xref: /linux/include/net/sock.h (revision cd371e0959a3f2d5df69d50000750f7eefc94659)
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 
62 #include <linux/filter.h>
63 #include <linux/rculist_nulls.h>
64 #include <linux/poll.h>
65 
66 #include <linux/atomic.h>
67 #include <net/dst.h>
68 #include <net/checksum.h>
69 #include <net/tcp_states.h>
70 #include <linux/net_tstamp.h>
71 
72 struct cgroup;
73 struct cgroup_subsys;
74 #ifdef CONFIG_NET
75 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
76 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
77 #else
78 static inline
79 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
80 {
81 	return 0;
82 }
83 static inline
84 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
85 {
86 }
87 #endif
88 /*
89  * This structure really needs to be cleaned up.
90  * Most of it is for TCP, and not used by any of
91  * the other protocols.
92  */
93 
94 /* Define this to get the SOCK_DBG debugging facility. */
95 #define SOCK_DEBUGGING
96 #ifdef SOCK_DEBUGGING
97 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
98 					printk(KERN_DEBUG msg); } while (0)
99 #else
100 /* Validate arguments and do nothing */
101 static inline __printf(2, 3)
102 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
103 {
104 }
105 #endif
106 
107 /* This is the per-socket lock.  The spinlock provides a synchronization
108  * between user contexts and software interrupt processing, whereas the
109  * mini-semaphore synchronizes multiple users amongst themselves.
110  */
111 typedef struct {
112 	spinlock_t		slock;
113 	int			owned;
114 	wait_queue_head_t	wq;
115 	/*
116 	 * We express the mutex-alike socket_lock semantics
117 	 * to the lock validator by explicitly managing
118 	 * the slock as a lock variant (in addition to
119 	 * the slock itself):
120 	 */
121 #ifdef CONFIG_DEBUG_LOCK_ALLOC
122 	struct lockdep_map dep_map;
123 #endif
124 } socket_lock_t;
125 
126 struct sock;
127 struct proto;
128 struct net;
129 
130 typedef __u32 __bitwise __portpair;
131 typedef __u64 __bitwise __addrpair;
132 
133 /**
134  *	struct sock_common - minimal network layer representation of sockets
135  *	@skc_daddr: Foreign IPv4 addr
136  *	@skc_rcv_saddr: Bound local IPv4 addr
137  *	@skc_hash: hash value used with various protocol lookup tables
138  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
139  *	@skc_dport: placeholder for inet_dport/tw_dport
140  *	@skc_num: placeholder for inet_num/tw_num
141  *	@skc_family: network address family
142  *	@skc_state: Connection state
143  *	@skc_reuse: %SO_REUSEADDR setting
144  *	@skc_reuseport: %SO_REUSEPORT setting
145  *	@skc_bound_dev_if: bound device index if != 0
146  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
147  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
148  *	@skc_prot: protocol handlers inside a network family
149  *	@skc_net: reference to the network namespace of this socket
150  *	@skc_node: main hash linkage for various protocol lookup tables
151  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
152  *	@skc_tx_queue_mapping: tx queue number for this connection
153  *	@skc_flags: place holder for sk_flags
154  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
155  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
156  *	@skc_incoming_cpu: record/match cpu processing incoming packets
157  *	@skc_refcnt: reference count
158  *
159  *	This is the minimal network layer representation of sockets, the header
160  *	for struct sock and struct inet_timewait_sock.
161  */
162 struct sock_common {
163 	/* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
164 	 * address on 64bit arches : cf INET_MATCH()
165 	 */
166 	union {
167 		__addrpair	skc_addrpair;
168 		struct {
169 			__be32	skc_daddr;
170 			__be32	skc_rcv_saddr;
171 		};
172 	};
173 	union  {
174 		unsigned int	skc_hash;
175 		__u16		skc_u16hashes[2];
176 	};
177 	/* skc_dport && skc_num must be grouped as well */
178 	union {
179 		__portpair	skc_portpair;
180 		struct {
181 			__be16	skc_dport;
182 			__u16	skc_num;
183 		};
184 	};
185 
186 	unsigned short		skc_family;
187 	volatile unsigned char	skc_state;
188 	unsigned char		skc_reuse:4;
189 	unsigned char		skc_reuseport:1;
190 	unsigned char		skc_ipv6only:1;
191 	unsigned char		skc_net_refcnt:1;
192 	int			skc_bound_dev_if;
193 	union {
194 		struct hlist_node	skc_bind_node;
195 		struct hlist_nulls_node skc_portaddr_node;
196 	};
197 	struct proto		*skc_prot;
198 	possible_net_t		skc_net;
199 
200 #if IS_ENABLED(CONFIG_IPV6)
201 	struct in6_addr		skc_v6_daddr;
202 	struct in6_addr		skc_v6_rcv_saddr;
203 #endif
204 
205 	atomic64_t		skc_cookie;
206 
207 	/* following fields are padding to force
208 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
209 	 * assuming IPV6 is enabled. We use this padding differently
210 	 * for different kind of 'sockets'
211 	 */
212 	union {
213 		unsigned long	skc_flags;
214 		struct sock	*skc_listener; /* request_sock */
215 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
216 	};
217 	/*
218 	 * fields between dontcopy_begin/dontcopy_end
219 	 * are not copied in sock_copy()
220 	 */
221 	/* private: */
222 	int			skc_dontcopy_begin[0];
223 	/* public: */
224 	union {
225 		struct hlist_node	skc_node;
226 		struct hlist_nulls_node skc_nulls_node;
227 	};
228 	int			skc_tx_queue_mapping;
229 	union {
230 		int		skc_incoming_cpu;
231 		u32		skc_rcv_wnd;
232 		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
233 	};
234 
235 	atomic_t		skc_refcnt;
236 	/* private: */
237 	int                     skc_dontcopy_end[0];
238 	union {
239 		u32		skc_rxhash;
240 		u32		skc_window_clamp;
241 		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
242 	};
243 	/* public: */
244 };
245 
246 struct cg_proto;
247 /**
248   *	struct sock - network layer representation of sockets
249   *	@__sk_common: shared layout with inet_timewait_sock
250   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
251   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
252   *	@sk_lock:	synchronizer
253   *	@sk_rcvbuf: size of receive buffer in bytes
254   *	@sk_wq: sock wait queue and async head
255   *	@sk_rx_dst: receive input route used by early demux
256   *	@sk_dst_cache: destination cache
257   *	@sk_policy: flow policy
258   *	@sk_receive_queue: incoming packets
259   *	@sk_wmem_alloc: transmit queue bytes committed
260   *	@sk_write_queue: Packet sending queue
261   *	@sk_omem_alloc: "o" is "option" or "other"
262   *	@sk_wmem_queued: persistent queue size
263   *	@sk_forward_alloc: space allocated forward
264   *	@sk_napi_id: id of the last napi context to receive data for sk
265   *	@sk_ll_usec: usecs to busypoll when there is no data
266   *	@sk_allocation: allocation mode
267   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
268   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
269   *	@sk_sndbuf: size of send buffer in bytes
270   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
271   *	@sk_no_check_rx: allow zero checksum in RX packets
272   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
273   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
274   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
275   *	@sk_gso_max_size: Maximum GSO segment size to build
276   *	@sk_gso_max_segs: Maximum number of GSO segments
277   *	@sk_lingertime: %SO_LINGER l_linger setting
278   *	@sk_backlog: always used with the per-socket spinlock held
279   *	@sk_callback_lock: used with the callbacks in the end of this struct
280   *	@sk_error_queue: rarely used
281   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
282   *			  IPV6_ADDRFORM for instance)
283   *	@sk_err: last error
284   *	@sk_err_soft: errors that don't cause failure but are the cause of a
285   *		      persistent failure not just 'timed out'
286   *	@sk_drops: raw/udp drops counter
287   *	@sk_ack_backlog: current listen backlog
288   *	@sk_max_ack_backlog: listen backlog set in listen()
289   *	@sk_priority: %SO_PRIORITY setting
290   *	@sk_cgrp_prioidx: socket group's priority map index
291   *	@sk_type: socket type (%SOCK_STREAM, etc)
292   *	@sk_protocol: which protocol this socket belongs in this network family
293   *	@sk_peer_pid: &struct pid for this socket's peer
294   *	@sk_peer_cred: %SO_PEERCRED setting
295   *	@sk_rcvlowat: %SO_RCVLOWAT setting
296   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
297   *	@sk_sndtimeo: %SO_SNDTIMEO setting
298   *	@sk_txhash: computed flow hash for use on transmit
299   *	@sk_filter: socket filtering instructions
300   *	@sk_timer: sock cleanup timer
301   *	@sk_stamp: time stamp of last packet received
302   *	@sk_tsflags: SO_TIMESTAMPING socket options
303   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
304   *	@sk_socket: Identd and reporting IO signals
305   *	@sk_user_data: RPC layer private data
306   *	@sk_frag: cached page frag
307   *	@sk_peek_off: current peek_offset value
308   *	@sk_send_head: front of stuff to transmit
309   *	@sk_security: used by security modules
310   *	@sk_mark: generic packet mark
311   *	@sk_classid: this socket's cgroup classid
312   *	@sk_cgrp: this socket's cgroup-specific proto data
313   *	@sk_write_pending: a write to stream socket waits to start
314   *	@sk_state_change: callback to indicate change in the state of the sock
315   *	@sk_data_ready: callback to indicate there is data to be processed
316   *	@sk_write_space: callback to indicate there is bf sending space available
317   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
318   *	@sk_backlog_rcv: callback to process the backlog
319   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
320  */
321 struct sock {
322 	/*
323 	 * Now struct inet_timewait_sock also uses sock_common, so please just
324 	 * don't add nothing before this first member (__sk_common) --acme
325 	 */
326 	struct sock_common	__sk_common;
327 #define sk_node			__sk_common.skc_node
328 #define sk_nulls_node		__sk_common.skc_nulls_node
329 #define sk_refcnt		__sk_common.skc_refcnt
330 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
331 
332 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
333 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
334 #define sk_hash			__sk_common.skc_hash
335 #define sk_portpair		__sk_common.skc_portpair
336 #define sk_num			__sk_common.skc_num
337 #define sk_dport		__sk_common.skc_dport
338 #define sk_addrpair		__sk_common.skc_addrpair
339 #define sk_daddr		__sk_common.skc_daddr
340 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
341 #define sk_family		__sk_common.skc_family
342 #define sk_state		__sk_common.skc_state
343 #define sk_reuse		__sk_common.skc_reuse
344 #define sk_reuseport		__sk_common.skc_reuseport
345 #define sk_ipv6only		__sk_common.skc_ipv6only
346 #define sk_net_refcnt		__sk_common.skc_net_refcnt
347 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
348 #define sk_bind_node		__sk_common.skc_bind_node
349 #define sk_prot			__sk_common.skc_prot
350 #define sk_net			__sk_common.skc_net
351 #define sk_v6_daddr		__sk_common.skc_v6_daddr
352 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
353 #define sk_cookie		__sk_common.skc_cookie
354 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
355 #define sk_flags		__sk_common.skc_flags
356 #define sk_rxhash		__sk_common.skc_rxhash
357 
358 	socket_lock_t		sk_lock;
359 	struct sk_buff_head	sk_receive_queue;
360 	/*
361 	 * The backlog queue is special, it is always used with
362 	 * the per-socket spinlock held and requires low latency
363 	 * access. Therefore we special case it's implementation.
364 	 * Note : rmem_alloc is in this structure to fill a hole
365 	 * on 64bit arches, not because its logically part of
366 	 * backlog.
367 	 */
368 	struct {
369 		atomic_t	rmem_alloc;
370 		int		len;
371 		struct sk_buff	*head;
372 		struct sk_buff	*tail;
373 	} sk_backlog;
374 #define sk_rmem_alloc sk_backlog.rmem_alloc
375 	int			sk_forward_alloc;
376 
377 	__u32			sk_txhash;
378 #ifdef CONFIG_NET_RX_BUSY_POLL
379 	unsigned int		sk_napi_id;
380 	unsigned int		sk_ll_usec;
381 #endif
382 	atomic_t		sk_drops;
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	*sk_policy[2];
392 #endif
393 	struct dst_entry	*sk_rx_dst;
394 	struct dst_entry __rcu	*sk_dst_cache;
395 	/* Note: 32bit hole on 64bit arches */
396 	atomic_t		sk_wmem_alloc;
397 	atomic_t		sk_omem_alloc;
398 	int			sk_sndbuf;
399 	struct sk_buff_head	sk_write_queue;
400 	kmemcheck_bitfield_begin(flags);
401 	unsigned int		sk_shutdown  : 2,
402 				sk_no_check_tx : 1,
403 				sk_no_check_rx : 1,
404 				sk_userlocks : 4,
405 				sk_protocol  : 8,
406 				sk_type      : 16;
407 	kmemcheck_bitfield_end(flags);
408 	int			sk_wmem_queued;
409 	gfp_t			sk_allocation;
410 	u32			sk_pacing_rate; /* bytes per second */
411 	u32			sk_max_pacing_rate;
412 	netdev_features_t	sk_route_caps;
413 	netdev_features_t	sk_route_nocaps;
414 	int			sk_gso_type;
415 	unsigned int		sk_gso_max_size;
416 	u16			sk_gso_max_segs;
417 	int			sk_rcvlowat;
418 	unsigned long	        sk_lingertime;
419 	struct sk_buff_head	sk_error_queue;
420 	struct proto		*sk_prot_creator;
421 	rwlock_t		sk_callback_lock;
422 	int			sk_err,
423 				sk_err_soft;
424 	u32			sk_ack_backlog;
425 	u32			sk_max_ack_backlog;
426 	__u32			sk_priority;
427 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
428 	__u32			sk_cgrp_prioidx;
429 #endif
430 	struct pid		*sk_peer_pid;
431 	const struct cred	*sk_peer_cred;
432 	long			sk_rcvtimeo;
433 	long			sk_sndtimeo;
434 	struct timer_list	sk_timer;
435 	ktime_t			sk_stamp;
436 	u16			sk_tsflags;
437 	u32			sk_tskey;
438 	struct socket		*sk_socket;
439 	void			*sk_user_data;
440 	struct page_frag	sk_frag;
441 	struct sk_buff		*sk_send_head;
442 	__s32			sk_peek_off;
443 	int			sk_write_pending;
444 #ifdef CONFIG_SECURITY
445 	void			*sk_security;
446 #endif
447 	__u32			sk_mark;
448 #ifdef CONFIG_CGROUP_NET_CLASSID
449 	u32			sk_classid;
450 #endif
451 	struct cg_proto		*sk_cgrp;
452 	void			(*sk_state_change)(struct sock *sk);
453 	void			(*sk_data_ready)(struct sock *sk);
454 	void			(*sk_write_space)(struct sock *sk);
455 	void			(*sk_error_report)(struct sock *sk);
456 	int			(*sk_backlog_rcv)(struct sock *sk,
457 						  struct sk_buff *skb);
458 	void                    (*sk_destruct)(struct sock *sk);
459 };
460 
461 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
462 
463 #define rcu_dereference_sk_user_data(sk)	rcu_dereference(__sk_user_data((sk)))
464 #define rcu_assign_sk_user_data(sk, ptr)	rcu_assign_pointer(__sk_user_data((sk)), ptr)
465 
466 /*
467  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
468  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
469  * on a socket means that the socket will reuse everybody else's port
470  * without looking at the other's sk_reuse value.
471  */
472 
473 #define SK_NO_REUSE	0
474 #define SK_CAN_REUSE	1
475 #define SK_FORCE_REUSE	2
476 
477 static inline int sk_peek_offset(struct sock *sk, int flags)
478 {
479 	if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
480 		return sk->sk_peek_off;
481 	else
482 		return 0;
483 }
484 
485 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
486 {
487 	if (sk->sk_peek_off >= 0) {
488 		if (sk->sk_peek_off >= val)
489 			sk->sk_peek_off -= val;
490 		else
491 			sk->sk_peek_off = 0;
492 	}
493 }
494 
495 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
496 {
497 	if (sk->sk_peek_off >= 0)
498 		sk->sk_peek_off += val;
499 }
500 
501 /*
502  * Hashed lists helper routines
503  */
504 static inline struct sock *sk_entry(const struct hlist_node *node)
505 {
506 	return hlist_entry(node, struct sock, sk_node);
507 }
508 
509 static inline struct sock *__sk_head(const struct hlist_head *head)
510 {
511 	return hlist_entry(head->first, struct sock, sk_node);
512 }
513 
514 static inline struct sock *sk_head(const struct hlist_head *head)
515 {
516 	return hlist_empty(head) ? NULL : __sk_head(head);
517 }
518 
519 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
520 {
521 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
522 }
523 
524 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
525 {
526 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
527 }
528 
529 static inline struct sock *sk_next(const struct sock *sk)
530 {
531 	return sk->sk_node.next ?
532 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
533 }
534 
535 static inline struct sock *sk_nulls_next(const struct sock *sk)
536 {
537 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
538 		hlist_nulls_entry(sk->sk_nulls_node.next,
539 				  struct sock, sk_nulls_node) :
540 		NULL;
541 }
542 
543 static inline bool sk_unhashed(const struct sock *sk)
544 {
545 	return hlist_unhashed(&sk->sk_node);
546 }
547 
548 static inline bool sk_hashed(const struct sock *sk)
549 {
550 	return !sk_unhashed(sk);
551 }
552 
553 static inline void sk_node_init(struct hlist_node *node)
554 {
555 	node->pprev = NULL;
556 }
557 
558 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
559 {
560 	node->pprev = NULL;
561 }
562 
563 static inline void __sk_del_node(struct sock *sk)
564 {
565 	__hlist_del(&sk->sk_node);
566 }
567 
568 /* NB: equivalent to hlist_del_init_rcu */
569 static inline bool __sk_del_node_init(struct sock *sk)
570 {
571 	if (sk_hashed(sk)) {
572 		__sk_del_node(sk);
573 		sk_node_init(&sk->sk_node);
574 		return true;
575 	}
576 	return false;
577 }
578 
579 /* Grab socket reference count. This operation is valid only
580    when sk is ALREADY grabbed f.e. it is found in hash table
581    or a list and the lookup is made under lock preventing hash table
582    modifications.
583  */
584 
585 static inline void sock_hold(struct sock *sk)
586 {
587 	atomic_inc(&sk->sk_refcnt);
588 }
589 
590 /* Ungrab socket in the context, which assumes that socket refcnt
591    cannot hit zero, f.e. it is true in context of any socketcall.
592  */
593 static inline void __sock_put(struct sock *sk)
594 {
595 	atomic_dec(&sk->sk_refcnt);
596 }
597 
598 static inline bool sk_del_node_init(struct sock *sk)
599 {
600 	bool rc = __sk_del_node_init(sk);
601 
602 	if (rc) {
603 		/* paranoid for a while -acme */
604 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
605 		__sock_put(sk);
606 	}
607 	return rc;
608 }
609 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
610 
611 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
612 {
613 	if (sk_hashed(sk)) {
614 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
615 		return true;
616 	}
617 	return false;
618 }
619 
620 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
621 {
622 	bool rc = __sk_nulls_del_node_init_rcu(sk);
623 
624 	if (rc) {
625 		/* paranoid for a while -acme */
626 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
627 		__sock_put(sk);
628 	}
629 	return rc;
630 }
631 
632 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
633 {
634 	hlist_add_head(&sk->sk_node, list);
635 }
636 
637 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
638 {
639 	sock_hold(sk);
640 	__sk_add_node(sk, list);
641 }
642 
643 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
644 {
645 	sock_hold(sk);
646 	hlist_add_head_rcu(&sk->sk_node, list);
647 }
648 
649 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
650 {
651 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
652 }
653 
654 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
655 {
656 	sock_hold(sk);
657 	__sk_nulls_add_node_rcu(sk, list);
658 }
659 
660 static inline void __sk_del_bind_node(struct sock *sk)
661 {
662 	__hlist_del(&sk->sk_bind_node);
663 }
664 
665 static inline void sk_add_bind_node(struct sock *sk,
666 					struct hlist_head *list)
667 {
668 	hlist_add_head(&sk->sk_bind_node, list);
669 }
670 
671 #define sk_for_each(__sk, list) \
672 	hlist_for_each_entry(__sk, list, sk_node)
673 #define sk_for_each_rcu(__sk, list) \
674 	hlist_for_each_entry_rcu(__sk, list, sk_node)
675 #define sk_nulls_for_each(__sk, node, list) \
676 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
677 #define sk_nulls_for_each_rcu(__sk, node, list) \
678 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
679 #define sk_for_each_from(__sk) \
680 	hlist_for_each_entry_from(__sk, sk_node)
681 #define sk_nulls_for_each_from(__sk, node) \
682 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
683 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
684 #define sk_for_each_safe(__sk, tmp, list) \
685 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
686 #define sk_for_each_bound(__sk, list) \
687 	hlist_for_each_entry(__sk, list, sk_bind_node)
688 
689 /**
690  * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
691  * @tpos:	the type * to use as a loop cursor.
692  * @pos:	the &struct hlist_node to use as a loop cursor.
693  * @head:	the head for your list.
694  * @offset:	offset of hlist_node within the struct.
695  *
696  */
697 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset)		       \
698 	for (pos = (head)->first;					       \
699 	     (!is_a_nulls(pos)) &&					       \
700 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
701 	     pos = pos->next)
702 
703 static inline struct user_namespace *sk_user_ns(struct sock *sk)
704 {
705 	/* Careful only use this in a context where these parameters
706 	 * can not change and must all be valid, such as recvmsg from
707 	 * userspace.
708 	 */
709 	return sk->sk_socket->file->f_cred->user_ns;
710 }
711 
712 /* Sock flags */
713 enum sock_flags {
714 	SOCK_DEAD,
715 	SOCK_DONE,
716 	SOCK_URGINLINE,
717 	SOCK_KEEPOPEN,
718 	SOCK_LINGER,
719 	SOCK_DESTROY,
720 	SOCK_BROADCAST,
721 	SOCK_TIMESTAMP,
722 	SOCK_ZAPPED,
723 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
724 	SOCK_DBG, /* %SO_DEBUG setting */
725 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
726 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
727 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
728 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
729 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
730 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
731 	SOCK_FASYNC, /* fasync() active */
732 	SOCK_RXQ_OVFL,
733 	SOCK_ZEROCOPY, /* buffers from userspace */
734 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
735 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
736 		     * Will use last 4 bytes of packet sent from
737 		     * user-space instead.
738 		     */
739 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
740 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
741 };
742 
743 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
744 {
745 	nsk->sk_flags = osk->sk_flags;
746 }
747 
748 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
749 {
750 	__set_bit(flag, &sk->sk_flags);
751 }
752 
753 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
754 {
755 	__clear_bit(flag, &sk->sk_flags);
756 }
757 
758 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
759 {
760 	return test_bit(flag, &sk->sk_flags);
761 }
762 
763 #ifdef CONFIG_NET
764 extern struct static_key memalloc_socks;
765 static inline int sk_memalloc_socks(void)
766 {
767 	return static_key_false(&memalloc_socks);
768 }
769 #else
770 
771 static inline int sk_memalloc_socks(void)
772 {
773 	return 0;
774 }
775 
776 #endif
777 
778 static inline gfp_t sk_gfp_atomic(const struct sock *sk, gfp_t gfp_mask)
779 {
780 	return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
781 }
782 
783 static inline void sk_acceptq_removed(struct sock *sk)
784 {
785 	sk->sk_ack_backlog--;
786 }
787 
788 static inline void sk_acceptq_added(struct sock *sk)
789 {
790 	sk->sk_ack_backlog++;
791 }
792 
793 static inline bool sk_acceptq_is_full(const struct sock *sk)
794 {
795 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
796 }
797 
798 /*
799  * Compute minimal free write space needed to queue new packets.
800  */
801 static inline int sk_stream_min_wspace(const struct sock *sk)
802 {
803 	return sk->sk_wmem_queued >> 1;
804 }
805 
806 static inline int sk_stream_wspace(const struct sock *sk)
807 {
808 	return sk->sk_sndbuf - sk->sk_wmem_queued;
809 }
810 
811 void sk_stream_write_space(struct sock *sk);
812 
813 /* OOB backlog add */
814 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
815 {
816 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
817 	skb_dst_force(skb);
818 
819 	if (!sk->sk_backlog.tail)
820 		sk->sk_backlog.head = skb;
821 	else
822 		sk->sk_backlog.tail->next = skb;
823 
824 	sk->sk_backlog.tail = skb;
825 	skb->next = NULL;
826 }
827 
828 /*
829  * Take into account size of receive queue and backlog queue
830  * Do not take into account this skb truesize,
831  * to allow even a single big packet to come.
832  */
833 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
834 {
835 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
836 
837 	return qsize > limit;
838 }
839 
840 /* The per-socket spinlock must be held here. */
841 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
842 					      unsigned int limit)
843 {
844 	if (sk_rcvqueues_full(sk, limit))
845 		return -ENOBUFS;
846 
847 	/*
848 	 * If the skb was allocated from pfmemalloc reserves, only
849 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
850 	 * helping free memory
851 	 */
852 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
853 		return -ENOMEM;
854 
855 	__sk_add_backlog(sk, skb);
856 	sk->sk_backlog.len += skb->truesize;
857 	return 0;
858 }
859 
860 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
861 
862 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
863 {
864 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
865 		return __sk_backlog_rcv(sk, skb);
866 
867 	return sk->sk_backlog_rcv(sk, skb);
868 }
869 
870 static inline void sk_incoming_cpu_update(struct sock *sk)
871 {
872 	sk->sk_incoming_cpu = raw_smp_processor_id();
873 }
874 
875 static inline void sock_rps_record_flow_hash(__u32 hash)
876 {
877 #ifdef CONFIG_RPS
878 	struct rps_sock_flow_table *sock_flow_table;
879 
880 	rcu_read_lock();
881 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
882 	rps_record_sock_flow(sock_flow_table, hash);
883 	rcu_read_unlock();
884 #endif
885 }
886 
887 static inline void sock_rps_record_flow(const struct sock *sk)
888 {
889 #ifdef CONFIG_RPS
890 	sock_rps_record_flow_hash(sk->sk_rxhash);
891 #endif
892 }
893 
894 static inline void sock_rps_save_rxhash(struct sock *sk,
895 					const struct sk_buff *skb)
896 {
897 #ifdef CONFIG_RPS
898 	if (unlikely(sk->sk_rxhash != skb->hash))
899 		sk->sk_rxhash = skb->hash;
900 #endif
901 }
902 
903 static inline void sock_rps_reset_rxhash(struct sock *sk)
904 {
905 #ifdef CONFIG_RPS
906 	sk->sk_rxhash = 0;
907 #endif
908 }
909 
910 #define sk_wait_event(__sk, __timeo, __condition)			\
911 	({	int __rc;						\
912 		release_sock(__sk);					\
913 		__rc = __condition;					\
914 		if (!__rc) {						\
915 			*(__timeo) = schedule_timeout(*(__timeo));	\
916 		}							\
917 		sched_annotate_sleep();						\
918 		lock_sock(__sk);					\
919 		__rc = __condition;					\
920 		__rc;							\
921 	})
922 
923 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
924 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
925 void sk_stream_wait_close(struct sock *sk, long timeo_p);
926 int sk_stream_error(struct sock *sk, int flags, int err);
927 void sk_stream_kill_queues(struct sock *sk);
928 void sk_set_memalloc(struct sock *sk);
929 void sk_clear_memalloc(struct sock *sk);
930 
931 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
932 
933 struct request_sock_ops;
934 struct timewait_sock_ops;
935 struct inet_hashinfo;
936 struct raw_hashinfo;
937 struct module;
938 
939 /*
940  * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
941  * un-modified. Special care is taken when initializing object to zero.
942  */
943 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
944 {
945 	if (offsetof(struct sock, sk_node.next) != 0)
946 		memset(sk, 0, offsetof(struct sock, sk_node.next));
947 	memset(&sk->sk_node.pprev, 0,
948 	       size - offsetof(struct sock, sk_node.pprev));
949 }
950 
951 /* Networking protocol blocks we attach to sockets.
952  * socket layer -> transport layer interface
953  */
954 struct proto {
955 	void			(*close)(struct sock *sk,
956 					long timeout);
957 	int			(*connect)(struct sock *sk,
958 					struct sockaddr *uaddr,
959 					int addr_len);
960 	int			(*disconnect)(struct sock *sk, int flags);
961 
962 	struct sock *		(*accept)(struct sock *sk, int flags, int *err);
963 
964 	int			(*ioctl)(struct sock *sk, int cmd,
965 					 unsigned long arg);
966 	int			(*init)(struct sock *sk);
967 	void			(*destroy)(struct sock *sk);
968 	void			(*shutdown)(struct sock *sk, int how);
969 	int			(*setsockopt)(struct sock *sk, int level,
970 					int optname, char __user *optval,
971 					unsigned int optlen);
972 	int			(*getsockopt)(struct sock *sk, int level,
973 					int optname, char __user *optval,
974 					int __user *option);
975 #ifdef CONFIG_COMPAT
976 	int			(*compat_setsockopt)(struct sock *sk,
977 					int level,
978 					int optname, char __user *optval,
979 					unsigned int optlen);
980 	int			(*compat_getsockopt)(struct sock *sk,
981 					int level,
982 					int optname, char __user *optval,
983 					int __user *option);
984 	int			(*compat_ioctl)(struct sock *sk,
985 					unsigned int cmd, unsigned long arg);
986 #endif
987 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
988 					   size_t len);
989 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
990 					   size_t len, int noblock, int flags,
991 					   int *addr_len);
992 	int			(*sendpage)(struct sock *sk, struct page *page,
993 					int offset, size_t size, int flags);
994 	int			(*bind)(struct sock *sk,
995 					struct sockaddr *uaddr, int addr_len);
996 
997 	int			(*backlog_rcv) (struct sock *sk,
998 						struct sk_buff *skb);
999 
1000 	void		(*release_cb)(struct sock *sk);
1001 
1002 	/* Keeping track of sk's, looking them up, and port selection methods. */
1003 	void			(*hash)(struct sock *sk);
1004 	void			(*unhash)(struct sock *sk);
1005 	void			(*rehash)(struct sock *sk);
1006 	int			(*get_port)(struct sock *sk, unsigned short snum);
1007 	void			(*clear_sk)(struct sock *sk, int size);
1008 
1009 	/* Keeping track of sockets in use */
1010 #ifdef CONFIG_PROC_FS
1011 	unsigned int		inuse_idx;
1012 #endif
1013 
1014 	bool			(*stream_memory_free)(const struct sock *sk);
1015 	/* Memory pressure */
1016 	void			(*enter_memory_pressure)(struct sock *sk);
1017 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1018 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1019 	/*
1020 	 * Pressure flag: try to collapse.
1021 	 * Technical note: it is used by multiple contexts non atomically.
1022 	 * All the __sk_mem_schedule() is of this nature: accounting
1023 	 * is strict, actions are advisory and have some latency.
1024 	 */
1025 	int			*memory_pressure;
1026 	long			*sysctl_mem;
1027 	int			*sysctl_wmem;
1028 	int			*sysctl_rmem;
1029 	int			max_header;
1030 	bool			no_autobind;
1031 
1032 	struct kmem_cache	*slab;
1033 	unsigned int		obj_size;
1034 	int			slab_flags;
1035 
1036 	struct percpu_counter	*orphan_count;
1037 
1038 	struct request_sock_ops	*rsk_prot;
1039 	struct timewait_sock_ops *twsk_prot;
1040 
1041 	union {
1042 		struct inet_hashinfo	*hashinfo;
1043 		struct udp_table	*udp_table;
1044 		struct raw_hashinfo	*raw_hash;
1045 	} h;
1046 
1047 	struct module		*owner;
1048 
1049 	char			name[32];
1050 
1051 	struct list_head	node;
1052 #ifdef SOCK_REFCNT_DEBUG
1053 	atomic_t		socks;
1054 #endif
1055 #ifdef CONFIG_MEMCG_KMEM
1056 	/*
1057 	 * cgroup specific init/deinit functions. Called once for all
1058 	 * protocols that implement it, from cgroups populate function.
1059 	 * This function has to setup any files the protocol want to
1060 	 * appear in the kmem cgroup filesystem.
1061 	 */
1062 	int			(*init_cgroup)(struct mem_cgroup *memcg,
1063 					       struct cgroup_subsys *ss);
1064 	void			(*destroy_cgroup)(struct mem_cgroup *memcg);
1065 	struct cg_proto		*(*proto_cgroup)(struct mem_cgroup *memcg);
1066 #endif
1067 };
1068 
1069 int proto_register(struct proto *prot, int alloc_slab);
1070 void proto_unregister(struct proto *prot);
1071 
1072 #ifdef SOCK_REFCNT_DEBUG
1073 static inline void sk_refcnt_debug_inc(struct sock *sk)
1074 {
1075 	atomic_inc(&sk->sk_prot->socks);
1076 }
1077 
1078 static inline void sk_refcnt_debug_dec(struct sock *sk)
1079 {
1080 	atomic_dec(&sk->sk_prot->socks);
1081 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1082 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1083 }
1084 
1085 static inline void sk_refcnt_debug_release(const struct sock *sk)
1086 {
1087 	if (atomic_read(&sk->sk_refcnt) != 1)
1088 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1089 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1090 }
1091 #else /* SOCK_REFCNT_DEBUG */
1092 #define sk_refcnt_debug_inc(sk) do { } while (0)
1093 #define sk_refcnt_debug_dec(sk) do { } while (0)
1094 #define sk_refcnt_debug_release(sk) do { } while (0)
1095 #endif /* SOCK_REFCNT_DEBUG */
1096 
1097 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1098 extern struct static_key memcg_socket_limit_enabled;
1099 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1100 					       struct cg_proto *cg_proto)
1101 {
1102 	return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1103 }
1104 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1105 #else
1106 #define mem_cgroup_sockets_enabled 0
1107 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1108 					       struct cg_proto *cg_proto)
1109 {
1110 	return NULL;
1111 }
1112 #endif
1113 
1114 static inline bool sk_stream_memory_free(const struct sock *sk)
1115 {
1116 	if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1117 		return false;
1118 
1119 	return sk->sk_prot->stream_memory_free ?
1120 		sk->sk_prot->stream_memory_free(sk) : true;
1121 }
1122 
1123 static inline bool sk_stream_is_writeable(const struct sock *sk)
1124 {
1125 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1126 	       sk_stream_memory_free(sk);
1127 }
1128 
1129 
1130 static inline bool sk_has_memory_pressure(const struct sock *sk)
1131 {
1132 	return sk->sk_prot->memory_pressure != NULL;
1133 }
1134 
1135 static inline bool sk_under_memory_pressure(const struct sock *sk)
1136 {
1137 	if (!sk->sk_prot->memory_pressure)
1138 		return false;
1139 
1140 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1141 		return !!sk->sk_cgrp->memory_pressure;
1142 
1143 	return !!*sk->sk_prot->memory_pressure;
1144 }
1145 
1146 static inline void sk_leave_memory_pressure(struct sock *sk)
1147 {
1148 	int *memory_pressure = sk->sk_prot->memory_pressure;
1149 
1150 	if (!memory_pressure)
1151 		return;
1152 
1153 	if (*memory_pressure)
1154 		*memory_pressure = 0;
1155 
1156 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1157 		struct cg_proto *cg_proto = sk->sk_cgrp;
1158 		struct proto *prot = sk->sk_prot;
1159 
1160 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1161 			cg_proto->memory_pressure = 0;
1162 	}
1163 
1164 }
1165 
1166 static inline void sk_enter_memory_pressure(struct sock *sk)
1167 {
1168 	if (!sk->sk_prot->enter_memory_pressure)
1169 		return;
1170 
1171 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1172 		struct cg_proto *cg_proto = sk->sk_cgrp;
1173 		struct proto *prot = sk->sk_prot;
1174 
1175 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1176 			cg_proto->memory_pressure = 1;
1177 	}
1178 
1179 	sk->sk_prot->enter_memory_pressure(sk);
1180 }
1181 
1182 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1183 {
1184 	long *prot = sk->sk_prot->sysctl_mem;
1185 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1186 		prot = sk->sk_cgrp->sysctl_mem;
1187 	return prot[index];
1188 }
1189 
1190 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1191 					      unsigned long amt,
1192 					      int *parent_status)
1193 {
1194 	page_counter_charge(&prot->memory_allocated, amt);
1195 
1196 	if (page_counter_read(&prot->memory_allocated) >
1197 	    prot->memory_allocated.limit)
1198 		*parent_status = OVER_LIMIT;
1199 }
1200 
1201 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1202 					      unsigned long amt)
1203 {
1204 	page_counter_uncharge(&prot->memory_allocated, amt);
1205 }
1206 
1207 static inline long
1208 sk_memory_allocated(const struct sock *sk)
1209 {
1210 	struct proto *prot = sk->sk_prot;
1211 
1212 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1213 		return page_counter_read(&sk->sk_cgrp->memory_allocated);
1214 
1215 	return atomic_long_read(prot->memory_allocated);
1216 }
1217 
1218 static inline long
1219 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1220 {
1221 	struct proto *prot = sk->sk_prot;
1222 
1223 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1224 		memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1225 		/* update the root cgroup regardless */
1226 		atomic_long_add_return(amt, prot->memory_allocated);
1227 		return page_counter_read(&sk->sk_cgrp->memory_allocated);
1228 	}
1229 
1230 	return atomic_long_add_return(amt, prot->memory_allocated);
1231 }
1232 
1233 static inline void
1234 sk_memory_allocated_sub(struct sock *sk, int amt)
1235 {
1236 	struct proto *prot = sk->sk_prot;
1237 
1238 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1239 		memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1240 
1241 	atomic_long_sub(amt, prot->memory_allocated);
1242 }
1243 
1244 static inline void sk_sockets_allocated_dec(struct sock *sk)
1245 {
1246 	struct proto *prot = sk->sk_prot;
1247 
1248 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1249 		struct cg_proto *cg_proto = sk->sk_cgrp;
1250 
1251 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1252 			percpu_counter_dec(&cg_proto->sockets_allocated);
1253 	}
1254 
1255 	percpu_counter_dec(prot->sockets_allocated);
1256 }
1257 
1258 static inline void sk_sockets_allocated_inc(struct sock *sk)
1259 {
1260 	struct proto *prot = sk->sk_prot;
1261 
1262 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1263 		struct cg_proto *cg_proto = sk->sk_cgrp;
1264 
1265 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1266 			percpu_counter_inc(&cg_proto->sockets_allocated);
1267 	}
1268 
1269 	percpu_counter_inc(prot->sockets_allocated);
1270 }
1271 
1272 static inline int
1273 sk_sockets_allocated_read_positive(struct sock *sk)
1274 {
1275 	struct proto *prot = sk->sk_prot;
1276 
1277 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1278 		return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1279 
1280 	return percpu_counter_read_positive(prot->sockets_allocated);
1281 }
1282 
1283 static inline int
1284 proto_sockets_allocated_sum_positive(struct proto *prot)
1285 {
1286 	return percpu_counter_sum_positive(prot->sockets_allocated);
1287 }
1288 
1289 static inline long
1290 proto_memory_allocated(struct proto *prot)
1291 {
1292 	return atomic_long_read(prot->memory_allocated);
1293 }
1294 
1295 static inline bool
1296 proto_memory_pressure(struct proto *prot)
1297 {
1298 	if (!prot->memory_pressure)
1299 		return false;
1300 	return !!*prot->memory_pressure;
1301 }
1302 
1303 
1304 #ifdef CONFIG_PROC_FS
1305 /* Called with local bh disabled */
1306 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1307 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1308 #else
1309 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1310 		int inc)
1311 {
1312 }
1313 #endif
1314 
1315 
1316 /* With per-bucket locks this operation is not-atomic, so that
1317  * this version is not worse.
1318  */
1319 static inline void __sk_prot_rehash(struct sock *sk)
1320 {
1321 	sk->sk_prot->unhash(sk);
1322 	sk->sk_prot->hash(sk);
1323 }
1324 
1325 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1326 
1327 /* About 10 seconds */
1328 #define SOCK_DESTROY_TIME (10*HZ)
1329 
1330 /* Sockets 0-1023 can't be bound to unless you are superuser */
1331 #define PROT_SOCK	1024
1332 
1333 #define SHUTDOWN_MASK	3
1334 #define RCV_SHUTDOWN	1
1335 #define SEND_SHUTDOWN	2
1336 
1337 #define SOCK_SNDBUF_LOCK	1
1338 #define SOCK_RCVBUF_LOCK	2
1339 #define SOCK_BINDADDR_LOCK	4
1340 #define SOCK_BINDPORT_LOCK	8
1341 
1342 struct socket_alloc {
1343 	struct socket socket;
1344 	struct inode vfs_inode;
1345 };
1346 
1347 static inline struct socket *SOCKET_I(struct inode *inode)
1348 {
1349 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1350 }
1351 
1352 static inline struct inode *SOCK_INODE(struct socket *socket)
1353 {
1354 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1355 }
1356 
1357 /*
1358  * Functions for memory accounting
1359  */
1360 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1361 void __sk_mem_reclaim(struct sock *sk, int amount);
1362 
1363 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1364 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1365 #define SK_MEM_SEND	0
1366 #define SK_MEM_RECV	1
1367 
1368 static inline int sk_mem_pages(int amt)
1369 {
1370 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1371 }
1372 
1373 static inline bool sk_has_account(struct sock *sk)
1374 {
1375 	/* return true if protocol supports memory accounting */
1376 	return !!sk->sk_prot->memory_allocated;
1377 }
1378 
1379 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1380 {
1381 	if (!sk_has_account(sk))
1382 		return true;
1383 	return size <= sk->sk_forward_alloc ||
1384 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1385 }
1386 
1387 static inline bool
1388 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1389 {
1390 	if (!sk_has_account(sk))
1391 		return true;
1392 	return size<= sk->sk_forward_alloc ||
1393 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1394 		skb_pfmemalloc(skb);
1395 }
1396 
1397 static inline void sk_mem_reclaim(struct sock *sk)
1398 {
1399 	if (!sk_has_account(sk))
1400 		return;
1401 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1402 		__sk_mem_reclaim(sk, sk->sk_forward_alloc);
1403 }
1404 
1405 static inline void sk_mem_reclaim_partial(struct sock *sk)
1406 {
1407 	if (!sk_has_account(sk))
1408 		return;
1409 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1410 		__sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1411 }
1412 
1413 static inline void sk_mem_charge(struct sock *sk, int size)
1414 {
1415 	if (!sk_has_account(sk))
1416 		return;
1417 	sk->sk_forward_alloc -= size;
1418 }
1419 
1420 static inline void sk_mem_uncharge(struct sock *sk, int size)
1421 {
1422 	if (!sk_has_account(sk))
1423 		return;
1424 	sk->sk_forward_alloc += size;
1425 }
1426 
1427 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1428 {
1429 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1430 	sk->sk_wmem_queued -= skb->truesize;
1431 	sk_mem_uncharge(sk, skb->truesize);
1432 	__kfree_skb(skb);
1433 }
1434 
1435 /* Used by processes to "lock" a socket state, so that
1436  * interrupts and bottom half handlers won't change it
1437  * from under us. It essentially blocks any incoming
1438  * packets, so that we won't get any new data or any
1439  * packets that change the state of the socket.
1440  *
1441  * While locked, BH processing will add new packets to
1442  * the backlog queue.  This queue is processed by the
1443  * owner of the socket lock right before it is released.
1444  *
1445  * Since ~2.3.5 it is also exclusive sleep lock serializing
1446  * accesses from user process context.
1447  */
1448 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1449 
1450 static inline void sock_release_ownership(struct sock *sk)
1451 {
1452 	sk->sk_lock.owned = 0;
1453 }
1454 
1455 /*
1456  * Macro so as to not evaluate some arguments when
1457  * lockdep is not enabled.
1458  *
1459  * Mark both the sk_lock and the sk_lock.slock as a
1460  * per-address-family lock class.
1461  */
1462 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1463 do {									\
1464 	sk->sk_lock.owned = 0;						\
1465 	init_waitqueue_head(&sk->sk_lock.wq);				\
1466 	spin_lock_init(&(sk)->sk_lock.slock);				\
1467 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1468 			sizeof((sk)->sk_lock));				\
1469 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1470 				(skey), (sname));				\
1471 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1472 } while (0)
1473 
1474 void lock_sock_nested(struct sock *sk, int subclass);
1475 
1476 static inline void lock_sock(struct sock *sk)
1477 {
1478 	lock_sock_nested(sk, 0);
1479 }
1480 
1481 void release_sock(struct sock *sk);
1482 
1483 /* BH context may only use the following locking interface. */
1484 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1485 #define bh_lock_sock_nested(__sk) \
1486 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1487 				SINGLE_DEPTH_NESTING)
1488 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1489 
1490 bool lock_sock_fast(struct sock *sk);
1491 /**
1492  * unlock_sock_fast - complement of lock_sock_fast
1493  * @sk: socket
1494  * @slow: slow mode
1495  *
1496  * fast unlock socket for user context.
1497  * If slow mode is on, we call regular release_sock()
1498  */
1499 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1500 {
1501 	if (slow)
1502 		release_sock(sk);
1503 	else
1504 		spin_unlock_bh(&sk->sk_lock.slock);
1505 }
1506 
1507 
1508 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1509 		      struct proto *prot, int kern);
1510 void sk_free(struct sock *sk);
1511 void sk_destruct(struct sock *sk);
1512 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1513 
1514 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1515 			     gfp_t priority);
1516 void sock_wfree(struct sk_buff *skb);
1517 void skb_orphan_partial(struct sk_buff *skb);
1518 void sock_rfree(struct sk_buff *skb);
1519 void sock_efree(struct sk_buff *skb);
1520 #ifdef CONFIG_INET
1521 void sock_edemux(struct sk_buff *skb);
1522 #else
1523 #define sock_edemux(skb) sock_efree(skb)
1524 #endif
1525 
1526 int sock_setsockopt(struct socket *sock, int level, int op,
1527 		    char __user *optval, unsigned int optlen);
1528 
1529 int sock_getsockopt(struct socket *sock, int level, int op,
1530 		    char __user *optval, int __user *optlen);
1531 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1532 				    int noblock, int *errcode);
1533 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1534 				     unsigned long data_len, int noblock,
1535 				     int *errcode, int max_page_order);
1536 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1537 void sock_kfree_s(struct sock *sk, void *mem, int size);
1538 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1539 void sk_send_sigurg(struct sock *sk);
1540 
1541 struct sockcm_cookie {
1542 	u32 mark;
1543 };
1544 
1545 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1546 		   struct sockcm_cookie *sockc);
1547 
1548 /*
1549  * Functions to fill in entries in struct proto_ops when a protocol
1550  * does not implement a particular function.
1551  */
1552 int sock_no_bind(struct socket *, struct sockaddr *, int);
1553 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1554 int sock_no_socketpair(struct socket *, struct socket *);
1555 int sock_no_accept(struct socket *, struct socket *, int);
1556 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1557 unsigned int sock_no_poll(struct file *, struct socket *,
1558 			  struct poll_table_struct *);
1559 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1560 int sock_no_listen(struct socket *, int);
1561 int sock_no_shutdown(struct socket *, int);
1562 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1563 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1564 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1565 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1566 int sock_no_mmap(struct file *file, struct socket *sock,
1567 		 struct vm_area_struct *vma);
1568 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1569 			 size_t size, int flags);
1570 
1571 /*
1572  * Functions to fill in entries in struct proto_ops when a protocol
1573  * uses the inet style.
1574  */
1575 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1576 				  char __user *optval, int __user *optlen);
1577 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1578 			int flags);
1579 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1580 				  char __user *optval, unsigned int optlen);
1581 int compat_sock_common_getsockopt(struct socket *sock, int level,
1582 		int optname, char __user *optval, int __user *optlen);
1583 int compat_sock_common_setsockopt(struct socket *sock, int level,
1584 		int optname, char __user *optval, unsigned int optlen);
1585 
1586 void sk_common_release(struct sock *sk);
1587 
1588 /*
1589  *	Default socket callbacks and setup code
1590  */
1591 
1592 /* Initialise core socket variables */
1593 void sock_init_data(struct socket *sock, struct sock *sk);
1594 
1595 /*
1596  * Socket reference counting postulates.
1597  *
1598  * * Each user of socket SHOULD hold a reference count.
1599  * * Each access point to socket (an hash table bucket, reference from a list,
1600  *   running timer, skb in flight MUST hold a reference count.
1601  * * When reference count hits 0, it means it will never increase back.
1602  * * When reference count hits 0, it means that no references from
1603  *   outside exist to this socket and current process on current CPU
1604  *   is last user and may/should destroy this socket.
1605  * * sk_free is called from any context: process, BH, IRQ. When
1606  *   it is called, socket has no references from outside -> sk_free
1607  *   may release descendant resources allocated by the socket, but
1608  *   to the time when it is called, socket is NOT referenced by any
1609  *   hash tables, lists etc.
1610  * * Packets, delivered from outside (from network or from another process)
1611  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1612  *   when they sit in queue. Otherwise, packets will leak to hole, when
1613  *   socket is looked up by one cpu and unhasing is made by another CPU.
1614  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1615  *   (leak to backlog). Packet socket does all the processing inside
1616  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1617  *   use separate SMP lock, so that they are prone too.
1618  */
1619 
1620 /* Ungrab socket and destroy it, if it was the last reference. */
1621 static inline void sock_put(struct sock *sk)
1622 {
1623 	if (atomic_dec_and_test(&sk->sk_refcnt))
1624 		sk_free(sk);
1625 }
1626 /* Generic version of sock_put(), dealing with all sockets
1627  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1628  */
1629 void sock_gen_put(struct sock *sk);
1630 
1631 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1632 
1633 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1634 {
1635 	sk->sk_tx_queue_mapping = tx_queue;
1636 }
1637 
1638 static inline void sk_tx_queue_clear(struct sock *sk)
1639 {
1640 	sk->sk_tx_queue_mapping = -1;
1641 }
1642 
1643 static inline int sk_tx_queue_get(const struct sock *sk)
1644 {
1645 	return sk ? sk->sk_tx_queue_mapping : -1;
1646 }
1647 
1648 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1649 {
1650 	sk_tx_queue_clear(sk);
1651 	sk->sk_socket = sock;
1652 }
1653 
1654 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1655 {
1656 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1657 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1658 }
1659 /* Detach socket from process context.
1660  * Announce socket dead, detach it from wait queue and inode.
1661  * Note that parent inode held reference count on this struct sock,
1662  * we do not release it in this function, because protocol
1663  * probably wants some additional cleanups or even continuing
1664  * to work with this socket (TCP).
1665  */
1666 static inline void sock_orphan(struct sock *sk)
1667 {
1668 	write_lock_bh(&sk->sk_callback_lock);
1669 	sock_set_flag(sk, SOCK_DEAD);
1670 	sk_set_socket(sk, NULL);
1671 	sk->sk_wq  = NULL;
1672 	write_unlock_bh(&sk->sk_callback_lock);
1673 }
1674 
1675 static inline void sock_graft(struct sock *sk, struct socket *parent)
1676 {
1677 	write_lock_bh(&sk->sk_callback_lock);
1678 	sk->sk_wq = parent->wq;
1679 	parent->sk = sk;
1680 	sk_set_socket(sk, parent);
1681 	security_sock_graft(sk, parent);
1682 	write_unlock_bh(&sk->sk_callback_lock);
1683 }
1684 
1685 kuid_t sock_i_uid(struct sock *sk);
1686 unsigned long sock_i_ino(struct sock *sk);
1687 
1688 static inline u32 net_tx_rndhash(void)
1689 {
1690 	u32 v = prandom_u32();
1691 
1692 	return v ?: 1;
1693 }
1694 
1695 static inline void sk_set_txhash(struct sock *sk)
1696 {
1697 	sk->sk_txhash = net_tx_rndhash();
1698 }
1699 
1700 static inline void sk_rethink_txhash(struct sock *sk)
1701 {
1702 	if (sk->sk_txhash)
1703 		sk_set_txhash(sk);
1704 }
1705 
1706 static inline struct dst_entry *
1707 __sk_dst_get(struct sock *sk)
1708 {
1709 	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1710 						       lockdep_is_held(&sk->sk_lock.slock));
1711 }
1712 
1713 static inline struct dst_entry *
1714 sk_dst_get(struct sock *sk)
1715 {
1716 	struct dst_entry *dst;
1717 
1718 	rcu_read_lock();
1719 	dst = rcu_dereference(sk->sk_dst_cache);
1720 	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1721 		dst = NULL;
1722 	rcu_read_unlock();
1723 	return dst;
1724 }
1725 
1726 static inline void dst_negative_advice(struct sock *sk)
1727 {
1728 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1729 
1730 	sk_rethink_txhash(sk);
1731 
1732 	if (dst && dst->ops->negative_advice) {
1733 		ndst = dst->ops->negative_advice(dst);
1734 
1735 		if (ndst != dst) {
1736 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1737 			sk_tx_queue_clear(sk);
1738 		}
1739 	}
1740 }
1741 
1742 static inline void
1743 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1744 {
1745 	struct dst_entry *old_dst;
1746 
1747 	sk_tx_queue_clear(sk);
1748 	/*
1749 	 * This can be called while sk is owned by the caller only,
1750 	 * with no state that can be checked in a rcu_dereference_check() cond
1751 	 */
1752 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1753 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1754 	dst_release(old_dst);
1755 }
1756 
1757 static inline void
1758 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1759 {
1760 	struct dst_entry *old_dst;
1761 
1762 	sk_tx_queue_clear(sk);
1763 	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1764 	dst_release(old_dst);
1765 }
1766 
1767 static inline void
1768 __sk_dst_reset(struct sock *sk)
1769 {
1770 	__sk_dst_set(sk, NULL);
1771 }
1772 
1773 static inline void
1774 sk_dst_reset(struct sock *sk)
1775 {
1776 	sk_dst_set(sk, NULL);
1777 }
1778 
1779 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1780 
1781 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1782 
1783 bool sk_mc_loop(struct sock *sk);
1784 
1785 static inline bool sk_can_gso(const struct sock *sk)
1786 {
1787 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1788 }
1789 
1790 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1791 
1792 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1793 {
1794 	sk->sk_route_nocaps |= flags;
1795 	sk->sk_route_caps &= ~flags;
1796 }
1797 
1798 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1799 					   struct iov_iter *from, char *to,
1800 					   int copy, int offset)
1801 {
1802 	if (skb->ip_summed == CHECKSUM_NONE) {
1803 		__wsum csum = 0;
1804 		if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1805 			return -EFAULT;
1806 		skb->csum = csum_block_add(skb->csum, csum, offset);
1807 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1808 		if (copy_from_iter_nocache(to, copy, from) != copy)
1809 			return -EFAULT;
1810 	} else if (copy_from_iter(to, copy, from) != copy)
1811 		return -EFAULT;
1812 
1813 	return 0;
1814 }
1815 
1816 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1817 				       struct iov_iter *from, int copy)
1818 {
1819 	int err, offset = skb->len;
1820 
1821 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1822 				       copy, offset);
1823 	if (err)
1824 		__skb_trim(skb, offset);
1825 
1826 	return err;
1827 }
1828 
1829 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1830 					   struct sk_buff *skb,
1831 					   struct page *page,
1832 					   int off, int copy)
1833 {
1834 	int err;
1835 
1836 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1837 				       copy, skb->len);
1838 	if (err)
1839 		return err;
1840 
1841 	skb->len	     += copy;
1842 	skb->data_len	     += copy;
1843 	skb->truesize	     += copy;
1844 	sk->sk_wmem_queued   += copy;
1845 	sk_mem_charge(sk, copy);
1846 	return 0;
1847 }
1848 
1849 /**
1850  * sk_wmem_alloc_get - returns write allocations
1851  * @sk: socket
1852  *
1853  * Returns sk_wmem_alloc minus initial offset of one
1854  */
1855 static inline int sk_wmem_alloc_get(const struct sock *sk)
1856 {
1857 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1858 }
1859 
1860 /**
1861  * sk_rmem_alloc_get - returns read allocations
1862  * @sk: socket
1863  *
1864  * Returns sk_rmem_alloc
1865  */
1866 static inline int sk_rmem_alloc_get(const struct sock *sk)
1867 {
1868 	return atomic_read(&sk->sk_rmem_alloc);
1869 }
1870 
1871 /**
1872  * sk_has_allocations - check if allocations are outstanding
1873  * @sk: socket
1874  *
1875  * Returns true if socket has write or read allocations
1876  */
1877 static inline bool sk_has_allocations(const struct sock *sk)
1878 {
1879 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1880 }
1881 
1882 /**
1883  * wq_has_sleeper - check if there are any waiting processes
1884  * @wq: struct socket_wq
1885  *
1886  * Returns true if socket_wq has waiting processes
1887  *
1888  * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1889  * barrier call. They were added due to the race found within the tcp code.
1890  *
1891  * Consider following tcp code paths:
1892  *
1893  * CPU1                  CPU2
1894  *
1895  * sys_select            receive packet
1896  *   ...                 ...
1897  *   __add_wait_queue    update tp->rcv_nxt
1898  *   ...                 ...
1899  *   tp->rcv_nxt check   sock_def_readable
1900  *   ...                 {
1901  *   schedule               rcu_read_lock();
1902  *                          wq = rcu_dereference(sk->sk_wq);
1903  *                          if (wq && waitqueue_active(&wq->wait))
1904  *                              wake_up_interruptible(&wq->wait)
1905  *                          ...
1906  *                       }
1907  *
1908  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1909  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1910  * could then endup calling schedule and sleep forever if there are no more
1911  * data on the socket.
1912  *
1913  */
1914 static inline bool wq_has_sleeper(struct socket_wq *wq)
1915 {
1916 	/* We need to be sure we are in sync with the
1917 	 * add_wait_queue modifications to the wait queue.
1918 	 *
1919 	 * This memory barrier is paired in the sock_poll_wait.
1920 	 */
1921 	smp_mb();
1922 	return wq && waitqueue_active(&wq->wait);
1923 }
1924 
1925 /**
1926  * sock_poll_wait - place memory barrier behind the poll_wait call.
1927  * @filp:           file
1928  * @wait_address:   socket wait queue
1929  * @p:              poll_table
1930  *
1931  * See the comments in the wq_has_sleeper function.
1932  */
1933 static inline void sock_poll_wait(struct file *filp,
1934 		wait_queue_head_t *wait_address, poll_table *p)
1935 {
1936 	if (!poll_does_not_wait(p) && wait_address) {
1937 		poll_wait(filp, wait_address, p);
1938 		/* We need to be sure we are in sync with the
1939 		 * socket flags modification.
1940 		 *
1941 		 * This memory barrier is paired in the wq_has_sleeper.
1942 		 */
1943 		smp_mb();
1944 	}
1945 }
1946 
1947 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1948 {
1949 	if (sk->sk_txhash) {
1950 		skb->l4_hash = 1;
1951 		skb->hash = sk->sk_txhash;
1952 	}
1953 }
1954 
1955 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
1956 
1957 /*
1958  *	Queue a received datagram if it will fit. Stream and sequenced
1959  *	protocols can't normally use this as they need to fit buffers in
1960  *	and play with them.
1961  *
1962  *	Inlined as it's very short and called for pretty much every
1963  *	packet ever received.
1964  */
1965 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1966 {
1967 	skb_orphan(skb);
1968 	skb->sk = sk;
1969 	skb->destructor = sock_rfree;
1970 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1971 	sk_mem_charge(sk, skb->truesize);
1972 }
1973 
1974 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1975 		    unsigned long expires);
1976 
1977 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1978 
1979 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1980 
1981 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1982 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1983 
1984 /*
1985  *	Recover an error report and clear atomically
1986  */
1987 
1988 static inline int sock_error(struct sock *sk)
1989 {
1990 	int err;
1991 	if (likely(!sk->sk_err))
1992 		return 0;
1993 	err = xchg(&sk->sk_err, 0);
1994 	return -err;
1995 }
1996 
1997 static inline unsigned long sock_wspace(struct sock *sk)
1998 {
1999 	int amt = 0;
2000 
2001 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2002 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2003 		if (amt < 0)
2004 			amt = 0;
2005 	}
2006 	return amt;
2007 }
2008 
2009 /* Note:
2010  *  We use sk->sk_wq_raw, from contexts knowing this
2011  *  pointer is not NULL and cannot disappear/change.
2012  */
2013 static inline void sk_set_bit(int nr, struct sock *sk)
2014 {
2015 	set_bit(nr, &sk->sk_wq_raw->flags);
2016 }
2017 
2018 static inline void sk_clear_bit(int nr, struct sock *sk)
2019 {
2020 	clear_bit(nr, &sk->sk_wq_raw->flags);
2021 }
2022 
2023 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2024 {
2025 	if (sock_flag(sk, SOCK_FASYNC)) {
2026 		rcu_read_lock();
2027 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2028 		rcu_read_unlock();
2029 	}
2030 }
2031 
2032 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2033  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2034  * Note: for send buffers, TCP works better if we can build two skbs at
2035  * minimum.
2036  */
2037 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2038 
2039 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2040 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2041 
2042 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2043 {
2044 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2045 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2046 		sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2047 	}
2048 }
2049 
2050 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2051 				    bool force_schedule);
2052 
2053 /**
2054  * sk_page_frag - return an appropriate page_frag
2055  * @sk: socket
2056  *
2057  * If socket allocation mode allows current thread to sleep, it means its
2058  * safe to use the per task page_frag instead of the per socket one.
2059  */
2060 static inline struct page_frag *sk_page_frag(struct sock *sk)
2061 {
2062 	if (gfpflags_allow_blocking(sk->sk_allocation))
2063 		return &current->task_frag;
2064 
2065 	return &sk->sk_frag;
2066 }
2067 
2068 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2069 
2070 /*
2071  *	Default write policy as shown to user space via poll/select/SIGIO
2072  */
2073 static inline bool sock_writeable(const struct sock *sk)
2074 {
2075 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2076 }
2077 
2078 static inline gfp_t gfp_any(void)
2079 {
2080 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2081 }
2082 
2083 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2084 {
2085 	return noblock ? 0 : sk->sk_rcvtimeo;
2086 }
2087 
2088 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2089 {
2090 	return noblock ? 0 : sk->sk_sndtimeo;
2091 }
2092 
2093 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2094 {
2095 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2096 }
2097 
2098 /* Alas, with timeout socket operations are not restartable.
2099  * Compare this to poll().
2100  */
2101 static inline int sock_intr_errno(long timeo)
2102 {
2103 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2104 }
2105 
2106 struct sock_skb_cb {
2107 	u32 dropcount;
2108 };
2109 
2110 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2111  * using skb->cb[] would keep using it directly and utilize its
2112  * alignement guarantee.
2113  */
2114 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2115 			    sizeof(struct sock_skb_cb)))
2116 
2117 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2118 			    SOCK_SKB_CB_OFFSET))
2119 
2120 #define sock_skb_cb_check_size(size) \
2121 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2122 
2123 static inline void
2124 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2125 {
2126 	SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2127 }
2128 
2129 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2130 			   struct sk_buff *skb);
2131 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2132 			     struct sk_buff *skb);
2133 
2134 static inline void
2135 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2136 {
2137 	ktime_t kt = skb->tstamp;
2138 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2139 
2140 	/*
2141 	 * generate control messages if
2142 	 * - receive time stamping in software requested
2143 	 * - software time stamp available and wanted
2144 	 * - hardware time stamps available and wanted
2145 	 */
2146 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2147 	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2148 	    (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2149 	    (hwtstamps->hwtstamp.tv64 &&
2150 	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2151 		__sock_recv_timestamp(msg, sk, skb);
2152 	else
2153 		sk->sk_stamp = kt;
2154 
2155 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2156 		__sock_recv_wifi_status(msg, sk, skb);
2157 }
2158 
2159 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2160 			      struct sk_buff *skb);
2161 
2162 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2163 					  struct sk_buff *skb)
2164 {
2165 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2166 			   (1UL << SOCK_RCVTSTAMP))
2167 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2168 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2169 
2170 	if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2171 		__sock_recv_ts_and_drops(msg, sk, skb);
2172 	else
2173 		sk->sk_stamp = skb->tstamp;
2174 }
2175 
2176 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2177 
2178 /**
2179  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2180  * @sk:		socket sending this packet
2181  * @tx_flags:	completed with instructions for time stamping
2182  *
2183  * Note : callers should take care of initial *tx_flags value (usually 0)
2184  */
2185 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2186 {
2187 	if (unlikely(sk->sk_tsflags))
2188 		__sock_tx_timestamp(sk, tx_flags);
2189 	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2190 		*tx_flags |= SKBTX_WIFI_STATUS;
2191 }
2192 
2193 /**
2194  * sk_eat_skb - Release a skb if it is no longer needed
2195  * @sk: socket to eat this skb from
2196  * @skb: socket buffer to eat
2197  *
2198  * This routine must be called with interrupts disabled or with the socket
2199  * locked so that the sk_buff queue operation is ok.
2200 */
2201 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2202 {
2203 	__skb_unlink(skb, &sk->sk_receive_queue);
2204 	__kfree_skb(skb);
2205 }
2206 
2207 static inline
2208 struct net *sock_net(const struct sock *sk)
2209 {
2210 	return read_pnet(&sk->sk_net);
2211 }
2212 
2213 static inline
2214 void sock_net_set(struct sock *sk, struct net *net)
2215 {
2216 	write_pnet(&sk->sk_net, net);
2217 }
2218 
2219 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2220 {
2221 	if (skb->sk) {
2222 		struct sock *sk = skb->sk;
2223 
2224 		skb->destructor = NULL;
2225 		skb->sk = NULL;
2226 		return sk;
2227 	}
2228 	return NULL;
2229 }
2230 
2231 /* This helper checks if a socket is a full socket,
2232  * ie _not_ a timewait or request socket.
2233  */
2234 static inline bool sk_fullsock(const struct sock *sk)
2235 {
2236 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2237 }
2238 
2239 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2240  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2241  */
2242 static inline bool sk_listener(const struct sock *sk)
2243 {
2244 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2245 }
2246 
2247 /**
2248  * sk_state_load - read sk->sk_state for lockless contexts
2249  * @sk: socket pointer
2250  *
2251  * Paired with sk_state_store(). Used in places we do not hold socket lock :
2252  * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2253  */
2254 static inline int sk_state_load(const struct sock *sk)
2255 {
2256 	return smp_load_acquire(&sk->sk_state);
2257 }
2258 
2259 /**
2260  * sk_state_store - update sk->sk_state
2261  * @sk: socket pointer
2262  * @newstate: new state
2263  *
2264  * Paired with sk_state_load(). Should be used in contexts where
2265  * state change might impact lockless readers.
2266  */
2267 static inline void sk_state_store(struct sock *sk, int newstate)
2268 {
2269 	smp_store_release(&sk->sk_state, newstate);
2270 }
2271 
2272 void sock_enable_timestamp(struct sock *sk, int flag);
2273 int sock_get_timestamp(struct sock *, struct timeval __user *);
2274 int sock_get_timestampns(struct sock *, struct timespec __user *);
2275 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2276 		       int type);
2277 
2278 bool sk_ns_capable(const struct sock *sk,
2279 		   struct user_namespace *user_ns, int cap);
2280 bool sk_capable(const struct sock *sk, int cap);
2281 bool sk_net_capable(const struct sock *sk, int cap);
2282 
2283 extern __u32 sysctl_wmem_max;
2284 extern __u32 sysctl_rmem_max;
2285 
2286 extern int sysctl_tstamp_allow_data;
2287 extern int sysctl_optmem_max;
2288 
2289 extern __u32 sysctl_wmem_default;
2290 extern __u32 sysctl_rmem_default;
2291 
2292 #endif	/* _SOCK_H */
2293