xref: /linux/include/net/sock.h (revision b2d0f5d5dc53532e6f07bc546a476a55ebdfe0f3)
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 
76 /*
77  * This structure really needs to be cleaned up.
78  * Most of it is for TCP, and not used by any of
79  * the other protocols.
80  */
81 
82 /* Define this to get the SOCK_DBG debugging facility. */
83 #define SOCK_DEBUGGING
84 #ifdef SOCK_DEBUGGING
85 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
86 					printk(KERN_DEBUG msg); } while (0)
87 #else
88 /* Validate arguments and do nothing */
89 static inline __printf(2, 3)
90 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
91 {
92 }
93 #endif
94 
95 /* This is the per-socket lock.  The spinlock provides a synchronization
96  * between user contexts and software interrupt processing, whereas the
97  * mini-semaphore synchronizes multiple users amongst themselves.
98  */
99 typedef struct {
100 	spinlock_t		slock;
101 	int			owned;
102 	wait_queue_head_t	wq;
103 	/*
104 	 * We express the mutex-alike socket_lock semantics
105 	 * to the lock validator by explicitly managing
106 	 * the slock as a lock variant (in addition to
107 	 * the slock itself):
108 	 */
109 #ifdef CONFIG_DEBUG_LOCK_ALLOC
110 	struct lockdep_map dep_map;
111 #endif
112 } socket_lock_t;
113 
114 struct sock;
115 struct proto;
116 struct net;
117 
118 typedef __u32 __bitwise __portpair;
119 typedef __u64 __bitwise __addrpair;
120 
121 /**
122  *	struct sock_common - minimal network layer representation of sockets
123  *	@skc_daddr: Foreign IPv4 addr
124  *	@skc_rcv_saddr: Bound local IPv4 addr
125  *	@skc_hash: hash value used with various protocol lookup tables
126  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
127  *	@skc_dport: placeholder for inet_dport/tw_dport
128  *	@skc_num: placeholder for inet_num/tw_num
129  *	@skc_family: network address family
130  *	@skc_state: Connection state
131  *	@skc_reuse: %SO_REUSEADDR setting
132  *	@skc_reuseport: %SO_REUSEPORT setting
133  *	@skc_bound_dev_if: bound device index if != 0
134  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
135  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
136  *	@skc_prot: protocol handlers inside a network family
137  *	@skc_net: reference to the network namespace of this socket
138  *	@skc_node: main hash linkage for various protocol lookup tables
139  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
140  *	@skc_tx_queue_mapping: tx queue number for this connection
141  *	@skc_flags: place holder for sk_flags
142  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
143  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
144  *	@skc_incoming_cpu: record/match cpu processing incoming packets
145  *	@skc_refcnt: reference count
146  *
147  *	This is the minimal network layer representation of sockets, the header
148  *	for struct sock and struct inet_timewait_sock.
149  */
150 struct sock_common {
151 	/* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
152 	 * address on 64bit arches : cf INET_MATCH()
153 	 */
154 	union {
155 		__addrpair	skc_addrpair;
156 		struct {
157 			__be32	skc_daddr;
158 			__be32	skc_rcv_saddr;
159 		};
160 	};
161 	union  {
162 		unsigned int	skc_hash;
163 		__u16		skc_u16hashes[2];
164 	};
165 	/* skc_dport && skc_num must be grouped as well */
166 	union {
167 		__portpair	skc_portpair;
168 		struct {
169 			__be16	skc_dport;
170 			__u16	skc_num;
171 		};
172 	};
173 
174 	unsigned short		skc_family;
175 	volatile unsigned char	skc_state;
176 	unsigned char		skc_reuse:4;
177 	unsigned char		skc_reuseport:1;
178 	unsigned char		skc_ipv6only:1;
179 	unsigned char		skc_net_refcnt:1;
180 	int			skc_bound_dev_if;
181 	union {
182 		struct hlist_node	skc_bind_node;
183 		struct hlist_node	skc_portaddr_node;
184 	};
185 	struct proto		*skc_prot;
186 	possible_net_t		skc_net;
187 
188 #if IS_ENABLED(CONFIG_IPV6)
189 	struct in6_addr		skc_v6_daddr;
190 	struct in6_addr		skc_v6_rcv_saddr;
191 #endif
192 
193 	atomic64_t		skc_cookie;
194 
195 	/* following fields are padding to force
196 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
197 	 * assuming IPV6 is enabled. We use this padding differently
198 	 * for different kind of 'sockets'
199 	 */
200 	union {
201 		unsigned long	skc_flags;
202 		struct sock	*skc_listener; /* request_sock */
203 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
204 	};
205 	/*
206 	 * fields between dontcopy_begin/dontcopy_end
207 	 * are not copied in sock_copy()
208 	 */
209 	/* private: */
210 	int			skc_dontcopy_begin[0];
211 	/* public: */
212 	union {
213 		struct hlist_node	skc_node;
214 		struct hlist_nulls_node skc_nulls_node;
215 	};
216 	int			skc_tx_queue_mapping;
217 	union {
218 		int		skc_incoming_cpu;
219 		u32		skc_rcv_wnd;
220 		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
221 	};
222 
223 	refcount_t		skc_refcnt;
224 	/* private: */
225 	int                     skc_dontcopy_end[0];
226 	union {
227 		u32		skc_rxhash;
228 		u32		skc_window_clamp;
229 		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
230 	};
231 	/* public: */
232 };
233 
234 /**
235   *	struct sock - network layer representation of sockets
236   *	@__sk_common: shared layout with inet_timewait_sock
237   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
238   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
239   *	@sk_lock:	synchronizer
240   *	@sk_kern_sock: True if sock is using kernel lock classes
241   *	@sk_rcvbuf: size of receive buffer in bytes
242   *	@sk_wq: sock wait queue and async head
243   *	@sk_rx_dst: receive input route used by early demux
244   *	@sk_dst_cache: destination cache
245   *	@sk_dst_pending_confirm: need to confirm neighbour
246   *	@sk_policy: flow policy
247   *	@sk_receive_queue: incoming packets
248   *	@sk_wmem_alloc: transmit queue bytes committed
249   *	@sk_tsq_flags: TCP Small Queues flags
250   *	@sk_write_queue: Packet sending queue
251   *	@sk_omem_alloc: "o" is "option" or "other"
252   *	@sk_wmem_queued: persistent queue size
253   *	@sk_forward_alloc: space allocated forward
254   *	@sk_napi_id: id of the last napi context to receive data for sk
255   *	@sk_ll_usec: usecs to busypoll when there is no data
256   *	@sk_allocation: allocation mode
257   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
258   *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
259   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
260   *	@sk_sndbuf: size of send buffer in bytes
261   *	@__sk_flags_offset: empty field used to determine location of bitfield
262   *	@sk_padding: unused element for alignment
263   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
264   *	@sk_no_check_rx: allow zero checksum in RX packets
265   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
266   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
267   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
268   *	@sk_gso_max_size: Maximum GSO segment size to build
269   *	@sk_gso_max_segs: Maximum number of GSO segments
270   *	@sk_lingertime: %SO_LINGER l_linger setting
271   *	@sk_backlog: always used with the per-socket spinlock held
272   *	@sk_callback_lock: used with the callbacks in the end of this struct
273   *	@sk_error_queue: rarely used
274   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
275   *			  IPV6_ADDRFORM for instance)
276   *	@sk_err: last error
277   *	@sk_err_soft: errors that don't cause failure but are the cause of a
278   *		      persistent failure not just 'timed out'
279   *	@sk_drops: raw/udp drops counter
280   *	@sk_ack_backlog: current listen backlog
281   *	@sk_max_ack_backlog: listen backlog set in listen()
282   *	@sk_uid: user id of owner
283   *	@sk_priority: %SO_PRIORITY setting
284   *	@sk_type: socket type (%SOCK_STREAM, etc)
285   *	@sk_protocol: which protocol this socket belongs in this network family
286   *	@sk_peer_pid: &struct pid for this socket's peer
287   *	@sk_peer_cred: %SO_PEERCRED setting
288   *	@sk_rcvlowat: %SO_RCVLOWAT setting
289   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
290   *	@sk_sndtimeo: %SO_SNDTIMEO setting
291   *	@sk_txhash: computed flow hash for use on transmit
292   *	@sk_filter: socket filtering instructions
293   *	@sk_timer: sock cleanup timer
294   *	@sk_stamp: time stamp of last packet received
295   *	@sk_tsflags: SO_TIMESTAMPING socket options
296   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
297   *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
298   *	@sk_socket: Identd and reporting IO signals
299   *	@sk_user_data: RPC layer private data
300   *	@sk_frag: cached page frag
301   *	@sk_peek_off: current peek_offset value
302   *	@sk_send_head: front of stuff to transmit
303   *	@sk_security: used by security modules
304   *	@sk_mark: generic packet mark
305   *	@sk_cgrp_data: cgroup data for this cgroup
306   *	@sk_memcg: this socket's memory cgroup association
307   *	@sk_write_pending: a write to stream socket waits to start
308   *	@sk_state_change: callback to indicate change in the state of the sock
309   *	@sk_data_ready: callback to indicate there is data to be processed
310   *	@sk_write_space: callback to indicate there is bf sending space available
311   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
312   *	@sk_backlog_rcv: callback to process the backlog
313   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
314   *	@sk_reuseport_cb: reuseport group container
315   *	@sk_rcu: used during RCU grace period
316   */
317 struct sock {
318 	/*
319 	 * Now struct inet_timewait_sock also uses sock_common, so please just
320 	 * don't add nothing before this first member (__sk_common) --acme
321 	 */
322 	struct sock_common	__sk_common;
323 #define sk_node			__sk_common.skc_node
324 #define sk_nulls_node		__sk_common.skc_nulls_node
325 #define sk_refcnt		__sk_common.skc_refcnt
326 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
327 
328 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
329 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
330 #define sk_hash			__sk_common.skc_hash
331 #define sk_portpair		__sk_common.skc_portpair
332 #define sk_num			__sk_common.skc_num
333 #define sk_dport		__sk_common.skc_dport
334 #define sk_addrpair		__sk_common.skc_addrpair
335 #define sk_daddr		__sk_common.skc_daddr
336 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
337 #define sk_family		__sk_common.skc_family
338 #define sk_state		__sk_common.skc_state
339 #define sk_reuse		__sk_common.skc_reuse
340 #define sk_reuseport		__sk_common.skc_reuseport
341 #define sk_ipv6only		__sk_common.skc_ipv6only
342 #define sk_net_refcnt		__sk_common.skc_net_refcnt
343 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
344 #define sk_bind_node		__sk_common.skc_bind_node
345 #define sk_prot			__sk_common.skc_prot
346 #define sk_net			__sk_common.skc_net
347 #define sk_v6_daddr		__sk_common.skc_v6_daddr
348 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
349 #define sk_cookie		__sk_common.skc_cookie
350 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
351 #define sk_flags		__sk_common.skc_flags
352 #define sk_rxhash		__sk_common.skc_rxhash
353 
354 	socket_lock_t		sk_lock;
355 	atomic_t		sk_drops;
356 	int			sk_rcvlowat;
357 	struct sk_buff_head	sk_error_queue;
358 	struct sk_buff_head	sk_receive_queue;
359 	/*
360 	 * The backlog queue is special, it is always used with
361 	 * the per-socket spinlock held and requires low latency
362 	 * access. Therefore we special case it's implementation.
363 	 * Note : rmem_alloc is in this structure to fill a hole
364 	 * on 64bit arches, not because its logically part of
365 	 * backlog.
366 	 */
367 	struct {
368 		atomic_t	rmem_alloc;
369 		int		len;
370 		struct sk_buff	*head;
371 		struct sk_buff	*tail;
372 	} sk_backlog;
373 #define sk_rmem_alloc sk_backlog.rmem_alloc
374 
375 	int			sk_forward_alloc;
376 #ifdef CONFIG_NET_RX_BUSY_POLL
377 	unsigned int		sk_ll_usec;
378 	/* ===== mostly read cache line ===== */
379 	unsigned int		sk_napi_id;
380 #endif
381 	int			sk_rcvbuf;
382 
383 	struct sk_filter __rcu	*sk_filter;
384 	union {
385 		struct socket_wq __rcu	*sk_wq;
386 		struct socket_wq	*sk_wq_raw;
387 	};
388 #ifdef CONFIG_XFRM
389 	struct xfrm_policy __rcu *sk_policy[2];
390 #endif
391 	struct dst_entry	*sk_rx_dst;
392 	struct dst_entry __rcu	*sk_dst_cache;
393 	atomic_t		sk_omem_alloc;
394 	int			sk_sndbuf;
395 
396 	/* ===== cache line for TX ===== */
397 	int			sk_wmem_queued;
398 	refcount_t		sk_wmem_alloc;
399 	unsigned long		sk_tsq_flags;
400 	union {
401 		struct sk_buff	*sk_send_head;
402 		struct rb_root	tcp_rtx_queue;
403 	};
404 	struct sk_buff_head	sk_write_queue;
405 	__s32			sk_peek_off;
406 	int			sk_write_pending;
407 	__u32			sk_dst_pending_confirm;
408 	u32			sk_pacing_status; /* see enum sk_pacing */
409 	long			sk_sndtimeo;
410 	struct timer_list	sk_timer;
411 	__u32			sk_priority;
412 	__u32			sk_mark;
413 	u32			sk_pacing_rate; /* bytes per second */
414 	u32			sk_max_pacing_rate;
415 	struct page_frag	sk_frag;
416 	netdev_features_t	sk_route_caps;
417 	netdev_features_t	sk_route_nocaps;
418 	int			sk_gso_type;
419 	unsigned int		sk_gso_max_size;
420 	gfp_t			sk_allocation;
421 	__u32			sk_txhash;
422 
423 	/*
424 	 * Because of non atomicity rules, all
425 	 * changes are protected by socket lock.
426 	 */
427 	unsigned int		__sk_flags_offset[0];
428 #ifdef __BIG_ENDIAN_BITFIELD
429 #define SK_FL_PROTO_SHIFT  16
430 #define SK_FL_PROTO_MASK   0x00ff0000
431 
432 #define SK_FL_TYPE_SHIFT   0
433 #define SK_FL_TYPE_MASK    0x0000ffff
434 #else
435 #define SK_FL_PROTO_SHIFT  8
436 #define SK_FL_PROTO_MASK   0x0000ff00
437 
438 #define SK_FL_TYPE_SHIFT   16
439 #define SK_FL_TYPE_MASK    0xffff0000
440 #endif
441 
442 	kmemcheck_bitfield_begin(flags);
443 	unsigned int		sk_padding : 1,
444 				sk_kern_sock : 1,
445 				sk_no_check_tx : 1,
446 				sk_no_check_rx : 1,
447 				sk_userlocks : 4,
448 				sk_protocol  : 8,
449 				sk_type      : 16;
450 #define SK_PROTOCOL_MAX U8_MAX
451 	kmemcheck_bitfield_end(flags);
452 
453 	u16			sk_gso_max_segs;
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 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
690 	    sk->sk_family == AF_INET6)
691 		hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
692 	else
693 		hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
694 }
695 
696 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
697 {
698 	sock_hold(sk);
699 	__sk_nulls_add_node_rcu(sk, list);
700 }
701 
702 static inline void __sk_del_bind_node(struct sock *sk)
703 {
704 	__hlist_del(&sk->sk_bind_node);
705 }
706 
707 static inline void sk_add_bind_node(struct sock *sk,
708 					struct hlist_head *list)
709 {
710 	hlist_add_head(&sk->sk_bind_node, list);
711 }
712 
713 #define sk_for_each(__sk, list) \
714 	hlist_for_each_entry(__sk, list, sk_node)
715 #define sk_for_each_rcu(__sk, list) \
716 	hlist_for_each_entry_rcu(__sk, list, sk_node)
717 #define sk_nulls_for_each(__sk, node, list) \
718 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
719 #define sk_nulls_for_each_rcu(__sk, node, list) \
720 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
721 #define sk_for_each_from(__sk) \
722 	hlist_for_each_entry_from(__sk, sk_node)
723 #define sk_nulls_for_each_from(__sk, node) \
724 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
725 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
726 #define sk_for_each_safe(__sk, tmp, list) \
727 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
728 #define sk_for_each_bound(__sk, list) \
729 	hlist_for_each_entry(__sk, list, sk_bind_node)
730 
731 /**
732  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
733  * @tpos:	the type * to use as a loop cursor.
734  * @pos:	the &struct hlist_node to use as a loop cursor.
735  * @head:	the head for your list.
736  * @offset:	offset of hlist_node within the struct.
737  *
738  */
739 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
740 	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
741 	     pos != NULL &&						       \
742 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
743 	     pos = rcu_dereference(hlist_next_rcu(pos)))
744 
745 static inline struct user_namespace *sk_user_ns(struct sock *sk)
746 {
747 	/* Careful only use this in a context where these parameters
748 	 * can not change and must all be valid, such as recvmsg from
749 	 * userspace.
750 	 */
751 	return sk->sk_socket->file->f_cred->user_ns;
752 }
753 
754 /* Sock flags */
755 enum sock_flags {
756 	SOCK_DEAD,
757 	SOCK_DONE,
758 	SOCK_URGINLINE,
759 	SOCK_KEEPOPEN,
760 	SOCK_LINGER,
761 	SOCK_DESTROY,
762 	SOCK_BROADCAST,
763 	SOCK_TIMESTAMP,
764 	SOCK_ZAPPED,
765 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
766 	SOCK_DBG, /* %SO_DEBUG setting */
767 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
768 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
769 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
770 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
771 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
772 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
773 	SOCK_FASYNC, /* fasync() active */
774 	SOCK_RXQ_OVFL,
775 	SOCK_ZEROCOPY, /* buffers from userspace */
776 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
777 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
778 		     * Will use last 4 bytes of packet sent from
779 		     * user-space instead.
780 		     */
781 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
782 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
783 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
784 };
785 
786 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
787 
788 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
789 {
790 	nsk->sk_flags = osk->sk_flags;
791 }
792 
793 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
794 {
795 	__set_bit(flag, &sk->sk_flags);
796 }
797 
798 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
799 {
800 	__clear_bit(flag, &sk->sk_flags);
801 }
802 
803 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
804 {
805 	return test_bit(flag, &sk->sk_flags);
806 }
807 
808 #ifdef CONFIG_NET
809 extern struct static_key memalloc_socks;
810 static inline int sk_memalloc_socks(void)
811 {
812 	return static_key_false(&memalloc_socks);
813 }
814 #else
815 
816 static inline int sk_memalloc_socks(void)
817 {
818 	return 0;
819 }
820 
821 #endif
822 
823 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
824 {
825 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
826 }
827 
828 static inline void sk_acceptq_removed(struct sock *sk)
829 {
830 	sk->sk_ack_backlog--;
831 }
832 
833 static inline void sk_acceptq_added(struct sock *sk)
834 {
835 	sk->sk_ack_backlog++;
836 }
837 
838 static inline bool sk_acceptq_is_full(const struct sock *sk)
839 {
840 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
841 }
842 
843 /*
844  * Compute minimal free write space needed to queue new packets.
845  */
846 static inline int sk_stream_min_wspace(const struct sock *sk)
847 {
848 	return sk->sk_wmem_queued >> 1;
849 }
850 
851 static inline int sk_stream_wspace(const struct sock *sk)
852 {
853 	return sk->sk_sndbuf - sk->sk_wmem_queued;
854 }
855 
856 void sk_stream_write_space(struct sock *sk);
857 
858 /* OOB backlog add */
859 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
860 {
861 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
862 	skb_dst_force(skb);
863 
864 	if (!sk->sk_backlog.tail)
865 		sk->sk_backlog.head = skb;
866 	else
867 		sk->sk_backlog.tail->next = skb;
868 
869 	sk->sk_backlog.tail = skb;
870 	skb->next = NULL;
871 }
872 
873 /*
874  * Take into account size of receive queue and backlog queue
875  * Do not take into account this skb truesize,
876  * to allow even a single big packet to come.
877  */
878 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
879 {
880 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
881 
882 	return qsize > limit;
883 }
884 
885 /* The per-socket spinlock must be held here. */
886 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
887 					      unsigned int limit)
888 {
889 	if (sk_rcvqueues_full(sk, limit))
890 		return -ENOBUFS;
891 
892 	/*
893 	 * If the skb was allocated from pfmemalloc reserves, only
894 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
895 	 * helping free memory
896 	 */
897 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
898 		return -ENOMEM;
899 
900 	__sk_add_backlog(sk, skb);
901 	sk->sk_backlog.len += skb->truesize;
902 	return 0;
903 }
904 
905 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
906 
907 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
908 {
909 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
910 		return __sk_backlog_rcv(sk, skb);
911 
912 	return sk->sk_backlog_rcv(sk, skb);
913 }
914 
915 static inline void sk_incoming_cpu_update(struct sock *sk)
916 {
917 	int cpu = raw_smp_processor_id();
918 
919 	if (unlikely(sk->sk_incoming_cpu != cpu))
920 		sk->sk_incoming_cpu = cpu;
921 }
922 
923 static inline void sock_rps_record_flow_hash(__u32 hash)
924 {
925 #ifdef CONFIG_RPS
926 	struct rps_sock_flow_table *sock_flow_table;
927 
928 	rcu_read_lock();
929 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
930 	rps_record_sock_flow(sock_flow_table, hash);
931 	rcu_read_unlock();
932 #endif
933 }
934 
935 static inline void sock_rps_record_flow(const struct sock *sk)
936 {
937 #ifdef CONFIG_RPS
938 	if (static_key_false(&rfs_needed)) {
939 		/* Reading sk->sk_rxhash might incur an expensive cache line
940 		 * miss.
941 		 *
942 		 * TCP_ESTABLISHED does cover almost all states where RFS
943 		 * might be useful, and is cheaper [1] than testing :
944 		 *	IPv4: inet_sk(sk)->inet_daddr
945 		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
946 		 * OR	an additional socket flag
947 		 * [1] : sk_state and sk_prot are in the same cache line.
948 		 */
949 		if (sk->sk_state == TCP_ESTABLISHED)
950 			sock_rps_record_flow_hash(sk->sk_rxhash);
951 	}
952 #endif
953 }
954 
955 static inline void sock_rps_save_rxhash(struct sock *sk,
956 					const struct sk_buff *skb)
957 {
958 #ifdef CONFIG_RPS
959 	if (unlikely(sk->sk_rxhash != skb->hash))
960 		sk->sk_rxhash = skb->hash;
961 #endif
962 }
963 
964 static inline void sock_rps_reset_rxhash(struct sock *sk)
965 {
966 #ifdef CONFIG_RPS
967 	sk->sk_rxhash = 0;
968 #endif
969 }
970 
971 #define sk_wait_event(__sk, __timeo, __condition, __wait)		\
972 	({	int __rc;						\
973 		release_sock(__sk);					\
974 		__rc = __condition;					\
975 		if (!__rc) {						\
976 			*(__timeo) = wait_woken(__wait,			\
977 						TASK_INTERRUPTIBLE,	\
978 						*(__timeo));		\
979 		}							\
980 		sched_annotate_sleep();					\
981 		lock_sock(__sk);					\
982 		__rc = __condition;					\
983 		__rc;							\
984 	})
985 
986 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
987 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
988 void sk_stream_wait_close(struct sock *sk, long timeo_p);
989 int sk_stream_error(struct sock *sk, int flags, int err);
990 void sk_stream_kill_queues(struct sock *sk);
991 void sk_set_memalloc(struct sock *sk);
992 void sk_clear_memalloc(struct sock *sk);
993 
994 void __sk_flush_backlog(struct sock *sk);
995 
996 static inline bool sk_flush_backlog(struct sock *sk)
997 {
998 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
999 		__sk_flush_backlog(sk);
1000 		return true;
1001 	}
1002 	return false;
1003 }
1004 
1005 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1006 
1007 struct request_sock_ops;
1008 struct timewait_sock_ops;
1009 struct inet_hashinfo;
1010 struct raw_hashinfo;
1011 struct smc_hashinfo;
1012 struct module;
1013 
1014 /*
1015  * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1016  * un-modified. Special care is taken when initializing object to zero.
1017  */
1018 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1019 {
1020 	if (offsetof(struct sock, sk_node.next) != 0)
1021 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1022 	memset(&sk->sk_node.pprev, 0,
1023 	       size - offsetof(struct sock, sk_node.pprev));
1024 }
1025 
1026 /* Networking protocol blocks we attach to sockets.
1027  * socket layer -> transport layer interface
1028  */
1029 struct proto {
1030 	void			(*close)(struct sock *sk,
1031 					long timeout);
1032 	int			(*connect)(struct sock *sk,
1033 					struct sockaddr *uaddr,
1034 					int addr_len);
1035 	int			(*disconnect)(struct sock *sk, int flags);
1036 
1037 	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1038 					  bool kern);
1039 
1040 	int			(*ioctl)(struct sock *sk, int cmd,
1041 					 unsigned long arg);
1042 	int			(*init)(struct sock *sk);
1043 	void			(*destroy)(struct sock *sk);
1044 	void			(*shutdown)(struct sock *sk, int how);
1045 	int			(*setsockopt)(struct sock *sk, int level,
1046 					int optname, char __user *optval,
1047 					unsigned int optlen);
1048 	int			(*getsockopt)(struct sock *sk, int level,
1049 					int optname, char __user *optval,
1050 					int __user *option);
1051 	void			(*keepalive)(struct sock *sk, int valbool);
1052 #ifdef CONFIG_COMPAT
1053 	int			(*compat_setsockopt)(struct sock *sk,
1054 					int level,
1055 					int optname, char __user *optval,
1056 					unsigned int optlen);
1057 	int			(*compat_getsockopt)(struct sock *sk,
1058 					int level,
1059 					int optname, char __user *optval,
1060 					int __user *option);
1061 	int			(*compat_ioctl)(struct sock *sk,
1062 					unsigned int cmd, unsigned long arg);
1063 #endif
1064 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1065 					   size_t len);
1066 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1067 					   size_t len, int noblock, int flags,
1068 					   int *addr_len);
1069 	int			(*sendpage)(struct sock *sk, struct page *page,
1070 					int offset, size_t size, int flags);
1071 	int			(*bind)(struct sock *sk,
1072 					struct sockaddr *uaddr, int addr_len);
1073 
1074 	int			(*backlog_rcv) (struct sock *sk,
1075 						struct sk_buff *skb);
1076 
1077 	void		(*release_cb)(struct sock *sk);
1078 
1079 	/* Keeping track of sk's, looking them up, and port selection methods. */
1080 	int			(*hash)(struct sock *sk);
1081 	void			(*unhash)(struct sock *sk);
1082 	void			(*rehash)(struct sock *sk);
1083 	int			(*get_port)(struct sock *sk, unsigned short snum);
1084 
1085 	/* Keeping track of sockets in use */
1086 #ifdef CONFIG_PROC_FS
1087 	unsigned int		inuse_idx;
1088 #endif
1089 
1090 	bool			(*stream_memory_free)(const struct sock *sk);
1091 	/* Memory pressure */
1092 	void			(*enter_memory_pressure)(struct sock *sk);
1093 	void			(*leave_memory_pressure)(struct sock *sk);
1094 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1095 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1096 	/*
1097 	 * Pressure flag: try to collapse.
1098 	 * Technical note: it is used by multiple contexts non atomically.
1099 	 * All the __sk_mem_schedule() is of this nature: accounting
1100 	 * is strict, actions are advisory and have some latency.
1101 	 */
1102 	unsigned long		*memory_pressure;
1103 	long			*sysctl_mem;
1104 	int			*sysctl_wmem;
1105 	int			*sysctl_rmem;
1106 	int			max_header;
1107 	bool			no_autobind;
1108 
1109 	struct kmem_cache	*slab;
1110 	unsigned int		obj_size;
1111 	int			slab_flags;
1112 
1113 	struct percpu_counter	*orphan_count;
1114 
1115 	struct request_sock_ops	*rsk_prot;
1116 	struct timewait_sock_ops *twsk_prot;
1117 
1118 	union {
1119 		struct inet_hashinfo	*hashinfo;
1120 		struct udp_table	*udp_table;
1121 		struct raw_hashinfo	*raw_hash;
1122 		struct smc_hashinfo	*smc_hash;
1123 	} h;
1124 
1125 	struct module		*owner;
1126 
1127 	char			name[32];
1128 
1129 	struct list_head	node;
1130 #ifdef SOCK_REFCNT_DEBUG
1131 	atomic_t		socks;
1132 #endif
1133 	int			(*diag_destroy)(struct sock *sk, int err);
1134 } __randomize_layout;
1135 
1136 int proto_register(struct proto *prot, int alloc_slab);
1137 void proto_unregister(struct proto *prot);
1138 
1139 #ifdef SOCK_REFCNT_DEBUG
1140 static inline void sk_refcnt_debug_inc(struct sock *sk)
1141 {
1142 	atomic_inc(&sk->sk_prot->socks);
1143 }
1144 
1145 static inline void sk_refcnt_debug_dec(struct sock *sk)
1146 {
1147 	atomic_dec(&sk->sk_prot->socks);
1148 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1149 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1150 }
1151 
1152 static inline void sk_refcnt_debug_release(const struct sock *sk)
1153 {
1154 	if (refcount_read(&sk->sk_refcnt) != 1)
1155 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1156 		       sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1157 }
1158 #else /* SOCK_REFCNT_DEBUG */
1159 #define sk_refcnt_debug_inc(sk) do { } while (0)
1160 #define sk_refcnt_debug_dec(sk) do { } while (0)
1161 #define sk_refcnt_debug_release(sk) do { } while (0)
1162 #endif /* SOCK_REFCNT_DEBUG */
1163 
1164 static inline bool sk_stream_memory_free(const struct sock *sk)
1165 {
1166 	if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1167 		return false;
1168 
1169 	return sk->sk_prot->stream_memory_free ?
1170 		sk->sk_prot->stream_memory_free(sk) : true;
1171 }
1172 
1173 static inline bool sk_stream_is_writeable(const struct sock *sk)
1174 {
1175 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1176 	       sk_stream_memory_free(sk);
1177 }
1178 
1179 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1180 					    struct cgroup *ancestor)
1181 {
1182 #ifdef CONFIG_SOCK_CGROUP_DATA
1183 	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1184 				    ancestor);
1185 #else
1186 	return -ENOTSUPP;
1187 #endif
1188 }
1189 
1190 static inline bool sk_has_memory_pressure(const struct sock *sk)
1191 {
1192 	return sk->sk_prot->memory_pressure != NULL;
1193 }
1194 
1195 static inline bool sk_under_memory_pressure(const struct sock *sk)
1196 {
1197 	if (!sk->sk_prot->memory_pressure)
1198 		return false;
1199 
1200 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1201 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1202 		return true;
1203 
1204 	return !!*sk->sk_prot->memory_pressure;
1205 }
1206 
1207 static inline long
1208 sk_memory_allocated(const struct sock *sk)
1209 {
1210 	return atomic_long_read(sk->sk_prot->memory_allocated);
1211 }
1212 
1213 static inline long
1214 sk_memory_allocated_add(struct sock *sk, int amt)
1215 {
1216 	return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1217 }
1218 
1219 static inline void
1220 sk_memory_allocated_sub(struct sock *sk, int amt)
1221 {
1222 	atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1223 }
1224 
1225 static inline void sk_sockets_allocated_dec(struct sock *sk)
1226 {
1227 	percpu_counter_dec(sk->sk_prot->sockets_allocated);
1228 }
1229 
1230 static inline void sk_sockets_allocated_inc(struct sock *sk)
1231 {
1232 	percpu_counter_inc(sk->sk_prot->sockets_allocated);
1233 }
1234 
1235 static inline int
1236 sk_sockets_allocated_read_positive(struct sock *sk)
1237 {
1238 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1239 }
1240 
1241 static inline int
1242 proto_sockets_allocated_sum_positive(struct proto *prot)
1243 {
1244 	return percpu_counter_sum_positive(prot->sockets_allocated);
1245 }
1246 
1247 static inline long
1248 proto_memory_allocated(struct proto *prot)
1249 {
1250 	return atomic_long_read(prot->memory_allocated);
1251 }
1252 
1253 static inline bool
1254 proto_memory_pressure(struct proto *prot)
1255 {
1256 	if (!prot->memory_pressure)
1257 		return false;
1258 	return !!*prot->memory_pressure;
1259 }
1260 
1261 
1262 #ifdef CONFIG_PROC_FS
1263 /* Called with local bh disabled */
1264 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1265 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1266 #else
1267 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1268 		int inc)
1269 {
1270 }
1271 #endif
1272 
1273 
1274 /* With per-bucket locks this operation is not-atomic, so that
1275  * this version is not worse.
1276  */
1277 static inline int __sk_prot_rehash(struct sock *sk)
1278 {
1279 	sk->sk_prot->unhash(sk);
1280 	return sk->sk_prot->hash(sk);
1281 }
1282 
1283 /* About 10 seconds */
1284 #define SOCK_DESTROY_TIME (10*HZ)
1285 
1286 /* Sockets 0-1023 can't be bound to unless you are superuser */
1287 #define PROT_SOCK	1024
1288 
1289 #define SHUTDOWN_MASK	3
1290 #define RCV_SHUTDOWN	1
1291 #define SEND_SHUTDOWN	2
1292 
1293 #define SOCK_SNDBUF_LOCK	1
1294 #define SOCK_RCVBUF_LOCK	2
1295 #define SOCK_BINDADDR_LOCK	4
1296 #define SOCK_BINDPORT_LOCK	8
1297 
1298 struct socket_alloc {
1299 	struct socket socket;
1300 	struct inode vfs_inode;
1301 };
1302 
1303 static inline struct socket *SOCKET_I(struct inode *inode)
1304 {
1305 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1306 }
1307 
1308 static inline struct inode *SOCK_INODE(struct socket *socket)
1309 {
1310 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1311 }
1312 
1313 /*
1314  * Functions for memory accounting
1315  */
1316 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1317 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1318 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1319 void __sk_mem_reclaim(struct sock *sk, int amount);
1320 
1321 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1322  * do not necessarily have 16x time more memory than 4KB ones.
1323  */
1324 #define SK_MEM_QUANTUM 4096
1325 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1326 #define SK_MEM_SEND	0
1327 #define SK_MEM_RECV	1
1328 
1329 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1330 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1331 {
1332 	long val = sk->sk_prot->sysctl_mem[index];
1333 
1334 #if PAGE_SIZE > SK_MEM_QUANTUM
1335 	val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1336 #elif PAGE_SIZE < SK_MEM_QUANTUM
1337 	val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1338 #endif
1339 	return val;
1340 }
1341 
1342 static inline int sk_mem_pages(int amt)
1343 {
1344 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1345 }
1346 
1347 static inline bool sk_has_account(struct sock *sk)
1348 {
1349 	/* return true if protocol supports memory accounting */
1350 	return !!sk->sk_prot->memory_allocated;
1351 }
1352 
1353 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1354 {
1355 	if (!sk_has_account(sk))
1356 		return true;
1357 	return size <= sk->sk_forward_alloc ||
1358 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1359 }
1360 
1361 static inline bool
1362 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1363 {
1364 	if (!sk_has_account(sk))
1365 		return true;
1366 	return size<= sk->sk_forward_alloc ||
1367 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1368 		skb_pfmemalloc(skb);
1369 }
1370 
1371 static inline void sk_mem_reclaim(struct sock *sk)
1372 {
1373 	if (!sk_has_account(sk))
1374 		return;
1375 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1376 		__sk_mem_reclaim(sk, sk->sk_forward_alloc);
1377 }
1378 
1379 static inline void sk_mem_reclaim_partial(struct sock *sk)
1380 {
1381 	if (!sk_has_account(sk))
1382 		return;
1383 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1384 		__sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1385 }
1386 
1387 static inline void sk_mem_charge(struct sock *sk, int size)
1388 {
1389 	if (!sk_has_account(sk))
1390 		return;
1391 	sk->sk_forward_alloc -= size;
1392 }
1393 
1394 static inline void sk_mem_uncharge(struct sock *sk, int size)
1395 {
1396 	if (!sk_has_account(sk))
1397 		return;
1398 	sk->sk_forward_alloc += size;
1399 
1400 	/* Avoid a possible overflow.
1401 	 * TCP send queues can make this happen, if sk_mem_reclaim()
1402 	 * is not called and more than 2 GBytes are released at once.
1403 	 *
1404 	 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1405 	 * no need to hold that much forward allocation anyway.
1406 	 */
1407 	if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1408 		__sk_mem_reclaim(sk, 1 << 20);
1409 }
1410 
1411 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1412 {
1413 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1414 	sk->sk_wmem_queued -= skb->truesize;
1415 	sk_mem_uncharge(sk, skb->truesize);
1416 	__kfree_skb(skb);
1417 }
1418 
1419 static inline void sock_release_ownership(struct sock *sk)
1420 {
1421 	if (sk->sk_lock.owned) {
1422 		sk->sk_lock.owned = 0;
1423 
1424 		/* The sk_lock has mutex_unlock() semantics: */
1425 		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1426 	}
1427 }
1428 
1429 /*
1430  * Macro so as to not evaluate some arguments when
1431  * lockdep is not enabled.
1432  *
1433  * Mark both the sk_lock and the sk_lock.slock as a
1434  * per-address-family lock class.
1435  */
1436 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1437 do {									\
1438 	sk->sk_lock.owned = 0;						\
1439 	init_waitqueue_head(&sk->sk_lock.wq);				\
1440 	spin_lock_init(&(sk)->sk_lock.slock);				\
1441 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1442 			sizeof((sk)->sk_lock));				\
1443 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1444 				(skey), (sname));				\
1445 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1446 } while (0)
1447 
1448 #ifdef CONFIG_LOCKDEP
1449 static inline bool lockdep_sock_is_held(const struct sock *csk)
1450 {
1451 	struct sock *sk = (struct sock *)csk;
1452 
1453 	return lockdep_is_held(&sk->sk_lock) ||
1454 	       lockdep_is_held(&sk->sk_lock.slock);
1455 }
1456 #endif
1457 
1458 void lock_sock_nested(struct sock *sk, int subclass);
1459 
1460 static inline void lock_sock(struct sock *sk)
1461 {
1462 	lock_sock_nested(sk, 0);
1463 }
1464 
1465 void release_sock(struct sock *sk);
1466 
1467 /* BH context may only use the following locking interface. */
1468 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1469 #define bh_lock_sock_nested(__sk) \
1470 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1471 				SINGLE_DEPTH_NESTING)
1472 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1473 
1474 bool lock_sock_fast(struct sock *sk);
1475 /**
1476  * unlock_sock_fast - complement of lock_sock_fast
1477  * @sk: socket
1478  * @slow: slow mode
1479  *
1480  * fast unlock socket for user context.
1481  * If slow mode is on, we call regular release_sock()
1482  */
1483 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1484 {
1485 	if (slow)
1486 		release_sock(sk);
1487 	else
1488 		spin_unlock_bh(&sk->sk_lock.slock);
1489 }
1490 
1491 /* Used by processes to "lock" a socket state, so that
1492  * interrupts and bottom half handlers won't change it
1493  * from under us. It essentially blocks any incoming
1494  * packets, so that we won't get any new data or any
1495  * packets that change the state of the socket.
1496  *
1497  * While locked, BH processing will add new packets to
1498  * the backlog queue.  This queue is processed by the
1499  * owner of the socket lock right before it is released.
1500  *
1501  * Since ~2.3.5 it is also exclusive sleep lock serializing
1502  * accesses from user process context.
1503  */
1504 
1505 static inline void sock_owned_by_me(const struct sock *sk)
1506 {
1507 #ifdef CONFIG_LOCKDEP
1508 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1509 #endif
1510 }
1511 
1512 static inline bool sock_owned_by_user(const struct sock *sk)
1513 {
1514 	sock_owned_by_me(sk);
1515 	return sk->sk_lock.owned;
1516 }
1517 
1518 /* no reclassification while locks are held */
1519 static inline bool sock_allow_reclassification(const struct sock *csk)
1520 {
1521 	struct sock *sk = (struct sock *)csk;
1522 
1523 	return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1524 }
1525 
1526 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1527 		      struct proto *prot, int kern);
1528 void sk_free(struct sock *sk);
1529 void sk_destruct(struct sock *sk);
1530 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1531 void sk_free_unlock_clone(struct sock *sk);
1532 
1533 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1534 			     gfp_t priority);
1535 void __sock_wfree(struct sk_buff *skb);
1536 void sock_wfree(struct sk_buff *skb);
1537 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1538 			     gfp_t priority);
1539 void skb_orphan_partial(struct sk_buff *skb);
1540 void sock_rfree(struct sk_buff *skb);
1541 void sock_efree(struct sk_buff *skb);
1542 #ifdef CONFIG_INET
1543 void sock_edemux(struct sk_buff *skb);
1544 #else
1545 #define sock_edemux sock_efree
1546 #endif
1547 
1548 int sock_setsockopt(struct socket *sock, int level, int op,
1549 		    char __user *optval, unsigned int optlen);
1550 
1551 int sock_getsockopt(struct socket *sock, int level, int op,
1552 		    char __user *optval, int __user *optlen);
1553 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1554 				    int noblock, int *errcode);
1555 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1556 				     unsigned long data_len, int noblock,
1557 				     int *errcode, int max_page_order);
1558 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1559 void sock_kfree_s(struct sock *sk, void *mem, int size);
1560 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1561 void sk_send_sigurg(struct sock *sk);
1562 
1563 struct sockcm_cookie {
1564 	u32 mark;
1565 	u16 tsflags;
1566 };
1567 
1568 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1569 		     struct sockcm_cookie *sockc);
1570 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1571 		   struct sockcm_cookie *sockc);
1572 
1573 /*
1574  * Functions to fill in entries in struct proto_ops when a protocol
1575  * does not implement a particular function.
1576  */
1577 int sock_no_bind(struct socket *, struct sockaddr *, int);
1578 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1579 int sock_no_socketpair(struct socket *, struct socket *);
1580 int sock_no_accept(struct socket *, struct socket *, int, bool);
1581 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1582 unsigned int sock_no_poll(struct file *, struct socket *,
1583 			  struct poll_table_struct *);
1584 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1585 int sock_no_listen(struct socket *, int);
1586 int sock_no_shutdown(struct socket *, int);
1587 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1588 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1589 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1590 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1591 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1592 int sock_no_mmap(struct file *file, struct socket *sock,
1593 		 struct vm_area_struct *vma);
1594 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1595 			 size_t size, int flags);
1596 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1597 				int offset, size_t size, int flags);
1598 
1599 /*
1600  * Functions to fill in entries in struct proto_ops when a protocol
1601  * uses the inet style.
1602  */
1603 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1604 				  char __user *optval, int __user *optlen);
1605 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1606 			int flags);
1607 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1608 				  char __user *optval, unsigned int optlen);
1609 int compat_sock_common_getsockopt(struct socket *sock, int level,
1610 		int optname, char __user *optval, int __user *optlen);
1611 int compat_sock_common_setsockopt(struct socket *sock, int level,
1612 		int optname, char __user *optval, unsigned int optlen);
1613 
1614 void sk_common_release(struct sock *sk);
1615 
1616 /*
1617  *	Default socket callbacks and setup code
1618  */
1619 
1620 /* Initialise core socket variables */
1621 void sock_init_data(struct socket *sock, struct sock *sk);
1622 
1623 /*
1624  * Socket reference counting postulates.
1625  *
1626  * * Each user of socket SHOULD hold a reference count.
1627  * * Each access point to socket (an hash table bucket, reference from a list,
1628  *   running timer, skb in flight MUST hold a reference count.
1629  * * When reference count hits 0, it means it will never increase back.
1630  * * When reference count hits 0, it means that no references from
1631  *   outside exist to this socket and current process on current CPU
1632  *   is last user and may/should destroy this socket.
1633  * * sk_free is called from any context: process, BH, IRQ. When
1634  *   it is called, socket has no references from outside -> sk_free
1635  *   may release descendant resources allocated by the socket, but
1636  *   to the time when it is called, socket is NOT referenced by any
1637  *   hash tables, lists etc.
1638  * * Packets, delivered from outside (from network or from another process)
1639  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1640  *   when they sit in queue. Otherwise, packets will leak to hole, when
1641  *   socket is looked up by one cpu and unhasing is made by another CPU.
1642  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1643  *   (leak to backlog). Packet socket does all the processing inside
1644  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1645  *   use separate SMP lock, so that they are prone too.
1646  */
1647 
1648 /* Ungrab socket and destroy it, if it was the last reference. */
1649 static inline void sock_put(struct sock *sk)
1650 {
1651 	if (refcount_dec_and_test(&sk->sk_refcnt))
1652 		sk_free(sk);
1653 }
1654 /* Generic version of sock_put(), dealing with all sockets
1655  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1656  */
1657 void sock_gen_put(struct sock *sk);
1658 
1659 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1660 		     unsigned int trim_cap, bool refcounted);
1661 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1662 				 const int nested)
1663 {
1664 	return __sk_receive_skb(sk, skb, nested, 1, true);
1665 }
1666 
1667 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1668 {
1669 	sk->sk_tx_queue_mapping = tx_queue;
1670 }
1671 
1672 static inline void sk_tx_queue_clear(struct sock *sk)
1673 {
1674 	sk->sk_tx_queue_mapping = -1;
1675 }
1676 
1677 static inline int sk_tx_queue_get(const struct sock *sk)
1678 {
1679 	return sk ? sk->sk_tx_queue_mapping : -1;
1680 }
1681 
1682 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1683 {
1684 	sk_tx_queue_clear(sk);
1685 	sk->sk_socket = sock;
1686 }
1687 
1688 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1689 {
1690 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1691 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1692 }
1693 /* Detach socket from process context.
1694  * Announce socket dead, detach it from wait queue and inode.
1695  * Note that parent inode held reference count on this struct sock,
1696  * we do not release it in this function, because protocol
1697  * probably wants some additional cleanups or even continuing
1698  * to work with this socket (TCP).
1699  */
1700 static inline void sock_orphan(struct sock *sk)
1701 {
1702 	write_lock_bh(&sk->sk_callback_lock);
1703 	sock_set_flag(sk, SOCK_DEAD);
1704 	sk_set_socket(sk, NULL);
1705 	sk->sk_wq  = NULL;
1706 	write_unlock_bh(&sk->sk_callback_lock);
1707 }
1708 
1709 static inline void sock_graft(struct sock *sk, struct socket *parent)
1710 {
1711 	WARN_ON(parent->sk);
1712 	write_lock_bh(&sk->sk_callback_lock);
1713 	sk->sk_wq = parent->wq;
1714 	parent->sk = sk;
1715 	sk_set_socket(sk, parent);
1716 	sk->sk_uid = SOCK_INODE(parent)->i_uid;
1717 	security_sock_graft(sk, parent);
1718 	write_unlock_bh(&sk->sk_callback_lock);
1719 }
1720 
1721 kuid_t sock_i_uid(struct sock *sk);
1722 unsigned long sock_i_ino(struct sock *sk);
1723 
1724 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1725 {
1726 	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1727 }
1728 
1729 static inline u32 net_tx_rndhash(void)
1730 {
1731 	u32 v = prandom_u32();
1732 
1733 	return v ?: 1;
1734 }
1735 
1736 static inline void sk_set_txhash(struct sock *sk)
1737 {
1738 	sk->sk_txhash = net_tx_rndhash();
1739 }
1740 
1741 static inline void sk_rethink_txhash(struct sock *sk)
1742 {
1743 	if (sk->sk_txhash)
1744 		sk_set_txhash(sk);
1745 }
1746 
1747 static inline struct dst_entry *
1748 __sk_dst_get(struct sock *sk)
1749 {
1750 	return rcu_dereference_check(sk->sk_dst_cache,
1751 				     lockdep_sock_is_held(sk));
1752 }
1753 
1754 static inline struct dst_entry *
1755 sk_dst_get(struct sock *sk)
1756 {
1757 	struct dst_entry *dst;
1758 
1759 	rcu_read_lock();
1760 	dst = rcu_dereference(sk->sk_dst_cache);
1761 	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1762 		dst = NULL;
1763 	rcu_read_unlock();
1764 	return dst;
1765 }
1766 
1767 static inline void dst_negative_advice(struct sock *sk)
1768 {
1769 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1770 
1771 	sk_rethink_txhash(sk);
1772 
1773 	if (dst && dst->ops->negative_advice) {
1774 		ndst = dst->ops->negative_advice(dst);
1775 
1776 		if (ndst != dst) {
1777 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1778 			sk_tx_queue_clear(sk);
1779 			sk->sk_dst_pending_confirm = 0;
1780 		}
1781 	}
1782 }
1783 
1784 static inline void
1785 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1786 {
1787 	struct dst_entry *old_dst;
1788 
1789 	sk_tx_queue_clear(sk);
1790 	sk->sk_dst_pending_confirm = 0;
1791 	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1792 					    lockdep_sock_is_held(sk));
1793 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1794 	dst_release(old_dst);
1795 }
1796 
1797 static inline void
1798 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1799 {
1800 	struct dst_entry *old_dst;
1801 
1802 	sk_tx_queue_clear(sk);
1803 	sk->sk_dst_pending_confirm = 0;
1804 	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1805 	dst_release(old_dst);
1806 }
1807 
1808 static inline void
1809 __sk_dst_reset(struct sock *sk)
1810 {
1811 	__sk_dst_set(sk, NULL);
1812 }
1813 
1814 static inline void
1815 sk_dst_reset(struct sock *sk)
1816 {
1817 	sk_dst_set(sk, NULL);
1818 }
1819 
1820 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1821 
1822 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1823 
1824 static inline void sk_dst_confirm(struct sock *sk)
1825 {
1826 	if (!sk->sk_dst_pending_confirm)
1827 		sk->sk_dst_pending_confirm = 1;
1828 }
1829 
1830 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1831 {
1832 	if (skb_get_dst_pending_confirm(skb)) {
1833 		struct sock *sk = skb->sk;
1834 		unsigned long now = jiffies;
1835 
1836 		/* avoid dirtying neighbour */
1837 		if (n->confirmed != now)
1838 			n->confirmed = now;
1839 		if (sk && sk->sk_dst_pending_confirm)
1840 			sk->sk_dst_pending_confirm = 0;
1841 	}
1842 }
1843 
1844 bool sk_mc_loop(struct sock *sk);
1845 
1846 static inline bool sk_can_gso(const struct sock *sk)
1847 {
1848 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1849 }
1850 
1851 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1852 
1853 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1854 {
1855 	sk->sk_route_nocaps |= flags;
1856 	sk->sk_route_caps &= ~flags;
1857 }
1858 
1859 static inline bool sk_check_csum_caps(struct sock *sk)
1860 {
1861 	return (sk->sk_route_caps & NETIF_F_HW_CSUM) ||
1862 	       (sk->sk_family == PF_INET &&
1863 		(sk->sk_route_caps & NETIF_F_IP_CSUM)) ||
1864 	       (sk->sk_family == PF_INET6 &&
1865 		(sk->sk_route_caps & NETIF_F_IPV6_CSUM));
1866 }
1867 
1868 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1869 					   struct iov_iter *from, char *to,
1870 					   int copy, int offset)
1871 {
1872 	if (skb->ip_summed == CHECKSUM_NONE) {
1873 		__wsum csum = 0;
1874 		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1875 			return -EFAULT;
1876 		skb->csum = csum_block_add(skb->csum, csum, offset);
1877 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1878 		if (!copy_from_iter_full_nocache(to, copy, from))
1879 			return -EFAULT;
1880 	} else if (!copy_from_iter_full(to, copy, from))
1881 		return -EFAULT;
1882 
1883 	return 0;
1884 }
1885 
1886 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1887 				       struct iov_iter *from, int copy)
1888 {
1889 	int err, offset = skb->len;
1890 
1891 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1892 				       copy, offset);
1893 	if (err)
1894 		__skb_trim(skb, offset);
1895 
1896 	return err;
1897 }
1898 
1899 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1900 					   struct sk_buff *skb,
1901 					   struct page *page,
1902 					   int off, int copy)
1903 {
1904 	int err;
1905 
1906 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1907 				       copy, skb->len);
1908 	if (err)
1909 		return err;
1910 
1911 	skb->len	     += copy;
1912 	skb->data_len	     += copy;
1913 	skb->truesize	     += copy;
1914 	sk->sk_wmem_queued   += copy;
1915 	sk_mem_charge(sk, copy);
1916 	return 0;
1917 }
1918 
1919 /**
1920  * sk_wmem_alloc_get - returns write allocations
1921  * @sk: socket
1922  *
1923  * Returns sk_wmem_alloc minus initial offset of one
1924  */
1925 static inline int sk_wmem_alloc_get(const struct sock *sk)
1926 {
1927 	return refcount_read(&sk->sk_wmem_alloc) - 1;
1928 }
1929 
1930 /**
1931  * sk_rmem_alloc_get - returns read allocations
1932  * @sk: socket
1933  *
1934  * Returns sk_rmem_alloc
1935  */
1936 static inline int sk_rmem_alloc_get(const struct sock *sk)
1937 {
1938 	return atomic_read(&sk->sk_rmem_alloc);
1939 }
1940 
1941 /**
1942  * sk_has_allocations - check if allocations are outstanding
1943  * @sk: socket
1944  *
1945  * Returns true if socket has write or read allocations
1946  */
1947 static inline bool sk_has_allocations(const struct sock *sk)
1948 {
1949 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1950 }
1951 
1952 /**
1953  * skwq_has_sleeper - check if there are any waiting processes
1954  * @wq: struct socket_wq
1955  *
1956  * Returns true if socket_wq has waiting processes
1957  *
1958  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1959  * barrier call. They were added due to the race found within the tcp code.
1960  *
1961  * Consider following tcp code paths::
1962  *
1963  *   CPU1                CPU2
1964  *   sys_select          receive packet
1965  *   ...                 ...
1966  *   __add_wait_queue    update tp->rcv_nxt
1967  *   ...                 ...
1968  *   tp->rcv_nxt check   sock_def_readable
1969  *   ...                 {
1970  *   schedule               rcu_read_lock();
1971  *                          wq = rcu_dereference(sk->sk_wq);
1972  *                          if (wq && waitqueue_active(&wq->wait))
1973  *                              wake_up_interruptible(&wq->wait)
1974  *                          ...
1975  *                       }
1976  *
1977  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1978  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1979  * could then endup calling schedule and sleep forever if there are no more
1980  * data on the socket.
1981  *
1982  */
1983 static inline bool skwq_has_sleeper(struct socket_wq *wq)
1984 {
1985 	return wq && wq_has_sleeper(&wq->wait);
1986 }
1987 
1988 /**
1989  * sock_poll_wait - place memory barrier behind the poll_wait call.
1990  * @filp:           file
1991  * @wait_address:   socket wait queue
1992  * @p:              poll_table
1993  *
1994  * See the comments in the wq_has_sleeper function.
1995  */
1996 static inline void sock_poll_wait(struct file *filp,
1997 		wait_queue_head_t *wait_address, poll_table *p)
1998 {
1999 	if (!poll_does_not_wait(p) && wait_address) {
2000 		poll_wait(filp, wait_address, p);
2001 		/* We need to be sure we are in sync with the
2002 		 * socket flags modification.
2003 		 *
2004 		 * This memory barrier is paired in the wq_has_sleeper.
2005 		 */
2006 		smp_mb();
2007 	}
2008 }
2009 
2010 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2011 {
2012 	if (sk->sk_txhash) {
2013 		skb->l4_hash = 1;
2014 		skb->hash = sk->sk_txhash;
2015 	}
2016 }
2017 
2018 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2019 
2020 /*
2021  *	Queue a received datagram if it will fit. Stream and sequenced
2022  *	protocols can't normally use this as they need to fit buffers in
2023  *	and play with them.
2024  *
2025  *	Inlined as it's very short and called for pretty much every
2026  *	packet ever received.
2027  */
2028 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2029 {
2030 	skb_orphan(skb);
2031 	skb->sk = sk;
2032 	skb->destructor = sock_rfree;
2033 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2034 	sk_mem_charge(sk, skb->truesize);
2035 }
2036 
2037 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2038 		    unsigned long expires);
2039 
2040 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2041 
2042 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2043 			struct sk_buff *skb, unsigned int flags,
2044 			void (*destructor)(struct sock *sk,
2045 					   struct sk_buff *skb));
2046 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2047 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2048 
2049 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2050 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2051 
2052 /*
2053  *	Recover an error report and clear atomically
2054  */
2055 
2056 static inline int sock_error(struct sock *sk)
2057 {
2058 	int err;
2059 	if (likely(!sk->sk_err))
2060 		return 0;
2061 	err = xchg(&sk->sk_err, 0);
2062 	return -err;
2063 }
2064 
2065 static inline unsigned long sock_wspace(struct sock *sk)
2066 {
2067 	int amt = 0;
2068 
2069 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2070 		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2071 		if (amt < 0)
2072 			amt = 0;
2073 	}
2074 	return amt;
2075 }
2076 
2077 /* Note:
2078  *  We use sk->sk_wq_raw, from contexts knowing this
2079  *  pointer is not NULL and cannot disappear/change.
2080  */
2081 static inline void sk_set_bit(int nr, struct sock *sk)
2082 {
2083 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2084 	    !sock_flag(sk, SOCK_FASYNC))
2085 		return;
2086 
2087 	set_bit(nr, &sk->sk_wq_raw->flags);
2088 }
2089 
2090 static inline void sk_clear_bit(int nr, struct sock *sk)
2091 {
2092 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2093 	    !sock_flag(sk, SOCK_FASYNC))
2094 		return;
2095 
2096 	clear_bit(nr, &sk->sk_wq_raw->flags);
2097 }
2098 
2099 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2100 {
2101 	if (sock_flag(sk, SOCK_FASYNC)) {
2102 		rcu_read_lock();
2103 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2104 		rcu_read_unlock();
2105 	}
2106 }
2107 
2108 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2109  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2110  * Note: for send buffers, TCP works better if we can build two skbs at
2111  * minimum.
2112  */
2113 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2114 
2115 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2116 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2117 
2118 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2119 {
2120 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2121 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2122 		sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2123 	}
2124 }
2125 
2126 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2127 				    bool force_schedule);
2128 
2129 /**
2130  * sk_page_frag - return an appropriate page_frag
2131  * @sk: socket
2132  *
2133  * If socket allocation mode allows current thread to sleep, it means its
2134  * safe to use the per task page_frag instead of the per socket one.
2135  */
2136 static inline struct page_frag *sk_page_frag(struct sock *sk)
2137 {
2138 	if (gfpflags_allow_blocking(sk->sk_allocation))
2139 		return &current->task_frag;
2140 
2141 	return &sk->sk_frag;
2142 }
2143 
2144 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2145 
2146 /*
2147  *	Default write policy as shown to user space via poll/select/SIGIO
2148  */
2149 static inline bool sock_writeable(const struct sock *sk)
2150 {
2151 	return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2152 }
2153 
2154 static inline gfp_t gfp_any(void)
2155 {
2156 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2157 }
2158 
2159 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2160 {
2161 	return noblock ? 0 : sk->sk_rcvtimeo;
2162 }
2163 
2164 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2165 {
2166 	return noblock ? 0 : sk->sk_sndtimeo;
2167 }
2168 
2169 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2170 {
2171 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2172 }
2173 
2174 /* Alas, with timeout socket operations are not restartable.
2175  * Compare this to poll().
2176  */
2177 static inline int sock_intr_errno(long timeo)
2178 {
2179 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2180 }
2181 
2182 struct sock_skb_cb {
2183 	u32 dropcount;
2184 };
2185 
2186 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2187  * using skb->cb[] would keep using it directly and utilize its
2188  * alignement guarantee.
2189  */
2190 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2191 			    sizeof(struct sock_skb_cb)))
2192 
2193 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2194 			    SOCK_SKB_CB_OFFSET))
2195 
2196 #define sock_skb_cb_check_size(size) \
2197 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2198 
2199 static inline void
2200 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2201 {
2202 	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2203 						atomic_read(&sk->sk_drops) : 0;
2204 }
2205 
2206 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2207 {
2208 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2209 
2210 	atomic_add(segs, &sk->sk_drops);
2211 }
2212 
2213 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2214 			   struct sk_buff *skb);
2215 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2216 			     struct sk_buff *skb);
2217 
2218 static inline void
2219 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2220 {
2221 	ktime_t kt = skb->tstamp;
2222 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2223 
2224 	/*
2225 	 * generate control messages if
2226 	 * - receive time stamping in software requested
2227 	 * - software time stamp available and wanted
2228 	 * - hardware time stamps available and wanted
2229 	 */
2230 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2231 	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2232 	    (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2233 	    (hwtstamps->hwtstamp &&
2234 	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2235 		__sock_recv_timestamp(msg, sk, skb);
2236 	else
2237 		sk->sk_stamp = kt;
2238 
2239 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2240 		__sock_recv_wifi_status(msg, sk, skb);
2241 }
2242 
2243 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2244 			      struct sk_buff *skb);
2245 
2246 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2247 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2248 					  struct sk_buff *skb)
2249 {
2250 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2251 			   (1UL << SOCK_RCVTSTAMP))
2252 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2253 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2254 
2255 	if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2256 		__sock_recv_ts_and_drops(msg, sk, skb);
2257 	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2258 		sk->sk_stamp = skb->tstamp;
2259 	else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2260 		sk->sk_stamp = 0;
2261 }
2262 
2263 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2264 
2265 /**
2266  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2267  * @sk:		socket sending this packet
2268  * @tsflags:	timestamping flags to use
2269  * @tx_flags:	completed with instructions for time stamping
2270  *
2271  * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2272  */
2273 static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2274 				     __u8 *tx_flags)
2275 {
2276 	if (unlikely(tsflags))
2277 		__sock_tx_timestamp(tsflags, tx_flags);
2278 	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2279 		*tx_flags |= SKBTX_WIFI_STATUS;
2280 }
2281 
2282 /**
2283  * sk_eat_skb - Release a skb if it is no longer needed
2284  * @sk: socket to eat this skb from
2285  * @skb: socket buffer to eat
2286  *
2287  * This routine must be called with interrupts disabled or with the socket
2288  * locked so that the sk_buff queue operation is ok.
2289 */
2290 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2291 {
2292 	__skb_unlink(skb, &sk->sk_receive_queue);
2293 	__kfree_skb(skb);
2294 }
2295 
2296 static inline
2297 struct net *sock_net(const struct sock *sk)
2298 {
2299 	return read_pnet(&sk->sk_net);
2300 }
2301 
2302 static inline
2303 void sock_net_set(struct sock *sk, struct net *net)
2304 {
2305 	write_pnet(&sk->sk_net, net);
2306 }
2307 
2308 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2309 {
2310 	if (skb->sk) {
2311 		struct sock *sk = skb->sk;
2312 
2313 		skb->destructor = NULL;
2314 		skb->sk = NULL;
2315 		return sk;
2316 	}
2317 	return NULL;
2318 }
2319 
2320 /* This helper checks if a socket is a full socket,
2321  * ie _not_ a timewait or request socket.
2322  */
2323 static inline bool sk_fullsock(const struct sock *sk)
2324 {
2325 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2326 }
2327 
2328 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2329  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2330  */
2331 static inline bool sk_listener(const struct sock *sk)
2332 {
2333 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2334 }
2335 
2336 /**
2337  * sk_state_load - read sk->sk_state for lockless contexts
2338  * @sk: socket pointer
2339  *
2340  * Paired with sk_state_store(). Used in places we do not hold socket lock :
2341  * tcp_diag_get_info(), tcp_get_info(), tcp_poll(), get_tcp4_sock() ...
2342  */
2343 static inline int sk_state_load(const struct sock *sk)
2344 {
2345 	return smp_load_acquire(&sk->sk_state);
2346 }
2347 
2348 /**
2349  * sk_state_store - update sk->sk_state
2350  * @sk: socket pointer
2351  * @newstate: new state
2352  *
2353  * Paired with sk_state_load(). Should be used in contexts where
2354  * state change might impact lockless readers.
2355  */
2356 static inline void sk_state_store(struct sock *sk, int newstate)
2357 {
2358 	smp_store_release(&sk->sk_state, newstate);
2359 }
2360 
2361 void sock_enable_timestamp(struct sock *sk, int flag);
2362 int sock_get_timestamp(struct sock *, struct timeval __user *);
2363 int sock_get_timestampns(struct sock *, struct timespec __user *);
2364 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2365 		       int type);
2366 
2367 bool sk_ns_capable(const struct sock *sk,
2368 		   struct user_namespace *user_ns, int cap);
2369 bool sk_capable(const struct sock *sk, int cap);
2370 bool sk_net_capable(const struct sock *sk, int cap);
2371 
2372 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2373 
2374 /* Take into consideration the size of the struct sk_buff overhead in the
2375  * determination of these values, since that is non-constant across
2376  * platforms.  This makes socket queueing behavior and performance
2377  * not depend upon such differences.
2378  */
2379 #define _SK_MEM_PACKETS		256
2380 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2381 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2382 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2383 
2384 extern __u32 sysctl_wmem_max;
2385 extern __u32 sysctl_rmem_max;
2386 
2387 extern int sysctl_tstamp_allow_data;
2388 extern int sysctl_optmem_max;
2389 
2390 extern __u32 sysctl_wmem_default;
2391 extern __u32 sysctl_rmem_default;
2392 
2393 #endif	/* _SOCK_H */
2394