(cpu)
3084 		res += *per_cpu_ptr(net->core.sock_inuse, cpu);
3085 
3086 	return res;
3087 }
3088 
3089 EXPORT_SYMBOL_GPL(sock_inuse_get);
3090 
3091 static int __net_init sock_inuse_init_net(struct net *net)
3092 {
3093 	net->core.prot_inuse = alloc_percpu(struct prot_inuse);
3094 	if (net->core.prot_inuse == NULL)
3095 		return -ENOMEM;
3096 
3097 	net->core.sock_inuse = alloc_percpu(int);
3098 	if (net->core.sock_inuse == NULL)
3099 		goto out;
3100 
3101 	return 0;
3102 
3103 out:
3104 	free_percpu(net->core.prot_inuse);
3105 	return -ENOMEM;
3106 }
3107 
3108 static void __net_exit sock_inuse_exit_net(struct net *net)
3109 {
3110 	free_percpu(net->core.prot_inuse);
3111 	free_percpu(net->core.sock_inuse);
3112 }
3113 
3114 static struct pernet_operations net_inuse_ops = {
3115 	.init = sock_inuse_init_net,
3116 	.exit = sock_inuse_exit_net,
3117 };
3118 
3119 static __init int net_inuse_init(void)
3120 {
3121 	if (register_pernet_subsys(&net_inuse_ops))
3122 		panic("Cannot initialize net inuse counters");
3123 
3124 	return 0;
3125 }
3126 
3127 core_initcall(net_inuse_init);
3128 
3129 static void assign_proto_idx(struct proto *prot)
3130 {
3131 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
3132 
3133 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
3134 		pr_err("PROTO_INUSE_NR exhausted\n");
3135 		return;
3136 	}
3137 
3138 	set_bit(prot->inuse_idx, proto_inuse_idx);
3139 }
3140 
3141 static void release_proto_idx(struct proto *prot)
3142 {
3143 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
3144 		clear_bit(prot->inuse_idx, proto_inuse_idx);
3145 }
3146 #else
3147 static inline void assign_proto_idx(struct proto *prot)
3148 {
3149 }
3150 
3151 static inline void release_proto_idx(struct proto *prot)
3152 {
3153 }
3154 
3155 static void sock_inuse_add(struct net *net, int val)
3156 {
3157 }
3158 #endif
3159 
3160 static void req_prot_cleanup(struct request_sock_ops *rsk_prot)
3161 {
3162 	if (!rsk_prot)
3163 		return;
3164 	kfree(rsk_prot->slab_name);
3165 	rsk_prot->slab_name = NULL;
3166 	kmem_cache_destroy(rsk_prot->slab);
3167 	rsk_prot->slab = NULL;
3168 }
3169 
3170 static int req_prot_init(const struct proto *prot)
3171 {
3172 	struct request_sock_ops *rsk_prot = prot->rsk_prot;
3173 
3174 	if (!rsk_prot)
3175 		return 0;
3176 
3177 	rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s",
3178 					prot->name);
3179 	if (!rsk_prot->slab_name)
3180 		return -ENOMEM;
3181 
3182 	rsk_prot->slab = kmem_cache_create(rsk_prot->slab_name,
3183 					   rsk_prot->obj_size, 0,
3184 					   prot->slab_flags, NULL);
3185 
3186 	if (!rsk_prot->slab) {
3187 		pr_crit("%s: Can't create request sock SLAB cache!\n",
3188 			prot->name);
3189 		return -ENOMEM;
3190 	}
3191 	return 0;
3192 }
3193 
3194 int proto_register(struct proto *prot, int alloc_slab)
3195 {
3196 	if (alloc_slab) {
3197 		prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
3198 					SLAB_HWCACHE_ALIGN | prot->slab_flags,
3199 					NULL);
3200 
3201 		if (prot->slab == NULL) {
3202 			pr_crit("%s: Can't create sock SLAB cache!\n",
3203 				prot->name);
3204 			goto out;
3205 		}
3206 
3207 		if (req_prot_init(prot))
3208 			goto out_free_request_sock_slab;
3209 
3210 		if (prot->twsk_prot != NULL) {
3211 			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
3212 
3213 			if (prot->twsk_prot->twsk_slab_name == NULL)
3214 				goto out_free_request_sock_slab;
3215 
3216 			prot->twsk_prot->twsk_slab =
3217 				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
3218 						  prot->twsk_prot->twsk_obj_size,
3219 						  0,
3220 						  prot->slab_flags,
3221 						  NULL);
3222 			if (prot->twsk_prot->twsk_slab == NULL)
3223 				goto out_free_timewait_sock_slab_name;
3224 		}
3225 	}
3226 
3227 	mutex_lock(&proto_list_mutex);
3228 	list_add(&prot->node, &proto_list);
3229 	assign_proto_idx(prot);
3230 	mutex_unlock(&proto_list_mutex);
3231 	return 0;
3232 
3233 out_free_timewait_sock_slab_name:
3234 	kfree(prot->twsk_prot->twsk_slab_name);
3235 out_free_request_sock_slab:
3236 	req_prot_cleanup(prot->rsk_prot);
3237 
3238 	kmem_cache_destroy(prot->slab);
3239 	prot->slab = NULL;
3240 out:
3241 	return -ENOBUFS;
3242 }
3243 EXPORT_SYMBOL(proto_register);
3244 
3245 void proto_unregister(struct proto *prot)
3246 {
3247 	mutex_lock(&proto_list_mutex);
3248 	release_proto_idx(prot);
3249 	list_del(&prot->node);
3250 	mutex_unlock(&proto_list_mutex);
3251 
3252 	kmem_cache_destroy(prot->slab);
3253 	prot->slab = NULL;
3254 
3255 	req_prot_cleanup(prot->rsk_prot);
3256 
3257 	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
3258 		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
3259 		kfree(prot->twsk_prot->twsk_slab_name);
3260 		prot->twsk_prot->twsk_slab = NULL;
3261 	}
3262 }
3263 EXPORT_SYMBOL(proto_unregister);
3264 
3265 #ifdef CONFIG_PROC_FS
3266 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
3267 	__acquires(proto_list_mutex)
3268 {
3269 	mutex_lock(&proto_list_mutex);
3270 	return seq_list_start_head(&proto_list, *pos);
3271 }
3272 
3273 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3274 {
3275 	return seq_list_next(v, &proto_list, pos);
3276 }
3277 
3278 static void proto_seq_stop(struct seq_file *seq, void *v)
3279 	__releases(proto_list_mutex)
3280 {
3281 	mutex_unlock(&proto_list_mutex);
3282 }
3283 
3284 static char proto_method_implemented(const void *method)
3285 {
3286 	return method == NULL ? 'n' : 'y';
3287 }
3288 static long sock_prot_memory_allocated(struct proto *proto)
3289 {
3290 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
3291 }
3292 
3293 static char *sock_prot_memory_pressure(struct proto *proto)
3294 {
3295 	return proto->memory_pressure != NULL ?
3296 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
3297 }
3298 
3299 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
3300 {
3301 
3302 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
3303 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
3304 		   proto->name,
3305 		   proto->obj_size,
3306 		   sock_prot_inuse_get(seq_file_net(seq), proto),
3307 		   sock_prot_memory_allocated(proto),
3308 		   sock_prot_memory_pressure(proto),
3309 		   proto->max_header,
3310 		   proto->slab == NULL ? "no" : "yes",
3311 		   module_name(proto->owner),
3312 		   proto_method_implemented(proto->close),
3313 		   proto_method_implemented(proto->connect),
3314 		   proto_method_implemented(proto->disconnect),
3315 		   proto_method_implemented(proto->accept),
3316 		   proto_method_implemented(proto->ioctl),
3317 		   proto_method_implemented(proto->init),
3318 		   proto_method_implemented(proto->destroy),
3319 		   proto_method_implemented(proto->shutdown),
3320 		   proto_method_implemented(proto->setsockopt),
3321 		   proto_method_implemented(proto->getsockopt),
3322 		   proto_method_implemented(proto->sendmsg),
3323 		   proto_method_implemented(proto->recvmsg),
3324 		   proto_method_implemented(proto->sendpage),
3325 		   proto_method_implemented(proto->bind),
3326 		   proto_method_implemented(proto->backlog_rcv),
3327 		   proto_method_implemented(proto->hash),
3328 		   proto_method_implemented(proto->unhash),
3329 		   proto_method_implemented(proto->get_port),
3330 		   proto_method_implemented(proto->enter_memory_pressure));
3331 }
3332 
3333 static int proto_seq_show(struct seq_file *seq, void *v)
3334 {
3335 	if (v == &proto_list)
3336 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
3337 			   "protocol",
3338 			   "size",
3339 			   "sockets",
3340 			   "memory",
3341 			   "press",
3342 			   "maxhdr",
3343 			   "slab",
3344 			   "module",
3345 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
3346 	else
3347 		proto_seq_printf(seq, list_entry(v, struct proto, node));
3348 	return 0;
3349 }
3350 
3351 static const struct seq_operations proto_seq_ops = {
3352 	.start  = proto_seq_start,
3353 	.next   = proto_seq_next,
3354 	.stop   = proto_seq_stop,
3355 	.show   = proto_seq_show,
3356 };
3357 
3358 static int proto_seq_open(struct inode *inode, struct file *file)
3359 {
3360 	return seq_open_net(inode, file, &proto_seq_ops,
3361 			    sizeof(struct seq_net_private));
3362 }
3363 
3364 static const struct file_operations proto_seq_fops = {
3365 	.owner		= THIS_MODULE,
3366 	.open		= proto_seq_open,
3367 	.read		= seq_read,
3368 	.llseek		= seq_lseek,
3369 	.release	= seq_release_net,
3370 };
3371 
3372 static __net_init int proto_init_net(struct net *net)
3373 {
3374 	if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
3375 		return -ENOMEM;
3376 
3377 	return 0;
3378 }
3379 
3380 static __net_exit void proto_exit_net(struct net *net)
3381 {
3382 	remove_proc_entry("protocols", net->proc_net);
3383 }
3384 
3385 
3386 static __net_initdata struct pernet_operations proto_net_ops = {
3387 	.init = proto_init_net,
3388 	.exit = proto_exit_net,
3389 };
3390 
3391 static int __init proto_init(void)
3392 {
3393 	return register_pernet_subsys(&proto_net_ops);
3394 }
3395 
3396 subsys_initcall(proto_init);
3397 
3398 #endif /* PROC_FS */
3399 
3400 #ifdef CONFIG_NET_RX_BUSY_POLL
3401 bool sk_busy_loop_end(void *p, unsigned long start_time)
3402 {
3403 	struct sock *sk = p;
3404 
3405 	return !skb_queue_empty(&sk->sk_receive_queue) ||
3406 	       sk_busy_loop_timeout(sk, start_time);
3407 }
3408 EXPORT_SYMBOL(sk_busy_loop_end);
3409 #endif /* CONFIG_NET_RX_BUSY_POLL */
3410