xref: /openbmc/linux/include/net/sock.h (revision 133f9794)
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 #include <linux/rbtree.h>
64 #include <linux/filter.h>
65 #include <linux/rculist_nulls.h>
66 #include <linux/poll.h>
67 
68 #include <linux/atomic.h>
69 #include <linux/refcount.h>
70 #include <net/dst.h>
71 #include <net/checksum.h>
72 #include <net/tcp_states.h>
73 #include <linux/net_tstamp.h>
74 #include <net/smc.h>
75 #include <net/l3mdev.h>
76 
77 /*
78  * This structure really needs to be cleaned up.
79  * Most of it is for TCP, and not used by any of
80  * the other protocols.
81  */
82 
83 /* Define this to get the SOCK_DBG debugging facility. */
84 #define SOCK_DEBUGGING
85 #ifdef SOCK_DEBUGGING
86 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
87 					printk(KERN_DEBUG msg); } while (0)
88 #else
89 /* Validate arguments and do nothing */
90 static inline __printf(2, 3)
91 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
92 {
93 }
94 #endif
95 
96 /* This is the per-socket lock.  The spinlock provides a synchronization
97  * between user contexts and software interrupt processing, whereas the
98  * mini-semaphore synchronizes multiple users amongst themselves.
99  */
100 typedef struct {
101 	spinlock_t		slock;
102 	int			owned;
103 	wait_queue_head_t	wq;
104 	/*
105 	 * We express the mutex-alike socket_lock semantics
106 	 * to the lock validator by explicitly managing
107 	 * the slock as a lock variant (in addition to
108 	 * the slock itself):
109 	 */
110 #ifdef CONFIG_DEBUG_LOCK_ALLOC
111 	struct lockdep_map dep_map;
112 #endif
113 } socket_lock_t;
114 
115 struct sock;
116 struct proto;
117 struct net;
118 
119 typedef __u32 __bitwise __portpair;
120 typedef __u64 __bitwise __addrpair;
121 
122 /**
123  *	struct sock_common - minimal network layer representation of sockets
124  *	@skc_daddr: Foreign IPv4 addr
125  *	@skc_rcv_saddr: Bound local IPv4 addr
126  *	@skc_hash: hash value used with various protocol lookup tables
127  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
128  *	@skc_dport: placeholder for inet_dport/tw_dport
129  *	@skc_num: placeholder for inet_num/tw_num
130  *	@skc_family: network address family
131  *	@skc_state: Connection state
132  *	@skc_reuse: %SO_REUSEADDR setting
133  *	@skc_reuseport: %SO_REUSEPORT setting
134  *	@skc_bound_dev_if: bound device index if != 0
135  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
136  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137  *	@skc_prot: protocol handlers inside a network family
138  *	@skc_net: reference to the network namespace of this socket
139  *	@skc_node: main hash linkage for various protocol lookup tables
140  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
141  *	@skc_tx_queue_mapping: tx queue number for this connection
142  *	@skc_flags: place holder for sk_flags
143  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
144  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
145  *	@skc_incoming_cpu: record/match cpu processing incoming packets
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 on a 8 bytes aligned
153 	 * address on 64bit arches : cf INET_MATCH()
154 	 */
155 	union {
156 		__addrpair	skc_addrpair;
157 		struct {
158 			__be32	skc_daddr;
159 			__be32	skc_rcv_saddr;
160 		};
161 	};
162 	union  {
163 		unsigned int	skc_hash;
164 		__u16		skc_u16hashes[2];
165 	};
166 	/* skc_dport && skc_num must be grouped as well */
167 	union {
168 		__portpair	skc_portpair;
169 		struct {
170 			__be16	skc_dport;
171 			__u16	skc_num;
172 		};
173 	};
174 
175 	unsigned short		skc_family;
176 	volatile unsigned char	skc_state;
177 	unsigned char		skc_reuse:4;
178 	unsigned char		skc_reuseport:1;
179 	unsigned char		skc_ipv6only:1;
180 	unsigned char		skc_net_refcnt:1;
181 	int			skc_bound_dev_if;
182 	union {
183 		struct hlist_node	skc_bind_node;
184 		struct hlist_node	skc_portaddr_node;
185 	};
186 	struct proto		*skc_prot;
187 	possible_net_t		skc_net;
188 
189 #if IS_ENABLED(CONFIG_IPV6)
190 	struct in6_addr		skc_v6_daddr;
191 	struct in6_addr		skc_v6_rcv_saddr;
192 #endif
193 
194 	atomic64_t		skc_cookie;
195 
196 	/* following fields are padding to force
197 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
198 	 * assuming IPV6 is enabled. We use this padding differently
199 	 * for different kind of 'sockets'
200 	 */
201 	union {
202 		unsigned long	skc_flags;
203 		struct sock	*skc_listener; /* request_sock */
204 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
205 	};
206 	/*
207 	 * fields between dontcopy_begin/dontcopy_end
208 	 * are not copied in sock_copy()
209 	 */
210 	/* private: */
211 	int			skc_dontcopy_begin[0];
212 	/* public: */
213 	union {
214 		struct hlist_node	skc_node;
215 		struct hlist_nulls_node skc_nulls_node;
216 	};
217 	int			skc_tx_queue_mapping;
218 	union {
219 		int		skc_incoming_cpu;
220 		u32		skc_rcv_wnd;
221 		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
222 	};
223 
224 	refcount_t		skc_refcnt;
225 	/* private: */
226 	int                     skc_dontcopy_end[0];
227 	union {
228 		u32		skc_rxhash;
229 		u32		skc_window_clamp;
230 		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
231 	};
232 	/* public: */
233 };
234 
235 /**
236   *	struct sock - network layer representation of sockets
237   *	@__sk_common: shared layout with inet_timewait_sock
238   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
239   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
240   *	@sk_lock:	synchronizer
241   *	@sk_kern_sock: True if sock is using kernel lock classes
242   *	@sk_rcvbuf: size of receive buffer in bytes
243   *	@sk_wq: sock wait queue and async head
244   *	@sk_rx_dst: receive input route used by early demux
245   *	@sk_dst_cache: destination cache
246   *	@sk_dst_pending_confirm: need to confirm neighbour
247   *	@sk_policy: flow policy
248   *	@sk_receive_queue: incoming packets
249   *	@sk_wmem_alloc: transmit queue bytes committed
250   *	@sk_tsq_flags: TCP Small Queues flags
251   *	@sk_write_queue: Packet sending queue
252   *	@sk_omem_alloc: "o" is "option" or "other"
253   *	@sk_wmem_queued: persistent queue size
254   *	@sk_forward_alloc: space allocated forward
255   *	@sk_napi_id: id of the last napi context to receive data for sk
256   *	@sk_ll_usec: usecs to busypoll when there is no data
257   *	@sk_allocation: allocation mode
258   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
259   *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
260   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
261   *	@sk_sndbuf: size of send buffer in bytes
262   *	@__sk_flags_offset: empty field used to determine location of bitfield
263   *	@sk_padding: unused element for alignment
264   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
265   *	@sk_no_check_rx: allow zero checksum in RX packets
266   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
267   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
268   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
269   *	@sk_gso_max_size: Maximum GSO segment size to build
270   *	@sk_gso_max_segs: Maximum number of GSO segments
271   *	@sk_pacing_shift: scaling factor for TCP Small Queues
272   *	@sk_lingertime: %SO_LINGER l_linger setting
273   *	@sk_backlog: always used with the per-socket spinlock held
274   *	@sk_callback_lock: used with the callbacks in the end of this struct
275   *	@sk_error_queue: rarely used
276   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
277   *			  IPV6_ADDRFORM for instance)
278   *	@sk_err: last error
279   *	@sk_err_soft: errors that don't cause failure but are the cause of a
280   *		      persistent failure not just 'timed out'
281   *	@sk_drops: raw/udp drops counter
282   *	@sk_ack_backlog: current listen backlog
283   *	@sk_max_ack_backlog: listen backlog set in listen()
284   *	@sk_uid: user id of owner
285   *	@sk_priority: %SO_PRIORITY setting
286   *	@sk_type: socket type (%SOCK_STREAM, etc)
287   *	@sk_protocol: which protocol this socket belongs in this network family
288   *	@sk_peer_pid: &struct pid for this socket's peer
289   *	@sk_peer_cred: %SO_PEERCRED setting
290   *	@sk_rcvlowat: %SO_RCVLOWAT setting
291   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
292   *	@sk_sndtimeo: %SO_SNDTIMEO setting
293   *	@sk_txhash: computed flow hash for use on transmit
294   *	@sk_filter: socket filtering instructions
295   *	@sk_timer: sock cleanup timer
296   *	@sk_stamp: time stamp of last packet received
297   *	@sk_tsflags: SO_TIMESTAMPING socket options
298   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
299   *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
300   *	@sk_socket: Identd and reporting IO signals
301   *	@sk_user_data: RPC layer private data
302   *	@sk_frag: cached page frag
303   *	@sk_peek_off: current peek_offset value
304   *	@sk_send_head: front of stuff to transmit
305   *	@sk_security: used by security modules
306   *	@sk_mark: generic packet mark
307   *	@sk_cgrp_data: cgroup data for this cgroup
308   *	@sk_memcg: this socket's memory cgroup association
309   *	@sk_write_pending: a write to stream socket waits to start
310   *	@sk_state_change: callback to indicate change in the state of the sock
311   *	@sk_data_ready: callback to indicate there is data to be processed
312   *	@sk_write_space: callback to indicate there is bf sending space available
313   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
314   *	@sk_backlog_rcv: callback to process the backlog
315   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
316   *	@sk_reuseport_cb: reuseport group container
317   *	@sk_rcu: used during RCU grace period
318   */
319 struct sock {
320 	/*
321 	 * Now struct inet_timewait_sock also uses sock_common, so please just
322 	 * don't add nothing before this first member (__sk_common) --acme
323 	 */
324 	struct sock_common	__sk_common;
325 #define sk_node			__sk_common.skc_node
326 #define sk_nulls_node		__sk_common.skc_nulls_node
327 #define sk_refcnt		__sk_common.skc_refcnt
328 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
329 
330 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
331 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
332 #define sk_hash			__sk_common.skc_hash
333 #define sk_portpair		__sk_common.skc_portpair
334 #define sk_num			__sk_common.skc_num
335 #define sk_dport		__sk_common.skc_dport
336 #define sk_addrpair		__sk_common.skc_addrpair
337 #define sk_daddr		__sk_common.skc_daddr
338 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
339 #define sk_family		__sk_common.skc_family
340 #define sk_state		__sk_common.skc_state
341 #define sk_reuse		__sk_common.skc_reuse
342 #define sk_reuseport		__sk_common.skc_reuseport
343 #define sk_ipv6only		__sk_common.skc_ipv6only
344 #define sk_net_refcnt		__sk_common.skc_net_refcnt
345 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
346 #define sk_bind_node		__sk_common.skc_bind_node
347 #define sk_prot			__sk_common.skc_prot
348 #define sk_net			__sk_common.skc_net
349 #define sk_v6_daddr		__sk_common.skc_v6_daddr
350 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
351 #define sk_cookie		__sk_common.skc_cookie
352 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
353 #define sk_flags		__sk_common.skc_flags
354 #define sk_rxhash		__sk_common.skc_rxhash
355 
356 	socket_lock_t		sk_lock;
357 	atomic_t		sk_drops;
358 	int			sk_rcvlowat;
359 	struct sk_buff_head	sk_error_queue;
360 	struct sk_buff_head	sk_receive_queue;
361 	/*
362 	 * The backlog queue is special, it is always used with
363 	 * the per-socket spinlock held and requires low latency
364 	 * access. Therefore we special case it's implementation.
365 	 * Note : rmem_alloc is in this structure to fill a hole
366 	 * on 64bit arches, not because its logically part of
367 	 * backlog.
368 	 */
369 	struct {
370 		atomic_t	rmem_alloc;
371 		int		len;
372 		struct sk_buff	*head;
373 		struct sk_buff	*tail;
374 	} sk_backlog;
375 #define sk_rmem_alloc sk_backlog.rmem_alloc
376 
377 	int			sk_forward_alloc;
378 #ifdef CONFIG_NET_RX_BUSY_POLL
379 	unsigned int		sk_ll_usec;
380 	/* ===== mostly read cache line ===== */
381 	unsigned int		sk_napi_id;
382 #endif
383 	int			sk_rcvbuf;
384 
385 	struct sk_filter __rcu	*sk_filter;
386 	union {
387 		struct socket_wq __rcu	*sk_wq;
388 		struct socket_wq	*sk_wq_raw;
389 	};
390 #ifdef CONFIG_XFRM
391 	struct xfrm_policy __rcu *sk_policy[2];
392 #endif
393 	struct dst_entry	*sk_rx_dst;
394 	struct dst_entry __rcu	*sk_dst_cache;
395 	atomic_t		sk_omem_alloc;
396 	int			sk_sndbuf;
397 
398 	/* ===== cache line for TX ===== */
399 	int			sk_wmem_queued;
400 	refcount_t		sk_wmem_alloc;
401 	unsigned long		sk_tsq_flags;
402 	union {
403 		struct sk_buff	*sk_send_head;
404 		struct rb_root	tcp_rtx_queue;
405 	};
406 	struct sk_buff_head	sk_write_queue;
407 	__s32			sk_peek_off;
408 	int			sk_write_pending;
409 	__u32			sk_dst_pending_confirm;
410 	u32			sk_pacing_status; /* see enum sk_pacing */
411 	long			sk_sndtimeo;
412 	struct timer_list	sk_timer;
413 	__u32			sk_priority;
414 	__u32			sk_mark;
415 	u32			sk_pacing_rate; /* bytes per second */
416 	u32			sk_max_pacing_rate;
417 	struct page_frag	sk_frag;
418 	netdev_features_t	sk_route_caps;
419 	netdev_features_t	sk_route_nocaps;
420 	int			sk_gso_type;
421 	unsigned int		sk_gso_max_size;
422 	gfp_t			sk_allocation;
423 	__u32			sk_txhash;
424 
425 	/*
426 	 * Because of non atomicity rules, all
427 	 * changes are protected by socket lock.
428 	 */
429 	unsigned int		__sk_flags_offset[0];
430 #ifdef __BIG_ENDIAN_BITFIELD
431 #define SK_FL_PROTO_SHIFT  16
432 #define SK_FL_PROTO_MASK   0x00ff0000
433 
434 #define SK_FL_TYPE_SHIFT   0
435 #define SK_FL_TYPE_MASK    0x0000ffff
436 #else
437 #define SK_FL_PROTO_SHIFT  8
438 #define SK_FL_PROTO_MASK   0x0000ff00
439 
440 #define SK_FL_TYPE_SHIFT   16
441 #define SK_FL_TYPE_MASK    0xffff0000
442 #endif
443 
444 	unsigned int		sk_padding : 1,
445 				sk_kern_sock : 1,
446 				sk_no_check_tx : 1,
447 				sk_no_check_rx : 1,
448 				sk_userlocks : 4,
449 				sk_protocol  : 8,
450 				sk_type      : 16;
451 #define SK_PROTOCOL_MAX U8_MAX
452 	u16			sk_gso_max_segs;
453 	u8			sk_pacing_shift;
454 	unsigned long	        sk_lingertime;
455 	struct proto		*sk_prot_creator;
456 	rwlock_t		sk_callback_lock;
457 	int			sk_err,
458 				sk_err_soft;
459 	u32			sk_ack_backlog;
460 	u32			sk_max_ack_backlog;
461 	kuid_t			sk_uid;
462 	struct pid		*sk_peer_pid;
463 	const struct cred	*sk_peer_cred;
464 	long			sk_rcvtimeo;
465 	ktime_t			sk_stamp;
466 	u16			sk_tsflags;
467 	u8			sk_shutdown;
468 	u32			sk_tskey;
469 	atomic_t		sk_zckey;
470 	struct socket		*sk_socket;
471 	void			*sk_user_data;
472 #ifdef CONFIG_SECURITY
473 	void			*sk_security;
474 #endif
475 	struct sock_cgroup_data	sk_cgrp_data;
476 	struct mem_cgroup	*sk_memcg;
477 	void			(*sk_state_change)(struct sock *sk);
478 	void			(*sk_data_ready)(struct sock *sk);
479 	void			(*sk_write_space)(struct sock *sk);
480 	void			(*sk_error_report)(struct sock *sk);
481 	int			(*sk_backlog_rcv)(struct sock *sk,
482 						  struct sk_buff *skb);
483 	void                    (*sk_destruct)(struct sock *sk);
484 	struct sock_reuseport __rcu	*sk_reuseport_cb;
485 	struct rcu_head		sk_rcu;
486 };
487 
488 enum sk_pacing {
489 	SK_PACING_NONE		= 0,
490 	SK_PACING_NEEDED	= 1,
491 	SK_PACING_FQ		= 2,
492 };
493 
494 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
495 
496 #define rcu_dereference_sk_user_data(sk)	rcu_dereference(__sk_user_data((sk)))
497 #define rcu_assign_sk_user_data(sk, ptr)	rcu_assign_pointer(__sk_user_data((sk)), ptr)
498 
499 /*
500  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
501  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
502  * on a socket means that the socket will reuse everybody else's port
503  * without looking at the other's sk_reuse value.
504  */
505 
506 #define SK_NO_REUSE	0
507 #define SK_CAN_REUSE	1
508 #define SK_FORCE_REUSE	2
509 
510 int sk_set_peek_off(struct sock *sk, int val);
511 
512 static inline int sk_peek_offset(struct sock *sk, int flags)
513 {
514 	if (unlikely(flags & MSG_PEEK)) {
515 		return READ_ONCE(sk->sk_peek_off);
516 	}
517 
518 	return 0;
519 }
520 
521 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
522 {
523 	s32 off = READ_ONCE(sk->sk_peek_off);
524 
525 	if (unlikely(off >= 0)) {
526 		off = max_t(s32, off - val, 0);
527 		WRITE_ONCE(sk->sk_peek_off, off);
528 	}
529 }
530 
531 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
532 {
533 	sk_peek_offset_bwd(sk, -val);
534 }
535 
536 /*
537  * Hashed lists helper routines
538  */
539 static inline struct sock *sk_entry(const struct hlist_node *node)
540 {
541 	return hlist_entry(node, struct sock, sk_node);
542 }
543 
544 static inline struct sock *__sk_head(const struct hlist_head *head)
545 {
546 	return hlist_entry(head->first, struct sock, sk_node);
547 }
548 
549 static inline struct sock *sk_head(const struct hlist_head *head)
550 {
551 	return hlist_empty(head) ? NULL : __sk_head(head);
552 }
553 
554 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
555 {
556 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
557 }
558 
559 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
560 {
561 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
562 }
563 
564 static inline struct sock *sk_next(const struct sock *sk)
565 {
566 	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
567 }
568 
569 static inline struct sock *sk_nulls_next(const struct sock *sk)
570 {
571 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
572 		hlist_nulls_entry(sk->sk_nulls_node.next,
573 				  struct sock, sk_nulls_node) :
574 		NULL;
575 }
576 
577 static inline bool sk_unhashed(const struct sock *sk)
578 {
579 	return hlist_unhashed(&sk->sk_node);
580 }
581 
582 static inline bool sk_hashed(const struct sock *sk)
583 {
584 	return !sk_unhashed(sk);
585 }
586 
587 static inline void sk_node_init(struct hlist_node *node)
588 {
589 	node->pprev = NULL;
590 }
591 
592 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
593 {
594 	node->pprev = NULL;
595 }
596 
597 static inline void __sk_del_node(struct sock *sk)
598 {
599 	__hlist_del(&sk->sk_node);
600 }
601 
602 /* NB: equivalent to hlist_del_init_rcu */
603 static inline bool __sk_del_node_init(struct sock *sk)
604 {
605 	if (sk_hashed(sk)) {
606 		__sk_del_node(sk);
607 		sk_node_init(&sk->sk_node);
608 		return true;
609 	}
610 	return false;
611 }
612 
613 /* Grab socket reference count. This operation is valid only
614    when sk is ALREADY grabbed f.e. it is found in hash table
615    or a list and the lookup is made under lock preventing hash table
616    modifications.
617  */
618 
619 static __always_inline void sock_hold(struct sock *sk)
620 {
621 	refcount_inc(&sk->sk_refcnt);
622 }
623 
624 /* Ungrab socket in the context, which assumes that socket refcnt
625    cannot hit zero, f.e. it is true in context of any socketcall.
626  */
627 static __always_inline void __sock_put(struct sock *sk)
628 {
629 	refcount_dec(&sk->sk_refcnt);
630 }
631 
632 static inline bool sk_del_node_init(struct sock *sk)
633 {
634 	bool rc = __sk_del_node_init(sk);
635 
636 	if (rc) {
637 		/* paranoid for a while -acme */
638 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
639 		__sock_put(sk);
640 	}
641 	return rc;
642 }
643 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
644 
645 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
646 {
647 	if (sk_hashed(sk)) {
648 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
649 		return true;
650 	}
651 	return false;
652 }
653 
654 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
655 {
656 	bool rc = __sk_nulls_del_node_init_rcu(sk);
657 
658 	if (rc) {
659 		/* paranoid for a while -acme */
660 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
661 		__sock_put(sk);
662 	}
663 	return rc;
664 }
665 
666 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
667 {
668 	hlist_add_head(&sk->sk_node, list);
669 }
670 
671 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
672 {
673 	sock_hold(sk);
674 	__sk_add_node(sk, list);
675 }
676 
677 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
678 {
679 	sock_hold(sk);
680 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
681 	    sk->sk_family == AF_INET6)
682 		hlist_add_tail_rcu(&sk->sk_node, list);
683 	else
684 		hlist_add_head_rcu(&sk->sk_node, list);
685 }
686 
687 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
688 {
689 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
690 }
691 
692 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
693 {
694 	sock_hold(sk);
695 	__sk_nulls_add_node_rcu(sk, list);
696 }
697 
698 static inline void __sk_del_bind_node(struct sock *sk)
699 {
700 	__hlist_del(&sk->sk_bind_node);
701 }
702 
703 static inline void sk_add_bind_node(struct sock *sk,
704 					struct hlist_head *list)
705 {
706 	hlist_add_head(&sk->sk_bind_node, list);
707 }
708 
709 #define sk_for_each(__sk, list) \
710 	hlist_for_each_entry(__sk, list, sk_node)
711 #define sk_for_each_rcu(__sk, list) \
712 	hlist_for_each_entry_rcu(__sk, list, sk_node)
713 #define sk_nulls_for_each(__sk, node, list) \
714 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
715 #define sk_nulls_for_each_rcu(__sk, node, list) \
716 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
717 #define sk_for_each_from(__sk) \
718 	hlist_for_each_entry_from(__sk, sk_node)
719 #define sk_nulls_for_each_from(__sk, node) \
720 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
721 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
722 #define sk_for_each_safe(__sk, tmp, list) \
723 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
724 #define sk_for_each_bound(__sk, list) \
725 	hlist_for_each_entry(__sk, list, sk_bind_node)
726 
727 /**
728  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
729  * @tpos:	the type * to use as a loop cursor.
730  * @pos:	the &struct hlist_node to use as a loop cursor.
731  * @head:	the head for your list.
732  * @offset:	offset of hlist_node within the struct.
733  *
734  */
735 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
736 	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
737 	     pos != NULL &&						       \
738 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
739 	     pos = rcu_dereference(hlist_next_rcu(pos)))
740 
741 static inline struct user_namespace *sk_user_ns(struct sock *sk)
742 {
743 	/* Careful only use this in a context where these parameters
744 	 * can not change and must all be valid, such as recvmsg from
745 	 * userspace.
746 	 */
747 	return sk->sk_socket->file->f_cred->user_ns;
748 }
749 
750 /* Sock flags */
751 enum sock_flags {
752 	SOCK_DEAD,
753 	SOCK_DONE,
754 	SOCK_URGINLINE,
755 	SOCK_KEEPOPEN,
756 	SOCK_LINGER,
757 	SOCK_DESTROY,
758 	SOCK_BROADCAST,
759 	SOCK_TIMESTAMP,
760 	SOCK_ZAPPED,
761 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
762 	SOCK_DBG, /* %SO_DEBUG setting */
763 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
764 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
765 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
766 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
767 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
768 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
769 	SOCK_FASYNC, /* fasync() active */
770 	SOCK_RXQ_OVFL,
771 	SOCK_ZEROCOPY, /* buffers from userspace */
772 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
773 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
774 		     * Will use last 4 bytes of packet sent from
775 		     * user-space instead.
776 		     */
777 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
778 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
779 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
780 };
781 
782 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
783 
784 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
785 {
786 	nsk->sk_flags = osk->sk_flags;
787 }
788 
789 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
790 {
791 	__set_bit(flag, &sk->sk_flags);
792 }
793 
794 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
795 {
796 	__clear_bit(flag, &sk->sk_flags);
797 }
798 
799 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
800 {
801 	return test_bit(flag, &sk->sk_flags);
802 }
803 
804 #ifdef CONFIG_NET
805 extern struct static_key memalloc_socks;
806 static inline int sk_memalloc_socks(void)
807 {
808 	return static_key_false(&memalloc_socks);
809 }
810 #else
811 
812 static inline int sk_memalloc_socks(void)
813 {
814 	return 0;
815 }
816 
817 #endif
818 
819 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
820 {
821 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
822 }
823 
824 static inline void sk_acceptq_removed(struct sock *sk)
825 {
826 	sk->sk_ack_backlog--;
827 }
828 
829 static inline void sk_acceptq_added(struct sock *sk)
830 {
831 	sk->sk_ack_backlog++;
832 }
833 
834 static inline bool sk_acceptq_is_full(const struct sock *sk)
835 {
836 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
837 }
838 
839 /*
840  * Compute minimal free write space needed to queue new packets.
841  */
842 static inline int sk_stream_min_wspace(const struct sock *sk)
843 {
844 	return sk->sk_wmem_queued >> 1;
845 }
846 
847 static inline int sk_stream_wspace(const struct sock *sk)
848 {
849 	return sk->sk_sndbuf - sk->sk_wmem_queued;
850 }
851 
852 void sk_stream_write_space(struct sock *sk);
853 
854 /* OOB backlog add */
855 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
856 {
857 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
858 	skb_dst_force(skb);
859 
860 	if (!sk->sk_backlog.tail)
861 		sk->sk_backlog.head = skb;
862 	else
863 		sk->sk_backlog.tail->next = skb;
864 
865 	sk->sk_backlog.tail = skb;
866 	skb->next = NULL;
867 }
868 
869 /*
870  * Take into account size of receive queue and backlog queue
871  * Do not take into account this skb truesize,
872  * to allow even a single big packet to come.
873  */
874 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
875 {
876 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
877 
878 	return qsize > limit;
879 }
880 
881 /* The per-socket spinlock must be held here. */
882 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
883 					      unsigned int limit)
884 {
885 	if (sk_rcvqueues_full(sk, limit))
886 		return -ENOBUFS;
887 
888 	/*
889 	 * If the skb was allocated from pfmemalloc reserves, only
890 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
891 	 * helping free memory
892 	 */
893 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
894 		return -ENOMEM;
895 
896 	__sk_add_backlog(sk, skb);
897 	sk->sk_backlog.len += skb->truesize;
898 	return 0;
899 }
900 
901 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
902 
903 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
904 {
905 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
906 		return __sk_backlog_rcv(sk, skb);
907 
908 	return sk->sk_backlog_rcv(sk, skb);
909 }
910 
911 static inline void sk_incoming_cpu_update(struct sock *sk)
912 {
913 	int cpu = raw_smp_processor_id();
914 
915 	if (unlikely(sk->sk_incoming_cpu != cpu))
916 		sk->sk_incoming_cpu = cpu;
917 }
918 
919 static inline void sock_rps_record_flow_hash(__u32 hash)
920 {
921 #ifdef CONFIG_RPS
922 	struct rps_sock_flow_table *sock_flow_table;
923 
924 	rcu_read_lock();
925 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
926 	rps_record_sock_flow(sock_flow_table, hash);
927 	rcu_read_unlock();
928 #endif
929 }
930 
931 static inline void sock_rps_record_flow(const struct sock *sk)
932 {
933 #ifdef CONFIG_RPS
934 	if (static_key_false(&rfs_needed)) {
935 		/* Reading sk->sk_rxhash might incur an expensive cache line
936 		 * miss.
937 		 *
938 		 * TCP_ESTABLISHED does cover almost all states where RFS
939 		 * might be useful, and is cheaper [1] than testing :
940 		 *	IPv4: inet_sk(sk)->inet_daddr
941 		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
942 		 * OR	an additional socket flag
943 		 * [1] : sk_state and sk_prot are in the same cache line.
944 		 */
945 		if (sk->sk_state == TCP_ESTABLISHED)
946 			sock_rps_record_flow_hash(sk->sk_rxhash);
947 	}
948 #endif
949 }
950 
951 static inline void sock_rps_save_rxhash(struct sock *sk,
952 					const struct sk_buff *skb)
953 {
954 #ifdef CONFIG_RPS
955 	if (unlikely(sk->sk_rxhash != skb->hash))
956 		sk->sk_rxhash = skb->hash;
957 #endif
958 }
959 
960 static inline void sock_rps_reset_rxhash(struct sock *sk)
961 {
962 #ifdef CONFIG_RPS
963 	sk->sk_rxhash = 0;
964 #endif
965 }
966 
967 #define sk_wait_event(__sk, __timeo, __condition, __wait)		\
968 	({	int __rc;						\
969 		release_sock(__sk);					\
970 		__rc = __condition;					\
971 		if (!__rc) {						\
972 			*(__timeo) = wait_woken(__wait,			\
973 						TASK_INTERRUPTIBLE,	\
974 						*(__timeo));		\
975 		}							\
976 		sched_annotate_sleep();					\
977 		lock_sock(__sk);					\
978 		__rc = __condition;					\
979 		__rc;							\
980 	})
981 
982 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
983 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
984 void sk_stream_wait_close(struct sock *sk, long timeo_p);
985 int sk_stream_error(struct sock *sk, int flags, int err);
986 void sk_stream_kill_queues(struct sock *sk);
987 void sk_set_memalloc(struct sock *sk);
988 void sk_clear_memalloc(struct sock *sk);
989 
990 void __sk_flush_backlog(struct sock *sk);
991 
992 static inline bool sk_flush_backlog(struct sock *sk)
993 {
994 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
995 		__sk_flush_backlog(sk);
996 		return true;
997 	}
998 	return false;
999 }
1000 
1001 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1002 
1003 struct request_sock_ops;
1004 struct timewait_sock_ops;
1005 struct inet_hashinfo;
1006 struct raw_hashinfo;
1007 struct smc_hashinfo;
1008 struct module;
1009 
1010 /*
1011  * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1012  * un-modified. Special care is taken when initializing object to zero.
1013  */
1014 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1015 {
1016 	if (offsetof(struct sock, sk_node.next) != 0)
1017 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1018 	memset(&sk->sk_node.pprev, 0,
1019 	       size - offsetof(struct sock, sk_node.pprev));
1020 }
1021 
1022 /* Networking protocol blocks we attach to sockets.
1023  * socket layer -> transport layer interface
1024  */
1025 struct proto {
1026 	void			(*close)(struct sock *sk,
1027 					long timeout);
1028 	int			(*connect)(struct sock *sk,
1029 					struct sockaddr *uaddr,
1030 					int addr_len);
1031 	int			(*disconnect)(struct sock *sk, int flags);
1032 
1033 	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1034 					  bool kern);
1035 
1036 	int			(*ioctl)(struct sock *sk, int cmd,
1037 					 unsigned long arg);
1038 	int			(*init)(struct sock *sk);
1039 	void			(*destroy)(struct sock *sk);
1040 	void			(*shutdown)(struct sock *sk, int how);
1041 	int			(*setsockopt)(struct sock *sk, int level,
1042 					int optname, char __user *optval,
1043 					unsigned int optlen);
1044 	int			(*getsockopt)(struct sock *sk, int level,
1045 					int optname, char __user *optval,
1046 					int __user *option);
1047 	void			(*keepalive)(struct sock *sk, int valbool);
1048 #ifdef CONFIG_COMPAT
1049 	int			(*compat_setsockopt)(struct sock *sk,
1050 					int level,
1051 					int optname, char __user *optval,
1052 					unsigned int optlen);
1053 	int			(*compat_getsockopt)(struct sock *sk,
1054 					int level,
1055 					int optname, char __user *optval,
1056 					int __user *option);
1057 	int			(*compat_ioctl)(struct sock *sk,
1058 					unsigned int cmd, unsigned long arg);
1059 #endif
1060 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1061 					   size_t len);
1062 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1063 					   size_t len, int noblock, int flags,
1064 					   int *addr_len);
1065 	int			(*sendpage)(struct sock *sk, struct page *page,
1066 					int offset, size_t size, int flags);
1067 	int			(*bind)(struct sock *sk,
1068 					struct sockaddr *uaddr, int addr_len);
1069 
1070 	int			(*backlog_rcv) (struct sock *sk,
1071 						struct sk_buff *skb);
1072 
1073 	void		(*release_cb)(struct sock *sk);
1074 
1075 	/* Keeping track of sk's, looking them up, and port selection methods. */
1076 	int			(*hash)(struct sock *sk);
1077 	void			(*unhash)(struct sock *sk);
1078 	void			(*rehash)(struct sock *sk);
1079 	int			(*get_port)(struct sock *sk, unsigned short snum);
1080 
1081 	/* Keeping track of sockets in use */
1082 #ifdef CONFIG_PROC_FS
1083 	unsigned int		inuse_idx;
1084 #endif
1085 
1086 	bool			(*stream_memory_free)(const struct sock *sk);
1087 	/* Memory pressure */
1088 	void			(*enter_memory_pressure)(struct sock *sk);
1089 	void			(*leave_memory_pressure)(struct sock *sk);
1090 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1091 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1092 	/*
1093 	 * Pressure flag: try to collapse.
1094 	 * Technical note: it is used by multiple contexts non atomically.
1095 	 * All the __sk_mem_schedule() is of this nature: accounting
1096 	 * is strict, actions are advisory and have some latency.
1097 	 */
1098 	unsigned long		*memory_pressure;
1099 	long			*sysctl_mem;
1100 
1101 	int			*sysctl_wmem;
1102 	int			*sysctl_rmem;
1103 	u32			sysctl_wmem_offset;
1104 	u32			sysctl_rmem_offset;
1105 
1106 	int			max_header;
1107 	bool			no_autobind;
1108 
1109 	struct kmem_cache	*slab;
1110 	unsigned int		obj_size;
1111 	slab_flags_t		slab_flags;
1112 	size_t			useroffset;	/* Usercopy region offset */
1113 	size_t			usersize;	/* Usercopy region size */
1114 
1115 	struct percpu_counter	*orphan_count;
1116 
1117 	struct request_sock_ops	*rsk_prot;
1118 	struct timewait_sock_ops *twsk_prot;
1119 
1120 	union {
1121 		struct inet_hashinfo	*hashinfo;
1122 		struct udp_table	*udp_table;
1123 		struct raw_hashinfo	*raw_hash;
1124 		struct smc_hashinfo	*smc_hash;
1125 	} h;
1126 
1127 	struct module		*owner;
1128 
1129 	char			name[32];
1130 
1131 	struct list_head	node;
1132 #ifdef SOCK_REFCNT_DEBUG
1133 	atomic_t		socks;
1134 #endif
1135 	int			(*diag_destroy)(struct sock *sk, int err);
1136 } __randomize_layout;
1137 
1138 int proto_register(struct proto *prot, int alloc_slab);
1139 void proto_unregister(struct proto *prot);
1140 
1141 #ifdef SOCK_REFCNT_DEBUG
1142 static inline void sk_refcnt_debug_inc(struct sock *sk)
1143 {
1144 	atomic_inc(&sk->sk_prot->socks);
1145 }
1146 
1147 static inline void sk_refcnt_debug_dec(struct sock *sk)
1148 {
1149 	atomic_dec(&sk->sk_prot->socks);
1150 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1151 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1152 }
1153 
1154 static inline void sk_refcnt_debug_release(const struct sock *sk)
1155 {
1156 	if (refcount_read(&sk->sk_refcnt) != 1)
1157 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1158 		       sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1159 }
1160 #else /* SOCK_REFCNT_DEBUG */
1161 #define sk_refcnt_debug_inc(sk) do { } while (0)
1162 #define sk_refcnt_debug_dec(sk) do { } while (0)
1163 #define sk_refcnt_debug_release(sk) do { } while (0)
1164 #endif /* SOCK_REFCNT_DEBUG */
1165 
1166 static inline bool sk_stream_memory_free(const struct sock *sk)
1167 {
1168 	if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1169 		return false;
1170 
1171 	return sk->sk_prot->stream_memory_free ?
1172 		sk->sk_prot->stream_memory_free(sk) : true;
1173 }
1174 
1175 static inline bool sk_stream_is_writeable(const struct sock *sk)
1176 {
1177 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1178 	       sk_stream_memory_free(sk);
1179 }
1180 
1181 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1182 					    struct cgroup *ancestor)
1183 {
1184 #ifdef CONFIG_SOCK_CGROUP_DATA
1185 	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1186 				    ancestor);
1187 #else
1188 	return -ENOTSUPP;
1189 #endif
1190 }
1191 
1192 static inline bool sk_has_memory_pressure(const struct sock *sk)
1193 {
1194 	return sk->sk_prot->memory_pressure != NULL;
1195 }
1196 
1197 static inline bool sk_under_memory_pressure(const struct sock *sk)
1198 {
1199 	if (!sk->sk_prot->memory_pressure)
1200 		return false;
1201 
1202 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1203 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1204 		return true;
1205 
1206 	return !!*sk->sk_prot->memory_pressure;
1207 }
1208 
1209 static inline long
1210 sk_memory_allocated(const struct sock *sk)
1211 {
1212 	return atomic_long_read(sk->sk_prot->memory_allocated);
1213 }
1214 
1215 static inline long
1216 sk_memory_allocated_add(struct sock *sk, int amt)
1217 {
1218 	return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1219 }
1220 
1221 static inline void
1222 sk_memory_allocated_sub(struct sock *sk, int amt)
1223 {
1224 	atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1225 }
1226 
1227 static inline void sk_sockets_allocated_dec(struct sock *sk)
1228 {
1229 	percpu_counter_dec(sk->sk_prot->sockets_allocated);
1230 }
1231 
1232 static inline void sk_sockets_allocated_inc(struct sock *sk)
1233 {
1234 	percpu_counter_inc(sk->sk_prot->sockets_allocated);
1235 }
1236 
1237 static inline int
1238 sk_sockets_allocated_read_positive(struct sock *sk)
1239 {
1240 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1241 }
1242 
1243 static inline int
1244 proto_sockets_allocated_sum_positive(struct proto *prot)
1245 {
1246 	return percpu_counter_sum_positive(prot->sockets_allocated);
1247 }
1248 
1249 static inline long
1250 proto_memory_allocated(struct proto *prot)
1251 {
1252 	return atomic_long_read(prot->memory_allocated);
1253 }
1254 
1255 static inline bool
1256 proto_memory_pressure(struct proto *prot)
1257 {
1258 	if (!prot->memory_pressure)
1259 		return false;
1260 	return !!*prot->memory_pressure;
1261 }
1262 
1263 
1264 #ifdef CONFIG_PROC_FS
1265 /* Called with local bh disabled */
1266 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1267 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1268 int sock_inuse_get(struct net *net);
1269 #else
1270 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1271 		int inc)
1272 {
1273 }
1274 #endif
1275 
1276 
1277 /* With per-bucket locks this operation is not-atomic, so that
1278  * this version is not worse.
1279  */
1280 static inline int __sk_prot_rehash(struct sock *sk)
1281 {
1282 	sk->sk_prot->unhash(sk);
1283 	return sk->sk_prot->hash(sk);
1284 }
1285 
1286 /* About 10 seconds */
1287 #define SOCK_DESTROY_TIME (10*HZ)
1288 
1289 /* Sockets 0-1023 can't be bound to unless you are superuser */
1290 #define PROT_SOCK	1024
1291 
1292 #define SHUTDOWN_MASK	3
1293 #define RCV_SHUTDOWN	1
1294 #define SEND_SHUTDOWN	2
1295 
1296 #define SOCK_SNDBUF_LOCK	1
1297 #define SOCK_RCVBUF_LOCK	2
1298 #define SOCK_BINDADDR_LOCK	4
1299 #define SOCK_BINDPORT_LOCK	8
1300 
1301 struct socket_alloc {
1302 	struct socket socket;
1303 	struct inode vfs_inode;
1304 };
1305 
1306 static inline struct socket *SOCKET_I(struct inode *inode)
1307 {
1308 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1309 }
1310 
1311 static inline struct inode *SOCK_INODE(struct socket *socket)
1312 {
1313 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1314 }
1315 
1316 /*
1317  * Functions for memory accounting
1318  */
1319 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1320 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1321 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1322 void __sk_mem_reclaim(struct sock *sk, int amount);
1323 
1324 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1325  * do not necessarily have 16x time more memory than 4KB ones.
1326  */
1327 #define SK_MEM_QUANTUM 4096
1328 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1329 #define SK_MEM_SEND	0
1330 #define SK_MEM_RECV	1
1331 
1332 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1333 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1334 {
1335 	long val = sk->sk_prot->sysctl_mem[index];
1336 
1337 #if PAGE_SIZE > SK_MEM_QUANTUM
1338 	val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1339 #elif PAGE_SIZE < SK_MEM_QUANTUM
1340 	val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1341 #endif
1342 	return val;
1343 }
1344 
1345 static inline int sk_mem_pages(int amt)
1346 {
1347 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1348 }
1349 
1350 static inline bool sk_has_account(struct sock *sk)
1351 {
1352 	/* return true if protocol supports memory accounting */
1353 	return !!sk->sk_prot->memory_allocated;
1354 }
1355 
1356 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1357 {
1358 	if (!sk_has_account(sk))
1359 		return true;
1360 	return size <= sk->sk_forward_alloc ||
1361 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1362 }
1363 
1364 static inline bool
1365 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1366 {
1367 	if (!sk_has_account(sk))
1368 		return true;
1369 	return size<= sk->sk_forward_alloc ||
1370 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1371 		skb_pfmemalloc(skb);
1372 }
1373 
1374 static inline void sk_mem_reclaim(struct sock *sk)
1375 {
1376 	if (!sk_has_account(sk))
1377 		return;
1378 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1379 		__sk_mem_reclaim(sk, sk->sk_forward_alloc);
1380 }
1381 
1382 static inline void sk_mem_reclaim_partial(struct sock *sk)
1383 {
1384 	if (!sk_has_account(sk))
1385 		return;
1386 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1387 		__sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1388 }
1389 
1390 static inline void sk_mem_charge(struct sock *sk, int size)
1391 {
1392 	if (!sk_has_account(sk))
1393 		return;
1394 	sk->sk_forward_alloc -= size;
1395 }
1396 
1397 static inline void sk_mem_uncharge(struct sock *sk, int size)
1398 {
1399 	if (!sk_has_account(sk))
1400 		return;
1401 	sk->sk_forward_alloc += size;
1402 
1403 	/* Avoid a possible overflow.
1404 	 * TCP send queues can make this happen, if sk_mem_reclaim()
1405 	 * is not called and more than 2 GBytes are released at once.
1406 	 *
1407 	 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1408 	 * no need to hold that much forward allocation anyway.
1409 	 */
1410 	if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1411 		__sk_mem_reclaim(sk, 1 << 20);
1412 }
1413 
1414 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1415 {
1416 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1417 	sk->sk_wmem_queued -= skb->truesize;
1418 	sk_mem_uncharge(sk, skb->truesize);
1419 	__kfree_skb(skb);
1420 }
1421 
1422 static inline void sock_release_ownership(struct sock *sk)
1423 {
1424 	if (sk->sk_lock.owned) {
1425 		sk->sk_lock.owned = 0;
1426 
1427 		/* The sk_lock has mutex_unlock() semantics: */
1428 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1429 	}
1430 }
1431 
1432 /*
1433  * Macro so as to not evaluate some arguments when
1434  * lockdep is not enabled.
1435  *
1436  * Mark both the sk_lock and the sk_lock.slock as a
1437  * per-address-family lock class.
1438  */
1439 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1440 do {									\
1441 	sk->sk_lock.owned = 0;						\
1442 	init_waitqueue_head(&sk->sk_lock.wq);				\
1443 	spin_lock_init(&(sk)->sk_lock.slock);				\
1444 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1445 			sizeof((sk)->sk_lock));				\
1446 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1447 				(skey), (sname));				\
1448 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1449 } while (0)
1450 
1451 #ifdef CONFIG_LOCKDEP
1452 static inline bool lockdep_sock_is_held(const struct sock *sk)
1453 {
1454 	return lockdep_is_held(&sk->sk_lock) ||
1455 	       lockdep_is_held(&sk->sk_lock.slock);
1456 }
1457 #endif
1458 
1459 void lock_sock_nested(struct sock *sk, int subclass);
1460 
1461 static inline void lock_sock(struct sock *sk)
1462 {
1463 	lock_sock_nested(sk, 0);
1464 }
1465 
1466 void release_sock(struct sock *sk);
1467 
1468 /* BH context may only use the following locking interface. */
1469 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1470 #define bh_lock_sock_nested(__sk) \
1471 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1472 				SINGLE_DEPTH_NESTING)
1473 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1474 
1475 bool lock_sock_fast(struct sock *sk);
1476 /**
1477  * unlock_sock_fast - complement of lock_sock_fast
1478  * @sk: socket
1479  * @slow: slow mode
1480  *
1481  * fast unlock socket for user context.
1482  * If slow mode is on, we call regular release_sock()
1483  */
1484 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1485 {
1486 	if (slow)
1487 		release_sock(sk);
1488 	else
1489 		spin_unlock_bh(&sk->sk_lock.slock);
1490 }
1491 
1492 /* Used by processes to "lock" a socket state, so that
1493  * interrupts and bottom half handlers won't change it
1494  * from under us. It essentially blocks any incoming
1495  * packets, so that we won't get any new data or any
1496  * packets that change the state of the socket.
1497  *
1498  * While locked, BH processing will add new packets to
1499  * the backlog queue.  This queue is processed by the
1500  * owner of the socket lock right before it is released.
1501  *
1502  * Since ~2.3.5 it is also exclusive sleep lock serializing
1503  * accesses from user process context.
1504  */
1505 
1506 static inline void sock_owned_by_me(const struct sock *sk)
1507 {
1508 #ifdef CONFIG_LOCKDEP
1509 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1510 #endif
1511 }
1512 
1513 static inline bool sock_owned_by_user(const struct sock *sk)
1514 {
1515 	sock_owned_by_me(sk);
1516 	return sk->sk_lock.owned;
1517 }
1518 
1519 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1520 {
1521 	return sk->sk_lock.owned;
1522 }
1523 
1524 /* no reclassification while locks are held */
1525 static inline bool sock_allow_reclassification(const struct sock *csk)
1526 {
1527 	struct sock *sk = (struct sock *)csk;
1528 
1529 	return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1530 }
1531 
1532 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1533 		      struct proto *prot, int kern);
1534 void sk_free(struct sock *sk);
1535 void sk_destruct(struct sock *sk);
1536 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1537 void sk_free_unlock_clone(struct sock *sk);
1538 
1539 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1540 			     gfp_t priority);
1541 void __sock_wfree(struct sk_buff *skb);
1542 void sock_wfree(struct sk_buff *skb);
1543 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1544 			     gfp_t priority);
1545 void skb_orphan_partial(struct sk_buff *skb);
1546 void sock_rfree(struct sk_buff *skb);
1547 void sock_efree(struct sk_buff *skb);
1548 #ifdef CONFIG_INET
1549 void sock_edemux(struct sk_buff *skb);
1550 #else
1551 #define sock_edemux sock_efree
1552 #endif
1553 
1554 int sock_setsockopt(struct socket *sock, int level, int op,
1555 		    char __user *optval, unsigned int optlen);
1556 
1557 int sock_getsockopt(struct socket *sock, int level, int op,
1558 		    char __user *optval, int __user *optlen);
1559 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1560 				    int noblock, int *errcode);
1561 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1562 				     unsigned long data_len, int noblock,
1563 				     int *errcode, int max_page_order);
1564 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1565 void sock_kfree_s(struct sock *sk, void *mem, int size);
1566 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1567 void sk_send_sigurg(struct sock *sk);
1568 
1569 struct sockcm_cookie {
1570 	u32 mark;
1571 	u16 tsflags;
1572 };
1573 
1574 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1575 		     struct sockcm_cookie *sockc);
1576 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1577 		   struct sockcm_cookie *sockc);
1578 
1579 /*
1580  * Functions to fill in entries in struct proto_ops when a protocol
1581  * does not implement a particular function.
1582  */
1583 int sock_no_bind(struct socket *, struct sockaddr *, int);
1584 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1585 int sock_no_socketpair(struct socket *, struct socket *);
1586 int sock_no_accept(struct socket *, struct socket *, int, bool);
1587 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1588 __poll_t sock_no_poll(struct file *, struct socket *,
1589 			  struct poll_table_struct *);
1590 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1591 int sock_no_listen(struct socket *, int);
1592 int sock_no_shutdown(struct socket *, int);
1593 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1594 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1595 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1596 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1597 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1598 int sock_no_mmap(struct file *file, struct socket *sock,
1599 		 struct vm_area_struct *vma);
1600 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1601 			 size_t size, int flags);
1602 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1603 				int offset, size_t size, int flags);
1604 
1605 /*
1606  * Functions to fill in entries in struct proto_ops when a protocol
1607  * uses the inet style.
1608  */
1609 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1610 				  char __user *optval, int __user *optlen);
1611 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1612 			int flags);
1613 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1614 				  char __user *optval, unsigned int optlen);
1615 int compat_sock_common_getsockopt(struct socket *sock, int level,
1616 		int optname, char __user *optval, int __user *optlen);
1617 int compat_sock_common_setsockopt(struct socket *sock, int level,
1618 		int optname, char __user *optval, unsigned int optlen);
1619 
1620 void sk_common_release(struct sock *sk);
1621 
1622 /*
1623  *	Default socket callbacks and setup code
1624  */
1625 
1626 /* Initialise core socket variables */
1627 void sock_init_data(struct socket *sock, struct sock *sk);
1628 
1629 /*
1630  * Socket reference counting postulates.
1631  *
1632  * * Each user of socket SHOULD hold a reference count.
1633  * * Each access point to socket (an hash table bucket, reference from a list,
1634  *   running timer, skb in flight MUST hold a reference count.
1635  * * When reference count hits 0, it means it will never increase back.
1636  * * When reference count hits 0, it means that no references from
1637  *   outside exist to this socket and current process on current CPU
1638  *   is last user and may/should destroy this socket.
1639  * * sk_free is called from any context: process, BH, IRQ. When
1640  *   it is called, socket has no references from outside -> sk_free
1641  *   may release descendant resources allocated by the socket, but
1642  *   to the time when it is called, socket is NOT referenced by any
1643  *   hash tables, lists etc.
1644  * * Packets, delivered from outside (from network or from another process)
1645  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1646  *   when they sit in queue. Otherwise, packets will leak to hole, when
1647  *   socket is looked up by one cpu and unhasing is made by another CPU.
1648  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1649  *   (leak to backlog). Packet socket does all the processing inside
1650  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1651  *   use separate SMP lock, so that they are prone too.
1652  */
1653 
1654 /* Ungrab socket and destroy it, if it was the last reference. */
1655 static inline void sock_put(struct sock *sk)
1656 {
1657 	if (refcount_dec_and_test(&sk->sk_refcnt))
1658 		sk_free(sk);
1659 }
1660 /* Generic version of sock_put(), dealing with all sockets
1661  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1662  */
1663 void sock_gen_put(struct sock *sk);
1664 
1665 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1666 		     unsigned int trim_cap, bool refcounted);
1667 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1668 				 const int nested)
1669 {
1670 	return __sk_receive_skb(sk, skb, nested, 1, true);
1671 }
1672 
1673 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1674 {
1675 	sk->sk_tx_queue_mapping = tx_queue;
1676 }
1677 
1678 static inline void sk_tx_queue_clear(struct sock *sk)
1679 {
1680 	sk->sk_tx_queue_mapping = -1;
1681 }
1682 
1683 static inline int sk_tx_queue_get(const struct sock *sk)
1684 {
1685 	return sk ? sk->sk_tx_queue_mapping : -1;
1686 }
1687 
1688 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1689 {
1690 	sk_tx_queue_clear(sk);
1691 	sk->sk_socket = sock;
1692 }
1693 
1694 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1695 {
1696 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1697 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1698 }
1699 /* Detach socket from process context.
1700  * Announce socket dead, detach it from wait queue and inode.
1701  * Note that parent inode held reference count on this struct sock,
1702  * we do not release it in this function, because protocol
1703  * probably wants some additional cleanups or even continuing
1704  * to work with this socket (TCP).
1705  */
1706 static inline void sock_orphan(struct sock *sk)
1707 {
1708 	write_lock_bh(&sk->sk_callback_lock);
1709 	sock_set_flag(sk, SOCK_DEAD);
1710 	sk_set_socket(sk, NULL);
1711 	sk->sk_wq  = NULL;
1712 	write_unlock_bh(&sk->sk_callback_lock);
1713 }
1714 
1715 static inline void sock_graft(struct sock *sk, struct socket *parent)
1716 {
1717 	WARN_ON(parent->sk);
1718 	write_lock_bh(&sk->sk_callback_lock);
1719 	sk->sk_wq = parent->wq;
1720 	parent->sk = sk;
1721 	sk_set_socket(sk, parent);
1722 	sk->sk_uid = SOCK_INODE(parent)->i_uid;
1723 	security_sock_graft(sk, parent);
1724 	write_unlock_bh(&sk->sk_callback_lock);
1725 }
1726 
1727 kuid_t sock_i_uid(struct sock *sk);
1728 unsigned long sock_i_ino(struct sock *sk);
1729 
1730 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1731 {
1732 	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1733 }
1734 
1735 static inline u32 net_tx_rndhash(void)
1736 {
1737 	u32 v = prandom_u32();
1738 
1739 	return v ?: 1;
1740 }
1741 
1742 static inline void sk_set_txhash(struct sock *sk)
1743 {
1744 	sk->sk_txhash = net_tx_rndhash();
1745 }
1746 
1747 static inline void sk_rethink_txhash(struct sock *sk)
1748 {
1749 	if (sk->sk_txhash)
1750 		sk_set_txhash(sk);
1751 }
1752 
1753 static inline struct dst_entry *
1754 __sk_dst_get(struct sock *sk)
1755 {
1756 	return rcu_dereference_check(sk->sk_dst_cache,
1757 				     lockdep_sock_is_held(sk));
1758 }
1759 
1760 static inline struct dst_entry *
1761 sk_dst_get(struct sock *sk)
1762 {
1763 	struct dst_entry *dst;
1764 
1765 	rcu_read_lock();
1766 	dst = rcu_dereference(sk->sk_dst_cache);
1767 	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1768 		dst = NULL;
1769 	rcu_read_unlock();
1770 	return dst;
1771 }
1772 
1773 static inline void dst_negative_advice(struct sock *sk)
1774 {
1775 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1776 
1777 	sk_rethink_txhash(sk);
1778 
1779 	if (dst && dst->ops->negative_advice) {
1780 		ndst = dst->ops->negative_advice(dst);
1781 
1782 		if (ndst != dst) {
1783 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1784 			sk_tx_queue_clear(sk);
1785 			sk->sk_dst_pending_confirm = 0;
1786 		}
1787 	}
1788 }
1789 
1790 static inline void
1791 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1792 {
1793 	struct dst_entry *old_dst;
1794 
1795 	sk_tx_queue_clear(sk);
1796 	sk->sk_dst_pending_confirm = 0;
1797 	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1798 					    lockdep_sock_is_held(sk));
1799 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1800 	dst_release(old_dst);
1801 }
1802 
1803 static inline void
1804 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1805 {
1806 	struct dst_entry *old_dst;
1807 
1808 	sk_tx_queue_clear(sk);
1809 	sk->sk_dst_pending_confirm = 0;
1810 	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1811 	dst_release(old_dst);
1812 }
1813 
1814 static inline void
1815 __sk_dst_reset(struct sock *sk)
1816 {
1817 	__sk_dst_set(sk, NULL);
1818 }
1819 
1820 static inline void
1821 sk_dst_reset(struct sock *sk)
1822 {
1823 	sk_dst_set(sk, NULL);
1824 }
1825 
1826 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1827 
1828 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1829 
1830 static inline void sk_dst_confirm(struct sock *sk)
1831 {
1832 	if (!sk->sk_dst_pending_confirm)
1833 		sk->sk_dst_pending_confirm = 1;
1834 }
1835 
1836 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1837 {
1838 	if (skb_get_dst_pending_confirm(skb)) {
1839 		struct sock *sk = skb->sk;
1840 		unsigned long now = jiffies;
1841 
1842 		/* avoid dirtying neighbour */
1843 		if (n->confirmed != now)
1844 			n->confirmed = now;
1845 		if (sk && sk->sk_dst_pending_confirm)
1846 			sk->sk_dst_pending_confirm = 0;
1847 	}
1848 }
1849 
1850 bool sk_mc_loop(struct sock *sk);
1851 
1852 static inline bool sk_can_gso(const struct sock *sk)
1853 {
1854 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1855 }
1856 
1857 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1858 
1859 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1860 {
1861 	sk->sk_route_nocaps |= flags;
1862 	sk->sk_route_caps &= ~flags;
1863 }
1864 
1865 static inline bool sk_check_csum_caps(struct sock *sk)
1866 {
1867 	return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1868 	       (sk->sk_family == PF_INET &&
1869 		(sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1870 	       (sk->sk_family == PF_INET6 &&
1871 		(sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1872 }
1873 
1874 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1875 					   struct iov_iter *from, char *to,
1876 					   int copy, int offset)
1877 {
1878 	if (skb->ip_summed == CHECKSUM_NONE) {
1879 		__wsum csum = 0;
1880 		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1881 			return -EFAULT;
1882 		skb->csum = csum_block_add(skb->csum, csum, offset);
1883 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1884 		if (!copy_from_iter_full_nocache(to, copy, from))
1885 			return -EFAULT;
1886 	} else if (!copy_from_iter_full(to, copy, from))
1887 		return -EFAULT;
1888 
1889 	return 0;
1890 }
1891 
1892 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1893 				       struct iov_iter *from, int copy)
1894 {
1895 	int err, offset = skb->len;
1896 
1897 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1898 				       copy, offset);
1899 	if (err)
1900 		__skb_trim(skb, offset);
1901 
1902 	return err;
1903 }
1904 
1905 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1906 					   struct sk_buff *skb,
1907 					   struct page *page,
1908 					   int off, int copy)
1909 {
1910 	int err;
1911 
1912 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1913 				       copy, skb->len);
1914 	if (err)
1915 		return err;
1916 
1917 	skb->len	     += copy;
1918 	skb->data_len	     += copy;
1919 	skb->truesize	     += copy;
1920 	sk->sk_wmem_queued   += copy;
1921 	sk_mem_charge(sk, copy);
1922 	return 0;
1923 }
1924 
1925 /**
1926  * sk_wmem_alloc_get - returns write allocations
1927  * @sk: socket
1928  *
1929  * Returns sk_wmem_alloc minus initial offset of one
1930  */
1931 static inline int sk_wmem_alloc_get(const struct sock *sk)
1932 {
1933 	return refcount_read(&sk->sk_wmem_alloc) - 1;
1934 }
1935 
1936 /**
1937  * sk_rmem_alloc_get - returns read allocations
1938  * @sk: socket
1939  *
1940  * Returns sk_rmem_alloc
1941  */
1942 static inline int sk_rmem_alloc_get(const struct sock *sk)
1943 {
1944 	return atomic_read(&sk->sk_rmem_alloc);
1945 }
1946 
1947 /**
1948  * sk_has_allocations - check if allocations are outstanding
1949  * @sk: socket
1950  *
1951  * Returns true if socket has write or read allocations
1952  */
1953 static inline bool sk_has_allocations(const struct sock *sk)
1954 {
1955 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1956 }
1957 
1958 /**
1959  * skwq_has_sleeper - check if there are any waiting processes
1960  * @wq: struct socket_wq
1961  *
1962  * Returns true if socket_wq has waiting processes
1963  *
1964  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1965  * barrier call. They were added due to the race found within the tcp code.
1966  *
1967  * Consider following tcp code paths::
1968  *
1969  *   CPU1                CPU2
1970  *   sys_select          receive packet
1971  *   ...                 ...
1972  *   __add_wait_queue    update tp->rcv_nxt
1973  *   ...                 ...
1974  *   tp->rcv_nxt check   sock_def_readable
1975  *   ...                 {
1976  *   schedule               rcu_read_lock();
1977  *                          wq = rcu_dereference(sk->sk_wq);
1978  *                          if (wq && waitqueue_active(&wq->wait))
1979  *                              wake_up_interruptible(&wq->wait)
1980  *                          ...
1981  *                       }
1982  *
1983  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1984  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1985  * could then endup calling schedule and sleep forever if there are no more
1986  * data on the socket.
1987  *
1988  */
1989 static inline bool skwq_has_sleeper(struct socket_wq *wq)
1990 {
1991 	return wq && wq_has_sleeper(&wq->wait);
1992 }
1993 
1994 /**
1995  * sock_poll_wait - place memory barrier behind the poll_wait call.
1996  * @filp:           file
1997  * @wait_address:   socket wait queue
1998  * @p:              poll_table
1999  *
2000  * See the comments in the wq_has_sleeper function.
2001  */
2002 static inline void sock_poll_wait(struct file *filp,
2003 		wait_queue_head_t *wait_address, poll_table *p)
2004 {
2005 	if (!poll_does_not_wait(p) && wait_address) {
2006 		poll_wait(filp, wait_address, p);
2007 		/* We need to be sure we are in sync with the
2008 		 * socket flags modification.
2009 		 *
2010 		 * This memory barrier is paired in the wq_has_sleeper.
2011 		 */
2012 		smp_mb();
2013 	}
2014 }
2015 
2016 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2017 {
2018 	if (sk->sk_txhash) {
2019 		skb->l4_hash = 1;
2020 		skb->hash = sk->sk_txhash;
2021 	}
2022 }
2023 
2024 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2025 
2026 /*
2027  *	Queue a received datagram if it will fit. Stream and sequenced
2028  *	protocols can't normally use this as they need to fit buffers in
2029  *	and play with them.
2030  *
2031  *	Inlined as it's very short and called for pretty much every
2032  *	packet ever received.
2033  */
2034 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2035 {
2036 	skb_orphan(skb);
2037 	skb->sk = sk;
2038 	skb->destructor = sock_rfree;
2039 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2040 	sk_mem_charge(sk, skb->truesize);
2041 }
2042 
2043 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2044 		    unsigned long expires);
2045 
2046 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2047 
2048 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2049 			struct sk_buff *skb, unsigned int flags,
2050 			void (*destructor)(struct sock *sk,
2051 					   struct sk_buff *skb));
2052 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2053 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2054 
2055 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2056 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2057 
2058 /*
2059  *	Recover an error report and clear atomically
2060  */
2061 
2062 static inline int sock_error(struct sock *sk)
2063 {
2064 	int err;
2065 	if (likely(!sk->sk_err))
2066 		return 0;
2067 	err = xchg(&sk->sk_err, 0);
2068 	return -err;
2069 }
2070 
2071 static inline unsigned long sock_wspace(struct sock *sk)
2072 {
2073 	int amt = 0;
2074 
2075 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2076 		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2077 		if (amt < 0)
2078 			amt = 0;
2079 	}
2080 	return amt;
2081 }
2082 
2083 /* Note:
2084  *  We use sk->sk_wq_raw, from contexts knowing this
2085  *  pointer is not NULL and cannot disappear/change.
2086  */
2087 static inline void sk_set_bit(int nr, struct sock *sk)
2088 {
2089 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2090 	    !sock_flag(sk, SOCK_FASYNC))
2091 		return;
2092 
2093 	set_bit(nr, &sk->sk_wq_raw->flags);
2094 }
2095 
2096 static inline void sk_clear_bit(int nr, struct sock *sk)
2097 {
2098 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2099 	    !sock_flag(sk, SOCK_FASYNC))
2100 		return;
2101 
2102 	clear_bit(nr, &sk->sk_wq_raw->flags);
2103 }
2104 
2105 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2106 {
2107 	if (sock_flag(sk, SOCK_FASYNC)) {
2108 		rcu_read_lock();
2109 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2110 		rcu_read_unlock();
2111 	}
2112 }
2113 
2114 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2115  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2116  * Note: for send buffers, TCP works better if we can build two skbs at
2117  * minimum.
2118  */
2119 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2120 
2121 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2122 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2123 
2124 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2125 {
2126 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2127 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2128 		sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2129 	}
2130 }
2131 
2132 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2133 				    bool force_schedule);
2134 
2135 /**
2136  * sk_page_frag - return an appropriate page_frag
2137  * @sk: socket
2138  *
2139  * If socket allocation mode allows current thread to sleep, it means its
2140  * safe to use the per task page_frag instead of the per socket one.
2141  */
2142 static inline struct page_frag *sk_page_frag(struct sock *sk)
2143 {
2144 	if (gfpflags_allow_blocking(sk->sk_allocation))
2145 		return &current->task_frag;
2146 
2147 	return &sk->sk_frag;
2148 }
2149 
2150 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2151 
2152 /*
2153  *	Default write policy as shown to user space via poll/select/SIGIO
2154  */
2155 static inline bool sock_writeable(const struct sock *sk)
2156 {
2157 	return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2158 }
2159 
2160 static inline gfp_t gfp_any(void)
2161 {
2162 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2163 }
2164 
2165 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2166 {
2167 	return noblock ? 0 : sk->sk_rcvtimeo;
2168 }
2169 
2170 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2171 {
2172 	return noblock ? 0 : sk->sk_sndtimeo;
2173 }
2174 
2175 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2176 {
2177 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2178 }
2179 
2180 /* Alas, with timeout socket operations are not restartable.
2181  * Compare this to poll().
2182  */
2183 static inline int sock_intr_errno(long timeo)
2184 {
2185 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2186 }
2187 
2188 struct sock_skb_cb {
2189 	u32 dropcount;
2190 };
2191 
2192 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2193  * using skb->cb[] would keep using it directly and utilize its
2194  * alignement guarantee.
2195  */
2196 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2197 			    sizeof(struct sock_skb_cb)))
2198 
2199 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2200 			    SOCK_SKB_CB_OFFSET))
2201 
2202 #define sock_skb_cb_check_size(size) \
2203 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2204 
2205 static inline void
2206 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2207 {
2208 	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2209 						atomic_read(&sk->sk_drops) : 0;
2210 }
2211 
2212 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2213 {
2214 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2215 
2216 	atomic_add(segs, &sk->sk_drops);
2217 }
2218 
2219 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2220 			   struct sk_buff *skb);
2221 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2222 			     struct sk_buff *skb);
2223 
2224 static inline void
2225 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2226 {
2227 	ktime_t kt = skb->tstamp;
2228 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2229 
2230 	/*
2231 	 * generate control messages if
2232 	 * - receive time stamping in software requested
2233 	 * - software time stamp available and wanted
2234 	 * - hardware time stamps available and wanted
2235 	 */
2236 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2237 	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2238 	    (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2239 	    (hwtstamps->hwtstamp &&
2240 	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2241 		__sock_recv_timestamp(msg, sk, skb);
2242 	else
2243 		sk->sk_stamp = kt;
2244 
2245 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2246 		__sock_recv_wifi_status(msg, sk, skb);
2247 }
2248 
2249 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2250 			      struct sk_buff *skb);
2251 
2252 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2253 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2254 					  struct sk_buff *skb)
2255 {
2256 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2257 			   (1UL << SOCK_RCVTSTAMP))
2258 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2259 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2260 
2261 	if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2262 		__sock_recv_ts_and_drops(msg, sk, skb);
2263 	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2264 		sk->sk_stamp = skb->tstamp;
2265 	else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2266 		sk->sk_stamp = 0;
2267 }
2268 
2269 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2270 
2271 /**
2272  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2273  * @sk:		socket sending this packet
2274  * @tsflags:	timestamping flags to use
2275  * @tx_flags:	completed with instructions for time stamping
2276  *
2277  * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2278  */
2279 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2280 				     __u8 *tx_flags)
2281 {
2282 	if (unlikely(tsflags))
2283 		__sock_tx_timestamp(tsflags, tx_flags);
2284 	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2285 		*tx_flags |= SKBTX_WIFI_STATUS;
2286 }
2287 
2288 /**
2289  * sk_eat_skb - Release a skb if it is no longer needed
2290  * @sk: socket to eat this skb from
2291  * @skb: socket buffer to eat
2292  *
2293  * This routine must be called with interrupts disabled or with the socket
2294  * locked so that the sk_buff queue operation is ok.
2295 */
2296 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2297 {
2298 	__skb_unlink(skb, &sk->sk_receive_queue);
2299 	__kfree_skb(skb);
2300 }
2301 
2302 static inline
2303 struct net *sock_net(const struct sock *sk)
2304 {
2305 	return read_pnet(&sk->sk_net);
2306 }
2307 
2308 static inline
2309 void sock_net_set(struct sock *sk, struct net *net)
2310 {
2311 	write_pnet(&sk->sk_net, net);
2312 }
2313 
2314 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2315 {
2316 	if (skb->sk) {
2317 		struct sock *sk = skb->sk;
2318 
2319 		skb->destructor = NULL;
2320 		skb->sk = NULL;
2321 		return sk;
2322 	}
2323 	return NULL;
2324 }
2325 
2326 /* This helper checks if a socket is a full socket,
2327  * ie _not_ a timewait or request socket.
2328  */
2329 static inline bool sk_fullsock(const struct sock *sk)
2330 {
2331 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2332 }
2333 
2334 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2335  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2336  */
2337 static inline bool sk_listener(const struct sock *sk)
2338 {
2339 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2340 }
2341 
2342 void sock_enable_timestamp(struct sock *sk, int flag);
2343 int sock_get_timestamp(struct sock *, struct timeval __user *);
2344 int sock_get_timestampns(struct sock *, struct timespec __user *);
2345 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2346 		       int type);
2347 
2348 bool sk_ns_capable(const struct sock *sk,
2349 		   struct user_namespace *user_ns, int cap);
2350 bool sk_capable(const struct sock *sk, int cap);
2351 bool sk_net_capable(const struct sock *sk, int cap);
2352 
2353 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2354 
2355 /* Take into consideration the size of the struct sk_buff overhead in the
2356  * determination of these values, since that is non-constant across
2357  * platforms.  This makes socket queueing behavior and performance
2358  * not depend upon such differences.
2359  */
2360 #define _SK_MEM_PACKETS		256
2361 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2362 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2363 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2364 
2365 extern __u32 sysctl_wmem_max;
2366 extern __u32 sysctl_rmem_max;
2367 
2368 extern int sysctl_tstamp_allow_data;
2369 extern int sysctl_optmem_max;
2370 
2371 extern __u32 sysctl_wmem_default;
2372 extern __u32 sysctl_rmem_default;
2373 
2374 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2375 {
2376 	/* Does this proto have per netns sysctl_wmem ? */
2377 	if (proto->sysctl_wmem_offset)
2378 		return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2379 
2380 	return *proto->sysctl_wmem;
2381 }
2382 
2383 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2384 {
2385 	/* Does this proto have per netns sysctl_rmem ? */
2386 	if (proto->sysctl_rmem_offset)
2387 		return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2388 
2389 	return *proto->sysctl_rmem;
2390 }
2391 
2392 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2393  * Some wifi drivers need to tweak it to get more chunks.
2394  * They can use this helper from their ndo_start_xmit()
2395  */
2396 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2397 {
2398 	if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2399 		return;
2400 	sk->sk_pacing_shift = val;
2401 }
2402 
2403 /* if a socket is bound to a device, check that the given device
2404  * index is either the same or that the socket is bound to an L3
2405  * master device and the given device index is also enslaved to
2406  * that L3 master
2407  */
2408 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2409 {
2410 	int mdif;
2411 
2412 	if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2413 		return true;
2414 
2415 	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2416 	if (mdif && mdif == sk->sk_bound_dev_if)
2417 		return true;
2418 
2419 	return false;
2420 }
2421 
2422 #endif	/* _SOCK_H */
2423