xref: /linux/net/core/sock.c (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
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 struct static_key memcg_socket_limit_enabled;
190 EXPORT_SYMBOL(memcg_socket_limit_enabled);
191 
192 /*
193  * Make lock validator output more readable. (we pre-construct these
194  * strings build-time, so that runtime initialization of socket
195  * locks is fast):
196  */
197 static const char *const af_family_key_strings[AF_MAX+1] = {
198   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
199   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
200   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
201   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
202   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
203   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
204   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
205   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
206   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
207   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
208   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
209   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
210   "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
211   "sk_lock-AF_NFC"   , "sk_lock-AF_MAX"
212 };
213 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
214   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
215   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
216   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
217   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
218   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
219   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
220   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
221   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
222   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
223   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
224   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
225   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
226   "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
227   "slock-AF_NFC"   , "slock-AF_MAX"
228 };
229 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
230   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
231   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
232   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
233   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
234   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
235   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
236   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
237   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
238   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
239   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
240   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
241   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
242   "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
243   "clock-AF_NFC"   , "clock-AF_MAX"
244 };
245 
246 /*
247  * sk_callback_lock locking rules are per-address-family,
248  * so split the lock classes by using a per-AF key:
249  */
250 static struct lock_class_key af_callback_keys[AF_MAX];
251 
252 /* Take into consideration the size of the struct sk_buff overhead in the
253  * determination of these values, since that is non-constant across
254  * platforms.  This makes socket queueing behavior and performance
255  * not depend upon such differences.
256  */
257 #define _SK_MEM_PACKETS		256
258 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
259 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
260 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
261 
262 /* Run time adjustable parameters. */
263 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
264 EXPORT_SYMBOL(sysctl_wmem_max);
265 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
266 EXPORT_SYMBOL(sysctl_rmem_max);
267 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
268 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
269 
270 /* Maximal space eaten by iovec or ancillary data plus some space */
271 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
272 EXPORT_SYMBOL(sysctl_optmem_max);
273 
274 struct static_key memalloc_socks = STATIC_KEY_INIT_FALSE;
275 EXPORT_SYMBOL_GPL(memalloc_socks);
276 
277 /**
278  * sk_set_memalloc - sets %SOCK_MEMALLOC
279  * @sk: socket to set it on
280  *
281  * Set %SOCK_MEMALLOC on a socket for access to emergency reserves.
282  * It's the responsibility of the admin to adjust min_free_kbytes
283  * to meet the requirements
284  */
285 void sk_set_memalloc(struct sock *sk)
286 {
287 	sock_set_flag(sk, SOCK_MEMALLOC);
288 	sk->sk_allocation |= __GFP_MEMALLOC;
289 	static_key_slow_inc(&memalloc_socks);
290 }
291 EXPORT_SYMBOL_GPL(sk_set_memalloc);
292 
293 void sk_clear_memalloc(struct sock *sk)
294 {
295 	sock_reset_flag(sk, SOCK_MEMALLOC);
296 	sk->sk_allocation &= ~__GFP_MEMALLOC;
297 	static_key_slow_dec(&memalloc_socks);
298 
299 	/*
300 	 * SOCK_MEMALLOC is allowed to ignore rmem limits to ensure forward
301 	 * progress of swapping. However, if SOCK_MEMALLOC is cleared while
302 	 * it has rmem allocations there is a risk that the user of the
303 	 * socket cannot make forward progress due to exceeding the rmem
304 	 * limits. By rights, sk_clear_memalloc() should only be called
305 	 * on sockets being torn down but warn and reset the accounting if
306 	 * that assumption breaks.
307 	 */
308 	if (WARN_ON(sk->sk_forward_alloc))
309 		sk_mem_reclaim(sk);
310 }
311 EXPORT_SYMBOL_GPL(sk_clear_memalloc);
312 
313 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
314 {
315 	int ret;
316 	unsigned long pflags = current->flags;
317 
318 	/* these should have been dropped before queueing */
319 	BUG_ON(!sock_flag(sk, SOCK_MEMALLOC));
320 
321 	current->flags |= PF_MEMALLOC;
322 	ret = sk->sk_backlog_rcv(sk, skb);
323 	tsk_restore_flags(current, pflags, PF_MEMALLOC);
324 
325 	return ret;
326 }
327 EXPORT_SYMBOL(__sk_backlog_rcv);
328 
329 #if defined(CONFIG_CGROUPS)
330 #if !defined(CONFIG_NET_CLS_CGROUP)
331 int net_cls_subsys_id = -1;
332 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
333 #endif
334 #if !defined(CONFIG_NETPRIO_CGROUP)
335 int net_prio_subsys_id = -1;
336 EXPORT_SYMBOL_GPL(net_prio_subsys_id);
337 #endif
338 #endif
339 
340 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
341 {
342 	struct timeval tv;
343 
344 	if (optlen < sizeof(tv))
345 		return -EINVAL;
346 	if (copy_from_user(&tv, optval, sizeof(tv)))
347 		return -EFAULT;
348 	if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
349 		return -EDOM;
350 
351 	if (tv.tv_sec < 0) {
352 		static int warned __read_mostly;
353 
354 		*timeo_p = 0;
355 		if (warned < 10 && net_ratelimit()) {
356 			warned++;
357 			pr_info("%s: `%s' (pid %d) tries to set negative timeout\n",
358 				__func__, current->comm, task_pid_nr(current));
359 		}
360 		return 0;
361 	}
362 	*timeo_p = MAX_SCHEDULE_TIMEOUT;
363 	if (tv.tv_sec == 0 && tv.tv_usec == 0)
364 		return 0;
365 	if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
366 		*timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
367 	return 0;
368 }
369 
370 static void sock_warn_obsolete_bsdism(const char *name)
371 {
372 	static int warned;
373 	static char warncomm[TASK_COMM_LEN];
374 	if (strcmp(warncomm, current->comm) && warned < 5) {
375 		strcpy(warncomm,  current->comm);
376 		pr_warn("process `%s' is using obsolete %s SO_BSDCOMPAT\n",
377 			warncomm, name);
378 		warned++;
379 	}
380 }
381 
382 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
383 
384 static void sock_disable_timestamp(struct sock *sk, unsigned long flags)
385 {
386 	if (sk->sk_flags & flags) {
387 		sk->sk_flags &= ~flags;
388 		if (!(sk->sk_flags & SK_FLAGS_TIMESTAMP))
389 			net_disable_timestamp();
390 	}
391 }
392 
393 
394 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
395 {
396 	int err;
397 	int skb_len;
398 	unsigned long flags;
399 	struct sk_buff_head *list = &sk->sk_receive_queue;
400 
401 	if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
402 		atomic_inc(&sk->sk_drops);
403 		trace_sock_rcvqueue_full(sk, skb);
404 		return -ENOMEM;
405 	}
406 
407 	err = sk_filter(sk, skb);
408 	if (err)
409 		return err;
410 
411 	if (!sk_rmem_schedule(sk, skb, skb->truesize)) {
412 		atomic_inc(&sk->sk_drops);
413 		return -ENOBUFS;
414 	}
415 
416 	skb->dev = NULL;
417 	skb_set_owner_r(skb, sk);
418 
419 	/* Cache the SKB length before we tack it onto the receive
420 	 * queue.  Once it is added it no longer belongs to us and
421 	 * may be freed by other threads of control pulling packets
422 	 * from the queue.
423 	 */
424 	skb_len = skb->len;
425 
426 	/* we escape from rcu protected region, make sure we dont leak
427 	 * a norefcounted dst
428 	 */
429 	skb_dst_force(skb);
430 
431 	spin_lock_irqsave(&list->lock, flags);
432 	skb->dropcount = atomic_read(&sk->sk_drops);
433 	__skb_queue_tail(list, skb);
434 	spin_unlock_irqrestore(&list->lock, flags);
435 
436 	if (!sock_flag(sk, SOCK_DEAD))
437 		sk->sk_data_ready(sk, skb_len);
438 	return 0;
439 }
440 EXPORT_SYMBOL(sock_queue_rcv_skb);
441 
442 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
443 {
444 	int rc = NET_RX_SUCCESS;
445 
446 	if (sk_filter(sk, skb))
447 		goto discard_and_relse;
448 
449 	skb->dev = NULL;
450 
451 	if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf)) {
452 		atomic_inc(&sk->sk_drops);
453 		goto discard_and_relse;
454 	}
455 	if (nested)
456 		bh_lock_sock_nested(sk);
457 	else
458 		bh_lock_sock(sk);
459 	if (!sock_owned_by_user(sk)) {
460 		/*
461 		 * trylock + unlock semantics:
462 		 */
463 		mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
464 
465 		rc = sk_backlog_rcv(sk, skb);
466 
467 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
468 	} else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
469 		bh_unlock_sock(sk);
470 		atomic_inc(&sk->sk_drops);
471 		goto discard_and_relse;
472 	}
473 
474 	bh_unlock_sock(sk);
475 out:
476 	sock_put(sk);
477 	return rc;
478 discard_and_relse:
479 	kfree_skb(skb);
480 	goto out;
481 }
482 EXPORT_SYMBOL(sk_receive_skb);
483 
484 void sk_reset_txq(struct sock *sk)
485 {
486 	sk_tx_queue_clear(sk);
487 }
488 EXPORT_SYMBOL(sk_reset_txq);
489 
490 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
491 {
492 	struct dst_entry *dst = __sk_dst_get(sk);
493 
494 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
495 		sk_tx_queue_clear(sk);
496 		RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
497 		dst_release(dst);
498 		return NULL;
499 	}
500 
501 	return dst;
502 }
503 EXPORT_SYMBOL(__sk_dst_check);
504 
505 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
506 {
507 	struct dst_entry *dst = sk_dst_get(sk);
508 
509 	if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
510 		sk_dst_reset(sk);
511 		dst_release(dst);
512 		return NULL;
513 	}
514 
515 	return dst;
516 }
517 EXPORT_SYMBOL(sk_dst_check);
518 
519 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
520 {
521 	int ret = -ENOPROTOOPT;
522 #ifdef CONFIG_NETDEVICES
523 	struct net *net = sock_net(sk);
524 	char devname[IFNAMSIZ];
525 	int index;
526 
527 	/* Sorry... */
528 	ret = -EPERM;
529 	if (!capable(CAP_NET_RAW))
530 		goto out;
531 
532 	ret = -EINVAL;
533 	if (optlen < 0)
534 		goto out;
535 
536 	/* Bind this socket to a particular device like "eth0",
537 	 * as specified in the passed interface name. If the
538 	 * name is "" or the option length is zero the socket
539 	 * is not bound.
540 	 */
541 	if (optlen > IFNAMSIZ - 1)
542 		optlen = IFNAMSIZ - 1;
543 	memset(devname, 0, sizeof(devname));
544 
545 	ret = -EFAULT;
546 	if (copy_from_user(devname, optval, optlen))
547 		goto out;
548 
549 	index = 0;
550 	if (devname[0] != '\0') {
551 		struct net_device *dev;
552 
553 		rcu_read_lock();
554 		dev = dev_get_by_name_rcu(net, devname);
555 		if (dev)
556 			index = dev->ifindex;
557 		rcu_read_unlock();
558 		ret = -ENODEV;
559 		if (!dev)
560 			goto out;
561 	}
562 
563 	lock_sock(sk);
564 	sk->sk_bound_dev_if = index;
565 	sk_dst_reset(sk);
566 	release_sock(sk);
567 
568 	ret = 0;
569 
570 out:
571 #endif
572 
573 	return ret;
574 }
575 
576 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
577 {
578 	if (valbool)
579 		sock_set_flag(sk, bit);
580 	else
581 		sock_reset_flag(sk, bit);
582 }
583 
584 /*
585  *	This is meant for all protocols to use and covers goings on
586  *	at the socket level. Everything here is generic.
587  */
588 
589 int sock_setsockopt(struct socket *sock, int level, int optname,
590 		    char __user *optval, unsigned int optlen)
591 {
592 	struct sock *sk = sock->sk;
593 	int val;
594 	int valbool;
595 	struct linger ling;
596 	int ret = 0;
597 
598 	/*
599 	 *	Options without arguments
600 	 */
601 
602 	if (optname == SO_BINDTODEVICE)
603 		return sock_bindtodevice(sk, optval, optlen);
604 
605 	if (optlen < sizeof(int))
606 		return -EINVAL;
607 
608 	if (get_user(val, (int __user *)optval))
609 		return -EFAULT;
610 
611 	valbool = val ? 1 : 0;
612 
613 	lock_sock(sk);
614 
615 	switch (optname) {
616 	case SO_DEBUG:
617 		if (val && !capable(CAP_NET_ADMIN))
618 			ret = -EACCES;
619 		else
620 			sock_valbool_flag(sk, SOCK_DBG, valbool);
621 		break;
622 	case SO_REUSEADDR:
623 		sk->sk_reuse = (valbool ? SK_CAN_REUSE : SK_NO_REUSE);
624 		break;
625 	case SO_TYPE:
626 	case SO_PROTOCOL:
627 	case SO_DOMAIN:
628 	case SO_ERROR:
629 		ret = -ENOPROTOOPT;
630 		break;
631 	case SO_DONTROUTE:
632 		sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
633 		break;
634 	case SO_BROADCAST:
635 		sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
636 		break;
637 	case SO_SNDBUF:
638 		/* Don't error on this BSD doesn't and if you think
639 		 * about it this is right. Otherwise apps have to
640 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
641 		 * are treated in BSD as hints
642 		 */
643 		val = min_t(u32, val, sysctl_wmem_max);
644 set_sndbuf:
645 		sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
646 		sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
647 		/* Wake up sending tasks if we upped the value. */
648 		sk->sk_write_space(sk);
649 		break;
650 
651 	case SO_SNDBUFFORCE:
652 		if (!capable(CAP_NET_ADMIN)) {
653 			ret = -EPERM;
654 			break;
655 		}
656 		goto set_sndbuf;
657 
658 	case SO_RCVBUF:
659 		/* Don't error on this BSD doesn't and if you think
660 		 * about it this is right. Otherwise apps have to
661 		 * play 'guess the biggest size' games. RCVBUF/SNDBUF
662 		 * are treated in BSD as hints
663 		 */
664 		val = min_t(u32, val, sysctl_rmem_max);
665 set_rcvbuf:
666 		sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
667 		/*
668 		 * We double it on the way in to account for
669 		 * "struct sk_buff" etc. overhead.   Applications
670 		 * assume that the SO_RCVBUF setting they make will
671 		 * allow that much actual data to be received on that
672 		 * socket.
673 		 *
674 		 * Applications are unaware that "struct sk_buff" and
675 		 * other overheads allocate from the receive buffer
676 		 * during socket buffer allocation.
677 		 *
678 		 * And after considering the possible alternatives,
679 		 * returning the value we actually used in getsockopt
680 		 * is the most desirable behavior.
681 		 */
682 		sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
683 		break;
684 
685 	case SO_RCVBUFFORCE:
686 		if (!capable(CAP_NET_ADMIN)) {
687 			ret = -EPERM;
688 			break;
689 		}
690 		goto set_rcvbuf;
691 
692 	case SO_KEEPALIVE:
693 #ifdef CONFIG_INET
694 		if (sk->sk_protocol == IPPROTO_TCP)
695 			tcp_set_keepalive(sk, valbool);
696 #endif
697 		sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
698 		break;
699 
700 	case SO_OOBINLINE:
701 		sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
702 		break;
703 
704 	case SO_NO_CHECK:
705 		sk->sk_no_check = valbool;
706 		break;
707 
708 	case SO_PRIORITY:
709 		if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
710 			sk->sk_priority = val;
711 		else
712 			ret = -EPERM;
713 		break;
714 
715 	case SO_LINGER:
716 		if (optlen < sizeof(ling)) {
717 			ret = -EINVAL;	/* 1003.1g */
718 			break;
719 		}
720 		if (copy_from_user(&ling, optval, sizeof(ling))) {
721 			ret = -EFAULT;
722 			break;
723 		}
724 		if (!ling.l_onoff)
725 			sock_reset_flag(sk, SOCK_LINGER);
726 		else {
727 #if (BITS_PER_LONG == 32)
728 			if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
729 				sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
730 			else
731 #endif
732 				sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
733 			sock_set_flag(sk, SOCK_LINGER);
734 		}
735 		break;
736 
737 	case SO_BSDCOMPAT:
738 		sock_warn_obsolete_bsdism("setsockopt");
739 		break;
740 
741 	case SO_PASSCRED:
742 		if (valbool)
743 			set_bit(SOCK_PASSCRED, &sock->flags);
744 		else
745 			clear_bit(SOCK_PASSCRED, &sock->flags);
746 		break;
747 
748 	case SO_TIMESTAMP:
749 	case SO_TIMESTAMPNS:
750 		if (valbool)  {
751 			if (optname == SO_TIMESTAMP)
752 				sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
753 			else
754 				sock_set_flag(sk, SOCK_RCVTSTAMPNS);
755 			sock_set_flag(sk, SOCK_RCVTSTAMP);
756 			sock_enable_timestamp(sk, SOCK_TIMESTAMP);
757 		} else {
758 			sock_reset_flag(sk, SOCK_RCVTSTAMP);
759 			sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
760 		}
761 		break;
762 
763 	case SO_TIMESTAMPING:
764 		if (val & ~SOF_TIMESTAMPING_MASK) {
765 			ret = -EINVAL;
766 			break;
767 		}
768 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
769 				  val & SOF_TIMESTAMPING_TX_HARDWARE);
770 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
771 				  val & SOF_TIMESTAMPING_TX_SOFTWARE);
772 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
773 				  val & SOF_TIMESTAMPING_RX_HARDWARE);
774 		if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
775 			sock_enable_timestamp(sk,
776 					      SOCK_TIMESTAMPING_RX_SOFTWARE);
777 		else
778 			sock_disable_timestamp(sk,
779 					       (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE));
780 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
781 				  val & SOF_TIMESTAMPING_SOFTWARE);
782 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
783 				  val & SOF_TIMESTAMPING_SYS_HARDWARE);
784 		sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
785 				  val & SOF_TIMESTAMPING_RAW_HARDWARE);
786 		break;
787 
788 	case SO_RCVLOWAT:
789 		if (val < 0)
790 			val = INT_MAX;
791 		sk->sk_rcvlowat = val ? : 1;
792 		break;
793 
794 	case SO_RCVTIMEO:
795 		ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
796 		break;
797 
798 	case SO_SNDTIMEO:
799 		ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
800 		break;
801 
802 	case SO_ATTACH_FILTER:
803 		ret = -EINVAL;
804 		if (optlen == sizeof(struct sock_fprog)) {
805 			struct sock_fprog fprog;
806 
807 			ret = -EFAULT;
808 			if (copy_from_user(&fprog, optval, sizeof(fprog)))
809 				break;
810 
811 			ret = sk_attach_filter(&fprog, sk);
812 		}
813 		break;
814 
815 	case SO_DETACH_FILTER:
816 		ret = sk_detach_filter(sk);
817 		break;
818 
819 	case SO_PASSSEC:
820 		if (valbool)
821 			set_bit(SOCK_PASSSEC, &sock->flags);
822 		else
823 			clear_bit(SOCK_PASSSEC, &sock->flags);
824 		break;
825 	case SO_MARK:
826 		if (!capable(CAP_NET_ADMIN))
827 			ret = -EPERM;
828 		else
829 			sk->sk_mark = val;
830 		break;
831 
832 		/* We implement the SO_SNDLOWAT etc to
833 		   not be settable (1003.1g 5.3) */
834 	case SO_RXQ_OVFL:
835 		sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
836 		break;
837 
838 	case SO_WIFI_STATUS:
839 		sock_valbool_flag(sk, SOCK_WIFI_STATUS, valbool);
840 		break;
841 
842 	case SO_PEEK_OFF:
843 		if (sock->ops->set_peek_off)
844 			sock->ops->set_peek_off(sk, val);
845 		else
846 			ret = -EOPNOTSUPP;
847 		break;
848 
849 	case SO_NOFCS:
850 		sock_valbool_flag(sk, SOCK_NOFCS, valbool);
851 		break;
852 
853 	default:
854 		ret = -ENOPROTOOPT;
855 		break;
856 	}
857 	release_sock(sk);
858 	return ret;
859 }
860 EXPORT_SYMBOL(sock_setsockopt);
861 
862 
863 void cred_to_ucred(struct pid *pid, const struct cred *cred,
864 		   struct ucred *ucred)
865 {
866 	ucred->pid = pid_vnr(pid);
867 	ucred->uid = ucred->gid = -1;
868 	if (cred) {
869 		struct user_namespace *current_ns = current_user_ns();
870 
871 		ucred->uid = from_kuid(current_ns, cred->euid);
872 		ucred->gid = from_kgid(current_ns, cred->egid);
873 	}
874 }
875 EXPORT_SYMBOL_GPL(cred_to_ucred);
876 
877 int sock_getsockopt(struct socket *sock, int level, int optname,
878 		    char __user *optval, int __user *optlen)
879 {
880 	struct sock *sk = sock->sk;
881 
882 	union {
883 		int val;
884 		struct linger ling;
885 		struct timeval tm;
886 	} v;
887 
888 	int lv = sizeof(int);
889 	int len;
890 
891 	if (get_user(len, optlen))
892 		return -EFAULT;
893 	if (len < 0)
894 		return -EINVAL;
895 
896 	memset(&v, 0, sizeof(v));
897 
898 	switch (optname) {
899 	case SO_DEBUG:
900 		v.val = sock_flag(sk, SOCK_DBG);
901 		break;
902 
903 	case SO_DONTROUTE:
904 		v.val = sock_flag(sk, SOCK_LOCALROUTE);
905 		break;
906 
907 	case SO_BROADCAST:
908 		v.val = sock_flag(sk, SOCK_BROADCAST);
909 		break;
910 
911 	case SO_SNDBUF:
912 		v.val = sk->sk_sndbuf;
913 		break;
914 
915 	case SO_RCVBUF:
916 		v.val = sk->sk_rcvbuf;
917 		break;
918 
919 	case SO_REUSEADDR:
920 		v.val = sk->sk_reuse;
921 		break;
922 
923 	case SO_KEEPALIVE:
924 		v.val = sock_flag(sk, SOCK_KEEPOPEN);
925 		break;
926 
927 	case SO_TYPE:
928 		v.val = sk->sk_type;
929 		break;
930 
931 	case SO_PROTOCOL:
932 		v.val = sk->sk_protocol;
933 		break;
934 
935 	case SO_DOMAIN:
936 		v.val = sk->sk_family;
937 		break;
938 
939 	case SO_ERROR:
940 		v.val = -sock_error(sk);
941 		if (v.val == 0)
942 			v.val = xchg(&sk->sk_err_soft, 0);
943 		break;
944 
945 	case SO_OOBINLINE:
946 		v.val = sock_flag(sk, SOCK_URGINLINE);
947 		break;
948 
949 	case SO_NO_CHECK:
950 		v.val = sk->sk_no_check;
951 		break;
952 
953 	case SO_PRIORITY:
954 		v.val = sk->sk_priority;
955 		break;
956 
957 	case SO_LINGER:
958 		lv		= sizeof(v.ling);
959 		v.ling.l_onoff	= sock_flag(sk, SOCK_LINGER);
960 		v.ling.l_linger	= sk->sk_lingertime / HZ;
961 		break;
962 
963 	case SO_BSDCOMPAT:
964 		sock_warn_obsolete_bsdism("getsockopt");
965 		break;
966 
967 	case SO_TIMESTAMP:
968 		v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
969 				!sock_flag(sk, SOCK_RCVTSTAMPNS);
970 		break;
971 
972 	case SO_TIMESTAMPNS:
973 		v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
974 		break;
975 
976 	case SO_TIMESTAMPING:
977 		v.val = 0;
978 		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
979 			v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
980 		if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
981 			v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
982 		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
983 			v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
984 		if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
985 			v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
986 		if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
987 			v.val |= SOF_TIMESTAMPING_SOFTWARE;
988 		if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
989 			v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
990 		if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
991 			v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
992 		break;
993 
994 	case SO_RCVTIMEO:
995 		lv = sizeof(struct timeval);
996 		if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
997 			v.tm.tv_sec = 0;
998 			v.tm.tv_usec = 0;
999 		} else {
1000 			v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
1001 			v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
1002 		}
1003 		break;
1004 
1005 	case SO_SNDTIMEO:
1006 		lv = sizeof(struct timeval);
1007 		if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
1008 			v.tm.tv_sec = 0;
1009 			v.tm.tv_usec = 0;
1010 		} else {
1011 			v.tm.tv_sec = sk->sk_sndtimeo / HZ;
1012 			v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
1013 		}
1014 		break;
1015 
1016 	case SO_RCVLOWAT:
1017 		v.val = sk->sk_rcvlowat;
1018 		break;
1019 
1020 	case SO_SNDLOWAT:
1021 		v.val = 1;
1022 		break;
1023 
1024 	case SO_PASSCRED:
1025 		v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
1026 		break;
1027 
1028 	case SO_PEERCRED:
1029 	{
1030 		struct ucred peercred;
1031 		if (len > sizeof(peercred))
1032 			len = sizeof(peercred);
1033 		cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
1034 		if (copy_to_user(optval, &peercred, len))
1035 			return -EFAULT;
1036 		goto lenout;
1037 	}
1038 
1039 	case SO_PEERNAME:
1040 	{
1041 		char address[128];
1042 
1043 		if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
1044 			return -ENOTCONN;
1045 		if (lv < len)
1046 			return -EINVAL;
1047 		if (copy_to_user(optval, address, len))
1048 			return -EFAULT;
1049 		goto lenout;
1050 	}
1051 
1052 	/* Dubious BSD thing... Probably nobody even uses it, but
1053 	 * the UNIX standard wants it for whatever reason... -DaveM
1054 	 */
1055 	case SO_ACCEPTCONN:
1056 		v.val = sk->sk_state == TCP_LISTEN;
1057 		break;
1058 
1059 	case SO_PASSSEC:
1060 		v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
1061 		break;
1062 
1063 	case SO_PEERSEC:
1064 		return security_socket_getpeersec_stream(sock, optval, optlen, len);
1065 
1066 	case SO_MARK:
1067 		v.val = sk->sk_mark;
1068 		break;
1069 
1070 	case SO_RXQ_OVFL:
1071 		v.val = sock_flag(sk, SOCK_RXQ_OVFL);
1072 		break;
1073 
1074 	case SO_WIFI_STATUS:
1075 		v.val = sock_flag(sk, SOCK_WIFI_STATUS);
1076 		break;
1077 
1078 	case SO_PEEK_OFF:
1079 		if (!sock->ops->set_peek_off)
1080 			return -EOPNOTSUPP;
1081 
1082 		v.val = sk->sk_peek_off;
1083 		break;
1084 	case SO_NOFCS:
1085 		v.val = sock_flag(sk, SOCK_NOFCS);
1086 		break;
1087 	default:
1088 		return -ENOPROTOOPT;
1089 	}
1090 
1091 	if (len > lv)
1092 		len = lv;
1093 	if (copy_to_user(optval, &v, len))
1094 		return -EFAULT;
1095 lenout:
1096 	if (put_user(len, optlen))
1097 		return -EFAULT;
1098 	return 0;
1099 }
1100 
1101 /*
1102  * Initialize an sk_lock.
1103  *
1104  * (We also register the sk_lock with the lock validator.)
1105  */
1106 static inline void sock_lock_init(struct sock *sk)
1107 {
1108 	sock_lock_init_class_and_name(sk,
1109 			af_family_slock_key_strings[sk->sk_family],
1110 			af_family_slock_keys + sk->sk_family,
1111 			af_family_key_strings[sk->sk_family],
1112 			af_family_keys + sk->sk_family);
1113 }
1114 
1115 /*
1116  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1117  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1118  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1119  */
1120 static void sock_copy(struct sock *nsk, const struct sock *osk)
1121 {
1122 #ifdef CONFIG_SECURITY_NETWORK
1123 	void *sptr = nsk->sk_security;
1124 #endif
1125 	memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1126 
1127 	memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1128 	       osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1129 
1130 #ifdef CONFIG_SECURITY_NETWORK
1131 	nsk->sk_security = sptr;
1132 	security_sk_clone(osk, nsk);
1133 #endif
1134 }
1135 
1136 /*
1137  * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
1138  * un-modified. Special care is taken when initializing object to zero.
1139  */
1140 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1141 {
1142 	if (offsetof(struct sock, sk_node.next) != 0)
1143 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1144 	memset(&sk->sk_node.pprev, 0,
1145 	       size - offsetof(struct sock, sk_node.pprev));
1146 }
1147 
1148 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1149 {
1150 	unsigned long nulls1, nulls2;
1151 
1152 	nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1153 	nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1154 	if (nulls1 > nulls2)
1155 		swap(nulls1, nulls2);
1156 
1157 	if (nulls1 != 0)
1158 		memset((char *)sk, 0, nulls1);
1159 	memset((char *)sk + nulls1 + sizeof(void *), 0,
1160 	       nulls2 - nulls1 - sizeof(void *));
1161 	memset((char *)sk + nulls2 + sizeof(void *), 0,
1162 	       size - nulls2 - sizeof(void *));
1163 }
1164 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1165 
1166 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1167 		int family)
1168 {
1169 	struct sock *sk;
1170 	struct kmem_cache *slab;
1171 
1172 	slab = prot->slab;
1173 	if (slab != NULL) {
1174 		sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1175 		if (!sk)
1176 			return sk;
1177 		if (priority & __GFP_ZERO) {
1178 			if (prot->clear_sk)
1179 				prot->clear_sk(sk, prot->obj_size);
1180 			else
1181 				sk_prot_clear_nulls(sk, prot->obj_size);
1182 		}
1183 	} else
1184 		sk = kmalloc(prot->obj_size, priority);
1185 
1186 	if (sk != NULL) {
1187 		kmemcheck_annotate_bitfield(sk, flags);
1188 
1189 		if (security_sk_alloc(sk, family, priority))
1190 			goto out_free;
1191 
1192 		if (!try_module_get(prot->owner))
1193 			goto out_free_sec;
1194 		sk_tx_queue_clear(sk);
1195 	}
1196 
1197 	return sk;
1198 
1199 out_free_sec:
1200 	security_sk_free(sk);
1201 out_free:
1202 	if (slab != NULL)
1203 		kmem_cache_free(slab, sk);
1204 	else
1205 		kfree(sk);
1206 	return NULL;
1207 }
1208 
1209 static void sk_prot_free(struct proto *prot, struct sock *sk)
1210 {
1211 	struct kmem_cache *slab;
1212 	struct module *owner;
1213 
1214 	owner = prot->owner;
1215 	slab = prot->slab;
1216 
1217 	security_sk_free(sk);
1218 	if (slab != NULL)
1219 		kmem_cache_free(slab, sk);
1220 	else
1221 		kfree(sk);
1222 	module_put(owner);
1223 }
1224 
1225 #ifdef CONFIG_CGROUPS
1226 void sock_update_classid(struct sock *sk)
1227 {
1228 	u32 classid;
1229 
1230 	rcu_read_lock();  /* doing current task, which cannot vanish. */
1231 	classid = task_cls_classid(current);
1232 	rcu_read_unlock();
1233 	if (classid && classid != sk->sk_classid)
1234 		sk->sk_classid = classid;
1235 }
1236 EXPORT_SYMBOL(sock_update_classid);
1237 
1238 void sock_update_netprioidx(struct sock *sk, struct task_struct *task)
1239 {
1240 	if (in_interrupt())
1241 		return;
1242 
1243 	sk->sk_cgrp_prioidx = task_netprioidx(task);
1244 }
1245 EXPORT_SYMBOL_GPL(sock_update_netprioidx);
1246 #endif
1247 
1248 /**
1249  *	sk_alloc - All socket objects are allocated here
1250  *	@net: the applicable net namespace
1251  *	@family: protocol family
1252  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1253  *	@prot: struct proto associated with this new sock instance
1254  */
1255 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1256 		      struct proto *prot)
1257 {
1258 	struct sock *sk;
1259 
1260 	sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1261 	if (sk) {
1262 		sk->sk_family = family;
1263 		/*
1264 		 * See comment in struct sock definition to understand
1265 		 * why we need sk_prot_creator -acme
1266 		 */
1267 		sk->sk_prot = sk->sk_prot_creator = prot;
1268 		sock_lock_init(sk);
1269 		sock_net_set(sk, get_net(net));
1270 		atomic_set(&sk->sk_wmem_alloc, 1);
1271 
1272 		sock_update_classid(sk);
1273 		sock_update_netprioidx(sk, current);
1274 	}
1275 
1276 	return sk;
1277 }
1278 EXPORT_SYMBOL(sk_alloc);
1279 
1280 static void __sk_free(struct sock *sk)
1281 {
1282 	struct sk_filter *filter;
1283 
1284 	if (sk->sk_destruct)
1285 		sk->sk_destruct(sk);
1286 
1287 	filter = rcu_dereference_check(sk->sk_filter,
1288 				       atomic_read(&sk->sk_wmem_alloc) == 0);
1289 	if (filter) {
1290 		sk_filter_uncharge(sk, filter);
1291 		RCU_INIT_POINTER(sk->sk_filter, NULL);
1292 	}
1293 
1294 	sock_disable_timestamp(sk, SK_FLAGS_TIMESTAMP);
1295 
1296 	if (atomic_read(&sk->sk_omem_alloc))
1297 		pr_debug("%s: optmem leakage (%d bytes) detected\n",
1298 			 __func__, atomic_read(&sk->sk_omem_alloc));
1299 
1300 	if (sk->sk_peer_cred)
1301 		put_cred(sk->sk_peer_cred);
1302 	put_pid(sk->sk_peer_pid);
1303 	put_net(sock_net(sk));
1304 	sk_prot_free(sk->sk_prot_creator, sk);
1305 }
1306 
1307 void sk_free(struct sock *sk)
1308 {
1309 	/*
1310 	 * We subtract one from sk_wmem_alloc and can know if
1311 	 * some packets are still in some tx queue.
1312 	 * If not null, sock_wfree() will call __sk_free(sk) later
1313 	 */
1314 	if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1315 		__sk_free(sk);
1316 }
1317 EXPORT_SYMBOL(sk_free);
1318 
1319 /*
1320  * Last sock_put should drop reference to sk->sk_net. It has already
1321  * been dropped in sk_change_net. Taking reference to stopping namespace
1322  * is not an option.
1323  * Take reference to a socket to remove it from hash _alive_ and after that
1324  * destroy it in the context of init_net.
1325  */
1326 void sk_release_kernel(struct sock *sk)
1327 {
1328 	if (sk == NULL || sk->sk_socket == NULL)
1329 		return;
1330 
1331 	sock_hold(sk);
1332 	sock_release(sk->sk_socket);
1333 	release_net(sock_net(sk));
1334 	sock_net_set(sk, get_net(&init_net));
1335 	sock_put(sk);
1336 }
1337 EXPORT_SYMBOL(sk_release_kernel);
1338 
1339 static void sk_update_clone(const struct sock *sk, struct sock *newsk)
1340 {
1341 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1342 		sock_update_memcg(newsk);
1343 }
1344 
1345 /**
1346  *	sk_clone_lock - clone a socket, and lock its clone
1347  *	@sk: the socket to clone
1348  *	@priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1349  *
1350  *	Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1351  */
1352 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority)
1353 {
1354 	struct sock *newsk;
1355 
1356 	newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1357 	if (newsk != NULL) {
1358 		struct sk_filter *filter;
1359 
1360 		sock_copy(newsk, sk);
1361 
1362 		/* SANITY */
1363 		get_net(sock_net(newsk));
1364 		sk_node_init(&newsk->sk_node);
1365 		sock_lock_init(newsk);
1366 		bh_lock_sock(newsk);
1367 		newsk->sk_backlog.head	= newsk->sk_backlog.tail = NULL;
1368 		newsk->sk_backlog.len = 0;
1369 
1370 		atomic_set(&newsk->sk_rmem_alloc, 0);
1371 		/*
1372 		 * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1373 		 */
1374 		atomic_set(&newsk->sk_wmem_alloc, 1);
1375 		atomic_set(&newsk->sk_omem_alloc, 0);
1376 		skb_queue_head_init(&newsk->sk_receive_queue);
1377 		skb_queue_head_init(&newsk->sk_write_queue);
1378 #ifdef CONFIG_NET_DMA
1379 		skb_queue_head_init(&newsk->sk_async_wait_queue);
1380 #endif
1381 
1382 		spin_lock_init(&newsk->sk_dst_lock);
1383 		rwlock_init(&newsk->sk_callback_lock);
1384 		lockdep_set_class_and_name(&newsk->sk_callback_lock,
1385 				af_callback_keys + newsk->sk_family,
1386 				af_family_clock_key_strings[newsk->sk_family]);
1387 
1388 		newsk->sk_dst_cache	= NULL;
1389 		newsk->sk_wmem_queued	= 0;
1390 		newsk->sk_forward_alloc = 0;
1391 		newsk->sk_send_head	= NULL;
1392 		newsk->sk_userlocks	= sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1393 
1394 		sock_reset_flag(newsk, SOCK_DONE);
1395 		skb_queue_head_init(&newsk->sk_error_queue);
1396 
1397 		filter = rcu_dereference_protected(newsk->sk_filter, 1);
1398 		if (filter != NULL)
1399 			sk_filter_charge(newsk, filter);
1400 
1401 		if (unlikely(xfrm_sk_clone_policy(newsk))) {
1402 			/* It is still raw copy of parent, so invalidate
1403 			 * destructor and make plain sk_free() */
1404 			newsk->sk_destruct = NULL;
1405 			bh_unlock_sock(newsk);
1406 			sk_free(newsk);
1407 			newsk = NULL;
1408 			goto out;
1409 		}
1410 
1411 		newsk->sk_err	   = 0;
1412 		newsk->sk_priority = 0;
1413 		/*
1414 		 * Before updating sk_refcnt, we must commit prior changes to memory
1415 		 * (Documentation/RCU/rculist_nulls.txt for details)
1416 		 */
1417 		smp_wmb();
1418 		atomic_set(&newsk->sk_refcnt, 2);
1419 
1420 		/*
1421 		 * Increment the counter in the same struct proto as the master
1422 		 * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1423 		 * is the same as sk->sk_prot->socks, as this field was copied
1424 		 * with memcpy).
1425 		 *
1426 		 * This _changes_ the previous behaviour, where
1427 		 * tcp_create_openreq_child always was incrementing the
1428 		 * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1429 		 * to be taken into account in all callers. -acme
1430 		 */
1431 		sk_refcnt_debug_inc(newsk);
1432 		sk_set_socket(newsk, NULL);
1433 		newsk->sk_wq = NULL;
1434 
1435 		sk_update_clone(sk, newsk);
1436 
1437 		if (newsk->sk_prot->sockets_allocated)
1438 			sk_sockets_allocated_inc(newsk);
1439 
1440 		if (newsk->sk_flags & SK_FLAGS_TIMESTAMP)
1441 			net_enable_timestamp();
1442 	}
1443 out:
1444 	return newsk;
1445 }
1446 EXPORT_SYMBOL_GPL(sk_clone_lock);
1447 
1448 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1449 {
1450 	__sk_dst_set(sk, dst);
1451 	sk->sk_route_caps = dst->dev->features;
1452 	if (sk->sk_route_caps & NETIF_F_GSO)
1453 		sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1454 	sk->sk_route_caps &= ~sk->sk_route_nocaps;
1455 	if (sk_can_gso(sk)) {
1456 		if (dst->header_len) {
1457 			sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1458 		} else {
1459 			sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1460 			sk->sk_gso_max_size = dst->dev->gso_max_size;
1461 		}
1462 	}
1463 }
1464 EXPORT_SYMBOL_GPL(sk_setup_caps);
1465 
1466 void __init sk_init(void)
1467 {
1468 	if (totalram_pages <= 4096) {
1469 		sysctl_wmem_max = 32767;
1470 		sysctl_rmem_max = 32767;
1471 		sysctl_wmem_default = 32767;
1472 		sysctl_rmem_default = 32767;
1473 	} else if (totalram_pages >= 131072) {
1474 		sysctl_wmem_max = 131071;
1475 		sysctl_rmem_max = 131071;
1476 	}
1477 }
1478 
1479 /*
1480  *	Simple resource managers for sockets.
1481  */
1482 
1483 
1484 /*
1485  * Write buffer destructor automatically called from kfree_skb.
1486  */
1487 void sock_wfree(struct sk_buff *skb)
1488 {
1489 	struct sock *sk = skb->sk;
1490 	unsigned int len = skb->truesize;
1491 
1492 	if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1493 		/*
1494 		 * Keep a reference on sk_wmem_alloc, this will be released
1495 		 * after sk_write_space() call
1496 		 */
1497 		atomic_sub(len - 1, &sk->sk_wmem_alloc);
1498 		sk->sk_write_space(sk);
1499 		len = 1;
1500 	}
1501 	/*
1502 	 * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1503 	 * could not do because of in-flight packets
1504 	 */
1505 	if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1506 		__sk_free(sk);
1507 }
1508 EXPORT_SYMBOL(sock_wfree);
1509 
1510 /*
1511  * Read buffer destructor automatically called from kfree_skb.
1512  */
1513 void sock_rfree(struct sk_buff *skb)
1514 {
1515 	struct sock *sk = skb->sk;
1516 	unsigned int len = skb->truesize;
1517 
1518 	atomic_sub(len, &sk->sk_rmem_alloc);
1519 	sk_mem_uncharge(sk, len);
1520 }
1521 EXPORT_SYMBOL(sock_rfree);
1522 
1523 void sock_edemux(struct sk_buff *skb)
1524 {
1525 	sock_put(skb->sk);
1526 }
1527 EXPORT_SYMBOL(sock_edemux);
1528 
1529 int sock_i_uid(struct sock *sk)
1530 {
1531 	int uid;
1532 
1533 	read_lock_bh(&sk->sk_callback_lock);
1534 	uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1535 	read_unlock_bh(&sk->sk_callback_lock);
1536 	return uid;
1537 }
1538 EXPORT_SYMBOL(sock_i_uid);
1539 
1540 unsigned long sock_i_ino(struct sock *sk)
1541 {
1542 	unsigned long ino;
1543 
1544 	read_lock_bh(&sk->sk_callback_lock);
1545 	ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1546 	read_unlock_bh(&sk->sk_callback_lock);
1547 	return ino;
1548 }
1549 EXPORT_SYMBOL(sock_i_ino);
1550 
1551 /*
1552  * Allocate a skb from the socket's send buffer.
1553  */
1554 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1555 			     gfp_t priority)
1556 {
1557 	if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1558 		struct sk_buff *skb = alloc_skb(size, priority);
1559 		if (skb) {
1560 			skb_set_owner_w(skb, sk);
1561 			return skb;
1562 		}
1563 	}
1564 	return NULL;
1565 }
1566 EXPORT_SYMBOL(sock_wmalloc);
1567 
1568 /*
1569  * Allocate a skb from the socket's receive buffer.
1570  */
1571 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1572 			     gfp_t priority)
1573 {
1574 	if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1575 		struct sk_buff *skb = alloc_skb(size, priority);
1576 		if (skb) {
1577 			skb_set_owner_r(skb, sk);
1578 			return skb;
1579 		}
1580 	}
1581 	return NULL;
1582 }
1583 
1584 /*
1585  * Allocate a memory block from the socket's option memory buffer.
1586  */
1587 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1588 {
1589 	if ((unsigned int)size <= sysctl_optmem_max &&
1590 	    atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1591 		void *mem;
1592 		/* First do the add, to avoid the race if kmalloc
1593 		 * might sleep.
1594 		 */
1595 		atomic_add(size, &sk->sk_omem_alloc);
1596 		mem = kmalloc(size, priority);
1597 		if (mem)
1598 			return mem;
1599 		atomic_sub(size, &sk->sk_omem_alloc);
1600 	}
1601 	return NULL;
1602 }
1603 EXPORT_SYMBOL(sock_kmalloc);
1604 
1605 /*
1606  * Free an option memory block.
1607  */
1608 void sock_kfree_s(struct sock *sk, void *mem, int size)
1609 {
1610 	kfree(mem);
1611 	atomic_sub(size, &sk->sk_omem_alloc);
1612 }
1613 EXPORT_SYMBOL(sock_kfree_s);
1614 
1615 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1616    I think, these locks should be removed for datagram sockets.
1617  */
1618 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1619 {
1620 	DEFINE_WAIT(wait);
1621 
1622 	clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1623 	for (;;) {
1624 		if (!timeo)
1625 			break;
1626 		if (signal_pending(current))
1627 			break;
1628 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1629 		prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1630 		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1631 			break;
1632 		if (sk->sk_shutdown & SEND_SHUTDOWN)
1633 			break;
1634 		if (sk->sk_err)
1635 			break;
1636 		timeo = schedule_timeout(timeo);
1637 	}
1638 	finish_wait(sk_sleep(sk), &wait);
1639 	return timeo;
1640 }
1641 
1642 
1643 /*
1644  *	Generic send/receive buffer handlers
1645  */
1646 
1647 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1648 				     unsigned long data_len, int noblock,
1649 				     int *errcode)
1650 {
1651 	struct sk_buff *skb;
1652 	gfp_t gfp_mask;
1653 	long timeo;
1654 	int err;
1655 	int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1656 
1657 	err = -EMSGSIZE;
1658 	if (npages > MAX_SKB_FRAGS)
1659 		goto failure;
1660 
1661 	gfp_mask = sk->sk_allocation;
1662 	if (gfp_mask & __GFP_WAIT)
1663 		gfp_mask |= __GFP_REPEAT;
1664 
1665 	timeo = sock_sndtimeo(sk, noblock);
1666 	while (1) {
1667 		err = sock_error(sk);
1668 		if (err != 0)
1669 			goto failure;
1670 
1671 		err = -EPIPE;
1672 		if (sk->sk_shutdown & SEND_SHUTDOWN)
1673 			goto failure;
1674 
1675 		if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1676 			skb = alloc_skb(header_len, gfp_mask);
1677 			if (skb) {
1678 				int i;
1679 
1680 				/* No pages, we're done... */
1681 				if (!data_len)
1682 					break;
1683 
1684 				skb->truesize += data_len;
1685 				skb_shinfo(skb)->nr_frags = npages;
1686 				for (i = 0; i < npages; i++) {
1687 					struct page *page;
1688 
1689 					page = alloc_pages(sk->sk_allocation, 0);
1690 					if (!page) {
1691 						err = -ENOBUFS;
1692 						skb_shinfo(skb)->nr_frags = i;
1693 						kfree_skb(skb);
1694 						goto failure;
1695 					}
1696 
1697 					__skb_fill_page_desc(skb, i,
1698 							page, 0,
1699 							(data_len >= PAGE_SIZE ?
1700 							 PAGE_SIZE :
1701 							 data_len));
1702 					data_len -= PAGE_SIZE;
1703 				}
1704 
1705 				/* Full success... */
1706 				break;
1707 			}
1708 			err = -ENOBUFS;
1709 			goto failure;
1710 		}
1711 		set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1712 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1713 		err = -EAGAIN;
1714 		if (!timeo)
1715 			goto failure;
1716 		if (signal_pending(current))
1717 			goto interrupted;
1718 		timeo = sock_wait_for_wmem(sk, timeo);
1719 	}
1720 
1721 	skb_set_owner_w(skb, sk);
1722 	return skb;
1723 
1724 interrupted:
1725 	err = sock_intr_errno(timeo);
1726 failure:
1727 	*errcode = err;
1728 	return NULL;
1729 }
1730 EXPORT_SYMBOL(sock_alloc_send_pskb);
1731 
1732 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1733 				    int noblock, int *errcode)
1734 {
1735 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1736 }
1737 EXPORT_SYMBOL(sock_alloc_send_skb);
1738 
1739 static void __lock_sock(struct sock *sk)
1740 	__releases(&sk->sk_lock.slock)
1741 	__acquires(&sk->sk_lock.slock)
1742 {
1743 	DEFINE_WAIT(wait);
1744 
1745 	for (;;) {
1746 		prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1747 					TASK_UNINTERRUPTIBLE);
1748 		spin_unlock_bh(&sk->sk_lock.slock);
1749 		schedule();
1750 		spin_lock_bh(&sk->sk_lock.slock);
1751 		if (!sock_owned_by_user(sk))
1752 			break;
1753 	}
1754 	finish_wait(&sk->sk_lock.wq, &wait);
1755 }
1756 
1757 static void __release_sock(struct sock *sk)
1758 	__releases(&sk->sk_lock.slock)
1759 	__acquires(&sk->sk_lock.slock)
1760 {
1761 	struct sk_buff *skb = sk->sk_backlog.head;
1762 
1763 	do {
1764 		sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1765 		bh_unlock_sock(sk);
1766 
1767 		do {
1768 			struct sk_buff *next = skb->next;
1769 
1770 			prefetch(next);
1771 			WARN_ON_ONCE(skb_dst_is_noref(skb));
1772 			skb->next = NULL;
1773 			sk_backlog_rcv(sk, skb);
1774 
1775 			/*
1776 			 * We are in process context here with softirqs
1777 			 * disabled, use cond_resched_softirq() to preempt.
1778 			 * This is safe to do because we've taken the backlog
1779 			 * queue private:
1780 			 */
1781 			cond_resched_softirq();
1782 
1783 			skb = next;
1784 		} while (skb != NULL);
1785 
1786 		bh_lock_sock(sk);
1787 	} while ((skb = sk->sk_backlog.head) != NULL);
1788 
1789 	/*
1790 	 * Doing the zeroing here guarantee we can not loop forever
1791 	 * while a wild producer attempts to flood us.
1792 	 */
1793 	sk->sk_backlog.len = 0;
1794 }
1795 
1796 /**
1797  * sk_wait_data - wait for data to arrive at sk_receive_queue
1798  * @sk:    sock to wait on
1799  * @timeo: for how long
1800  *
1801  * Now socket state including sk->sk_err is changed only under lock,
1802  * hence we may omit checks after joining wait queue.
1803  * We check receive queue before schedule() only as optimization;
1804  * it is very likely that release_sock() added new data.
1805  */
1806 int sk_wait_data(struct sock *sk, long *timeo)
1807 {
1808 	int rc;
1809 	DEFINE_WAIT(wait);
1810 
1811 	prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1812 	set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1813 	rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1814 	clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1815 	finish_wait(sk_sleep(sk), &wait);
1816 	return rc;
1817 }
1818 EXPORT_SYMBOL(sk_wait_data);
1819 
1820 /**
1821  *	__sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1822  *	@sk: socket
1823  *	@size: memory size to allocate
1824  *	@kind: allocation type
1825  *
1826  *	If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1827  *	rmem allocation. This function assumes that protocols which have
1828  *	memory_pressure use sk_wmem_queued as write buffer accounting.
1829  */
1830 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1831 {
1832 	struct proto *prot = sk->sk_prot;
1833 	int amt = sk_mem_pages(size);
1834 	long allocated;
1835 	int parent_status = UNDER_LIMIT;
1836 
1837 	sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1838 
1839 	allocated = sk_memory_allocated_add(sk, amt, &parent_status);
1840 
1841 	/* Under limit. */
1842 	if (parent_status == UNDER_LIMIT &&
1843 			allocated <= sk_prot_mem_limits(sk, 0)) {
1844 		sk_leave_memory_pressure(sk);
1845 		return 1;
1846 	}
1847 
1848 	/* Under pressure. (we or our parents) */
1849 	if ((parent_status > SOFT_LIMIT) ||
1850 			allocated > sk_prot_mem_limits(sk, 1))
1851 		sk_enter_memory_pressure(sk);
1852 
1853 	/* Over hard limit (we or our parents) */
1854 	if ((parent_status == OVER_LIMIT) ||
1855 			(allocated > sk_prot_mem_limits(sk, 2)))
1856 		goto suppress_allocation;
1857 
1858 	/* guarantee minimum buffer size under pressure */
1859 	if (kind == SK_MEM_RECV) {
1860 		if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1861 			return 1;
1862 
1863 	} else { /* SK_MEM_SEND */
1864 		if (sk->sk_type == SOCK_STREAM) {
1865 			if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1866 				return 1;
1867 		} else if (atomic_read(&sk->sk_wmem_alloc) <
1868 			   prot->sysctl_wmem[0])
1869 				return 1;
1870 	}
1871 
1872 	if (sk_has_memory_pressure(sk)) {
1873 		int alloc;
1874 
1875 		if (!sk_under_memory_pressure(sk))
1876 			return 1;
1877 		alloc = sk_sockets_allocated_read_positive(sk);
1878 		if (sk_prot_mem_limits(sk, 2) > alloc *
1879 		    sk_mem_pages(sk->sk_wmem_queued +
1880 				 atomic_read(&sk->sk_rmem_alloc) +
1881 				 sk->sk_forward_alloc))
1882 			return 1;
1883 	}
1884 
1885 suppress_allocation:
1886 
1887 	if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1888 		sk_stream_moderate_sndbuf(sk);
1889 
1890 		/* Fail only if socket is _under_ its sndbuf.
1891 		 * In this case we cannot block, so that we have to fail.
1892 		 */
1893 		if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1894 			return 1;
1895 	}
1896 
1897 	trace_sock_exceed_buf_limit(sk, prot, allocated);
1898 
1899 	/* Alas. Undo changes. */
1900 	sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1901 
1902 	sk_memory_allocated_sub(sk, amt);
1903 
1904 	return 0;
1905 }
1906 EXPORT_SYMBOL(__sk_mem_schedule);
1907 
1908 /**
1909  *	__sk_reclaim - reclaim memory_allocated
1910  *	@sk: socket
1911  */
1912 void __sk_mem_reclaim(struct sock *sk)
1913 {
1914 	sk_memory_allocated_sub(sk,
1915 				sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT);
1916 	sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1917 
1918 	if (sk_under_memory_pressure(sk) &&
1919 	    (sk_memory_allocated(sk) < sk_prot_mem_limits(sk, 0)))
1920 		sk_leave_memory_pressure(sk);
1921 }
1922 EXPORT_SYMBOL(__sk_mem_reclaim);
1923 
1924 
1925 /*
1926  * Set of default routines for initialising struct proto_ops when
1927  * the protocol does not support a particular function. In certain
1928  * cases where it makes no sense for a protocol to have a "do nothing"
1929  * function, some default processing is provided.
1930  */
1931 
1932 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1933 {
1934 	return -EOPNOTSUPP;
1935 }
1936 EXPORT_SYMBOL(sock_no_bind);
1937 
1938 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1939 		    int len, int flags)
1940 {
1941 	return -EOPNOTSUPP;
1942 }
1943 EXPORT_SYMBOL(sock_no_connect);
1944 
1945 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1946 {
1947 	return -EOPNOTSUPP;
1948 }
1949 EXPORT_SYMBOL(sock_no_socketpair);
1950 
1951 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1952 {
1953 	return -EOPNOTSUPP;
1954 }
1955 EXPORT_SYMBOL(sock_no_accept);
1956 
1957 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1958 		    int *len, int peer)
1959 {
1960 	return -EOPNOTSUPP;
1961 }
1962 EXPORT_SYMBOL(sock_no_getname);
1963 
1964 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1965 {
1966 	return 0;
1967 }
1968 EXPORT_SYMBOL(sock_no_poll);
1969 
1970 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1971 {
1972 	return -EOPNOTSUPP;
1973 }
1974 EXPORT_SYMBOL(sock_no_ioctl);
1975 
1976 int sock_no_listen(struct socket *sock, int backlog)
1977 {
1978 	return -EOPNOTSUPP;
1979 }
1980 EXPORT_SYMBOL(sock_no_listen);
1981 
1982 int sock_no_shutdown(struct socket *sock, int how)
1983 {
1984 	return -EOPNOTSUPP;
1985 }
1986 EXPORT_SYMBOL(sock_no_shutdown);
1987 
1988 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1989 		    char __user *optval, unsigned int optlen)
1990 {
1991 	return -EOPNOTSUPP;
1992 }
1993 EXPORT_SYMBOL(sock_no_setsockopt);
1994 
1995 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1996 		    char __user *optval, int __user *optlen)
1997 {
1998 	return -EOPNOTSUPP;
1999 }
2000 EXPORT_SYMBOL(sock_no_getsockopt);
2001 
2002 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2003 		    size_t len)
2004 {
2005 	return -EOPNOTSUPP;
2006 }
2007 EXPORT_SYMBOL(sock_no_sendmsg);
2008 
2009 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
2010 		    size_t len, int flags)
2011 {
2012 	return -EOPNOTSUPP;
2013 }
2014 EXPORT_SYMBOL(sock_no_recvmsg);
2015 
2016 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
2017 {
2018 	/* Mirror missing mmap method error code */
2019 	return -ENODEV;
2020 }
2021 EXPORT_SYMBOL(sock_no_mmap);
2022 
2023 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
2024 {
2025 	ssize_t res;
2026 	struct msghdr msg = {.msg_flags = flags};
2027 	struct kvec iov;
2028 	char *kaddr = kmap(page);
2029 	iov.iov_base = kaddr + offset;
2030 	iov.iov_len = size;
2031 	res = kernel_sendmsg(sock, &msg, &iov, 1, size);
2032 	kunmap(page);
2033 	return res;
2034 }
2035 EXPORT_SYMBOL(sock_no_sendpage);
2036 
2037 /*
2038  *	Default Socket Callbacks
2039  */
2040 
2041 static void sock_def_wakeup(struct sock *sk)
2042 {
2043 	struct socket_wq *wq;
2044 
2045 	rcu_read_lock();
2046 	wq = rcu_dereference(sk->sk_wq);
2047 	if (wq_has_sleeper(wq))
2048 		wake_up_interruptible_all(&wq->wait);
2049 	rcu_read_unlock();
2050 }
2051 
2052 static void sock_def_error_report(struct sock *sk)
2053 {
2054 	struct socket_wq *wq;
2055 
2056 	rcu_read_lock();
2057 	wq = rcu_dereference(sk->sk_wq);
2058 	if (wq_has_sleeper(wq))
2059 		wake_up_interruptible_poll(&wq->wait, POLLERR);
2060 	sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
2061 	rcu_read_unlock();
2062 }
2063 
2064 static void sock_def_readable(struct sock *sk, int len)
2065 {
2066 	struct socket_wq *wq;
2067 
2068 	rcu_read_lock();
2069 	wq = rcu_dereference(sk->sk_wq);
2070 	if (wq_has_sleeper(wq))
2071 		wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
2072 						POLLRDNORM | POLLRDBAND);
2073 	sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
2074 	rcu_read_unlock();
2075 }
2076 
2077 static void sock_def_write_space(struct sock *sk)
2078 {
2079 	struct socket_wq *wq;
2080 
2081 	rcu_read_lock();
2082 
2083 	/* Do not wake up a writer until he can make "significant"
2084 	 * progress.  --DaveM
2085 	 */
2086 	if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
2087 		wq = rcu_dereference(sk->sk_wq);
2088 		if (wq_has_sleeper(wq))
2089 			wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
2090 						POLLWRNORM | POLLWRBAND);
2091 
2092 		/* Should agree with poll, otherwise some programs break */
2093 		if (sock_writeable(sk))
2094 			sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
2095 	}
2096 
2097 	rcu_read_unlock();
2098 }
2099 
2100 static void sock_def_destruct(struct sock *sk)
2101 {
2102 	kfree(sk->sk_protinfo);
2103 }
2104 
2105 void sk_send_sigurg(struct sock *sk)
2106 {
2107 	if (sk->sk_socket && sk->sk_socket->file)
2108 		if (send_sigurg(&sk->sk_socket->file->f_owner))
2109 			sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
2110 }
2111 EXPORT_SYMBOL(sk_send_sigurg);
2112 
2113 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
2114 		    unsigned long expires)
2115 {
2116 	if (!mod_timer(timer, expires))
2117 		sock_hold(sk);
2118 }
2119 EXPORT_SYMBOL(sk_reset_timer);
2120 
2121 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
2122 {
2123 	if (timer_pending(timer) && del_timer(timer))
2124 		__sock_put(sk);
2125 }
2126 EXPORT_SYMBOL(sk_stop_timer);
2127 
2128 void sock_init_data(struct socket *sock, struct sock *sk)
2129 {
2130 	skb_queue_head_init(&sk->sk_receive_queue);
2131 	skb_queue_head_init(&sk->sk_write_queue);
2132 	skb_queue_head_init(&sk->sk_error_queue);
2133 #ifdef CONFIG_NET_DMA
2134 	skb_queue_head_init(&sk->sk_async_wait_queue);
2135 #endif
2136 
2137 	sk->sk_send_head	=	NULL;
2138 
2139 	init_timer(&sk->sk_timer);
2140 
2141 	sk->sk_allocation	=	GFP_KERNEL;
2142 	sk->sk_rcvbuf		=	sysctl_rmem_default;
2143 	sk->sk_sndbuf		=	sysctl_wmem_default;
2144 	sk->sk_state		=	TCP_CLOSE;
2145 	sk_set_socket(sk, sock);
2146 
2147 	sock_set_flag(sk, SOCK_ZAPPED);
2148 
2149 	if (sock) {
2150 		sk->sk_type	=	sock->type;
2151 		sk->sk_wq	=	sock->wq;
2152 		sock->sk	=	sk;
2153 	} else
2154 		sk->sk_wq	=	NULL;
2155 
2156 	spin_lock_init(&sk->sk_dst_lock);
2157 	rwlock_init(&sk->sk_callback_lock);
2158 	lockdep_set_class_and_name(&sk->sk_callback_lock,
2159 			af_callback_keys + sk->sk_family,
2160 			af_family_clock_key_strings[sk->sk_family]);
2161 
2162 	sk->sk_state_change	=	sock_def_wakeup;
2163 	sk->sk_data_ready	=	sock_def_readable;
2164 	sk->sk_write_space	=	sock_def_write_space;
2165 	sk->sk_error_report	=	sock_def_error_report;
2166 	sk->sk_destruct		=	sock_def_destruct;
2167 
2168 	sk->sk_sndmsg_page	=	NULL;
2169 	sk->sk_sndmsg_off	=	0;
2170 	sk->sk_peek_off		=	-1;
2171 
2172 	sk->sk_peer_pid 	=	NULL;
2173 	sk->sk_peer_cred	=	NULL;
2174 	sk->sk_write_pending	=	0;
2175 	sk->sk_rcvlowat		=	1;
2176 	sk->sk_rcvtimeo		=	MAX_SCHEDULE_TIMEOUT;
2177 	sk->sk_sndtimeo		=	MAX_SCHEDULE_TIMEOUT;
2178 
2179 	sk->sk_stamp = ktime_set(-1L, 0);
2180 
2181 	/*
2182 	 * Before updating sk_refcnt, we must commit prior changes to memory
2183 	 * (Documentation/RCU/rculist_nulls.txt for details)
2184 	 */
2185 	smp_wmb();
2186 	atomic_set(&sk->sk_refcnt, 1);
2187 	atomic_set(&sk->sk_drops, 0);
2188 }
2189 EXPORT_SYMBOL(sock_init_data);
2190 
2191 void lock_sock_nested(struct sock *sk, int subclass)
2192 {
2193 	might_sleep();
2194 	spin_lock_bh(&sk->sk_lock.slock);
2195 	if (sk->sk_lock.owned)
2196 		__lock_sock(sk);
2197 	sk->sk_lock.owned = 1;
2198 	spin_unlock(&sk->sk_lock.slock);
2199 	/*
2200 	 * The sk_lock has mutex_lock() semantics here:
2201 	 */
2202 	mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2203 	local_bh_enable();
2204 }
2205 EXPORT_SYMBOL(lock_sock_nested);
2206 
2207 void release_sock(struct sock *sk)
2208 {
2209 	/*
2210 	 * The sk_lock has mutex_unlock() semantics:
2211 	 */
2212 	mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2213 
2214 	spin_lock_bh(&sk->sk_lock.slock);
2215 	if (sk->sk_backlog.tail)
2216 		__release_sock(sk);
2217 
2218 	if (sk->sk_prot->release_cb)
2219 		sk->sk_prot->release_cb(sk);
2220 
2221 	sk->sk_lock.owned = 0;
2222 	if (waitqueue_active(&sk->sk_lock.wq))
2223 		wake_up(&sk->sk_lock.wq);
2224 	spin_unlock_bh(&sk->sk_lock.slock);
2225 }
2226 EXPORT_SYMBOL(release_sock);
2227 
2228 /**
2229  * lock_sock_fast - fast version of lock_sock
2230  * @sk: socket
2231  *
2232  * This version should be used for very small section, where process wont block
2233  * return false if fast path is taken
2234  *   sk_lock.slock locked, owned = 0, BH disabled
2235  * return true if slow path is taken
2236  *   sk_lock.slock unlocked, owned = 1, BH enabled
2237  */
2238 bool lock_sock_fast(struct sock *sk)
2239 {
2240 	might_sleep();
2241 	spin_lock_bh(&sk->sk_lock.slock);
2242 
2243 	if (!sk->sk_lock.owned)
2244 		/*
2245 		 * Note : We must disable BH
2246 		 */
2247 		return false;
2248 
2249 	__lock_sock(sk);
2250 	sk->sk_lock.owned = 1;
2251 	spin_unlock(&sk->sk_lock.slock);
2252 	/*
2253 	 * The sk_lock has mutex_lock() semantics here:
2254 	 */
2255 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2256 	local_bh_enable();
2257 	return true;
2258 }
2259 EXPORT_SYMBOL(lock_sock_fast);
2260 
2261 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2262 {
2263 	struct timeval tv;
2264 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2265 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2266 	tv = ktime_to_timeval(sk->sk_stamp);
2267 	if (tv.tv_sec == -1)
2268 		return -ENOENT;
2269 	if (tv.tv_sec == 0) {
2270 		sk->sk_stamp = ktime_get_real();
2271 		tv = ktime_to_timeval(sk->sk_stamp);
2272 	}
2273 	return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2274 }
2275 EXPORT_SYMBOL(sock_get_timestamp);
2276 
2277 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2278 {
2279 	struct timespec ts;
2280 	if (!sock_flag(sk, SOCK_TIMESTAMP))
2281 		sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2282 	ts = ktime_to_timespec(sk->sk_stamp);
2283 	if (ts.tv_sec == -1)
2284 		return -ENOENT;
2285 	if (ts.tv_sec == 0) {
2286 		sk->sk_stamp = ktime_get_real();
2287 		ts = ktime_to_timespec(sk->sk_stamp);
2288 	}
2289 	return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2290 }
2291 EXPORT_SYMBOL(sock_get_timestampns);
2292 
2293 void sock_enable_timestamp(struct sock *sk, int flag)
2294 {
2295 	if (!sock_flag(sk, flag)) {
2296 		unsigned long previous_flags = sk->sk_flags;
2297 
2298 		sock_set_flag(sk, flag);
2299 		/*
2300 		 * we just set one of the two flags which require net
2301 		 * time stamping, but time stamping might have been on
2302 		 * already because of the other one
2303 		 */
2304 		if (!(previous_flags & SK_FLAGS_TIMESTAMP))
2305 			net_enable_timestamp();
2306 	}
2307 }
2308 
2309 /*
2310  *	Get a socket option on an socket.
2311  *
2312  *	FIX: POSIX 1003.1g is very ambiguous here. It states that
2313  *	asynchronous errors should be reported by getsockopt. We assume
2314  *	this means if you specify SO_ERROR (otherwise whats the point of it).
2315  */
2316 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2317 			   char __user *optval, int __user *optlen)
2318 {
2319 	struct sock *sk = sock->sk;
2320 
2321 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2322 }
2323 EXPORT_SYMBOL(sock_common_getsockopt);
2324 
2325 #ifdef CONFIG_COMPAT
2326 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2327 				  char __user *optval, int __user *optlen)
2328 {
2329 	struct sock *sk = sock->sk;
2330 
2331 	if (sk->sk_prot->compat_getsockopt != NULL)
2332 		return sk->sk_prot->compat_getsockopt(sk, level, optname,
2333 						      optval, optlen);
2334 	return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2335 }
2336 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2337 #endif
2338 
2339 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2340 			struct msghdr *msg, size_t size, int flags)
2341 {
2342 	struct sock *sk = sock->sk;
2343 	int addr_len = 0;
2344 	int err;
2345 
2346 	err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2347 				   flags & ~MSG_DONTWAIT, &addr_len);
2348 	if (err >= 0)
2349 		msg->msg_namelen = addr_len;
2350 	return err;
2351 }
2352 EXPORT_SYMBOL(sock_common_recvmsg);
2353 
2354 /*
2355  *	Set socket options on an inet socket.
2356  */
2357 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2358 			   char __user *optval, unsigned int optlen)
2359 {
2360 	struct sock *sk = sock->sk;
2361 
2362 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2363 }
2364 EXPORT_SYMBOL(sock_common_setsockopt);
2365 
2366 #ifdef CONFIG_COMPAT
2367 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2368 				  char __user *optval, unsigned int optlen)
2369 {
2370 	struct sock *sk = sock->sk;
2371 
2372 	if (sk->sk_prot->compat_setsockopt != NULL)
2373 		return sk->sk_prot->compat_setsockopt(sk, level, optname,
2374 						      optval, optlen);
2375 	return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2376 }
2377 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2378 #endif
2379 
2380 void sk_common_release(struct sock *sk)
2381 {
2382 	if (sk->sk_prot->destroy)
2383 		sk->sk_prot->destroy(sk);
2384 
2385 	/*
2386 	 * Observation: when sock_common_release is called, processes have
2387 	 * no access to socket. But net still has.
2388 	 * Step one, detach it from networking:
2389 	 *
2390 	 * A. Remove from hash tables.
2391 	 */
2392 
2393 	sk->sk_prot->unhash(sk);
2394 
2395 	/*
2396 	 * In this point socket cannot receive new packets, but it is possible
2397 	 * that some packets are in flight because some CPU runs receiver and
2398 	 * did hash table lookup before we unhashed socket. They will achieve
2399 	 * receive queue and will be purged by socket destructor.
2400 	 *
2401 	 * Also we still have packets pending on receive queue and probably,
2402 	 * our own packets waiting in device queues. sock_destroy will drain
2403 	 * receive queue, but transmitted packets will delay socket destruction
2404 	 * until the last reference will be released.
2405 	 */
2406 
2407 	sock_orphan(sk);
2408 
2409 	xfrm_sk_free_policy(sk);
2410 
2411 	sk_refcnt_debug_release(sk);
2412 	sock_put(sk);
2413 }
2414 EXPORT_SYMBOL(sk_common_release);
2415 
2416 #ifdef CONFIG_PROC_FS
2417 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
2418 struct prot_inuse {
2419 	int val[PROTO_INUSE_NR];
2420 };
2421 
2422 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2423 
2424 #ifdef CONFIG_NET_NS
2425 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2426 {
2427 	__this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2428 }
2429 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2430 
2431 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2432 {
2433 	int cpu, idx = prot->inuse_idx;
2434 	int res = 0;
2435 
2436 	for_each_possible_cpu(cpu)
2437 		res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2438 
2439 	return res >= 0 ? res : 0;
2440 }
2441 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2442 
2443 static int __net_init sock_inuse_init_net(struct net *net)
2444 {
2445 	net->core.inuse = alloc_percpu(struct prot_inuse);
2446 	return net->core.inuse ? 0 : -ENOMEM;
2447 }
2448 
2449 static void __net_exit sock_inuse_exit_net(struct net *net)
2450 {
2451 	free_percpu(net->core.inuse);
2452 }
2453 
2454 static struct pernet_operations net_inuse_ops = {
2455 	.init = sock_inuse_init_net,
2456 	.exit = sock_inuse_exit_net,
2457 };
2458 
2459 static __init int net_inuse_init(void)
2460 {
2461 	if (register_pernet_subsys(&net_inuse_ops))
2462 		panic("Cannot initialize net inuse counters");
2463 
2464 	return 0;
2465 }
2466 
2467 core_initcall(net_inuse_init);
2468 #else
2469 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2470 
2471 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2472 {
2473 	__this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2474 }
2475 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2476 
2477 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2478 {
2479 	int cpu, idx = prot->inuse_idx;
2480 	int res = 0;
2481 
2482 	for_each_possible_cpu(cpu)
2483 		res += per_cpu(prot_inuse, cpu).val[idx];
2484 
2485 	return res >= 0 ? res : 0;
2486 }
2487 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2488 #endif
2489 
2490 static void assign_proto_idx(struct proto *prot)
2491 {
2492 	prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2493 
2494 	if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2495 		pr_err("PROTO_INUSE_NR exhausted\n");
2496 		return;
2497 	}
2498 
2499 	set_bit(prot->inuse_idx, proto_inuse_idx);
2500 }
2501 
2502 static void release_proto_idx(struct proto *prot)
2503 {
2504 	if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2505 		clear_bit(prot->inuse_idx, proto_inuse_idx);
2506 }
2507 #else
2508 static inline void assign_proto_idx(struct proto *prot)
2509 {
2510 }
2511 
2512 static inline void release_proto_idx(struct proto *prot)
2513 {
2514 }
2515 #endif
2516 
2517 int proto_register(struct proto *prot, int alloc_slab)
2518 {
2519 	if (alloc_slab) {
2520 		prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2521 					SLAB_HWCACHE_ALIGN | prot->slab_flags,
2522 					NULL);
2523 
2524 		if (prot->slab == NULL) {
2525 			pr_crit("%s: Can't create sock SLAB cache!\n",
2526 				prot->name);
2527 			goto out;
2528 		}
2529 
2530 		if (prot->rsk_prot != NULL) {
2531 			prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2532 			if (prot->rsk_prot->slab_name == NULL)
2533 				goto out_free_sock_slab;
2534 
2535 			prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2536 								 prot->rsk_prot->obj_size, 0,
2537 								 SLAB_HWCACHE_ALIGN, NULL);
2538 
2539 			if (prot->rsk_prot->slab == NULL) {
2540 				pr_crit("%s: Can't create request sock SLAB cache!\n",
2541 					prot->name);
2542 				goto out_free_request_sock_slab_name;
2543 			}
2544 		}
2545 
2546 		if (prot->twsk_prot != NULL) {
2547 			prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2548 
2549 			if (prot->twsk_prot->twsk_slab_name == NULL)
2550 				goto out_free_request_sock_slab;
2551 
2552 			prot->twsk_prot->twsk_slab =
2553 				kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2554 						  prot->twsk_prot->twsk_obj_size,
2555 						  0,
2556 						  SLAB_HWCACHE_ALIGN |
2557 							prot->slab_flags,
2558 						  NULL);
2559 			if (prot->twsk_prot->twsk_slab == NULL)
2560 				goto out_free_timewait_sock_slab_name;
2561 		}
2562 	}
2563 
2564 	mutex_lock(&proto_list_mutex);
2565 	list_add(&prot->node, &proto_list);
2566 	assign_proto_idx(prot);
2567 	mutex_unlock(&proto_list_mutex);
2568 	return 0;
2569 
2570 out_free_timewait_sock_slab_name:
2571 	kfree(prot->twsk_prot->twsk_slab_name);
2572 out_free_request_sock_slab:
2573 	if (prot->rsk_prot && prot->rsk_prot->slab) {
2574 		kmem_cache_destroy(prot->rsk_prot->slab);
2575 		prot->rsk_prot->slab = NULL;
2576 	}
2577 out_free_request_sock_slab_name:
2578 	if (prot->rsk_prot)
2579 		kfree(prot->rsk_prot->slab_name);
2580 out_free_sock_slab:
2581 	kmem_cache_destroy(prot->slab);
2582 	prot->slab = NULL;
2583 out:
2584 	return -ENOBUFS;
2585 }
2586 EXPORT_SYMBOL(proto_register);
2587 
2588 void proto_unregister(struct proto *prot)
2589 {
2590 	mutex_lock(&proto_list_mutex);
2591 	release_proto_idx(prot);
2592 	list_del(&prot->node);
2593 	mutex_unlock(&proto_list_mutex);
2594 
2595 	if (prot->slab != NULL) {
2596 		kmem_cache_destroy(prot->slab);
2597 		prot->slab = NULL;
2598 	}
2599 
2600 	if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2601 		kmem_cache_destroy(prot->rsk_prot->slab);
2602 		kfree(prot->rsk_prot->slab_name);
2603 		prot->rsk_prot->slab = NULL;
2604 	}
2605 
2606 	if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2607 		kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2608 		kfree(prot->twsk_prot->twsk_slab_name);
2609 		prot->twsk_prot->twsk_slab = NULL;
2610 	}
2611 }
2612 EXPORT_SYMBOL(proto_unregister);
2613 
2614 #ifdef CONFIG_PROC_FS
2615 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2616 	__acquires(proto_list_mutex)
2617 {
2618 	mutex_lock(&proto_list_mutex);
2619 	return seq_list_start_head(&proto_list, *pos);
2620 }
2621 
2622 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2623 {
2624 	return seq_list_next(v, &proto_list, pos);
2625 }
2626 
2627 static void proto_seq_stop(struct seq_file *seq, void *v)
2628 	__releases(proto_list_mutex)
2629 {
2630 	mutex_unlock(&proto_list_mutex);
2631 }
2632 
2633 static char proto_method_implemented(const void *method)
2634 {
2635 	return method == NULL ? 'n' : 'y';
2636 }
2637 static long sock_prot_memory_allocated(struct proto *proto)
2638 {
2639 	return proto->memory_allocated != NULL ? proto_memory_allocated(proto) : -1L;
2640 }
2641 
2642 static char *sock_prot_memory_pressure(struct proto *proto)
2643 {
2644 	return proto->memory_pressure != NULL ?
2645 	proto_memory_pressure(proto) ? "yes" : "no" : "NI";
2646 }
2647 
2648 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2649 {
2650 
2651 	seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2652 			"%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2653 		   proto->name,
2654 		   proto->obj_size,
2655 		   sock_prot_inuse_get(seq_file_net(seq), proto),
2656 		   sock_prot_memory_allocated(proto),
2657 		   sock_prot_memory_pressure(proto),
2658 		   proto->max_header,
2659 		   proto->slab == NULL ? "no" : "yes",
2660 		   module_name(proto->owner),
2661 		   proto_method_implemented(proto->close),
2662 		   proto_method_implemented(proto->connect),
2663 		   proto_method_implemented(proto->disconnect),
2664 		   proto_method_implemented(proto->accept),
2665 		   proto_method_implemented(proto->ioctl),
2666 		   proto_method_implemented(proto->init),
2667 		   proto_method_implemented(proto->destroy),
2668 		   proto_method_implemented(proto->shutdown),
2669 		   proto_method_implemented(proto->setsockopt),
2670 		   proto_method_implemented(proto->getsockopt),
2671 		   proto_method_implemented(proto->sendmsg),
2672 		   proto_method_implemented(proto->recvmsg),
2673 		   proto_method_implemented(proto->sendpage),
2674 		   proto_method_implemented(proto->bind),
2675 		   proto_method_implemented(proto->backlog_rcv),
2676 		   proto_method_implemented(proto->hash),
2677 		   proto_method_implemented(proto->unhash),
2678 		   proto_method_implemented(proto->get_port),
2679 		   proto_method_implemented(proto->enter_memory_pressure));
2680 }
2681 
2682 static int proto_seq_show(struct seq_file *seq, void *v)
2683 {
2684 	if (v == &proto_list)
2685 		seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2686 			   "protocol",
2687 			   "size",
2688 			   "sockets",
2689 			   "memory",
2690 			   "press",
2691 			   "maxhdr",
2692 			   "slab",
2693 			   "module",
2694 			   "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2695 	else
2696 		proto_seq_printf(seq, list_entry(v, struct proto, node));
2697 	return 0;
2698 }
2699 
2700 static const struct seq_operations proto_seq_ops = {
2701 	.start  = proto_seq_start,
2702 	.next   = proto_seq_next,
2703 	.stop   = proto_seq_stop,
2704 	.show   = proto_seq_show,
2705 };
2706 
2707 static int proto_seq_open(struct inode *inode, struct file *file)
2708 {
2709 	return seq_open_net(inode, file, &proto_seq_ops,
2710 			    sizeof(struct seq_net_private));
2711 }
2712 
2713 static const struct file_operations proto_seq_fops = {
2714 	.owner		= THIS_MODULE,
2715 	.open		= proto_seq_open,
2716 	.read		= seq_read,
2717 	.llseek		= seq_lseek,
2718 	.release	= seq_release_net,
2719 };
2720 
2721 static __net_init int proto_init_net(struct net *net)
2722 {
2723 	if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2724 		return -ENOMEM;
2725 
2726 	return 0;
2727 }
2728 
2729 static __net_exit void proto_exit_net(struct net *net)
2730 {
2731 	proc_net_remove(net, "protocols");
2732 }
2733 
2734 
2735 static __net_initdata struct pernet_operations proto_net_ops = {
2736 	.init = proto_init_net,
2737 	.exit = proto_exit_net,
2738 };
2739 
2740 static int __init proto_init(void)
2741 {
2742 	return register_pernet_subsys(&proto_net_ops);
2743 }
2744 
2745 subsys_initcall(proto_init);
2746 
2747 #endif /* PROC_FS */
2748