xref: /linux/include/net/sock.h (revision d89dffa976bcd13fd87eb76e02e3b71c3a7868e3)
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/hardirq.h>
44 #include <linux/kernel.h>
45 #include <linux/list.h>
46 #include <linux/list_nulls.h>
47 #include <linux/timer.h>
48 #include <linux/cache.h>
49 #include <linux/bitops.h>
50 #include <linux/lockdep.h>
51 #include <linux/netdevice.h>
52 #include <linux/skbuff.h>	/* struct sk_buff */
53 #include <linux/mm.h>
54 #include <linux/security.h>
55 #include <linux/slab.h>
56 #include <linux/uaccess.h>
57 #include <linux/memcontrol.h>
58 #include <linux/res_counter.h>
59 #include <linux/static_key.h>
60 #include <linux/aio.h>
61 #include <linux/sched.h>
62 
63 #include <linux/filter.h>
64 #include <linux/rculist_nulls.h>
65 #include <linux/poll.h>
66 
67 #include <linux/atomic.h>
68 #include <net/dst.h>
69 #include <net/checksum.h>
70 
71 struct cgroup;
72 struct cgroup_subsys;
73 #ifdef CONFIG_NET
74 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
75 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
76 #else
77 static inline
78 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
79 {
80 	return 0;
81 }
82 static inline
83 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
84 {
85 }
86 #endif
87 /*
88  * This structure really needs to be cleaned up.
89  * Most of it is for TCP, and not used by any of
90  * the other protocols.
91  */
92 
93 /* Define this to get the SOCK_DBG debugging facility. */
94 #define SOCK_DEBUGGING
95 #ifdef SOCK_DEBUGGING
96 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
97 					printk(KERN_DEBUG msg); } while (0)
98 #else
99 /* Validate arguments and do nothing */
100 static inline __printf(2, 3)
101 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
102 {
103 }
104 #endif
105 
106 /* This is the per-socket lock.  The spinlock provides a synchronization
107  * between user contexts and software interrupt processing, whereas the
108  * mini-semaphore synchronizes multiple users amongst themselves.
109  */
110 typedef struct {
111 	spinlock_t		slock;
112 	int			owned;
113 	wait_queue_head_t	wq;
114 	/*
115 	 * We express the mutex-alike socket_lock semantics
116 	 * to the lock validator by explicitly managing
117 	 * the slock as a lock variant (in addition to
118 	 * the slock itself):
119 	 */
120 #ifdef CONFIG_DEBUG_LOCK_ALLOC
121 	struct lockdep_map dep_map;
122 #endif
123 } socket_lock_t;
124 
125 struct sock;
126 struct proto;
127 struct net;
128 
129 /**
130  *	struct sock_common - minimal network layer representation of sockets
131  *	@skc_daddr: Foreign IPv4 addr
132  *	@skc_rcv_saddr: Bound local IPv4 addr
133  *	@skc_hash: hash value used with various protocol lookup tables
134  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
135  *	@skc_family: network address family
136  *	@skc_state: Connection state
137  *	@skc_reuse: %SO_REUSEADDR setting
138  *	@skc_bound_dev_if: bound device index if != 0
139  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
140  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
141  *	@skc_prot: protocol handlers inside a network family
142  *	@skc_net: reference to the network namespace of this socket
143  *	@skc_node: main hash linkage for various protocol lookup tables
144  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
145  *	@skc_tx_queue_mapping: tx queue number for this connection
146  *	@skc_refcnt: reference count
147  *
148  *	This is the minimal network layer representation of sockets, the header
149  *	for struct sock and struct inet_timewait_sock.
150  */
151 struct sock_common {
152 	/* skc_daddr and skc_rcv_saddr must be grouped :
153 	 * cf INET_MATCH() and INET_TW_MATCH()
154 	 */
155 	__be32			skc_daddr;
156 	__be32			skc_rcv_saddr;
157 
158 	union  {
159 		unsigned int	skc_hash;
160 		__u16		skc_u16hashes[2];
161 	};
162 	unsigned short		skc_family;
163 	volatile unsigned char	skc_state;
164 	unsigned char		skc_reuse;
165 	int			skc_bound_dev_if;
166 	union {
167 		struct hlist_node	skc_bind_node;
168 		struct hlist_nulls_node skc_portaddr_node;
169 	};
170 	struct proto		*skc_prot;
171 #ifdef CONFIG_NET_NS
172 	struct net	 	*skc_net;
173 #endif
174 	/*
175 	 * fields between dontcopy_begin/dontcopy_end
176 	 * are not copied in sock_copy()
177 	 */
178 	/* private: */
179 	int			skc_dontcopy_begin[0];
180 	/* public: */
181 	union {
182 		struct hlist_node	skc_node;
183 		struct hlist_nulls_node skc_nulls_node;
184 	};
185 	int			skc_tx_queue_mapping;
186 	atomic_t		skc_refcnt;
187 	/* private: */
188 	int                     skc_dontcopy_end[0];
189 	/* public: */
190 };
191 
192 struct cg_proto;
193 /**
194   *	struct sock - network layer representation of sockets
195   *	@__sk_common: shared layout with inet_timewait_sock
196   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
197   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
198   *	@sk_lock:	synchronizer
199   *	@sk_rcvbuf: size of receive buffer in bytes
200   *	@sk_wq: sock wait queue and async head
201   *	@sk_rx_dst: receive input route used by early tcp demux
202   *	@sk_dst_cache: destination cache
203   *	@sk_dst_lock: destination cache lock
204   *	@sk_policy: flow policy
205   *	@sk_receive_queue: incoming packets
206   *	@sk_wmem_alloc: transmit queue bytes committed
207   *	@sk_write_queue: Packet sending queue
208   *	@sk_async_wait_queue: DMA copied packets
209   *	@sk_omem_alloc: "o" is "option" or "other"
210   *	@sk_wmem_queued: persistent queue size
211   *	@sk_forward_alloc: space allocated forward
212   *	@sk_allocation: allocation mode
213   *	@sk_sndbuf: size of send buffer in bytes
214   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
215   *		   %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
216   *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
217   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
218   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
219   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
220   *	@sk_gso_max_size: Maximum GSO segment size to build
221   *	@sk_lingertime: %SO_LINGER l_linger setting
222   *	@sk_backlog: always used with the per-socket spinlock held
223   *	@sk_callback_lock: used with the callbacks in the end of this struct
224   *	@sk_error_queue: rarely used
225   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
226   *			  IPV6_ADDRFORM for instance)
227   *	@sk_err: last error
228   *	@sk_err_soft: errors that don't cause failure but are the cause of a
229   *		      persistent failure not just 'timed out'
230   *	@sk_drops: raw/udp drops counter
231   *	@sk_ack_backlog: current listen backlog
232   *	@sk_max_ack_backlog: listen backlog set in listen()
233   *	@sk_priority: %SO_PRIORITY setting
234   *	@sk_cgrp_prioidx: socket group's priority map index
235   *	@sk_type: socket type (%SOCK_STREAM, etc)
236   *	@sk_protocol: which protocol this socket belongs in this network family
237   *	@sk_peer_pid: &struct pid for this socket's peer
238   *	@sk_peer_cred: %SO_PEERCRED setting
239   *	@sk_rcvlowat: %SO_RCVLOWAT setting
240   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
241   *	@sk_sndtimeo: %SO_SNDTIMEO setting
242   *	@sk_rxhash: flow hash received from netif layer
243   *	@sk_filter: socket filtering instructions
244   *	@sk_protinfo: private area, net family specific, when not using slab
245   *	@sk_timer: sock cleanup timer
246   *	@sk_stamp: time stamp of last packet received
247   *	@sk_socket: Identd and reporting IO signals
248   *	@sk_user_data: RPC layer private data
249   *	@sk_sndmsg_page: cached page for sendmsg
250   *	@sk_sndmsg_off: cached offset for sendmsg
251   *	@sk_peek_off: current peek_offset value
252   *	@sk_send_head: front of stuff to transmit
253   *	@sk_security: used by security modules
254   *	@sk_mark: generic packet mark
255   *	@sk_classid: this socket's cgroup classid
256   *	@sk_cgrp: this socket's cgroup-specific proto data
257   *	@sk_write_pending: a write to stream socket waits to start
258   *	@sk_state_change: callback to indicate change in the state of the sock
259   *	@sk_data_ready: callback to indicate there is data to be processed
260   *	@sk_write_space: callback to indicate there is bf sending space available
261   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
262   *	@sk_backlog_rcv: callback to process the backlog
263   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
264  */
265 struct sock {
266 	/*
267 	 * Now struct inet_timewait_sock also uses sock_common, so please just
268 	 * don't add nothing before this first member (__sk_common) --acme
269 	 */
270 	struct sock_common	__sk_common;
271 #define sk_node			__sk_common.skc_node
272 #define sk_nulls_node		__sk_common.skc_nulls_node
273 #define sk_refcnt		__sk_common.skc_refcnt
274 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
275 
276 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
277 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
278 #define sk_hash			__sk_common.skc_hash
279 #define sk_family		__sk_common.skc_family
280 #define sk_state		__sk_common.skc_state
281 #define sk_reuse		__sk_common.skc_reuse
282 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
283 #define sk_bind_node		__sk_common.skc_bind_node
284 #define sk_prot			__sk_common.skc_prot
285 #define sk_net			__sk_common.skc_net
286 	socket_lock_t		sk_lock;
287 	struct sk_buff_head	sk_receive_queue;
288 	/*
289 	 * The backlog queue is special, it is always used with
290 	 * the per-socket spinlock held and requires low latency
291 	 * access. Therefore we special case it's implementation.
292 	 * Note : rmem_alloc is in this structure to fill a hole
293 	 * on 64bit arches, not because its logically part of
294 	 * backlog.
295 	 */
296 	struct {
297 		atomic_t	rmem_alloc;
298 		int		len;
299 		struct sk_buff	*head;
300 		struct sk_buff	*tail;
301 	} sk_backlog;
302 #define sk_rmem_alloc sk_backlog.rmem_alloc
303 	int			sk_forward_alloc;
304 #ifdef CONFIG_RPS
305 	__u32			sk_rxhash;
306 #endif
307 	atomic_t		sk_drops;
308 	int			sk_rcvbuf;
309 
310 	struct sk_filter __rcu	*sk_filter;
311 	struct socket_wq __rcu	*sk_wq;
312 
313 #ifdef CONFIG_NET_DMA
314 	struct sk_buff_head	sk_async_wait_queue;
315 #endif
316 
317 #ifdef CONFIG_XFRM
318 	struct xfrm_policy	*sk_policy[2];
319 #endif
320 	unsigned long 		sk_flags;
321 	struct dst_entry	*sk_rx_dst;
322 	struct dst_entry	*sk_dst_cache;
323 	spinlock_t		sk_dst_lock;
324 	atomic_t		sk_wmem_alloc;
325 	atomic_t		sk_omem_alloc;
326 	int			sk_sndbuf;
327 	struct sk_buff_head	sk_write_queue;
328 	kmemcheck_bitfield_begin(flags);
329 	unsigned int		sk_shutdown  : 2,
330 				sk_no_check  : 2,
331 				sk_userlocks : 4,
332 				sk_protocol  : 8,
333 				sk_type      : 16;
334 	kmemcheck_bitfield_end(flags);
335 	int			sk_wmem_queued;
336 	gfp_t			sk_allocation;
337 	netdev_features_t	sk_route_caps;
338 	netdev_features_t	sk_route_nocaps;
339 	int			sk_gso_type;
340 	unsigned int		sk_gso_max_size;
341 	int			sk_rcvlowat;
342 	unsigned long	        sk_lingertime;
343 	struct sk_buff_head	sk_error_queue;
344 	struct proto		*sk_prot_creator;
345 	rwlock_t		sk_callback_lock;
346 	int			sk_err,
347 				sk_err_soft;
348 	unsigned short		sk_ack_backlog;
349 	unsigned short		sk_max_ack_backlog;
350 	__u32			sk_priority;
351 #ifdef CONFIG_CGROUPS
352 	__u32			sk_cgrp_prioidx;
353 #endif
354 	struct pid		*sk_peer_pid;
355 	const struct cred	*sk_peer_cred;
356 	long			sk_rcvtimeo;
357 	long			sk_sndtimeo;
358 	void			*sk_protinfo;
359 	struct timer_list	sk_timer;
360 	ktime_t			sk_stamp;
361 	struct socket		*sk_socket;
362 	void			*sk_user_data;
363 	struct page		*sk_sndmsg_page;
364 	struct sk_buff		*sk_send_head;
365 	__u32			sk_sndmsg_off;
366 	__s32			sk_peek_off;
367 	int			sk_write_pending;
368 #ifdef CONFIG_SECURITY
369 	void			*sk_security;
370 #endif
371 	__u32			sk_mark;
372 	u32			sk_classid;
373 	struct cg_proto		*sk_cgrp;
374 	void			(*sk_state_change)(struct sock *sk);
375 	void			(*sk_data_ready)(struct sock *sk, int bytes);
376 	void			(*sk_write_space)(struct sock *sk);
377 	void			(*sk_error_report)(struct sock *sk);
378 	int			(*sk_backlog_rcv)(struct sock *sk,
379 						  struct sk_buff *skb);
380 	void                    (*sk_destruct)(struct sock *sk);
381 };
382 
383 /*
384  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
385  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
386  * on a socket means that the socket will reuse everybody else's port
387  * without looking at the other's sk_reuse value.
388  */
389 
390 #define SK_NO_REUSE	0
391 #define SK_CAN_REUSE	1
392 #define SK_FORCE_REUSE	2
393 
394 static inline int sk_peek_offset(struct sock *sk, int flags)
395 {
396 	if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
397 		return sk->sk_peek_off;
398 	else
399 		return 0;
400 }
401 
402 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
403 {
404 	if (sk->sk_peek_off >= 0) {
405 		if (sk->sk_peek_off >= val)
406 			sk->sk_peek_off -= val;
407 		else
408 			sk->sk_peek_off = 0;
409 	}
410 }
411 
412 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
413 {
414 	if (sk->sk_peek_off >= 0)
415 		sk->sk_peek_off += val;
416 }
417 
418 /*
419  * Hashed lists helper routines
420  */
421 static inline struct sock *sk_entry(const struct hlist_node *node)
422 {
423 	return hlist_entry(node, struct sock, sk_node);
424 }
425 
426 static inline struct sock *__sk_head(const struct hlist_head *head)
427 {
428 	return hlist_entry(head->first, struct sock, sk_node);
429 }
430 
431 static inline struct sock *sk_head(const struct hlist_head *head)
432 {
433 	return hlist_empty(head) ? NULL : __sk_head(head);
434 }
435 
436 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
437 {
438 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
439 }
440 
441 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
442 {
443 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
444 }
445 
446 static inline struct sock *sk_next(const struct sock *sk)
447 {
448 	return sk->sk_node.next ?
449 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
450 }
451 
452 static inline struct sock *sk_nulls_next(const struct sock *sk)
453 {
454 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
455 		hlist_nulls_entry(sk->sk_nulls_node.next,
456 				  struct sock, sk_nulls_node) :
457 		NULL;
458 }
459 
460 static inline bool sk_unhashed(const struct sock *sk)
461 {
462 	return hlist_unhashed(&sk->sk_node);
463 }
464 
465 static inline bool sk_hashed(const struct sock *sk)
466 {
467 	return !sk_unhashed(sk);
468 }
469 
470 static inline void sk_node_init(struct hlist_node *node)
471 {
472 	node->pprev = NULL;
473 }
474 
475 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
476 {
477 	node->pprev = NULL;
478 }
479 
480 static inline void __sk_del_node(struct sock *sk)
481 {
482 	__hlist_del(&sk->sk_node);
483 }
484 
485 /* NB: equivalent to hlist_del_init_rcu */
486 static inline bool __sk_del_node_init(struct sock *sk)
487 {
488 	if (sk_hashed(sk)) {
489 		__sk_del_node(sk);
490 		sk_node_init(&sk->sk_node);
491 		return true;
492 	}
493 	return false;
494 }
495 
496 /* Grab socket reference count. This operation is valid only
497    when sk is ALREADY grabbed f.e. it is found in hash table
498    or a list and the lookup is made under lock preventing hash table
499    modifications.
500  */
501 
502 static inline void sock_hold(struct sock *sk)
503 {
504 	atomic_inc(&sk->sk_refcnt);
505 }
506 
507 /* Ungrab socket in the context, which assumes that socket refcnt
508    cannot hit zero, f.e. it is true in context of any socketcall.
509  */
510 static inline void __sock_put(struct sock *sk)
511 {
512 	atomic_dec(&sk->sk_refcnt);
513 }
514 
515 static inline bool sk_del_node_init(struct sock *sk)
516 {
517 	bool rc = __sk_del_node_init(sk);
518 
519 	if (rc) {
520 		/* paranoid for a while -acme */
521 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
522 		__sock_put(sk);
523 	}
524 	return rc;
525 }
526 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
527 
528 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
529 {
530 	if (sk_hashed(sk)) {
531 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
532 		return true;
533 	}
534 	return false;
535 }
536 
537 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
538 {
539 	bool rc = __sk_nulls_del_node_init_rcu(sk);
540 
541 	if (rc) {
542 		/* paranoid for a while -acme */
543 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
544 		__sock_put(sk);
545 	}
546 	return rc;
547 }
548 
549 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
550 {
551 	hlist_add_head(&sk->sk_node, list);
552 }
553 
554 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
555 {
556 	sock_hold(sk);
557 	__sk_add_node(sk, list);
558 }
559 
560 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
561 {
562 	sock_hold(sk);
563 	hlist_add_head_rcu(&sk->sk_node, list);
564 }
565 
566 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
567 {
568 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
569 }
570 
571 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
572 {
573 	sock_hold(sk);
574 	__sk_nulls_add_node_rcu(sk, list);
575 }
576 
577 static inline void __sk_del_bind_node(struct sock *sk)
578 {
579 	__hlist_del(&sk->sk_bind_node);
580 }
581 
582 static inline void sk_add_bind_node(struct sock *sk,
583 					struct hlist_head *list)
584 {
585 	hlist_add_head(&sk->sk_bind_node, list);
586 }
587 
588 #define sk_for_each(__sk, node, list) \
589 	hlist_for_each_entry(__sk, node, list, sk_node)
590 #define sk_for_each_rcu(__sk, node, list) \
591 	hlist_for_each_entry_rcu(__sk, node, list, sk_node)
592 #define sk_nulls_for_each(__sk, node, list) \
593 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
594 #define sk_nulls_for_each_rcu(__sk, node, list) \
595 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
596 #define sk_for_each_from(__sk, node) \
597 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
598 		hlist_for_each_entry_from(__sk, node, sk_node)
599 #define sk_nulls_for_each_from(__sk, node) \
600 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
601 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
602 #define sk_for_each_safe(__sk, node, tmp, list) \
603 	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
604 #define sk_for_each_bound(__sk, node, list) \
605 	hlist_for_each_entry(__sk, node, list, sk_bind_node)
606 
607 /* Sock flags */
608 enum sock_flags {
609 	SOCK_DEAD,
610 	SOCK_DONE,
611 	SOCK_URGINLINE,
612 	SOCK_KEEPOPEN,
613 	SOCK_LINGER,
614 	SOCK_DESTROY,
615 	SOCK_BROADCAST,
616 	SOCK_TIMESTAMP,
617 	SOCK_ZAPPED,
618 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
619 	SOCK_DBG, /* %SO_DEBUG setting */
620 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
621 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
622 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
623 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
624 	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
625 	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
626 	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
627 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
628 	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
629 	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
630 	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
631 	SOCK_FASYNC, /* fasync() active */
632 	SOCK_RXQ_OVFL,
633 	SOCK_ZEROCOPY, /* buffers from userspace */
634 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
635 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
636 		     * Will use last 4 bytes of packet sent from
637 		     * user-space instead.
638 		     */
639 };
640 
641 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
642 {
643 	nsk->sk_flags = osk->sk_flags;
644 }
645 
646 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
647 {
648 	__set_bit(flag, &sk->sk_flags);
649 }
650 
651 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
652 {
653 	__clear_bit(flag, &sk->sk_flags);
654 }
655 
656 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
657 {
658 	return test_bit(flag, &sk->sk_flags);
659 }
660 
661 static inline void sk_acceptq_removed(struct sock *sk)
662 {
663 	sk->sk_ack_backlog--;
664 }
665 
666 static inline void sk_acceptq_added(struct sock *sk)
667 {
668 	sk->sk_ack_backlog++;
669 }
670 
671 static inline bool sk_acceptq_is_full(const struct sock *sk)
672 {
673 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
674 }
675 
676 /*
677  * Compute minimal free write space needed to queue new packets.
678  */
679 static inline int sk_stream_min_wspace(const struct sock *sk)
680 {
681 	return sk->sk_wmem_queued >> 1;
682 }
683 
684 static inline int sk_stream_wspace(const struct sock *sk)
685 {
686 	return sk->sk_sndbuf - sk->sk_wmem_queued;
687 }
688 
689 extern void sk_stream_write_space(struct sock *sk);
690 
691 static inline bool sk_stream_memory_free(const struct sock *sk)
692 {
693 	return sk->sk_wmem_queued < sk->sk_sndbuf;
694 }
695 
696 /* OOB backlog add */
697 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
698 {
699 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
700 	skb_dst_force(skb);
701 
702 	if (!sk->sk_backlog.tail)
703 		sk->sk_backlog.head = skb;
704 	else
705 		sk->sk_backlog.tail->next = skb;
706 
707 	sk->sk_backlog.tail = skb;
708 	skb->next = NULL;
709 }
710 
711 /*
712  * Take into account size of receive queue and backlog queue
713  * Do not take into account this skb truesize,
714  * to allow even a single big packet to come.
715  */
716 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
717 				     unsigned int limit)
718 {
719 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
720 
721 	return qsize > limit;
722 }
723 
724 /* The per-socket spinlock must be held here. */
725 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
726 					      unsigned int limit)
727 {
728 	if (sk_rcvqueues_full(sk, skb, limit))
729 		return -ENOBUFS;
730 
731 	__sk_add_backlog(sk, skb);
732 	sk->sk_backlog.len += skb->truesize;
733 	return 0;
734 }
735 
736 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
737 {
738 	return sk->sk_backlog_rcv(sk, skb);
739 }
740 
741 static inline void sock_rps_record_flow(const struct sock *sk)
742 {
743 #ifdef CONFIG_RPS
744 	struct rps_sock_flow_table *sock_flow_table;
745 
746 	rcu_read_lock();
747 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
748 	rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
749 	rcu_read_unlock();
750 #endif
751 }
752 
753 static inline void sock_rps_reset_flow(const struct sock *sk)
754 {
755 #ifdef CONFIG_RPS
756 	struct rps_sock_flow_table *sock_flow_table;
757 
758 	rcu_read_lock();
759 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
760 	rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
761 	rcu_read_unlock();
762 #endif
763 }
764 
765 static inline void sock_rps_save_rxhash(struct sock *sk,
766 					const struct sk_buff *skb)
767 {
768 #ifdef CONFIG_RPS
769 	if (unlikely(sk->sk_rxhash != skb->rxhash)) {
770 		sock_rps_reset_flow(sk);
771 		sk->sk_rxhash = skb->rxhash;
772 	}
773 #endif
774 }
775 
776 static inline void sock_rps_reset_rxhash(struct sock *sk)
777 {
778 #ifdef CONFIG_RPS
779 	sock_rps_reset_flow(sk);
780 	sk->sk_rxhash = 0;
781 #endif
782 }
783 
784 #define sk_wait_event(__sk, __timeo, __condition)			\
785 	({	int __rc;						\
786 		release_sock(__sk);					\
787 		__rc = __condition;					\
788 		if (!__rc) {						\
789 			*(__timeo) = schedule_timeout(*(__timeo));	\
790 		}							\
791 		lock_sock(__sk);					\
792 		__rc = __condition;					\
793 		__rc;							\
794 	})
795 
796 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
797 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
798 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
799 extern int sk_stream_error(struct sock *sk, int flags, int err);
800 extern void sk_stream_kill_queues(struct sock *sk);
801 
802 extern int sk_wait_data(struct sock *sk, long *timeo);
803 
804 struct request_sock_ops;
805 struct timewait_sock_ops;
806 struct inet_hashinfo;
807 struct raw_hashinfo;
808 struct module;
809 
810 /* Networking protocol blocks we attach to sockets.
811  * socket layer -> transport layer interface
812  * transport -> network interface is defined by struct inet_proto
813  */
814 struct proto {
815 	void			(*close)(struct sock *sk,
816 					long timeout);
817 	int			(*connect)(struct sock *sk,
818 					struct sockaddr *uaddr,
819 					int addr_len);
820 	int			(*disconnect)(struct sock *sk, int flags);
821 
822 	struct sock *		(*accept)(struct sock *sk, int flags, int *err);
823 
824 	int			(*ioctl)(struct sock *sk, int cmd,
825 					 unsigned long arg);
826 	int			(*init)(struct sock *sk);
827 	void			(*destroy)(struct sock *sk);
828 	void			(*shutdown)(struct sock *sk, int how);
829 	int			(*setsockopt)(struct sock *sk, int level,
830 					int optname, char __user *optval,
831 					unsigned int optlen);
832 	int			(*getsockopt)(struct sock *sk, int level,
833 					int optname, char __user *optval,
834 					int __user *option);
835 #ifdef CONFIG_COMPAT
836 	int			(*compat_setsockopt)(struct sock *sk,
837 					int level,
838 					int optname, char __user *optval,
839 					unsigned int optlen);
840 	int			(*compat_getsockopt)(struct sock *sk,
841 					int level,
842 					int optname, char __user *optval,
843 					int __user *option);
844 	int			(*compat_ioctl)(struct sock *sk,
845 					unsigned int cmd, unsigned long arg);
846 #endif
847 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
848 					   struct msghdr *msg, size_t len);
849 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
850 					   struct msghdr *msg,
851 					   size_t len, int noblock, int flags,
852 					   int *addr_len);
853 	int			(*sendpage)(struct sock *sk, struct page *page,
854 					int offset, size_t size, int flags);
855 	int			(*bind)(struct sock *sk,
856 					struct sockaddr *uaddr, int addr_len);
857 
858 	int			(*backlog_rcv) (struct sock *sk,
859 						struct sk_buff *skb);
860 
861 	void		(*release_cb)(struct sock *sk);
862 	void		(*mtu_reduced)(struct sock *sk);
863 
864 	/* Keeping track of sk's, looking them up, and port selection methods. */
865 	void			(*hash)(struct sock *sk);
866 	void			(*unhash)(struct sock *sk);
867 	void			(*rehash)(struct sock *sk);
868 	int			(*get_port)(struct sock *sk, unsigned short snum);
869 	void			(*clear_sk)(struct sock *sk, int size);
870 
871 	/* Keeping track of sockets in use */
872 #ifdef CONFIG_PROC_FS
873 	unsigned int		inuse_idx;
874 #endif
875 
876 	/* Memory pressure */
877 	void			(*enter_memory_pressure)(struct sock *sk);
878 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
879 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
880 	/*
881 	 * Pressure flag: try to collapse.
882 	 * Technical note: it is used by multiple contexts non atomically.
883 	 * All the __sk_mem_schedule() is of this nature: accounting
884 	 * is strict, actions are advisory and have some latency.
885 	 */
886 	int			*memory_pressure;
887 	long			*sysctl_mem;
888 	int			*sysctl_wmem;
889 	int			*sysctl_rmem;
890 	int			max_header;
891 	bool			no_autobind;
892 
893 	struct kmem_cache	*slab;
894 	unsigned int		obj_size;
895 	int			slab_flags;
896 
897 	struct percpu_counter	*orphan_count;
898 
899 	struct request_sock_ops	*rsk_prot;
900 	struct timewait_sock_ops *twsk_prot;
901 
902 	union {
903 		struct inet_hashinfo	*hashinfo;
904 		struct udp_table	*udp_table;
905 		struct raw_hashinfo	*raw_hash;
906 	} h;
907 
908 	struct module		*owner;
909 
910 	char			name[32];
911 
912 	struct list_head	node;
913 #ifdef SOCK_REFCNT_DEBUG
914 	atomic_t		socks;
915 #endif
916 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
917 	/*
918 	 * cgroup specific init/deinit functions. Called once for all
919 	 * protocols that implement it, from cgroups populate function.
920 	 * This function has to setup any files the protocol want to
921 	 * appear in the kmem cgroup filesystem.
922 	 */
923 	int			(*init_cgroup)(struct mem_cgroup *memcg,
924 					       struct cgroup_subsys *ss);
925 	void			(*destroy_cgroup)(struct mem_cgroup *memcg);
926 	struct cg_proto		*(*proto_cgroup)(struct mem_cgroup *memcg);
927 #endif
928 };
929 
930 /*
931  * Bits in struct cg_proto.flags
932  */
933 enum cg_proto_flags {
934 	/* Currently active and new sockets should be assigned to cgroups */
935 	MEMCG_SOCK_ACTIVE,
936 	/* It was ever activated; we must disarm static keys on destruction */
937 	MEMCG_SOCK_ACTIVATED,
938 };
939 
940 struct cg_proto {
941 	void			(*enter_memory_pressure)(struct sock *sk);
942 	struct res_counter	*memory_allocated;	/* Current allocated memory. */
943 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
944 	int			*memory_pressure;
945 	long			*sysctl_mem;
946 	unsigned long		flags;
947 	/*
948 	 * memcg field is used to find which memcg we belong directly
949 	 * Each memcg struct can hold more than one cg_proto, so container_of
950 	 * won't really cut.
951 	 *
952 	 * The elegant solution would be having an inverse function to
953 	 * proto_cgroup in struct proto, but that means polluting the structure
954 	 * for everybody, instead of just for memcg users.
955 	 */
956 	struct mem_cgroup	*memcg;
957 };
958 
959 extern int proto_register(struct proto *prot, int alloc_slab);
960 extern void proto_unregister(struct proto *prot);
961 
962 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
963 {
964 	return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
965 }
966 
967 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
968 {
969 	return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
970 }
971 
972 #ifdef SOCK_REFCNT_DEBUG
973 static inline void sk_refcnt_debug_inc(struct sock *sk)
974 {
975 	atomic_inc(&sk->sk_prot->socks);
976 }
977 
978 static inline void sk_refcnt_debug_dec(struct sock *sk)
979 {
980 	atomic_dec(&sk->sk_prot->socks);
981 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
982 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
983 }
984 
985 inline void sk_refcnt_debug_release(const struct sock *sk)
986 {
987 	if (atomic_read(&sk->sk_refcnt) != 1)
988 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
989 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
990 }
991 #else /* SOCK_REFCNT_DEBUG */
992 #define sk_refcnt_debug_inc(sk) do { } while (0)
993 #define sk_refcnt_debug_dec(sk) do { } while (0)
994 #define sk_refcnt_debug_release(sk) do { } while (0)
995 #endif /* SOCK_REFCNT_DEBUG */
996 
997 #if defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM) && defined(CONFIG_NET)
998 extern struct static_key memcg_socket_limit_enabled;
999 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1000 					       struct cg_proto *cg_proto)
1001 {
1002 	return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1003 }
1004 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1005 #else
1006 #define mem_cgroup_sockets_enabled 0
1007 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1008 					       struct cg_proto *cg_proto)
1009 {
1010 	return NULL;
1011 }
1012 #endif
1013 
1014 
1015 static inline bool sk_has_memory_pressure(const struct sock *sk)
1016 {
1017 	return sk->sk_prot->memory_pressure != NULL;
1018 }
1019 
1020 static inline bool sk_under_memory_pressure(const struct sock *sk)
1021 {
1022 	if (!sk->sk_prot->memory_pressure)
1023 		return false;
1024 
1025 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1026 		return !!*sk->sk_cgrp->memory_pressure;
1027 
1028 	return !!*sk->sk_prot->memory_pressure;
1029 }
1030 
1031 static inline void sk_leave_memory_pressure(struct sock *sk)
1032 {
1033 	int *memory_pressure = sk->sk_prot->memory_pressure;
1034 
1035 	if (!memory_pressure)
1036 		return;
1037 
1038 	if (*memory_pressure)
1039 		*memory_pressure = 0;
1040 
1041 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1042 		struct cg_proto *cg_proto = sk->sk_cgrp;
1043 		struct proto *prot = sk->sk_prot;
1044 
1045 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1046 			if (*cg_proto->memory_pressure)
1047 				*cg_proto->memory_pressure = 0;
1048 	}
1049 
1050 }
1051 
1052 static inline void sk_enter_memory_pressure(struct sock *sk)
1053 {
1054 	if (!sk->sk_prot->enter_memory_pressure)
1055 		return;
1056 
1057 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1058 		struct cg_proto *cg_proto = sk->sk_cgrp;
1059 		struct proto *prot = sk->sk_prot;
1060 
1061 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1062 			cg_proto->enter_memory_pressure(sk);
1063 	}
1064 
1065 	sk->sk_prot->enter_memory_pressure(sk);
1066 }
1067 
1068 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1069 {
1070 	long *prot = sk->sk_prot->sysctl_mem;
1071 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1072 		prot = sk->sk_cgrp->sysctl_mem;
1073 	return prot[index];
1074 }
1075 
1076 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1077 					      unsigned long amt,
1078 					      int *parent_status)
1079 {
1080 	struct res_counter *fail;
1081 	int ret;
1082 
1083 	ret = res_counter_charge_nofail(prot->memory_allocated,
1084 					amt << PAGE_SHIFT, &fail);
1085 	if (ret < 0)
1086 		*parent_status = OVER_LIMIT;
1087 }
1088 
1089 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1090 					      unsigned long amt)
1091 {
1092 	res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1093 }
1094 
1095 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1096 {
1097 	u64 ret;
1098 	ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1099 	return ret >> PAGE_SHIFT;
1100 }
1101 
1102 static inline long
1103 sk_memory_allocated(const struct sock *sk)
1104 {
1105 	struct proto *prot = sk->sk_prot;
1106 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1107 		return memcg_memory_allocated_read(sk->sk_cgrp);
1108 
1109 	return atomic_long_read(prot->memory_allocated);
1110 }
1111 
1112 static inline long
1113 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1114 {
1115 	struct proto *prot = sk->sk_prot;
1116 
1117 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1118 		memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1119 		/* update the root cgroup regardless */
1120 		atomic_long_add_return(amt, prot->memory_allocated);
1121 		return memcg_memory_allocated_read(sk->sk_cgrp);
1122 	}
1123 
1124 	return atomic_long_add_return(amt, prot->memory_allocated);
1125 }
1126 
1127 static inline void
1128 sk_memory_allocated_sub(struct sock *sk, int amt)
1129 {
1130 	struct proto *prot = sk->sk_prot;
1131 
1132 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1133 		memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1134 
1135 	atomic_long_sub(amt, prot->memory_allocated);
1136 }
1137 
1138 static inline void sk_sockets_allocated_dec(struct sock *sk)
1139 {
1140 	struct proto *prot = sk->sk_prot;
1141 
1142 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1143 		struct cg_proto *cg_proto = sk->sk_cgrp;
1144 
1145 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1146 			percpu_counter_dec(cg_proto->sockets_allocated);
1147 	}
1148 
1149 	percpu_counter_dec(prot->sockets_allocated);
1150 }
1151 
1152 static inline void sk_sockets_allocated_inc(struct sock *sk)
1153 {
1154 	struct proto *prot = sk->sk_prot;
1155 
1156 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1157 		struct cg_proto *cg_proto = sk->sk_cgrp;
1158 
1159 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1160 			percpu_counter_inc(cg_proto->sockets_allocated);
1161 	}
1162 
1163 	percpu_counter_inc(prot->sockets_allocated);
1164 }
1165 
1166 static inline int
1167 sk_sockets_allocated_read_positive(struct sock *sk)
1168 {
1169 	struct proto *prot = sk->sk_prot;
1170 
1171 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1172 		return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1173 
1174 	return percpu_counter_read_positive(prot->sockets_allocated);
1175 }
1176 
1177 static inline int
1178 proto_sockets_allocated_sum_positive(struct proto *prot)
1179 {
1180 	return percpu_counter_sum_positive(prot->sockets_allocated);
1181 }
1182 
1183 static inline long
1184 proto_memory_allocated(struct proto *prot)
1185 {
1186 	return atomic_long_read(prot->memory_allocated);
1187 }
1188 
1189 static inline bool
1190 proto_memory_pressure(struct proto *prot)
1191 {
1192 	if (!prot->memory_pressure)
1193 		return false;
1194 	return !!*prot->memory_pressure;
1195 }
1196 
1197 
1198 #ifdef CONFIG_PROC_FS
1199 /* Called with local bh disabled */
1200 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1201 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1202 #else
1203 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1204 		int inc)
1205 {
1206 }
1207 #endif
1208 
1209 
1210 /* With per-bucket locks this operation is not-atomic, so that
1211  * this version is not worse.
1212  */
1213 static inline void __sk_prot_rehash(struct sock *sk)
1214 {
1215 	sk->sk_prot->unhash(sk);
1216 	sk->sk_prot->hash(sk);
1217 }
1218 
1219 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1220 
1221 /* About 10 seconds */
1222 #define SOCK_DESTROY_TIME (10*HZ)
1223 
1224 /* Sockets 0-1023 can't be bound to unless you are superuser */
1225 #define PROT_SOCK	1024
1226 
1227 #define SHUTDOWN_MASK	3
1228 #define RCV_SHUTDOWN	1
1229 #define SEND_SHUTDOWN	2
1230 
1231 #define SOCK_SNDBUF_LOCK	1
1232 #define SOCK_RCVBUF_LOCK	2
1233 #define SOCK_BINDADDR_LOCK	4
1234 #define SOCK_BINDPORT_LOCK	8
1235 
1236 /* sock_iocb: used to kick off async processing of socket ios */
1237 struct sock_iocb {
1238 	struct list_head	list;
1239 
1240 	int			flags;
1241 	int			size;
1242 	struct socket		*sock;
1243 	struct sock		*sk;
1244 	struct scm_cookie	*scm;
1245 	struct msghdr		*msg, async_msg;
1246 	struct kiocb		*kiocb;
1247 };
1248 
1249 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1250 {
1251 	return (struct sock_iocb *)iocb->private;
1252 }
1253 
1254 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1255 {
1256 	return si->kiocb;
1257 }
1258 
1259 struct socket_alloc {
1260 	struct socket socket;
1261 	struct inode vfs_inode;
1262 };
1263 
1264 static inline struct socket *SOCKET_I(struct inode *inode)
1265 {
1266 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1267 }
1268 
1269 static inline struct inode *SOCK_INODE(struct socket *socket)
1270 {
1271 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1272 }
1273 
1274 /*
1275  * Functions for memory accounting
1276  */
1277 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1278 extern void __sk_mem_reclaim(struct sock *sk);
1279 
1280 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1281 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1282 #define SK_MEM_SEND	0
1283 #define SK_MEM_RECV	1
1284 
1285 static inline int sk_mem_pages(int amt)
1286 {
1287 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1288 }
1289 
1290 static inline bool sk_has_account(struct sock *sk)
1291 {
1292 	/* return true if protocol supports memory accounting */
1293 	return !!sk->sk_prot->memory_allocated;
1294 }
1295 
1296 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1297 {
1298 	if (!sk_has_account(sk))
1299 		return true;
1300 	return size <= sk->sk_forward_alloc ||
1301 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1302 }
1303 
1304 static inline bool sk_rmem_schedule(struct sock *sk, int size)
1305 {
1306 	if (!sk_has_account(sk))
1307 		return true;
1308 	return size <= sk->sk_forward_alloc ||
1309 		__sk_mem_schedule(sk, size, SK_MEM_RECV);
1310 }
1311 
1312 static inline void sk_mem_reclaim(struct sock *sk)
1313 {
1314 	if (!sk_has_account(sk))
1315 		return;
1316 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1317 		__sk_mem_reclaim(sk);
1318 }
1319 
1320 static inline void sk_mem_reclaim_partial(struct sock *sk)
1321 {
1322 	if (!sk_has_account(sk))
1323 		return;
1324 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1325 		__sk_mem_reclaim(sk);
1326 }
1327 
1328 static inline void sk_mem_charge(struct sock *sk, int size)
1329 {
1330 	if (!sk_has_account(sk))
1331 		return;
1332 	sk->sk_forward_alloc -= size;
1333 }
1334 
1335 static inline void sk_mem_uncharge(struct sock *sk, int size)
1336 {
1337 	if (!sk_has_account(sk))
1338 		return;
1339 	sk->sk_forward_alloc += size;
1340 }
1341 
1342 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1343 {
1344 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1345 	sk->sk_wmem_queued -= skb->truesize;
1346 	sk_mem_uncharge(sk, skb->truesize);
1347 	__kfree_skb(skb);
1348 }
1349 
1350 /* Used by processes to "lock" a socket state, so that
1351  * interrupts and bottom half handlers won't change it
1352  * from under us. It essentially blocks any incoming
1353  * packets, so that we won't get any new data or any
1354  * packets that change the state of the socket.
1355  *
1356  * While locked, BH processing will add new packets to
1357  * the backlog queue.  This queue is processed by the
1358  * owner of the socket lock right before it is released.
1359  *
1360  * Since ~2.3.5 it is also exclusive sleep lock serializing
1361  * accesses from user process context.
1362  */
1363 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1364 
1365 /*
1366  * Macro so as to not evaluate some arguments when
1367  * lockdep is not enabled.
1368  *
1369  * Mark both the sk_lock and the sk_lock.slock as a
1370  * per-address-family lock class.
1371  */
1372 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1373 do {									\
1374 	sk->sk_lock.owned = 0;						\
1375 	init_waitqueue_head(&sk->sk_lock.wq);				\
1376 	spin_lock_init(&(sk)->sk_lock.slock);				\
1377 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1378 			sizeof((sk)->sk_lock));				\
1379 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1380 				(skey), (sname));				\
1381 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1382 } while (0)
1383 
1384 extern void lock_sock_nested(struct sock *sk, int subclass);
1385 
1386 static inline void lock_sock(struct sock *sk)
1387 {
1388 	lock_sock_nested(sk, 0);
1389 }
1390 
1391 extern void release_sock(struct sock *sk);
1392 
1393 /* BH context may only use the following locking interface. */
1394 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1395 #define bh_lock_sock_nested(__sk) \
1396 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1397 				SINGLE_DEPTH_NESTING)
1398 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1399 
1400 extern bool lock_sock_fast(struct sock *sk);
1401 /**
1402  * unlock_sock_fast - complement of lock_sock_fast
1403  * @sk: socket
1404  * @slow: slow mode
1405  *
1406  * fast unlock socket for user context.
1407  * If slow mode is on, we call regular release_sock()
1408  */
1409 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1410 {
1411 	if (slow)
1412 		release_sock(sk);
1413 	else
1414 		spin_unlock_bh(&sk->sk_lock.slock);
1415 }
1416 
1417 
1418 extern struct sock		*sk_alloc(struct net *net, int family,
1419 					  gfp_t priority,
1420 					  struct proto *prot);
1421 extern void			sk_free(struct sock *sk);
1422 extern void			sk_release_kernel(struct sock *sk);
1423 extern struct sock		*sk_clone_lock(const struct sock *sk,
1424 					       const gfp_t priority);
1425 
1426 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
1427 					      unsigned long size, int force,
1428 					      gfp_t priority);
1429 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
1430 					      unsigned long size, int force,
1431 					      gfp_t priority);
1432 extern void			sock_wfree(struct sk_buff *skb);
1433 extern void			sock_rfree(struct sk_buff *skb);
1434 extern void			sock_edemux(struct sk_buff *skb);
1435 
1436 extern int			sock_setsockopt(struct socket *sock, int level,
1437 						int op, char __user *optval,
1438 						unsigned int optlen);
1439 
1440 extern int			sock_getsockopt(struct socket *sock, int level,
1441 						int op, char __user *optval,
1442 						int __user *optlen);
1443 extern struct sk_buff		*sock_alloc_send_skb(struct sock *sk,
1444 						     unsigned long size,
1445 						     int noblock,
1446 						     int *errcode);
1447 extern struct sk_buff		*sock_alloc_send_pskb(struct sock *sk,
1448 						      unsigned long header_len,
1449 						      unsigned long data_len,
1450 						      int noblock,
1451 						      int *errcode);
1452 extern void *sock_kmalloc(struct sock *sk, int size,
1453 			  gfp_t priority);
1454 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1455 extern void sk_send_sigurg(struct sock *sk);
1456 
1457 #ifdef CONFIG_CGROUPS
1458 extern void sock_update_classid(struct sock *sk);
1459 #else
1460 static inline void sock_update_classid(struct sock *sk)
1461 {
1462 }
1463 #endif
1464 
1465 /*
1466  * Functions to fill in entries in struct proto_ops when a protocol
1467  * does not implement a particular function.
1468  */
1469 extern int                      sock_no_bind(struct socket *,
1470 					     struct sockaddr *, int);
1471 extern int                      sock_no_connect(struct socket *,
1472 						struct sockaddr *, int, int);
1473 extern int                      sock_no_socketpair(struct socket *,
1474 						   struct socket *);
1475 extern int                      sock_no_accept(struct socket *,
1476 					       struct socket *, int);
1477 extern int                      sock_no_getname(struct socket *,
1478 						struct sockaddr *, int *, int);
1479 extern unsigned int             sock_no_poll(struct file *, struct socket *,
1480 					     struct poll_table_struct *);
1481 extern int                      sock_no_ioctl(struct socket *, unsigned int,
1482 					      unsigned long);
1483 extern int			sock_no_listen(struct socket *, int);
1484 extern int                      sock_no_shutdown(struct socket *, int);
1485 extern int			sock_no_getsockopt(struct socket *, int , int,
1486 						   char __user *, int __user *);
1487 extern int			sock_no_setsockopt(struct socket *, int, int,
1488 						   char __user *, unsigned int);
1489 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1490 						struct msghdr *, size_t);
1491 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1492 						struct msghdr *, size_t, int);
1493 extern int			sock_no_mmap(struct file *file,
1494 					     struct socket *sock,
1495 					     struct vm_area_struct *vma);
1496 extern ssize_t			sock_no_sendpage(struct socket *sock,
1497 						struct page *page,
1498 						int offset, size_t size,
1499 						int flags);
1500 
1501 /*
1502  * Functions to fill in entries in struct proto_ops when a protocol
1503  * uses the inet style.
1504  */
1505 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1506 				  char __user *optval, int __user *optlen);
1507 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1508 			       struct msghdr *msg, size_t size, int flags);
1509 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1510 				  char __user *optval, unsigned int optlen);
1511 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1512 		int optname, char __user *optval, int __user *optlen);
1513 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1514 		int optname, char __user *optval, unsigned int optlen);
1515 
1516 extern void sk_common_release(struct sock *sk);
1517 
1518 /*
1519  *	Default socket callbacks and setup code
1520  */
1521 
1522 /* Initialise core socket variables */
1523 extern void sock_init_data(struct socket *sock, struct sock *sk);
1524 
1525 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1526 
1527 /**
1528  *	sk_filter_release - release a socket filter
1529  *	@fp: filter to remove
1530  *
1531  *	Remove a filter from a socket and release its resources.
1532  */
1533 
1534 static inline void sk_filter_release(struct sk_filter *fp)
1535 {
1536 	if (atomic_dec_and_test(&fp->refcnt))
1537 		call_rcu(&fp->rcu, sk_filter_release_rcu);
1538 }
1539 
1540 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1541 {
1542 	unsigned int size = sk_filter_len(fp);
1543 
1544 	atomic_sub(size, &sk->sk_omem_alloc);
1545 	sk_filter_release(fp);
1546 }
1547 
1548 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1549 {
1550 	atomic_inc(&fp->refcnt);
1551 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1552 }
1553 
1554 /*
1555  * Socket reference counting postulates.
1556  *
1557  * * Each user of socket SHOULD hold a reference count.
1558  * * Each access point to socket (an hash table bucket, reference from a list,
1559  *   running timer, skb in flight MUST hold a reference count.
1560  * * When reference count hits 0, it means it will never increase back.
1561  * * When reference count hits 0, it means that no references from
1562  *   outside exist to this socket and current process on current CPU
1563  *   is last user and may/should destroy this socket.
1564  * * sk_free is called from any context: process, BH, IRQ. When
1565  *   it is called, socket has no references from outside -> sk_free
1566  *   may release descendant resources allocated by the socket, but
1567  *   to the time when it is called, socket is NOT referenced by any
1568  *   hash tables, lists etc.
1569  * * Packets, delivered from outside (from network or from another process)
1570  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1571  *   when they sit in queue. Otherwise, packets will leak to hole, when
1572  *   socket is looked up by one cpu and unhasing is made by another CPU.
1573  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1574  *   (leak to backlog). Packet socket does all the processing inside
1575  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1576  *   use separate SMP lock, so that they are prone too.
1577  */
1578 
1579 /* Ungrab socket and destroy it, if it was the last reference. */
1580 static inline void sock_put(struct sock *sk)
1581 {
1582 	if (atomic_dec_and_test(&sk->sk_refcnt))
1583 		sk_free(sk);
1584 }
1585 
1586 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1587 			  const int nested);
1588 
1589 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1590 {
1591 	sk->sk_tx_queue_mapping = tx_queue;
1592 }
1593 
1594 static inline void sk_tx_queue_clear(struct sock *sk)
1595 {
1596 	sk->sk_tx_queue_mapping = -1;
1597 }
1598 
1599 static inline int sk_tx_queue_get(const struct sock *sk)
1600 {
1601 	return sk ? sk->sk_tx_queue_mapping : -1;
1602 }
1603 
1604 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1605 {
1606 	sk_tx_queue_clear(sk);
1607 	sk->sk_socket = sock;
1608 }
1609 
1610 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1611 {
1612 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1613 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1614 }
1615 /* Detach socket from process context.
1616  * Announce socket dead, detach it from wait queue and inode.
1617  * Note that parent inode held reference count on this struct sock,
1618  * we do not release it in this function, because protocol
1619  * probably wants some additional cleanups or even continuing
1620  * to work with this socket (TCP).
1621  */
1622 static inline void sock_orphan(struct sock *sk)
1623 {
1624 	write_lock_bh(&sk->sk_callback_lock);
1625 	sock_set_flag(sk, SOCK_DEAD);
1626 	sk_set_socket(sk, NULL);
1627 	sk->sk_wq  = NULL;
1628 	write_unlock_bh(&sk->sk_callback_lock);
1629 }
1630 
1631 static inline void sock_graft(struct sock *sk, struct socket *parent)
1632 {
1633 	write_lock_bh(&sk->sk_callback_lock);
1634 	sk->sk_wq = parent->wq;
1635 	parent->sk = sk;
1636 	sk_set_socket(sk, parent);
1637 	security_sock_graft(sk, parent);
1638 	write_unlock_bh(&sk->sk_callback_lock);
1639 }
1640 
1641 extern int sock_i_uid(struct sock *sk);
1642 extern unsigned long sock_i_ino(struct sock *sk);
1643 
1644 static inline struct dst_entry *
1645 __sk_dst_get(struct sock *sk)
1646 {
1647 	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1648 						       lockdep_is_held(&sk->sk_lock.slock));
1649 }
1650 
1651 static inline struct dst_entry *
1652 sk_dst_get(struct sock *sk)
1653 {
1654 	struct dst_entry *dst;
1655 
1656 	rcu_read_lock();
1657 	dst = rcu_dereference(sk->sk_dst_cache);
1658 	if (dst)
1659 		dst_hold(dst);
1660 	rcu_read_unlock();
1661 	return dst;
1662 }
1663 
1664 extern void sk_reset_txq(struct sock *sk);
1665 
1666 static inline void dst_negative_advice(struct sock *sk)
1667 {
1668 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1669 
1670 	if (dst && dst->ops->negative_advice) {
1671 		ndst = dst->ops->negative_advice(dst);
1672 
1673 		if (ndst != dst) {
1674 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1675 			sk_reset_txq(sk);
1676 		}
1677 	}
1678 }
1679 
1680 static inline void
1681 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1682 {
1683 	struct dst_entry *old_dst;
1684 
1685 	sk_tx_queue_clear(sk);
1686 	/*
1687 	 * This can be called while sk is owned by the caller only,
1688 	 * with no state that can be checked in a rcu_dereference_check() cond
1689 	 */
1690 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1691 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1692 	dst_release(old_dst);
1693 }
1694 
1695 static inline void
1696 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1697 {
1698 	spin_lock(&sk->sk_dst_lock);
1699 	__sk_dst_set(sk, dst);
1700 	spin_unlock(&sk->sk_dst_lock);
1701 }
1702 
1703 static inline void
1704 __sk_dst_reset(struct sock *sk)
1705 {
1706 	__sk_dst_set(sk, NULL);
1707 }
1708 
1709 static inline void
1710 sk_dst_reset(struct sock *sk)
1711 {
1712 	spin_lock(&sk->sk_dst_lock);
1713 	__sk_dst_reset(sk);
1714 	spin_unlock(&sk->sk_dst_lock);
1715 }
1716 
1717 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1718 
1719 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1720 
1721 static inline bool sk_can_gso(const struct sock *sk)
1722 {
1723 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1724 }
1725 
1726 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1727 
1728 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1729 {
1730 	sk->sk_route_nocaps |= flags;
1731 	sk->sk_route_caps &= ~flags;
1732 }
1733 
1734 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1735 					   char __user *from, char *to,
1736 					   int copy, int offset)
1737 {
1738 	if (skb->ip_summed == CHECKSUM_NONE) {
1739 		int err = 0;
1740 		__wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1741 		if (err)
1742 			return err;
1743 		skb->csum = csum_block_add(skb->csum, csum, offset);
1744 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1745 		if (!access_ok(VERIFY_READ, from, copy) ||
1746 		    __copy_from_user_nocache(to, from, copy))
1747 			return -EFAULT;
1748 	} else if (copy_from_user(to, from, copy))
1749 		return -EFAULT;
1750 
1751 	return 0;
1752 }
1753 
1754 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1755 				       char __user *from, int copy)
1756 {
1757 	int err, offset = skb->len;
1758 
1759 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1760 				       copy, offset);
1761 	if (err)
1762 		__skb_trim(skb, offset);
1763 
1764 	return err;
1765 }
1766 
1767 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1768 					   struct sk_buff *skb,
1769 					   struct page *page,
1770 					   int off, int copy)
1771 {
1772 	int err;
1773 
1774 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1775 				       copy, skb->len);
1776 	if (err)
1777 		return err;
1778 
1779 	skb->len	     += copy;
1780 	skb->data_len	     += copy;
1781 	skb->truesize	     += copy;
1782 	sk->sk_wmem_queued   += copy;
1783 	sk_mem_charge(sk, copy);
1784 	return 0;
1785 }
1786 
1787 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1788 				   struct sk_buff *skb, struct page *page,
1789 				   int off, int copy)
1790 {
1791 	if (skb->ip_summed == CHECKSUM_NONE) {
1792 		int err = 0;
1793 		__wsum csum = csum_and_copy_from_user(from,
1794 						     page_address(page) + off,
1795 							    copy, 0, &err);
1796 		if (err)
1797 			return err;
1798 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1799 	} else if (copy_from_user(page_address(page) + off, from, copy))
1800 		return -EFAULT;
1801 
1802 	skb->len	     += copy;
1803 	skb->data_len	     += copy;
1804 	skb->truesize	     += copy;
1805 	sk->sk_wmem_queued   += copy;
1806 	sk_mem_charge(sk, copy);
1807 	return 0;
1808 }
1809 
1810 /**
1811  * sk_wmem_alloc_get - returns write allocations
1812  * @sk: socket
1813  *
1814  * Returns sk_wmem_alloc minus initial offset of one
1815  */
1816 static inline int sk_wmem_alloc_get(const struct sock *sk)
1817 {
1818 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1819 }
1820 
1821 /**
1822  * sk_rmem_alloc_get - returns read allocations
1823  * @sk: socket
1824  *
1825  * Returns sk_rmem_alloc
1826  */
1827 static inline int sk_rmem_alloc_get(const struct sock *sk)
1828 {
1829 	return atomic_read(&sk->sk_rmem_alloc);
1830 }
1831 
1832 /**
1833  * sk_has_allocations - check if allocations are outstanding
1834  * @sk: socket
1835  *
1836  * Returns true if socket has write or read allocations
1837  */
1838 static inline bool sk_has_allocations(const struct sock *sk)
1839 {
1840 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1841 }
1842 
1843 /**
1844  * wq_has_sleeper - check if there are any waiting processes
1845  * @wq: struct socket_wq
1846  *
1847  * Returns true if socket_wq has waiting processes
1848  *
1849  * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1850  * barrier call. They were added due to the race found within the tcp code.
1851  *
1852  * Consider following tcp code paths:
1853  *
1854  * CPU1                  CPU2
1855  *
1856  * sys_select            receive packet
1857  *   ...                 ...
1858  *   __add_wait_queue    update tp->rcv_nxt
1859  *   ...                 ...
1860  *   tp->rcv_nxt check   sock_def_readable
1861  *   ...                 {
1862  *   schedule               rcu_read_lock();
1863  *                          wq = rcu_dereference(sk->sk_wq);
1864  *                          if (wq && waitqueue_active(&wq->wait))
1865  *                              wake_up_interruptible(&wq->wait)
1866  *                          ...
1867  *                       }
1868  *
1869  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1870  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1871  * could then endup calling schedule and sleep forever if there are no more
1872  * data on the socket.
1873  *
1874  */
1875 static inline bool wq_has_sleeper(struct socket_wq *wq)
1876 {
1877 	/* We need to be sure we are in sync with the
1878 	 * add_wait_queue modifications to the wait queue.
1879 	 *
1880 	 * This memory barrier is paired in the sock_poll_wait.
1881 	 */
1882 	smp_mb();
1883 	return wq && waitqueue_active(&wq->wait);
1884 }
1885 
1886 /**
1887  * sock_poll_wait - place memory barrier behind the poll_wait call.
1888  * @filp:           file
1889  * @wait_address:   socket wait queue
1890  * @p:              poll_table
1891  *
1892  * See the comments in the wq_has_sleeper function.
1893  */
1894 static inline void sock_poll_wait(struct file *filp,
1895 		wait_queue_head_t *wait_address, poll_table *p)
1896 {
1897 	if (!poll_does_not_wait(p) && wait_address) {
1898 		poll_wait(filp, wait_address, p);
1899 		/* We need to be sure we are in sync with the
1900 		 * socket flags modification.
1901 		 *
1902 		 * This memory barrier is paired in the wq_has_sleeper.
1903 		 */
1904 		smp_mb();
1905 	}
1906 }
1907 
1908 /*
1909  *	Queue a received datagram if it will fit. Stream and sequenced
1910  *	protocols can't normally use this as they need to fit buffers in
1911  *	and play with them.
1912  *
1913  *	Inlined as it's very short and called for pretty much every
1914  *	packet ever received.
1915  */
1916 
1917 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1918 {
1919 	skb_orphan(skb);
1920 	skb->sk = sk;
1921 	skb->destructor = sock_wfree;
1922 	/*
1923 	 * We used to take a refcount on sk, but following operation
1924 	 * is enough to guarantee sk_free() wont free this sock until
1925 	 * all in-flight packets are completed
1926 	 */
1927 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1928 }
1929 
1930 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1931 {
1932 	skb_orphan(skb);
1933 	skb->sk = sk;
1934 	skb->destructor = sock_rfree;
1935 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1936 	sk_mem_charge(sk, skb->truesize);
1937 }
1938 
1939 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1940 			   unsigned long expires);
1941 
1942 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1943 
1944 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1945 
1946 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1947 
1948 /*
1949  *	Recover an error report and clear atomically
1950  */
1951 
1952 static inline int sock_error(struct sock *sk)
1953 {
1954 	int err;
1955 	if (likely(!sk->sk_err))
1956 		return 0;
1957 	err = xchg(&sk->sk_err, 0);
1958 	return -err;
1959 }
1960 
1961 static inline unsigned long sock_wspace(struct sock *sk)
1962 {
1963 	int amt = 0;
1964 
1965 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1966 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1967 		if (amt < 0)
1968 			amt = 0;
1969 	}
1970 	return amt;
1971 }
1972 
1973 static inline void sk_wake_async(struct sock *sk, int how, int band)
1974 {
1975 	if (sock_flag(sk, SOCK_FASYNC))
1976 		sock_wake_async(sk->sk_socket, how, band);
1977 }
1978 
1979 #define SOCK_MIN_SNDBUF 2048
1980 /*
1981  * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1982  * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1983  */
1984 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1985 
1986 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1987 {
1988 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1989 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1990 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1991 	}
1992 }
1993 
1994 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1995 
1996 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1997 {
1998 	struct page *page = NULL;
1999 
2000 	page = alloc_pages(sk->sk_allocation, 0);
2001 	if (!page) {
2002 		sk_enter_memory_pressure(sk);
2003 		sk_stream_moderate_sndbuf(sk);
2004 	}
2005 	return page;
2006 }
2007 
2008 /*
2009  *	Default write policy as shown to user space via poll/select/SIGIO
2010  */
2011 static inline bool sock_writeable(const struct sock *sk)
2012 {
2013 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2014 }
2015 
2016 static inline gfp_t gfp_any(void)
2017 {
2018 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2019 }
2020 
2021 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2022 {
2023 	return noblock ? 0 : sk->sk_rcvtimeo;
2024 }
2025 
2026 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2027 {
2028 	return noblock ? 0 : sk->sk_sndtimeo;
2029 }
2030 
2031 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2032 {
2033 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2034 }
2035 
2036 /* Alas, with timeout socket operations are not restartable.
2037  * Compare this to poll().
2038  */
2039 static inline int sock_intr_errno(long timeo)
2040 {
2041 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2042 }
2043 
2044 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2045 	struct sk_buff *skb);
2046 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2047 	struct sk_buff *skb);
2048 
2049 static inline void
2050 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2051 {
2052 	ktime_t kt = skb->tstamp;
2053 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2054 
2055 	/*
2056 	 * generate control messages if
2057 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2058 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
2059 	 * - software time stamp available and wanted
2060 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
2061 	 * - hardware time stamps available and wanted
2062 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
2063 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
2064 	 */
2065 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2066 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2067 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2068 	    (hwtstamps->hwtstamp.tv64 &&
2069 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2070 	    (hwtstamps->syststamp.tv64 &&
2071 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2072 		__sock_recv_timestamp(msg, sk, skb);
2073 	else
2074 		sk->sk_stamp = kt;
2075 
2076 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2077 		__sock_recv_wifi_status(msg, sk, skb);
2078 }
2079 
2080 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2081 				     struct sk_buff *skb);
2082 
2083 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2084 					  struct sk_buff *skb)
2085 {
2086 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2087 			   (1UL << SOCK_RCVTSTAMP)			| \
2088 			   (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)	| \
2089 			   (1UL << SOCK_TIMESTAMPING_SOFTWARE)		| \
2090 			   (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE)	| \
2091 			   (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2092 
2093 	if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2094 		__sock_recv_ts_and_drops(msg, sk, skb);
2095 	else
2096 		sk->sk_stamp = skb->tstamp;
2097 }
2098 
2099 /**
2100  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2101  * @sk:		socket sending this packet
2102  * @tx_flags:	filled with instructions for time stamping
2103  *
2104  * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2105  * parameters are invalid.
2106  */
2107 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2108 
2109 /**
2110  * sk_eat_skb - Release a skb if it is no longer needed
2111  * @sk: socket to eat this skb from
2112  * @skb: socket buffer to eat
2113  * @copied_early: flag indicating whether DMA operations copied this data early
2114  *
2115  * This routine must be called with interrupts disabled or with the socket
2116  * locked so that the sk_buff queue operation is ok.
2117 */
2118 #ifdef CONFIG_NET_DMA
2119 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2120 {
2121 	__skb_unlink(skb, &sk->sk_receive_queue);
2122 	if (!copied_early)
2123 		__kfree_skb(skb);
2124 	else
2125 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
2126 }
2127 #else
2128 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2129 {
2130 	__skb_unlink(skb, &sk->sk_receive_queue);
2131 	__kfree_skb(skb);
2132 }
2133 #endif
2134 
2135 static inline
2136 struct net *sock_net(const struct sock *sk)
2137 {
2138 	return read_pnet(&sk->sk_net);
2139 }
2140 
2141 static inline
2142 void sock_net_set(struct sock *sk, struct net *net)
2143 {
2144 	write_pnet(&sk->sk_net, net);
2145 }
2146 
2147 /*
2148  * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2149  * They should not hold a reference to a namespace in order to allow
2150  * to stop it.
2151  * Sockets after sk_change_net should be released using sk_release_kernel
2152  */
2153 static inline void sk_change_net(struct sock *sk, struct net *net)
2154 {
2155 	put_net(sock_net(sk));
2156 	sock_net_set(sk, hold_net(net));
2157 }
2158 
2159 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2160 {
2161 	if (skb->sk) {
2162 		struct sock *sk = skb->sk;
2163 
2164 		skb->destructor = NULL;
2165 		skb->sk = NULL;
2166 		return sk;
2167 	}
2168 	return NULL;
2169 }
2170 
2171 extern void sock_enable_timestamp(struct sock *sk, int flag);
2172 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2173 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2174 
2175 /*
2176  *	Enable debug/info messages
2177  */
2178 extern int net_msg_warn;
2179 #define NETDEBUG(fmt, args...) \
2180 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
2181 
2182 #define LIMIT_NETDEBUG(fmt, args...) \
2183 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2184 
2185 extern __u32 sysctl_wmem_max;
2186 extern __u32 sysctl_rmem_max;
2187 
2188 extern void sk_init(void);
2189 
2190 extern int sysctl_optmem_max;
2191 
2192 extern __u32 sysctl_wmem_default;
2193 extern __u32 sysctl_rmem_default;
2194 
2195 #endif	/* _SOCK_H */
2196