xref: /linux/include/net/sock.h (revision 25aee3debe0464f6c680173041fa3de30ec9ff54)
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_MEMALLOC, /* VM depends on this socket for swapping */
625 	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
626 	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
627 	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
628 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
629 	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
630 	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
631 	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
632 	SOCK_FASYNC, /* fasync() active */
633 	SOCK_RXQ_OVFL,
634 	SOCK_ZEROCOPY, /* buffers from userspace */
635 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
636 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
637 		     * Will use last 4 bytes of packet sent from
638 		     * user-space instead.
639 		     */
640 };
641 
642 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
643 {
644 	nsk->sk_flags = osk->sk_flags;
645 }
646 
647 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
648 {
649 	__set_bit(flag, &sk->sk_flags);
650 }
651 
652 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
653 {
654 	__clear_bit(flag, &sk->sk_flags);
655 }
656 
657 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
658 {
659 	return test_bit(flag, &sk->sk_flags);
660 }
661 
662 #ifdef CONFIG_NET
663 extern struct static_key memalloc_socks;
664 static inline int sk_memalloc_socks(void)
665 {
666 	return static_key_false(&memalloc_socks);
667 }
668 #else
669 
670 static inline int sk_memalloc_socks(void)
671 {
672 	return 0;
673 }
674 
675 #endif
676 
677 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
678 {
679 	return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
680 }
681 
682 static inline void sk_acceptq_removed(struct sock *sk)
683 {
684 	sk->sk_ack_backlog--;
685 }
686 
687 static inline void sk_acceptq_added(struct sock *sk)
688 {
689 	sk->sk_ack_backlog++;
690 }
691 
692 static inline bool sk_acceptq_is_full(const struct sock *sk)
693 {
694 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
695 }
696 
697 /*
698  * Compute minimal free write space needed to queue new packets.
699  */
700 static inline int sk_stream_min_wspace(const struct sock *sk)
701 {
702 	return sk->sk_wmem_queued >> 1;
703 }
704 
705 static inline int sk_stream_wspace(const struct sock *sk)
706 {
707 	return sk->sk_sndbuf - sk->sk_wmem_queued;
708 }
709 
710 extern void sk_stream_write_space(struct sock *sk);
711 
712 static inline bool sk_stream_memory_free(const struct sock *sk)
713 {
714 	return sk->sk_wmem_queued < sk->sk_sndbuf;
715 }
716 
717 /* OOB backlog add */
718 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
719 {
720 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
721 	skb_dst_force(skb);
722 
723 	if (!sk->sk_backlog.tail)
724 		sk->sk_backlog.head = skb;
725 	else
726 		sk->sk_backlog.tail->next = skb;
727 
728 	sk->sk_backlog.tail = skb;
729 	skb->next = NULL;
730 }
731 
732 /*
733  * Take into account size of receive queue and backlog queue
734  * Do not take into account this skb truesize,
735  * to allow even a single big packet to come.
736  */
737 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
738 				     unsigned int limit)
739 {
740 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
741 
742 	return qsize > limit;
743 }
744 
745 /* The per-socket spinlock must be held here. */
746 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
747 					      unsigned int limit)
748 {
749 	if (sk_rcvqueues_full(sk, skb, limit))
750 		return -ENOBUFS;
751 
752 	__sk_add_backlog(sk, skb);
753 	sk->sk_backlog.len += skb->truesize;
754 	return 0;
755 }
756 
757 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
758 
759 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
760 {
761 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
762 		return __sk_backlog_rcv(sk, skb);
763 
764 	return sk->sk_backlog_rcv(sk, skb);
765 }
766 
767 static inline void sock_rps_record_flow(const struct sock *sk)
768 {
769 #ifdef CONFIG_RPS
770 	struct rps_sock_flow_table *sock_flow_table;
771 
772 	rcu_read_lock();
773 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
774 	rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
775 	rcu_read_unlock();
776 #endif
777 }
778 
779 static inline void sock_rps_reset_flow(const struct sock *sk)
780 {
781 #ifdef CONFIG_RPS
782 	struct rps_sock_flow_table *sock_flow_table;
783 
784 	rcu_read_lock();
785 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
786 	rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
787 	rcu_read_unlock();
788 #endif
789 }
790 
791 static inline void sock_rps_save_rxhash(struct sock *sk,
792 					const struct sk_buff *skb)
793 {
794 #ifdef CONFIG_RPS
795 	if (unlikely(sk->sk_rxhash != skb->rxhash)) {
796 		sock_rps_reset_flow(sk);
797 		sk->sk_rxhash = skb->rxhash;
798 	}
799 #endif
800 }
801 
802 static inline void sock_rps_reset_rxhash(struct sock *sk)
803 {
804 #ifdef CONFIG_RPS
805 	sock_rps_reset_flow(sk);
806 	sk->sk_rxhash = 0;
807 #endif
808 }
809 
810 #define sk_wait_event(__sk, __timeo, __condition)			\
811 	({	int __rc;						\
812 		release_sock(__sk);					\
813 		__rc = __condition;					\
814 		if (!__rc) {						\
815 			*(__timeo) = schedule_timeout(*(__timeo));	\
816 		}							\
817 		lock_sock(__sk);					\
818 		__rc = __condition;					\
819 		__rc;							\
820 	})
821 
822 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
823 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
824 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
825 extern int sk_stream_error(struct sock *sk, int flags, int err);
826 extern void sk_stream_kill_queues(struct sock *sk);
827 extern void sk_set_memalloc(struct sock *sk);
828 extern void sk_clear_memalloc(struct sock *sk);
829 
830 extern int sk_wait_data(struct sock *sk, long *timeo);
831 
832 struct request_sock_ops;
833 struct timewait_sock_ops;
834 struct inet_hashinfo;
835 struct raw_hashinfo;
836 struct module;
837 
838 /* Networking protocol blocks we attach to sockets.
839  * socket layer -> transport layer interface
840  * transport -> network interface is defined by struct inet_proto
841  */
842 struct proto {
843 	void			(*close)(struct sock *sk,
844 					long timeout);
845 	int			(*connect)(struct sock *sk,
846 					struct sockaddr *uaddr,
847 					int addr_len);
848 	int			(*disconnect)(struct sock *sk, int flags);
849 
850 	struct sock *		(*accept)(struct sock *sk, int flags, int *err);
851 
852 	int			(*ioctl)(struct sock *sk, int cmd,
853 					 unsigned long arg);
854 	int			(*init)(struct sock *sk);
855 	void			(*destroy)(struct sock *sk);
856 	void			(*shutdown)(struct sock *sk, int how);
857 	int			(*setsockopt)(struct sock *sk, int level,
858 					int optname, char __user *optval,
859 					unsigned int optlen);
860 	int			(*getsockopt)(struct sock *sk, int level,
861 					int optname, char __user *optval,
862 					int __user *option);
863 #ifdef CONFIG_COMPAT
864 	int			(*compat_setsockopt)(struct sock *sk,
865 					int level,
866 					int optname, char __user *optval,
867 					unsigned int optlen);
868 	int			(*compat_getsockopt)(struct sock *sk,
869 					int level,
870 					int optname, char __user *optval,
871 					int __user *option);
872 	int			(*compat_ioctl)(struct sock *sk,
873 					unsigned int cmd, unsigned long arg);
874 #endif
875 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
876 					   struct msghdr *msg, size_t len);
877 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
878 					   struct msghdr *msg,
879 					   size_t len, int noblock, int flags,
880 					   int *addr_len);
881 	int			(*sendpage)(struct sock *sk, struct page *page,
882 					int offset, size_t size, int flags);
883 	int			(*bind)(struct sock *sk,
884 					struct sockaddr *uaddr, int addr_len);
885 
886 	int			(*backlog_rcv) (struct sock *sk,
887 						struct sk_buff *skb);
888 
889 	void		(*release_cb)(struct sock *sk);
890 	void		(*mtu_reduced)(struct sock *sk);
891 
892 	/* Keeping track of sk's, looking them up, and port selection methods. */
893 	void			(*hash)(struct sock *sk);
894 	void			(*unhash)(struct sock *sk);
895 	void			(*rehash)(struct sock *sk);
896 	int			(*get_port)(struct sock *sk, unsigned short snum);
897 	void			(*clear_sk)(struct sock *sk, int size);
898 
899 	/* Keeping track of sockets in use */
900 #ifdef CONFIG_PROC_FS
901 	unsigned int		inuse_idx;
902 #endif
903 
904 	/* Memory pressure */
905 	void			(*enter_memory_pressure)(struct sock *sk);
906 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
907 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
908 	/*
909 	 * Pressure flag: try to collapse.
910 	 * Technical note: it is used by multiple contexts non atomically.
911 	 * All the __sk_mem_schedule() is of this nature: accounting
912 	 * is strict, actions are advisory and have some latency.
913 	 */
914 	int			*memory_pressure;
915 	long			*sysctl_mem;
916 	int			*sysctl_wmem;
917 	int			*sysctl_rmem;
918 	int			max_header;
919 	bool			no_autobind;
920 
921 	struct kmem_cache	*slab;
922 	unsigned int		obj_size;
923 	int			slab_flags;
924 
925 	struct percpu_counter	*orphan_count;
926 
927 	struct request_sock_ops	*rsk_prot;
928 	struct timewait_sock_ops *twsk_prot;
929 
930 	union {
931 		struct inet_hashinfo	*hashinfo;
932 		struct udp_table	*udp_table;
933 		struct raw_hashinfo	*raw_hash;
934 	} h;
935 
936 	struct module		*owner;
937 
938 	char			name[32];
939 
940 	struct list_head	node;
941 #ifdef SOCK_REFCNT_DEBUG
942 	atomic_t		socks;
943 #endif
944 #ifdef CONFIG_MEMCG_KMEM
945 	/*
946 	 * cgroup specific init/deinit functions. Called once for all
947 	 * protocols that implement it, from cgroups populate function.
948 	 * This function has to setup any files the protocol want to
949 	 * appear in the kmem cgroup filesystem.
950 	 */
951 	int			(*init_cgroup)(struct mem_cgroup *memcg,
952 					       struct cgroup_subsys *ss);
953 	void			(*destroy_cgroup)(struct mem_cgroup *memcg);
954 	struct cg_proto		*(*proto_cgroup)(struct mem_cgroup *memcg);
955 #endif
956 };
957 
958 /*
959  * Bits in struct cg_proto.flags
960  */
961 enum cg_proto_flags {
962 	/* Currently active and new sockets should be assigned to cgroups */
963 	MEMCG_SOCK_ACTIVE,
964 	/* It was ever activated; we must disarm static keys on destruction */
965 	MEMCG_SOCK_ACTIVATED,
966 };
967 
968 struct cg_proto {
969 	void			(*enter_memory_pressure)(struct sock *sk);
970 	struct res_counter	*memory_allocated;	/* Current allocated memory. */
971 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
972 	int			*memory_pressure;
973 	long			*sysctl_mem;
974 	unsigned long		flags;
975 	/*
976 	 * memcg field is used to find which memcg we belong directly
977 	 * Each memcg struct can hold more than one cg_proto, so container_of
978 	 * won't really cut.
979 	 *
980 	 * The elegant solution would be having an inverse function to
981 	 * proto_cgroup in struct proto, but that means polluting the structure
982 	 * for everybody, instead of just for memcg users.
983 	 */
984 	struct mem_cgroup	*memcg;
985 };
986 
987 extern int proto_register(struct proto *prot, int alloc_slab);
988 extern void proto_unregister(struct proto *prot);
989 
990 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
991 {
992 	return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
993 }
994 
995 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
996 {
997 	return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
998 }
999 
1000 #ifdef SOCK_REFCNT_DEBUG
1001 static inline void sk_refcnt_debug_inc(struct sock *sk)
1002 {
1003 	atomic_inc(&sk->sk_prot->socks);
1004 }
1005 
1006 static inline void sk_refcnt_debug_dec(struct sock *sk)
1007 {
1008 	atomic_dec(&sk->sk_prot->socks);
1009 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1010 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1011 }
1012 
1013 inline void sk_refcnt_debug_release(const struct sock *sk)
1014 {
1015 	if (atomic_read(&sk->sk_refcnt) != 1)
1016 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1017 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1018 }
1019 #else /* SOCK_REFCNT_DEBUG */
1020 #define sk_refcnt_debug_inc(sk) do { } while (0)
1021 #define sk_refcnt_debug_dec(sk) do { } while (0)
1022 #define sk_refcnt_debug_release(sk) do { } while (0)
1023 #endif /* SOCK_REFCNT_DEBUG */
1024 
1025 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1026 extern struct static_key memcg_socket_limit_enabled;
1027 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1028 					       struct cg_proto *cg_proto)
1029 {
1030 	return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1031 }
1032 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1033 #else
1034 #define mem_cgroup_sockets_enabled 0
1035 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1036 					       struct cg_proto *cg_proto)
1037 {
1038 	return NULL;
1039 }
1040 #endif
1041 
1042 
1043 static inline bool sk_has_memory_pressure(const struct sock *sk)
1044 {
1045 	return sk->sk_prot->memory_pressure != NULL;
1046 }
1047 
1048 static inline bool sk_under_memory_pressure(const struct sock *sk)
1049 {
1050 	if (!sk->sk_prot->memory_pressure)
1051 		return false;
1052 
1053 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1054 		return !!*sk->sk_cgrp->memory_pressure;
1055 
1056 	return !!*sk->sk_prot->memory_pressure;
1057 }
1058 
1059 static inline void sk_leave_memory_pressure(struct sock *sk)
1060 {
1061 	int *memory_pressure = sk->sk_prot->memory_pressure;
1062 
1063 	if (!memory_pressure)
1064 		return;
1065 
1066 	if (*memory_pressure)
1067 		*memory_pressure = 0;
1068 
1069 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1070 		struct cg_proto *cg_proto = sk->sk_cgrp;
1071 		struct proto *prot = sk->sk_prot;
1072 
1073 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1074 			if (*cg_proto->memory_pressure)
1075 				*cg_proto->memory_pressure = 0;
1076 	}
1077 
1078 }
1079 
1080 static inline void sk_enter_memory_pressure(struct sock *sk)
1081 {
1082 	if (!sk->sk_prot->enter_memory_pressure)
1083 		return;
1084 
1085 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1086 		struct cg_proto *cg_proto = sk->sk_cgrp;
1087 		struct proto *prot = sk->sk_prot;
1088 
1089 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1090 			cg_proto->enter_memory_pressure(sk);
1091 	}
1092 
1093 	sk->sk_prot->enter_memory_pressure(sk);
1094 }
1095 
1096 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1097 {
1098 	long *prot = sk->sk_prot->sysctl_mem;
1099 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1100 		prot = sk->sk_cgrp->sysctl_mem;
1101 	return prot[index];
1102 }
1103 
1104 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1105 					      unsigned long amt,
1106 					      int *parent_status)
1107 {
1108 	struct res_counter *fail;
1109 	int ret;
1110 
1111 	ret = res_counter_charge_nofail(prot->memory_allocated,
1112 					amt << PAGE_SHIFT, &fail);
1113 	if (ret < 0)
1114 		*parent_status = OVER_LIMIT;
1115 }
1116 
1117 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1118 					      unsigned long amt)
1119 {
1120 	res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1121 }
1122 
1123 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1124 {
1125 	u64 ret;
1126 	ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1127 	return ret >> PAGE_SHIFT;
1128 }
1129 
1130 static inline long
1131 sk_memory_allocated(const struct sock *sk)
1132 {
1133 	struct proto *prot = sk->sk_prot;
1134 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1135 		return memcg_memory_allocated_read(sk->sk_cgrp);
1136 
1137 	return atomic_long_read(prot->memory_allocated);
1138 }
1139 
1140 static inline long
1141 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1142 {
1143 	struct proto *prot = sk->sk_prot;
1144 
1145 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1146 		memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1147 		/* update the root cgroup regardless */
1148 		atomic_long_add_return(amt, prot->memory_allocated);
1149 		return memcg_memory_allocated_read(sk->sk_cgrp);
1150 	}
1151 
1152 	return atomic_long_add_return(amt, prot->memory_allocated);
1153 }
1154 
1155 static inline void
1156 sk_memory_allocated_sub(struct sock *sk, int amt)
1157 {
1158 	struct proto *prot = sk->sk_prot;
1159 
1160 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1161 		memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1162 
1163 	atomic_long_sub(amt, prot->memory_allocated);
1164 }
1165 
1166 static inline void sk_sockets_allocated_dec(struct sock *sk)
1167 {
1168 	struct proto *prot = sk->sk_prot;
1169 
1170 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1171 		struct cg_proto *cg_proto = sk->sk_cgrp;
1172 
1173 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1174 			percpu_counter_dec(cg_proto->sockets_allocated);
1175 	}
1176 
1177 	percpu_counter_dec(prot->sockets_allocated);
1178 }
1179 
1180 static inline void sk_sockets_allocated_inc(struct sock *sk)
1181 {
1182 	struct proto *prot = sk->sk_prot;
1183 
1184 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1185 		struct cg_proto *cg_proto = sk->sk_cgrp;
1186 
1187 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1188 			percpu_counter_inc(cg_proto->sockets_allocated);
1189 	}
1190 
1191 	percpu_counter_inc(prot->sockets_allocated);
1192 }
1193 
1194 static inline int
1195 sk_sockets_allocated_read_positive(struct sock *sk)
1196 {
1197 	struct proto *prot = sk->sk_prot;
1198 
1199 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1200 		return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1201 
1202 	return percpu_counter_read_positive(prot->sockets_allocated);
1203 }
1204 
1205 static inline int
1206 proto_sockets_allocated_sum_positive(struct proto *prot)
1207 {
1208 	return percpu_counter_sum_positive(prot->sockets_allocated);
1209 }
1210 
1211 static inline long
1212 proto_memory_allocated(struct proto *prot)
1213 {
1214 	return atomic_long_read(prot->memory_allocated);
1215 }
1216 
1217 static inline bool
1218 proto_memory_pressure(struct proto *prot)
1219 {
1220 	if (!prot->memory_pressure)
1221 		return false;
1222 	return !!*prot->memory_pressure;
1223 }
1224 
1225 
1226 #ifdef CONFIG_PROC_FS
1227 /* Called with local bh disabled */
1228 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1229 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1230 #else
1231 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1232 		int inc)
1233 {
1234 }
1235 #endif
1236 
1237 
1238 /* With per-bucket locks this operation is not-atomic, so that
1239  * this version is not worse.
1240  */
1241 static inline void __sk_prot_rehash(struct sock *sk)
1242 {
1243 	sk->sk_prot->unhash(sk);
1244 	sk->sk_prot->hash(sk);
1245 }
1246 
1247 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1248 
1249 /* About 10 seconds */
1250 #define SOCK_DESTROY_TIME (10*HZ)
1251 
1252 /* Sockets 0-1023 can't be bound to unless you are superuser */
1253 #define PROT_SOCK	1024
1254 
1255 #define SHUTDOWN_MASK	3
1256 #define RCV_SHUTDOWN	1
1257 #define SEND_SHUTDOWN	2
1258 
1259 #define SOCK_SNDBUF_LOCK	1
1260 #define SOCK_RCVBUF_LOCK	2
1261 #define SOCK_BINDADDR_LOCK	4
1262 #define SOCK_BINDPORT_LOCK	8
1263 
1264 /* sock_iocb: used to kick off async processing of socket ios */
1265 struct sock_iocb {
1266 	struct list_head	list;
1267 
1268 	int			flags;
1269 	int			size;
1270 	struct socket		*sock;
1271 	struct sock		*sk;
1272 	struct scm_cookie	*scm;
1273 	struct msghdr		*msg, async_msg;
1274 	struct kiocb		*kiocb;
1275 };
1276 
1277 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1278 {
1279 	return (struct sock_iocb *)iocb->private;
1280 }
1281 
1282 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1283 {
1284 	return si->kiocb;
1285 }
1286 
1287 struct socket_alloc {
1288 	struct socket socket;
1289 	struct inode vfs_inode;
1290 };
1291 
1292 static inline struct socket *SOCKET_I(struct inode *inode)
1293 {
1294 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1295 }
1296 
1297 static inline struct inode *SOCK_INODE(struct socket *socket)
1298 {
1299 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1300 }
1301 
1302 /*
1303  * Functions for memory accounting
1304  */
1305 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1306 extern void __sk_mem_reclaim(struct sock *sk);
1307 
1308 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1309 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1310 #define SK_MEM_SEND	0
1311 #define SK_MEM_RECV	1
1312 
1313 static inline int sk_mem_pages(int amt)
1314 {
1315 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1316 }
1317 
1318 static inline bool sk_has_account(struct sock *sk)
1319 {
1320 	/* return true if protocol supports memory accounting */
1321 	return !!sk->sk_prot->memory_allocated;
1322 }
1323 
1324 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1325 {
1326 	if (!sk_has_account(sk))
1327 		return true;
1328 	return size <= sk->sk_forward_alloc ||
1329 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1330 }
1331 
1332 static inline bool
1333 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, unsigned int size)
1334 {
1335 	if (!sk_has_account(sk))
1336 		return true;
1337 	return size<= sk->sk_forward_alloc ||
1338 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1339 		skb_pfmemalloc(skb);
1340 }
1341 
1342 static inline void sk_mem_reclaim(struct sock *sk)
1343 {
1344 	if (!sk_has_account(sk))
1345 		return;
1346 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1347 		__sk_mem_reclaim(sk);
1348 }
1349 
1350 static inline void sk_mem_reclaim_partial(struct sock *sk)
1351 {
1352 	if (!sk_has_account(sk))
1353 		return;
1354 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1355 		__sk_mem_reclaim(sk);
1356 }
1357 
1358 static inline void sk_mem_charge(struct sock *sk, int size)
1359 {
1360 	if (!sk_has_account(sk))
1361 		return;
1362 	sk->sk_forward_alloc -= size;
1363 }
1364 
1365 static inline void sk_mem_uncharge(struct sock *sk, int size)
1366 {
1367 	if (!sk_has_account(sk))
1368 		return;
1369 	sk->sk_forward_alloc += size;
1370 }
1371 
1372 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1373 {
1374 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1375 	sk->sk_wmem_queued -= skb->truesize;
1376 	sk_mem_uncharge(sk, skb->truesize);
1377 	__kfree_skb(skb);
1378 }
1379 
1380 /* Used by processes to "lock" a socket state, so that
1381  * interrupts and bottom half handlers won't change it
1382  * from under us. It essentially blocks any incoming
1383  * packets, so that we won't get any new data or any
1384  * packets that change the state of the socket.
1385  *
1386  * While locked, BH processing will add new packets to
1387  * the backlog queue.  This queue is processed by the
1388  * owner of the socket lock right before it is released.
1389  *
1390  * Since ~2.3.5 it is also exclusive sleep lock serializing
1391  * accesses from user process context.
1392  */
1393 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1394 
1395 /*
1396  * Macro so as to not evaluate some arguments when
1397  * lockdep is not enabled.
1398  *
1399  * Mark both the sk_lock and the sk_lock.slock as a
1400  * per-address-family lock class.
1401  */
1402 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1403 do {									\
1404 	sk->sk_lock.owned = 0;						\
1405 	init_waitqueue_head(&sk->sk_lock.wq);				\
1406 	spin_lock_init(&(sk)->sk_lock.slock);				\
1407 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1408 			sizeof((sk)->sk_lock));				\
1409 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1410 				(skey), (sname));				\
1411 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1412 } while (0)
1413 
1414 extern void lock_sock_nested(struct sock *sk, int subclass);
1415 
1416 static inline void lock_sock(struct sock *sk)
1417 {
1418 	lock_sock_nested(sk, 0);
1419 }
1420 
1421 extern void release_sock(struct sock *sk);
1422 
1423 /* BH context may only use the following locking interface. */
1424 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1425 #define bh_lock_sock_nested(__sk) \
1426 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1427 				SINGLE_DEPTH_NESTING)
1428 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1429 
1430 extern bool lock_sock_fast(struct sock *sk);
1431 /**
1432  * unlock_sock_fast - complement of lock_sock_fast
1433  * @sk: socket
1434  * @slow: slow mode
1435  *
1436  * fast unlock socket for user context.
1437  * If slow mode is on, we call regular release_sock()
1438  */
1439 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1440 {
1441 	if (slow)
1442 		release_sock(sk);
1443 	else
1444 		spin_unlock_bh(&sk->sk_lock.slock);
1445 }
1446 
1447 
1448 extern struct sock		*sk_alloc(struct net *net, int family,
1449 					  gfp_t priority,
1450 					  struct proto *prot);
1451 extern void			sk_free(struct sock *sk);
1452 extern void			sk_release_kernel(struct sock *sk);
1453 extern struct sock		*sk_clone_lock(const struct sock *sk,
1454 					       const gfp_t priority);
1455 
1456 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
1457 					      unsigned long size, int force,
1458 					      gfp_t priority);
1459 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
1460 					      unsigned long size, int force,
1461 					      gfp_t priority);
1462 extern void			sock_wfree(struct sk_buff *skb);
1463 extern void			sock_rfree(struct sk_buff *skb);
1464 extern void			sock_edemux(struct sk_buff *skb);
1465 
1466 extern int			sock_setsockopt(struct socket *sock, int level,
1467 						int op, char __user *optval,
1468 						unsigned int optlen);
1469 
1470 extern int			sock_getsockopt(struct socket *sock, int level,
1471 						int op, char __user *optval,
1472 						int __user *optlen);
1473 extern struct sk_buff		*sock_alloc_send_skb(struct sock *sk,
1474 						     unsigned long size,
1475 						     int noblock,
1476 						     int *errcode);
1477 extern struct sk_buff		*sock_alloc_send_pskb(struct sock *sk,
1478 						      unsigned long header_len,
1479 						      unsigned long data_len,
1480 						      int noblock,
1481 						      int *errcode);
1482 extern void *sock_kmalloc(struct sock *sk, int size,
1483 			  gfp_t priority);
1484 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1485 extern void sk_send_sigurg(struct sock *sk);
1486 
1487 #ifdef CONFIG_CGROUPS
1488 extern void sock_update_classid(struct sock *sk);
1489 #else
1490 static inline void sock_update_classid(struct sock *sk)
1491 {
1492 }
1493 #endif
1494 
1495 /*
1496  * Functions to fill in entries in struct proto_ops when a protocol
1497  * does not implement a particular function.
1498  */
1499 extern int                      sock_no_bind(struct socket *,
1500 					     struct sockaddr *, int);
1501 extern int                      sock_no_connect(struct socket *,
1502 						struct sockaddr *, int, int);
1503 extern int                      sock_no_socketpair(struct socket *,
1504 						   struct socket *);
1505 extern int                      sock_no_accept(struct socket *,
1506 					       struct socket *, int);
1507 extern int                      sock_no_getname(struct socket *,
1508 						struct sockaddr *, int *, int);
1509 extern unsigned int             sock_no_poll(struct file *, struct socket *,
1510 					     struct poll_table_struct *);
1511 extern int                      sock_no_ioctl(struct socket *, unsigned int,
1512 					      unsigned long);
1513 extern int			sock_no_listen(struct socket *, int);
1514 extern int                      sock_no_shutdown(struct socket *, int);
1515 extern int			sock_no_getsockopt(struct socket *, int , int,
1516 						   char __user *, int __user *);
1517 extern int			sock_no_setsockopt(struct socket *, int, int,
1518 						   char __user *, unsigned int);
1519 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1520 						struct msghdr *, size_t);
1521 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1522 						struct msghdr *, size_t, int);
1523 extern int			sock_no_mmap(struct file *file,
1524 					     struct socket *sock,
1525 					     struct vm_area_struct *vma);
1526 extern ssize_t			sock_no_sendpage(struct socket *sock,
1527 						struct page *page,
1528 						int offset, size_t size,
1529 						int flags);
1530 
1531 /*
1532  * Functions to fill in entries in struct proto_ops when a protocol
1533  * uses the inet style.
1534  */
1535 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1536 				  char __user *optval, int __user *optlen);
1537 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1538 			       struct msghdr *msg, size_t size, int flags);
1539 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1540 				  char __user *optval, unsigned int optlen);
1541 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1542 		int optname, char __user *optval, int __user *optlen);
1543 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1544 		int optname, char __user *optval, unsigned int optlen);
1545 
1546 extern void sk_common_release(struct sock *sk);
1547 
1548 /*
1549  *	Default socket callbacks and setup code
1550  */
1551 
1552 /* Initialise core socket variables */
1553 extern void sock_init_data(struct socket *sock, struct sock *sk);
1554 
1555 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1556 
1557 /**
1558  *	sk_filter_release - release a socket filter
1559  *	@fp: filter to remove
1560  *
1561  *	Remove a filter from a socket and release its resources.
1562  */
1563 
1564 static inline void sk_filter_release(struct sk_filter *fp)
1565 {
1566 	if (atomic_dec_and_test(&fp->refcnt))
1567 		call_rcu(&fp->rcu, sk_filter_release_rcu);
1568 }
1569 
1570 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1571 {
1572 	unsigned int size = sk_filter_len(fp);
1573 
1574 	atomic_sub(size, &sk->sk_omem_alloc);
1575 	sk_filter_release(fp);
1576 }
1577 
1578 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1579 {
1580 	atomic_inc(&fp->refcnt);
1581 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1582 }
1583 
1584 /*
1585  * Socket reference counting postulates.
1586  *
1587  * * Each user of socket SHOULD hold a reference count.
1588  * * Each access point to socket (an hash table bucket, reference from a list,
1589  *   running timer, skb in flight MUST hold a reference count.
1590  * * When reference count hits 0, it means it will never increase back.
1591  * * When reference count hits 0, it means that no references from
1592  *   outside exist to this socket and current process on current CPU
1593  *   is last user and may/should destroy this socket.
1594  * * sk_free is called from any context: process, BH, IRQ. When
1595  *   it is called, socket has no references from outside -> sk_free
1596  *   may release descendant resources allocated by the socket, but
1597  *   to the time when it is called, socket is NOT referenced by any
1598  *   hash tables, lists etc.
1599  * * Packets, delivered from outside (from network or from another process)
1600  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1601  *   when they sit in queue. Otherwise, packets will leak to hole, when
1602  *   socket is looked up by one cpu and unhasing is made by another CPU.
1603  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1604  *   (leak to backlog). Packet socket does all the processing inside
1605  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1606  *   use separate SMP lock, so that they are prone too.
1607  */
1608 
1609 /* Ungrab socket and destroy it, if it was the last reference. */
1610 static inline void sock_put(struct sock *sk)
1611 {
1612 	if (atomic_dec_and_test(&sk->sk_refcnt))
1613 		sk_free(sk);
1614 }
1615 
1616 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1617 			  const int nested);
1618 
1619 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1620 {
1621 	sk->sk_tx_queue_mapping = tx_queue;
1622 }
1623 
1624 static inline void sk_tx_queue_clear(struct sock *sk)
1625 {
1626 	sk->sk_tx_queue_mapping = -1;
1627 }
1628 
1629 static inline int sk_tx_queue_get(const struct sock *sk)
1630 {
1631 	return sk ? sk->sk_tx_queue_mapping : -1;
1632 }
1633 
1634 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1635 {
1636 	sk_tx_queue_clear(sk);
1637 	sk->sk_socket = sock;
1638 }
1639 
1640 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1641 {
1642 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1643 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1644 }
1645 /* Detach socket from process context.
1646  * Announce socket dead, detach it from wait queue and inode.
1647  * Note that parent inode held reference count on this struct sock,
1648  * we do not release it in this function, because protocol
1649  * probably wants some additional cleanups or even continuing
1650  * to work with this socket (TCP).
1651  */
1652 static inline void sock_orphan(struct sock *sk)
1653 {
1654 	write_lock_bh(&sk->sk_callback_lock);
1655 	sock_set_flag(sk, SOCK_DEAD);
1656 	sk_set_socket(sk, NULL);
1657 	sk->sk_wq  = NULL;
1658 	write_unlock_bh(&sk->sk_callback_lock);
1659 }
1660 
1661 static inline void sock_graft(struct sock *sk, struct socket *parent)
1662 {
1663 	write_lock_bh(&sk->sk_callback_lock);
1664 	sk->sk_wq = parent->wq;
1665 	parent->sk = sk;
1666 	sk_set_socket(sk, parent);
1667 	security_sock_graft(sk, parent);
1668 	write_unlock_bh(&sk->sk_callback_lock);
1669 }
1670 
1671 extern int sock_i_uid(struct sock *sk);
1672 extern unsigned long sock_i_ino(struct sock *sk);
1673 
1674 static inline struct dst_entry *
1675 __sk_dst_get(struct sock *sk)
1676 {
1677 	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1678 						       lockdep_is_held(&sk->sk_lock.slock));
1679 }
1680 
1681 static inline struct dst_entry *
1682 sk_dst_get(struct sock *sk)
1683 {
1684 	struct dst_entry *dst;
1685 
1686 	rcu_read_lock();
1687 	dst = rcu_dereference(sk->sk_dst_cache);
1688 	if (dst)
1689 		dst_hold(dst);
1690 	rcu_read_unlock();
1691 	return dst;
1692 }
1693 
1694 extern void sk_reset_txq(struct sock *sk);
1695 
1696 static inline void dst_negative_advice(struct sock *sk)
1697 {
1698 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1699 
1700 	if (dst && dst->ops->negative_advice) {
1701 		ndst = dst->ops->negative_advice(dst);
1702 
1703 		if (ndst != dst) {
1704 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1705 			sk_reset_txq(sk);
1706 		}
1707 	}
1708 }
1709 
1710 static inline void
1711 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1712 {
1713 	struct dst_entry *old_dst;
1714 
1715 	sk_tx_queue_clear(sk);
1716 	/*
1717 	 * This can be called while sk is owned by the caller only,
1718 	 * with no state that can be checked in a rcu_dereference_check() cond
1719 	 */
1720 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1721 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1722 	dst_release(old_dst);
1723 }
1724 
1725 static inline void
1726 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1727 {
1728 	spin_lock(&sk->sk_dst_lock);
1729 	__sk_dst_set(sk, dst);
1730 	spin_unlock(&sk->sk_dst_lock);
1731 }
1732 
1733 static inline void
1734 __sk_dst_reset(struct sock *sk)
1735 {
1736 	__sk_dst_set(sk, NULL);
1737 }
1738 
1739 static inline void
1740 sk_dst_reset(struct sock *sk)
1741 {
1742 	spin_lock(&sk->sk_dst_lock);
1743 	__sk_dst_reset(sk);
1744 	spin_unlock(&sk->sk_dst_lock);
1745 }
1746 
1747 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1748 
1749 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1750 
1751 static inline bool sk_can_gso(const struct sock *sk)
1752 {
1753 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1754 }
1755 
1756 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1757 
1758 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1759 {
1760 	sk->sk_route_nocaps |= flags;
1761 	sk->sk_route_caps &= ~flags;
1762 }
1763 
1764 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1765 					   char __user *from, char *to,
1766 					   int copy, int offset)
1767 {
1768 	if (skb->ip_summed == CHECKSUM_NONE) {
1769 		int err = 0;
1770 		__wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1771 		if (err)
1772 			return err;
1773 		skb->csum = csum_block_add(skb->csum, csum, offset);
1774 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1775 		if (!access_ok(VERIFY_READ, from, copy) ||
1776 		    __copy_from_user_nocache(to, from, copy))
1777 			return -EFAULT;
1778 	} else if (copy_from_user(to, from, copy))
1779 		return -EFAULT;
1780 
1781 	return 0;
1782 }
1783 
1784 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1785 				       char __user *from, int copy)
1786 {
1787 	int err, offset = skb->len;
1788 
1789 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1790 				       copy, offset);
1791 	if (err)
1792 		__skb_trim(skb, offset);
1793 
1794 	return err;
1795 }
1796 
1797 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1798 					   struct sk_buff *skb,
1799 					   struct page *page,
1800 					   int off, int copy)
1801 {
1802 	int err;
1803 
1804 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1805 				       copy, skb->len);
1806 	if (err)
1807 		return err;
1808 
1809 	skb->len	     += copy;
1810 	skb->data_len	     += copy;
1811 	skb->truesize	     += copy;
1812 	sk->sk_wmem_queued   += copy;
1813 	sk_mem_charge(sk, copy);
1814 	return 0;
1815 }
1816 
1817 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1818 				   struct sk_buff *skb, struct page *page,
1819 				   int off, int copy)
1820 {
1821 	if (skb->ip_summed == CHECKSUM_NONE) {
1822 		int err = 0;
1823 		__wsum csum = csum_and_copy_from_user(from,
1824 						     page_address(page) + off,
1825 							    copy, 0, &err);
1826 		if (err)
1827 			return err;
1828 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1829 	} else if (copy_from_user(page_address(page) + off, from, copy))
1830 		return -EFAULT;
1831 
1832 	skb->len	     += copy;
1833 	skb->data_len	     += copy;
1834 	skb->truesize	     += copy;
1835 	sk->sk_wmem_queued   += copy;
1836 	sk_mem_charge(sk, copy);
1837 	return 0;
1838 }
1839 
1840 /**
1841  * sk_wmem_alloc_get - returns write allocations
1842  * @sk: socket
1843  *
1844  * Returns sk_wmem_alloc minus initial offset of one
1845  */
1846 static inline int sk_wmem_alloc_get(const struct sock *sk)
1847 {
1848 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1849 }
1850 
1851 /**
1852  * sk_rmem_alloc_get - returns read allocations
1853  * @sk: socket
1854  *
1855  * Returns sk_rmem_alloc
1856  */
1857 static inline int sk_rmem_alloc_get(const struct sock *sk)
1858 {
1859 	return atomic_read(&sk->sk_rmem_alloc);
1860 }
1861 
1862 /**
1863  * sk_has_allocations - check if allocations are outstanding
1864  * @sk: socket
1865  *
1866  * Returns true if socket has write or read allocations
1867  */
1868 static inline bool sk_has_allocations(const struct sock *sk)
1869 {
1870 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1871 }
1872 
1873 /**
1874  * wq_has_sleeper - check if there are any waiting processes
1875  * @wq: struct socket_wq
1876  *
1877  * Returns true if socket_wq has waiting processes
1878  *
1879  * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1880  * barrier call. They were added due to the race found within the tcp code.
1881  *
1882  * Consider following tcp code paths:
1883  *
1884  * CPU1                  CPU2
1885  *
1886  * sys_select            receive packet
1887  *   ...                 ...
1888  *   __add_wait_queue    update tp->rcv_nxt
1889  *   ...                 ...
1890  *   tp->rcv_nxt check   sock_def_readable
1891  *   ...                 {
1892  *   schedule               rcu_read_lock();
1893  *                          wq = rcu_dereference(sk->sk_wq);
1894  *                          if (wq && waitqueue_active(&wq->wait))
1895  *                              wake_up_interruptible(&wq->wait)
1896  *                          ...
1897  *                       }
1898  *
1899  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1900  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1901  * could then endup calling schedule and sleep forever if there are no more
1902  * data on the socket.
1903  *
1904  */
1905 static inline bool wq_has_sleeper(struct socket_wq *wq)
1906 {
1907 	/* We need to be sure we are in sync with the
1908 	 * add_wait_queue modifications to the wait queue.
1909 	 *
1910 	 * This memory barrier is paired in the sock_poll_wait.
1911 	 */
1912 	smp_mb();
1913 	return wq && waitqueue_active(&wq->wait);
1914 }
1915 
1916 /**
1917  * sock_poll_wait - place memory barrier behind the poll_wait call.
1918  * @filp:           file
1919  * @wait_address:   socket wait queue
1920  * @p:              poll_table
1921  *
1922  * See the comments in the wq_has_sleeper function.
1923  */
1924 static inline void sock_poll_wait(struct file *filp,
1925 		wait_queue_head_t *wait_address, poll_table *p)
1926 {
1927 	if (!poll_does_not_wait(p) && wait_address) {
1928 		poll_wait(filp, wait_address, p);
1929 		/* We need to be sure we are in sync with the
1930 		 * socket flags modification.
1931 		 *
1932 		 * This memory barrier is paired in the wq_has_sleeper.
1933 		 */
1934 		smp_mb();
1935 	}
1936 }
1937 
1938 /*
1939  *	Queue a received datagram if it will fit. Stream and sequenced
1940  *	protocols can't normally use this as they need to fit buffers in
1941  *	and play with them.
1942  *
1943  *	Inlined as it's very short and called for pretty much every
1944  *	packet ever received.
1945  */
1946 
1947 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1948 {
1949 	skb_orphan(skb);
1950 	skb->sk = sk;
1951 	skb->destructor = sock_wfree;
1952 	/*
1953 	 * We used to take a refcount on sk, but following operation
1954 	 * is enough to guarantee sk_free() wont free this sock until
1955 	 * all in-flight packets are completed
1956 	 */
1957 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1958 }
1959 
1960 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1961 {
1962 	skb_orphan(skb);
1963 	skb->sk = sk;
1964 	skb->destructor = sock_rfree;
1965 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1966 	sk_mem_charge(sk, skb->truesize);
1967 }
1968 
1969 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1970 			   unsigned long expires);
1971 
1972 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1973 
1974 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1975 
1976 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1977 
1978 /*
1979  *	Recover an error report and clear atomically
1980  */
1981 
1982 static inline int sock_error(struct sock *sk)
1983 {
1984 	int err;
1985 	if (likely(!sk->sk_err))
1986 		return 0;
1987 	err = xchg(&sk->sk_err, 0);
1988 	return -err;
1989 }
1990 
1991 static inline unsigned long sock_wspace(struct sock *sk)
1992 {
1993 	int amt = 0;
1994 
1995 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1996 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1997 		if (amt < 0)
1998 			amt = 0;
1999 	}
2000 	return amt;
2001 }
2002 
2003 static inline void sk_wake_async(struct sock *sk, int how, int band)
2004 {
2005 	if (sock_flag(sk, SOCK_FASYNC))
2006 		sock_wake_async(sk->sk_socket, how, band);
2007 }
2008 
2009 #define SOCK_MIN_SNDBUF 2048
2010 /*
2011  * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
2012  * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
2013  */
2014 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
2015 
2016 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2017 {
2018 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2019 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2020 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2021 	}
2022 }
2023 
2024 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2025 
2026 static inline struct page *sk_stream_alloc_page(struct sock *sk)
2027 {
2028 	struct page *page = NULL;
2029 
2030 	page = alloc_pages(sk->sk_allocation, 0);
2031 	if (!page) {
2032 		sk_enter_memory_pressure(sk);
2033 		sk_stream_moderate_sndbuf(sk);
2034 	}
2035 	return page;
2036 }
2037 
2038 /*
2039  *	Default write policy as shown to user space via poll/select/SIGIO
2040  */
2041 static inline bool sock_writeable(const struct sock *sk)
2042 {
2043 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2044 }
2045 
2046 static inline gfp_t gfp_any(void)
2047 {
2048 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2049 }
2050 
2051 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2052 {
2053 	return noblock ? 0 : sk->sk_rcvtimeo;
2054 }
2055 
2056 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2057 {
2058 	return noblock ? 0 : sk->sk_sndtimeo;
2059 }
2060 
2061 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2062 {
2063 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2064 }
2065 
2066 /* Alas, with timeout socket operations are not restartable.
2067  * Compare this to poll().
2068  */
2069 static inline int sock_intr_errno(long timeo)
2070 {
2071 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2072 }
2073 
2074 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2075 	struct sk_buff *skb);
2076 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2077 	struct sk_buff *skb);
2078 
2079 static inline void
2080 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2081 {
2082 	ktime_t kt = skb->tstamp;
2083 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2084 
2085 	/*
2086 	 * generate control messages if
2087 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2088 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
2089 	 * - software time stamp available and wanted
2090 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
2091 	 * - hardware time stamps available and wanted
2092 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
2093 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
2094 	 */
2095 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2096 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2097 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2098 	    (hwtstamps->hwtstamp.tv64 &&
2099 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2100 	    (hwtstamps->syststamp.tv64 &&
2101 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2102 		__sock_recv_timestamp(msg, sk, skb);
2103 	else
2104 		sk->sk_stamp = kt;
2105 
2106 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2107 		__sock_recv_wifi_status(msg, sk, skb);
2108 }
2109 
2110 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2111 				     struct sk_buff *skb);
2112 
2113 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2114 					  struct sk_buff *skb)
2115 {
2116 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2117 			   (1UL << SOCK_RCVTSTAMP)			| \
2118 			   (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)	| \
2119 			   (1UL << SOCK_TIMESTAMPING_SOFTWARE)		| \
2120 			   (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE)	| \
2121 			   (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2122 
2123 	if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2124 		__sock_recv_ts_and_drops(msg, sk, skb);
2125 	else
2126 		sk->sk_stamp = skb->tstamp;
2127 }
2128 
2129 /**
2130  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2131  * @sk:		socket sending this packet
2132  * @tx_flags:	filled with instructions for time stamping
2133  *
2134  * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2135  * parameters are invalid.
2136  */
2137 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2138 
2139 /**
2140  * sk_eat_skb - Release a skb if it is no longer needed
2141  * @sk: socket to eat this skb from
2142  * @skb: socket buffer to eat
2143  * @copied_early: flag indicating whether DMA operations copied this data early
2144  *
2145  * This routine must be called with interrupts disabled or with the socket
2146  * locked so that the sk_buff queue operation is ok.
2147 */
2148 #ifdef CONFIG_NET_DMA
2149 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2150 {
2151 	__skb_unlink(skb, &sk->sk_receive_queue);
2152 	if (!copied_early)
2153 		__kfree_skb(skb);
2154 	else
2155 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
2156 }
2157 #else
2158 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2159 {
2160 	__skb_unlink(skb, &sk->sk_receive_queue);
2161 	__kfree_skb(skb);
2162 }
2163 #endif
2164 
2165 static inline
2166 struct net *sock_net(const struct sock *sk)
2167 {
2168 	return read_pnet(&sk->sk_net);
2169 }
2170 
2171 static inline
2172 void sock_net_set(struct sock *sk, struct net *net)
2173 {
2174 	write_pnet(&sk->sk_net, net);
2175 }
2176 
2177 /*
2178  * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2179  * They should not hold a reference to a namespace in order to allow
2180  * to stop it.
2181  * Sockets after sk_change_net should be released using sk_release_kernel
2182  */
2183 static inline void sk_change_net(struct sock *sk, struct net *net)
2184 {
2185 	put_net(sock_net(sk));
2186 	sock_net_set(sk, hold_net(net));
2187 }
2188 
2189 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2190 {
2191 	if (skb->sk) {
2192 		struct sock *sk = skb->sk;
2193 
2194 		skb->destructor = NULL;
2195 		skb->sk = NULL;
2196 		return sk;
2197 	}
2198 	return NULL;
2199 }
2200 
2201 extern void sock_enable_timestamp(struct sock *sk, int flag);
2202 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2203 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2204 
2205 /*
2206  *	Enable debug/info messages
2207  */
2208 extern int net_msg_warn;
2209 #define NETDEBUG(fmt, args...) \
2210 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
2211 
2212 #define LIMIT_NETDEBUG(fmt, args...) \
2213 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2214 
2215 extern __u32 sysctl_wmem_max;
2216 extern __u32 sysctl_rmem_max;
2217 
2218 extern void sk_init(void);
2219 
2220 extern int sysctl_optmem_max;
2221 
2222 extern __u32 sysctl_wmem_default;
2223 extern __u32 sysctl_rmem_default;
2224 
2225 #endif	/* _SOCK_H */
2226