xref: /linux/include/net/sock.h (revision a33f32244d8550da8b4a26e277ce07d5c6d158b5)
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/list_nulls.h>
46 #include <linux/timer.h>
47 #include <linux/cache.h>
48 #include <linux/module.h>
49 #include <linux/lockdep.h>
50 #include <linux/netdevice.h>
51 #include <linux/skbuff.h>	/* struct sk_buff */
52 #include <linux/mm.h>
53 #include <linux/security.h>
54 #include <linux/slab.h>
55 
56 #include <linux/filter.h>
57 #include <linux/rculist_nulls.h>
58 #include <linux/poll.h>
59 
60 #include <asm/atomic.h>
61 #include <net/dst.h>
62 #include <net/checksum.h>
63 
64 /*
65  * This structure really needs to be cleaned up.
66  * Most of it is for TCP, and not used by any of
67  * the other protocols.
68  */
69 
70 /* Define this to get the SOCK_DBG debugging facility. */
71 #define SOCK_DEBUGGING
72 #ifdef SOCK_DEBUGGING
73 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
74 					printk(KERN_DEBUG msg); } while (0)
75 #else
76 /* Validate arguments and do nothing */
77 static void inline int __attribute__ ((format (printf, 2, 3)))
78 SOCK_DEBUG(struct sock *sk, const char *msg, ...)
79 {
80 }
81 #endif
82 
83 /* This is the per-socket lock.  The spinlock provides a synchronization
84  * between user contexts and software interrupt processing, whereas the
85  * mini-semaphore synchronizes multiple users amongst themselves.
86  */
87 typedef struct {
88 	spinlock_t		slock;
89 	int			owned;
90 	wait_queue_head_t	wq;
91 	/*
92 	 * We express the mutex-alike socket_lock semantics
93 	 * to the lock validator by explicitly managing
94 	 * the slock as a lock variant (in addition to
95 	 * the slock itself):
96 	 */
97 #ifdef CONFIG_DEBUG_LOCK_ALLOC
98 	struct lockdep_map dep_map;
99 #endif
100 } socket_lock_t;
101 
102 struct sock;
103 struct proto;
104 struct net;
105 
106 /**
107  *	struct sock_common - minimal network layer representation of sockets
108  *	@skc_node: main hash linkage for various protocol lookup tables
109  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
110  *	@skc_refcnt: reference count
111  *	@skc_tx_queue_mapping: tx queue number for this connection
112  *	@skc_hash: hash value used with various protocol lookup tables
113  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
114  *	@skc_family: network address family
115  *	@skc_state: Connection state
116  *	@skc_reuse: %SO_REUSEADDR setting
117  *	@skc_bound_dev_if: bound device index if != 0
118  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
119  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
120  *	@skc_prot: protocol handlers inside a network family
121  *	@skc_net: reference to the network namespace of this socket
122  *
123  *	This is the minimal network layer representation of sockets, the header
124  *	for struct sock and struct inet_timewait_sock.
125  */
126 struct sock_common {
127 	/*
128 	 * first fields are not copied in sock_copy()
129 	 */
130 	union {
131 		struct hlist_node	skc_node;
132 		struct hlist_nulls_node skc_nulls_node;
133 	};
134 	atomic_t		skc_refcnt;
135 	int			skc_tx_queue_mapping;
136 
137 	union  {
138 		unsigned int	skc_hash;
139 		__u16		skc_u16hashes[2];
140 	};
141 	unsigned short		skc_family;
142 	volatile unsigned char	skc_state;
143 	unsigned char		skc_reuse;
144 	int			skc_bound_dev_if;
145 	union {
146 		struct hlist_node	skc_bind_node;
147 		struct hlist_nulls_node skc_portaddr_node;
148 	};
149 	struct proto		*skc_prot;
150 #ifdef CONFIG_NET_NS
151 	struct net	 	*skc_net;
152 #endif
153 };
154 
155 /**
156   *	struct sock - network layer representation of sockets
157   *	@__sk_common: shared layout with inet_timewait_sock
158   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
159   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
160   *	@sk_lock:	synchronizer
161   *	@sk_rcvbuf: size of receive buffer in bytes
162   *	@sk_sleep: sock wait queue
163   *	@sk_dst_cache: destination cache
164   *	@sk_dst_lock: destination cache lock
165   *	@sk_policy: flow policy
166   *	@sk_rmem_alloc: receive queue bytes committed
167   *	@sk_receive_queue: incoming packets
168   *	@sk_wmem_alloc: transmit queue bytes committed
169   *	@sk_write_queue: Packet sending queue
170   *	@sk_async_wait_queue: DMA copied packets
171   *	@sk_omem_alloc: "o" is "option" or "other"
172   *	@sk_wmem_queued: persistent queue size
173   *	@sk_forward_alloc: space allocated forward
174   *	@sk_allocation: allocation mode
175   *	@sk_sndbuf: size of send buffer in bytes
176   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
177   *		   %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
178   *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
179   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
180   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
181   *	@sk_gso_max_size: Maximum GSO segment size to build
182   *	@sk_lingertime: %SO_LINGER l_linger setting
183   *	@sk_backlog: always used with the per-socket spinlock held
184   *	@sk_callback_lock: used with the callbacks in the end of this struct
185   *	@sk_error_queue: rarely used
186   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
187   *			  IPV6_ADDRFORM for instance)
188   *	@sk_err: last error
189   *	@sk_err_soft: errors that don't cause failure but are the cause of a
190   *		      persistent failure not just 'timed out'
191   *	@sk_drops: raw/udp drops counter
192   *	@sk_ack_backlog: current listen backlog
193   *	@sk_max_ack_backlog: listen backlog set in listen()
194   *	@sk_priority: %SO_PRIORITY setting
195   *	@sk_type: socket type (%SOCK_STREAM, etc)
196   *	@sk_protocol: which protocol this socket belongs in this network family
197   *	@sk_peercred: %SO_PEERCRED setting
198   *	@sk_rcvlowat: %SO_RCVLOWAT setting
199   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
200   *	@sk_sndtimeo: %SO_SNDTIMEO setting
201   *	@sk_filter: socket filtering instructions
202   *	@sk_protinfo: private area, net family specific, when not using slab
203   *	@sk_timer: sock cleanup timer
204   *	@sk_stamp: time stamp of last packet received
205   *	@sk_socket: Identd and reporting IO signals
206   *	@sk_user_data: RPC layer private data
207   *	@sk_sndmsg_page: cached page for sendmsg
208   *	@sk_sndmsg_off: cached offset for sendmsg
209   *	@sk_send_head: front of stuff to transmit
210   *	@sk_security: used by security modules
211   *	@sk_mark: generic packet mark
212   *	@sk_write_pending: a write to stream socket waits to start
213   *	@sk_state_change: callback to indicate change in the state of the sock
214   *	@sk_data_ready: callback to indicate there is data to be processed
215   *	@sk_write_space: callback to indicate there is bf sending space available
216   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
217   *	@sk_backlog_rcv: callback to process the backlog
218   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
219  */
220 struct sock {
221 	/*
222 	 * Now struct inet_timewait_sock also uses sock_common, so please just
223 	 * don't add nothing before this first member (__sk_common) --acme
224 	 */
225 	struct sock_common	__sk_common;
226 #define sk_node			__sk_common.skc_node
227 #define sk_nulls_node		__sk_common.skc_nulls_node
228 #define sk_refcnt		__sk_common.skc_refcnt
229 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
230 
231 #define sk_copy_start		__sk_common.skc_hash
232 #define sk_hash			__sk_common.skc_hash
233 #define sk_family		__sk_common.skc_family
234 #define sk_state		__sk_common.skc_state
235 #define sk_reuse		__sk_common.skc_reuse
236 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
237 #define sk_bind_node		__sk_common.skc_bind_node
238 #define sk_prot			__sk_common.skc_prot
239 #define sk_net			__sk_common.skc_net
240 	kmemcheck_bitfield_begin(flags);
241 	unsigned int		sk_shutdown  : 2,
242 				sk_no_check  : 2,
243 				sk_userlocks : 4,
244 				sk_protocol  : 8,
245 				sk_type      : 16;
246 	kmemcheck_bitfield_end(flags);
247 	int			sk_rcvbuf;
248 	socket_lock_t		sk_lock;
249 	/*
250 	 * The backlog queue is special, it is always used with
251 	 * the per-socket spinlock held and requires low latency
252 	 * access. Therefore we special case it's implementation.
253 	 */
254 	struct {
255 		struct sk_buff *head;
256 		struct sk_buff *tail;
257 		int len;
258 		int limit;
259 	} sk_backlog;
260 	wait_queue_head_t	*sk_sleep;
261 	struct dst_entry	*sk_dst_cache;
262 #ifdef CONFIG_XFRM
263 	struct xfrm_policy	*sk_policy[2];
264 #endif
265 	rwlock_t		sk_dst_lock;
266 	atomic_t		sk_rmem_alloc;
267 	atomic_t		sk_wmem_alloc;
268 	atomic_t		sk_omem_alloc;
269 	int			sk_sndbuf;
270 	struct sk_buff_head	sk_receive_queue;
271 	struct sk_buff_head	sk_write_queue;
272 #ifdef CONFIG_NET_DMA
273 	struct sk_buff_head	sk_async_wait_queue;
274 #endif
275 	int			sk_wmem_queued;
276 	int			sk_forward_alloc;
277 	gfp_t			sk_allocation;
278 	int			sk_route_caps;
279 	int			sk_gso_type;
280 	unsigned int		sk_gso_max_size;
281 	int			sk_rcvlowat;
282 	unsigned long 		sk_flags;
283 	unsigned long	        sk_lingertime;
284 	struct sk_buff_head	sk_error_queue;
285 	struct proto		*sk_prot_creator;
286 	rwlock_t		sk_callback_lock;
287 	int			sk_err,
288 				sk_err_soft;
289 	atomic_t		sk_drops;
290 	unsigned short		sk_ack_backlog;
291 	unsigned short		sk_max_ack_backlog;
292 	__u32			sk_priority;
293 	struct ucred		sk_peercred;
294 	long			sk_rcvtimeo;
295 	long			sk_sndtimeo;
296 	struct sk_filter      	*sk_filter;
297 	void			*sk_protinfo;
298 	struct timer_list	sk_timer;
299 	ktime_t			sk_stamp;
300 	struct socket		*sk_socket;
301 	void			*sk_user_data;
302 	struct page		*sk_sndmsg_page;
303 	struct sk_buff		*sk_send_head;
304 	__u32			sk_sndmsg_off;
305 	int			sk_write_pending;
306 #ifdef CONFIG_SECURITY
307 	void			*sk_security;
308 #endif
309 	__u32			sk_mark;
310 	/* XXX 4 bytes hole on 64 bit */
311 	void			(*sk_state_change)(struct sock *sk);
312 	void			(*sk_data_ready)(struct sock *sk, int bytes);
313 	void			(*sk_write_space)(struct sock *sk);
314 	void			(*sk_error_report)(struct sock *sk);
315   	int			(*sk_backlog_rcv)(struct sock *sk,
316 						  struct sk_buff *skb);
317 	void                    (*sk_destruct)(struct sock *sk);
318 };
319 
320 /*
321  * Hashed lists helper routines
322  */
323 static inline struct sock *sk_entry(const struct hlist_node *node)
324 {
325 	return hlist_entry(node, struct sock, sk_node);
326 }
327 
328 static inline struct sock *__sk_head(const struct hlist_head *head)
329 {
330 	return hlist_entry(head->first, struct sock, sk_node);
331 }
332 
333 static inline struct sock *sk_head(const struct hlist_head *head)
334 {
335 	return hlist_empty(head) ? NULL : __sk_head(head);
336 }
337 
338 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
339 {
340 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
341 }
342 
343 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
344 {
345 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
346 }
347 
348 static inline struct sock *sk_next(const struct sock *sk)
349 {
350 	return sk->sk_node.next ?
351 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
352 }
353 
354 static inline struct sock *sk_nulls_next(const struct sock *sk)
355 {
356 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
357 		hlist_nulls_entry(sk->sk_nulls_node.next,
358 				  struct sock, sk_nulls_node) :
359 		NULL;
360 }
361 
362 static inline int sk_unhashed(const struct sock *sk)
363 {
364 	return hlist_unhashed(&sk->sk_node);
365 }
366 
367 static inline int sk_hashed(const struct sock *sk)
368 {
369 	return !sk_unhashed(sk);
370 }
371 
372 static __inline__ void sk_node_init(struct hlist_node *node)
373 {
374 	node->pprev = NULL;
375 }
376 
377 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
378 {
379 	node->pprev = NULL;
380 }
381 
382 static __inline__ void __sk_del_node(struct sock *sk)
383 {
384 	__hlist_del(&sk->sk_node);
385 }
386 
387 /* NB: equivalent to hlist_del_init_rcu */
388 static __inline__ int __sk_del_node_init(struct sock *sk)
389 {
390 	if (sk_hashed(sk)) {
391 		__sk_del_node(sk);
392 		sk_node_init(&sk->sk_node);
393 		return 1;
394 	}
395 	return 0;
396 }
397 
398 /* Grab socket reference count. This operation is valid only
399    when sk is ALREADY grabbed f.e. it is found in hash table
400    or a list and the lookup is made under lock preventing hash table
401    modifications.
402  */
403 
404 static inline void sock_hold(struct sock *sk)
405 {
406 	atomic_inc(&sk->sk_refcnt);
407 }
408 
409 /* Ungrab socket in the context, which assumes that socket refcnt
410    cannot hit zero, f.e. it is true in context of any socketcall.
411  */
412 static inline void __sock_put(struct sock *sk)
413 {
414 	atomic_dec(&sk->sk_refcnt);
415 }
416 
417 static __inline__ int sk_del_node_init(struct sock *sk)
418 {
419 	int rc = __sk_del_node_init(sk);
420 
421 	if (rc) {
422 		/* paranoid for a while -acme */
423 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
424 		__sock_put(sk);
425 	}
426 	return rc;
427 }
428 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
429 
430 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
431 {
432 	if (sk_hashed(sk)) {
433 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
434 		return 1;
435 	}
436 	return 0;
437 }
438 
439 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
440 {
441 	int rc = __sk_nulls_del_node_init_rcu(sk);
442 
443 	if (rc) {
444 		/* paranoid for a while -acme */
445 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
446 		__sock_put(sk);
447 	}
448 	return rc;
449 }
450 
451 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
452 {
453 	hlist_add_head(&sk->sk_node, list);
454 }
455 
456 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
457 {
458 	sock_hold(sk);
459 	__sk_add_node(sk, list);
460 }
461 
462 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
463 {
464 	sock_hold(sk);
465 	hlist_add_head_rcu(&sk->sk_node, list);
466 }
467 
468 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
469 {
470 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
471 }
472 
473 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
474 {
475 	sock_hold(sk);
476 	__sk_nulls_add_node_rcu(sk, list);
477 }
478 
479 static __inline__ void __sk_del_bind_node(struct sock *sk)
480 {
481 	__hlist_del(&sk->sk_bind_node);
482 }
483 
484 static __inline__ void sk_add_bind_node(struct sock *sk,
485 					struct hlist_head *list)
486 {
487 	hlist_add_head(&sk->sk_bind_node, list);
488 }
489 
490 #define sk_for_each(__sk, node, list) \
491 	hlist_for_each_entry(__sk, node, list, sk_node)
492 #define sk_for_each_rcu(__sk, node, list) \
493 	hlist_for_each_entry_rcu(__sk, node, list, sk_node)
494 #define sk_nulls_for_each(__sk, node, list) \
495 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
496 #define sk_nulls_for_each_rcu(__sk, node, list) \
497 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
498 #define sk_for_each_from(__sk, node) \
499 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
500 		hlist_for_each_entry_from(__sk, node, sk_node)
501 #define sk_nulls_for_each_from(__sk, node) \
502 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
503 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
504 #define sk_for_each_continue(__sk, node) \
505 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
506 		hlist_for_each_entry_continue(__sk, node, sk_node)
507 #define sk_for_each_safe(__sk, node, tmp, list) \
508 	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
509 #define sk_for_each_bound(__sk, node, list) \
510 	hlist_for_each_entry(__sk, node, list, sk_bind_node)
511 
512 /* Sock flags */
513 enum sock_flags {
514 	SOCK_DEAD,
515 	SOCK_DONE,
516 	SOCK_URGINLINE,
517 	SOCK_KEEPOPEN,
518 	SOCK_LINGER,
519 	SOCK_DESTROY,
520 	SOCK_BROADCAST,
521 	SOCK_TIMESTAMP,
522 	SOCK_ZAPPED,
523 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
524 	SOCK_DBG, /* %SO_DEBUG setting */
525 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
526 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
527 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
528 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
529 	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
530 	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
531 	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
532 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
533 	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
534 	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
535 	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
536 	SOCK_FASYNC, /* fasync() active */
537 	SOCK_RXQ_OVFL,
538 };
539 
540 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
541 {
542 	nsk->sk_flags = osk->sk_flags;
543 }
544 
545 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
546 {
547 	__set_bit(flag, &sk->sk_flags);
548 }
549 
550 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
551 {
552 	__clear_bit(flag, &sk->sk_flags);
553 }
554 
555 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
556 {
557 	return test_bit(flag, &sk->sk_flags);
558 }
559 
560 static inline void sk_acceptq_removed(struct sock *sk)
561 {
562 	sk->sk_ack_backlog--;
563 }
564 
565 static inline void sk_acceptq_added(struct sock *sk)
566 {
567 	sk->sk_ack_backlog++;
568 }
569 
570 static inline int sk_acceptq_is_full(struct sock *sk)
571 {
572 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
573 }
574 
575 /*
576  * Compute minimal free write space needed to queue new packets.
577  */
578 static inline int sk_stream_min_wspace(struct sock *sk)
579 {
580 	return sk->sk_wmem_queued >> 1;
581 }
582 
583 static inline int sk_stream_wspace(struct sock *sk)
584 {
585 	return sk->sk_sndbuf - sk->sk_wmem_queued;
586 }
587 
588 extern void sk_stream_write_space(struct sock *sk);
589 
590 static inline int sk_stream_memory_free(struct sock *sk)
591 {
592 	return sk->sk_wmem_queued < sk->sk_sndbuf;
593 }
594 
595 /* OOB backlog add */
596 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
597 {
598 	if (!sk->sk_backlog.tail) {
599 		sk->sk_backlog.head = sk->sk_backlog.tail = skb;
600 	} else {
601 		sk->sk_backlog.tail->next = skb;
602 		sk->sk_backlog.tail = skb;
603 	}
604 	skb->next = NULL;
605 }
606 
607 /* The per-socket spinlock must be held here. */
608 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
609 {
610 	if (sk->sk_backlog.len >= max(sk->sk_backlog.limit, sk->sk_rcvbuf << 1))
611 		return -ENOBUFS;
612 
613 	__sk_add_backlog(sk, skb);
614 	sk->sk_backlog.len += skb->truesize;
615 	return 0;
616 }
617 
618 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
619 {
620 	return sk->sk_backlog_rcv(sk, skb);
621 }
622 
623 #define sk_wait_event(__sk, __timeo, __condition)			\
624 	({	int __rc;						\
625 		release_sock(__sk);					\
626 		__rc = __condition;					\
627 		if (!__rc) {						\
628 			*(__timeo) = schedule_timeout(*(__timeo));	\
629 		}							\
630 		lock_sock(__sk);					\
631 		__rc = __condition;					\
632 		__rc;							\
633 	})
634 
635 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
636 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
637 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
638 extern int sk_stream_error(struct sock *sk, int flags, int err);
639 extern void sk_stream_kill_queues(struct sock *sk);
640 
641 extern int sk_wait_data(struct sock *sk, long *timeo);
642 
643 struct request_sock_ops;
644 struct timewait_sock_ops;
645 struct inet_hashinfo;
646 struct raw_hashinfo;
647 
648 /* Networking protocol blocks we attach to sockets.
649  * socket layer -> transport layer interface
650  * transport -> network interface is defined by struct inet_proto
651  */
652 struct proto {
653 	void			(*close)(struct sock *sk,
654 					long timeout);
655 	int			(*connect)(struct sock *sk,
656 				        struct sockaddr *uaddr,
657 					int addr_len);
658 	int			(*disconnect)(struct sock *sk, int flags);
659 
660 	struct sock *		(*accept) (struct sock *sk, int flags, int *err);
661 
662 	int			(*ioctl)(struct sock *sk, int cmd,
663 					 unsigned long arg);
664 	int			(*init)(struct sock *sk);
665 	void			(*destroy)(struct sock *sk);
666 	void			(*shutdown)(struct sock *sk, int how);
667 	int			(*setsockopt)(struct sock *sk, int level,
668 					int optname, char __user *optval,
669 					unsigned int optlen);
670 	int			(*getsockopt)(struct sock *sk, int level,
671 					int optname, char __user *optval,
672 					int __user *option);
673 #ifdef CONFIG_COMPAT
674 	int			(*compat_setsockopt)(struct sock *sk,
675 					int level,
676 					int optname, char __user *optval,
677 					unsigned int optlen);
678 	int			(*compat_getsockopt)(struct sock *sk,
679 					int level,
680 					int optname, char __user *optval,
681 					int __user *option);
682 #endif
683 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
684 					   struct msghdr *msg, size_t len);
685 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
686 					   struct msghdr *msg,
687 					size_t len, int noblock, int flags,
688 					int *addr_len);
689 	int			(*sendpage)(struct sock *sk, struct page *page,
690 					int offset, size_t size, int flags);
691 	int			(*bind)(struct sock *sk,
692 					struct sockaddr *uaddr, int addr_len);
693 
694 	int			(*backlog_rcv) (struct sock *sk,
695 						struct sk_buff *skb);
696 
697 	/* Keeping track of sk's, looking them up, and port selection methods. */
698 	void			(*hash)(struct sock *sk);
699 	void			(*unhash)(struct sock *sk);
700 	int			(*get_port)(struct sock *sk, unsigned short snum);
701 
702 	/* Keeping track of sockets in use */
703 #ifdef CONFIG_PROC_FS
704 	unsigned int		inuse_idx;
705 #endif
706 
707 	/* Memory pressure */
708 	void			(*enter_memory_pressure)(struct sock *sk);
709 	atomic_t		*memory_allocated;	/* Current allocated memory. */
710 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
711 	/*
712 	 * Pressure flag: try to collapse.
713 	 * Technical note: it is used by multiple contexts non atomically.
714 	 * All the __sk_mem_schedule() is of this nature: accounting
715 	 * is strict, actions are advisory and have some latency.
716 	 */
717 	int			*memory_pressure;
718 	int			*sysctl_mem;
719 	int			*sysctl_wmem;
720 	int			*sysctl_rmem;
721 	int			max_header;
722 
723 	struct kmem_cache	*slab;
724 	unsigned int		obj_size;
725 	int			slab_flags;
726 
727 	struct percpu_counter	*orphan_count;
728 
729 	struct request_sock_ops	*rsk_prot;
730 	struct timewait_sock_ops *twsk_prot;
731 
732 	union {
733 		struct inet_hashinfo	*hashinfo;
734 		struct udp_table	*udp_table;
735 		struct raw_hashinfo	*raw_hash;
736 	} h;
737 
738 	struct module		*owner;
739 
740 	char			name[32];
741 
742 	struct list_head	node;
743 #ifdef SOCK_REFCNT_DEBUG
744 	atomic_t		socks;
745 #endif
746 };
747 
748 extern int proto_register(struct proto *prot, int alloc_slab);
749 extern void proto_unregister(struct proto *prot);
750 
751 #ifdef SOCK_REFCNT_DEBUG
752 static inline void sk_refcnt_debug_inc(struct sock *sk)
753 {
754 	atomic_inc(&sk->sk_prot->socks);
755 }
756 
757 static inline void sk_refcnt_debug_dec(struct sock *sk)
758 {
759 	atomic_dec(&sk->sk_prot->socks);
760 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
761 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
762 }
763 
764 static inline void sk_refcnt_debug_release(const struct sock *sk)
765 {
766 	if (atomic_read(&sk->sk_refcnt) != 1)
767 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
768 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
769 }
770 #else /* SOCK_REFCNT_DEBUG */
771 #define sk_refcnt_debug_inc(sk) do { } while (0)
772 #define sk_refcnt_debug_dec(sk) do { } while (0)
773 #define sk_refcnt_debug_release(sk) do { } while (0)
774 #endif /* SOCK_REFCNT_DEBUG */
775 
776 
777 #ifdef CONFIG_PROC_FS
778 /* Called with local bh disabled */
779 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
780 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
781 #else
782 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
783 		int inc)
784 {
785 }
786 #endif
787 
788 
789 /* With per-bucket locks this operation is not-atomic, so that
790  * this version is not worse.
791  */
792 static inline void __sk_prot_rehash(struct sock *sk)
793 {
794 	sk->sk_prot->unhash(sk);
795 	sk->sk_prot->hash(sk);
796 }
797 
798 /* About 10 seconds */
799 #define SOCK_DESTROY_TIME (10*HZ)
800 
801 /* Sockets 0-1023 can't be bound to unless you are superuser */
802 #define PROT_SOCK	1024
803 
804 #define SHUTDOWN_MASK	3
805 #define RCV_SHUTDOWN	1
806 #define SEND_SHUTDOWN	2
807 
808 #define SOCK_SNDBUF_LOCK	1
809 #define SOCK_RCVBUF_LOCK	2
810 #define SOCK_BINDADDR_LOCK	4
811 #define SOCK_BINDPORT_LOCK	8
812 
813 /* sock_iocb: used to kick off async processing of socket ios */
814 struct sock_iocb {
815 	struct list_head	list;
816 
817 	int			flags;
818 	int			size;
819 	struct socket		*sock;
820 	struct sock		*sk;
821 	struct scm_cookie	*scm;
822 	struct msghdr		*msg, async_msg;
823 	struct kiocb		*kiocb;
824 };
825 
826 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
827 {
828 	return (struct sock_iocb *)iocb->private;
829 }
830 
831 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
832 {
833 	return si->kiocb;
834 }
835 
836 struct socket_alloc {
837 	struct socket socket;
838 	struct inode vfs_inode;
839 };
840 
841 static inline struct socket *SOCKET_I(struct inode *inode)
842 {
843 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
844 }
845 
846 static inline struct inode *SOCK_INODE(struct socket *socket)
847 {
848 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
849 }
850 
851 /*
852  * Functions for memory accounting
853  */
854 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
855 extern void __sk_mem_reclaim(struct sock *sk);
856 
857 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
858 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
859 #define SK_MEM_SEND	0
860 #define SK_MEM_RECV	1
861 
862 static inline int sk_mem_pages(int amt)
863 {
864 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
865 }
866 
867 static inline int sk_has_account(struct sock *sk)
868 {
869 	/* return true if protocol supports memory accounting */
870 	return !!sk->sk_prot->memory_allocated;
871 }
872 
873 static inline int sk_wmem_schedule(struct sock *sk, int size)
874 {
875 	if (!sk_has_account(sk))
876 		return 1;
877 	return size <= sk->sk_forward_alloc ||
878 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
879 }
880 
881 static inline int sk_rmem_schedule(struct sock *sk, int size)
882 {
883 	if (!sk_has_account(sk))
884 		return 1;
885 	return size <= sk->sk_forward_alloc ||
886 		__sk_mem_schedule(sk, size, SK_MEM_RECV);
887 }
888 
889 static inline void sk_mem_reclaim(struct sock *sk)
890 {
891 	if (!sk_has_account(sk))
892 		return;
893 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
894 		__sk_mem_reclaim(sk);
895 }
896 
897 static inline void sk_mem_reclaim_partial(struct sock *sk)
898 {
899 	if (!sk_has_account(sk))
900 		return;
901 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
902 		__sk_mem_reclaim(sk);
903 }
904 
905 static inline void sk_mem_charge(struct sock *sk, int size)
906 {
907 	if (!sk_has_account(sk))
908 		return;
909 	sk->sk_forward_alloc -= size;
910 }
911 
912 static inline void sk_mem_uncharge(struct sock *sk, int size)
913 {
914 	if (!sk_has_account(sk))
915 		return;
916 	sk->sk_forward_alloc += size;
917 }
918 
919 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
920 {
921 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
922 	sk->sk_wmem_queued -= skb->truesize;
923 	sk_mem_uncharge(sk, skb->truesize);
924 	__kfree_skb(skb);
925 }
926 
927 /* Used by processes to "lock" a socket state, so that
928  * interrupts and bottom half handlers won't change it
929  * from under us. It essentially blocks any incoming
930  * packets, so that we won't get any new data or any
931  * packets that change the state of the socket.
932  *
933  * While locked, BH processing will add new packets to
934  * the backlog queue.  This queue is processed by the
935  * owner of the socket lock right before it is released.
936  *
937  * Since ~2.3.5 it is also exclusive sleep lock serializing
938  * accesses from user process context.
939  */
940 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
941 
942 /*
943  * Macro so as to not evaluate some arguments when
944  * lockdep is not enabled.
945  *
946  * Mark both the sk_lock and the sk_lock.slock as a
947  * per-address-family lock class.
948  */
949 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) 	\
950 do {									\
951 	sk->sk_lock.owned = 0;						\
952 	init_waitqueue_head(&sk->sk_lock.wq);				\
953 	spin_lock_init(&(sk)->sk_lock.slock);				\
954 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
955 			sizeof((sk)->sk_lock));				\
956 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
957 		       	(skey), (sname));				\
958 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
959 } while (0)
960 
961 extern void lock_sock_nested(struct sock *sk, int subclass);
962 
963 static inline void lock_sock(struct sock *sk)
964 {
965 	lock_sock_nested(sk, 0);
966 }
967 
968 extern void release_sock(struct sock *sk);
969 
970 /* BH context may only use the following locking interface. */
971 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
972 #define bh_lock_sock_nested(__sk) \
973 				spin_lock_nested(&((__sk)->sk_lock.slock), \
974 				SINGLE_DEPTH_NESTING)
975 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
976 
977 extern struct sock		*sk_alloc(struct net *net, int family,
978 					  gfp_t priority,
979 					  struct proto *prot);
980 extern void			sk_free(struct sock *sk);
981 extern void			sk_release_kernel(struct sock *sk);
982 extern struct sock		*sk_clone(const struct sock *sk,
983 					  const gfp_t priority);
984 
985 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
986 					      unsigned long size, int force,
987 					      gfp_t priority);
988 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
989 					      unsigned long size, int force,
990 					      gfp_t priority);
991 extern void			sock_wfree(struct sk_buff *skb);
992 extern void			sock_rfree(struct sk_buff *skb);
993 
994 extern int			sock_setsockopt(struct socket *sock, int level,
995 						int op, char __user *optval,
996 						unsigned int optlen);
997 
998 extern int			sock_getsockopt(struct socket *sock, int level,
999 						int op, char __user *optval,
1000 						int __user *optlen);
1001 extern struct sk_buff 		*sock_alloc_send_skb(struct sock *sk,
1002 						     unsigned long size,
1003 						     int noblock,
1004 						     int *errcode);
1005 extern struct sk_buff 		*sock_alloc_send_pskb(struct sock *sk,
1006 						      unsigned long header_len,
1007 						      unsigned long data_len,
1008 						      int noblock,
1009 						      int *errcode);
1010 extern void *sock_kmalloc(struct sock *sk, int size,
1011 			  gfp_t priority);
1012 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1013 extern void sk_send_sigurg(struct sock *sk);
1014 
1015 /*
1016  * Functions to fill in entries in struct proto_ops when a protocol
1017  * does not implement a particular function.
1018  */
1019 extern int                      sock_no_bind(struct socket *,
1020 					     struct sockaddr *, int);
1021 extern int                      sock_no_connect(struct socket *,
1022 						struct sockaddr *, int, int);
1023 extern int                      sock_no_socketpair(struct socket *,
1024 						   struct socket *);
1025 extern int                      sock_no_accept(struct socket *,
1026 					       struct socket *, int);
1027 extern int                      sock_no_getname(struct socket *,
1028 						struct sockaddr *, int *, int);
1029 extern unsigned int             sock_no_poll(struct file *, struct socket *,
1030 					     struct poll_table_struct *);
1031 extern int                      sock_no_ioctl(struct socket *, unsigned int,
1032 					      unsigned long);
1033 extern int			sock_no_listen(struct socket *, int);
1034 extern int                      sock_no_shutdown(struct socket *, int);
1035 extern int			sock_no_getsockopt(struct socket *, int , int,
1036 						   char __user *, int __user *);
1037 extern int			sock_no_setsockopt(struct socket *, int, int,
1038 						   char __user *, unsigned int);
1039 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1040 						struct msghdr *, size_t);
1041 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1042 						struct msghdr *, size_t, int);
1043 extern int			sock_no_mmap(struct file *file,
1044 					     struct socket *sock,
1045 					     struct vm_area_struct *vma);
1046 extern ssize_t			sock_no_sendpage(struct socket *sock,
1047 						struct page *page,
1048 						int offset, size_t size,
1049 						int flags);
1050 
1051 /*
1052  * Functions to fill in entries in struct proto_ops when a protocol
1053  * uses the inet style.
1054  */
1055 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1056 				  char __user *optval, int __user *optlen);
1057 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1058 			       struct msghdr *msg, size_t size, int flags);
1059 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1060 				  char __user *optval, unsigned int optlen);
1061 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1062 		int optname, char __user *optval, int __user *optlen);
1063 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1064 		int optname, char __user *optval, unsigned int optlen);
1065 
1066 extern void sk_common_release(struct sock *sk);
1067 
1068 /*
1069  *	Default socket callbacks and setup code
1070  */
1071 
1072 /* Initialise core socket variables */
1073 extern void sock_init_data(struct socket *sock, struct sock *sk);
1074 
1075 /**
1076  *	sk_filter_release - release a socket filter
1077  *	@fp: filter to remove
1078  *
1079  *	Remove a filter from a socket and release its resources.
1080  */
1081 
1082 static inline void sk_filter_release(struct sk_filter *fp)
1083 {
1084 	if (atomic_dec_and_test(&fp->refcnt))
1085 		kfree(fp);
1086 }
1087 
1088 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1089 {
1090 	unsigned int size = sk_filter_len(fp);
1091 
1092 	atomic_sub(size, &sk->sk_omem_alloc);
1093 	sk_filter_release(fp);
1094 }
1095 
1096 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1097 {
1098 	atomic_inc(&fp->refcnt);
1099 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1100 }
1101 
1102 /*
1103  * Socket reference counting postulates.
1104  *
1105  * * Each user of socket SHOULD hold a reference count.
1106  * * Each access point to socket (an hash table bucket, reference from a list,
1107  *   running timer, skb in flight MUST hold a reference count.
1108  * * When reference count hits 0, it means it will never increase back.
1109  * * When reference count hits 0, it means that no references from
1110  *   outside exist to this socket and current process on current CPU
1111  *   is last user and may/should destroy this socket.
1112  * * sk_free is called from any context: process, BH, IRQ. When
1113  *   it is called, socket has no references from outside -> sk_free
1114  *   may release descendant resources allocated by the socket, but
1115  *   to the time when it is called, socket is NOT referenced by any
1116  *   hash tables, lists etc.
1117  * * Packets, delivered from outside (from network or from another process)
1118  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1119  *   when they sit in queue. Otherwise, packets will leak to hole, when
1120  *   socket is looked up by one cpu and unhasing is made by another CPU.
1121  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1122  *   (leak to backlog). Packet socket does all the processing inside
1123  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1124  *   use separate SMP lock, so that they are prone too.
1125  */
1126 
1127 /* Ungrab socket and destroy it, if it was the last reference. */
1128 static inline void sock_put(struct sock *sk)
1129 {
1130 	if (atomic_dec_and_test(&sk->sk_refcnt))
1131 		sk_free(sk);
1132 }
1133 
1134 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1135 			  const int nested);
1136 
1137 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1138 {
1139 	sk->sk_tx_queue_mapping = tx_queue;
1140 }
1141 
1142 static inline void sk_tx_queue_clear(struct sock *sk)
1143 {
1144 	sk->sk_tx_queue_mapping = -1;
1145 }
1146 
1147 static inline int sk_tx_queue_get(const struct sock *sk)
1148 {
1149 	return sk->sk_tx_queue_mapping;
1150 }
1151 
1152 static inline bool sk_tx_queue_recorded(const struct sock *sk)
1153 {
1154 	return (sk && sk->sk_tx_queue_mapping >= 0);
1155 }
1156 
1157 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1158 {
1159 	sk_tx_queue_clear(sk);
1160 	sk->sk_socket = sock;
1161 }
1162 
1163 /* Detach socket from process context.
1164  * Announce socket dead, detach it from wait queue and inode.
1165  * Note that parent inode held reference count on this struct sock,
1166  * we do not release it in this function, because protocol
1167  * probably wants some additional cleanups or even continuing
1168  * to work with this socket (TCP).
1169  */
1170 static inline void sock_orphan(struct sock *sk)
1171 {
1172 	write_lock_bh(&sk->sk_callback_lock);
1173 	sock_set_flag(sk, SOCK_DEAD);
1174 	sk_set_socket(sk, NULL);
1175 	sk->sk_sleep  = NULL;
1176 	write_unlock_bh(&sk->sk_callback_lock);
1177 }
1178 
1179 static inline void sock_graft(struct sock *sk, struct socket *parent)
1180 {
1181 	write_lock_bh(&sk->sk_callback_lock);
1182 	sk->sk_sleep = &parent->wait;
1183 	parent->sk = sk;
1184 	sk_set_socket(sk, parent);
1185 	security_sock_graft(sk, parent);
1186 	write_unlock_bh(&sk->sk_callback_lock);
1187 }
1188 
1189 extern int sock_i_uid(struct sock *sk);
1190 extern unsigned long sock_i_ino(struct sock *sk);
1191 
1192 static inline struct dst_entry *
1193 __sk_dst_get(struct sock *sk)
1194 {
1195 	return sk->sk_dst_cache;
1196 }
1197 
1198 static inline struct dst_entry *
1199 sk_dst_get(struct sock *sk)
1200 {
1201 	struct dst_entry *dst;
1202 
1203 	read_lock(&sk->sk_dst_lock);
1204 	dst = sk->sk_dst_cache;
1205 	if (dst)
1206 		dst_hold(dst);
1207 	read_unlock(&sk->sk_dst_lock);
1208 	return dst;
1209 }
1210 
1211 static inline void
1212 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1213 {
1214 	struct dst_entry *old_dst;
1215 
1216 	sk_tx_queue_clear(sk);
1217 	old_dst = sk->sk_dst_cache;
1218 	sk->sk_dst_cache = dst;
1219 	dst_release(old_dst);
1220 }
1221 
1222 static inline void
1223 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1224 {
1225 	write_lock(&sk->sk_dst_lock);
1226 	__sk_dst_set(sk, dst);
1227 	write_unlock(&sk->sk_dst_lock);
1228 }
1229 
1230 static inline void
1231 __sk_dst_reset(struct sock *sk)
1232 {
1233 	struct dst_entry *old_dst;
1234 
1235 	sk_tx_queue_clear(sk);
1236 	old_dst = sk->sk_dst_cache;
1237 	sk->sk_dst_cache = NULL;
1238 	dst_release(old_dst);
1239 }
1240 
1241 static inline void
1242 sk_dst_reset(struct sock *sk)
1243 {
1244 	write_lock(&sk->sk_dst_lock);
1245 	__sk_dst_reset(sk);
1246 	write_unlock(&sk->sk_dst_lock);
1247 }
1248 
1249 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1250 
1251 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1252 
1253 static inline int sk_can_gso(const struct sock *sk)
1254 {
1255 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1256 }
1257 
1258 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1259 
1260 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1261 				   struct sk_buff *skb, struct page *page,
1262 				   int off, int copy)
1263 {
1264 	if (skb->ip_summed == CHECKSUM_NONE) {
1265 		int err = 0;
1266 		__wsum csum = csum_and_copy_from_user(from,
1267 						     page_address(page) + off,
1268 							    copy, 0, &err);
1269 		if (err)
1270 			return err;
1271 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1272 	} else if (copy_from_user(page_address(page) + off, from, copy))
1273 		return -EFAULT;
1274 
1275 	skb->len	     += copy;
1276 	skb->data_len	     += copy;
1277 	skb->truesize	     += copy;
1278 	sk->sk_wmem_queued   += copy;
1279 	sk_mem_charge(sk, copy);
1280 	return 0;
1281 }
1282 
1283 /**
1284  * sk_wmem_alloc_get - returns write allocations
1285  * @sk: socket
1286  *
1287  * Returns sk_wmem_alloc minus initial offset of one
1288  */
1289 static inline int sk_wmem_alloc_get(const struct sock *sk)
1290 {
1291 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1292 }
1293 
1294 /**
1295  * sk_rmem_alloc_get - returns read allocations
1296  * @sk: socket
1297  *
1298  * Returns sk_rmem_alloc
1299  */
1300 static inline int sk_rmem_alloc_get(const struct sock *sk)
1301 {
1302 	return atomic_read(&sk->sk_rmem_alloc);
1303 }
1304 
1305 /**
1306  * sk_has_allocations - check if allocations are outstanding
1307  * @sk: socket
1308  *
1309  * Returns true if socket has write or read allocations
1310  */
1311 static inline int sk_has_allocations(const struct sock *sk)
1312 {
1313 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1314 }
1315 
1316 /**
1317  * sk_has_sleeper - check if there are any waiting processes
1318  * @sk: socket
1319  *
1320  * Returns true if socket has waiting processes
1321  *
1322  * The purpose of the sk_has_sleeper and sock_poll_wait is to wrap the memory
1323  * barrier call. They were added due to the race found within the tcp code.
1324  *
1325  * Consider following tcp code paths:
1326  *
1327  * CPU1                  CPU2
1328  *
1329  * sys_select            receive packet
1330  *   ...                 ...
1331  *   __add_wait_queue    update tp->rcv_nxt
1332  *   ...                 ...
1333  *   tp->rcv_nxt check   sock_def_readable
1334  *   ...                 {
1335  *   schedule               ...
1336  *                          if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1337  *                              wake_up_interruptible(sk->sk_sleep)
1338  *                          ...
1339  *                       }
1340  *
1341  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1342  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1343  * could then endup calling schedule and sleep forever if there are no more
1344  * data on the socket.
1345  *
1346  * The sk_has_sleeper is always called right after a call to read_lock, so we
1347  * can use smp_mb__after_lock barrier.
1348  */
1349 static inline int sk_has_sleeper(struct sock *sk)
1350 {
1351 	/*
1352 	 * We need to be sure we are in sync with the
1353 	 * add_wait_queue modifications to the wait queue.
1354 	 *
1355 	 * This memory barrier is paired in the sock_poll_wait.
1356 	 */
1357 	smp_mb__after_lock();
1358 	return sk->sk_sleep && waitqueue_active(sk->sk_sleep);
1359 }
1360 
1361 /**
1362  * sock_poll_wait - place memory barrier behind the poll_wait call.
1363  * @filp:           file
1364  * @wait_address:   socket wait queue
1365  * @p:              poll_table
1366  *
1367  * See the comments in the sk_has_sleeper function.
1368  */
1369 static inline void sock_poll_wait(struct file *filp,
1370 		wait_queue_head_t *wait_address, poll_table *p)
1371 {
1372 	if (p && wait_address) {
1373 		poll_wait(filp, wait_address, p);
1374 		/*
1375 		 * We need to be sure we are in sync with the
1376 		 * socket flags modification.
1377 		 *
1378 		 * This memory barrier is paired in the sk_has_sleeper.
1379 		*/
1380 		smp_mb();
1381 	}
1382 }
1383 
1384 /*
1385  * 	Queue a received datagram if it will fit. Stream and sequenced
1386  *	protocols can't normally use this as they need to fit buffers in
1387  *	and play with them.
1388  *
1389  * 	Inlined as it's very short and called for pretty much every
1390  *	packet ever received.
1391  */
1392 
1393 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1394 {
1395 	skb_orphan(skb);
1396 	skb->sk = sk;
1397 	skb->destructor = sock_wfree;
1398 	/*
1399 	 * We used to take a refcount on sk, but following operation
1400 	 * is enough to guarantee sk_free() wont free this sock until
1401 	 * all in-flight packets are completed
1402 	 */
1403 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1404 }
1405 
1406 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1407 {
1408 	skb_orphan(skb);
1409 	skb->sk = sk;
1410 	skb->destructor = sock_rfree;
1411 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1412 	sk_mem_charge(sk, skb->truesize);
1413 }
1414 
1415 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1416 			   unsigned long expires);
1417 
1418 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1419 
1420 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1421 
1422 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
1423 {
1424 	/* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
1425 	   number of warnings when compiling with -W --ANK
1426 	 */
1427 	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1428 	    (unsigned)sk->sk_rcvbuf)
1429 		return -ENOMEM;
1430 	skb_set_owner_r(skb, sk);
1431 	skb_queue_tail(&sk->sk_error_queue, skb);
1432 	if (!sock_flag(sk, SOCK_DEAD))
1433 		sk->sk_data_ready(sk, skb->len);
1434 	return 0;
1435 }
1436 
1437 /*
1438  *	Recover an error report and clear atomically
1439  */
1440 
1441 static inline int sock_error(struct sock *sk)
1442 {
1443 	int err;
1444 	if (likely(!sk->sk_err))
1445 		return 0;
1446 	err = xchg(&sk->sk_err, 0);
1447 	return -err;
1448 }
1449 
1450 static inline unsigned long sock_wspace(struct sock *sk)
1451 {
1452 	int amt = 0;
1453 
1454 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1455 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1456 		if (amt < 0)
1457 			amt = 0;
1458 	}
1459 	return amt;
1460 }
1461 
1462 static inline void sk_wake_async(struct sock *sk, int how, int band)
1463 {
1464 	if (sock_flag(sk, SOCK_FASYNC))
1465 		sock_wake_async(sk->sk_socket, how, band);
1466 }
1467 
1468 #define SOCK_MIN_SNDBUF 2048
1469 #define SOCK_MIN_RCVBUF 256
1470 
1471 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1472 {
1473 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1474 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1475 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1476 	}
1477 }
1478 
1479 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1480 
1481 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1482 {
1483 	struct page *page = NULL;
1484 
1485 	page = alloc_pages(sk->sk_allocation, 0);
1486 	if (!page) {
1487 		sk->sk_prot->enter_memory_pressure(sk);
1488 		sk_stream_moderate_sndbuf(sk);
1489 	}
1490 	return page;
1491 }
1492 
1493 /*
1494  *	Default write policy as shown to user space via poll/select/SIGIO
1495  */
1496 static inline int sock_writeable(const struct sock *sk)
1497 {
1498 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1499 }
1500 
1501 static inline gfp_t gfp_any(void)
1502 {
1503 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1504 }
1505 
1506 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1507 {
1508 	return noblock ? 0 : sk->sk_rcvtimeo;
1509 }
1510 
1511 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1512 {
1513 	return noblock ? 0 : sk->sk_sndtimeo;
1514 }
1515 
1516 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1517 {
1518 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1519 }
1520 
1521 /* Alas, with timeout socket operations are not restartable.
1522  * Compare this to poll().
1523  */
1524 static inline int sock_intr_errno(long timeo)
1525 {
1526 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1527 }
1528 
1529 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1530 	struct sk_buff *skb);
1531 
1532 static __inline__ void
1533 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1534 {
1535 	ktime_t kt = skb->tstamp;
1536 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1537 
1538 	/*
1539 	 * generate control messages if
1540 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
1541 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
1542 	 * - software time stamp available and wanted
1543 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
1544 	 * - hardware time stamps available and wanted
1545 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
1546 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
1547 	 */
1548 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
1549 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
1550 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
1551 	    (hwtstamps->hwtstamp.tv64 &&
1552 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
1553 	    (hwtstamps->syststamp.tv64 &&
1554 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
1555 		__sock_recv_timestamp(msg, sk, skb);
1556 	else
1557 		sk->sk_stamp = kt;
1558 }
1559 
1560 extern void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk, struct sk_buff *skb);
1561 
1562 /**
1563  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
1564  * @msg:	outgoing packet
1565  * @sk:		socket sending this packet
1566  * @shtx:	filled with instructions for time stamping
1567  *
1568  * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
1569  * parameters are invalid.
1570  */
1571 extern int sock_tx_timestamp(struct msghdr *msg,
1572 			     struct sock *sk,
1573 			     union skb_shared_tx *shtx);
1574 
1575 
1576 /**
1577  * sk_eat_skb - Release a skb if it is no longer needed
1578  * @sk: socket to eat this skb from
1579  * @skb: socket buffer to eat
1580  * @copied_early: flag indicating whether DMA operations copied this data early
1581  *
1582  * This routine must be called with interrupts disabled or with the socket
1583  * locked so that the sk_buff queue operation is ok.
1584 */
1585 #ifdef CONFIG_NET_DMA
1586 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1587 {
1588 	__skb_unlink(skb, &sk->sk_receive_queue);
1589 	if (!copied_early)
1590 		__kfree_skb(skb);
1591 	else
1592 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
1593 }
1594 #else
1595 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1596 {
1597 	__skb_unlink(skb, &sk->sk_receive_queue);
1598 	__kfree_skb(skb);
1599 }
1600 #endif
1601 
1602 static inline
1603 struct net *sock_net(const struct sock *sk)
1604 {
1605 #ifdef CONFIG_NET_NS
1606 	return sk->sk_net;
1607 #else
1608 	return &init_net;
1609 #endif
1610 }
1611 
1612 static inline
1613 void sock_net_set(struct sock *sk, struct net *net)
1614 {
1615 #ifdef CONFIG_NET_NS
1616 	sk->sk_net = net;
1617 #endif
1618 }
1619 
1620 /*
1621  * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
1622  * They should not hold a referrence to a namespace in order to allow
1623  * to stop it.
1624  * Sockets after sk_change_net should be released using sk_release_kernel
1625  */
1626 static inline void sk_change_net(struct sock *sk, struct net *net)
1627 {
1628 	put_net(sock_net(sk));
1629 	sock_net_set(sk, hold_net(net));
1630 }
1631 
1632 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
1633 {
1634 	if (unlikely(skb->sk)) {
1635 		struct sock *sk = skb->sk;
1636 
1637 		skb->destructor = NULL;
1638 		skb->sk = NULL;
1639 		return sk;
1640 	}
1641 	return NULL;
1642 }
1643 
1644 extern void sock_enable_timestamp(struct sock *sk, int flag);
1645 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1646 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1647 
1648 /*
1649  *	Enable debug/info messages
1650  */
1651 extern int net_msg_warn;
1652 #define NETDEBUG(fmt, args...) \
1653 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
1654 
1655 #define LIMIT_NETDEBUG(fmt, args...) \
1656 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
1657 
1658 extern __u32 sysctl_wmem_max;
1659 extern __u32 sysctl_rmem_max;
1660 
1661 extern void sk_init(void);
1662 
1663 extern int sysctl_optmem_max;
1664 
1665 extern __u32 sysctl_wmem_default;
1666 extern __u32 sysctl_rmem_default;
1667 
1668 #endif	/* _SOCK_H */
1669