xref: /linux/include/net/sock.h (revision a0d3c7c5c07cfbe00ab89438ddf82482f5a99422)
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/page_counter.h>
58 #include <linux/memcontrol.h>
59 #include <linux/static_key.h>
60 #include <linux/sched.h>
61 #include <linux/wait.h>
62 #include <linux/cgroup-defs.h>
63 
64 #include <linux/filter.h>
65 #include <linux/rculist_nulls.h>
66 #include <linux/poll.h>
67 
68 #include <linux/atomic.h>
69 #include <net/dst.h>
70 #include <net/checksum.h>
71 #include <net/tcp_states.h>
72 #include <linux/net_tstamp.h>
73 
74 /*
75  * This structure really needs to be cleaned up.
76  * Most of it is for TCP, and not used by any of
77  * the other protocols.
78  */
79 
80 /* Define this to get the SOCK_DBG debugging facility. */
81 #define SOCK_DEBUGGING
82 #ifdef SOCK_DEBUGGING
83 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
84 					printk(KERN_DEBUG msg); } while (0)
85 #else
86 /* Validate arguments and do nothing */
87 static inline __printf(2, 3)
88 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
89 {
90 }
91 #endif
92 
93 /* This is the per-socket lock.  The spinlock provides a synchronization
94  * between user contexts and software interrupt processing, whereas the
95  * mini-semaphore synchronizes multiple users amongst themselves.
96  */
97 typedef struct {
98 	spinlock_t		slock;
99 	int			owned;
100 	wait_queue_head_t	wq;
101 	/*
102 	 * We express the mutex-alike socket_lock semantics
103 	 * to the lock validator by explicitly managing
104 	 * the slock as a lock variant (in addition to
105 	 * the slock itself):
106 	 */
107 #ifdef CONFIG_DEBUG_LOCK_ALLOC
108 	struct lockdep_map dep_map;
109 #endif
110 } socket_lock_t;
111 
112 struct sock;
113 struct proto;
114 struct net;
115 
116 typedef __u32 __bitwise __portpair;
117 typedef __u64 __bitwise __addrpair;
118 
119 /**
120  *	struct sock_common - minimal network layer representation of sockets
121  *	@skc_daddr: Foreign IPv4 addr
122  *	@skc_rcv_saddr: Bound local IPv4 addr
123  *	@skc_hash: hash value used with various protocol lookup tables
124  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
125  *	@skc_dport: placeholder for inet_dport/tw_dport
126  *	@skc_num: placeholder for inet_num/tw_num
127  *	@skc_family: network address family
128  *	@skc_state: Connection state
129  *	@skc_reuse: %SO_REUSEADDR setting
130  *	@skc_reuseport: %SO_REUSEPORT setting
131  *	@skc_bound_dev_if: bound device index if != 0
132  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
133  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
134  *	@skc_prot: protocol handlers inside a network family
135  *	@skc_net: reference to the network namespace of this socket
136  *	@skc_node: main hash linkage for various protocol lookup tables
137  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
138  *	@skc_tx_queue_mapping: tx queue number for this connection
139  *	@skc_flags: place holder for sk_flags
140  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
141  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
142  *	@skc_incoming_cpu: record/match cpu processing incoming packets
143  *	@skc_refcnt: reference count
144  *
145  *	This is the minimal network layer representation of sockets, the header
146  *	for struct sock and struct inet_timewait_sock.
147  */
148 struct sock_common {
149 	/* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
150 	 * address on 64bit arches : cf INET_MATCH()
151 	 */
152 	union {
153 		__addrpair	skc_addrpair;
154 		struct {
155 			__be32	skc_daddr;
156 			__be32	skc_rcv_saddr;
157 		};
158 	};
159 	union  {
160 		unsigned int	skc_hash;
161 		__u16		skc_u16hashes[2];
162 	};
163 	/* skc_dport && skc_num must be grouped as well */
164 	union {
165 		__portpair	skc_portpair;
166 		struct {
167 			__be16	skc_dport;
168 			__u16	skc_num;
169 		};
170 	};
171 
172 	unsigned short		skc_family;
173 	volatile unsigned char	skc_state;
174 	unsigned char		skc_reuse:4;
175 	unsigned char		skc_reuseport:1;
176 	unsigned char		skc_ipv6only:1;
177 	unsigned char		skc_net_refcnt:1;
178 	int			skc_bound_dev_if;
179 	union {
180 		struct hlist_node	skc_bind_node;
181 		struct hlist_node	skc_portaddr_node;
182 	};
183 	struct proto		*skc_prot;
184 	possible_net_t		skc_net;
185 
186 #if IS_ENABLED(CONFIG_IPV6)
187 	struct in6_addr		skc_v6_daddr;
188 	struct in6_addr		skc_v6_rcv_saddr;
189 #endif
190 
191 	atomic64_t		skc_cookie;
192 
193 	/* following fields are padding to force
194 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
195 	 * assuming IPV6 is enabled. We use this padding differently
196 	 * for different kind of 'sockets'
197 	 */
198 	union {
199 		unsigned long	skc_flags;
200 		struct sock	*skc_listener; /* request_sock */
201 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
202 	};
203 	/*
204 	 * fields between dontcopy_begin/dontcopy_end
205 	 * are not copied in sock_copy()
206 	 */
207 	/* private: */
208 	int			skc_dontcopy_begin[0];
209 	/* public: */
210 	union {
211 		struct hlist_node	skc_node;
212 		struct hlist_nulls_node skc_nulls_node;
213 	};
214 	int			skc_tx_queue_mapping;
215 	union {
216 		int		skc_incoming_cpu;
217 		u32		skc_rcv_wnd;
218 		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
219 	};
220 
221 	atomic_t		skc_refcnt;
222 	/* private: */
223 	int                     skc_dontcopy_end[0];
224 	union {
225 		u32		skc_rxhash;
226 		u32		skc_window_clamp;
227 		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
228 	};
229 	/* public: */
230 };
231 
232 /**
233   *	struct sock - network layer representation of sockets
234   *	@__sk_common: shared layout with inet_timewait_sock
235   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
236   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
237   *	@sk_lock:	synchronizer
238   *	@sk_rcvbuf: size of receive buffer in bytes
239   *	@sk_wq: sock wait queue and async head
240   *	@sk_rx_dst: receive input route used by early demux
241   *	@sk_dst_cache: destination cache
242   *	@sk_policy: flow policy
243   *	@sk_receive_queue: incoming packets
244   *	@sk_wmem_alloc: transmit queue bytes committed
245   *	@sk_write_queue: Packet sending queue
246   *	@sk_omem_alloc: "o" is "option" or "other"
247   *	@sk_wmem_queued: persistent queue size
248   *	@sk_forward_alloc: space allocated forward
249   *	@sk_napi_id: id of the last napi context to receive data for sk
250   *	@sk_ll_usec: usecs to busypoll when there is no data
251   *	@sk_allocation: allocation mode
252   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
253   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
254   *	@sk_sndbuf: size of send buffer in bytes
255   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
256   *	@sk_no_check_rx: allow zero checksum in RX packets
257   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
258   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
259   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
260   *	@sk_gso_max_size: Maximum GSO segment size to build
261   *	@sk_gso_max_segs: Maximum number of GSO segments
262   *	@sk_lingertime: %SO_LINGER l_linger setting
263   *	@sk_backlog: always used with the per-socket spinlock held
264   *	@sk_callback_lock: used with the callbacks in the end of this struct
265   *	@sk_error_queue: rarely used
266   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
267   *			  IPV6_ADDRFORM for instance)
268   *	@sk_err: last error
269   *	@sk_err_soft: errors that don't cause failure but are the cause of a
270   *		      persistent failure not just 'timed out'
271   *	@sk_drops: raw/udp drops counter
272   *	@sk_ack_backlog: current listen backlog
273   *	@sk_max_ack_backlog: listen backlog set in listen()
274   *	@sk_priority: %SO_PRIORITY setting
275   *	@sk_type: socket type (%SOCK_STREAM, etc)
276   *	@sk_protocol: which protocol this socket belongs in this network family
277   *	@sk_peer_pid: &struct pid for this socket's peer
278   *	@sk_peer_cred: %SO_PEERCRED setting
279   *	@sk_rcvlowat: %SO_RCVLOWAT setting
280   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
281   *	@sk_sndtimeo: %SO_SNDTIMEO setting
282   *	@sk_txhash: computed flow hash for use on transmit
283   *	@sk_filter: socket filtering instructions
284   *	@sk_timer: sock cleanup timer
285   *	@sk_stamp: time stamp of last packet received
286   *	@sk_tsflags: SO_TIMESTAMPING socket options
287   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
288   *	@sk_socket: Identd and reporting IO signals
289   *	@sk_user_data: RPC layer private data
290   *	@sk_frag: cached page frag
291   *	@sk_peek_off: current peek_offset value
292   *	@sk_send_head: front of stuff to transmit
293   *	@sk_security: used by security modules
294   *	@sk_mark: generic packet mark
295   *	@sk_cgrp_data: cgroup data for this cgroup
296   *	@sk_memcg: this socket's memory cgroup association
297   *	@sk_write_pending: a write to stream socket waits to start
298   *	@sk_state_change: callback to indicate change in the state of the sock
299   *	@sk_data_ready: callback to indicate there is data to be processed
300   *	@sk_write_space: callback to indicate there is bf sending space available
301   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
302   *	@sk_backlog_rcv: callback to process the backlog
303   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
304   *	@sk_reuseport_cb: reuseport group container
305  */
306 struct sock {
307 	/*
308 	 * Now struct inet_timewait_sock also uses sock_common, so please just
309 	 * don't add nothing before this first member (__sk_common) --acme
310 	 */
311 	struct sock_common	__sk_common;
312 #define sk_node			__sk_common.skc_node
313 #define sk_nulls_node		__sk_common.skc_nulls_node
314 #define sk_refcnt		__sk_common.skc_refcnt
315 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
316 
317 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
318 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
319 #define sk_hash			__sk_common.skc_hash
320 #define sk_portpair		__sk_common.skc_portpair
321 #define sk_num			__sk_common.skc_num
322 #define sk_dport		__sk_common.skc_dport
323 #define sk_addrpair		__sk_common.skc_addrpair
324 #define sk_daddr		__sk_common.skc_daddr
325 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
326 #define sk_family		__sk_common.skc_family
327 #define sk_state		__sk_common.skc_state
328 #define sk_reuse		__sk_common.skc_reuse
329 #define sk_reuseport		__sk_common.skc_reuseport
330 #define sk_ipv6only		__sk_common.skc_ipv6only
331 #define sk_net_refcnt		__sk_common.skc_net_refcnt
332 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
333 #define sk_bind_node		__sk_common.skc_bind_node
334 #define sk_prot			__sk_common.skc_prot
335 #define sk_net			__sk_common.skc_net
336 #define sk_v6_daddr		__sk_common.skc_v6_daddr
337 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
338 #define sk_cookie		__sk_common.skc_cookie
339 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
340 #define sk_flags		__sk_common.skc_flags
341 #define sk_rxhash		__sk_common.skc_rxhash
342 
343 	socket_lock_t		sk_lock;
344 	struct sk_buff_head	sk_receive_queue;
345 	/*
346 	 * The backlog queue is special, it is always used with
347 	 * the per-socket spinlock held and requires low latency
348 	 * access. Therefore we special case it's implementation.
349 	 * Note : rmem_alloc is in this structure to fill a hole
350 	 * on 64bit arches, not because its logically part of
351 	 * backlog.
352 	 */
353 	struct {
354 		atomic_t	rmem_alloc;
355 		int		len;
356 		struct sk_buff	*head;
357 		struct sk_buff	*tail;
358 	} sk_backlog;
359 #define sk_rmem_alloc sk_backlog.rmem_alloc
360 	int			sk_forward_alloc;
361 
362 	__u32			sk_txhash;
363 #ifdef CONFIG_NET_RX_BUSY_POLL
364 	unsigned int		sk_napi_id;
365 	unsigned int		sk_ll_usec;
366 #endif
367 	atomic_t		sk_drops;
368 	int			sk_rcvbuf;
369 
370 	struct sk_filter __rcu	*sk_filter;
371 	union {
372 		struct socket_wq __rcu	*sk_wq;
373 		struct socket_wq	*sk_wq_raw;
374 	};
375 #ifdef CONFIG_XFRM
376 	struct xfrm_policy __rcu *sk_policy[2];
377 #endif
378 	struct dst_entry	*sk_rx_dst;
379 	struct dst_entry __rcu	*sk_dst_cache;
380 	/* Note: 32bit hole on 64bit arches */
381 	atomic_t		sk_wmem_alloc;
382 	atomic_t		sk_omem_alloc;
383 	int			sk_sndbuf;
384 	struct sk_buff_head	sk_write_queue;
385 	kmemcheck_bitfield_begin(flags);
386 	unsigned int		sk_shutdown  : 2,
387 				sk_no_check_tx : 1,
388 				sk_no_check_rx : 1,
389 				sk_userlocks : 4,
390 				sk_protocol  : 8,
391 				sk_type      : 16;
392 #define SK_PROTOCOL_MAX U8_MAX
393 	kmemcheck_bitfield_end(flags);
394 	int			sk_wmem_queued;
395 	gfp_t			sk_allocation;
396 	u32			sk_pacing_rate; /* bytes per second */
397 	u32			sk_max_pacing_rate;
398 	netdev_features_t	sk_route_caps;
399 	netdev_features_t	sk_route_nocaps;
400 	int			sk_gso_type;
401 	unsigned int		sk_gso_max_size;
402 	u16			sk_gso_max_segs;
403 	int			sk_rcvlowat;
404 	unsigned long	        sk_lingertime;
405 	struct sk_buff_head	sk_error_queue;
406 	struct proto		*sk_prot_creator;
407 	rwlock_t		sk_callback_lock;
408 	int			sk_err,
409 				sk_err_soft;
410 	u32			sk_ack_backlog;
411 	u32			sk_max_ack_backlog;
412 	__u32			sk_priority;
413 	__u32			sk_mark;
414 	struct pid		*sk_peer_pid;
415 	const struct cred	*sk_peer_cred;
416 	long			sk_rcvtimeo;
417 	long			sk_sndtimeo;
418 	struct timer_list	sk_timer;
419 	ktime_t			sk_stamp;
420 	u16			sk_tsflags;
421 	u32			sk_tskey;
422 	struct socket		*sk_socket;
423 	void			*sk_user_data;
424 	struct page_frag	sk_frag;
425 	struct sk_buff		*sk_send_head;
426 	__s32			sk_peek_off;
427 	int			sk_write_pending;
428 #ifdef CONFIG_SECURITY
429 	void			*sk_security;
430 #endif
431 	struct sock_cgroup_data	sk_cgrp_data;
432 	struct mem_cgroup	*sk_memcg;
433 	void			(*sk_state_change)(struct sock *sk);
434 	void			(*sk_data_ready)(struct sock *sk);
435 	void			(*sk_write_space)(struct sock *sk);
436 	void			(*sk_error_report)(struct sock *sk);
437 	int			(*sk_backlog_rcv)(struct sock *sk,
438 						  struct sk_buff *skb);
439 	void                    (*sk_destruct)(struct sock *sk);
440 	struct sock_reuseport __rcu	*sk_reuseport_cb;
441 	struct rcu_head		sk_rcu;
442 };
443 
444 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
445 
446 #define rcu_dereference_sk_user_data(sk)	rcu_dereference(__sk_user_data((sk)))
447 #define rcu_assign_sk_user_data(sk, ptr)	rcu_assign_pointer(__sk_user_data((sk)), ptr)
448 
449 /*
450  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
451  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
452  * on a socket means that the socket will reuse everybody else's port
453  * without looking at the other's sk_reuse value.
454  */
455 
456 #define SK_NO_REUSE	0
457 #define SK_CAN_REUSE	1
458 #define SK_FORCE_REUSE	2
459 
460 int sk_set_peek_off(struct sock *sk, int val);
461 
462 static inline int sk_peek_offset(struct sock *sk, int flags)
463 {
464 	if (unlikely(flags & MSG_PEEK)) {
465 		s32 off = READ_ONCE(sk->sk_peek_off);
466 		if (off >= 0)
467 			return off;
468 	}
469 
470 	return 0;
471 }
472 
473 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
474 {
475 	s32 off = READ_ONCE(sk->sk_peek_off);
476 
477 	if (unlikely(off >= 0)) {
478 		off = max_t(s32, off - val, 0);
479 		WRITE_ONCE(sk->sk_peek_off, off);
480 	}
481 }
482 
483 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
484 {
485 	sk_peek_offset_bwd(sk, -val);
486 }
487 
488 /*
489  * Hashed lists helper routines
490  */
491 static inline struct sock *sk_entry(const struct hlist_node *node)
492 {
493 	return hlist_entry(node, struct sock, sk_node);
494 }
495 
496 static inline struct sock *__sk_head(const struct hlist_head *head)
497 {
498 	return hlist_entry(head->first, struct sock, sk_node);
499 }
500 
501 static inline struct sock *sk_head(const struct hlist_head *head)
502 {
503 	return hlist_empty(head) ? NULL : __sk_head(head);
504 }
505 
506 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
507 {
508 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
509 }
510 
511 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
512 {
513 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
514 }
515 
516 static inline struct sock *sk_next(const struct sock *sk)
517 {
518 	return sk->sk_node.next ?
519 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
520 }
521 
522 static inline struct sock *sk_nulls_next(const struct sock *sk)
523 {
524 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
525 		hlist_nulls_entry(sk->sk_nulls_node.next,
526 				  struct sock, sk_nulls_node) :
527 		NULL;
528 }
529 
530 static inline bool sk_unhashed(const struct sock *sk)
531 {
532 	return hlist_unhashed(&sk->sk_node);
533 }
534 
535 static inline bool sk_hashed(const struct sock *sk)
536 {
537 	return !sk_unhashed(sk);
538 }
539 
540 static inline void sk_node_init(struct hlist_node *node)
541 {
542 	node->pprev = NULL;
543 }
544 
545 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
546 {
547 	node->pprev = NULL;
548 }
549 
550 static inline void __sk_del_node(struct sock *sk)
551 {
552 	__hlist_del(&sk->sk_node);
553 }
554 
555 /* NB: equivalent to hlist_del_init_rcu */
556 static inline bool __sk_del_node_init(struct sock *sk)
557 {
558 	if (sk_hashed(sk)) {
559 		__sk_del_node(sk);
560 		sk_node_init(&sk->sk_node);
561 		return true;
562 	}
563 	return false;
564 }
565 
566 /* Grab socket reference count. This operation is valid only
567    when sk is ALREADY grabbed f.e. it is found in hash table
568    or a list and the lookup is made under lock preventing hash table
569    modifications.
570  */
571 
572 static __always_inline void sock_hold(struct sock *sk)
573 {
574 	atomic_inc(&sk->sk_refcnt);
575 }
576 
577 /* Ungrab socket in the context, which assumes that socket refcnt
578    cannot hit zero, f.e. it is true in context of any socketcall.
579  */
580 static __always_inline void __sock_put(struct sock *sk)
581 {
582 	atomic_dec(&sk->sk_refcnt);
583 }
584 
585 static inline bool sk_del_node_init(struct sock *sk)
586 {
587 	bool rc = __sk_del_node_init(sk);
588 
589 	if (rc) {
590 		/* paranoid for a while -acme */
591 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
592 		__sock_put(sk);
593 	}
594 	return rc;
595 }
596 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
597 
598 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
599 {
600 	if (sk_hashed(sk)) {
601 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
602 		return true;
603 	}
604 	return false;
605 }
606 
607 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
608 {
609 	bool rc = __sk_nulls_del_node_init_rcu(sk);
610 
611 	if (rc) {
612 		/* paranoid for a while -acme */
613 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
614 		__sock_put(sk);
615 	}
616 	return rc;
617 }
618 
619 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
620 {
621 	hlist_add_head(&sk->sk_node, list);
622 }
623 
624 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
625 {
626 	sock_hold(sk);
627 	__sk_add_node(sk, list);
628 }
629 
630 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
631 {
632 	sock_hold(sk);
633 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
634 	    sk->sk_family == AF_INET6)
635 		hlist_add_tail_rcu(&sk->sk_node, list);
636 	else
637 		hlist_add_head_rcu(&sk->sk_node, list);
638 }
639 
640 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
641 {
642 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
643 	    sk->sk_family == AF_INET6)
644 		hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
645 	else
646 		hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
647 }
648 
649 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
650 {
651 	sock_hold(sk);
652 	__sk_nulls_add_node_rcu(sk, list);
653 }
654 
655 static inline void __sk_del_bind_node(struct sock *sk)
656 {
657 	__hlist_del(&sk->sk_bind_node);
658 }
659 
660 static inline void sk_add_bind_node(struct sock *sk,
661 					struct hlist_head *list)
662 {
663 	hlist_add_head(&sk->sk_bind_node, list);
664 }
665 
666 #define sk_for_each(__sk, list) \
667 	hlist_for_each_entry(__sk, list, sk_node)
668 #define sk_for_each_rcu(__sk, list) \
669 	hlist_for_each_entry_rcu(__sk, list, sk_node)
670 #define sk_nulls_for_each(__sk, node, list) \
671 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
672 #define sk_nulls_for_each_rcu(__sk, node, list) \
673 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
674 #define sk_for_each_from(__sk) \
675 	hlist_for_each_entry_from(__sk, sk_node)
676 #define sk_nulls_for_each_from(__sk, node) \
677 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
678 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
679 #define sk_for_each_safe(__sk, tmp, list) \
680 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
681 #define sk_for_each_bound(__sk, list) \
682 	hlist_for_each_entry(__sk, list, sk_bind_node)
683 
684 /**
685  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
686  * @tpos:	the type * to use as a loop cursor.
687  * @pos:	the &struct hlist_node to use as a loop cursor.
688  * @head:	the head for your list.
689  * @offset:	offset of hlist_node within the struct.
690  *
691  */
692 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
693 	for (pos = rcu_dereference((head)->first);			       \
694 	     pos != NULL &&						       \
695 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
696 	     pos = rcu_dereference(pos->next))
697 
698 static inline struct user_namespace *sk_user_ns(struct sock *sk)
699 {
700 	/* Careful only use this in a context where these parameters
701 	 * can not change and must all be valid, such as recvmsg from
702 	 * userspace.
703 	 */
704 	return sk->sk_socket->file->f_cred->user_ns;
705 }
706 
707 /* Sock flags */
708 enum sock_flags {
709 	SOCK_DEAD,
710 	SOCK_DONE,
711 	SOCK_URGINLINE,
712 	SOCK_KEEPOPEN,
713 	SOCK_LINGER,
714 	SOCK_DESTROY,
715 	SOCK_BROADCAST,
716 	SOCK_TIMESTAMP,
717 	SOCK_ZAPPED,
718 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
719 	SOCK_DBG, /* %SO_DEBUG setting */
720 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
721 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
722 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
723 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
724 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
725 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
726 	SOCK_FASYNC, /* fasync() active */
727 	SOCK_RXQ_OVFL,
728 	SOCK_ZEROCOPY, /* buffers from userspace */
729 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
730 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
731 		     * Will use last 4 bytes of packet sent from
732 		     * user-space instead.
733 		     */
734 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
735 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
736 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
737 };
738 
739 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
740 
741 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
742 {
743 	nsk->sk_flags = osk->sk_flags;
744 }
745 
746 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
747 {
748 	__set_bit(flag, &sk->sk_flags);
749 }
750 
751 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
752 {
753 	__clear_bit(flag, &sk->sk_flags);
754 }
755 
756 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
757 {
758 	return test_bit(flag, &sk->sk_flags);
759 }
760 
761 #ifdef CONFIG_NET
762 extern struct static_key memalloc_socks;
763 static inline int sk_memalloc_socks(void)
764 {
765 	return static_key_false(&memalloc_socks);
766 }
767 #else
768 
769 static inline int sk_memalloc_socks(void)
770 {
771 	return 0;
772 }
773 
774 #endif
775 
776 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
777 {
778 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
779 }
780 
781 static inline void sk_acceptq_removed(struct sock *sk)
782 {
783 	sk->sk_ack_backlog--;
784 }
785 
786 static inline void sk_acceptq_added(struct sock *sk)
787 {
788 	sk->sk_ack_backlog++;
789 }
790 
791 static inline bool sk_acceptq_is_full(const struct sock *sk)
792 {
793 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
794 }
795 
796 /*
797  * Compute minimal free write space needed to queue new packets.
798  */
799 static inline int sk_stream_min_wspace(const struct sock *sk)
800 {
801 	return sk->sk_wmem_queued >> 1;
802 }
803 
804 static inline int sk_stream_wspace(const struct sock *sk)
805 {
806 	return sk->sk_sndbuf - sk->sk_wmem_queued;
807 }
808 
809 void sk_stream_write_space(struct sock *sk);
810 
811 /* OOB backlog add */
812 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
813 {
814 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
815 	skb_dst_force_safe(skb);
816 
817 	if (!sk->sk_backlog.tail)
818 		sk->sk_backlog.head = skb;
819 	else
820 		sk->sk_backlog.tail->next = skb;
821 
822 	sk->sk_backlog.tail = skb;
823 	skb->next = NULL;
824 }
825 
826 /*
827  * Take into account size of receive queue and backlog queue
828  * Do not take into account this skb truesize,
829  * to allow even a single big packet to come.
830  */
831 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
832 {
833 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
834 
835 	return qsize > limit;
836 }
837 
838 /* The per-socket spinlock must be held here. */
839 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
840 					      unsigned int limit)
841 {
842 	if (sk_rcvqueues_full(sk, limit))
843 		return -ENOBUFS;
844 
845 	/*
846 	 * If the skb was allocated from pfmemalloc reserves, only
847 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
848 	 * helping free memory
849 	 */
850 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
851 		return -ENOMEM;
852 
853 	__sk_add_backlog(sk, skb);
854 	sk->sk_backlog.len += skb->truesize;
855 	return 0;
856 }
857 
858 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
859 
860 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
861 {
862 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
863 		return __sk_backlog_rcv(sk, skb);
864 
865 	return sk->sk_backlog_rcv(sk, skb);
866 }
867 
868 static inline void sk_incoming_cpu_update(struct sock *sk)
869 {
870 	sk->sk_incoming_cpu = raw_smp_processor_id();
871 }
872 
873 static inline void sock_rps_record_flow_hash(__u32 hash)
874 {
875 #ifdef CONFIG_RPS
876 	struct rps_sock_flow_table *sock_flow_table;
877 
878 	rcu_read_lock();
879 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
880 	rps_record_sock_flow(sock_flow_table, hash);
881 	rcu_read_unlock();
882 #endif
883 }
884 
885 static inline void sock_rps_record_flow(const struct sock *sk)
886 {
887 #ifdef CONFIG_RPS
888 	sock_rps_record_flow_hash(sk->sk_rxhash);
889 #endif
890 }
891 
892 static inline void sock_rps_save_rxhash(struct sock *sk,
893 					const struct sk_buff *skb)
894 {
895 #ifdef CONFIG_RPS
896 	if (unlikely(sk->sk_rxhash != skb->hash))
897 		sk->sk_rxhash = skb->hash;
898 #endif
899 }
900 
901 static inline void sock_rps_reset_rxhash(struct sock *sk)
902 {
903 #ifdef CONFIG_RPS
904 	sk->sk_rxhash = 0;
905 #endif
906 }
907 
908 #define sk_wait_event(__sk, __timeo, __condition)			\
909 	({	int __rc;						\
910 		release_sock(__sk);					\
911 		__rc = __condition;					\
912 		if (!__rc) {						\
913 			*(__timeo) = schedule_timeout(*(__timeo));	\
914 		}							\
915 		sched_annotate_sleep();						\
916 		lock_sock(__sk);					\
917 		__rc = __condition;					\
918 		__rc;							\
919 	})
920 
921 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
922 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
923 void sk_stream_wait_close(struct sock *sk, long timeo_p);
924 int sk_stream_error(struct sock *sk, int flags, int err);
925 void sk_stream_kill_queues(struct sock *sk);
926 void sk_set_memalloc(struct sock *sk);
927 void sk_clear_memalloc(struct sock *sk);
928 
929 void __sk_flush_backlog(struct sock *sk);
930 
931 static inline bool sk_flush_backlog(struct sock *sk)
932 {
933 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
934 		__sk_flush_backlog(sk);
935 		return true;
936 	}
937 	return false;
938 }
939 
940 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
941 
942 struct request_sock_ops;
943 struct timewait_sock_ops;
944 struct inet_hashinfo;
945 struct raw_hashinfo;
946 struct module;
947 
948 /*
949  * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
950  * un-modified. Special care is taken when initializing object to zero.
951  */
952 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
953 {
954 	if (offsetof(struct sock, sk_node.next) != 0)
955 		memset(sk, 0, offsetof(struct sock, sk_node.next));
956 	memset(&sk->sk_node.pprev, 0,
957 	       size - offsetof(struct sock, sk_node.pprev));
958 }
959 
960 /* Networking protocol blocks we attach to sockets.
961  * socket layer -> transport layer interface
962  */
963 struct proto {
964 	void			(*close)(struct sock *sk,
965 					long timeout);
966 	int			(*connect)(struct sock *sk,
967 					struct sockaddr *uaddr,
968 					int addr_len);
969 	int			(*disconnect)(struct sock *sk, int flags);
970 
971 	struct sock *		(*accept)(struct sock *sk, int flags, int *err);
972 
973 	int			(*ioctl)(struct sock *sk, int cmd,
974 					 unsigned long arg);
975 	int			(*init)(struct sock *sk);
976 	void			(*destroy)(struct sock *sk);
977 	void			(*shutdown)(struct sock *sk, int how);
978 	int			(*setsockopt)(struct sock *sk, int level,
979 					int optname, char __user *optval,
980 					unsigned int optlen);
981 	int			(*getsockopt)(struct sock *sk, int level,
982 					int optname, char __user *optval,
983 					int __user *option);
984 #ifdef CONFIG_COMPAT
985 	int			(*compat_setsockopt)(struct sock *sk,
986 					int level,
987 					int optname, char __user *optval,
988 					unsigned int optlen);
989 	int			(*compat_getsockopt)(struct sock *sk,
990 					int level,
991 					int optname, char __user *optval,
992 					int __user *option);
993 	int			(*compat_ioctl)(struct sock *sk,
994 					unsigned int cmd, unsigned long arg);
995 #endif
996 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
997 					   size_t len);
998 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
999 					   size_t len, int noblock, int flags,
1000 					   int *addr_len);
1001 	int			(*sendpage)(struct sock *sk, struct page *page,
1002 					int offset, size_t size, int flags);
1003 	int			(*bind)(struct sock *sk,
1004 					struct sockaddr *uaddr, int addr_len);
1005 
1006 	int			(*backlog_rcv) (struct sock *sk,
1007 						struct sk_buff *skb);
1008 
1009 	void		(*release_cb)(struct sock *sk);
1010 
1011 	/* Keeping track of sk's, looking them up, and port selection methods. */
1012 	int			(*hash)(struct sock *sk);
1013 	void			(*unhash)(struct sock *sk);
1014 	void			(*rehash)(struct sock *sk);
1015 	int			(*get_port)(struct sock *sk, unsigned short snum);
1016 	void			(*clear_sk)(struct sock *sk, int size);
1017 
1018 	/* Keeping track of sockets in use */
1019 #ifdef CONFIG_PROC_FS
1020 	unsigned int		inuse_idx;
1021 #endif
1022 
1023 	bool			(*stream_memory_free)(const struct sock *sk);
1024 	/* Memory pressure */
1025 	void			(*enter_memory_pressure)(struct sock *sk);
1026 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1027 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1028 	/*
1029 	 * Pressure flag: try to collapse.
1030 	 * Technical note: it is used by multiple contexts non atomically.
1031 	 * All the __sk_mem_schedule() is of this nature: accounting
1032 	 * is strict, actions are advisory and have some latency.
1033 	 */
1034 	int			*memory_pressure;
1035 	long			*sysctl_mem;
1036 	int			*sysctl_wmem;
1037 	int			*sysctl_rmem;
1038 	int			max_header;
1039 	bool			no_autobind;
1040 
1041 	struct kmem_cache	*slab;
1042 	unsigned int		obj_size;
1043 	int			slab_flags;
1044 
1045 	struct percpu_counter	*orphan_count;
1046 
1047 	struct request_sock_ops	*rsk_prot;
1048 	struct timewait_sock_ops *twsk_prot;
1049 
1050 	union {
1051 		struct inet_hashinfo	*hashinfo;
1052 		struct udp_table	*udp_table;
1053 		struct raw_hashinfo	*raw_hash;
1054 	} h;
1055 
1056 	struct module		*owner;
1057 
1058 	char			name[32];
1059 
1060 	struct list_head	node;
1061 #ifdef SOCK_REFCNT_DEBUG
1062 	atomic_t		socks;
1063 #endif
1064 	int			(*diag_destroy)(struct sock *sk, int err);
1065 };
1066 
1067 int proto_register(struct proto *prot, int alloc_slab);
1068 void proto_unregister(struct proto *prot);
1069 
1070 #ifdef SOCK_REFCNT_DEBUG
1071 static inline void sk_refcnt_debug_inc(struct sock *sk)
1072 {
1073 	atomic_inc(&sk->sk_prot->socks);
1074 }
1075 
1076 static inline void sk_refcnt_debug_dec(struct sock *sk)
1077 {
1078 	atomic_dec(&sk->sk_prot->socks);
1079 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1080 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1081 }
1082 
1083 static inline void sk_refcnt_debug_release(const struct sock *sk)
1084 {
1085 	if (atomic_read(&sk->sk_refcnt) != 1)
1086 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1087 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1088 }
1089 #else /* SOCK_REFCNT_DEBUG */
1090 #define sk_refcnt_debug_inc(sk) do { } while (0)
1091 #define sk_refcnt_debug_dec(sk) do { } while (0)
1092 #define sk_refcnt_debug_release(sk) do { } while (0)
1093 #endif /* SOCK_REFCNT_DEBUG */
1094 
1095 static inline bool sk_stream_memory_free(const struct sock *sk)
1096 {
1097 	if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1098 		return false;
1099 
1100 	return sk->sk_prot->stream_memory_free ?
1101 		sk->sk_prot->stream_memory_free(sk) : true;
1102 }
1103 
1104 static inline bool sk_stream_is_writeable(const struct sock *sk)
1105 {
1106 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1107 	       sk_stream_memory_free(sk);
1108 }
1109 
1110 
1111 static inline bool sk_has_memory_pressure(const struct sock *sk)
1112 {
1113 	return sk->sk_prot->memory_pressure != NULL;
1114 }
1115 
1116 static inline bool sk_under_memory_pressure(const struct sock *sk)
1117 {
1118 	if (!sk->sk_prot->memory_pressure)
1119 		return false;
1120 
1121 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1122 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1123 		return true;
1124 
1125 	return !!*sk->sk_prot->memory_pressure;
1126 }
1127 
1128 static inline void sk_leave_memory_pressure(struct sock *sk)
1129 {
1130 	int *memory_pressure = sk->sk_prot->memory_pressure;
1131 
1132 	if (!memory_pressure)
1133 		return;
1134 
1135 	if (*memory_pressure)
1136 		*memory_pressure = 0;
1137 }
1138 
1139 static inline void sk_enter_memory_pressure(struct sock *sk)
1140 {
1141 	if (!sk->sk_prot->enter_memory_pressure)
1142 		return;
1143 
1144 	sk->sk_prot->enter_memory_pressure(sk);
1145 }
1146 
1147 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1148 {
1149 	return sk->sk_prot->sysctl_mem[index];
1150 }
1151 
1152 static inline long
1153 sk_memory_allocated(const struct sock *sk)
1154 {
1155 	return atomic_long_read(sk->sk_prot->memory_allocated);
1156 }
1157 
1158 static inline long
1159 sk_memory_allocated_add(struct sock *sk, int amt)
1160 {
1161 	return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1162 }
1163 
1164 static inline void
1165 sk_memory_allocated_sub(struct sock *sk, int amt)
1166 {
1167 	atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1168 }
1169 
1170 static inline void sk_sockets_allocated_dec(struct sock *sk)
1171 {
1172 	percpu_counter_dec(sk->sk_prot->sockets_allocated);
1173 }
1174 
1175 static inline void sk_sockets_allocated_inc(struct sock *sk)
1176 {
1177 	percpu_counter_inc(sk->sk_prot->sockets_allocated);
1178 }
1179 
1180 static inline int
1181 sk_sockets_allocated_read_positive(struct sock *sk)
1182 {
1183 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1184 }
1185 
1186 static inline int
1187 proto_sockets_allocated_sum_positive(struct proto *prot)
1188 {
1189 	return percpu_counter_sum_positive(prot->sockets_allocated);
1190 }
1191 
1192 static inline long
1193 proto_memory_allocated(struct proto *prot)
1194 {
1195 	return atomic_long_read(prot->memory_allocated);
1196 }
1197 
1198 static inline bool
1199 proto_memory_pressure(struct proto *prot)
1200 {
1201 	if (!prot->memory_pressure)
1202 		return false;
1203 	return !!*prot->memory_pressure;
1204 }
1205 
1206 
1207 #ifdef CONFIG_PROC_FS
1208 /* Called with local bh disabled */
1209 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1210 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1211 #else
1212 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1213 		int inc)
1214 {
1215 }
1216 #endif
1217 
1218 
1219 /* With per-bucket locks this operation is not-atomic, so that
1220  * this version is not worse.
1221  */
1222 static inline int __sk_prot_rehash(struct sock *sk)
1223 {
1224 	sk->sk_prot->unhash(sk);
1225 	return sk->sk_prot->hash(sk);
1226 }
1227 
1228 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1229 
1230 /* About 10 seconds */
1231 #define SOCK_DESTROY_TIME (10*HZ)
1232 
1233 /* Sockets 0-1023 can't be bound to unless you are superuser */
1234 #define PROT_SOCK	1024
1235 
1236 #define SHUTDOWN_MASK	3
1237 #define RCV_SHUTDOWN	1
1238 #define SEND_SHUTDOWN	2
1239 
1240 #define SOCK_SNDBUF_LOCK	1
1241 #define SOCK_RCVBUF_LOCK	2
1242 #define SOCK_BINDADDR_LOCK	4
1243 #define SOCK_BINDPORT_LOCK	8
1244 
1245 struct socket_alloc {
1246 	struct socket socket;
1247 	struct inode vfs_inode;
1248 };
1249 
1250 static inline struct socket *SOCKET_I(struct inode *inode)
1251 {
1252 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1253 }
1254 
1255 static inline struct inode *SOCK_INODE(struct socket *socket)
1256 {
1257 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1258 }
1259 
1260 /*
1261  * Functions for memory accounting
1262  */
1263 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1264 void __sk_mem_reclaim(struct sock *sk, int amount);
1265 
1266 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1267 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1268 #define SK_MEM_SEND	0
1269 #define SK_MEM_RECV	1
1270 
1271 static inline int sk_mem_pages(int amt)
1272 {
1273 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1274 }
1275 
1276 static inline bool sk_has_account(struct sock *sk)
1277 {
1278 	/* return true if protocol supports memory accounting */
1279 	return !!sk->sk_prot->memory_allocated;
1280 }
1281 
1282 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1283 {
1284 	if (!sk_has_account(sk))
1285 		return true;
1286 	return size <= sk->sk_forward_alloc ||
1287 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1288 }
1289 
1290 static inline bool
1291 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1292 {
1293 	if (!sk_has_account(sk))
1294 		return true;
1295 	return size<= sk->sk_forward_alloc ||
1296 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1297 		skb_pfmemalloc(skb);
1298 }
1299 
1300 static inline void sk_mem_reclaim(struct sock *sk)
1301 {
1302 	if (!sk_has_account(sk))
1303 		return;
1304 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1305 		__sk_mem_reclaim(sk, sk->sk_forward_alloc);
1306 }
1307 
1308 static inline void sk_mem_reclaim_partial(struct sock *sk)
1309 {
1310 	if (!sk_has_account(sk))
1311 		return;
1312 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1313 		__sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1314 }
1315 
1316 static inline void sk_mem_charge(struct sock *sk, int size)
1317 {
1318 	if (!sk_has_account(sk))
1319 		return;
1320 	sk->sk_forward_alloc -= size;
1321 }
1322 
1323 static inline void sk_mem_uncharge(struct sock *sk, int size)
1324 {
1325 	if (!sk_has_account(sk))
1326 		return;
1327 	sk->sk_forward_alloc += size;
1328 }
1329 
1330 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1331 {
1332 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1333 	sk->sk_wmem_queued -= skb->truesize;
1334 	sk_mem_uncharge(sk, skb->truesize);
1335 	__kfree_skb(skb);
1336 }
1337 
1338 static inline void sock_release_ownership(struct sock *sk)
1339 {
1340 	if (sk->sk_lock.owned) {
1341 		sk->sk_lock.owned = 0;
1342 
1343 		/* The sk_lock has mutex_unlock() semantics: */
1344 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1345 	}
1346 }
1347 
1348 /*
1349  * Macro so as to not evaluate some arguments when
1350  * lockdep is not enabled.
1351  *
1352  * Mark both the sk_lock and the sk_lock.slock as a
1353  * per-address-family lock class.
1354  */
1355 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1356 do {									\
1357 	sk->sk_lock.owned = 0;						\
1358 	init_waitqueue_head(&sk->sk_lock.wq);				\
1359 	spin_lock_init(&(sk)->sk_lock.slock);				\
1360 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1361 			sizeof((sk)->sk_lock));				\
1362 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1363 				(skey), (sname));				\
1364 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1365 } while (0)
1366 
1367 #ifdef CONFIG_LOCKDEP
1368 static inline bool lockdep_sock_is_held(const struct sock *csk)
1369 {
1370 	struct sock *sk = (struct sock *)csk;
1371 
1372 	return lockdep_is_held(&sk->sk_lock) ||
1373 	       lockdep_is_held(&sk->sk_lock.slock);
1374 }
1375 #endif
1376 
1377 void lock_sock_nested(struct sock *sk, int subclass);
1378 
1379 static inline void lock_sock(struct sock *sk)
1380 {
1381 	lock_sock_nested(sk, 0);
1382 }
1383 
1384 void release_sock(struct sock *sk);
1385 
1386 /* BH context may only use the following locking interface. */
1387 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1388 #define bh_lock_sock_nested(__sk) \
1389 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1390 				SINGLE_DEPTH_NESTING)
1391 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1392 
1393 bool lock_sock_fast(struct sock *sk);
1394 /**
1395  * unlock_sock_fast - complement of lock_sock_fast
1396  * @sk: socket
1397  * @slow: slow mode
1398  *
1399  * fast unlock socket for user context.
1400  * If slow mode is on, we call regular release_sock()
1401  */
1402 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1403 {
1404 	if (slow)
1405 		release_sock(sk);
1406 	else
1407 		spin_unlock_bh(&sk->sk_lock.slock);
1408 }
1409 
1410 /* Used by processes to "lock" a socket state, so that
1411  * interrupts and bottom half handlers won't change it
1412  * from under us. It essentially blocks any incoming
1413  * packets, so that we won't get any new data or any
1414  * packets that change the state of the socket.
1415  *
1416  * While locked, BH processing will add new packets to
1417  * the backlog queue.  This queue is processed by the
1418  * owner of the socket lock right before it is released.
1419  *
1420  * Since ~2.3.5 it is also exclusive sleep lock serializing
1421  * accesses from user process context.
1422  */
1423 
1424 static inline void sock_owned_by_me(const struct sock *sk)
1425 {
1426 #ifdef CONFIG_LOCKDEP
1427 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1428 #endif
1429 }
1430 
1431 static inline bool sock_owned_by_user(const struct sock *sk)
1432 {
1433 	sock_owned_by_me(sk);
1434 	return sk->sk_lock.owned;
1435 }
1436 
1437 /* no reclassification while locks are held */
1438 static inline bool sock_allow_reclassification(const struct sock *csk)
1439 {
1440 	struct sock *sk = (struct sock *)csk;
1441 
1442 	return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1443 }
1444 
1445 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1446 		      struct proto *prot, int kern);
1447 void sk_free(struct sock *sk);
1448 void sk_destruct(struct sock *sk);
1449 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1450 
1451 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1452 			     gfp_t priority);
1453 void __sock_wfree(struct sk_buff *skb);
1454 void sock_wfree(struct sk_buff *skb);
1455 void skb_orphan_partial(struct sk_buff *skb);
1456 void sock_rfree(struct sk_buff *skb);
1457 void sock_efree(struct sk_buff *skb);
1458 #ifdef CONFIG_INET
1459 void sock_edemux(struct sk_buff *skb);
1460 #else
1461 #define sock_edemux(skb) sock_efree(skb)
1462 #endif
1463 
1464 int sock_setsockopt(struct socket *sock, int level, int op,
1465 		    char __user *optval, unsigned int optlen);
1466 
1467 int sock_getsockopt(struct socket *sock, int level, int op,
1468 		    char __user *optval, int __user *optlen);
1469 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1470 				    int noblock, int *errcode);
1471 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1472 				     unsigned long data_len, int noblock,
1473 				     int *errcode, int max_page_order);
1474 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1475 void sock_kfree_s(struct sock *sk, void *mem, int size);
1476 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1477 void sk_send_sigurg(struct sock *sk);
1478 
1479 struct sockcm_cookie {
1480 	u32 mark;
1481 	u16 tsflags;
1482 };
1483 
1484 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1485 		     struct sockcm_cookie *sockc);
1486 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1487 		   struct sockcm_cookie *sockc);
1488 
1489 /*
1490  * Functions to fill in entries in struct proto_ops when a protocol
1491  * does not implement a particular function.
1492  */
1493 int sock_no_bind(struct socket *, struct sockaddr *, int);
1494 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1495 int sock_no_socketpair(struct socket *, struct socket *);
1496 int sock_no_accept(struct socket *, struct socket *, int);
1497 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1498 unsigned int sock_no_poll(struct file *, struct socket *,
1499 			  struct poll_table_struct *);
1500 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1501 int sock_no_listen(struct socket *, int);
1502 int sock_no_shutdown(struct socket *, int);
1503 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1504 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1505 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1506 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1507 int sock_no_mmap(struct file *file, struct socket *sock,
1508 		 struct vm_area_struct *vma);
1509 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1510 			 size_t size, int flags);
1511 
1512 /*
1513  * Functions to fill in entries in struct proto_ops when a protocol
1514  * uses the inet style.
1515  */
1516 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1517 				  char __user *optval, int __user *optlen);
1518 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1519 			int flags);
1520 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1521 				  char __user *optval, unsigned int optlen);
1522 int compat_sock_common_getsockopt(struct socket *sock, int level,
1523 		int optname, char __user *optval, int __user *optlen);
1524 int compat_sock_common_setsockopt(struct socket *sock, int level,
1525 		int optname, char __user *optval, unsigned int optlen);
1526 
1527 void sk_common_release(struct sock *sk);
1528 
1529 /*
1530  *	Default socket callbacks and setup code
1531  */
1532 
1533 /* Initialise core socket variables */
1534 void sock_init_data(struct socket *sock, struct sock *sk);
1535 
1536 /*
1537  * Socket reference counting postulates.
1538  *
1539  * * Each user of socket SHOULD hold a reference count.
1540  * * Each access point to socket (an hash table bucket, reference from a list,
1541  *   running timer, skb in flight MUST hold a reference count.
1542  * * When reference count hits 0, it means it will never increase back.
1543  * * When reference count hits 0, it means that no references from
1544  *   outside exist to this socket and current process on current CPU
1545  *   is last user and may/should destroy this socket.
1546  * * sk_free is called from any context: process, BH, IRQ. When
1547  *   it is called, socket has no references from outside -> sk_free
1548  *   may release descendant resources allocated by the socket, but
1549  *   to the time when it is called, socket is NOT referenced by any
1550  *   hash tables, lists etc.
1551  * * Packets, delivered from outside (from network or from another process)
1552  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1553  *   when they sit in queue. Otherwise, packets will leak to hole, when
1554  *   socket is looked up by one cpu and unhasing is made by another CPU.
1555  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1556  *   (leak to backlog). Packet socket does all the processing inside
1557  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1558  *   use separate SMP lock, so that they are prone too.
1559  */
1560 
1561 /* Ungrab socket and destroy it, if it was the last reference. */
1562 static inline void sock_put(struct sock *sk)
1563 {
1564 	if (atomic_dec_and_test(&sk->sk_refcnt))
1565 		sk_free(sk);
1566 }
1567 /* Generic version of sock_put(), dealing with all sockets
1568  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1569  */
1570 void sock_gen_put(struct sock *sk);
1571 
1572 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1573 
1574 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1575 {
1576 	sk->sk_tx_queue_mapping = tx_queue;
1577 }
1578 
1579 static inline void sk_tx_queue_clear(struct sock *sk)
1580 {
1581 	sk->sk_tx_queue_mapping = -1;
1582 }
1583 
1584 static inline int sk_tx_queue_get(const struct sock *sk)
1585 {
1586 	return sk ? sk->sk_tx_queue_mapping : -1;
1587 }
1588 
1589 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1590 {
1591 	sk_tx_queue_clear(sk);
1592 	sk->sk_socket = sock;
1593 }
1594 
1595 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1596 {
1597 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1598 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1599 }
1600 /* Detach socket from process context.
1601  * Announce socket dead, detach it from wait queue and inode.
1602  * Note that parent inode held reference count on this struct sock,
1603  * we do not release it in this function, because protocol
1604  * probably wants some additional cleanups or even continuing
1605  * to work with this socket (TCP).
1606  */
1607 static inline void sock_orphan(struct sock *sk)
1608 {
1609 	write_lock_bh(&sk->sk_callback_lock);
1610 	sock_set_flag(sk, SOCK_DEAD);
1611 	sk_set_socket(sk, NULL);
1612 	sk->sk_wq  = NULL;
1613 	write_unlock_bh(&sk->sk_callback_lock);
1614 }
1615 
1616 static inline void sock_graft(struct sock *sk, struct socket *parent)
1617 {
1618 	write_lock_bh(&sk->sk_callback_lock);
1619 	sk->sk_wq = parent->wq;
1620 	parent->sk = sk;
1621 	sk_set_socket(sk, parent);
1622 	security_sock_graft(sk, parent);
1623 	write_unlock_bh(&sk->sk_callback_lock);
1624 }
1625 
1626 kuid_t sock_i_uid(struct sock *sk);
1627 unsigned long sock_i_ino(struct sock *sk);
1628 
1629 static inline u32 net_tx_rndhash(void)
1630 {
1631 	u32 v = prandom_u32();
1632 
1633 	return v ?: 1;
1634 }
1635 
1636 static inline void sk_set_txhash(struct sock *sk)
1637 {
1638 	sk->sk_txhash = net_tx_rndhash();
1639 }
1640 
1641 static inline void sk_rethink_txhash(struct sock *sk)
1642 {
1643 	if (sk->sk_txhash)
1644 		sk_set_txhash(sk);
1645 }
1646 
1647 static inline struct dst_entry *
1648 __sk_dst_get(struct sock *sk)
1649 {
1650 	return rcu_dereference_check(sk->sk_dst_cache,
1651 				     lockdep_sock_is_held(sk));
1652 }
1653 
1654 static inline struct dst_entry *
1655 sk_dst_get(struct sock *sk)
1656 {
1657 	struct dst_entry *dst;
1658 
1659 	rcu_read_lock();
1660 	dst = rcu_dereference(sk->sk_dst_cache);
1661 	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1662 		dst = NULL;
1663 	rcu_read_unlock();
1664 	return dst;
1665 }
1666 
1667 static inline void dst_negative_advice(struct sock *sk)
1668 {
1669 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1670 
1671 	sk_rethink_txhash(sk);
1672 
1673 	if (dst && dst->ops->negative_advice) {
1674 		ndst = dst->ops->negative_advice(dst);
1675 
1676 		if (ndst != dst) {
1677 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1678 			sk_tx_queue_clear(sk);
1679 		}
1680 	}
1681 }
1682 
1683 static inline void
1684 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1685 {
1686 	struct dst_entry *old_dst;
1687 
1688 	sk_tx_queue_clear(sk);
1689 	/*
1690 	 * This can be called while sk is owned by the caller only,
1691 	 * with no state that can be checked in a rcu_dereference_check() cond
1692 	 */
1693 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1694 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1695 	dst_release(old_dst);
1696 }
1697 
1698 static inline void
1699 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1700 {
1701 	struct dst_entry *old_dst;
1702 
1703 	sk_tx_queue_clear(sk);
1704 	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1705 	dst_release(old_dst);
1706 }
1707 
1708 static inline void
1709 __sk_dst_reset(struct sock *sk)
1710 {
1711 	__sk_dst_set(sk, NULL);
1712 }
1713 
1714 static inline void
1715 sk_dst_reset(struct sock *sk)
1716 {
1717 	sk_dst_set(sk, NULL);
1718 }
1719 
1720 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1721 
1722 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1723 
1724 bool sk_mc_loop(struct sock *sk);
1725 
1726 static inline bool sk_can_gso(const struct sock *sk)
1727 {
1728 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1729 }
1730 
1731 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1732 
1733 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1734 {
1735 	sk->sk_route_nocaps |= flags;
1736 	sk->sk_route_caps &= ~flags;
1737 }
1738 
1739 static inline bool sk_check_csum_caps(struct sock *sk)
1740 {
1741 	return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1742 	       (sk->sk_family == PF_INET &&
1743 		(sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1744 	       (sk->sk_family == PF_INET6 &&
1745 		(sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1746 }
1747 
1748 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1749 					   struct iov_iter *from, char *to,
1750 					   int copy, int offset)
1751 {
1752 	if (skb->ip_summed == CHECKSUM_NONE) {
1753 		__wsum csum = 0;
1754 		if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1755 			return -EFAULT;
1756 		skb->csum = csum_block_add(skb->csum, csum, offset);
1757 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1758 		if (copy_from_iter_nocache(to, copy, from) != copy)
1759 			return -EFAULT;
1760 	} else if (copy_from_iter(to, copy, from) != copy)
1761 		return -EFAULT;
1762 
1763 	return 0;
1764 }
1765 
1766 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1767 				       struct iov_iter *from, int copy)
1768 {
1769 	int err, offset = skb->len;
1770 
1771 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1772 				       copy, offset);
1773 	if (err)
1774 		__skb_trim(skb, offset);
1775 
1776 	return err;
1777 }
1778 
1779 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1780 					   struct sk_buff *skb,
1781 					   struct page *page,
1782 					   int off, int copy)
1783 {
1784 	int err;
1785 
1786 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1787 				       copy, skb->len);
1788 	if (err)
1789 		return err;
1790 
1791 	skb->len	     += copy;
1792 	skb->data_len	     += copy;
1793 	skb->truesize	     += copy;
1794 	sk->sk_wmem_queued   += copy;
1795 	sk_mem_charge(sk, copy);
1796 	return 0;
1797 }
1798 
1799 /**
1800  * sk_wmem_alloc_get - returns write allocations
1801  * @sk: socket
1802  *
1803  * Returns sk_wmem_alloc minus initial offset of one
1804  */
1805 static inline int sk_wmem_alloc_get(const struct sock *sk)
1806 {
1807 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1808 }
1809 
1810 /**
1811  * sk_rmem_alloc_get - returns read allocations
1812  * @sk: socket
1813  *
1814  * Returns sk_rmem_alloc
1815  */
1816 static inline int sk_rmem_alloc_get(const struct sock *sk)
1817 {
1818 	return atomic_read(&sk->sk_rmem_alloc);
1819 }
1820 
1821 /**
1822  * sk_has_allocations - check if allocations are outstanding
1823  * @sk: socket
1824  *
1825  * Returns true if socket has write or read allocations
1826  */
1827 static inline bool sk_has_allocations(const struct sock *sk)
1828 {
1829 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1830 }
1831 
1832 /**
1833  * skwq_has_sleeper - check if there are any waiting processes
1834  * @wq: struct socket_wq
1835  *
1836  * Returns true if socket_wq has waiting processes
1837  *
1838  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1839  * barrier call. They were added due to the race found within the tcp code.
1840  *
1841  * Consider following tcp code paths:
1842  *
1843  * CPU1                  CPU2
1844  *
1845  * sys_select            receive packet
1846  *   ...                 ...
1847  *   __add_wait_queue    update tp->rcv_nxt
1848  *   ...                 ...
1849  *   tp->rcv_nxt check   sock_def_readable
1850  *   ...                 {
1851  *   schedule               rcu_read_lock();
1852  *                          wq = rcu_dereference(sk->sk_wq);
1853  *                          if (wq && waitqueue_active(&wq->wait))
1854  *                              wake_up_interruptible(&wq->wait)
1855  *                          ...
1856  *                       }
1857  *
1858  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1859  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1860  * could then endup calling schedule and sleep forever if there are no more
1861  * data on the socket.
1862  *
1863  */
1864 static inline bool skwq_has_sleeper(struct socket_wq *wq)
1865 {
1866 	return wq && wq_has_sleeper(&wq->wait);
1867 }
1868 
1869 /**
1870  * sock_poll_wait - place memory barrier behind the poll_wait call.
1871  * @filp:           file
1872  * @wait_address:   socket wait queue
1873  * @p:              poll_table
1874  *
1875  * See the comments in the wq_has_sleeper function.
1876  */
1877 static inline void sock_poll_wait(struct file *filp,
1878 		wait_queue_head_t *wait_address, poll_table *p)
1879 {
1880 	if (!poll_does_not_wait(p) && wait_address) {
1881 		poll_wait(filp, wait_address, p);
1882 		/* We need to be sure we are in sync with the
1883 		 * socket flags modification.
1884 		 *
1885 		 * This memory barrier is paired in the wq_has_sleeper.
1886 		 */
1887 		smp_mb();
1888 	}
1889 }
1890 
1891 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1892 {
1893 	if (sk->sk_txhash) {
1894 		skb->l4_hash = 1;
1895 		skb->hash = sk->sk_txhash;
1896 	}
1897 }
1898 
1899 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
1900 
1901 /*
1902  *	Queue a received datagram if it will fit. Stream and sequenced
1903  *	protocols can't normally use this as they need to fit buffers in
1904  *	and play with them.
1905  *
1906  *	Inlined as it's very short and called for pretty much every
1907  *	packet ever received.
1908  */
1909 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1910 {
1911 	skb_orphan(skb);
1912 	skb->sk = sk;
1913 	skb->destructor = sock_rfree;
1914 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1915 	sk_mem_charge(sk, skb->truesize);
1916 }
1917 
1918 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1919 		    unsigned long expires);
1920 
1921 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1922 
1923 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1924 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1925 
1926 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1927 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1928 
1929 /*
1930  *	Recover an error report and clear atomically
1931  */
1932 
1933 static inline int sock_error(struct sock *sk)
1934 {
1935 	int err;
1936 	if (likely(!sk->sk_err))
1937 		return 0;
1938 	err = xchg(&sk->sk_err, 0);
1939 	return -err;
1940 }
1941 
1942 static inline unsigned long sock_wspace(struct sock *sk)
1943 {
1944 	int amt = 0;
1945 
1946 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1947 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1948 		if (amt < 0)
1949 			amt = 0;
1950 	}
1951 	return amt;
1952 }
1953 
1954 /* Note:
1955  *  We use sk->sk_wq_raw, from contexts knowing this
1956  *  pointer is not NULL and cannot disappear/change.
1957  */
1958 static inline void sk_set_bit(int nr, struct sock *sk)
1959 {
1960 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
1961 	    !sock_flag(sk, SOCK_FASYNC))
1962 		return;
1963 
1964 	set_bit(nr, &sk->sk_wq_raw->flags);
1965 }
1966 
1967 static inline void sk_clear_bit(int nr, struct sock *sk)
1968 {
1969 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
1970 	    !sock_flag(sk, SOCK_FASYNC))
1971 		return;
1972 
1973 	clear_bit(nr, &sk->sk_wq_raw->flags);
1974 }
1975 
1976 static inline void sk_wake_async(const struct sock *sk, int how, int band)
1977 {
1978 	if (sock_flag(sk, SOCK_FASYNC)) {
1979 		rcu_read_lock();
1980 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
1981 		rcu_read_unlock();
1982 	}
1983 }
1984 
1985 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
1986  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
1987  * Note: for send buffers, TCP works better if we can build two skbs at
1988  * minimum.
1989  */
1990 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
1991 
1992 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
1993 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
1994 
1995 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1996 {
1997 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1998 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1999 		sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2000 	}
2001 }
2002 
2003 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2004 				    bool force_schedule);
2005 
2006 /**
2007  * sk_page_frag - return an appropriate page_frag
2008  * @sk: socket
2009  *
2010  * If socket allocation mode allows current thread to sleep, it means its
2011  * safe to use the per task page_frag instead of the per socket one.
2012  */
2013 static inline struct page_frag *sk_page_frag(struct sock *sk)
2014 {
2015 	if (gfpflags_allow_blocking(sk->sk_allocation))
2016 		return &current->task_frag;
2017 
2018 	return &sk->sk_frag;
2019 }
2020 
2021 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2022 
2023 /*
2024  *	Default write policy as shown to user space via poll/select/SIGIO
2025  */
2026 static inline bool sock_writeable(const struct sock *sk)
2027 {
2028 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2029 }
2030 
2031 static inline gfp_t gfp_any(void)
2032 {
2033 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2034 }
2035 
2036 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2037 {
2038 	return noblock ? 0 : sk->sk_rcvtimeo;
2039 }
2040 
2041 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2042 {
2043 	return noblock ? 0 : sk->sk_sndtimeo;
2044 }
2045 
2046 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2047 {
2048 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2049 }
2050 
2051 /* Alas, with timeout socket operations are not restartable.
2052  * Compare this to poll().
2053  */
2054 static inline int sock_intr_errno(long timeo)
2055 {
2056 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2057 }
2058 
2059 struct sock_skb_cb {
2060 	u32 dropcount;
2061 };
2062 
2063 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2064  * using skb->cb[] would keep using it directly and utilize its
2065  * alignement guarantee.
2066  */
2067 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2068 			    sizeof(struct sock_skb_cb)))
2069 
2070 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2071 			    SOCK_SKB_CB_OFFSET))
2072 
2073 #define sock_skb_cb_check_size(size) \
2074 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2075 
2076 static inline void
2077 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2078 {
2079 	SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2080 }
2081 
2082 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2083 {
2084 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2085 
2086 	atomic_add(segs, &sk->sk_drops);
2087 }
2088 
2089 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2090 			   struct sk_buff *skb);
2091 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2092 			     struct sk_buff *skb);
2093 
2094 static inline void
2095 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2096 {
2097 	ktime_t kt = skb->tstamp;
2098 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2099 
2100 	/*
2101 	 * generate control messages if
2102 	 * - receive time stamping in software requested
2103 	 * - software time stamp available and wanted
2104 	 * - hardware time stamps available and wanted
2105 	 */
2106 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2107 	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2108 	    (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2109 	    (hwtstamps->hwtstamp.tv64 &&
2110 	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2111 		__sock_recv_timestamp(msg, sk, skb);
2112 	else
2113 		sk->sk_stamp = kt;
2114 
2115 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2116 		__sock_recv_wifi_status(msg, sk, skb);
2117 }
2118 
2119 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2120 			      struct sk_buff *skb);
2121 
2122 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2123 					  struct sk_buff *skb)
2124 {
2125 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2126 			   (1UL << SOCK_RCVTSTAMP))
2127 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2128 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2129 
2130 	if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2131 		__sock_recv_ts_and_drops(msg, sk, skb);
2132 	else
2133 		sk->sk_stamp = skb->tstamp;
2134 }
2135 
2136 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2137 
2138 /**
2139  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2140  * @sk:		socket sending this packet
2141  * @tsflags:	timestamping flags to use
2142  * @tx_flags:	completed with instructions for time stamping
2143  *
2144  * Note : callers should take care of initial *tx_flags value (usually 0)
2145  */
2146 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2147 				     __u8 *tx_flags)
2148 {
2149 	if (unlikely(tsflags))
2150 		__sock_tx_timestamp(tsflags, tx_flags);
2151 	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2152 		*tx_flags |= SKBTX_WIFI_STATUS;
2153 }
2154 
2155 /**
2156  * sk_eat_skb - Release a skb if it is no longer needed
2157  * @sk: socket to eat this skb from
2158  * @skb: socket buffer to eat
2159  *
2160  * This routine must be called with interrupts disabled or with the socket
2161  * locked so that the sk_buff queue operation is ok.
2162 */
2163 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2164 {
2165 	__skb_unlink(skb, &sk->sk_receive_queue);
2166 	__kfree_skb(skb);
2167 }
2168 
2169 static inline
2170 struct net *sock_net(const struct sock *sk)
2171 {
2172 	return read_pnet(&sk->sk_net);
2173 }
2174 
2175 static inline
2176 void sock_net_set(struct sock *sk, struct net *net)
2177 {
2178 	write_pnet(&sk->sk_net, net);
2179 }
2180 
2181 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2182 {
2183 	if (skb->sk) {
2184 		struct sock *sk = skb->sk;
2185 
2186 		skb->destructor = NULL;
2187 		skb->sk = NULL;
2188 		return sk;
2189 	}
2190 	return NULL;
2191 }
2192 
2193 /* This helper checks if a socket is a full socket,
2194  * ie _not_ a timewait or request socket.
2195  */
2196 static inline bool sk_fullsock(const struct sock *sk)
2197 {
2198 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2199 }
2200 
2201 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2202  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2203  */
2204 static inline bool sk_listener(const struct sock *sk)
2205 {
2206 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2207 }
2208 
2209 /**
2210  * sk_state_load - read sk->sk_state for lockless contexts
2211  * @sk: socket pointer
2212  *
2213  * Paired with sk_state_store(). Used in places we do not hold socket lock :
2214  * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2215  */
2216 static inline int sk_state_load(const struct sock *sk)
2217 {
2218 	return smp_load_acquire(&sk->sk_state);
2219 }
2220 
2221 /**
2222  * sk_state_store - update sk->sk_state
2223  * @sk: socket pointer
2224  * @newstate: new state
2225  *
2226  * Paired with sk_state_load(). Should be used in contexts where
2227  * state change might impact lockless readers.
2228  */
2229 static inline void sk_state_store(struct sock *sk, int newstate)
2230 {
2231 	smp_store_release(&sk->sk_state, newstate);
2232 }
2233 
2234 void sock_enable_timestamp(struct sock *sk, int flag);
2235 int sock_get_timestamp(struct sock *, struct timeval __user *);
2236 int sock_get_timestampns(struct sock *, struct timespec __user *);
2237 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2238 		       int type);
2239 
2240 bool sk_ns_capable(const struct sock *sk,
2241 		   struct user_namespace *user_ns, int cap);
2242 bool sk_capable(const struct sock *sk, int cap);
2243 bool sk_net_capable(const struct sock *sk, int cap);
2244 
2245 extern __u32 sysctl_wmem_max;
2246 extern __u32 sysctl_rmem_max;
2247 
2248 extern int sysctl_tstamp_allow_data;
2249 extern int sysctl_optmem_max;
2250 
2251 extern __u32 sysctl_wmem_default;
2252 extern __u32 sysctl_rmem_default;
2253 
2254 #endif	/* _SOCK_H */
2255