xref: /linux/net/core/sock.c (revision f884ab15afdc5514e88105c92a4e2e1e6539869a)
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/types.h>
97 #include <linux/socket.h>
98 #include <linux/in.h>
99 #include <linux/kernel.h>
100 #include <linux/module.h>
101 #include <linux/proc_fs.h>
102 #include <linux/seq_file.h>
103 #include <linux/sched.h>
104 #include <linux/timer.h>
105 #include <linux/string.h>
106 #include <linux/sockios.h>
107 #include <linux/net.h>
108 #include <linux/mm.h>
109 #include <linux/slab.h>
110 #include <linux/interrupt.h>
111 #include <linux/poll.h>
112 #include <linux/tcp.h>
113 #include <linux/init.h>
114 #include <linux/highmem.h>
115 #include <linux/user_namespace.h>
116 #include <linux/static_key.h>
117 #include <linux/memcontrol.h>
118 #include <linux/prefetch.h>
119 
120 #include <asm/uaccess.h>
121 
122 #include <linux/netdevice.h>
123 #include <net/protocol.h>
124 #include <linux/skbuff.h>
125 #include <net/net_namespace.h>
126 #include <net/request_sock.h>
127 #include <net/sock.h>
128 #include <linux/net_tstamp.h>
129 #include <net/xfrm.h>
130 #include <linux/ipsec.h>
131 #include <net/cls_cgroup.h>
132 #include <net/netprio_cgroup.h>
133 
134 #include <linux/filter.h>
135 
136 #include <trace/events/sock.h>
137 
138 #ifdef CONFIG_INET
139 #include <net/tcp.h>
140 #endif
141 
142 static DEFINE_MUTEX(proto_list_mutex);
143 static LIST_HEAD(proto_list);
144 
145 #ifdef CONFIG_MEMCG_KMEM
146 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
147 {
148 	struct proto *proto;
149 	int ret = 0;
150 
151 	mutex_lock(&proto_list_mutex);
152 	list_for_each_entry(proto, &proto_list, node) {
153 		if (proto->init_cgroup) {
154 			ret = proto->init_cgroup(memcg, ss);
155 			if (ret)
156 				goto out;
157 		}
158 	}
159 
160 	mutex_unlock(&proto_list_mutex);
161 	return ret;
162 out:
163 	list_for_each_entry_continue_reverse(proto, &proto_list, node)
164 		if (proto->destroy_cgroup)
165 			proto->destroy_cgroup(memcg);
166 	mutex_unlock(&proto_list_mutex);
167 	return ret;
168 }
169 
170 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
171 {
172 	struct proto *proto;
173 
174 	mutex_lock(&proto_list_mutex);
175 	list_for_each_entry_reverse(proto, &proto_list, node)
176 		if (proto->destroy_cgroup)
177 			proto->destroy_cgroup(memcg);
178 	mutex_unlock(&proto_list_mutex);
179 }
180 #endif
181 
182 /*
183  * Each address family might have different locking rules, so we have
184  * one slock key per address family:
185  */
186 static struct lock_class_key af_family_keys[AF_MAX];
187 static struct lock_class_key af_family_slock_keys[AF_MAX];
188 
189 #if defined(CONFIG_MEMCG_KMEM)
190 struct static_key memcg_socket_limit_enabled;
191 EXPORT_SYMBOL(memcg_socket_limit_enabled);
192 #endif
193 
194 /*
195  * Make lock validator output more readable. (we pre-construct these
196  * strings build-time, so that runtime initialization of socket
197  * locks is fast):
198  */
199 static const char *const af_family_key_strings[AF_MAX+1] = {
200   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
201   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
202   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
203   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
204   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
205   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
206   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
207   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
208   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
209   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
210   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
211   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
212   "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
213   "sk_lock-AF_NFC"   , "sk_lock-AF_MAX"
214 };
215 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
216   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
217   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
218   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
219   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
220   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
221   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
222   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
223   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
224   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
225   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
226   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
227   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
228   "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
229   "slock-AF_NFC"   , "slock-AF_MAX"
230 };
231 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
232   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
233   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
234   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
235   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
236   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
237   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
238   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
239   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
240   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
241   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
242   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
243   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
244   "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
245   "clock-AF_NFC"   , "clock-AF_MAX"
246 };
247 
248 /*
249  * sk_callback_lock locking rules are per-address-family,
250  * so split the lock classes by using a per-AF key:
251  */
252 static struct lock_class_key af_callback_keys[AF_MAX];
253 
254 /* Take into consideration the size of the struct sk_buff overhead in the
255  * determination of these values, since that is non-constant across
256  * platforms.  This makes socket queueing behavior and performance
257  * not depend upon such differences.
258  */
259 #define _SK_MEM_PACKETS		256
260 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
261 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
262 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
263 
264 /* Run time adjustable parameters. */
265 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
266 EXPORT_SYMBOL(sysctl_wmem_max);
267 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
268 EXPORT_SYMBOL(sysctl_rmem_max);
269 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
270 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
271 
272 /* Maximal space eaten by iovec or ancillary data plus some space */
273 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
274 EXPORT_SYMBOL(sysctl_optmem_max);
275 
276 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
277 EXPORT_SYMBOL_GPL(memalloc_socks);
278 
279 /**
280  * sk_set_memalloc - sets %SOCK_MEMALLOC
281  * @sk: socket to set it on
282  *
283  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
284  * It's the responsibility of the admin to adjust min_free_kbytes
285  * to meet the requirements
286  */
287 void sk_set_memalloc(struct sock *sk)
288 {
289 	sock_set_flag(sk, SOCK_MEMALLOC);
290 	sk->sk_allocation |= __GFP_MEMALLOC;
291 	static_key_slow_inc(&memalloc_socks);
292 }
293 EXPORT_SYMBOL_GPL(sk_set_memalloc);
294 
295 void sk_clear_memalloc(struct sock *sk)
296 {
297 	sock_reset_flag(sk, SOCK_MEMALLOC);
298 	sk->sk_allocation &= ~__GFP_MEMALLOC;
299 	static_key_slow_dec(&memalloc_socks);
300 
301 	/*
302 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
303 	 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
304 	 * it has rmem allocations there is a risk that the user of the
305 	 * socket cannot make forward progress due to exceeding the rmem
306 	 * limits. By rights, sk_clear_memalloc() should only be called
307 	 * on sockets being torn down but warn and reset the accounting if
308 	 * that assumption breaks.
309 	 */
310 	if (WARN_ON(sk->sk_forward_alloc))
311 		sk_mem_reclaim(sk);
312 }
313 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
314 
315 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
316 {
317 	int ret;
318 	unsigned long pflags = current->flags;
319 
320 	/* these should have been dropped before queueing */
321 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
322 
323 	current->flags |= PF_MEMALLOC;
324 	ret = sk->sk_backlog_rcv(sk, skb);
325 	tsk_restore_flags(current, pflags, PF_MEMALLOC);
326 
327 	return ret;
328 }
329 EXPORT_SYMBOL(__sk_backlog_rcv);
330 
331 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
332 {
333 	struct timeval tv;
334 
335 	if (optlen < sizeof(tv))
336 		return -EINVAL;
337 	if (copy_from_user(&tv, optval, sizeof(tv)))
338 		return -EFAULT;
339 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
340 		return -EDOM;
341 
342 	if (tv.tv_sec < 0) {
343 		static int warned __read_mostly;
344 
345 		*timeo_p = 0;
346 		if (warned < 10 && net_ratelimit()) {
347 			warned++;
348 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
349 				__func__, current->comm, task_pid_nr(current));
350 		}
351 		return 0;
352 	}
353 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
354 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
355 		return 0;
356 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
357 		*timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
358 	return 0;
359 }
360 
361 static void sock_warn_obsolete_bsdism(const char *name)
362 {
363 	static int warned;
364 	static char warncomm[TASK_COMM_LEN];
365 	if (strcmp(warncomm, current->comm) && warned < 5) {
366 		strcpy(warncomm,  current->comm);
367 		pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
368 			warncomm, name);
369 		warned++;
370 	}
371 }
372 
373 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
374 
375 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
376 {
377 	if (sk->sk_flags & flags) {
378 		sk->sk_flags &= ~flags;
379 		if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
380 			net_disable_timestamp();
381 	}
382 }
383 
384 
385 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
386 {
387 	int err;
388 	int skb_len;
389 	unsigned long flags;
390 	struct sk_buff_head *list = &sk->sk_receive_queue;
391 
392 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
393 		atomic_inc(&sk->sk_drops);
394 		trace_sock_rcvqueue_full(sk, skb);
395 		return -ENOMEM;
396 	}
397 
398 	err = sk_filter(sk, skb);
399 	if (err)
400 		return err;
401 
402 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
403 		atomic_inc(&sk->sk_drops);
404 		return -ENOBUFS;
405 	}
406 
407 	skb->dev = NULL;
408 	skb_set_owner_r(skb, sk);
409 
410 	/* Cache the SKB length before we tack it onto the receive
411 	 * queue.  Once it is added it no longer belongs to us and
412 	 * may be freed by other threads of control pulling packets
413 	 * from the queue.
414 	 */
415 	skb_len = skb->len;
416 
417 	/* we escape from rcu protected region, make sure we dont leak
418 	 * a norefcounted dst
419 	 */
420 	skb_dst_force(skb);
421 
422 	spin_lock_irqsave(&list->lock, flags);
423 	skb->dropcount = atomic_read(&sk->sk_drops);
424 	__skb_queue_tail(list, skb);
425 	spin_unlock_irqrestore(&list->lock, flags);
426 
427 	if (!sock_flag(sk, SOCK_DEAD))
428 		sk->sk_data_ready(sk, skb_len);
429 	return 0;
430 }
431 EXPORT_SYMBOL(sock_queue_rcv_skb);
432 
433 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
434 {
435 	int rc = NET_RX_SUCCESS;
436 
437 	if (sk_filter(sk, skb))
438 		goto discard_and_relse;
439 
440 	skb->dev = NULL;
441 
442 	if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
443 		atomic_inc(&sk->sk_drops);
444 		goto discard_and_relse;
445 	}
446 	if (nested)
447 		bh_lock_sock_nested(sk);
448 	else
449 		bh_lock_sock(sk);
450 	if (!sock_owned_by_user(sk)) {
451 		/*
452 		 * trylock + unlock semantics:
453 		 */
454 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
455 
456 		rc = sk_backlog_rcv(sk, skb);
457 
458 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
459 	} else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
460 		bh_unlock_sock(sk);
461 		atomic_inc(&sk->sk_drops);
462 		goto discard_and_relse;
463 	}
464 
465 	bh_unlock_sock(sk);
466 out:
467 	sock_put(sk);
468 	return rc;
469 discard_and_relse:
470 	kfree_skb(skb);
471 	goto out;
472 }
473 EXPORT_SYMBOL(sk_receive_skb);
474 
475 void sk_reset_txq(struct sock *sk)
476 {
477 	sk_tx_queue_clear(sk);
478 }
479 EXPORT_SYMBOL(sk_reset_txq);
480 
481 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
482 {
483 	struct dst_entry *dst = __sk_dst_get(sk);
484 
485 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
486 		sk_tx_queue_clear(sk);
487 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
488 		dst_release(dst);
489 		return NULL;
490 	}
491 
492 	return dst;
493 }
494 EXPORT_SYMBOL(__sk_dst_check);
495 
496 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
497 {
498 	struct dst_entry *dst = sk_dst_get(sk);
499 
500 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
501 		sk_dst_reset(sk);
502 		dst_release(dst);
503 		return NULL;
504 	}
505 
506 	return dst;
507 }
508 EXPORT_SYMBOL(sk_dst_check);
509 
510 static int sock_setbindtodevice(struct sock *sk, char __user *optval,
511 				int optlen)
512 {
513 	int ret = -ENOPROTOOPT;
514 #ifdef CONFIG_NETDEVICES
515 	struct net *net = sock_net(sk);
516 	char devname[IFNAMSIZ];
517 	int index;
518 
519 	/* Sorry... */
520 	ret = -EPERM;
521 	if (!ns_capable(net->user_ns, CAP_NET_RAW))
522 		goto out;
523 
524 	ret = -EINVAL;
525 	if (optlen < 0)
526 		goto out;
527 
528 	/* Bind this socket to a particular device like "eth0",
529 	 * as specified in the passed interface name. If the
530 	 * name is "" or the option length is zero the socket
531 	 * is not bound.
532 	 */
533 	if (optlen > IFNAMSIZ - 1)
534 		optlen = IFNAMSIZ - 1;
535 	memset(devname, 0, sizeof(devname));
536 
537 	ret = -EFAULT;
538 	if (copy_from_user(devname, optval, optlen))
539 		goto out;
540 
541 	index = 0;
542 	if (devname[0] != '\0') {
543 		struct net_device *dev;
544 
545 		rcu_read_lock();
546 		dev = dev_get_by_name_rcu(net, devname);
547 		if (dev)
548 			index = dev->ifindex;
549 		rcu_read_unlock();
550 		ret = -ENODEV;
551 		if (!dev)
552 			goto out;
553 	}
554 
555 	lock_sock(sk);
556 	sk->sk_bound_dev_if = index;
557 	sk_dst_reset(sk);
558 	release_sock(sk);
559 
560 	ret = 0;
561 
562 out:
563 #endif
564 
565 	return ret;
566 }
567 
568 static int sock_getbindtodevice(struct sock *sk, char __user *optval,
569 				int __user *optlen, int len)
570 {
571 	int ret = -ENOPROTOOPT;
572 #ifdef CONFIG_NETDEVICES
573 	struct net *net = sock_net(sk);
574 	struct net_device *dev;
575 	char devname[IFNAMSIZ];
576 	unsigned seq;
577 
578 	if (sk->sk_bound_dev_if == 0) {
579 		len = 0;
580 		goto zero;
581 	}
582 
583 	ret = -EINVAL;
584 	if (len < IFNAMSIZ)
585 		goto out;
586 
587 retry:
588 	seq = read_seqcount_begin(&devnet_rename_seq);
589 	rcu_read_lock();
590 	dev = dev_get_by_index_rcu(net, sk->sk_bound_dev_if);
591 	ret = -ENODEV;
592 	if (!dev) {
593 		rcu_read_unlock();
594 		goto out;
595 	}
596 
597 	strcpy(devname, dev->name);
598 	rcu_read_unlock();
599 	if (read_seqcount_retry(&devnet_rename_seq, seq))
600 		goto retry;
601 
602 	len = strlen(devname) + 1;
603 
604 	ret = -EFAULT;
605 	if (copy_to_user(optval, devname, len))
606 		goto out;
607 
608 zero:
609 	ret = -EFAULT;
610 	if (put_user(len, optlen))
611 		goto out;
612 
613 	ret = 0;
614 
615 out:
616 #endif
617 
618 	return ret;
619 }
620 
621 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
622 {
623 	if (valbool)
624 		sock_set_flag(sk, bit);
625 	else
626 		sock_reset_flag(sk, bit);
627 }
628 
629 /*
630  *	This is meant for all protocols to use and covers goings on
631  *	at the socket level. Everything here is generic.
632  */
633 
634 int sock_setsockopt(struct socket *sock, int level, int optname,
635 		    char __user *optval, unsigned int optlen)
636 {
637 	struct sock *sk = sock->sk;
638 	int val;
639 	int valbool;
640 	struct linger ling;
641 	int ret = 0;
642 
643 	/*
644 	 *	Options without arguments
645 	 */
646 
647 	if (optname == SO_BINDTODEVICE)
648 		return sock_setbindtodevice(sk, optval, optlen);
649 
650 	if (optlen < sizeof(int))
651 		return -EINVAL;
652 
653 	if (get_user(val, (int __user *)optval))
654 		return -EFAULT;
655 
656 	valbool = val ? 1 : 0;
657 
658 	lock_sock(sk);
659 
660 	switch (optname) {
661 	case SO_DEBUG:
662 		if (val && !capable(CAP_NET_ADMIN))
663 			ret = -EACCES;
664 		else
665 			sock_valbool_flag(sk, SOCK_DBG, valbool);
666 		break;
667 	case SO_REUSEADDR:
668 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
669 		break;
670 	case SO_REUSEPORT:
671 		sk->sk_reuseport = valbool;
672 		break;
673 	case SO_TYPE:
674 	case SO_PROTOCOL:
675 	case SO_DOMAIN:
676 	case SO_ERROR:
677 		ret = -ENOPROTOOPT;
678 		break;
679 	case SO_DONTROUTE:
680 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
681 		break;
682 	case SO_BROADCAST:
683 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
684 		break;
685 	case SO_SNDBUF:
686 		/* Don't error on this BSD doesn't and if you think
687 		 * about it this is right. Otherwise apps have to
688 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
689 		 * are treated in BSD as hints
690 		 */
691 		val = min_t(u32, val, sysctl_wmem_max);
692 set_sndbuf:
693 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
694 		sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
695 		/* Wake up sending tasks if we upped the value. */
696 		sk->sk_write_space(sk);
697 		break;
698 
699 	case SO_SNDBUFFORCE:
700 		if (!capable(CAP_NET_ADMIN)) {
701 			ret = -EPERM;
702 			break;
703 		}
704 		goto set_sndbuf;
705 
706 	case SO_RCVBUF:
707 		/* Don't error on this BSD doesn't and if you think
708 		 * about it this is right. Otherwise apps have to
709 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
710 		 * are treated in BSD as hints
711 		 */
712 		val = min_t(u32, val, sysctl_rmem_max);
713 set_rcvbuf:
714 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
715 		/*
716 		 * We double it on the way in to account for
717 		 * "struct sk_buff" etc. overhead.   Applications
718 		 * assume that the SO_RCVBUF setting they make will
719 		 * allow that much actual data to be received on that
720 		 * socket.
721 		 *
722 		 * Applications are unaware that "struct sk_buff" and
723 		 * other overheads allocate from the receive buffer
724 		 * during socket buffer allocation.
725 		 *
726 		 * And after considering the possible alternatives,
727 		 * returning the value we actually used in getsockopt
728 		 * is the most desirable behavior.
729 		 */
730 		sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
731 		break;
732 
733 	case SO_RCVBUFFORCE:
734 		if (!capable(CAP_NET_ADMIN)) {
735 			ret = -EPERM;
736 			break;
737 		}
738 		goto set_rcvbuf;
739 
740 	case SO_KEEPALIVE:
741 #ifdef CONFIG_INET
742 		if (sk->sk_protocol == IPPROTO_TCP &&
743 		    sk->sk_type == SOCK_STREAM)
744 			tcp_set_keepalive(sk, valbool);
745 #endif
746 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
747 		break;
748 
749 	case SO_OOBINLINE:
750 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
751 		break;
752 
753 	case SO_NO_CHECK:
754 		sk->sk_no_check = valbool;
755 		break;
756 
757 	case SO_PRIORITY:
758 		if ((val >= 0 && val <= 6) ||
759 		    ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
760 			sk->sk_priority = val;
761 		else
762 			ret = -EPERM;
763 		break;
764 
765 	case SO_LINGER:
766 		if (optlen < sizeof(ling)) {
767 			ret = -EINVAL;	/* 1003.1g */
768 			break;
769 		}
770 		if (copy_from_user(&ling, optval, sizeof(ling))) {
771 			ret = -EFAULT;
772 			break;
773 		}
774 		if (!ling.l_onoff)
775 			sock_reset_flag(sk, SOCK_LINGER);
776 		else {
777 #if (BITS_PER_LONG == 32)
778 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
779 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
780 			else
781 #endif
782 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
783 			sock_set_flag(sk, SOCK_LINGER);
784 		}
785 		break;
786 
787 	case SO_BSDCOMPAT:
788 		sock_warn_obsolete_bsdism("setsockopt");
789 		break;
790 
791 	case SO_PASSCRED:
792 		if (valbool)
793 			set_bit(SOCK_PASSCRED, &sock->flags);
794 		else
795 			clear_bit(SOCK_PASSCRED, &sock->flags);
796 		break;
797 
798 	case SO_TIMESTAMP:
799 	case SO_TIMESTAMPNS:
800 		if (valbool)  {
801 			if (optname == SO_TIMESTAMP)
802 				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
803 			else
804 				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
805 			sock_set_flag(sk, SOCK_RCVTSTAMP);
806 			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
807 		} else {
808 			sock_reset_flag(sk, SOCK_RCVTSTAMP);
809 			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
810 		}
811 		break;
812 
813 	case SO_TIMESTAMPING:
814 		if (val & ~SOF_TIMESTAMPING_MASK) {
815 			ret = -EINVAL;
816 			break;
817 		}
818 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
819 				  val & SOF_TIMESTAMPING_TX_HARDWARE);
820 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
821 				  val & SOF_TIMESTAMPING_TX_SOFTWARE);
822 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
823 				  val & SOF_TIMESTAMPING_RX_HARDWARE);
824 		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
825 			sock_enable_timestamp(sk,
826 					      SOCK_TIMESTAMPING_RX_SOFTWARE);
827 		else
828 			sock_disable_timestamp(sk,
829 					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
830 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
831 				  val & SOF_TIMESTAMPING_SOFTWARE);
832 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
833 				  val & SOF_TIMESTAMPING_SYS_HARDWARE);
834 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
835 				  val & SOF_TIMESTAMPING_RAW_HARDWARE);
836 		break;
837 
838 	case SO_RCVLOWAT:
839 		if (val < 0)
840 			val = INT_MAX;
841 		sk->sk_rcvlowat = val ? : 1;
842 		break;
843 
844 	case SO_RCVTIMEO:
845 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
846 		break;
847 
848 	case SO_SNDTIMEO:
849 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
850 		break;
851 
852 	case SO_ATTACH_FILTER:
853 		ret = -EINVAL;
854 		if (optlen == sizeof(struct sock_fprog)) {
855 			struct sock_fprog fprog;
856 
857 			ret = -EFAULT;
858 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
859 				break;
860 
861 			ret = sk_attach_filter(&fprog, sk);
862 		}
863 		break;
864 
865 	case SO_DETACH_FILTER:
866 		ret = sk_detach_filter(sk);
867 		break;
868 
869 	case SO_LOCK_FILTER:
870 		if (sock_flag(sk, SOCK_FILTER_LOCKED) && !valbool)
871 			ret = -EPERM;
872 		else
873 			sock_valbool_flag(sk, SOCK_FILTER_LOCKED, valbool);
874 		break;
875 
876 	case SO_PASSSEC:
877 		if (valbool)
878 			set_bit(SOCK_PASSSEC, &sock->flags);
879 		else
880 			clear_bit(SOCK_PASSSEC, &sock->flags);
881 		break;
882 	case SO_MARK:
883 		if (!ns_capable(sock_net(sk)->user_ns, CAP_NET_ADMIN))
884 			ret = -EPERM;
885 		else
886 			sk->sk_mark = val;
887 		break;
888 
889 		/* We implement the SO_SNDLOWAT etc to
890 		   not be settable (1003.1g 5.3) */
891 	case SO_RXQ_OVFL:
892 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
893 		break;
894 
895 	case SO_WIFI_STATUS:
896 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
897 		break;
898 
899 	case SO_PEEK_OFF:
900 		if (sock->ops->set_peek_off)
901 			sock->ops->set_peek_off(sk, val);
902 		else
903 			ret = -EOPNOTSUPP;
904 		break;
905 
906 	case SO_NOFCS:
907 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
908 		break;
909 
910 	case SO_SELECT_ERR_QUEUE:
911 		sock_valbool_flag(sk, SOCK_SELECT_ERR_QUEUE, valbool);
912 		break;
913 
914 	default:
915 		ret = -ENOPROTOOPT;
916 		break;
917 	}
918 	release_sock(sk);
919 	return ret;
920 }
921 EXPORT_SYMBOL(sock_setsockopt);
922 
923 
924 void cred_to_ucred(struct pid *pid, const struct cred *cred,
925 		   struct ucred *ucred)
926 {
927 	ucred->pid = pid_vnr(pid);
928 	ucred->uid = ucred->gid = -1;
929 	if (cred) {
930 		struct user_namespace *current_ns = current_user_ns();
931 
932 		ucred->uid = from_kuid_munged(current_ns, cred->euid);
933 		ucred->gid = from_kgid_munged(current_ns, cred->egid);
934 	}
935 }
936 EXPORT_SYMBOL_GPL(cred_to_ucred);
937 
938 int sock_getsockopt(struct socket *sock, int level, int optname,
939 		    char __user *optval, int __user *optlen)
940 {
941 	struct sock *sk = sock->sk;
942 
943 	union {
944 		int val;
945 		struct linger ling;
946 		struct timeval tm;
947 	} v;
948 
949 	int lv = sizeof(int);
950 	int len;
951 
952 	if (get_user(len, optlen))
953 		return -EFAULT;
954 	if (len < 0)
955 		return -EINVAL;
956 
957 	memset(&v, 0, sizeof(v));
958 
959 	switch (optname) {
960 	case SO_DEBUG:
961 		v.val = sock_flag(sk, SOCK_DBG);
962 		break;
963 
964 	case SO_DONTROUTE:
965 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
966 		break;
967 
968 	case SO_BROADCAST:
969 		v.val = sock_flag(sk, SOCK_BROADCAST);
970 		break;
971 
972 	case SO_SNDBUF:
973 		v.val = sk->sk_sndbuf;
974 		break;
975 
976 	case SO_RCVBUF:
977 		v.val = sk->sk_rcvbuf;
978 		break;
979 
980 	case SO_REUSEADDR:
981 		v.val = sk->sk_reuse;
982 		break;
983 
984 	case SO_REUSEPORT:
985 		v.val = sk->sk_reuseport;
986 		break;
987 
988 	case SO_KEEPALIVE:
989 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
990 		break;
991 
992 	case SO_TYPE:
993 		v.val = sk->sk_type;
994 		break;
995 
996 	case SO_PROTOCOL:
997 		v.val = sk->sk_protocol;
998 		break;
999 
1000 	case SO_DOMAIN:
1001 		v.val = sk->sk_family;
1002 		break;
1003 
1004 	case SO_ERROR:
1005 		v.val = -sock_error(sk);
1006 		if (v.val == 0)
1007 			v.val = xchg(&sk->sk_err_soft, 0);
1008 		break;
1009 
1010 	case SO_OOBINLINE:
1011 		v.val = sock_flag(sk, SOCK_URGINLINE);
1012 		break;
1013 
1014 	case SO_NO_CHECK:
1015 		v.val = sk->sk_no_check;
1016 		break;
1017 
1018 	case SO_PRIORITY:
1019 		v.val = sk->sk_priority;
1020 		break;
1021 
1022 	case SO_LINGER:
1023 		lv		= sizeof(v.ling);
1024 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
1025 		v.ling.l_linger	= sk->sk_lingertime / HZ;
1026 		break;
1027 
1028 	case SO_BSDCOMPAT:
1029 		sock_warn_obsolete_bsdism("getsockopt");
1030 		break;
1031 
1032 	case SO_TIMESTAMP:
1033 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
1034 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
1035 		break;
1036 
1037 	case SO_TIMESTAMPNS:
1038 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
1039 		break;
1040 
1041 	case SO_TIMESTAMPING:
1042 		v.val = 0;
1043 		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
1044 			v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
1045 		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
1046 			v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
1047 		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
1048 			v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
1049 		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1050 			v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
1051 		if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
1052 			v.val |= SOF_TIMESTAMPING_SOFTWARE;
1053 		if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
1054 			v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
1055 		if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
1056 			v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
1057 		break;
1058 
1059 	case SO_RCVTIMEO:
1060 		lv = sizeof(struct timeval);
1061 		if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
1062 			v.tm.tv_sec = 0;
1063 			v.tm.tv_usec = 0;
1064 		} else {
1065 			v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1066 			v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1067 		}
1068 		break;
1069 
1070 	case SO_SNDTIMEO:
1071 		lv = sizeof(struct timeval);
1072 		if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1073 			v.tm.tv_sec = 0;
1074 			v.tm.tv_usec = 0;
1075 		} else {
1076 			v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1077 			v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1078 		}
1079 		break;
1080 
1081 	case SO_RCVLOWAT:
1082 		v.val = sk->sk_rcvlowat;
1083 		break;
1084 
1085 	case SO_SNDLOWAT:
1086 		v.val = 1;
1087 		break;
1088 
1089 	case SO_PASSCRED:
1090 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1091 		break;
1092 
1093 	case SO_PEERCRED:
1094 	{
1095 		struct ucred peercred;
1096 		if (len > sizeof(peercred))
1097 			len = sizeof(peercred);
1098 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1099 		if (copy_to_user(optval, &peercred, len))
1100 			return -EFAULT;
1101 		goto lenout;
1102 	}
1103 
1104 	case SO_PEERNAME:
1105 	{
1106 		char address[128];
1107 
1108 		if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1109 			return -ENOTCONN;
1110 		if (lv < len)
1111 			return -EINVAL;
1112 		if (copy_to_user(optval, address, len))
1113 			return -EFAULT;
1114 		goto lenout;
1115 	}
1116 
1117 	/* Dubious BSD thing... Probably nobody even uses it, but
1118 	 * the UNIX standard wants it for whatever reason... -DaveM
1119 	 */
1120 	case SO_ACCEPTCONN:
1121 		v.val = sk->sk_state == TCP_LISTEN;
1122 		break;
1123 
1124 	case SO_PASSSEC:
1125 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1126 		break;
1127 
1128 	case SO_PEERSEC:
1129 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1130 
1131 	case SO_MARK:
1132 		v.val = sk->sk_mark;
1133 		break;
1134 
1135 	case SO_RXQ_OVFL:
1136 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1137 		break;
1138 
1139 	case SO_WIFI_STATUS:
1140 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1141 		break;
1142 
1143 	case SO_PEEK_OFF:
1144 		if (!sock->ops->set_peek_off)
1145 			return -EOPNOTSUPP;
1146 
1147 		v.val = sk->sk_peek_off;
1148 		break;
1149 	case SO_NOFCS:
1150 		v.val = sock_flag(sk, SOCK_NOFCS);
1151 		break;
1152 
1153 	case SO_BINDTODEVICE:
1154 		return sock_getbindtodevice(sk, optval, optlen, len);
1155 
1156 	case SO_GET_FILTER:
1157 		len = sk_get_filter(sk, (struct sock_filter __user *)optval, len);
1158 		if (len < 0)
1159 			return len;
1160 
1161 		goto lenout;
1162 
1163 	case SO_LOCK_FILTER:
1164 		v.val = sock_flag(sk, SOCK_FILTER_LOCKED);
1165 		break;
1166 
1167 	case SO_SELECT_ERR_QUEUE:
1168 		v.val = sock_flag(sk, SOCK_SELECT_ERR_QUEUE);
1169 		break;
1170 
1171 	default:
1172 		return -ENOPROTOOPT;
1173 	}
1174 
1175 	if (len > lv)
1176 		len = lv;
1177 	if (copy_to_user(optval, &v, len))
1178 		return -EFAULT;
1179 lenout:
1180 	if (put_user(len, optlen))
1181 		return -EFAULT;
1182 	return 0;
1183 }
1184 
1185 /*
1186  * Initialize an sk_lock.
1187  *
1188  * (We also register the sk_lock with the lock validator.)
1189  */
1190 static inline void sock_lock_init(struct sock *sk)
1191 {
1192 	sock_lock_init_class_and_name(sk,
1193 			af_family_slock_key_strings[sk->sk_family],
1194 			af_family_slock_keys + sk->sk_family,
1195 			af_family_key_strings[sk->sk_family],
1196 			af_family_keys + sk->sk_family);
1197 }
1198 
1199 /*
1200  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1201  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1202  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1203  */
1204 static void sock_copy(struct sock *nsk, const struct sock *osk)
1205 {
1206 #ifdef CONFIG_SECURITY_NETWORK
1207 	void *sptr = nsk->sk_security;
1208 #endif
1209 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1210 
1211 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1212 	       osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1213 
1214 #ifdef CONFIG_SECURITY_NETWORK
1215 	nsk->sk_security = sptr;
1216 	security_sk_clone(osk, nsk);
1217 #endif
1218 }
1219 
1220 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1221 {
1222 	unsigned long nulls1, nulls2;
1223 
1224 	nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1225 	nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1226 	if (nulls1 > nulls2)
1227 		swap(nulls1, nulls2);
1228 
1229 	if (nulls1 != 0)
1230 		memset((char *)sk, 0, nulls1);
1231 	memset((char *)sk + nulls1 + sizeof(void *), 0,
1232 	       nulls2 - nulls1 - sizeof(void *));
1233 	memset((char *)sk + nulls2 + sizeof(void *), 0,
1234 	       size - nulls2 - sizeof(void *));
1235 }
1236 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1237 
1238 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1239 		int family)
1240 {
1241 	struct sock *sk;
1242 	struct kmem_cache *slab;
1243 
1244 	slab = prot->slab;
1245 	if (slab != NULL) {
1246 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1247 		if (!sk)
1248 			return sk;
1249 		if (priority & __GFP_ZERO) {
1250 			if (prot->clear_sk)
1251 				prot->clear_sk(sk, prot->obj_size);
1252 			else
1253 				sk_prot_clear_nulls(sk, prot->obj_size);
1254 		}
1255 	} else
1256 		sk = kmalloc(prot->obj_size, priority);
1257 
1258 	if (sk != NULL) {
1259 		kmemcheck_annotate_bitfield(sk, flags);
1260 
1261 		if (security_sk_alloc(sk, family, priority))
1262 			goto out_free;
1263 
1264 		if (!try_module_get(prot->owner))
1265 			goto out_free_sec;
1266 		sk_tx_queue_clear(sk);
1267 	}
1268 
1269 	return sk;
1270 
1271 out_free_sec:
1272 	security_sk_free(sk);
1273 out_free:
1274 	if (slab != NULL)
1275 		kmem_cache_free(slab, sk);
1276 	else
1277 		kfree(sk);
1278 	return NULL;
1279 }
1280 
1281 static void sk_prot_free(struct proto *prot, struct sock *sk)
1282 {
1283 	struct kmem_cache *slab;
1284 	struct module *owner;
1285 
1286 	owner = prot->owner;
1287 	slab = prot->slab;
1288 
1289 	security_sk_free(sk);
1290 	if (slab != NULL)
1291 		kmem_cache_free(slab, sk);
1292 	else
1293 		kfree(sk);
1294 	module_put(owner);
1295 }
1296 
1297 #if IS_ENABLED(CONFIG_NET_CLS_CGROUP)
1298 void sock_update_classid(struct sock *sk)
1299 {
1300 	u32 classid;
1301 
1302 	classid = task_cls_classid(current);
1303 	if (classid != sk->sk_classid)
1304 		sk->sk_classid = classid;
1305 }
1306 EXPORT_SYMBOL(sock_update_classid);
1307 #endif
1308 
1309 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
1310 void sock_update_netprioidx(struct sock *sk)
1311 {
1312 	if (in_interrupt())
1313 		return;
1314 
1315 	sk->sk_cgrp_prioidx = task_netprioidx(current);
1316 }
1317 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1318 #endif
1319 
1320 /**
1321  *	sk_alloc - All socket objects are allocated here
1322  *	@net: the applicable net namespace
1323  *	@family: protocol family
1324  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1325  *	@prot: struct proto associated with this new sock instance
1326  */
1327 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1328 		      struct proto *prot)
1329 {
1330 	struct sock *sk;
1331 
1332 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1333 	if (sk) {
1334 		sk->sk_family = family;
1335 		/*
1336 		 * See comment in struct sock definition to understand
1337 		 * why we need sk_prot_creator -acme
1338 		 */
1339 		sk->sk_prot = sk->sk_prot_creator = prot;
1340 		sock_lock_init(sk);
1341 		sock_net_set(sk, get_net(net));
1342 		atomic_set(&sk->sk_wmem_alloc, 1);
1343 
1344 		sock_update_classid(sk);
1345 		sock_update_netprioidx(sk);
1346 	}
1347 
1348 	return sk;
1349 }
1350 EXPORT_SYMBOL(sk_alloc);
1351 
1352 static void __sk_free(struct sock *sk)
1353 {
1354 	struct sk_filter *filter;
1355 
1356 	if (sk->sk_destruct)
1357 		sk->sk_destruct(sk);
1358 
1359 	filter = rcu_dereference_check(sk->sk_filter,
1360 				       atomic_read(&sk->sk_wmem_alloc) == 0);
1361 	if (filter) {
1362 		sk_filter_uncharge(sk, filter);
1363 		RCU_INIT_POINTER(sk->sk_filter, NULL);
1364 	}
1365 
1366 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1367 
1368 	if (atomic_read(&sk->sk_omem_alloc))
1369 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1370 			 __func__, atomic_read(&sk->sk_omem_alloc));
1371 
1372 	if (sk->sk_peer_cred)
1373 		put_cred(sk->sk_peer_cred);
1374 	put_pid(sk->sk_peer_pid);
1375 	put_net(sock_net(sk));
1376 	sk_prot_free(sk->sk_prot_creator, sk);
1377 }
1378 
1379 void sk_free(struct sock *sk)
1380 {
1381 	/*
1382 	 * We subtract one from sk_wmem_alloc and can know if
1383 	 * some packets are still in some tx queue.
1384 	 * If not null, sock_wfree() will call __sk_free(sk) later
1385 	 */
1386 	if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1387 		__sk_free(sk);
1388 }
1389 EXPORT_SYMBOL(sk_free);
1390 
1391 /*
1392  * Last sock_put should drop reference to sk->sk_net. It has already
1393  * been dropped in sk_change_net. Taking reference to stopping namespace
1394  * is not an option.
1395  * Take reference to a socket to remove it from hash _alive_ and after that
1396  * destroy it in the context of init_net.
1397  */
1398 void sk_release_kernel(struct sock *sk)
1399 {
1400 	if (sk == NULL || sk->sk_socket == NULL)
1401 		return;
1402 
1403 	sock_hold(sk);
1404 	sock_release(sk->sk_socket);
1405 	release_net(sock_net(sk));
1406 	sock_net_set(sk, get_net(&init_net));
1407 	sock_put(sk);
1408 }
1409 EXPORT_SYMBOL(sk_release_kernel);
1410 
1411 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1412 {
1413 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1414 		sock_update_memcg(newsk);
1415 }
1416 
1417 /**
1418  *	sk_clone_lock - clone a socket, and lock its clone
1419  *	@sk: the socket to clone
1420  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1421  *
1422  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1423  */
1424 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1425 {
1426 	struct sock *newsk;
1427 
1428 	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1429 	if (newsk != NULL) {
1430 		struct sk_filter *filter;
1431 
1432 		sock_copy(newsk, sk);
1433 
1434 		/* SANITY */
1435 		get_net(sock_net(newsk));
1436 		sk_node_init(&newsk->sk_node);
1437 		sock_lock_init(newsk);
1438 		bh_lock_sock(newsk);
1439 		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1440 		newsk->sk_backlog.len = 0;
1441 
1442 		atomic_set(&newsk->sk_rmem_alloc, 0);
1443 		/*
1444 		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1445 		 */
1446 		atomic_set(&newsk->sk_wmem_alloc, 1);
1447 		atomic_set(&newsk->sk_omem_alloc, 0);
1448 		skb_queue_head_init(&newsk->sk_receive_queue);
1449 		skb_queue_head_init(&newsk->sk_write_queue);
1450 #ifdef CONFIG_NET_DMA
1451 		skb_queue_head_init(&newsk->sk_async_wait_queue);
1452 #endif
1453 
1454 		spin_lock_init(&newsk->sk_dst_lock);
1455 		rwlock_init(&newsk->sk_callback_lock);
1456 		lockdep_set_class_and_name(&newsk->sk_callback_lock,
1457 				af_callback_keys + newsk->sk_family,
1458 				af_family_clock_key_strings[newsk->sk_family]);
1459 
1460 		newsk->sk_dst_cache	= NULL;
1461 		newsk->sk_wmem_queued	= 0;
1462 		newsk->sk_forward_alloc = 0;
1463 		newsk->sk_send_head	= NULL;
1464 		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1465 
1466 		sock_reset_flag(newsk, SOCK_DONE);
1467 		skb_queue_head_init(&newsk->sk_error_queue);
1468 
1469 		filter = rcu_dereference_protected(newsk->sk_filter, 1);
1470 		if (filter != NULL)
1471 			sk_filter_charge(newsk, filter);
1472 
1473 		if (unlikely(xfrm_sk_clone_policy(newsk))) {
1474 			/* It is still raw copy of parent, so invalidate
1475 			 * destructor and make plain sk_free() */
1476 			newsk->sk_destruct = NULL;
1477 			bh_unlock_sock(newsk);
1478 			sk_free(newsk);
1479 			newsk = NULL;
1480 			goto out;
1481 		}
1482 
1483 		newsk->sk_err	   = 0;
1484 		newsk->sk_priority = 0;
1485 		/*
1486 		 * Before updating sk_refcnt, we must commit prior changes to memory
1487 		 * (Documentation/RCU/rculist_nulls.txt for details)
1488 		 */
1489 		smp_wmb();
1490 		atomic_set(&newsk->sk_refcnt, 2);
1491 
1492 		/*
1493 		 * Increment the counter in the same struct proto as the master
1494 		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1495 		 * is the same as sk->sk_prot->socks, as this field was copied
1496 		 * with memcpy).
1497 		 *
1498 		 * This _changes_ the previous behaviour, where
1499 		 * tcp_create_openreq_child always was incrementing the
1500 		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1501 		 * to be taken into account in all callers. -acme
1502 		 */
1503 		sk_refcnt_debug_inc(newsk);
1504 		sk_set_socket(newsk, NULL);
1505 		newsk->sk_wq = NULL;
1506 
1507 		sk_update_clone(sk, newsk);
1508 
1509 		if (newsk->sk_prot->sockets_allocated)
1510 			sk_sockets_allocated_inc(newsk);
1511 
1512 		if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1513 			net_enable_timestamp();
1514 	}
1515 out:
1516 	return newsk;
1517 }
1518 EXPORT_SYMBOL_GPL(sk_clone_lock);
1519 
1520 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1521 {
1522 	__sk_dst_set(sk, dst);
1523 	sk->sk_route_caps = dst->dev->features;
1524 	if (sk->sk_route_caps & NETIF_F_GSO)
1525 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1526 	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1527 	if (sk_can_gso(sk)) {
1528 		if (dst->header_len) {
1529 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1530 		} else {
1531 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1532 			sk->sk_gso_max_size = dst->dev->gso_max_size;
1533 			sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1534 		}
1535 	}
1536 }
1537 EXPORT_SYMBOL_GPL(sk_setup_caps);
1538 
1539 /*
1540  *	Simple resource managers for sockets.
1541  */
1542 
1543 
1544 /*
1545  * Write buffer destructor automatically called from kfree_skb.
1546  */
1547 void sock_wfree(struct sk_buff *skb)
1548 {
1549 	struct sock *sk = skb->sk;
1550 	unsigned int len = skb->truesize;
1551 
1552 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1553 		/*
1554 		 * Keep a reference on sk_wmem_alloc, this will be released
1555 		 * after sk_write_space() call
1556 		 */
1557 		atomic_sub(len - 1, &sk->sk_wmem_alloc);
1558 		sk->sk_write_space(sk);
1559 		len = 1;
1560 	}
1561 	/*
1562 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1563 	 * could not do because of in-flight packets
1564 	 */
1565 	if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1566 		__sk_free(sk);
1567 }
1568 EXPORT_SYMBOL(sock_wfree);
1569 
1570 /*
1571  * Read buffer destructor automatically called from kfree_skb.
1572  */
1573 void sock_rfree(struct sk_buff *skb)
1574 {
1575 	struct sock *sk = skb->sk;
1576 	unsigned int len = skb->truesize;
1577 
1578 	atomic_sub(len, &sk->sk_rmem_alloc);
1579 	sk_mem_uncharge(sk, len);
1580 }
1581 EXPORT_SYMBOL(sock_rfree);
1582 
1583 void sock_edemux(struct sk_buff *skb)
1584 {
1585 	struct sock *sk = skb->sk;
1586 
1587 #ifdef CONFIG_INET
1588 	if (sk->sk_state == TCP_TIME_WAIT)
1589 		inet_twsk_put(inet_twsk(sk));
1590 	else
1591 #endif
1592 		sock_put(sk);
1593 }
1594 EXPORT_SYMBOL(sock_edemux);
1595 
1596 kuid_t sock_i_uid(struct sock *sk)
1597 {
1598 	kuid_t uid;
1599 
1600 	read_lock_bh(&sk->sk_callback_lock);
1601 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : GLOBAL_ROOT_UID;
1602 	read_unlock_bh(&sk->sk_callback_lock);
1603 	return uid;
1604 }
1605 EXPORT_SYMBOL(sock_i_uid);
1606 
1607 unsigned long sock_i_ino(struct sock *sk)
1608 {
1609 	unsigned long ino;
1610 
1611 	read_lock_bh(&sk->sk_callback_lock);
1612 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1613 	read_unlock_bh(&sk->sk_callback_lock);
1614 	return ino;
1615 }
1616 EXPORT_SYMBOL(sock_i_ino);
1617 
1618 /*
1619  * Allocate a skb from the socket's send buffer.
1620  */
1621 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1622 			     gfp_t priority)
1623 {
1624 	if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1625 		struct sk_buff *skb = alloc_skb(size, priority);
1626 		if (skb) {
1627 			skb_set_owner_w(skb, sk);
1628 			return skb;
1629 		}
1630 	}
1631 	return NULL;
1632 }
1633 EXPORT_SYMBOL(sock_wmalloc);
1634 
1635 /*
1636  * Allocate a skb from the socket's receive buffer.
1637  */
1638 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1639 			     gfp_t priority)
1640 {
1641 	if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1642 		struct sk_buff *skb = alloc_skb(size, priority);
1643 		if (skb) {
1644 			skb_set_owner_r(skb, sk);
1645 			return skb;
1646 		}
1647 	}
1648 	return NULL;
1649 }
1650 
1651 /*
1652  * Allocate a memory block from the socket's option memory buffer.
1653  */
1654 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1655 {
1656 	if ((unsigned int)size <= sysctl_optmem_max &&
1657 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1658 		void *mem;
1659 		/* First do the add, to avoid the race if kmalloc
1660 		 * might sleep.
1661 		 */
1662 		atomic_add(size, &sk->sk_omem_alloc);
1663 		mem = kmalloc(size, priority);
1664 		if (mem)
1665 			return mem;
1666 		atomic_sub(size, &sk->sk_omem_alloc);
1667 	}
1668 	return NULL;
1669 }
1670 EXPORT_SYMBOL(sock_kmalloc);
1671 
1672 /*
1673  * Free an option memory block.
1674  */
1675 void sock_kfree_s(struct sock *sk, void *mem, int size)
1676 {
1677 	kfree(mem);
1678 	atomic_sub(size, &sk->sk_omem_alloc);
1679 }
1680 EXPORT_SYMBOL(sock_kfree_s);
1681 
1682 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1683    I think, these locks should be removed for datagram sockets.
1684  */
1685 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1686 {
1687 	DEFINE_WAIT(wait);
1688 
1689 	clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1690 	for (;;) {
1691 		if (!timeo)
1692 			break;
1693 		if (signal_pending(current))
1694 			break;
1695 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1696 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1697 		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1698 			break;
1699 		if (sk->sk_shutdown & SEND_SHUTDOWN)
1700 			break;
1701 		if (sk->sk_err)
1702 			break;
1703 		timeo = schedule_timeout(timeo);
1704 	}
1705 	finish_wait(sk_sleep(sk), &wait);
1706 	return timeo;
1707 }
1708 
1709 
1710 /*
1711  *	Generic send/receive buffer handlers
1712  */
1713 
1714 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1715 				     unsigned long data_len, int noblock,
1716 				     int *errcode)
1717 {
1718 	struct sk_buff *skb;
1719 	gfp_t gfp_mask;
1720 	long timeo;
1721 	int err;
1722 	int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1723 
1724 	err = -EMSGSIZE;
1725 	if (npages > MAX_SKB_FRAGS)
1726 		goto failure;
1727 
1728 	gfp_mask = sk->sk_allocation;
1729 	if (gfp_mask & __GFP_WAIT)
1730 		gfp_mask |= __GFP_REPEAT;
1731 
1732 	timeo = sock_sndtimeo(sk, noblock);
1733 	while (1) {
1734 		err = sock_error(sk);
1735 		if (err != 0)
1736 			goto failure;
1737 
1738 		err = -EPIPE;
1739 		if (sk->sk_shutdown & SEND_SHUTDOWN)
1740 			goto failure;
1741 
1742 		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1743 			skb = alloc_skb(header_len, gfp_mask);
1744 			if (skb) {
1745 				int i;
1746 
1747 				/* No pages, we're done... */
1748 				if (!data_len)
1749 					break;
1750 
1751 				skb->truesize += data_len;
1752 				skb_shinfo(skb)->nr_frags = npages;
1753 				for (i = 0; i < npages; i++) {
1754 					struct page *page;
1755 
1756 					page = alloc_pages(sk->sk_allocation, 0);
1757 					if (!page) {
1758 						err = -ENOBUFS;
1759 						skb_shinfo(skb)->nr_frags = i;
1760 						kfree_skb(skb);
1761 						goto failure;
1762 					}
1763 
1764 					__skb_fill_page_desc(skb, i,
1765 							page, 0,
1766 							(data_len >= PAGE_SIZE ?
1767 							 PAGE_SIZE :
1768 							 data_len));
1769 					data_len -= PAGE_SIZE;
1770 				}
1771 
1772 				/* Full success... */
1773 				break;
1774 			}
1775 			err = -ENOBUFS;
1776 			goto failure;
1777 		}
1778 		set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1779 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1780 		err = -EAGAIN;
1781 		if (!timeo)
1782 			goto failure;
1783 		if (signal_pending(current))
1784 			goto interrupted;
1785 		timeo = sock_wait_for_wmem(sk, timeo);
1786 	}
1787 
1788 	skb_set_owner_w(skb, sk);
1789 	return skb;
1790 
1791 interrupted:
1792 	err = sock_intr_errno(timeo);
1793 failure:
1794 	*errcode = err;
1795 	return NULL;
1796 }
1797 EXPORT_SYMBOL(sock_alloc_send_pskb);
1798 
1799 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1800 				    int noblock, int *errcode)
1801 {
1802 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1803 }
1804 EXPORT_SYMBOL(sock_alloc_send_skb);
1805 
1806 /* On 32bit arches, an skb frag is limited to 2^15 */
1807 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
1808 
1809 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag)
1810 {
1811 	int order;
1812 
1813 	if (pfrag->page) {
1814 		if (atomic_read(&pfrag->page->_count) == 1) {
1815 			pfrag->offset = 0;
1816 			return true;
1817 		}
1818 		if (pfrag->offset < pfrag->size)
1819 			return true;
1820 		put_page(pfrag->page);
1821 	}
1822 
1823 	/* We restrict high order allocations to users that can afford to wait */
1824 	order = (sk->sk_allocation & __GFP_WAIT) ? SKB_FRAG_PAGE_ORDER : 0;
1825 
1826 	do {
1827 		gfp_t gfp = sk->sk_allocation;
1828 
1829 		if (order)
1830 			gfp |= __GFP_COMP | __GFP_NOWARN;
1831 		pfrag->page = alloc_pages(gfp, order);
1832 		if (likely(pfrag->page)) {
1833 			pfrag->offset = 0;
1834 			pfrag->size = PAGE_SIZE << order;
1835 			return true;
1836 		}
1837 	} while (--order >= 0);
1838 
1839 	sk_enter_memory_pressure(sk);
1840 	sk_stream_moderate_sndbuf(sk);
1841 	return false;
1842 }
1843 EXPORT_SYMBOL(sk_page_frag_refill);
1844 
1845 static void __lock_sock(struct sock *sk)
1846 	__releases(&sk->sk_lock.slock)
1847 	__acquires(&sk->sk_lock.slock)
1848 {
1849 	DEFINE_WAIT(wait);
1850 
1851 	for (;;) {
1852 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1853 					TASK_UNINTERRUPTIBLE);
1854 		spin_unlock_bh(&sk->sk_lock.slock);
1855 		schedule();
1856 		spin_lock_bh(&sk->sk_lock.slock);
1857 		if (!sock_owned_by_user(sk))
1858 			break;
1859 	}
1860 	finish_wait(&sk->sk_lock.wq, &wait);
1861 }
1862 
1863 static void __release_sock(struct sock *sk)
1864 	__releases(&sk->sk_lock.slock)
1865 	__acquires(&sk->sk_lock.slock)
1866 {
1867 	struct sk_buff *skb = sk->sk_backlog.head;
1868 
1869 	do {
1870 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1871 		bh_unlock_sock(sk);
1872 
1873 		do {
1874 			struct sk_buff *next = skb->next;
1875 
1876 			prefetch(next);
1877 			WARN_ON_ONCE(skb_dst_is_noref(skb));
1878 			skb->next = NULL;
1879 			sk_backlog_rcv(sk, skb);
1880 
1881 			/*
1882 			 * We are in process context here with softirqs
1883 			 * disabled, use cond_resched_softirq() to preempt.
1884 			 * This is safe to do because we've taken the backlog
1885 			 * queue private:
1886 			 */
1887 			cond_resched_softirq();
1888 
1889 			skb = next;
1890 		} while (skb != NULL);
1891 
1892 		bh_lock_sock(sk);
1893 	} while ((skb = sk->sk_backlog.head) != NULL);
1894 
1895 	/*
1896 	 * Doing the zeroing here guarantee we can not loop forever
1897 	 * while a wild producer attempts to flood us.
1898 	 */
1899 	sk->sk_backlog.len = 0;
1900 }
1901 
1902 /**
1903  * sk_wait_data - wait for data to arrive at sk_receive_queue
1904  * @sk:    sock to wait on
1905  * @timeo: for how long
1906  *
1907  * Now socket state including sk->sk_err is changed only under lock,
1908  * hence we may omit checks after joining wait queue.
1909  * We check receive queue before schedule() only as optimization;
1910  * it is very likely that release_sock() added new data.
1911  */
1912 int sk_wait_data(struct sock *sk, long *timeo)
1913 {
1914 	int rc;
1915 	DEFINE_WAIT(wait);
1916 
1917 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1918 	set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1919 	rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1920 	clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1921 	finish_wait(sk_sleep(sk), &wait);
1922 	return rc;
1923 }
1924 EXPORT_SYMBOL(sk_wait_data);
1925 
1926 /**
1927  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1928  *	@sk: socket
1929  *	@size: memory size to allocate
1930  *	@kind: allocation type
1931  *
1932  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1933  *	rmem allocation. This function assumes that protocols which have
1934  *	memory_pressure use sk_wmem_queued as write buffer accounting.
1935  */
1936 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1937 {
1938 	struct proto *prot = sk->sk_prot;
1939 	int amt = sk_mem_pages(size);
1940 	long allocated;
1941 	int parent_status = UNDER_LIMIT;
1942 
1943 	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1944 
1945 	allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1946 
1947 	/* Under limit. */
1948 	if (parent_status == UNDER_LIMIT &&
1949 			allocated <= sk_prot_mem_limits(sk, 0)) {
1950 		sk_leave_memory_pressure(sk);
1951 		return 1;
1952 	}
1953 
1954 	/* Under pressure. (we or our parents) */
1955 	if ((parent_status > SOFT_LIMIT) ||
1956 			allocated > sk_prot_mem_limits(sk, 1))
1957 		sk_enter_memory_pressure(sk);
1958 
1959 	/* Over hard limit (we or our parents) */
1960 	if ((parent_status == OVER_LIMIT) ||
1961 			(allocated > sk_prot_mem_limits(sk, 2)))
1962 		goto suppress_allocation;
1963 
1964 	/* guarantee minimum buffer size under pressure */
1965 	if (kind == SK_MEM_RECV) {
1966 		if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1967 			return 1;
1968 
1969 	} else { /* SK_MEM_SEND */
1970 		if (sk->sk_type == SOCK_STREAM) {
1971 			if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1972 				return 1;
1973 		} else if (atomic_read(&sk->sk_wmem_alloc) <
1974 			   prot->sysctl_wmem[0])
1975 				return 1;
1976 	}
1977 
1978 	if (sk_has_memory_pressure(sk)) {
1979 		int alloc;
1980 
1981 		if (!sk_under_memory_pressure(sk))
1982 			return 1;
1983 		alloc = sk_sockets_allocated_read_positive(sk);
1984 		if (sk_prot_mem_limits(sk, 2) > alloc *
1985 		    sk_mem_pages(sk->sk_wmem_queued +
1986 				 atomic_read(&sk->sk_rmem_alloc) +
1987 				 sk->sk_forward_alloc))
1988 			return 1;
1989 	}
1990 
1991 suppress_allocation:
1992 
1993 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1994 		sk_stream_moderate_sndbuf(sk);
1995 
1996 		/* Fail only if socket is _under_ its sndbuf.
1997 		 * In this case we cannot block, so that we have to fail.
1998 		 */
1999 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
2000 			return 1;
2001 	}
2002 
2003 	trace_sock_exceed_buf_limit(sk, prot, allocated);
2004 
2005 	/* Alas. Undo changes. */
2006 	sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
2007 
2008 	sk_memory_allocated_sub(sk, amt);
2009 
2010 	return 0;
2011 }
2012 EXPORT_SYMBOL(__sk_mem_schedule);
2013 
2014 /**
2015  *	__sk_reclaim - reclaim memory_allocated
2016  *	@sk: socket
2017  */
2018 void __sk_mem_reclaim(struct sock *sk)
2019 {
2020 	sk_memory_allocated_sub(sk,
2021 				sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
2022 	sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
2023 
2024 	if (sk_under_memory_pressure(sk) &&
2025 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
2026 		sk_leave_memory_pressure(sk);
2027 }
2028 EXPORT_SYMBOL(__sk_mem_reclaim);
2029 
2030 
2031 /*
2032  * Set of default routines for initialising struct proto_ops when
2033  * the protocol does not support a particular function. In certain
2034  * cases where it makes no sense for a protocol to have a "do nothing"
2035  * function, some default processing is provided.
2036  */
2037 
2038 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
2039 {
2040 	return -EOPNOTSUPP;
2041 }
2042 EXPORT_SYMBOL(sock_no_bind);
2043 
2044 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
2045 		    int len, int flags)
2046 {
2047 	return -EOPNOTSUPP;
2048 }
2049 EXPORT_SYMBOL(sock_no_connect);
2050 
2051 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
2052 {
2053 	return -EOPNOTSUPP;
2054 }
2055 EXPORT_SYMBOL(sock_no_socketpair);
2056 
2057 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
2058 {
2059 	return -EOPNOTSUPP;
2060 }
2061 EXPORT_SYMBOL(sock_no_accept);
2062 
2063 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
2064 		    int *len, int peer)
2065 {
2066 	return -EOPNOTSUPP;
2067 }
2068 EXPORT_SYMBOL(sock_no_getname);
2069 
2070 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
2071 {
2072 	return 0;
2073 }
2074 EXPORT_SYMBOL(sock_no_poll);
2075 
2076 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
2077 {
2078 	return -EOPNOTSUPP;
2079 }
2080 EXPORT_SYMBOL(sock_no_ioctl);
2081 
2082 int sock_no_listen(struct socket *sock, int backlog)
2083 {
2084 	return -EOPNOTSUPP;
2085 }
2086 EXPORT_SYMBOL(sock_no_listen);
2087 
2088 int sock_no_shutdown(struct socket *sock, int how)
2089 {
2090 	return -EOPNOTSUPP;
2091 }
2092 EXPORT_SYMBOL(sock_no_shutdown);
2093 
2094 int sock_no_setsockopt(struct socket *sock, int level, int optname,
2095 		    char __user *optval, unsigned int optlen)
2096 {
2097 	return -EOPNOTSUPP;
2098 }
2099 EXPORT_SYMBOL(sock_no_setsockopt);
2100 
2101 int sock_no_getsockopt(struct socket *sock, int level, int optname,
2102 		    char __user *optval, int __user *optlen)
2103 {
2104 	return -EOPNOTSUPP;
2105 }
2106 EXPORT_SYMBOL(sock_no_getsockopt);
2107 
2108 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2109 		    size_t len)
2110 {
2111 	return -EOPNOTSUPP;
2112 }
2113 EXPORT_SYMBOL(sock_no_sendmsg);
2114 
2115 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2116 		    size_t len, int flags)
2117 {
2118 	return -EOPNOTSUPP;
2119 }
2120 EXPORT_SYMBOL(sock_no_recvmsg);
2121 
2122 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2123 {
2124 	/* Mirror missing mmap method error code */
2125 	return -ENODEV;
2126 }
2127 EXPORT_SYMBOL(sock_no_mmap);
2128 
2129 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2130 {
2131 	ssize_t res;
2132 	struct msghdr msg = {.msg_flags = flags};
2133 	struct kvec iov;
2134 	char *kaddr = kmap(page);
2135 	iov.iov_base = kaddr + offset;
2136 	iov.iov_len = size;
2137 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2138 	kunmap(page);
2139 	return res;
2140 }
2141 EXPORT_SYMBOL(sock_no_sendpage);
2142 
2143 /*
2144  *	Default Socket Callbacks
2145  */
2146 
2147 static void sock_def_wakeup(struct sock *sk)
2148 {
2149 	struct socket_wq *wq;
2150 
2151 	rcu_read_lock();
2152 	wq = rcu_dereference(sk->sk_wq);
2153 	if (wq_has_sleeper(wq))
2154 		wake_up_interruptible_all(&wq->wait);
2155 	rcu_read_unlock();
2156 }
2157 
2158 static void sock_def_error_report(struct sock *sk)
2159 {
2160 	struct socket_wq *wq;
2161 
2162 	rcu_read_lock();
2163 	wq = rcu_dereference(sk->sk_wq);
2164 	if (wq_has_sleeper(wq))
2165 		wake_up_interruptible_poll(&wq->wait, POLLERR);
2166 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2167 	rcu_read_unlock();
2168 }
2169 
2170 static void sock_def_readable(struct sock *sk, int len)
2171 {
2172 	struct socket_wq *wq;
2173 
2174 	rcu_read_lock();
2175 	wq = rcu_dereference(sk->sk_wq);
2176 	if (wq_has_sleeper(wq))
2177 		wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2178 						POLLRDNORM | POLLRDBAND);
2179 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2180 	rcu_read_unlock();
2181 }
2182 
2183 static void sock_def_write_space(struct sock *sk)
2184 {
2185 	struct socket_wq *wq;
2186 
2187 	rcu_read_lock();
2188 
2189 	/* Do not wake up a writer until he can make "significant"
2190 	 * progress.  --DaveM
2191 	 */
2192 	if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2193 		wq = rcu_dereference(sk->sk_wq);
2194 		if (wq_has_sleeper(wq))
2195 			wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2196 						POLLWRNORM | POLLWRBAND);
2197 
2198 		/* Should agree with poll, otherwise some programs break */
2199 		if (sock_writeable(sk))
2200 			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2201 	}
2202 
2203 	rcu_read_unlock();
2204 }
2205 
2206 static void sock_def_destruct(struct sock *sk)
2207 {
2208 	kfree(sk->sk_protinfo);
2209 }
2210 
2211 void sk_send_sigurg(struct sock *sk)
2212 {
2213 	if (sk->sk_socket && sk->sk_socket->file)
2214 		if (send_sigurg(&sk->sk_socket->file->f_owner))
2215 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2216 }
2217 EXPORT_SYMBOL(sk_send_sigurg);
2218 
2219 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2220 		    unsigned long expires)
2221 {
2222 	if (!mod_timer(timer, expires))
2223 		sock_hold(sk);
2224 }
2225 EXPORT_SYMBOL(sk_reset_timer);
2226 
2227 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2228 {
2229 	if (del_timer(timer))
2230 		__sock_put(sk);
2231 }
2232 EXPORT_SYMBOL(sk_stop_timer);
2233 
2234 void sock_init_data(struct socket *sock, struct sock *sk)
2235 {
2236 	skb_queue_head_init(&sk->sk_receive_queue);
2237 	skb_queue_head_init(&sk->sk_write_queue);
2238 	skb_queue_head_init(&sk->sk_error_queue);
2239 #ifdef CONFIG_NET_DMA
2240 	skb_queue_head_init(&sk->sk_async_wait_queue);
2241 #endif
2242 
2243 	sk->sk_send_head	=	NULL;
2244 
2245 	init_timer(&sk->sk_timer);
2246 
2247 	sk->sk_allocation	=	GFP_KERNEL;
2248 	sk->sk_rcvbuf		=	sysctl_rmem_default;
2249 	sk->sk_sndbuf		=	sysctl_wmem_default;
2250 	sk->sk_state		=	TCP_CLOSE;
2251 	sk_set_socket(sk, sock);
2252 
2253 	sock_set_flag(sk, SOCK_ZAPPED);
2254 
2255 	if (sock) {
2256 		sk->sk_type	=	sock->type;
2257 		sk->sk_wq	=	sock->wq;
2258 		sock->sk	=	sk;
2259 	} else
2260 		sk->sk_wq	=	NULL;
2261 
2262 	spin_lock_init(&sk->sk_dst_lock);
2263 	rwlock_init(&sk->sk_callback_lock);
2264 	lockdep_set_class_and_name(&sk->sk_callback_lock,
2265 			af_callback_keys + sk->sk_family,
2266 			af_family_clock_key_strings[sk->sk_family]);
2267 
2268 	sk->sk_state_change	=	sock_def_wakeup;
2269 	sk->sk_data_ready	=	sock_def_readable;
2270 	sk->sk_write_space	=	sock_def_write_space;
2271 	sk->sk_error_report	=	sock_def_error_report;
2272 	sk->sk_destruct		=	sock_def_destruct;
2273 
2274 	sk->sk_frag.page	=	NULL;
2275 	sk->sk_frag.offset	=	0;
2276 	sk->sk_peek_off		=	-1;
2277 
2278 	sk->sk_peer_pid 	=	NULL;
2279 	sk->sk_peer_cred	=	NULL;
2280 	sk->sk_write_pending	=	0;
2281 	sk->sk_rcvlowat		=	1;
2282 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
2283 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
2284 
2285 	sk->sk_stamp = ktime_set(-1L, 0);
2286 
2287 	/*
2288 	 * Before updating sk_refcnt, we must commit prior changes to memory
2289 	 * (Documentation/RCU/rculist_nulls.txt for details)
2290 	 */
2291 	smp_wmb();
2292 	atomic_set(&sk->sk_refcnt, 1);
2293 	atomic_set(&sk->sk_drops, 0);
2294 }
2295 EXPORT_SYMBOL(sock_init_data);
2296 
2297 void lock_sock_nested(struct sock *sk, int subclass)
2298 {
2299 	might_sleep();
2300 	spin_lock_bh(&sk->sk_lock.slock);
2301 	if (sk->sk_lock.owned)
2302 		__lock_sock(sk);
2303 	sk->sk_lock.owned = 1;
2304 	spin_unlock(&sk->sk_lock.slock);
2305 	/*
2306 	 * The sk_lock has mutex_lock() semantics here:
2307 	 */
2308 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2309 	local_bh_enable();
2310 }
2311 EXPORT_SYMBOL(lock_sock_nested);
2312 
2313 void release_sock(struct sock *sk)
2314 {
2315 	/*
2316 	 * The sk_lock has mutex_unlock() semantics:
2317 	 */
2318 	mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2319 
2320 	spin_lock_bh(&sk->sk_lock.slock);
2321 	if (sk->sk_backlog.tail)
2322 		__release_sock(sk);
2323 
2324 	if (sk->sk_prot->release_cb)
2325 		sk->sk_prot->release_cb(sk);
2326 
2327 	sk->sk_lock.owned = 0;
2328 	if (waitqueue_active(&sk->sk_lock.wq))
2329 		wake_up(&sk->sk_lock.wq);
2330 	spin_unlock_bh(&sk->sk_lock.slock);
2331 }
2332 EXPORT_SYMBOL(release_sock);
2333 
2334 /**
2335  * lock_sock_fast - fast version of lock_sock
2336  * @sk: socket
2337  *
2338  * This version should be used for very small section, where process wont block
2339  * return false if fast path is taken
2340  *   sk_lock.slock locked, owned = 0, BH disabled
2341  * return true if slow path is taken
2342  *   sk_lock.slock unlocked, owned = 1, BH enabled
2343  */
2344 bool lock_sock_fast(struct sock *sk)
2345 {
2346 	might_sleep();
2347 	spin_lock_bh(&sk->sk_lock.slock);
2348 
2349 	if (!sk->sk_lock.owned)
2350 		/*
2351 		 * Note : We must disable BH
2352 		 */
2353 		return false;
2354 
2355 	__lock_sock(sk);
2356 	sk->sk_lock.owned = 1;
2357 	spin_unlock(&sk->sk_lock.slock);
2358 	/*
2359 	 * The sk_lock has mutex_lock() semantics here:
2360 	 */
2361 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2362 	local_bh_enable();
2363 	return true;
2364 }
2365 EXPORT_SYMBOL(lock_sock_fast);
2366 
2367 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2368 {
2369 	struct timeval tv;
2370 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2371 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2372 	tv = ktime_to_timeval(sk->sk_stamp);
2373 	if (tv.tv_sec == -1)
2374 		return -ENOENT;
2375 	if (tv.tv_sec == 0) {
2376 		sk->sk_stamp = ktime_get_real();
2377 		tv = ktime_to_timeval(sk->sk_stamp);
2378 	}
2379 	return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2380 }
2381 EXPORT_SYMBOL(sock_get_timestamp);
2382 
2383 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2384 {
2385 	struct timespec ts;
2386 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2387 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2388 	ts = ktime_to_timespec(sk->sk_stamp);
2389 	if (ts.tv_sec == -1)
2390 		return -ENOENT;
2391 	if (ts.tv_sec == 0) {
2392 		sk->sk_stamp = ktime_get_real();
2393 		ts = ktime_to_timespec(sk->sk_stamp);
2394 	}
2395 	return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2396 }
2397 EXPORT_SYMBOL(sock_get_timestampns);
2398 
2399 void sock_enable_timestamp(struct sock *sk, int flag)
2400 {
2401 	if (!sock_flag(sk, flag)) {
2402 		unsigned long previous_flags = sk->sk_flags;
2403 
2404 		sock_set_flag(sk, flag);
2405 		/*
2406 		 * we just set one of the two flags which require net
2407 		 * time stamping, but time stamping might have been on
2408 		 * already because of the other one
2409 		 */
2410 		if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2411 			net_enable_timestamp();
2412 	}
2413 }
2414 
2415 /*
2416  *	Get a socket option on an socket.
2417  *
2418  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
2419  *	asynchronous errors should be reported by getsockopt. We assume
2420  *	this means if you specify SO_ERROR (otherwise whats the point of it).
2421  */
2422 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2423 			   char __user *optval, int __user *optlen)
2424 {
2425 	struct sock *sk = sock->sk;
2426 
2427 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2428 }
2429 EXPORT_SYMBOL(sock_common_getsockopt);
2430 
2431 #ifdef CONFIG_COMPAT
2432 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2433 				  char __user *optval, int __user *optlen)
2434 {
2435 	struct sock *sk = sock->sk;
2436 
2437 	if (sk->sk_prot->compat_getsockopt != NULL)
2438 		return sk->sk_prot->compat_getsockopt(sk, level, optname,
2439 						      optval, optlen);
2440 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2441 }
2442 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2443 #endif
2444 
2445 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2446 			struct msghdr *msg, size_t size, int flags)
2447 {
2448 	struct sock *sk = sock->sk;
2449 	int addr_len = 0;
2450 	int err;
2451 
2452 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2453 				   flags & ~MSG_DONTWAIT, &addr_len);
2454 	if (err >= 0)
2455 		msg->msg_namelen = addr_len;
2456 	return err;
2457 }
2458 EXPORT_SYMBOL(sock_common_recvmsg);
2459 
2460 /*
2461  *	Set socket options on an inet socket.
2462  */
2463 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2464 			   char __user *optval, unsigned int optlen)
2465 {
2466 	struct sock *sk = sock->sk;
2467 
2468 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2469 }
2470 EXPORT_SYMBOL(sock_common_setsockopt);
2471 
2472 #ifdef CONFIG_COMPAT
2473 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2474 				  char __user *optval, unsigned int optlen)
2475 {
2476 	struct sock *sk = sock->sk;
2477 
2478 	if (sk->sk_prot->compat_setsockopt != NULL)
2479 		return sk->sk_prot->compat_setsockopt(sk, level, optname,
2480 						      optval, optlen);
2481 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2482 }
2483 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2484 #endif
2485 
2486 void sk_common_release(struct sock *sk)
2487 {
2488 	if (sk->sk_prot->destroy)
2489 		sk->sk_prot->destroy(sk);
2490 
2491 	/*
2492 	 * Observation: when sock_common_release is called, processes have
2493 	 * no access to socket. But net still has.
2494 	 * Step one, detach it from networking:
2495 	 *
2496 	 * A. Remove from hash tables.
2497 	 */
2498 
2499 	sk->sk_prot->unhash(sk);
2500 
2501 	/*
2502 	 * In this point socket cannot receive new packets, but it is possible
2503 	 * that some packets are in flight because some CPU runs receiver and
2504 	 * did hash table lookup before we unhashed socket. They will achieve
2505 	 * receive queue and will be purged by socket destructor.
2506 	 *
2507 	 * Also we still have packets pending on receive queue and probably,
2508 	 * our own packets waiting in device queues. sock_destroy will drain
2509 	 * receive queue, but transmitted packets will delay socket destruction
2510 	 * until the last reference will be released.
2511 	 */
2512 
2513 	sock_orphan(sk);
2514 
2515 	xfrm_sk_free_policy(sk);
2516 
2517 	sk_refcnt_debug_release(sk);
2518 
2519 	if (sk->sk_frag.page) {
2520 		put_page(sk->sk_frag.page);
2521 		sk->sk_frag.page = NULL;
2522 	}
2523 
2524 	sock_put(sk);
2525 }
2526 EXPORT_SYMBOL(sk_common_release);
2527 
2528 #ifdef CONFIG_PROC_FS
2529 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
2530 struct prot_inuse {
2531 	int val[PROTO_INUSE_NR];
2532 };
2533 
2534 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2535 
2536 #ifdef CONFIG_NET_NS
2537 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2538 {
2539 	__this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2540 }
2541 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2542 
2543 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2544 {
2545 	int cpu, idx = prot->inuse_idx;
2546 	int res = 0;
2547 
2548 	for_each_possible_cpu(cpu)
2549 		res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2550 
2551 	return res >= 0 ? res : 0;
2552 }
2553 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2554 
2555 static int __net_init sock_inuse_init_net(struct net *net)
2556 {
2557 	net->core.inuse = alloc_percpu(struct prot_inuse);
2558 	return net->core.inuse ? 0 : -ENOMEM;
2559 }
2560 
2561 static void __net_exit sock_inuse_exit_net(struct net *net)
2562 {
2563 	free_percpu(net->core.inuse);
2564 }
2565 
2566 static struct pernet_operations net_inuse_ops = {
2567 	.init = sock_inuse_init_net,
2568 	.exit = sock_inuse_exit_net,
2569 };
2570 
2571 static __init int net_inuse_init(void)
2572 {
2573 	if (register_pernet_subsys(&net_inuse_ops))
2574 		panic("Cannot initialize net inuse counters");
2575 
2576 	return 0;
2577 }
2578 
2579 core_initcall(net_inuse_init);
2580 #else
2581 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2582 
2583 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2584 {
2585 	__this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2586 }
2587 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2588 
2589 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2590 {
2591 	int cpu, idx = prot->inuse_idx;
2592 	int res = 0;
2593 
2594 	for_each_possible_cpu(cpu)
2595 		res += per_cpu(prot_inuse, cpu).val[idx];
2596 
2597 	return res >= 0 ? res : 0;
2598 }
2599 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2600 #endif
2601 
2602 static void assign_proto_idx(struct proto *prot)
2603 {
2604 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2605 
2606 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2607 		pr_err("PROTO_INUSE_NR exhausted\n");
2608 		return;
2609 	}
2610 
2611 	set_bit(prot->inuse_idx, proto_inuse_idx);
2612 }
2613 
2614 static void release_proto_idx(struct proto *prot)
2615 {
2616 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2617 		clear_bit(prot->inuse_idx, proto_inuse_idx);
2618 }
2619 #else
2620 static inline void assign_proto_idx(struct proto *prot)
2621 {
2622 }
2623 
2624 static inline void release_proto_idx(struct proto *prot)
2625 {
2626 }
2627 #endif
2628 
2629 int proto_register(struct proto *prot, int alloc_slab)
2630 {
2631 	if (alloc_slab) {
2632 		prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2633 					SLAB_HWCACHE_ALIGN | prot->slab_flags,
2634 					NULL);
2635 
2636 		if (prot->slab == NULL) {
2637 			pr_crit("%s: Can't create sock SLAB cache!\n",
2638 				prot->name);
2639 			goto out;
2640 		}
2641 
2642 		if (prot->rsk_prot != NULL) {
2643 			prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2644 			if (prot->rsk_prot->slab_name == NULL)
2645 				goto out_free_sock_slab;
2646 
2647 			prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2648 								 prot->rsk_prot->obj_size, 0,
2649 								 SLAB_HWCACHE_ALIGN, NULL);
2650 
2651 			if (prot->rsk_prot->slab == NULL) {
2652 				pr_crit("%s: Can't create request sock SLAB cache!\n",
2653 					prot->name);
2654 				goto out_free_request_sock_slab_name;
2655 			}
2656 		}
2657 
2658 		if (prot->twsk_prot != NULL) {
2659 			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2660 
2661 			if (prot->twsk_prot->twsk_slab_name == NULL)
2662 				goto out_free_request_sock_slab;
2663 
2664 			prot->twsk_prot->twsk_slab =
2665 				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2666 						  prot->twsk_prot->twsk_obj_size,
2667 						  0,
2668 						  SLAB_HWCACHE_ALIGN |
2669 							prot->slab_flags,
2670 						  NULL);
2671 			if (prot->twsk_prot->twsk_slab == NULL)
2672 				goto out_free_timewait_sock_slab_name;
2673 		}
2674 	}
2675 
2676 	mutex_lock(&proto_list_mutex);
2677 	list_add(&prot->node, &proto_list);
2678 	assign_proto_idx(prot);
2679 	mutex_unlock(&proto_list_mutex);
2680 	return 0;
2681 
2682 out_free_timewait_sock_slab_name:
2683 	kfree(prot->twsk_prot->twsk_slab_name);
2684 out_free_request_sock_slab:
2685 	if (prot->rsk_prot && prot->rsk_prot->slab) {
2686 		kmem_cache_destroy(prot->rsk_prot->slab);
2687 		prot->rsk_prot->slab = NULL;
2688 	}
2689 out_free_request_sock_slab_name:
2690 	if (prot->rsk_prot)
2691 		kfree(prot->rsk_prot->slab_name);
2692 out_free_sock_slab:
2693 	kmem_cache_destroy(prot->slab);
2694 	prot->slab = NULL;
2695 out:
2696 	return -ENOBUFS;
2697 }
2698 EXPORT_SYMBOL(proto_register);
2699 
2700 void proto_unregister(struct proto *prot)
2701 {
2702 	mutex_lock(&proto_list_mutex);
2703 	release_proto_idx(prot);
2704 	list_del(&prot->node);
2705 	mutex_unlock(&proto_list_mutex);
2706 
2707 	if (prot->slab != NULL) {
2708 		kmem_cache_destroy(prot->slab);
2709 		prot->slab = NULL;
2710 	}
2711 
2712 	if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2713 		kmem_cache_destroy(prot->rsk_prot->slab);
2714 		kfree(prot->rsk_prot->slab_name);
2715 		prot->rsk_prot->slab = NULL;
2716 	}
2717 
2718 	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2719 		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2720 		kfree(prot->twsk_prot->twsk_slab_name);
2721 		prot->twsk_prot->twsk_slab = NULL;
2722 	}
2723 }
2724 EXPORT_SYMBOL(proto_unregister);
2725 
2726 #ifdef CONFIG_PROC_FS
2727 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2728 	__acquires(proto_list_mutex)
2729 {
2730 	mutex_lock(&proto_list_mutex);
2731 	return seq_list_start_head(&proto_list, *pos);
2732 }
2733 
2734 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2735 {
2736 	return seq_list_next(v, &proto_list, pos);
2737 }
2738 
2739 static void proto_seq_stop(struct seq_file *seq, void *v)
2740 	__releases(proto_list_mutex)
2741 {
2742 	mutex_unlock(&proto_list_mutex);
2743 }
2744 
2745 static char proto_method_implemented(const void *method)
2746 {
2747 	return method == NULL ? 'n' : 'y';
2748 }
2749 static long sock_prot_memory_allocated(struct proto *proto)
2750 {
2751 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2752 }
2753 
2754 static char *sock_prot_memory_pressure(struct proto *proto)
2755 {
2756 	return proto->memory_pressure != NULL ?
2757 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2758 }
2759 
2760 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2761 {
2762 
2763 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2764 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2765 		   proto->name,
2766 		   proto->obj_size,
2767 		   sock_prot_inuse_get(seq_file_net(seq), proto),
2768 		   sock_prot_memory_allocated(proto),
2769 		   sock_prot_memory_pressure(proto),
2770 		   proto->max_header,
2771 		   proto->slab == NULL ? "no" : "yes",
2772 		   module_name(proto->owner),
2773 		   proto_method_implemented(proto->close),
2774 		   proto_method_implemented(proto->connect),
2775 		   proto_method_implemented(proto->disconnect),
2776 		   proto_method_implemented(proto->accept),
2777 		   proto_method_implemented(proto->ioctl),
2778 		   proto_method_implemented(proto->init),
2779 		   proto_method_implemented(proto->destroy),
2780 		   proto_method_implemented(proto->shutdown),
2781 		   proto_method_implemented(proto->setsockopt),
2782 		   proto_method_implemented(proto->getsockopt),
2783 		   proto_method_implemented(proto->sendmsg),
2784 		   proto_method_implemented(proto->recvmsg),
2785 		   proto_method_implemented(proto->sendpage),
2786 		   proto_method_implemented(proto->bind),
2787 		   proto_method_implemented(proto->backlog_rcv),
2788 		   proto_method_implemented(proto->hash),
2789 		   proto_method_implemented(proto->unhash),
2790 		   proto_method_implemented(proto->get_port),
2791 		   proto_method_implemented(proto->enter_memory_pressure));
2792 }
2793 
2794 static int proto_seq_show(struct seq_file *seq, void *v)
2795 {
2796 	if (v == &proto_list)
2797 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2798 			   "protocol",
2799 			   "size",
2800 			   "sockets",
2801 			   "memory",
2802 			   "press",
2803 			   "maxhdr",
2804 			   "slab",
2805 			   "module",
2806 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2807 	else
2808 		proto_seq_printf(seq, list_entry(v, struct proto, node));
2809 	return 0;
2810 }
2811 
2812 static const struct seq_operations proto_seq_ops = {
2813 	.start  = proto_seq_start,
2814 	.next   = proto_seq_next,
2815 	.stop   = proto_seq_stop,
2816 	.show   = proto_seq_show,
2817 };
2818 
2819 static int proto_seq_open(struct inode *inode, struct file *file)
2820 {
2821 	return seq_open_net(inode, file, &proto_seq_ops,
2822 			    sizeof(struct seq_net_private));
2823 }
2824 
2825 static const struct file_operations proto_seq_fops = {
2826 	.owner		= THIS_MODULE,
2827 	.open		= proto_seq_open,
2828 	.read		= seq_read,
2829 	.llseek		= seq_lseek,
2830 	.release	= seq_release_net,
2831 };
2832 
2833 static __net_init int proto_init_net(struct net *net)
2834 {
2835 	if (!proc_create("protocols", S_IRUGO, net->proc_net, &proto_seq_fops))
2836 		return -ENOMEM;
2837 
2838 	return 0;
2839 }
2840 
2841 static __net_exit void proto_exit_net(struct net *net)
2842 {
2843 	remove_proc_entry("protocols", net->proc_net);
2844 }
2845 
2846 
2847 static __net_initdata struct pernet_operations proto_net_ops = {
2848 	.init = proto_init_net,
2849 	.exit = proto_exit_net,
2850 };
2851 
2852 static int __init proto_init(void)
2853 {
2854 	return register_pernet_subsys(&proto_net_ops);
2855 }
2856 
2857 subsys_initcall(proto_init);
2858 
2859 #endif /* PROC_FS */
2860