xref: /linux/include/net/sock.h (revision 6b2d2cec1081a979e0efd6a1e9559e5a01a3c10e)
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/skbuff.h>	/* struct sk_buff */
51 #include <linux/mm.h>
52 #include <linux/security.h>
53 
54 #include <linux/filter.h>
55 
56 #include <asm/atomic.h>
57 #include <net/dst.h>
58 #include <net/checksum.h>
59 #include <net/net_namespace.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 
98 /**
99  *	struct sock_common - minimal network layer representation of sockets
100  *	@skc_family: network address family
101  *	@skc_state: Connection state
102  *	@skc_reuse: %SO_REUSEADDR setting
103  *	@skc_bound_dev_if: bound device index if != 0
104  *	@skc_node: main hash linkage for various protocol lookup tables
105  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
106  *	@skc_refcnt: reference count
107  *	@skc_hash: hash value used with various protocol lookup tables
108  *	@skc_prot: protocol handlers inside a network family
109  *	@skc_net: reference to the network namespace of this socket
110  *
111  *	This is the minimal network layer representation of sockets, the header
112  *	for struct sock and struct inet_timewait_sock.
113  */
114 struct sock_common {
115 	unsigned short		skc_family;
116 	volatile unsigned char	skc_state;
117 	unsigned char		skc_reuse;
118 	int			skc_bound_dev_if;
119 	struct hlist_node	skc_node;
120 	struct hlist_node	skc_bind_node;
121 	atomic_t		skc_refcnt;
122 	unsigned int		skc_hash;
123 	struct proto		*skc_prot;
124 	struct net	 	*skc_net;
125 };
126 
127 /**
128   *	struct sock - network layer representation of sockets
129   *	@__sk_common: shared layout with inet_timewait_sock
130   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
131   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
132   *	@sk_lock:	synchronizer
133   *	@sk_rcvbuf: size of receive buffer in bytes
134   *	@sk_sleep: sock wait queue
135   *	@sk_dst_cache: destination cache
136   *	@sk_dst_lock: destination cache lock
137   *	@sk_policy: flow policy
138   *	@sk_rmem_alloc: receive queue bytes committed
139   *	@sk_receive_queue: incoming packets
140   *	@sk_wmem_alloc: transmit queue bytes committed
141   *	@sk_write_queue: Packet sending queue
142   *	@sk_async_wait_queue: DMA copied packets
143   *	@sk_omem_alloc: "o" is "option" or "other"
144   *	@sk_wmem_queued: persistent queue size
145   *	@sk_forward_alloc: space allocated forward
146   *	@sk_allocation: allocation mode
147   *	@sk_sndbuf: size of send buffer in bytes
148   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE, %SO_OOBINLINE settings
149   *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
150   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
151   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
152   *	@sk_lingertime: %SO_LINGER l_linger setting
153   *	@sk_backlog: always used with the per-socket spinlock held
154   *	@sk_callback_lock: used with the callbacks in the end of this struct
155   *	@sk_error_queue: rarely used
156   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt, IPV6_ADDRFORM for instance)
157   *	@sk_err: last error
158   *	@sk_err_soft: errors that don't cause failure but are the cause of a persistent failure not just 'timed out'
159   *	@sk_ack_backlog: current listen backlog
160   *	@sk_max_ack_backlog: listen backlog set in listen()
161   *	@sk_priority: %SO_PRIORITY setting
162   *	@sk_type: socket type (%SOCK_STREAM, etc)
163   *	@sk_protocol: which protocol this socket belongs in this network family
164   *	@sk_peercred: %SO_PEERCRED setting
165   *	@sk_rcvlowat: %SO_RCVLOWAT setting
166   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
167   *	@sk_sndtimeo: %SO_SNDTIMEO setting
168   *	@sk_filter: socket filtering instructions
169   *	@sk_protinfo: private area, net family specific, when not using slab
170   *	@sk_timer: sock cleanup timer
171   *	@sk_stamp: time stamp of last packet received
172   *	@sk_socket: Identd and reporting IO signals
173   *	@sk_user_data: RPC layer private data
174   *	@sk_sndmsg_page: cached page for sendmsg
175   *	@sk_sndmsg_off: cached offset for sendmsg
176   *	@sk_send_head: front of stuff to transmit
177   *	@sk_security: used by security modules
178   *	@sk_write_pending: a write to stream socket waits to start
179   *	@sk_state_change: callback to indicate change in the state of the sock
180   *	@sk_data_ready: callback to indicate there is data to be processed
181   *	@sk_write_space: callback to indicate there is bf sending space available
182   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
183   *	@sk_backlog_rcv: callback to process the backlog
184   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
185  */
186 struct sock {
187 	/*
188 	 * Now struct inet_timewait_sock also uses sock_common, so please just
189 	 * don't add nothing before this first member (__sk_common) --acme
190 	 */
191 	struct sock_common	__sk_common;
192 #define sk_family		__sk_common.skc_family
193 #define sk_state		__sk_common.skc_state
194 #define sk_reuse		__sk_common.skc_reuse
195 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
196 #define sk_node			__sk_common.skc_node
197 #define sk_bind_node		__sk_common.skc_bind_node
198 #define sk_refcnt		__sk_common.skc_refcnt
199 #define sk_hash			__sk_common.skc_hash
200 #define sk_prot			__sk_common.skc_prot
201 #define sk_net			__sk_common.skc_net
202 	unsigned char		sk_shutdown : 2,
203 				sk_no_check : 2,
204 				sk_userlocks : 4;
205 	unsigned char		sk_protocol;
206 	unsigned short		sk_type;
207 	int			sk_rcvbuf;
208 	socket_lock_t		sk_lock;
209 	/*
210 	 * The backlog queue is special, it is always used with
211 	 * the per-socket spinlock held and requires low latency
212 	 * access. Therefore we special case it's implementation.
213 	 */
214 	struct {
215 		struct sk_buff *head;
216 		struct sk_buff *tail;
217 	} sk_backlog;
218 	wait_queue_head_t	*sk_sleep;
219 	struct dst_entry	*sk_dst_cache;
220 	struct xfrm_policy	*sk_policy[2];
221 	rwlock_t		sk_dst_lock;
222 	atomic_t		sk_rmem_alloc;
223 	atomic_t		sk_wmem_alloc;
224 	atomic_t		sk_omem_alloc;
225 	int			sk_sndbuf;
226 	struct sk_buff_head	sk_receive_queue;
227 	struct sk_buff_head	sk_write_queue;
228 	struct sk_buff_head	sk_async_wait_queue;
229 	int			sk_wmem_queued;
230 	int			sk_forward_alloc;
231 	gfp_t			sk_allocation;
232 	int			sk_route_caps;
233 	int			sk_gso_type;
234 	int			sk_rcvlowat;
235 	unsigned long 		sk_flags;
236 	unsigned long	        sk_lingertime;
237 	struct sk_buff_head	sk_error_queue;
238 	struct proto		*sk_prot_creator;
239 	rwlock_t		sk_callback_lock;
240 	int			sk_err,
241 				sk_err_soft;
242 	unsigned short		sk_ack_backlog;
243 	unsigned short		sk_max_ack_backlog;
244 	__u32			sk_priority;
245 	struct ucred		sk_peercred;
246 	long			sk_rcvtimeo;
247 	long			sk_sndtimeo;
248 	struct sk_filter      	*sk_filter;
249 	void			*sk_protinfo;
250 	struct timer_list	sk_timer;
251 	ktime_t			sk_stamp;
252 	struct socket		*sk_socket;
253 	void			*sk_user_data;
254 	struct page		*sk_sndmsg_page;
255 	struct sk_buff		*sk_send_head;
256 	__u32			sk_sndmsg_off;
257 	int			sk_write_pending;
258 	void			*sk_security;
259 	void			(*sk_state_change)(struct sock *sk);
260 	void			(*sk_data_ready)(struct sock *sk, int bytes);
261 	void			(*sk_write_space)(struct sock *sk);
262 	void			(*sk_error_report)(struct sock *sk);
263   	int			(*sk_backlog_rcv)(struct sock *sk,
264 						  struct sk_buff *skb);
265 	void                    (*sk_destruct)(struct sock *sk);
266 };
267 
268 /*
269  * Hashed lists helper routines
270  */
271 static inline struct sock *__sk_head(const struct hlist_head *head)
272 {
273 	return hlist_entry(head->first, struct sock, sk_node);
274 }
275 
276 static inline struct sock *sk_head(const struct hlist_head *head)
277 {
278 	return hlist_empty(head) ? NULL : __sk_head(head);
279 }
280 
281 static inline struct sock *sk_next(const struct sock *sk)
282 {
283 	return sk->sk_node.next ?
284 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
285 }
286 
287 static inline int sk_unhashed(const struct sock *sk)
288 {
289 	return hlist_unhashed(&sk->sk_node);
290 }
291 
292 static inline int sk_hashed(const struct sock *sk)
293 {
294 	return !sk_unhashed(sk);
295 }
296 
297 static __inline__ void sk_node_init(struct hlist_node *node)
298 {
299 	node->pprev = NULL;
300 }
301 
302 static __inline__ void __sk_del_node(struct sock *sk)
303 {
304 	__hlist_del(&sk->sk_node);
305 }
306 
307 static __inline__ int __sk_del_node_init(struct sock *sk)
308 {
309 	if (sk_hashed(sk)) {
310 		__sk_del_node(sk);
311 		sk_node_init(&sk->sk_node);
312 		return 1;
313 	}
314 	return 0;
315 }
316 
317 /* Grab socket reference count. This operation is valid only
318    when sk is ALREADY grabbed f.e. it is found in hash table
319    or a list and the lookup is made under lock preventing hash table
320    modifications.
321  */
322 
323 static inline void sock_hold(struct sock *sk)
324 {
325 	atomic_inc(&sk->sk_refcnt);
326 }
327 
328 /* Ungrab socket in the context, which assumes that socket refcnt
329    cannot hit zero, f.e. it is true in context of any socketcall.
330  */
331 static inline void __sock_put(struct sock *sk)
332 {
333 	atomic_dec(&sk->sk_refcnt);
334 }
335 
336 static __inline__ int sk_del_node_init(struct sock *sk)
337 {
338 	int rc = __sk_del_node_init(sk);
339 
340 	if (rc) {
341 		/* paranoid for a while -acme */
342 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
343 		__sock_put(sk);
344 	}
345 	return rc;
346 }
347 
348 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
349 {
350 	hlist_add_head(&sk->sk_node, list);
351 }
352 
353 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
354 {
355 	sock_hold(sk);
356 	__sk_add_node(sk, list);
357 }
358 
359 static __inline__ void __sk_del_bind_node(struct sock *sk)
360 {
361 	__hlist_del(&sk->sk_bind_node);
362 }
363 
364 static __inline__ void sk_add_bind_node(struct sock *sk,
365 					struct hlist_head *list)
366 {
367 	hlist_add_head(&sk->sk_bind_node, list);
368 }
369 
370 #define sk_for_each(__sk, node, list) \
371 	hlist_for_each_entry(__sk, node, list, sk_node)
372 #define sk_for_each_from(__sk, node) \
373 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
374 		hlist_for_each_entry_from(__sk, node, sk_node)
375 #define sk_for_each_continue(__sk, node) \
376 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
377 		hlist_for_each_entry_continue(__sk, node, sk_node)
378 #define sk_for_each_safe(__sk, node, tmp, list) \
379 	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
380 #define sk_for_each_bound(__sk, node, list) \
381 	hlist_for_each_entry(__sk, node, list, sk_bind_node)
382 
383 /* Sock flags */
384 enum sock_flags {
385 	SOCK_DEAD,
386 	SOCK_DONE,
387 	SOCK_URGINLINE,
388 	SOCK_KEEPOPEN,
389 	SOCK_LINGER,
390 	SOCK_DESTROY,
391 	SOCK_BROADCAST,
392 	SOCK_TIMESTAMP,
393 	SOCK_ZAPPED,
394 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
395 	SOCK_DBG, /* %SO_DEBUG setting */
396 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
397 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
398 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
399 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
400 };
401 
402 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
403 {
404 	nsk->sk_flags = osk->sk_flags;
405 }
406 
407 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
408 {
409 	__set_bit(flag, &sk->sk_flags);
410 }
411 
412 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
413 {
414 	__clear_bit(flag, &sk->sk_flags);
415 }
416 
417 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
418 {
419 	return test_bit(flag, &sk->sk_flags);
420 }
421 
422 static inline void sk_acceptq_removed(struct sock *sk)
423 {
424 	sk->sk_ack_backlog--;
425 }
426 
427 static inline void sk_acceptq_added(struct sock *sk)
428 {
429 	sk->sk_ack_backlog++;
430 }
431 
432 static inline int sk_acceptq_is_full(struct sock *sk)
433 {
434 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
435 }
436 
437 /*
438  * Compute minimal free write space needed to queue new packets.
439  */
440 static inline int sk_stream_min_wspace(struct sock *sk)
441 {
442 	return sk->sk_wmem_queued / 2;
443 }
444 
445 static inline int sk_stream_wspace(struct sock *sk)
446 {
447 	return sk->sk_sndbuf - sk->sk_wmem_queued;
448 }
449 
450 extern void sk_stream_write_space(struct sock *sk);
451 
452 static inline int sk_stream_memory_free(struct sock *sk)
453 {
454 	return sk->sk_wmem_queued < sk->sk_sndbuf;
455 }
456 
457 extern void sk_stream_rfree(struct sk_buff *skb);
458 
459 static inline void sk_stream_set_owner_r(struct sk_buff *skb, struct sock *sk)
460 {
461 	skb->sk = sk;
462 	skb->destructor = sk_stream_rfree;
463 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
464 	sk->sk_forward_alloc -= skb->truesize;
465 }
466 
467 static inline void sk_stream_free_skb(struct sock *sk, struct sk_buff *skb)
468 {
469 	skb_truesize_check(skb);
470 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
471 	sk->sk_wmem_queued   -= skb->truesize;
472 	sk->sk_forward_alloc += skb->truesize;
473 	__kfree_skb(skb);
474 }
475 
476 /* The per-socket spinlock must be held here. */
477 static inline void sk_add_backlog(struct sock *sk, struct sk_buff *skb)
478 {
479 	if (!sk->sk_backlog.tail) {
480 		sk->sk_backlog.head = sk->sk_backlog.tail = skb;
481 	} else {
482 		sk->sk_backlog.tail->next = skb;
483 		sk->sk_backlog.tail = skb;
484 	}
485 	skb->next = NULL;
486 }
487 
488 #define sk_wait_event(__sk, __timeo, __condition)			\
489 	({	int __rc;						\
490 		release_sock(__sk);					\
491 		__rc = __condition;					\
492 		if (!__rc) {						\
493 			*(__timeo) = schedule_timeout(*(__timeo));	\
494 		}							\
495 		lock_sock(__sk);					\
496 		__rc = __condition;					\
497 		__rc;							\
498 	})
499 
500 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
501 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
502 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
503 extern int sk_stream_error(struct sock *sk, int flags, int err);
504 extern void sk_stream_kill_queues(struct sock *sk);
505 
506 extern int sk_wait_data(struct sock *sk, long *timeo);
507 
508 struct request_sock_ops;
509 struct timewait_sock_ops;
510 
511 /* Networking protocol blocks we attach to sockets.
512  * socket layer -> transport layer interface
513  * transport -> network interface is defined by struct inet_proto
514  */
515 struct proto {
516 	void			(*close)(struct sock *sk,
517 					long timeout);
518 	int			(*connect)(struct sock *sk,
519 				        struct sockaddr *uaddr,
520 					int addr_len);
521 	int			(*disconnect)(struct sock *sk, int flags);
522 
523 	struct sock *		(*accept) (struct sock *sk, int flags, int *err);
524 
525 	int			(*ioctl)(struct sock *sk, int cmd,
526 					 unsigned long arg);
527 	int			(*init)(struct sock *sk);
528 	int			(*destroy)(struct sock *sk);
529 	void			(*shutdown)(struct sock *sk, int how);
530 	int			(*setsockopt)(struct sock *sk, int level,
531 					int optname, char __user *optval,
532 					int optlen);
533 	int			(*getsockopt)(struct sock *sk, int level,
534 					int optname, char __user *optval,
535 					int __user *option);
536 	int			(*compat_setsockopt)(struct sock *sk,
537 					int level,
538 					int optname, char __user *optval,
539 					int optlen);
540 	int			(*compat_getsockopt)(struct sock *sk,
541 					int level,
542 					int optname, char __user *optval,
543 					int __user *option);
544 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
545 					   struct msghdr *msg, size_t len);
546 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
547 					   struct msghdr *msg,
548 					size_t len, int noblock, int flags,
549 					int *addr_len);
550 	int			(*sendpage)(struct sock *sk, struct page *page,
551 					int offset, size_t size, int flags);
552 	int			(*bind)(struct sock *sk,
553 					struct sockaddr *uaddr, int addr_len);
554 
555 	int			(*backlog_rcv) (struct sock *sk,
556 						struct sk_buff *skb);
557 
558 	/* Keeping track of sk's, looking them up, and port selection methods. */
559 	void			(*hash)(struct sock *sk);
560 	void			(*unhash)(struct sock *sk);
561 	int			(*get_port)(struct sock *sk, unsigned short snum);
562 
563 #ifdef CONFIG_SMP
564 	/* Keeping track of sockets in use */
565 	void			(*inuse_add)(struct proto *prot, int inc);
566 	int			(*inuse_getval)(const struct proto *prot);
567 	int			*inuse_ptr;
568 #else
569 	int			inuse;
570 #endif
571 	/* Memory pressure */
572 	void			(*enter_memory_pressure)(void);
573 	atomic_t		*memory_allocated;	/* Current allocated memory. */
574 	atomic_t		*sockets_allocated;	/* Current number of sockets. */
575 	/*
576 	 * Pressure flag: try to collapse.
577 	 * Technical note: it is used by multiple contexts non atomically.
578 	 * All the sk_stream_mem_schedule() is of this nature: accounting
579 	 * is strict, actions are advisory and have some latency.
580 	 */
581 	int			*memory_pressure;
582 	int			*sysctl_mem;
583 	int			*sysctl_wmem;
584 	int			*sysctl_rmem;
585 	int			max_header;
586 
587 	struct kmem_cache		*slab;
588 	unsigned int		obj_size;
589 
590 	atomic_t		*orphan_count;
591 
592 	struct request_sock_ops	*rsk_prot;
593 	struct timewait_sock_ops *twsk_prot;
594 
595 	struct module		*owner;
596 
597 	char			name[32];
598 
599 	struct list_head	node;
600 #ifdef SOCK_REFCNT_DEBUG
601 	atomic_t		socks;
602 #endif
603 };
604 
605 /*
606  * Special macros to let protos use a fast version of inuse{get|add}
607  * using a static percpu variable per proto instead of an allocated one,
608  * saving one dereference.
609  * This might be changed if/when dynamic percpu vars become fast.
610  */
611 #ifdef CONFIG_SMP
612 # define DEFINE_PROTO_INUSE(NAME)			\
613 static DEFINE_PER_CPU(int, NAME##_inuse);		\
614 static void NAME##_inuse_add(struct proto *prot, int inc)	\
615 {							\
616 	__get_cpu_var(NAME##_inuse) += inc;		\
617 }							\
618 							\
619 static int NAME##_inuse_getval(const struct proto *prot)\
620 {							\
621 	int res = 0, cpu;				\
622 							\
623 	for_each_possible_cpu(cpu)			\
624 		res += per_cpu(NAME##_inuse, cpu);	\
625 	return res;					\
626 }
627 # define REF_PROTO_INUSE(NAME)				\
628 	.inuse_add = NAME##_inuse_add,			\
629 	.inuse_getval = NAME##_inuse_getval,
630 #else
631 # define DEFINE_PROTO_INUSE(NAME)
632 # define REF_PROTO_INUSE(NAME)
633 #endif
634 
635 extern int proto_register(struct proto *prot, int alloc_slab);
636 extern void proto_unregister(struct proto *prot);
637 
638 #ifdef SOCK_REFCNT_DEBUG
639 static inline void sk_refcnt_debug_inc(struct sock *sk)
640 {
641 	atomic_inc(&sk->sk_prot->socks);
642 }
643 
644 static inline void sk_refcnt_debug_dec(struct sock *sk)
645 {
646 	atomic_dec(&sk->sk_prot->socks);
647 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
648 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
649 }
650 
651 static inline void sk_refcnt_debug_release(const struct sock *sk)
652 {
653 	if (atomic_read(&sk->sk_refcnt) != 1)
654 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
655 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
656 }
657 #else /* SOCK_REFCNT_DEBUG */
658 #define sk_refcnt_debug_inc(sk) do { } while (0)
659 #define sk_refcnt_debug_dec(sk) do { } while (0)
660 #define sk_refcnt_debug_release(sk) do { } while (0)
661 #endif /* SOCK_REFCNT_DEBUG */
662 
663 /* Called with local bh disabled */
664 static __inline__ void sock_prot_inc_use(struct proto *prot)
665 {
666 #ifdef CONFIG_SMP
667 	prot->inuse_add(prot, 1);
668 #else
669 	prot->inuse++;
670 #endif
671 }
672 
673 static __inline__ void sock_prot_dec_use(struct proto *prot)
674 {
675 #ifdef CONFIG_SMP
676 	prot->inuse_add(prot, -1);
677 #else
678 	prot->inuse--;
679 #endif
680 }
681 
682 static __inline__ int sock_prot_inuse(struct proto *proto)
683 {
684 #ifdef CONFIG_SMP
685 	return proto->inuse_getval(proto);
686 #else
687 	return proto->inuse;
688 #endif
689 }
690 
691 /* With per-bucket locks this operation is not-atomic, so that
692  * this version is not worse.
693  */
694 static inline void __sk_prot_rehash(struct sock *sk)
695 {
696 	sk->sk_prot->unhash(sk);
697 	sk->sk_prot->hash(sk);
698 }
699 
700 /* About 10 seconds */
701 #define SOCK_DESTROY_TIME (10*HZ)
702 
703 /* Sockets 0-1023 can't be bound to unless you are superuser */
704 #define PROT_SOCK	1024
705 
706 #define SHUTDOWN_MASK	3
707 #define RCV_SHUTDOWN	1
708 #define SEND_SHUTDOWN	2
709 
710 #define SOCK_SNDBUF_LOCK	1
711 #define SOCK_RCVBUF_LOCK	2
712 #define SOCK_BINDADDR_LOCK	4
713 #define SOCK_BINDPORT_LOCK	8
714 
715 /* sock_iocb: used to kick off async processing of socket ios */
716 struct sock_iocb {
717 	struct list_head	list;
718 
719 	int			flags;
720 	int			size;
721 	struct socket		*sock;
722 	struct sock		*sk;
723 	struct scm_cookie	*scm;
724 	struct msghdr		*msg, async_msg;
725 	struct kiocb		*kiocb;
726 };
727 
728 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
729 {
730 	return (struct sock_iocb *)iocb->private;
731 }
732 
733 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
734 {
735 	return si->kiocb;
736 }
737 
738 struct socket_alloc {
739 	struct socket socket;
740 	struct inode vfs_inode;
741 };
742 
743 static inline struct socket *SOCKET_I(struct inode *inode)
744 {
745 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
746 }
747 
748 static inline struct inode *SOCK_INODE(struct socket *socket)
749 {
750 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
751 }
752 
753 extern void __sk_stream_mem_reclaim(struct sock *sk);
754 extern int sk_stream_mem_schedule(struct sock *sk, int size, int kind);
755 
756 #define SK_STREAM_MEM_QUANTUM ((int)PAGE_SIZE)
757 
758 static inline int sk_stream_pages(int amt)
759 {
760 	return DIV_ROUND_UP(amt, SK_STREAM_MEM_QUANTUM);
761 }
762 
763 static inline void sk_stream_mem_reclaim(struct sock *sk)
764 {
765 	if (sk->sk_forward_alloc >= SK_STREAM_MEM_QUANTUM)
766 		__sk_stream_mem_reclaim(sk);
767 }
768 
769 static inline int sk_stream_rmem_schedule(struct sock *sk, struct sk_buff *skb)
770 {
771 	return (int)skb->truesize <= sk->sk_forward_alloc ||
772 		sk_stream_mem_schedule(sk, skb->truesize, 1);
773 }
774 
775 static inline int sk_stream_wmem_schedule(struct sock *sk, int size)
776 {
777 	return size <= sk->sk_forward_alloc ||
778 	       sk_stream_mem_schedule(sk, size, 0);
779 }
780 
781 /* Used by processes to "lock" a socket state, so that
782  * interrupts and bottom half handlers won't change it
783  * from under us. It essentially blocks any incoming
784  * packets, so that we won't get any new data or any
785  * packets that change the state of the socket.
786  *
787  * While locked, BH processing will add new packets to
788  * the backlog queue.  This queue is processed by the
789  * owner of the socket lock right before it is released.
790  *
791  * Since ~2.3.5 it is also exclusive sleep lock serializing
792  * accesses from user process context.
793  */
794 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
795 
796 /*
797  * Macro so as to not evaluate some arguments when
798  * lockdep is not enabled.
799  *
800  * Mark both the sk_lock and the sk_lock.slock as a
801  * per-address-family lock class.
802  */
803 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) 	\
804 do {									\
805 	sk->sk_lock.owned = 0;					\
806 	init_waitqueue_head(&sk->sk_lock.wq);				\
807 	spin_lock_init(&(sk)->sk_lock.slock);				\
808 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
809 			sizeof((sk)->sk_lock));				\
810 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
811 		       	(skey), (sname));				\
812 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
813 } while (0)
814 
815 extern void FASTCALL(lock_sock_nested(struct sock *sk, int subclass));
816 
817 static inline void lock_sock(struct sock *sk)
818 {
819 	lock_sock_nested(sk, 0);
820 }
821 
822 extern void FASTCALL(release_sock(struct sock *sk));
823 
824 /* BH context may only use the following locking interface. */
825 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
826 #define bh_lock_sock_nested(__sk) \
827 				spin_lock_nested(&((__sk)->sk_lock.slock), \
828 				SINGLE_DEPTH_NESTING)
829 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
830 
831 extern struct sock		*sk_alloc(struct net *net, int family,
832 					  gfp_t priority,
833 					  struct proto *prot);
834 extern void			sk_free(struct sock *sk);
835 extern struct sock		*sk_clone(const struct sock *sk,
836 					  const gfp_t priority);
837 
838 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
839 					      unsigned long size, int force,
840 					      gfp_t priority);
841 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
842 					      unsigned long size, int force,
843 					      gfp_t priority);
844 extern void			sock_wfree(struct sk_buff *skb);
845 extern void			sock_rfree(struct sk_buff *skb);
846 
847 extern int			sock_setsockopt(struct socket *sock, int level,
848 						int op, char __user *optval,
849 						int optlen);
850 
851 extern int			sock_getsockopt(struct socket *sock, int level,
852 						int op, char __user *optval,
853 						int __user *optlen);
854 extern struct sk_buff 		*sock_alloc_send_skb(struct sock *sk,
855 						     unsigned long size,
856 						     int noblock,
857 						     int *errcode);
858 extern void *sock_kmalloc(struct sock *sk, int size,
859 			  gfp_t priority);
860 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
861 extern void sk_send_sigurg(struct sock *sk);
862 
863 /*
864  * Functions to fill in entries in struct proto_ops when a protocol
865  * does not implement a particular function.
866  */
867 extern int                      sock_no_bind(struct socket *,
868 					     struct sockaddr *, int);
869 extern int                      sock_no_connect(struct socket *,
870 						struct sockaddr *, int, int);
871 extern int                      sock_no_socketpair(struct socket *,
872 						   struct socket *);
873 extern int                      sock_no_accept(struct socket *,
874 					       struct socket *, int);
875 extern int                      sock_no_getname(struct socket *,
876 						struct sockaddr *, int *, int);
877 extern unsigned int             sock_no_poll(struct file *, struct socket *,
878 					     struct poll_table_struct *);
879 extern int                      sock_no_ioctl(struct socket *, unsigned int,
880 					      unsigned long);
881 extern int			sock_no_listen(struct socket *, int);
882 extern int                      sock_no_shutdown(struct socket *, int);
883 extern int			sock_no_getsockopt(struct socket *, int , int,
884 						   char __user *, int __user *);
885 extern int			sock_no_setsockopt(struct socket *, int, int,
886 						   char __user *, int);
887 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
888 						struct msghdr *, size_t);
889 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
890 						struct msghdr *, size_t, int);
891 extern int			sock_no_mmap(struct file *file,
892 					     struct socket *sock,
893 					     struct vm_area_struct *vma);
894 extern ssize_t			sock_no_sendpage(struct socket *sock,
895 						struct page *page,
896 						int offset, size_t size,
897 						int flags);
898 
899 /*
900  * Functions to fill in entries in struct proto_ops when a protocol
901  * uses the inet style.
902  */
903 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
904 				  char __user *optval, int __user *optlen);
905 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
906 			       struct msghdr *msg, size_t size, int flags);
907 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
908 				  char __user *optval, int optlen);
909 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
910 		int optname, char __user *optval, int __user *optlen);
911 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
912 		int optname, char __user *optval, int optlen);
913 
914 extern void sk_common_release(struct sock *sk);
915 
916 /*
917  *	Default socket callbacks and setup code
918  */
919 
920 /* Initialise core socket variables */
921 extern void sock_init_data(struct socket *sock, struct sock *sk);
922 
923 /**
924  *	sk_filter - run a packet through a socket filter
925  *	@sk: sock associated with &sk_buff
926  *	@skb: buffer to filter
927  *	@needlock: set to 1 if the sock is not locked by caller.
928  *
929  * Run the filter code and then cut skb->data to correct size returned by
930  * sk_run_filter. If pkt_len is 0 we toss packet. If skb->len is smaller
931  * than pkt_len we keep whole skb->data. This is the socket level
932  * wrapper to sk_run_filter. It returns 0 if the packet should
933  * be accepted or -EPERM if the packet should be tossed.
934  *
935  */
936 
937 static inline int sk_filter(struct sock *sk, struct sk_buff *skb)
938 {
939 	int err;
940 	struct sk_filter *filter;
941 
942 	err = security_sock_rcv_skb(sk, skb);
943 	if (err)
944 		return err;
945 
946 	rcu_read_lock_bh();
947 	filter = rcu_dereference(sk->sk_filter);
948 	if (filter) {
949 		unsigned int pkt_len = sk_run_filter(skb, filter->insns,
950 				filter->len);
951 		err = pkt_len ? pskb_trim(skb, pkt_len) : -EPERM;
952 	}
953  	rcu_read_unlock_bh();
954 
955 	return err;
956 }
957 
958 /**
959  *	sk_filter_release: Release a socket filter
960  *	@sk: socket
961  *	@fp: filter to remove
962  *
963  *	Remove a filter from a socket and release its resources.
964  */
965 
966 static inline void sk_filter_release(struct sk_filter *fp)
967 {
968 	if (atomic_dec_and_test(&fp->refcnt))
969 		kfree(fp);
970 }
971 
972 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
973 {
974 	unsigned int size = sk_filter_len(fp);
975 
976 	atomic_sub(size, &sk->sk_omem_alloc);
977 	sk_filter_release(fp);
978 }
979 
980 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
981 {
982 	atomic_inc(&fp->refcnt);
983 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
984 }
985 
986 /*
987  * Socket reference counting postulates.
988  *
989  * * Each user of socket SHOULD hold a reference count.
990  * * Each access point to socket (an hash table bucket, reference from a list,
991  *   running timer, skb in flight MUST hold a reference count.
992  * * When reference count hits 0, it means it will never increase back.
993  * * When reference count hits 0, it means that no references from
994  *   outside exist to this socket and current process on current CPU
995  *   is last user and may/should destroy this socket.
996  * * sk_free is called from any context: process, BH, IRQ. When
997  *   it is called, socket has no references from outside -> sk_free
998  *   may release descendant resources allocated by the socket, but
999  *   to the time when it is called, socket is NOT referenced by any
1000  *   hash tables, lists etc.
1001  * * Packets, delivered from outside (from network or from another process)
1002  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1003  *   when they sit in queue. Otherwise, packets will leak to hole, when
1004  *   socket is looked up by one cpu and unhasing is made by another CPU.
1005  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1006  *   (leak to backlog). Packet socket does all the processing inside
1007  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1008  *   use separate SMP lock, so that they are prone too.
1009  */
1010 
1011 /* Ungrab socket and destroy it, if it was the last reference. */
1012 static inline void sock_put(struct sock *sk)
1013 {
1014 	if (atomic_dec_and_test(&sk->sk_refcnt))
1015 		sk_free(sk);
1016 }
1017 
1018 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1019 			  const int nested);
1020 
1021 /* Detach socket from process context.
1022  * Announce socket dead, detach it from wait queue and inode.
1023  * Note that parent inode held reference count on this struct sock,
1024  * we do not release it in this function, because protocol
1025  * probably wants some additional cleanups or even continuing
1026  * to work with this socket (TCP).
1027  */
1028 static inline void sock_orphan(struct sock *sk)
1029 {
1030 	write_lock_bh(&sk->sk_callback_lock);
1031 	sock_set_flag(sk, SOCK_DEAD);
1032 	sk->sk_socket = NULL;
1033 	sk->sk_sleep  = NULL;
1034 	write_unlock_bh(&sk->sk_callback_lock);
1035 }
1036 
1037 static inline void sock_graft(struct sock *sk, struct socket *parent)
1038 {
1039 	write_lock_bh(&sk->sk_callback_lock);
1040 	sk->sk_sleep = &parent->wait;
1041 	parent->sk = sk;
1042 	sk->sk_socket = parent;
1043 	security_sock_graft(sk, parent);
1044 	write_unlock_bh(&sk->sk_callback_lock);
1045 }
1046 
1047 extern int sock_i_uid(struct sock *sk);
1048 extern unsigned long sock_i_ino(struct sock *sk);
1049 
1050 static inline struct dst_entry *
1051 __sk_dst_get(struct sock *sk)
1052 {
1053 	return sk->sk_dst_cache;
1054 }
1055 
1056 static inline struct dst_entry *
1057 sk_dst_get(struct sock *sk)
1058 {
1059 	struct dst_entry *dst;
1060 
1061 	read_lock(&sk->sk_dst_lock);
1062 	dst = sk->sk_dst_cache;
1063 	if (dst)
1064 		dst_hold(dst);
1065 	read_unlock(&sk->sk_dst_lock);
1066 	return dst;
1067 }
1068 
1069 static inline void
1070 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1071 {
1072 	struct dst_entry *old_dst;
1073 
1074 	old_dst = sk->sk_dst_cache;
1075 	sk->sk_dst_cache = dst;
1076 	dst_release(old_dst);
1077 }
1078 
1079 static inline void
1080 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1081 {
1082 	write_lock(&sk->sk_dst_lock);
1083 	__sk_dst_set(sk, dst);
1084 	write_unlock(&sk->sk_dst_lock);
1085 }
1086 
1087 static inline void
1088 __sk_dst_reset(struct sock *sk)
1089 {
1090 	struct dst_entry *old_dst;
1091 
1092 	old_dst = sk->sk_dst_cache;
1093 	sk->sk_dst_cache = NULL;
1094 	dst_release(old_dst);
1095 }
1096 
1097 static inline void
1098 sk_dst_reset(struct sock *sk)
1099 {
1100 	write_lock(&sk->sk_dst_lock);
1101 	__sk_dst_reset(sk);
1102 	write_unlock(&sk->sk_dst_lock);
1103 }
1104 
1105 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1106 
1107 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1108 
1109 static inline int sk_can_gso(const struct sock *sk)
1110 {
1111 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1112 }
1113 
1114 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1115 
1116 static inline void sk_charge_skb(struct sock *sk, struct sk_buff *skb)
1117 {
1118 	sk->sk_wmem_queued   += skb->truesize;
1119 	sk->sk_forward_alloc -= skb->truesize;
1120 }
1121 
1122 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1123 				   struct sk_buff *skb, struct page *page,
1124 				   int off, int copy)
1125 {
1126 	if (skb->ip_summed == CHECKSUM_NONE) {
1127 		int err = 0;
1128 		__wsum csum = csum_and_copy_from_user(from,
1129 						     page_address(page) + off,
1130 							    copy, 0, &err);
1131 		if (err)
1132 			return err;
1133 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1134 	} else if (copy_from_user(page_address(page) + off, from, copy))
1135 		return -EFAULT;
1136 
1137 	skb->len	     += copy;
1138 	skb->data_len	     += copy;
1139 	skb->truesize	     += copy;
1140 	sk->sk_wmem_queued   += copy;
1141 	sk->sk_forward_alloc -= copy;
1142 	return 0;
1143 }
1144 
1145 /*
1146  * 	Queue a received datagram if it will fit. Stream and sequenced
1147  *	protocols can't normally use this as they need to fit buffers in
1148  *	and play with them.
1149  *
1150  * 	Inlined as it's very short and called for pretty much every
1151  *	packet ever received.
1152  */
1153 
1154 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1155 {
1156 	sock_hold(sk);
1157 	skb->sk = sk;
1158 	skb->destructor = sock_wfree;
1159 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1160 }
1161 
1162 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1163 {
1164 	skb->sk = sk;
1165 	skb->destructor = sock_rfree;
1166 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1167 }
1168 
1169 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1170 			   unsigned long expires);
1171 
1172 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1173 
1174 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1175 
1176 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
1177 {
1178 	/* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
1179 	   number of warnings when compiling with -W --ANK
1180 	 */
1181 	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1182 	    (unsigned)sk->sk_rcvbuf)
1183 		return -ENOMEM;
1184 	skb_set_owner_r(skb, sk);
1185 	skb_queue_tail(&sk->sk_error_queue, skb);
1186 	if (!sock_flag(sk, SOCK_DEAD))
1187 		sk->sk_data_ready(sk, skb->len);
1188 	return 0;
1189 }
1190 
1191 /*
1192  *	Recover an error report and clear atomically
1193  */
1194 
1195 static inline int sock_error(struct sock *sk)
1196 {
1197 	int err;
1198 	if (likely(!sk->sk_err))
1199 		return 0;
1200 	err = xchg(&sk->sk_err, 0);
1201 	return -err;
1202 }
1203 
1204 static inline unsigned long sock_wspace(struct sock *sk)
1205 {
1206 	int amt = 0;
1207 
1208 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1209 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1210 		if (amt < 0)
1211 			amt = 0;
1212 	}
1213 	return amt;
1214 }
1215 
1216 static inline void sk_wake_async(struct sock *sk, int how, int band)
1217 {
1218 	if (sk->sk_socket && sk->sk_socket->fasync_list)
1219 		sock_wake_async(sk->sk_socket, how, band);
1220 }
1221 
1222 #define SOCK_MIN_SNDBUF 2048
1223 #define SOCK_MIN_RCVBUF 256
1224 
1225 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1226 {
1227 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1228 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued / 2);
1229 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1230 	}
1231 }
1232 
1233 static inline struct sk_buff *sk_stream_alloc_pskb(struct sock *sk,
1234 						   int size, int mem,
1235 						   gfp_t gfp)
1236 {
1237 	struct sk_buff *skb;
1238 
1239 	/* The TCP header must be at least 32-bit aligned.  */
1240 	size = ALIGN(size, 4);
1241 
1242 	skb = alloc_skb_fclone(size + sk->sk_prot->max_header, gfp);
1243 	if (skb) {
1244 		skb->truesize += mem;
1245 		if (sk_stream_wmem_schedule(sk, skb->truesize)) {
1246 			/*
1247 			 * Make sure that we have exactly size bytes
1248 			 * available to the caller, no more, no less.
1249 			 */
1250 			skb_reserve(skb, skb_tailroom(skb) - size);
1251 			return skb;
1252 		}
1253 		__kfree_skb(skb);
1254 	} else {
1255 		sk->sk_prot->enter_memory_pressure();
1256 		sk_stream_moderate_sndbuf(sk);
1257 	}
1258 	return NULL;
1259 }
1260 
1261 static inline struct sk_buff *sk_stream_alloc_skb(struct sock *sk,
1262 						  int size,
1263 						  gfp_t gfp)
1264 {
1265 	return sk_stream_alloc_pskb(sk, size, 0, gfp);
1266 }
1267 
1268 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1269 {
1270 	struct page *page = NULL;
1271 
1272 	page = alloc_pages(sk->sk_allocation, 0);
1273 	if (!page) {
1274 		sk->sk_prot->enter_memory_pressure();
1275 		sk_stream_moderate_sndbuf(sk);
1276 	}
1277 	return page;
1278 }
1279 
1280 /*
1281  *	Default write policy as shown to user space via poll/select/SIGIO
1282  */
1283 static inline int sock_writeable(const struct sock *sk)
1284 {
1285 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf / 2);
1286 }
1287 
1288 static inline gfp_t gfp_any(void)
1289 {
1290 	return in_atomic() ? GFP_ATOMIC : GFP_KERNEL;
1291 }
1292 
1293 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1294 {
1295 	return noblock ? 0 : sk->sk_rcvtimeo;
1296 }
1297 
1298 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1299 {
1300 	return noblock ? 0 : sk->sk_sndtimeo;
1301 }
1302 
1303 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1304 {
1305 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1306 }
1307 
1308 /* Alas, with timeout socket operations are not restartable.
1309  * Compare this to poll().
1310  */
1311 static inline int sock_intr_errno(long timeo)
1312 {
1313 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1314 }
1315 
1316 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1317 	struct sk_buff *skb);
1318 
1319 static __inline__ void
1320 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1321 {
1322 	ktime_t kt = skb->tstamp;
1323 
1324 	if (sock_flag(sk, SOCK_RCVTSTAMP))
1325 		__sock_recv_timestamp(msg, sk, skb);
1326 	else
1327 		sk->sk_stamp = kt;
1328 }
1329 
1330 /**
1331  * sk_eat_skb - Release a skb if it is no longer needed
1332  * @sk: socket to eat this skb from
1333  * @skb: socket buffer to eat
1334  * @copied_early: flag indicating whether DMA operations copied this data early
1335  *
1336  * This routine must be called with interrupts disabled or with the socket
1337  * locked so that the sk_buff queue operation is ok.
1338 */
1339 #ifdef CONFIG_NET_DMA
1340 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1341 {
1342 	__skb_unlink(skb, &sk->sk_receive_queue);
1343 	if (!copied_early)
1344 		__kfree_skb(skb);
1345 	else
1346 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
1347 }
1348 #else
1349 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1350 {
1351 	__skb_unlink(skb, &sk->sk_receive_queue);
1352 	__kfree_skb(skb);
1353 }
1354 #endif
1355 
1356 extern void sock_enable_timestamp(struct sock *sk);
1357 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1358 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1359 
1360 /*
1361  *	Enable debug/info messages
1362  */
1363 extern int net_msg_warn;
1364 #define NETDEBUG(fmt, args...) \
1365 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
1366 
1367 #define LIMIT_NETDEBUG(fmt, args...) \
1368 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
1369 
1370 /*
1371  * Macros for sleeping on a socket. Use them like this:
1372  *
1373  * SOCK_SLEEP_PRE(sk)
1374  * if (condition)
1375  * 	schedule();
1376  * SOCK_SLEEP_POST(sk)
1377  *
1378  * N.B. These are now obsolete and were, afaik, only ever used in DECnet
1379  * and when the last use of them in DECnet has gone, I'm intending to
1380  * remove them.
1381  */
1382 
1383 #define SOCK_SLEEP_PRE(sk) 	{ struct task_struct *tsk = current; \
1384 				DECLARE_WAITQUEUE(wait, tsk); \
1385 				tsk->state = TASK_INTERRUPTIBLE; \
1386 				add_wait_queue((sk)->sk_sleep, &wait); \
1387 				release_sock(sk);
1388 
1389 #define SOCK_SLEEP_POST(sk)	tsk->state = TASK_RUNNING; \
1390 				remove_wait_queue((sk)->sk_sleep, &wait); \
1391 				lock_sock(sk); \
1392 				}
1393 
1394 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
1395 {
1396 	if (valbool)
1397 		sock_set_flag(sk, bit);
1398 	else
1399 		sock_reset_flag(sk, bit);
1400 }
1401 
1402 extern __u32 sysctl_wmem_max;
1403 extern __u32 sysctl_rmem_max;
1404 
1405 extern void sk_init(void);
1406 
1407 #ifdef CONFIG_SYSCTL
1408 extern struct ctl_table core_table[];
1409 #endif
1410 
1411 extern int sysctl_optmem_max;
1412 
1413 extern __u32 sysctl_wmem_default;
1414 extern __u32 sysctl_rmem_default;
1415 
1416 #endif	/* _SOCK_H */
1417