xref: /linux/include/net/sock.h (revision 2277ab4a1df50e05bc732fe9488d4e902bb8399a)
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 
55 #include <linux/filter.h>
56 #include <linux/rculist_nulls.h>
57 #include <linux/poll.h>
58 
59 #include <asm/atomic.h>
60 #include <net/dst.h>
61 #include <net/checksum.h>
62 
63 /*
64  * This structure really needs to be cleaned up.
65  * Most of it is for TCP, and not used by any of
66  * the other protocols.
67  */
68 
69 /* Define this to get the SOCK_DBG debugging facility. */
70 #define SOCK_DEBUGGING
71 #ifdef SOCK_DEBUGGING
72 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
73 					printk(KERN_DEBUG msg); } while (0)
74 #else
75 /* Validate arguments and do nothing */
76 static void inline int __attribute__ ((format (printf, 2, 3)))
77 SOCK_DEBUG(struct sock *sk, const char *msg, ...)
78 {
79 }
80 #endif
81 
82 /* This is the per-socket lock.  The spinlock provides a synchronization
83  * between user contexts and software interrupt processing, whereas the
84  * mini-semaphore synchronizes multiple users amongst themselves.
85  */
86 typedef struct {
87 	spinlock_t		slock;
88 	int			owned;
89 	wait_queue_head_t	wq;
90 	/*
91 	 * We express the mutex-alike socket_lock semantics
92 	 * to the lock validator by explicitly managing
93 	 * the slock as a lock variant (in addition to
94 	 * the slock itself):
95 	 */
96 #ifdef CONFIG_DEBUG_LOCK_ALLOC
97 	struct lockdep_map dep_map;
98 #endif
99 } socket_lock_t;
100 
101 struct sock;
102 struct proto;
103 struct net;
104 
105 /**
106  *	struct sock_common - minimal network layer representation of sockets
107  *	@skc_family: network address family
108  *	@skc_state: Connection state
109  *	@skc_reuse: %SO_REUSEADDR setting
110  *	@skc_bound_dev_if: bound device index if != 0
111  *	@skc_node: main hash linkage for various protocol lookup tables
112  *	@skc_nulls_node: main hash linkage for UDP/UDP-Lite protocol
113  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
114  *	@skc_refcnt: reference count
115  *	@skc_hash: hash value used with various protocol lookup tables
116  *	@skc_prot: protocol handlers inside a network family
117  *	@skc_net: reference to the network namespace of this socket
118  *
119  *	This is the minimal network layer representation of sockets, the header
120  *	for struct sock and struct inet_timewait_sock.
121  */
122 struct sock_common {
123 	unsigned short		skc_family;
124 	volatile unsigned char	skc_state;
125 	unsigned char		skc_reuse;
126 	int			skc_bound_dev_if;
127 	union {
128 		struct hlist_node	skc_node;
129 		struct hlist_nulls_node skc_nulls_node;
130 	};
131 	struct hlist_node	skc_bind_node;
132 	atomic_t		skc_refcnt;
133 	unsigned int		skc_hash;
134 	struct proto		*skc_prot;
135 #ifdef CONFIG_NET_NS
136 	struct net	 	*skc_net;
137 #endif
138 };
139 
140 /**
141   *	struct sock - network layer representation of sockets
142   *	@__sk_common: shared layout with inet_timewait_sock
143   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
144   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
145   *	@sk_lock:	synchronizer
146   *	@sk_rcvbuf: size of receive buffer in bytes
147   *	@sk_sleep: sock wait queue
148   *	@sk_dst_cache: destination cache
149   *	@sk_dst_lock: destination cache lock
150   *	@sk_policy: flow policy
151   *	@sk_rmem_alloc: receive queue bytes committed
152   *	@sk_receive_queue: incoming packets
153   *	@sk_wmem_alloc: transmit queue bytes committed
154   *	@sk_write_queue: Packet sending queue
155   *	@sk_async_wait_queue: DMA copied packets
156   *	@sk_omem_alloc: "o" is "option" or "other"
157   *	@sk_wmem_queued: persistent queue size
158   *	@sk_forward_alloc: space allocated forward
159   *	@sk_allocation: allocation mode
160   *	@sk_sndbuf: size of send buffer in bytes
161   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
162   *		   %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
163   *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
164   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
165   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
166   *	@sk_gso_max_size: Maximum GSO segment size to build
167   *	@sk_lingertime: %SO_LINGER l_linger setting
168   *	@sk_backlog: always used with the per-socket spinlock held
169   *	@sk_callback_lock: used with the callbacks in the end of this struct
170   *	@sk_error_queue: rarely used
171   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
172   *			  IPV6_ADDRFORM for instance)
173   *	@sk_err: last error
174   *	@sk_err_soft: errors that don't cause failure but are the cause of a
175   *		      persistent failure not just 'timed out'
176   *	@sk_drops: raw/udp drops counter
177   *	@sk_ack_backlog: current listen backlog
178   *	@sk_max_ack_backlog: listen backlog set in listen()
179   *	@sk_priority: %SO_PRIORITY setting
180   *	@sk_type: socket type (%SOCK_STREAM, etc)
181   *	@sk_protocol: which protocol this socket belongs in this network family
182   *	@sk_peercred: %SO_PEERCRED setting
183   *	@sk_rcvlowat: %SO_RCVLOWAT setting
184   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
185   *	@sk_sndtimeo: %SO_SNDTIMEO setting
186   *	@sk_filter: socket filtering instructions
187   *	@sk_protinfo: private area, net family specific, when not using slab
188   *	@sk_timer: sock cleanup timer
189   *	@sk_stamp: time stamp of last packet received
190   *	@sk_socket: Identd and reporting IO signals
191   *	@sk_user_data: RPC layer private data
192   *	@sk_sndmsg_page: cached page for sendmsg
193   *	@sk_sndmsg_off: cached offset for sendmsg
194   *	@sk_send_head: front of stuff to transmit
195   *	@sk_security: used by security modules
196   *	@sk_mark: generic packet mark
197   *	@sk_write_pending: a write to stream socket waits to start
198   *	@sk_state_change: callback to indicate change in the state of the sock
199   *	@sk_data_ready: callback to indicate there is data to be processed
200   *	@sk_write_space: callback to indicate there is bf sending space available
201   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
202   *	@sk_backlog_rcv: callback to process the backlog
203   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
204  */
205 struct sock {
206 	/*
207 	 * Now struct inet_timewait_sock also uses sock_common, so please just
208 	 * don't add nothing before this first member (__sk_common) --acme
209 	 */
210 	struct sock_common	__sk_common;
211 #define sk_family		__sk_common.skc_family
212 #define sk_state		__sk_common.skc_state
213 #define sk_reuse		__sk_common.skc_reuse
214 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
215 #define sk_node			__sk_common.skc_node
216 #define sk_nulls_node		__sk_common.skc_nulls_node
217 #define sk_bind_node		__sk_common.skc_bind_node
218 #define sk_refcnt		__sk_common.skc_refcnt
219 #define sk_hash			__sk_common.skc_hash
220 #define sk_prot			__sk_common.skc_prot
221 #define sk_net			__sk_common.skc_net
222 	kmemcheck_bitfield_begin(flags);
223 	unsigned char		sk_shutdown : 2,
224 				sk_no_check : 2,
225 				sk_userlocks : 4;
226 	kmemcheck_bitfield_end(flags);
227 	unsigned char		sk_protocol;
228 	unsigned short		sk_type;
229 	int			sk_rcvbuf;
230 	socket_lock_t		sk_lock;
231 	/*
232 	 * The backlog queue is special, it is always used with
233 	 * the per-socket spinlock held and requires low latency
234 	 * access. Therefore we special case it's implementation.
235 	 */
236 	struct {
237 		struct sk_buff *head;
238 		struct sk_buff *tail;
239 	} sk_backlog;
240 	wait_queue_head_t	*sk_sleep;
241 	struct dst_entry	*sk_dst_cache;
242 #ifdef CONFIG_XFRM
243 	struct xfrm_policy	*sk_policy[2];
244 #endif
245 	rwlock_t		sk_dst_lock;
246 	atomic_t		sk_rmem_alloc;
247 	atomic_t		sk_wmem_alloc;
248 	atomic_t		sk_omem_alloc;
249 	int			sk_sndbuf;
250 	struct sk_buff_head	sk_receive_queue;
251 	struct sk_buff_head	sk_write_queue;
252 #ifdef CONFIG_NET_DMA
253 	struct sk_buff_head	sk_async_wait_queue;
254 #endif
255 	int			sk_wmem_queued;
256 	int			sk_forward_alloc;
257 	gfp_t			sk_allocation;
258 	int			sk_route_caps;
259 	int			sk_gso_type;
260 	unsigned int		sk_gso_max_size;
261 	int			sk_rcvlowat;
262 	unsigned long 		sk_flags;
263 	unsigned long	        sk_lingertime;
264 	struct sk_buff_head	sk_error_queue;
265 	struct proto		*sk_prot_creator;
266 	rwlock_t		sk_callback_lock;
267 	int			sk_err,
268 				sk_err_soft;
269 	atomic_t		sk_drops;
270 	unsigned short		sk_ack_backlog;
271 	unsigned short		sk_max_ack_backlog;
272 	__u32			sk_priority;
273 	struct ucred		sk_peercred;
274 	long			sk_rcvtimeo;
275 	long			sk_sndtimeo;
276 	struct sk_filter      	*sk_filter;
277 	void			*sk_protinfo;
278 	struct timer_list	sk_timer;
279 	ktime_t			sk_stamp;
280 	struct socket		*sk_socket;
281 	void			*sk_user_data;
282 	struct page		*sk_sndmsg_page;
283 	struct sk_buff		*sk_send_head;
284 	__u32			sk_sndmsg_off;
285 	int			sk_write_pending;
286 #ifdef CONFIG_SECURITY
287 	void			*sk_security;
288 #endif
289 	__u32			sk_mark;
290 	/* XXX 4 bytes hole on 64 bit */
291 	void			(*sk_state_change)(struct sock *sk);
292 	void			(*sk_data_ready)(struct sock *sk, int bytes);
293 	void			(*sk_write_space)(struct sock *sk);
294 	void			(*sk_error_report)(struct sock *sk);
295   	int			(*sk_backlog_rcv)(struct sock *sk,
296 						  struct sk_buff *skb);
297 	void                    (*sk_destruct)(struct sock *sk);
298 };
299 
300 /*
301  * Hashed lists helper routines
302  */
303 static inline struct sock *__sk_head(const struct hlist_head *head)
304 {
305 	return hlist_entry(head->first, struct sock, sk_node);
306 }
307 
308 static inline struct sock *sk_head(const struct hlist_head *head)
309 {
310 	return hlist_empty(head) ? NULL : __sk_head(head);
311 }
312 
313 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
314 {
315 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
316 }
317 
318 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
319 {
320 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
321 }
322 
323 static inline struct sock *sk_next(const struct sock *sk)
324 {
325 	return sk->sk_node.next ?
326 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
327 }
328 
329 static inline struct sock *sk_nulls_next(const struct sock *sk)
330 {
331 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
332 		hlist_nulls_entry(sk->sk_nulls_node.next,
333 				  struct sock, sk_nulls_node) :
334 		NULL;
335 }
336 
337 static inline int sk_unhashed(const struct sock *sk)
338 {
339 	return hlist_unhashed(&sk->sk_node);
340 }
341 
342 static inline int sk_hashed(const struct sock *sk)
343 {
344 	return !sk_unhashed(sk);
345 }
346 
347 static __inline__ void sk_node_init(struct hlist_node *node)
348 {
349 	node->pprev = NULL;
350 }
351 
352 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
353 {
354 	node->pprev = NULL;
355 }
356 
357 static __inline__ void __sk_del_node(struct sock *sk)
358 {
359 	__hlist_del(&sk->sk_node);
360 }
361 
362 static __inline__ int __sk_del_node_init(struct sock *sk)
363 {
364 	if (sk_hashed(sk)) {
365 		__sk_del_node(sk);
366 		sk_node_init(&sk->sk_node);
367 		return 1;
368 	}
369 	return 0;
370 }
371 
372 /* Grab socket reference count. This operation is valid only
373    when sk is ALREADY grabbed f.e. it is found in hash table
374    or a list and the lookup is made under lock preventing hash table
375    modifications.
376  */
377 
378 static inline void sock_hold(struct sock *sk)
379 {
380 	atomic_inc(&sk->sk_refcnt);
381 }
382 
383 /* Ungrab socket in the context, which assumes that socket refcnt
384    cannot hit zero, f.e. it is true in context of any socketcall.
385  */
386 static inline void __sock_put(struct sock *sk)
387 {
388 	atomic_dec(&sk->sk_refcnt);
389 }
390 
391 static __inline__ int sk_del_node_init(struct sock *sk)
392 {
393 	int rc = __sk_del_node_init(sk);
394 
395 	if (rc) {
396 		/* paranoid for a while -acme */
397 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
398 		__sock_put(sk);
399 	}
400 	return rc;
401 }
402 
403 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
404 {
405 	if (sk_hashed(sk)) {
406 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
407 		return 1;
408 	}
409 	return 0;
410 }
411 
412 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
413 {
414 	int rc = __sk_nulls_del_node_init_rcu(sk);
415 
416 	if (rc) {
417 		/* paranoid for a while -acme */
418 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
419 		__sock_put(sk);
420 	}
421 	return rc;
422 }
423 
424 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
425 {
426 	hlist_add_head(&sk->sk_node, list);
427 }
428 
429 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
430 {
431 	sock_hold(sk);
432 	__sk_add_node(sk, list);
433 }
434 
435 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
436 {
437 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
438 }
439 
440 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
441 {
442 	sock_hold(sk);
443 	__sk_nulls_add_node_rcu(sk, list);
444 }
445 
446 static __inline__ void __sk_del_bind_node(struct sock *sk)
447 {
448 	__hlist_del(&sk->sk_bind_node);
449 }
450 
451 static __inline__ void sk_add_bind_node(struct sock *sk,
452 					struct hlist_head *list)
453 {
454 	hlist_add_head(&sk->sk_bind_node, list);
455 }
456 
457 #define sk_for_each(__sk, node, list) \
458 	hlist_for_each_entry(__sk, node, list, sk_node)
459 #define sk_nulls_for_each(__sk, node, list) \
460 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
461 #define sk_nulls_for_each_rcu(__sk, node, list) \
462 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
463 #define sk_for_each_from(__sk, node) \
464 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
465 		hlist_for_each_entry_from(__sk, node, sk_node)
466 #define sk_nulls_for_each_from(__sk, node) \
467 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
468 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
469 #define sk_for_each_continue(__sk, node) \
470 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
471 		hlist_for_each_entry_continue(__sk, node, sk_node)
472 #define sk_for_each_safe(__sk, node, tmp, list) \
473 	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
474 #define sk_for_each_bound(__sk, node, list) \
475 	hlist_for_each_entry(__sk, node, list, sk_bind_node)
476 
477 /* Sock flags */
478 enum sock_flags {
479 	SOCK_DEAD,
480 	SOCK_DONE,
481 	SOCK_URGINLINE,
482 	SOCK_KEEPOPEN,
483 	SOCK_LINGER,
484 	SOCK_DESTROY,
485 	SOCK_BROADCAST,
486 	SOCK_TIMESTAMP,
487 	SOCK_ZAPPED,
488 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
489 	SOCK_DBG, /* %SO_DEBUG setting */
490 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
491 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
492 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
493 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
494 	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
495 	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
496 	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
497 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
498 	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
499 	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
500 	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
501 };
502 
503 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
504 {
505 	nsk->sk_flags = osk->sk_flags;
506 }
507 
508 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
509 {
510 	__set_bit(flag, &sk->sk_flags);
511 }
512 
513 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
514 {
515 	__clear_bit(flag, &sk->sk_flags);
516 }
517 
518 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
519 {
520 	return test_bit(flag, &sk->sk_flags);
521 }
522 
523 static inline void sk_acceptq_removed(struct sock *sk)
524 {
525 	sk->sk_ack_backlog--;
526 }
527 
528 static inline void sk_acceptq_added(struct sock *sk)
529 {
530 	sk->sk_ack_backlog++;
531 }
532 
533 static inline int sk_acceptq_is_full(struct sock *sk)
534 {
535 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
536 }
537 
538 /*
539  * Compute minimal free write space needed to queue new packets.
540  */
541 static inline int sk_stream_min_wspace(struct sock *sk)
542 {
543 	return sk->sk_wmem_queued >> 1;
544 }
545 
546 static inline int sk_stream_wspace(struct sock *sk)
547 {
548 	return sk->sk_sndbuf - sk->sk_wmem_queued;
549 }
550 
551 extern void sk_stream_write_space(struct sock *sk);
552 
553 static inline int sk_stream_memory_free(struct sock *sk)
554 {
555 	return sk->sk_wmem_queued < sk->sk_sndbuf;
556 }
557 
558 /* The per-socket spinlock must be held here. */
559 static inline void sk_add_backlog(struct sock *sk, struct sk_buff *skb)
560 {
561 	if (!sk->sk_backlog.tail) {
562 		sk->sk_backlog.head = sk->sk_backlog.tail = skb;
563 	} else {
564 		sk->sk_backlog.tail->next = skb;
565 		sk->sk_backlog.tail = skb;
566 	}
567 	skb->next = NULL;
568 }
569 
570 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
571 {
572 	return sk->sk_backlog_rcv(sk, skb);
573 }
574 
575 #define sk_wait_event(__sk, __timeo, __condition)			\
576 	({	int __rc;						\
577 		release_sock(__sk);					\
578 		__rc = __condition;					\
579 		if (!__rc) {						\
580 			*(__timeo) = schedule_timeout(*(__timeo));	\
581 		}							\
582 		lock_sock(__sk);					\
583 		__rc = __condition;					\
584 		__rc;							\
585 	})
586 
587 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
588 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
589 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
590 extern int sk_stream_error(struct sock *sk, int flags, int err);
591 extern void sk_stream_kill_queues(struct sock *sk);
592 
593 extern int sk_wait_data(struct sock *sk, long *timeo);
594 
595 struct request_sock_ops;
596 struct timewait_sock_ops;
597 struct inet_hashinfo;
598 struct raw_hashinfo;
599 
600 /* Networking protocol blocks we attach to sockets.
601  * socket layer -> transport layer interface
602  * transport -> network interface is defined by struct inet_proto
603  */
604 struct proto {
605 	void			(*close)(struct sock *sk,
606 					long timeout);
607 	int			(*connect)(struct sock *sk,
608 				        struct sockaddr *uaddr,
609 					int addr_len);
610 	int			(*disconnect)(struct sock *sk, int flags);
611 
612 	struct sock *		(*accept) (struct sock *sk, int flags, int *err);
613 
614 	int			(*ioctl)(struct sock *sk, int cmd,
615 					 unsigned long arg);
616 	int			(*init)(struct sock *sk);
617 	void			(*destroy)(struct sock *sk);
618 	void			(*shutdown)(struct sock *sk, int how);
619 	int			(*setsockopt)(struct sock *sk, int level,
620 					int optname, char __user *optval,
621 					int optlen);
622 	int			(*getsockopt)(struct sock *sk, int level,
623 					int optname, char __user *optval,
624 					int __user *option);
625 #ifdef CONFIG_COMPAT
626 	int			(*compat_setsockopt)(struct sock *sk,
627 					int level,
628 					int optname, char __user *optval,
629 					int optlen);
630 	int			(*compat_getsockopt)(struct sock *sk,
631 					int level,
632 					int optname, char __user *optval,
633 					int __user *option);
634 #endif
635 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
636 					   struct msghdr *msg, size_t len);
637 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
638 					   struct msghdr *msg,
639 					size_t len, int noblock, int flags,
640 					int *addr_len);
641 	int			(*sendpage)(struct sock *sk, struct page *page,
642 					int offset, size_t size, int flags);
643 	int			(*bind)(struct sock *sk,
644 					struct sockaddr *uaddr, int addr_len);
645 
646 	int			(*backlog_rcv) (struct sock *sk,
647 						struct sk_buff *skb);
648 
649 	/* Keeping track of sk's, looking them up, and port selection methods. */
650 	void			(*hash)(struct sock *sk);
651 	void			(*unhash)(struct sock *sk);
652 	int			(*get_port)(struct sock *sk, unsigned short snum);
653 
654 	/* Keeping track of sockets in use */
655 #ifdef CONFIG_PROC_FS
656 	unsigned int		inuse_idx;
657 #endif
658 
659 	/* Memory pressure */
660 	void			(*enter_memory_pressure)(struct sock *sk);
661 	atomic_t		*memory_allocated;	/* Current allocated memory. */
662 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
663 	/*
664 	 * Pressure flag: try to collapse.
665 	 * Technical note: it is used by multiple contexts non atomically.
666 	 * All the __sk_mem_schedule() is of this nature: accounting
667 	 * is strict, actions are advisory and have some latency.
668 	 */
669 	int			*memory_pressure;
670 	int			*sysctl_mem;
671 	int			*sysctl_wmem;
672 	int			*sysctl_rmem;
673 	int			max_header;
674 
675 	struct kmem_cache	*slab;
676 	unsigned int		obj_size;
677 	int			slab_flags;
678 
679 	struct percpu_counter	*orphan_count;
680 
681 	struct request_sock_ops	*rsk_prot;
682 	struct timewait_sock_ops *twsk_prot;
683 
684 	union {
685 		struct inet_hashinfo	*hashinfo;
686 		struct udp_table	*udp_table;
687 		struct raw_hashinfo	*raw_hash;
688 	} h;
689 
690 	struct module		*owner;
691 
692 	char			name[32];
693 
694 	struct list_head	node;
695 #ifdef SOCK_REFCNT_DEBUG
696 	atomic_t		socks;
697 #endif
698 };
699 
700 extern int proto_register(struct proto *prot, int alloc_slab);
701 extern void proto_unregister(struct proto *prot);
702 
703 #ifdef SOCK_REFCNT_DEBUG
704 static inline void sk_refcnt_debug_inc(struct sock *sk)
705 {
706 	atomic_inc(&sk->sk_prot->socks);
707 }
708 
709 static inline void sk_refcnt_debug_dec(struct sock *sk)
710 {
711 	atomic_dec(&sk->sk_prot->socks);
712 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
713 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
714 }
715 
716 static inline void sk_refcnt_debug_release(const struct sock *sk)
717 {
718 	if (atomic_read(&sk->sk_refcnt) != 1)
719 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
720 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
721 }
722 #else /* SOCK_REFCNT_DEBUG */
723 #define sk_refcnt_debug_inc(sk) do { } while (0)
724 #define sk_refcnt_debug_dec(sk) do { } while (0)
725 #define sk_refcnt_debug_release(sk) do { } while (0)
726 #endif /* SOCK_REFCNT_DEBUG */
727 
728 
729 #ifdef CONFIG_PROC_FS
730 /* Called with local bh disabled */
731 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
732 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
733 #else
734 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
735 		int inc)
736 {
737 }
738 #endif
739 
740 
741 /* With per-bucket locks this operation is not-atomic, so that
742  * this version is not worse.
743  */
744 static inline void __sk_prot_rehash(struct sock *sk)
745 {
746 	sk->sk_prot->unhash(sk);
747 	sk->sk_prot->hash(sk);
748 }
749 
750 /* About 10 seconds */
751 #define SOCK_DESTROY_TIME (10*HZ)
752 
753 /* Sockets 0-1023 can't be bound to unless you are superuser */
754 #define PROT_SOCK	1024
755 
756 #define SHUTDOWN_MASK	3
757 #define RCV_SHUTDOWN	1
758 #define SEND_SHUTDOWN	2
759 
760 #define SOCK_SNDBUF_LOCK	1
761 #define SOCK_RCVBUF_LOCK	2
762 #define SOCK_BINDADDR_LOCK	4
763 #define SOCK_BINDPORT_LOCK	8
764 
765 /* sock_iocb: used to kick off async processing of socket ios */
766 struct sock_iocb {
767 	struct list_head	list;
768 
769 	int			flags;
770 	int			size;
771 	struct socket		*sock;
772 	struct sock		*sk;
773 	struct scm_cookie	*scm;
774 	struct msghdr		*msg, async_msg;
775 	struct kiocb		*kiocb;
776 };
777 
778 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
779 {
780 	return (struct sock_iocb *)iocb->private;
781 }
782 
783 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
784 {
785 	return si->kiocb;
786 }
787 
788 struct socket_alloc {
789 	struct socket socket;
790 	struct inode vfs_inode;
791 };
792 
793 static inline struct socket *SOCKET_I(struct inode *inode)
794 {
795 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
796 }
797 
798 static inline struct inode *SOCK_INODE(struct socket *socket)
799 {
800 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
801 }
802 
803 /*
804  * Functions for memory accounting
805  */
806 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
807 extern void __sk_mem_reclaim(struct sock *sk);
808 
809 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
810 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
811 #define SK_MEM_SEND	0
812 #define SK_MEM_RECV	1
813 
814 static inline int sk_mem_pages(int amt)
815 {
816 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
817 }
818 
819 static inline int sk_has_account(struct sock *sk)
820 {
821 	/* return true if protocol supports memory accounting */
822 	return !!sk->sk_prot->memory_allocated;
823 }
824 
825 static inline int sk_wmem_schedule(struct sock *sk, int size)
826 {
827 	if (!sk_has_account(sk))
828 		return 1;
829 	return size <= sk->sk_forward_alloc ||
830 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
831 }
832 
833 static inline int sk_rmem_schedule(struct sock *sk, int size)
834 {
835 	if (!sk_has_account(sk))
836 		return 1;
837 	return size <= sk->sk_forward_alloc ||
838 		__sk_mem_schedule(sk, size, SK_MEM_RECV);
839 }
840 
841 static inline void sk_mem_reclaim(struct sock *sk)
842 {
843 	if (!sk_has_account(sk))
844 		return;
845 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
846 		__sk_mem_reclaim(sk);
847 }
848 
849 static inline void sk_mem_reclaim_partial(struct sock *sk)
850 {
851 	if (!sk_has_account(sk))
852 		return;
853 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
854 		__sk_mem_reclaim(sk);
855 }
856 
857 static inline void sk_mem_charge(struct sock *sk, int size)
858 {
859 	if (!sk_has_account(sk))
860 		return;
861 	sk->sk_forward_alloc -= size;
862 }
863 
864 static inline void sk_mem_uncharge(struct sock *sk, int size)
865 {
866 	if (!sk_has_account(sk))
867 		return;
868 	sk->sk_forward_alloc += size;
869 }
870 
871 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
872 {
873 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
874 	sk->sk_wmem_queued -= skb->truesize;
875 	sk_mem_uncharge(sk, skb->truesize);
876 	__kfree_skb(skb);
877 }
878 
879 /* Used by processes to "lock" a socket state, so that
880  * interrupts and bottom half handlers won't change it
881  * from under us. It essentially blocks any incoming
882  * packets, so that we won't get any new data or any
883  * packets that change the state of the socket.
884  *
885  * While locked, BH processing will add new packets to
886  * the backlog queue.  This queue is processed by the
887  * owner of the socket lock right before it is released.
888  *
889  * Since ~2.3.5 it is also exclusive sleep lock serializing
890  * accesses from user process context.
891  */
892 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
893 
894 /*
895  * Macro so as to not evaluate some arguments when
896  * lockdep is not enabled.
897  *
898  * Mark both the sk_lock and the sk_lock.slock as a
899  * per-address-family lock class.
900  */
901 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) 	\
902 do {									\
903 	sk->sk_lock.owned = 0;						\
904 	init_waitqueue_head(&sk->sk_lock.wq);				\
905 	spin_lock_init(&(sk)->sk_lock.slock);				\
906 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
907 			sizeof((sk)->sk_lock));				\
908 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
909 		       	(skey), (sname));				\
910 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
911 } while (0)
912 
913 extern void lock_sock_nested(struct sock *sk, int subclass);
914 
915 static inline void lock_sock(struct sock *sk)
916 {
917 	lock_sock_nested(sk, 0);
918 }
919 
920 extern void release_sock(struct sock *sk);
921 
922 /* BH context may only use the following locking interface. */
923 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
924 #define bh_lock_sock_nested(__sk) \
925 				spin_lock_nested(&((__sk)->sk_lock.slock), \
926 				SINGLE_DEPTH_NESTING)
927 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
928 
929 extern struct sock		*sk_alloc(struct net *net, int family,
930 					  gfp_t priority,
931 					  struct proto *prot);
932 extern void			sk_free(struct sock *sk);
933 extern void			sk_release_kernel(struct sock *sk);
934 extern struct sock		*sk_clone(const struct sock *sk,
935 					  const gfp_t priority);
936 
937 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
938 					      unsigned long size, int force,
939 					      gfp_t priority);
940 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
941 					      unsigned long size, int force,
942 					      gfp_t priority);
943 extern void			sock_wfree(struct sk_buff *skb);
944 extern void			sock_rfree(struct sk_buff *skb);
945 
946 extern int			sock_setsockopt(struct socket *sock, int level,
947 						int op, char __user *optval,
948 						int optlen);
949 
950 extern int			sock_getsockopt(struct socket *sock, int level,
951 						int op, char __user *optval,
952 						int __user *optlen);
953 extern struct sk_buff 		*sock_alloc_send_skb(struct sock *sk,
954 						     unsigned long size,
955 						     int noblock,
956 						     int *errcode);
957 extern struct sk_buff 		*sock_alloc_send_pskb(struct sock *sk,
958 						      unsigned long header_len,
959 						      unsigned long data_len,
960 						      int noblock,
961 						      int *errcode);
962 extern void *sock_kmalloc(struct sock *sk, int size,
963 			  gfp_t priority);
964 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
965 extern void sk_send_sigurg(struct sock *sk);
966 
967 /*
968  * Functions to fill in entries in struct proto_ops when a protocol
969  * does not implement a particular function.
970  */
971 extern int                      sock_no_bind(struct socket *,
972 					     struct sockaddr *, int);
973 extern int                      sock_no_connect(struct socket *,
974 						struct sockaddr *, int, int);
975 extern int                      sock_no_socketpair(struct socket *,
976 						   struct socket *);
977 extern int                      sock_no_accept(struct socket *,
978 					       struct socket *, int);
979 extern int                      sock_no_getname(struct socket *,
980 						struct sockaddr *, int *, int);
981 extern unsigned int             sock_no_poll(struct file *, struct socket *,
982 					     struct poll_table_struct *);
983 extern int                      sock_no_ioctl(struct socket *, unsigned int,
984 					      unsigned long);
985 extern int			sock_no_listen(struct socket *, int);
986 extern int                      sock_no_shutdown(struct socket *, int);
987 extern int			sock_no_getsockopt(struct socket *, int , int,
988 						   char __user *, int __user *);
989 extern int			sock_no_setsockopt(struct socket *, int, int,
990 						   char __user *, int);
991 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
992 						struct msghdr *, size_t);
993 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
994 						struct msghdr *, size_t, int);
995 extern int			sock_no_mmap(struct file *file,
996 					     struct socket *sock,
997 					     struct vm_area_struct *vma);
998 extern ssize_t			sock_no_sendpage(struct socket *sock,
999 						struct page *page,
1000 						int offset, size_t size,
1001 						int flags);
1002 
1003 /*
1004  * Functions to fill in entries in struct proto_ops when a protocol
1005  * uses the inet style.
1006  */
1007 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1008 				  char __user *optval, int __user *optlen);
1009 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1010 			       struct msghdr *msg, size_t size, int flags);
1011 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1012 				  char __user *optval, int optlen);
1013 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1014 		int optname, char __user *optval, int __user *optlen);
1015 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1016 		int optname, char __user *optval, int optlen);
1017 
1018 extern void sk_common_release(struct sock *sk);
1019 
1020 /*
1021  *	Default socket callbacks and setup code
1022  */
1023 
1024 /* Initialise core socket variables */
1025 extern void sock_init_data(struct socket *sock, struct sock *sk);
1026 
1027 /**
1028  *	sk_filter_release: Release a socket filter
1029  *	@fp: filter to remove
1030  *
1031  *	Remove a filter from a socket and release its resources.
1032  */
1033 
1034 static inline void sk_filter_release(struct sk_filter *fp)
1035 {
1036 	if (atomic_dec_and_test(&fp->refcnt))
1037 		kfree(fp);
1038 }
1039 
1040 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1041 {
1042 	unsigned int size = sk_filter_len(fp);
1043 
1044 	atomic_sub(size, &sk->sk_omem_alloc);
1045 	sk_filter_release(fp);
1046 }
1047 
1048 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1049 {
1050 	atomic_inc(&fp->refcnt);
1051 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1052 }
1053 
1054 /*
1055  * Socket reference counting postulates.
1056  *
1057  * * Each user of socket SHOULD hold a reference count.
1058  * * Each access point to socket (an hash table bucket, reference from a list,
1059  *   running timer, skb in flight MUST hold a reference count.
1060  * * When reference count hits 0, it means it will never increase back.
1061  * * When reference count hits 0, it means that no references from
1062  *   outside exist to this socket and current process on current CPU
1063  *   is last user and may/should destroy this socket.
1064  * * sk_free is called from any context: process, BH, IRQ. When
1065  *   it is called, socket has no references from outside -> sk_free
1066  *   may release descendant resources allocated by the socket, but
1067  *   to the time when it is called, socket is NOT referenced by any
1068  *   hash tables, lists etc.
1069  * * Packets, delivered from outside (from network or from another process)
1070  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1071  *   when they sit in queue. Otherwise, packets will leak to hole, when
1072  *   socket is looked up by one cpu and unhasing is made by another CPU.
1073  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1074  *   (leak to backlog). Packet socket does all the processing inside
1075  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1076  *   use separate SMP lock, so that they are prone too.
1077  */
1078 
1079 /* Ungrab socket and destroy it, if it was the last reference. */
1080 static inline void sock_put(struct sock *sk)
1081 {
1082 	if (atomic_dec_and_test(&sk->sk_refcnt))
1083 		sk_free(sk);
1084 }
1085 
1086 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1087 			  const int nested);
1088 
1089 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1090 {
1091 	sk->sk_socket = sock;
1092 }
1093 
1094 /* Detach socket from process context.
1095  * Announce socket dead, detach it from wait queue and inode.
1096  * Note that parent inode held reference count on this struct sock,
1097  * we do not release it in this function, because protocol
1098  * probably wants some additional cleanups or even continuing
1099  * to work with this socket (TCP).
1100  */
1101 static inline void sock_orphan(struct sock *sk)
1102 {
1103 	write_lock_bh(&sk->sk_callback_lock);
1104 	sock_set_flag(sk, SOCK_DEAD);
1105 	sk_set_socket(sk, NULL);
1106 	sk->sk_sleep  = NULL;
1107 	write_unlock_bh(&sk->sk_callback_lock);
1108 }
1109 
1110 static inline void sock_graft(struct sock *sk, struct socket *parent)
1111 {
1112 	write_lock_bh(&sk->sk_callback_lock);
1113 	sk->sk_sleep = &parent->wait;
1114 	parent->sk = sk;
1115 	sk_set_socket(sk, parent);
1116 	security_sock_graft(sk, parent);
1117 	write_unlock_bh(&sk->sk_callback_lock);
1118 }
1119 
1120 extern int sock_i_uid(struct sock *sk);
1121 extern unsigned long sock_i_ino(struct sock *sk);
1122 
1123 static inline struct dst_entry *
1124 __sk_dst_get(struct sock *sk)
1125 {
1126 	return sk->sk_dst_cache;
1127 }
1128 
1129 static inline struct dst_entry *
1130 sk_dst_get(struct sock *sk)
1131 {
1132 	struct dst_entry *dst;
1133 
1134 	read_lock(&sk->sk_dst_lock);
1135 	dst = sk->sk_dst_cache;
1136 	if (dst)
1137 		dst_hold(dst);
1138 	read_unlock(&sk->sk_dst_lock);
1139 	return dst;
1140 }
1141 
1142 static inline void
1143 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1144 {
1145 	struct dst_entry *old_dst;
1146 
1147 	old_dst = sk->sk_dst_cache;
1148 	sk->sk_dst_cache = dst;
1149 	dst_release(old_dst);
1150 }
1151 
1152 static inline void
1153 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1154 {
1155 	write_lock(&sk->sk_dst_lock);
1156 	__sk_dst_set(sk, dst);
1157 	write_unlock(&sk->sk_dst_lock);
1158 }
1159 
1160 static inline void
1161 __sk_dst_reset(struct sock *sk)
1162 {
1163 	struct dst_entry *old_dst;
1164 
1165 	old_dst = sk->sk_dst_cache;
1166 	sk->sk_dst_cache = NULL;
1167 	dst_release(old_dst);
1168 }
1169 
1170 static inline void
1171 sk_dst_reset(struct sock *sk)
1172 {
1173 	write_lock(&sk->sk_dst_lock);
1174 	__sk_dst_reset(sk);
1175 	write_unlock(&sk->sk_dst_lock);
1176 }
1177 
1178 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1179 
1180 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1181 
1182 static inline int sk_can_gso(const struct sock *sk)
1183 {
1184 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1185 }
1186 
1187 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1188 
1189 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1190 				   struct sk_buff *skb, struct page *page,
1191 				   int off, int copy)
1192 {
1193 	if (skb->ip_summed == CHECKSUM_NONE) {
1194 		int err = 0;
1195 		__wsum csum = csum_and_copy_from_user(from,
1196 						     page_address(page) + off,
1197 							    copy, 0, &err);
1198 		if (err)
1199 			return err;
1200 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1201 	} else if (copy_from_user(page_address(page) + off, from, copy))
1202 		return -EFAULT;
1203 
1204 	skb->len	     += copy;
1205 	skb->data_len	     += copy;
1206 	skb->truesize	     += copy;
1207 	sk->sk_wmem_queued   += copy;
1208 	sk_mem_charge(sk, copy);
1209 	return 0;
1210 }
1211 
1212 /**
1213  * sk_wmem_alloc_get - returns write allocations
1214  * @sk: socket
1215  *
1216  * Returns sk_wmem_alloc minus initial offset of one
1217  */
1218 static inline int sk_wmem_alloc_get(const struct sock *sk)
1219 {
1220 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1221 }
1222 
1223 /**
1224  * sk_rmem_alloc_get - returns read allocations
1225  * @sk: socket
1226  *
1227  * Returns sk_rmem_alloc
1228  */
1229 static inline int sk_rmem_alloc_get(const struct sock *sk)
1230 {
1231 	return atomic_read(&sk->sk_rmem_alloc);
1232 }
1233 
1234 /**
1235  * sk_has_allocations - check if allocations are outstanding
1236  * @sk: socket
1237  *
1238  * Returns true if socket has write or read allocations
1239  */
1240 static inline int sk_has_allocations(const struct sock *sk)
1241 {
1242 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1243 }
1244 
1245 /**
1246  * sk_has_sleeper - check if there are any waiting processes
1247  * @sk: socket
1248  *
1249  * Returns true if socket has waiting processes
1250  *
1251  * The purpose of the sk_has_sleeper and sock_poll_wait is to wrap the memory
1252  * barrier call. They were added due to the race found within the tcp code.
1253  *
1254  * Consider following tcp code paths:
1255  *
1256  * CPU1                  CPU2
1257  *
1258  * sys_select            receive packet
1259  *   ...                 ...
1260  *   __add_wait_queue    update tp->rcv_nxt
1261  *   ...                 ...
1262  *   tp->rcv_nxt check   sock_def_readable
1263  *   ...                 {
1264  *   schedule               ...
1265  *                          if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1266  *                              wake_up_interruptible(sk->sk_sleep)
1267  *                          ...
1268  *                       }
1269  *
1270  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1271  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1272  * could then endup calling schedule and sleep forever if there are no more
1273  * data on the socket.
1274  *
1275  * The sk_has_sleeper is always called right after a call to read_lock, so we
1276  * can use smp_mb__after_lock barrier.
1277  */
1278 static inline int sk_has_sleeper(struct sock *sk)
1279 {
1280 	/*
1281 	 * We need to be sure we are in sync with the
1282 	 * add_wait_queue modifications to the wait queue.
1283 	 *
1284 	 * This memory barrier is paired in the sock_poll_wait.
1285 	 */
1286 	smp_mb__after_lock();
1287 	return sk->sk_sleep && waitqueue_active(sk->sk_sleep);
1288 }
1289 
1290 /**
1291  * sock_poll_wait - place memory barrier behind the poll_wait call.
1292  * @filp:           file
1293  * @wait_address:   socket wait queue
1294  * @p:              poll_table
1295  *
1296  * See the comments in the sk_has_sleeper function.
1297  */
1298 static inline void sock_poll_wait(struct file *filp,
1299 		wait_queue_head_t *wait_address, poll_table *p)
1300 {
1301 	if (p && wait_address) {
1302 		poll_wait(filp, wait_address, p);
1303 		/*
1304 		 * We need to be sure we are in sync with the
1305 		 * socket flags modification.
1306 		 *
1307 		 * This memory barrier is paired in the sk_has_sleeper.
1308 		*/
1309 		smp_mb();
1310 	}
1311 }
1312 
1313 /*
1314  * 	Queue a received datagram if it will fit. Stream and sequenced
1315  *	protocols can't normally use this as they need to fit buffers in
1316  *	and play with them.
1317  *
1318  * 	Inlined as it's very short and called for pretty much every
1319  *	packet ever received.
1320  */
1321 
1322 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1323 {
1324 	skb_orphan(skb);
1325 	skb->sk = sk;
1326 	skb->destructor = sock_wfree;
1327 	/*
1328 	 * We used to take a refcount on sk, but following operation
1329 	 * is enough to guarantee sk_free() wont free this sock until
1330 	 * all in-flight packets are completed
1331 	 */
1332 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1333 }
1334 
1335 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1336 {
1337 	skb_orphan(skb);
1338 	skb->sk = sk;
1339 	skb->destructor = sock_rfree;
1340 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1341 	sk_mem_charge(sk, skb->truesize);
1342 }
1343 
1344 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1345 			   unsigned long expires);
1346 
1347 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1348 
1349 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1350 
1351 static inline int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
1352 {
1353 	/* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
1354 	   number of warnings when compiling with -W --ANK
1355 	 */
1356 	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
1357 	    (unsigned)sk->sk_rcvbuf)
1358 		return -ENOMEM;
1359 	skb_set_owner_r(skb, sk);
1360 	skb_queue_tail(&sk->sk_error_queue, skb);
1361 	if (!sock_flag(sk, SOCK_DEAD))
1362 		sk->sk_data_ready(sk, skb->len);
1363 	return 0;
1364 }
1365 
1366 /*
1367  *	Recover an error report and clear atomically
1368  */
1369 
1370 static inline int sock_error(struct sock *sk)
1371 {
1372 	int err;
1373 	if (likely(!sk->sk_err))
1374 		return 0;
1375 	err = xchg(&sk->sk_err, 0);
1376 	return -err;
1377 }
1378 
1379 static inline unsigned long sock_wspace(struct sock *sk)
1380 {
1381 	int amt = 0;
1382 
1383 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1384 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1385 		if (amt < 0)
1386 			amt = 0;
1387 	}
1388 	return amt;
1389 }
1390 
1391 static inline void sk_wake_async(struct sock *sk, int how, int band)
1392 {
1393 	if (sk->sk_socket && sk->sk_socket->fasync_list)
1394 		sock_wake_async(sk->sk_socket, how, band);
1395 }
1396 
1397 #define SOCK_MIN_SNDBUF 2048
1398 #define SOCK_MIN_RCVBUF 256
1399 
1400 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1401 {
1402 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1403 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1404 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1405 	}
1406 }
1407 
1408 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1409 
1410 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1411 {
1412 	struct page *page = NULL;
1413 
1414 	page = alloc_pages(sk->sk_allocation, 0);
1415 	if (!page) {
1416 		sk->sk_prot->enter_memory_pressure(sk);
1417 		sk_stream_moderate_sndbuf(sk);
1418 	}
1419 	return page;
1420 }
1421 
1422 /*
1423  *	Default write policy as shown to user space via poll/select/SIGIO
1424  */
1425 static inline int sock_writeable(const struct sock *sk)
1426 {
1427 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1428 }
1429 
1430 static inline gfp_t gfp_any(void)
1431 {
1432 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1433 }
1434 
1435 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1436 {
1437 	return noblock ? 0 : sk->sk_rcvtimeo;
1438 }
1439 
1440 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1441 {
1442 	return noblock ? 0 : sk->sk_sndtimeo;
1443 }
1444 
1445 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1446 {
1447 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1448 }
1449 
1450 /* Alas, with timeout socket operations are not restartable.
1451  * Compare this to poll().
1452  */
1453 static inline int sock_intr_errno(long timeo)
1454 {
1455 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1456 }
1457 
1458 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1459 	struct sk_buff *skb);
1460 
1461 static __inline__ void
1462 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1463 {
1464 	ktime_t kt = skb->tstamp;
1465 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1466 
1467 	/*
1468 	 * generate control messages if
1469 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
1470 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
1471 	 * - software time stamp available and wanted
1472 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
1473 	 * - hardware time stamps available and wanted
1474 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
1475 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
1476 	 */
1477 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
1478 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
1479 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
1480 	    (hwtstamps->hwtstamp.tv64 &&
1481 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
1482 	    (hwtstamps->syststamp.tv64 &&
1483 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
1484 		__sock_recv_timestamp(msg, sk, skb);
1485 	else
1486 		sk->sk_stamp = kt;
1487 }
1488 
1489 /**
1490  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
1491  * @msg:	outgoing packet
1492  * @sk:		socket sending this packet
1493  * @shtx:	filled with instructions for time stamping
1494  *
1495  * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
1496  * parameters are invalid.
1497  */
1498 extern int sock_tx_timestamp(struct msghdr *msg,
1499 			     struct sock *sk,
1500 			     union skb_shared_tx *shtx);
1501 
1502 
1503 /**
1504  * sk_eat_skb - Release a skb if it is no longer needed
1505  * @sk: socket to eat this skb from
1506  * @skb: socket buffer to eat
1507  * @copied_early: flag indicating whether DMA operations copied this data early
1508  *
1509  * This routine must be called with interrupts disabled or with the socket
1510  * locked so that the sk_buff queue operation is ok.
1511 */
1512 #ifdef CONFIG_NET_DMA
1513 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1514 {
1515 	__skb_unlink(skb, &sk->sk_receive_queue);
1516 	if (!copied_early)
1517 		__kfree_skb(skb);
1518 	else
1519 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
1520 }
1521 #else
1522 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1523 {
1524 	__skb_unlink(skb, &sk->sk_receive_queue);
1525 	__kfree_skb(skb);
1526 }
1527 #endif
1528 
1529 static inline
1530 struct net *sock_net(const struct sock *sk)
1531 {
1532 #ifdef CONFIG_NET_NS
1533 	return sk->sk_net;
1534 #else
1535 	return &init_net;
1536 #endif
1537 }
1538 
1539 static inline
1540 void sock_net_set(struct sock *sk, struct net *net)
1541 {
1542 #ifdef CONFIG_NET_NS
1543 	sk->sk_net = net;
1544 #endif
1545 }
1546 
1547 /*
1548  * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
1549  * They should not hold a referrence to a namespace in order to allow
1550  * to stop it.
1551  * Sockets after sk_change_net should be released using sk_release_kernel
1552  */
1553 static inline void sk_change_net(struct sock *sk, struct net *net)
1554 {
1555 	put_net(sock_net(sk));
1556 	sock_net_set(sk, hold_net(net));
1557 }
1558 
1559 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
1560 {
1561 	if (unlikely(skb->sk)) {
1562 		struct sock *sk = skb->sk;
1563 
1564 		skb->destructor = NULL;
1565 		skb->sk = NULL;
1566 		return sk;
1567 	}
1568 	return NULL;
1569 }
1570 
1571 extern void sock_enable_timestamp(struct sock *sk, int flag);
1572 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1573 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1574 
1575 /*
1576  *	Enable debug/info messages
1577  */
1578 extern int net_msg_warn;
1579 #define NETDEBUG(fmt, args...) \
1580 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
1581 
1582 #define LIMIT_NETDEBUG(fmt, args...) \
1583 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
1584 
1585 extern __u32 sysctl_wmem_max;
1586 extern __u32 sysctl_rmem_max;
1587 
1588 extern void sk_init(void);
1589 
1590 extern int sysctl_optmem_max;
1591 
1592 extern __u32 sysctl_wmem_default;
1593 extern __u32 sysctl_rmem_default;
1594 
1595 #endif	/* _SOCK_H */
1596