xref: /linux/include/net/sock.h (revision ed3174d93c342b8b2eeba6bbd124707d55304a7b)
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  *		Definitions for the AF_INET socket handler.
7  *
8  * Version:	@(#)sock.h	1.0.4	05/13/93
9  *
10  * Authors:	Ross Biro
11  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12  *		Corey Minyard <wf-rch!minyard@relay.EU.net>
13  *		Florian La Roche <flla@stud.uni-sb.de>
14  *
15  * Fixes:
16  *		Alan Cox	:	Volatiles in skbuff pointers. See
17  *					skbuff comments. May be overdone,
18  *					better to prove they can be removed
19  *					than the reverse.
20  *		Alan Cox	:	Added a zapped field for tcp to note
21  *					a socket is reset and must stay shut up
22  *		Alan Cox	:	New fields for options
23  *	Pauline Middelink	:	identd support
24  *		Alan Cox	:	Eliminate low level recv/recvfrom
25  *		David S. Miller	:	New socket lookup architecture.
26  *              Steve Whitehouse:       Default routines for sock_ops
27  *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
28  *              			protinfo be just a void pointer, as the
29  *              			protocol specific parts were moved to
30  *              			respective headers and ipv4/v6, etc now
31  *              			use private slabcaches for its socks
32  *              Pedro Hortas	:	New flags field for socket options
33  *
34  *
35  *		This program is free software; you can redistribute it and/or
36  *		modify it under the terms of the GNU General Public License
37  *		as published by the Free Software Foundation; either version
38  *		2 of the License, or (at your option) any later version.
39  */
40 #ifndef _SOCK_H
41 #define _SOCK_H
42 
43 #include <linux/kernel.h>
44 #include <linux/list.h>
45 #include <linux/timer.h>
46 #include <linux/cache.h>
47 #include <linux/module.h>
48 #include <linux/lockdep.h>
49 #include <linux/netdevice.h>
50 #include <linux/pcounter.h>
51 #include <linux/skbuff.h>	/* struct sk_buff */
52 #include <linux/mm.h>
53 #include <linux/security.h>
54 
55 #include <linux/filter.h>
56 
57 #include <asm/atomic.h>
58 #include <net/dst.h>
59 #include <net/checksum.h>
60 
61 /*
62  * This structure really needs to be cleaned up.
63  * Most of it is for TCP, and not used by any of
64  * the other protocols.
65  */
66 
67 /* Define this to get the SOCK_DBG debugging facility. */
68 #define SOCK_DEBUGGING
69 #ifdef SOCK_DEBUGGING
70 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
71 					printk(KERN_DEBUG msg); } while (0)
72 #else
73 #define SOCK_DEBUG(sk, msg...) do { } while (0)
74 #endif
75 
76 /* This is the per-socket lock.  The spinlock provides a synchronization
77  * between user contexts and software interrupt processing, whereas the
78  * mini-semaphore synchronizes multiple users amongst themselves.
79  */
80 typedef struct {
81 	spinlock_t		slock;
82 	int			owned;
83 	wait_queue_head_t	wq;
84 	/*
85 	 * We express the mutex-alike socket_lock semantics
86 	 * to the lock validator by explicitly managing
87 	 * the slock as a lock variant (in addition to
88 	 * the slock itself):
89 	 */
90 #ifdef CONFIG_DEBUG_LOCK_ALLOC
91 	struct lockdep_map dep_map;
92 #endif
93 } socket_lock_t;
94 
95 struct sock;
96 struct proto;
97 struct net;
98 
99 /**
100  *	struct sock_common - minimal network layer representation of sockets
101  *	@skc_family: network address family
102  *	@skc_state: Connection state
103  *	@skc_reuse: %SO_REUSEADDR setting
104  *	@skc_bound_dev_if: bound device index if != 0
105  *	@skc_node: main hash linkage for various protocol lookup tables
106  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
107  *	@skc_refcnt: reference count
108  *	@skc_hash: hash value used with various protocol lookup tables
109  *	@skc_prot: protocol handlers inside a network family
110  *	@skc_net: reference to the network namespace of this socket
111  *
112  *	This is the minimal network layer representation of sockets, the header
113  *	for struct sock and struct inet_timewait_sock.
114  */
115 struct sock_common {
116 	unsigned short		skc_family;
117 	volatile unsigned char	skc_state;
118 	unsigned char		skc_reuse;
119 	int			skc_bound_dev_if;
120 	struct hlist_node	skc_node;
121 	struct hlist_node	skc_bind_node;
122 	atomic_t		skc_refcnt;
123 	unsigned int		skc_hash;
124 	struct proto		*skc_prot;
125 	struct net	 	*skc_net;
126 };
127 
128 /**
129   *	struct sock - network layer representation of sockets
130   *	@__sk_common: shared layout with inet_timewait_sock
131   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
132   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
133   *	@sk_lock:	synchronizer
134   *	@sk_rcvbuf: size of receive buffer in bytes
135   *	@sk_sleep: sock wait queue
136   *	@sk_dst_cache: destination cache
137   *	@sk_dst_lock: destination cache lock
138   *	@sk_policy: flow policy
139   *	@sk_rmem_alloc: receive queue bytes committed
140   *	@sk_receive_queue: incoming packets
141   *	@sk_wmem_alloc: transmit queue bytes committed
142   *	@sk_write_queue: Packet sending queue
143   *	@sk_async_wait_queue: DMA copied packets
144   *	@sk_omem_alloc: "o" is "option" or "other"
145   *	@sk_wmem_queued: persistent queue size
146   *	@sk_forward_alloc: space allocated forward
147   *	@sk_allocation: allocation mode
148   *	@sk_sndbuf: size of send buffer in bytes
149   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
150   *		   %SO_OOBINLINE settings
151   *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
152   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
153   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
154   *	@sk_lingertime: %SO_LINGER l_linger setting
155   *	@sk_backlog: always used with the per-socket spinlock held
156   *	@sk_callback_lock: used with the callbacks in the end of this struct
157   *	@sk_error_queue: rarely used
158   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
159   *			  IPV6_ADDRFORM for instance)
160   *	@sk_err: last error
161   *	@sk_err_soft: errors that don't cause failure but are the cause of a
162   *		      persistent failure not just 'timed out'
163   *	@sk_drops: raw drops counter
164   *	@sk_ack_backlog: current listen backlog
165   *	@sk_max_ack_backlog: listen backlog set in listen()
166   *	@sk_priority: %SO_PRIORITY setting
167   *	@sk_type: socket type (%SOCK_STREAM, etc)
168   *	@sk_protocol: which protocol this socket belongs in this network family
169   *	@sk_peercred: %SO_PEERCRED setting
170   *	@sk_rcvlowat: %SO_RCVLOWAT setting
171   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
172   *	@sk_sndtimeo: %SO_SNDTIMEO setting
173   *	@sk_filter: socket filtering instructions
174   *	@sk_protinfo: private area, net family specific, when not using slab
175   *	@sk_timer: sock cleanup timer
176   *	@sk_stamp: time stamp of last packet received
177   *	@sk_socket: Identd and reporting IO signals
178   *	@sk_user_data: RPC layer private data
179   *	@sk_sndmsg_page: cached page for sendmsg
180   *	@sk_sndmsg_off: cached offset for sendmsg
181   *	@sk_send_head: front of stuff to transmit
182   *	@sk_security: used by security modules
183   *	@sk_write_pending: a write to stream socket waits to start
184   *	@sk_state_change: callback to indicate change in the state of the sock
185   *	@sk_data_ready: callback to indicate there is data to be processed
186   *	@sk_write_space: callback to indicate there is bf sending space available
187   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
188   *	@sk_backlog_rcv: callback to process the backlog
189   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
190  */
191 struct sock {
192 	/*
193 	 * Now struct inet_timewait_sock also uses sock_common, so please just
194 	 * don't add nothing before this first member (__sk_common) --acme
195 	 */
196 	struct sock_common	__sk_common;
197 #define sk_family		__sk_common.skc_family
198 #define sk_state		__sk_common.skc_state
199 #define sk_reuse		__sk_common.skc_reuse
200 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
201 #define sk_node			__sk_common.skc_node
202 #define sk_bind_node		__sk_common.skc_bind_node
203 #define sk_refcnt		__sk_common.skc_refcnt
204 #define sk_hash			__sk_common.skc_hash
205 #define sk_prot			__sk_common.skc_prot
206 #define sk_net			__sk_common.skc_net
207 	unsigned char		sk_shutdown : 2,
208 				sk_no_check : 2,
209 				sk_userlocks : 4;
210 	unsigned char		sk_protocol;
211 	unsigned short		sk_type;
212 	int			sk_rcvbuf;
213 	socket_lock_t		sk_lock;
214 	/*
215 	 * The backlog queue is special, it is always used with
216 	 * the per-socket spinlock held and requires low latency
217 	 * access. Therefore we special case it's implementation.
218 	 */
219 	struct {
220 		struct sk_buff *head;
221 		struct sk_buff *tail;
222 	} sk_backlog;
223 	wait_queue_head_t	*sk_sleep;
224 	struct dst_entry	*sk_dst_cache;
225 	struct xfrm_policy	*sk_policy[2];
226 	rwlock_t		sk_dst_lock;
227 	atomic_t		sk_rmem_alloc;
228 	atomic_t		sk_wmem_alloc;
229 	atomic_t		sk_omem_alloc;
230 	int			sk_sndbuf;
231 	struct sk_buff_head	sk_receive_queue;
232 	struct sk_buff_head	sk_write_queue;
233 	struct sk_buff_head	sk_async_wait_queue;
234 	int			sk_wmem_queued;
235 	int			sk_forward_alloc;
236 	gfp_t			sk_allocation;
237 	int			sk_route_caps;
238 	int			sk_gso_type;
239 	int			sk_rcvlowat;
240 	unsigned long 		sk_flags;
241 	unsigned long	        sk_lingertime;
242 	struct sk_buff_head	sk_error_queue;
243 	struct proto		*sk_prot_creator;
244 	rwlock_t		sk_callback_lock;
245 	int			sk_err,
246 				sk_err_soft;
247 	atomic_t		sk_drops;
248 	unsigned short		sk_ack_backlog;
249 	unsigned short		sk_max_ack_backlog;
250 	__u32			sk_priority;
251 	struct ucred		sk_peercred;
252 	long			sk_rcvtimeo;
253 	long			sk_sndtimeo;
254 	struct sk_filter      	*sk_filter;
255 	void			*sk_protinfo;
256 	struct timer_list	sk_timer;
257 	ktime_t			sk_stamp;
258 	struct socket		*sk_socket;
259 	void			*sk_user_data;
260 	struct page		*sk_sndmsg_page;
261 	struct sk_buff		*sk_send_head;
262 	__u32			sk_sndmsg_off;
263 	int			sk_write_pending;
264 	void			*sk_security;
265 	__u32			sk_mark;
266 	/* XXX 4 bytes hole on 64 bit */
267 	void			(*sk_state_change)(struct sock *sk);
268 	void			(*sk_data_ready)(struct sock *sk, int bytes);
269 	void			(*sk_write_space)(struct sock *sk);
270 	void			(*sk_error_report)(struct sock *sk);
271   	int			(*sk_backlog_rcv)(struct sock *sk,
272 						  struct sk_buff *skb);
273 	void                    (*sk_destruct)(struct sock *sk);
274 };
275 
276 /*
277  * Hashed lists helper routines
278  */
279 static inline struct sock *__sk_head(const struct hlist_head *head)
280 {
281 	return hlist_entry(head->first, struct sock, sk_node);
282 }
283 
284 static inline struct sock *sk_head(const struct hlist_head *head)
285 {
286 	return hlist_empty(head) ? NULL : __sk_head(head);
287 }
288 
289 static inline struct sock *sk_next(const struct sock *sk)
290 {
291 	return sk->sk_node.next ?
292 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
293 }
294 
295 static inline int sk_unhashed(const struct sock *sk)
296 {
297 	return hlist_unhashed(&sk->sk_node);
298 }
299 
300 static inline int sk_hashed(const struct sock *sk)
301 {
302 	return !sk_unhashed(sk);
303 }
304 
305 static __inline__ void sk_node_init(struct hlist_node *node)
306 {
307 	node->pprev = NULL;
308 }
309 
310 static __inline__ void __sk_del_node(struct sock *sk)
311 {
312 	__hlist_del(&sk->sk_node);
313 }
314 
315 static __inline__ int __sk_del_node_init(struct sock *sk)
316 {
317 	if (sk_hashed(sk)) {
318 		__sk_del_node(sk);
319 		sk_node_init(&sk->sk_node);
320 		return 1;
321 	}
322 	return 0;
323 }
324 
325 /* Grab socket reference count. This operation is valid only
326    when sk is ALREADY grabbed f.e. it is found in hash table
327    or a list and the lookup is made under lock preventing hash table
328    modifications.
329  */
330 
331 static inline void sock_hold(struct sock *sk)
332 {
333 	atomic_inc(&sk->sk_refcnt);
334 }
335 
336 /* Ungrab socket in the context, which assumes that socket refcnt
337    cannot hit zero, f.e. it is true in context of any socketcall.
338  */
339 static inline void __sock_put(struct sock *sk)
340 {
341 	atomic_dec(&sk->sk_refcnt);
342 }
343 
344 static __inline__ int sk_del_node_init(struct sock *sk)
345 {
346 	int rc = __sk_del_node_init(sk);
347 
348 	if (rc) {
349 		/* paranoid for a while -acme */
350 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
351 		__sock_put(sk);
352 	}
353 	return rc;
354 }
355 
356 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
357 {
358 	hlist_add_head(&sk->sk_node, list);
359 }
360 
361 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
362 {
363 	sock_hold(sk);
364 	__sk_add_node(sk, list);
365 }
366 
367 static __inline__ void __sk_del_bind_node(struct sock *sk)
368 {
369 	__hlist_del(&sk->sk_bind_node);
370 }
371 
372 static __inline__ void sk_add_bind_node(struct sock *sk,
373 					struct hlist_head *list)
374 {
375 	hlist_add_head(&sk->sk_bind_node, list);
376 }
377 
378 #define sk_for_each(__sk, node, list) \
379 	hlist_for_each_entry(__sk, node, list, sk_node)
380 #define sk_for_each_from(__sk, node) \
381 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
382 		hlist_for_each_entry_from(__sk, node, sk_node)
383 #define sk_for_each_continue(__sk, node) \
384 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
385 		hlist_for_each_entry_continue(__sk, node, sk_node)
386 #define sk_for_each_safe(__sk, node, tmp, list) \
387 	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
388 #define sk_for_each_bound(__sk, node, list) \
389 	hlist_for_each_entry(__sk, node, list, sk_bind_node)
390 
391 /* Sock flags */
392 enum sock_flags {
393 	SOCK_DEAD,
394 	SOCK_DONE,
395 	SOCK_URGINLINE,
396 	SOCK_KEEPOPEN,
397 	SOCK_LINGER,
398 	SOCK_DESTROY,
399 	SOCK_BROADCAST,
400 	SOCK_TIMESTAMP,
401 	SOCK_ZAPPED,
402 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
403 	SOCK_DBG, /* %SO_DEBUG setting */
404 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
405 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
406 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
407 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
408 };
409 
410 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
411 {
412 	nsk->sk_flags = osk->sk_flags;
413 }
414 
415 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
416 {
417 	__set_bit(flag, &sk->sk_flags);
418 }
419 
420 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
421 {
422 	__clear_bit(flag, &sk->sk_flags);
423 }
424 
425 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
426 {
427 	return test_bit(flag, &sk->sk_flags);
428 }
429 
430 static inline void sk_acceptq_removed(struct sock *sk)
431 {
432 	sk->sk_ack_backlog--;
433 }
434 
435 static inline void sk_acceptq_added(struct sock *sk)
436 {
437 	sk->sk_ack_backlog++;
438 }
439 
440 static inline int sk_acceptq_is_full(struct sock *sk)
441 {
442 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
443 }
444 
445 /*
446  * Compute minimal free write space needed to queue new packets.
447  */
448 static inline int sk_stream_min_wspace(struct sock *sk)
449 {
450 	return sk->sk_wmem_queued >> 1;
451 }
452 
453 static inline int sk_stream_wspace(struct sock *sk)
454 {
455 	return sk->sk_sndbuf - sk->sk_wmem_queued;
456 }
457 
458 extern void sk_stream_write_space(struct sock *sk);
459 
460 static inline int sk_stream_memory_free(struct sock *sk)
461 {
462 	return sk->sk_wmem_queued < sk->sk_sndbuf;
463 }
464 
465 /* The per-socket spinlock must be held here. */
466 static inline void sk_add_backlog(struct sock *sk, struct sk_buff *skb)
467 {
468 	if (!sk->sk_backlog.tail) {
469 		sk->sk_backlog.head = sk->sk_backlog.tail = skb;
470 	} else {
471 		sk->sk_backlog.tail->next = skb;
472 		sk->sk_backlog.tail = skb;
473 	}
474 	skb->next = NULL;
475 }
476 
477 #define sk_wait_event(__sk, __timeo, __condition)			\
478 	({	int __rc;						\
479 		release_sock(__sk);					\
480 		__rc = __condition;					\
481 		if (!__rc) {						\
482 			*(__timeo) = schedule_timeout(*(__timeo));	\
483 		}							\
484 		lock_sock(__sk);					\
485 		__rc = __condition;					\
486 		__rc;							\
487 	})
488 
489 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
490 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
491 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
492 extern int sk_stream_error(struct sock *sk, int flags, int err);
493 extern void sk_stream_kill_queues(struct sock *sk);
494 
495 extern int sk_wait_data(struct sock *sk, long *timeo);
496 
497 struct request_sock_ops;
498 struct timewait_sock_ops;
499 struct inet_hashinfo;
500 
501 /* Networking protocol blocks we attach to sockets.
502  * socket layer -> transport layer interface
503  * transport -> network interface is defined by struct inet_proto
504  */
505 struct proto {
506 	void			(*close)(struct sock *sk,
507 					long timeout);
508 	int			(*connect)(struct sock *sk,
509 				        struct sockaddr *uaddr,
510 					int addr_len);
511 	int			(*disconnect)(struct sock *sk, int flags);
512 
513 	struct sock *		(*accept) (struct sock *sk, int flags, int *err);
514 
515 	int			(*ioctl)(struct sock *sk, int cmd,
516 					 unsigned long arg);
517 	int			(*init)(struct sock *sk);
518 	int			(*destroy)(struct sock *sk);
519 	void			(*shutdown)(struct sock *sk, int how);
520 	int			(*setsockopt)(struct sock *sk, int level,
521 					int optname, char __user *optval,
522 					int optlen);
523 	int			(*getsockopt)(struct sock *sk, int level,
524 					int optname, char __user *optval,
525 					int __user *option);
526 	int			(*compat_setsockopt)(struct sock *sk,
527 					int level,
528 					int optname, char __user *optval,
529 					int optlen);
530 	int			(*compat_getsockopt)(struct sock *sk,
531 					int level,
532 					int optname, char __user *optval,
533 					int __user *option);
534 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
535 					   struct msghdr *msg, size_t len);
536 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
537 					   struct msghdr *msg,
538 					size_t len, int noblock, int flags,
539 					int *addr_len);
540 	int			(*sendpage)(struct sock *sk, struct page *page,
541 					int offset, size_t size, int flags);
542 	int			(*bind)(struct sock *sk,
543 					struct sockaddr *uaddr, int addr_len);
544 
545 	int			(*backlog_rcv) (struct sock *sk,
546 						struct sk_buff *skb);
547 
548 	/* Keeping track of sk's, looking them up, and port selection methods. */
549 	void			(*hash)(struct sock *sk);
550 	void			(*unhash)(struct sock *sk);
551 	int			(*get_port)(struct sock *sk, unsigned short snum);
552 
553 	/* Keeping track of sockets in use */
554 #ifdef CONFIG_PROC_FS
555 	struct pcounter		inuse;
556 #endif
557 
558 	/* Memory pressure */
559 	void			(*enter_memory_pressure)(void);
560 	atomic_t		*memory_allocated;	/* Current allocated memory. */
561 	atomic_t		*sockets_allocated;	/* Current number of sockets. */
562 	/*
563 	 * Pressure flag: try to collapse.
564 	 * Technical note: it is used by multiple contexts non atomically.
565 	 * All the __sk_mem_schedule() is of this nature: accounting
566 	 * is strict, actions are advisory and have some latency.
567 	 */
568 	int			*memory_pressure;
569 	int			*sysctl_mem;
570 	int			*sysctl_wmem;
571 	int			*sysctl_rmem;
572 	int			max_header;
573 
574 	struct kmem_cache		*slab;
575 	unsigned int		obj_size;
576 
577 	atomic_t		*orphan_count;
578 
579 	struct request_sock_ops	*rsk_prot;
580 	struct timewait_sock_ops *twsk_prot;
581 
582 	struct inet_hashinfo	*hashinfo;
583 
584 	struct module		*owner;
585 
586 	char			name[32];
587 
588 	struct list_head	node;
589 #ifdef SOCK_REFCNT_DEBUG
590 	atomic_t		socks;
591 #endif
592 };
593 
594 extern int proto_register(struct proto *prot, int alloc_slab);
595 extern void proto_unregister(struct proto *prot);
596 
597 #ifdef SOCK_REFCNT_DEBUG
598 static inline void sk_refcnt_debug_inc(struct sock *sk)
599 {
600 	atomic_inc(&sk->sk_prot->socks);
601 }
602 
603 static inline void sk_refcnt_debug_dec(struct sock *sk)
604 {
605 	atomic_dec(&sk->sk_prot->socks);
606 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
607 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
608 }
609 
610 static inline void sk_refcnt_debug_release(const struct sock *sk)
611 {
612 	if (atomic_read(&sk->sk_refcnt) != 1)
613 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
614 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
615 }
616 #else /* SOCK_REFCNT_DEBUG */
617 #define sk_refcnt_debug_inc(sk) do { } while (0)
618 #define sk_refcnt_debug_dec(sk) do { } while (0)
619 #define sk_refcnt_debug_release(sk) do { } while (0)
620 #endif /* SOCK_REFCNT_DEBUG */
621 
622 
623 #ifdef CONFIG_PROC_FS
624 # define DEFINE_PROTO_INUSE(NAME) DEFINE_PCOUNTER(NAME)
625 # define REF_PROTO_INUSE(NAME) PCOUNTER_MEMBER_INITIALIZER(NAME, .inuse)
626 /* Called with local bh disabled */
627 static inline void sock_prot_inuse_add(struct proto *prot, int inc)
628 {
629 	pcounter_add(&prot->inuse, inc);
630 }
631 static inline int sock_prot_inuse_init(struct proto *proto)
632 {
633 	return pcounter_alloc(&proto->inuse);
634 }
635 static inline int sock_prot_inuse_get(struct proto *proto)
636 {
637 	return pcounter_getval(&proto->inuse);
638 }
639 static inline void sock_prot_inuse_free(struct proto *proto)
640 {
641 	pcounter_free(&proto->inuse);
642 }
643 #else
644 # define DEFINE_PROTO_INUSE(NAME)
645 # define REF_PROTO_INUSE(NAME)
646 static void inline sock_prot_inuse_add(struct proto *prot, int inc)
647 {
648 }
649 static int inline sock_prot_inuse_init(struct proto *proto)
650 {
651 	return 0;
652 }
653 static void inline sock_prot_inuse_free(struct proto *proto)
654 {
655 }
656 #endif
657 
658 
659 /* With per-bucket locks this operation is not-atomic, so that
660  * this version is not worse.
661  */
662 static inline void __sk_prot_rehash(struct sock *sk)
663 {
664 	sk->sk_prot->unhash(sk);
665 	sk->sk_prot->hash(sk);
666 }
667 
668 /* About 10 seconds */
669 #define SOCK_DESTROY_TIME (10*HZ)
670 
671 /* Sockets 0-1023 can't be bound to unless you are superuser */
672 #define PROT_SOCK	1024
673 
674 #define SHUTDOWN_MASK	3
675 #define RCV_SHUTDOWN	1
676 #define SEND_SHUTDOWN	2
677 
678 #define SOCK_SNDBUF_LOCK	1
679 #define SOCK_RCVBUF_LOCK	2
680 #define SOCK_BINDADDR_LOCK	4
681 #define SOCK_BINDPORT_LOCK	8
682 
683 /* sock_iocb: used to kick off async processing of socket ios */
684 struct sock_iocb {
685 	struct list_head	list;
686 
687 	int			flags;
688 	int			size;
689 	struct socket		*sock;
690 	struct sock		*sk;
691 	struct scm_cookie	*scm;
692 	struct msghdr		*msg, async_msg;
693 	struct kiocb		*kiocb;
694 };
695 
696 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
697 {
698 	return (struct sock_iocb *)iocb->private;
699 }
700 
701 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
702 {
703 	return si->kiocb;
704 }
705 
706 struct socket_alloc {
707 	struct socket socket;
708 	struct inode vfs_inode;
709 };
710 
711 static inline struct socket *SOCKET_I(struct inode *inode)
712 {
713 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
714 }
715 
716 static inline struct inode *SOCK_INODE(struct socket *socket)
717 {
718 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
719 }
720 
721 /*
722  * Functions for memory accounting
723  */
724 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
725 extern void __sk_mem_reclaim(struct sock *sk);
726 
727 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
728 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
729 #define SK_MEM_SEND	0
730 #define SK_MEM_RECV	1
731 
732 static inline int sk_mem_pages(int amt)
733 {
734 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
735 }
736 
737 static inline int sk_has_account(struct sock *sk)
738 {
739 	/* return true if protocol supports memory accounting */
740 	return !!sk->sk_prot->memory_allocated;
741 }
742 
743 static inline int sk_wmem_schedule(struct sock *sk, int size)
744 {
745 	if (!sk_has_account(sk))
746 		return 1;
747 	return size <= sk->sk_forward_alloc ||
748 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
749 }
750 
751 static inline int sk_rmem_schedule(struct sock *sk, int size)
752 {
753 	if (!sk_has_account(sk))
754 		return 1;
755 	return size <= sk->sk_forward_alloc ||
756 		__sk_mem_schedule(sk, size, SK_MEM_RECV);
757 }
758 
759 static inline void sk_mem_reclaim(struct sock *sk)
760 {
761 	if (!sk_has_account(sk))
762 		return;
763 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
764 		__sk_mem_reclaim(sk);
765 }
766 
767 static inline void sk_mem_reclaim_partial(struct sock *sk)
768 {
769 	if (!sk_has_account(sk))
770 		return;
771 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
772 		__sk_mem_reclaim(sk);
773 }
774 
775 static inline void sk_mem_charge(struct sock *sk, int size)
776 {
777 	if (!sk_has_account(sk))
778 		return;
779 	sk->sk_forward_alloc -= size;
780 }
781 
782 static inline void sk_mem_uncharge(struct sock *sk, int size)
783 {
784 	if (!sk_has_account(sk))
785 		return;
786 	sk->sk_forward_alloc += size;
787 }
788 
789 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
790 {
791 	skb_truesize_check(skb);
792 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
793 	sk->sk_wmem_queued -= skb->truesize;
794 	sk_mem_uncharge(sk, skb->truesize);
795 	__kfree_skb(skb);
796 }
797 
798 /* Used by processes to "lock" a socket state, so that
799  * interrupts and bottom half handlers won't change it
800  * from under us. It essentially blocks any incoming
801  * packets, so that we won't get any new data or any
802  * packets that change the state of the socket.
803  *
804  * While locked, BH processing will add new packets to
805  * the backlog queue.  This queue is processed by the
806  * owner of the socket lock right before it is released.
807  *
808  * Since ~2.3.5 it is also exclusive sleep lock serializing
809  * accesses from user process context.
810  */
811 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
812 
813 /*
814  * Macro so as to not evaluate some arguments when
815  * lockdep is not enabled.
816  *
817  * Mark both the sk_lock and the sk_lock.slock as a
818  * per-address-family lock class.
819  */
820 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) 	\
821 do {									\
822 	sk->sk_lock.owned = 0;					\
823 	init_waitqueue_head(&sk->sk_lock.wq);				\
824 	spin_lock_init(&(sk)->sk_lock.slock);				\
825 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
826 			sizeof((sk)->sk_lock));				\
827 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
828 		       	(skey), (sname));				\
829 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
830 } while (0)
831 
832 extern void lock_sock_nested(struct sock *sk, int subclass);
833 
834 static inline void lock_sock(struct sock *sk)
835 {
836 	lock_sock_nested(sk, 0);
837 }
838 
839 extern void release_sock(struct sock *sk);
840 
841 /* BH context may only use the following locking interface. */
842 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
843 #define bh_lock_sock_nested(__sk) \
844 				spin_lock_nested(&((__sk)->sk_lock.slock), \
845 				SINGLE_DEPTH_NESTING)
846 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
847 
848 extern struct sock		*sk_alloc(struct net *net, int family,
849 					  gfp_t priority,
850 					  struct proto *prot);
851 extern void			sk_free(struct sock *sk);
852 extern struct sock		*sk_clone(const struct sock *sk,
853 					  const gfp_t priority);
854 
855 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
856 					      unsigned long size, int force,
857 					      gfp_t priority);
858 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
859 					      unsigned long size, int force,
860 					      gfp_t priority);
861 extern void			sock_wfree(struct sk_buff *skb);
862 extern void			sock_rfree(struct sk_buff *skb);
863 
864 extern int			sock_setsockopt(struct socket *sock, int level,
865 						int op, char __user *optval,
866 						int optlen);
867 
868 extern int			sock_getsockopt(struct socket *sock, int level,
869 						int op, char __user *optval,
870 						int __user *optlen);
871 extern struct sk_buff 		*sock_alloc_send_skb(struct sock *sk,
872 						     unsigned long size,
873 						     int noblock,
874 						     int *errcode);
875 extern void *sock_kmalloc(struct sock *sk, int size,
876 			  gfp_t priority);
877 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
878 extern void sk_send_sigurg(struct sock *sk);
879 
880 /*
881  * Functions to fill in entries in struct proto_ops when a protocol
882  * does not implement a particular function.
883  */
884 extern int                      sock_no_bind(struct socket *,
885 					     struct sockaddr *, int);
886 extern int                      sock_no_connect(struct socket *,
887 						struct sockaddr *, int, int);
888 extern int                      sock_no_socketpair(struct socket *,
889 						   struct socket *);
890 extern int                      sock_no_accept(struct socket *,
891 					       struct socket *, int);
892 extern int                      sock_no_getname(struct socket *,
893 						struct sockaddr *, int *, int);
894 extern unsigned int             sock_no_poll(struct file *, struct socket *,
895 					     struct poll_table_struct *);
896 extern int                      sock_no_ioctl(struct socket *, unsigned int,
897 					      unsigned long);
898 extern int			sock_no_listen(struct socket *, int);
899 extern int                      sock_no_shutdown(struct socket *, int);
900 extern int			sock_no_getsockopt(struct socket *, int , int,
901 						   char __user *, int __user *);
902 extern int			sock_no_setsockopt(struct socket *, int, int,
903 						   char __user *, int);
904 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
905 						struct msghdr *, size_t);
906 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
907 						struct msghdr *, size_t, int);
908 extern int			sock_no_mmap(struct file *file,
909 					     struct socket *sock,
910 					     struct vm_area_struct *vma);
911 extern ssize_t			sock_no_sendpage(struct socket *sock,
912 						struct page *page,
913 						int offset, size_t size,
914 						int flags);
915 
916 /*
917  * Functions to fill in entries in struct proto_ops when a protocol
918  * uses the inet style.
919  */
920 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
921 				  char __user *optval, int __user *optlen);
922 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
923 			       struct msghdr *msg, size_t size, int flags);
924 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
925 				  char __user *optval, int optlen);
926 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
927 		int optname, char __user *optval, int __user *optlen);
928 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
929 		int optname, char __user *optval, int optlen);
930 
931 extern void sk_common_release(struct sock *sk);
932 
933 /*
934  *	Default socket callbacks and setup code
935  */
936 
937 /* Initialise core socket variables */
938 extern void sock_init_data(struct socket *sock, struct sock *sk);
939 
940 /**
941  *	sk_filter - run a packet through a socket filter
942  *	@sk: sock associated with &sk_buff
943  *	@skb: buffer to filter
944  *	@needlock: set to 1 if the sock is not locked by caller.
945  *
946  * Run the filter code and then cut skb->data to correct size returned by
947  * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
948  * than pkt_len we keep whole skb->data. This is the socket level
949  * wrapper to sk_run_filter. It returns 0 if the packet should
950  * be accepted or -EPERM if the packet should be tossed.
951  *
952  */
953 
954 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
955 {
956 	int err;
957 	struct sk_filter *filter;
958 
959 	err = security_sock_rcv_skb(sk, skb);
960 	if (err)
961 		return err;
962 
963 	rcu_read_lock_bh();
964 	filter = rcu_dereference(sk->sk_filter);
965 	if (filter) {
966 		unsigned int pkt_len = sk_run_filter(skb, filter->insns,
967 				filter->len);
968 		err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
969 	}
970  	rcu_read_unlock_bh();
971 
972 	return err;
973 }
974 
975 /**
976  *	sk_filter_release: Release a socket filter
977  *	@sk: socket
978  *	@fp: filter to remove
979  *
980  *	Remove a filter from a socket and release its resources.
981  */
982 
983 static inline void sk_filter_release(struct sk_filter *fp)
984 {
985 	if (atomic_dec_and_test(&fp->refcnt))
986 		kfree(fp);
987 }
988 
989 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
990 {
991 	unsigned int size = sk_filter_len(fp);
992 
993 	atomic_sub(size, &sk->sk_omem_alloc);
994 	sk_filter_release(fp);
995 }
996 
997 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
998 {
999 	atomic_inc(&fp->refcnt);
1000 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1001 }
1002 
1003 /*
1004  * Socket reference counting postulates.
1005  *
1006  * * Each user of socket SHOULD hold a reference count.
1007  * * Each access point to socket (an hash table bucket, reference from a list,
1008  *   running timer, skb in flight MUST hold a reference count.
1009  * * When reference count hits 0, it means it will never increase back.
1010  * * When reference count hits 0, it means that no references from
1011  *   outside exist to this socket and current process on current CPU
1012  *   is last user and may/should destroy this socket.
1013  * * sk_free is called from any context: process, BH, IRQ. When
1014  *   it is called, socket has no references from outside -> sk_free
1015  *   may release descendant resources allocated by the socket, but
1016  *   to the time when it is called, socket is NOT referenced by any
1017  *   hash tables, lists etc.
1018  * * Packets, delivered from outside (from network or from another process)
1019  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1020  *   when they sit in queue. Otherwise, packets will leak to hole, when
1021  *   socket is looked up by one cpu and unhasing is made by another CPU.
1022  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1023  *   (leak to backlog). Packet socket does all the processing inside
1024  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1025  *   use separate SMP lock, so that they are prone too.
1026  */
1027 
1028 /* Ungrab socket and destroy it, if it was the last reference. */
1029 static inline void sock_put(struct sock *sk)
1030 {
1031 	if (atomic_dec_and_test(&sk->sk_refcnt))
1032 		sk_free(sk);
1033 }
1034 
1035 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1036 			  const int nested);
1037 
1038 /* Detach socket from process context.
1039  * Announce socket dead, detach it from wait queue and inode.
1040  * Note that parent inode held reference count on this struct sock,
1041  * we do not release it in this function, because protocol
1042  * probably wants some additional cleanups or even continuing
1043  * to work with this socket (TCP).
1044  */
1045 static inline void sock_orphan(struct sock *sk)
1046 {
1047 	write_lock_bh(&sk->sk_callback_lock);
1048 	sock_set_flag(sk, SOCK_DEAD);
1049 	sk->sk_socket = NULL;
1050 	sk->sk_sleep  = NULL;
1051 	write_unlock_bh(&sk->sk_callback_lock);
1052 }
1053 
1054 static inline void sock_graft(struct sock *sk, struct socket *parent)
1055 {
1056 	write_lock_bh(&sk->sk_callback_lock);
1057 	sk->sk_sleep = &parent->wait;
1058 	parent->sk = sk;
1059 	sk->sk_socket = parent;
1060 	security_sock_graft(sk, parent);
1061 	write_unlock_bh(&sk->sk_callback_lock);
1062 }
1063 
1064 extern int sock_i_uid(struct sock *sk);
1065 extern unsigned long sock_i_ino(struct sock *sk);
1066 
1067 static inline struct dst_entry *
1068 __sk_dst_get(struct sock *sk)
1069 {
1070 	return sk->sk_dst_cache;
1071 }
1072 
1073 static inline struct dst_entry *
1074 sk_dst_get(struct sock *sk)
1075 {
1076 	struct dst_entry *dst;
1077 
1078 	read_lock(&sk->sk_dst_lock);
1079 	dst = sk->sk_dst_cache;
1080 	if (dst)
1081 		dst_hold(dst);
1082 	read_unlock(&sk->sk_dst_lock);
1083 	return dst;
1084 }
1085 
1086 static inline void
1087 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1088 {
1089 	struct dst_entry *old_dst;
1090 
1091 	old_dst = sk->sk_dst_cache;
1092 	sk->sk_dst_cache = dst;
1093 	dst_release(old_dst);
1094 }
1095 
1096 static inline void
1097 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1098 {
1099 	write_lock(&sk->sk_dst_lock);
1100 	__sk_dst_set(sk, dst);
1101 	write_unlock(&sk->sk_dst_lock);
1102 }
1103 
1104 static inline void
1105 __sk_dst_reset(struct sock *sk)
1106 {
1107 	struct dst_entry *old_dst;
1108 
1109 	old_dst = sk->sk_dst_cache;
1110 	sk->sk_dst_cache = NULL;
1111 	dst_release(old_dst);
1112 }
1113 
1114 static inline void
1115 sk_dst_reset(struct sock *sk)
1116 {
1117 	write_lock(&sk->sk_dst_lock);
1118 	__sk_dst_reset(sk);
1119 	write_unlock(&sk->sk_dst_lock);
1120 }
1121 
1122 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1123 
1124 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1125 
1126 static inline int sk_can_gso(const struct sock *sk)
1127 {
1128 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1129 }
1130 
1131 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1132 
1133 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1134 				   struct sk_buff *skb, struct page *page,
1135 				   int off, int copy)
1136 {
1137 	if (skb->ip_summed == CHECKSUM_NONE) {
1138 		int err = 0;
1139 		__wsum csum = csum_and_copy_from_user(from,
1140 						     page_address(page) + off,
1141 							    copy, 0, &err);
1142 		if (err)
1143 			return err;
1144 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1145 	} else if (copy_from_user(page_address(page) + off, from, copy))
1146 		return -EFAULT;
1147 
1148 	skb->len	     += copy;
1149 	skb->data_len	     += copy;
1150 	skb->truesize	     += copy;
1151 	sk->sk_wmem_queued   += copy;
1152 	sk_mem_charge(sk, copy);
1153 	return 0;
1154 }
1155 
1156 /*
1157  * 	Queue a received datagram if it will fit. Stream and sequenced
1158  *	protocols can't normally use this as they need to fit buffers in
1159  *	and play with them.
1160  *
1161  * 	Inlined as it's very short and called for pretty much every
1162  *	packet ever received.
1163  */
1164 
1165 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1166 {
1167 	sock_hold(sk);
1168 	skb->sk = sk;
1169 	skb->destructor = sock_wfree;
1170 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1171 }
1172 
1173 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1174 {
1175 	skb->sk = sk;
1176 	skb->destructor = sock_rfree;
1177 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1178 	sk_mem_charge(sk, skb->truesize);
1179 }
1180 
1181 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1182 			   unsigned long expires);
1183 
1184 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1185 
1186 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1187 
1188 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
1189 {
1190 	/* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
1191 	   number of warnings when compiling with -W --ANK
1192 	 */
1193 	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1194 	    (unsigned)sk->sk_rcvbuf)
1195 		return -ENOMEM;
1196 	skb_set_owner_r(skb, sk);
1197 	skb_queue_tail(&sk->sk_error_queue, skb);
1198 	if (!sock_flag(sk, SOCK_DEAD))
1199 		sk->sk_data_ready(sk, skb->len);
1200 	return 0;
1201 }
1202 
1203 /*
1204  *	Recover an error report and clear atomically
1205  */
1206 
1207 static inline int sock_error(struct sock *sk)
1208 {
1209 	int err;
1210 	if (likely(!sk->sk_err))
1211 		return 0;
1212 	err = xchg(&sk->sk_err, 0);
1213 	return -err;
1214 }
1215 
1216 static inline unsigned long sock_wspace(struct sock *sk)
1217 {
1218 	int amt = 0;
1219 
1220 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1221 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1222 		if (amt < 0)
1223 			amt = 0;
1224 	}
1225 	return amt;
1226 }
1227 
1228 static inline void sk_wake_async(struct sock *sk, int how, int band)
1229 {
1230 	if (sk->sk_socket && sk->sk_socket->fasync_list)
1231 		sock_wake_async(sk->sk_socket, how, band);
1232 }
1233 
1234 #define SOCK_MIN_SNDBUF 2048
1235 #define SOCK_MIN_RCVBUF 256
1236 
1237 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1238 {
1239 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1240 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1241 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1242 	}
1243 }
1244 
1245 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1246 
1247 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1248 {
1249 	struct page *page = NULL;
1250 
1251 	page = alloc_pages(sk->sk_allocation, 0);
1252 	if (!page) {
1253 		sk->sk_prot->enter_memory_pressure();
1254 		sk_stream_moderate_sndbuf(sk);
1255 	}
1256 	return page;
1257 }
1258 
1259 /*
1260  *	Default write policy as shown to user space via poll/select/SIGIO
1261  */
1262 static inline int sock_writeable(const struct sock *sk)
1263 {
1264 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1265 }
1266 
1267 static inline gfp_t gfp_any(void)
1268 {
1269 	return in_atomic() ? GFP_ATOMIC : GFP_KERNEL;
1270 }
1271 
1272 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1273 {
1274 	return noblock ? 0 : sk->sk_rcvtimeo;
1275 }
1276 
1277 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1278 {
1279 	return noblock ? 0 : sk->sk_sndtimeo;
1280 }
1281 
1282 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1283 {
1284 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1285 }
1286 
1287 /* Alas, with timeout socket operations are not restartable.
1288  * Compare this to poll().
1289  */
1290 static inline int sock_intr_errno(long timeo)
1291 {
1292 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1293 }
1294 
1295 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1296 	struct sk_buff *skb);
1297 
1298 static __inline__ void
1299 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1300 {
1301 	ktime_t kt = skb->tstamp;
1302 
1303 	if (sock_flag(sk, SOCK_RCVTSTAMP))
1304 		__sock_recv_timestamp(msg, sk, skb);
1305 	else
1306 		sk->sk_stamp = kt;
1307 }
1308 
1309 /**
1310  * sk_eat_skb - Release a skb if it is no longer needed
1311  * @sk: socket to eat this skb from
1312  * @skb: socket buffer to eat
1313  * @copied_early: flag indicating whether DMA operations copied this data early
1314  *
1315  * This routine must be called with interrupts disabled or with the socket
1316  * locked so that the sk_buff queue operation is ok.
1317 */
1318 #ifdef CONFIG_NET_DMA
1319 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1320 {
1321 	__skb_unlink(skb, &sk->sk_receive_queue);
1322 	if (!copied_early)
1323 		__kfree_skb(skb);
1324 	else
1325 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
1326 }
1327 #else
1328 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1329 {
1330 	__skb_unlink(skb, &sk->sk_receive_queue);
1331 	__kfree_skb(skb);
1332 }
1333 #endif
1334 
1335 extern void sock_enable_timestamp(struct sock *sk);
1336 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1337 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1338 
1339 /*
1340  *	Enable debug/info messages
1341  */
1342 extern int net_msg_warn;
1343 #define NETDEBUG(fmt, args...) \
1344 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
1345 
1346 #define LIMIT_NETDEBUG(fmt, args...) \
1347 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
1348 
1349 /*
1350  * Macros for sleeping on a socket. Use them like this:
1351  *
1352  * SOCK_SLEEP_PRE(sk)
1353  * if (condition)
1354  * 	schedule();
1355  * SOCK_SLEEP_POST(sk)
1356  *
1357  * N.B. These are now obsolete and were, afaik, only ever used in DECnet
1358  * and when the last use of them in DECnet has gone, I'm intending to
1359  * remove them.
1360  */
1361 
1362 #define SOCK_SLEEP_PRE(sk) 	{ struct task_struct *tsk = current; \
1363 				DECLARE_WAITQUEUE(wait, tsk); \
1364 				tsk->state = TASK_INTERRUPTIBLE; \
1365 				add_wait_queue((sk)->sk_sleep, &wait); \
1366 				release_sock(sk);
1367 
1368 #define SOCK_SLEEP_POST(sk)	tsk->state = TASK_RUNNING; \
1369 				remove_wait_queue((sk)->sk_sleep, &wait); \
1370 				lock_sock(sk); \
1371 				}
1372 
1373 extern __u32 sysctl_wmem_max;
1374 extern __u32 sysctl_rmem_max;
1375 
1376 extern void sk_init(void);
1377 
1378 extern int sysctl_optmem_max;
1379 
1380 extern __u32 sysctl_wmem_default;
1381 extern __u32 sysctl_rmem_default;
1382 
1383 #endif	/* _SOCK_H */
1384