xref: /linux/include/net/sock.h (revision add452d09a38c7a7c44aea55c1015392cebf9fa7)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Definitions for the AF_INET socket handler.
8  *
9  * Version:	@(#)sock.h	1.0.4	05/13/93
10  *
11  * Authors:	Ross Biro
12  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13  *		Corey Minyard <wf-rch!minyard@relay.EU.net>
14  *		Florian La Roche <flla@stud.uni-sb.de>
15  *
16  * Fixes:
17  *		Alan Cox	:	Volatiles in skbuff pointers. See
18  *					skbuff comments. May be overdone,
19  *					better to prove they can be removed
20  *					than the reverse.
21  *		Alan Cox	:	Added a zapped field for tcp to note
22  *					a socket is reset and must stay shut up
23  *		Alan Cox	:	New fields for options
24  *	Pauline Middelink	:	identd support
25  *		Alan Cox	:	Eliminate low level recv/recvfrom
26  *		David S. Miller	:	New socket lookup architecture.
27  *              Steve Whitehouse:       Default routines for sock_ops
28  *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
29  *              			protinfo be just a void pointer, as the
30  *              			protocol specific parts were moved to
31  *              			respective headers and ipv4/v6, etc now
32  *              			use private slabcaches for its socks
33  *              Pedro Hortas	:	New flags field for socket options
34  */
35 #ifndef _SOCK_H
36 #define _SOCK_H
37 
38 #include <linux/hardirq.h>
39 #include <linux/kernel.h>
40 #include <linux/list.h>
41 #include <linux/list_nulls.h>
42 #include <linux/timer.h>
43 #include <linux/cache.h>
44 #include <linux/bitops.h>
45 #include <linux/lockdep.h>
46 #include <linux/netdevice.h>
47 #include <linux/skbuff.h>	/* struct sk_buff */
48 #include <linux/mm.h>
49 #include <linux/security.h>
50 #include <linux/slab.h>
51 #include <linux/uaccess.h>
52 #include <linux/page_counter.h>
53 #include <linux/memcontrol.h>
54 #include <linux/static_key.h>
55 #include <linux/sched.h>
56 #include <linux/wait.h>
57 #include <linux/cgroup-defs.h>
58 #include <linux/rbtree.h>
59 #include <linux/rculist_nulls.h>
60 #include <linux/poll.h>
61 #include <linux/sockptr.h>
62 #include <linux/indirect_call_wrapper.h>
63 #include <linux/atomic.h>
64 #include <linux/refcount.h>
65 #include <linux/llist.h>
66 #include <net/dst.h>
67 #include <net/checksum.h>
68 #include <net/tcp_states.h>
69 #include <linux/net_tstamp.h>
70 #include <net/l3mdev.h>
71 #include <uapi/linux/socket.h>
72 
73 /*
74  * This structure really needs to be cleaned up.
75  * Most of it is for TCP, and not used by any of
76  * the other protocols.
77  */
78 
79 /* This is the per-socket lock.  The spinlock provides a synchronization
80  * between user contexts and software interrupt processing, whereas the
81  * mini-semaphore synchronizes multiple users amongst themselves.
82  */
83 typedef struct {
84 	spinlock_t		slock;
85 	int			owned;
86 	wait_queue_head_t	wq;
87 	/*
88 	 * We express the mutex-alike socket_lock semantics
89 	 * to the lock validator by explicitly managing
90 	 * the slock as a lock variant (in addition to
91 	 * the slock itself):
92 	 */
93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
94 	struct lockdep_map dep_map;
95 #endif
96 } socket_lock_t;
97 
98 struct sock;
99 struct proto;
100 struct net;
101 
102 typedef __u32 __bitwise __portpair;
103 typedef __u64 __bitwise __addrpair;
104 
105 /**
106  *	struct sock_common - minimal network layer representation of sockets
107  *	@skc_daddr: Foreign IPv4 addr
108  *	@skc_rcv_saddr: Bound local IPv4 addr
109  *	@skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
110  *	@skc_hash: hash value used with various protocol lookup tables
111  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
112  *	@skc_dport: placeholder for inet_dport/tw_dport
113  *	@skc_num: placeholder for inet_num/tw_num
114  *	@skc_portpair: __u32 union of @skc_dport & @skc_num
115  *	@skc_family: network address family
116  *	@skc_state: Connection state
117  *	@skc_reuse: %SO_REUSEADDR setting
118  *	@skc_reuseport: %SO_REUSEPORT setting
119  *	@skc_ipv6only: socket is IPV6 only
120  *	@skc_net_refcnt: socket is using net ref counting
121  *	@skc_bound_dev_if: bound device index if != 0
122  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
123  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
124  *	@skc_prot: protocol handlers inside a network family
125  *	@skc_net: reference to the network namespace of this socket
126  *	@skc_v6_daddr: IPV6 destination address
127  *	@skc_v6_rcv_saddr: IPV6 source address
128  *	@skc_cookie: socket's cookie value
129  *	@skc_node: main hash linkage for various protocol lookup tables
130  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
131  *	@skc_tx_queue_mapping: tx queue number for this connection
132  *	@skc_rx_queue_mapping: rx queue number for this connection
133  *	@skc_flags: place holder for sk_flags
134  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
135  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
136  *	@skc_listener: connection request listener socket (aka rsk_listener)
137  *		[union with @skc_flags]
138  *	@skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
139  *		[union with @skc_flags]
140  *	@skc_incoming_cpu: record/match cpu processing incoming packets
141  *	@skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
142  *		[union with @skc_incoming_cpu]
143  *	@skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
144  *		[union with @skc_incoming_cpu]
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 	union {
152 		__addrpair	skc_addrpair;
153 		struct {
154 			__be32	skc_daddr;
155 			__be32	skc_rcv_saddr;
156 		};
157 	};
158 	union  {
159 		unsigned int	skc_hash;
160 		__u16		skc_u16hashes[2];
161 	};
162 	/* skc_dport && skc_num must be grouped as well */
163 	union {
164 		__portpair	skc_portpair;
165 		struct {
166 			__be16	skc_dport;
167 			__u16	skc_num;
168 		};
169 	};
170 
171 	unsigned short		skc_family;
172 	volatile unsigned char	skc_state;
173 	unsigned char		skc_reuse:4;
174 	unsigned char		skc_reuseport:1;
175 	unsigned char		skc_ipv6only:1;
176 	unsigned char		skc_net_refcnt:1;
177 	int			skc_bound_dev_if;
178 	union {
179 		struct hlist_node	skc_bind_node;
180 		struct hlist_node	skc_portaddr_node;
181 	};
182 	struct proto		*skc_prot;
183 	possible_net_t		skc_net;
184 
185 #if IS_ENABLED(CONFIG_IPV6)
186 	struct in6_addr		skc_v6_daddr;
187 	struct in6_addr		skc_v6_rcv_saddr;
188 #endif
189 
190 	atomic64_t		skc_cookie;
191 
192 	/* following fields are padding to force
193 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
194 	 * assuming IPV6 is enabled. We use this padding differently
195 	 * for different kind of 'sockets'
196 	 */
197 	union {
198 		unsigned long	skc_flags;
199 		struct sock	*skc_listener; /* request_sock */
200 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
201 	};
202 	/*
203 	 * fields between dontcopy_begin/dontcopy_end
204 	 * are not copied in sock_copy()
205 	 */
206 	/* private: */
207 	int			skc_dontcopy_begin[0];
208 	/* public: */
209 	union {
210 		struct hlist_node	skc_node;
211 		struct hlist_nulls_node skc_nulls_node;
212 	};
213 	unsigned short		skc_tx_queue_mapping;
214 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
215 	unsigned short		skc_rx_queue_mapping;
216 #endif
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 struct bpf_local_storage;
235 struct sk_filter;
236 
237 /**
238   *	struct sock - network layer representation of sockets
239   *	@__sk_common: shared layout with inet_timewait_sock
240   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
241   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
242   *	@sk_lock:	synchronizer
243   *	@sk_kern_sock: True if sock is using kernel lock classes
244   *	@sk_rcvbuf: size of receive buffer in bytes
245   *	@sk_wq: sock wait queue and async head
246   *	@sk_rx_dst: receive input route used by early demux
247   *	@sk_rx_dst_ifindex: ifindex for @sk_rx_dst
248   *	@sk_rx_dst_cookie: cookie for @sk_rx_dst
249   *	@sk_dst_cache: destination cache
250   *	@sk_dst_pending_confirm: need to confirm neighbour
251   *	@sk_policy: flow policy
252   *	@sk_receive_queue: incoming packets
253   *	@sk_wmem_alloc: transmit queue bytes committed
254   *	@sk_tsq_flags: TCP Small Queues flags
255   *	@sk_write_queue: Packet sending queue
256   *	@sk_omem_alloc: "o" is "option" or "other"
257   *	@sk_wmem_queued: persistent queue size
258   *	@sk_forward_alloc: space allocated forward
259   *	@sk_reserved_mem: space reserved and non-reclaimable for the socket
260   *	@sk_napi_id: id of the last napi context to receive data for sk
261   *	@sk_ll_usec: usecs to busypoll when there is no data
262   *	@sk_allocation: allocation mode
263   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
264   *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
265   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
266   *	@sk_sndbuf: size of send buffer in bytes
267   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
268   *	@sk_no_check_rx: allow zero checksum in RX packets
269   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
270   *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
271   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
272   *	@sk_gso_max_size: Maximum GSO segment size to build
273   *	@sk_gso_max_segs: Maximum number of GSO segments
274   *	@sk_pacing_shift: scaling factor for TCP Small Queues
275   *	@sk_lingertime: %SO_LINGER l_linger setting
276   *	@sk_backlog: always used with the per-socket spinlock held
277   *	@sk_callback_lock: used with the callbacks in the end of this struct
278   *	@sk_error_queue: rarely used
279   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
280   *			  IPV6_ADDRFORM for instance)
281   *	@sk_err: last error
282   *	@sk_err_soft: errors that don't cause failure but are the cause of a
283   *		      persistent failure not just 'timed out'
284   *	@sk_drops: raw/udp drops counter
285   *	@sk_ack_backlog: current listen backlog
286   *	@sk_max_ack_backlog: listen backlog set in listen()
287   *	@sk_uid: user id of owner
288   *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
289   *	@sk_busy_poll_budget: napi processing budget when busypolling
290   *	@sk_priority: %SO_PRIORITY setting
291   *	@sk_type: socket type (%SOCK_STREAM, etc)
292   *	@sk_protocol: which protocol this socket belongs in this network family
293   *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
294   *	@sk_peer_pid: &struct pid for this socket's peer
295   *	@sk_peer_cred: %SO_PEERCRED setting
296   *	@sk_rcvlowat: %SO_RCVLOWAT setting
297   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
298   *	@sk_sndtimeo: %SO_SNDTIMEO setting
299   *	@sk_txhash: computed flow hash for use on transmit
300   *	@sk_txrehash: enable TX hash rethink
301   *	@sk_filter: socket filtering instructions
302   *	@sk_timer: sock cleanup timer
303   *	@sk_stamp: time stamp of last packet received
304   *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
305   *	@sk_tsflags: SO_TIMESTAMPING flags
306   *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
307   *			   Sockets that can be used under memory reclaim should
308   *			   set this to false.
309   *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
310   *	              for timestamping
311   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
312   *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
313   *	@sk_socket: Identd and reporting IO signals
314   *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
315   *	@sk_frag: cached page frag
316   *	@sk_peek_off: current peek_offset value
317   *	@sk_send_head: front of stuff to transmit
318   *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
319   *	@sk_security: used by security modules
320   *	@sk_mark: generic packet mark
321   *	@sk_cgrp_data: cgroup data for this cgroup
322   *	@sk_memcg: this socket's memory cgroup association
323   *	@sk_write_pending: a write to stream socket waits to start
324   *	@sk_disconnects: number of disconnect operations performed on this sock
325   *	@sk_state_change: callback to indicate change in the state of the sock
326   *	@sk_data_ready: callback to indicate there is data to be processed
327   *	@sk_write_space: callback to indicate there is bf sending space available
328   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
329   *	@sk_backlog_rcv: callback to process the backlog
330   *	@sk_validate_xmit_skb: ptr to an optional validate function
331   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
332   *	@sk_reuseport_cb: reuseport group container
333   *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
334   *	@sk_rcu: used during RCU grace period
335   *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
336   *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
337   *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
338   *	@sk_txtime_unused: unused txtime flags
339   *	@ns_tracker: tracker for netns reference
340   *	@sk_user_frags: xarray of pages the user is holding a reference on.
341   */
342 struct sock {
343 	/*
344 	 * Now struct inet_timewait_sock also uses sock_common, so please just
345 	 * don't add nothing before this first member (__sk_common) --acme
346 	 */
347 	struct sock_common	__sk_common;
348 #define sk_node			__sk_common.skc_node
349 #define sk_nulls_node		__sk_common.skc_nulls_node
350 #define sk_refcnt		__sk_common.skc_refcnt
351 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
352 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
353 #define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
354 #endif
355 
356 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
357 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
358 #define sk_hash			__sk_common.skc_hash
359 #define sk_portpair		__sk_common.skc_portpair
360 #define sk_num			__sk_common.skc_num
361 #define sk_dport		__sk_common.skc_dport
362 #define sk_addrpair		__sk_common.skc_addrpair
363 #define sk_daddr		__sk_common.skc_daddr
364 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
365 #define sk_family		__sk_common.skc_family
366 #define sk_state		__sk_common.skc_state
367 #define sk_reuse		__sk_common.skc_reuse
368 #define sk_reuseport		__sk_common.skc_reuseport
369 #define sk_ipv6only		__sk_common.skc_ipv6only
370 #define sk_net_refcnt		__sk_common.skc_net_refcnt
371 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
372 #define sk_bind_node		__sk_common.skc_bind_node
373 #define sk_prot			__sk_common.skc_prot
374 #define sk_net			__sk_common.skc_net
375 #define sk_v6_daddr		__sk_common.skc_v6_daddr
376 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
377 #define sk_cookie		__sk_common.skc_cookie
378 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
379 #define sk_flags		__sk_common.skc_flags
380 #define sk_rxhash		__sk_common.skc_rxhash
381 
382 	__cacheline_group_begin(sock_write_rx);
383 
384 	atomic_t		sk_drops;
385 	__s32			sk_peek_off;
386 	struct sk_buff_head	sk_error_queue;
387 	struct sk_buff_head	sk_receive_queue;
388 	/*
389 	 * The backlog queue is special, it is always used with
390 	 * the per-socket spinlock held and requires low latency
391 	 * access. Therefore we special case it's implementation.
392 	 * Note : rmem_alloc is in this structure to fill a hole
393 	 * on 64bit arches, not because its logically part of
394 	 * backlog.
395 	 */
396 	struct {
397 		atomic_t	rmem_alloc;
398 		int		len;
399 		struct sk_buff	*head;
400 		struct sk_buff	*tail;
401 	} sk_backlog;
402 #define sk_rmem_alloc sk_backlog.rmem_alloc
403 
404 	__cacheline_group_end(sock_write_rx);
405 
406 	__cacheline_group_begin(sock_read_rx);
407 	/* early demux fields */
408 	struct dst_entry __rcu	*sk_rx_dst;
409 	int			sk_rx_dst_ifindex;
410 	u32			sk_rx_dst_cookie;
411 
412 #ifdef CONFIG_NET_RX_BUSY_POLL
413 	unsigned int		sk_ll_usec;
414 	unsigned int		sk_napi_id;
415 	u16			sk_busy_poll_budget;
416 	u8			sk_prefer_busy_poll;
417 #endif
418 	u8			sk_userlocks;
419 	int			sk_rcvbuf;
420 
421 	struct sk_filter __rcu	*sk_filter;
422 	union {
423 		struct socket_wq __rcu	*sk_wq;
424 		/* private: */
425 		struct socket_wq	*sk_wq_raw;
426 		/* public: */
427 	};
428 
429 	void			(*sk_data_ready)(struct sock *sk);
430 	long			sk_rcvtimeo;
431 	int			sk_rcvlowat;
432 	__cacheline_group_end(sock_read_rx);
433 
434 	__cacheline_group_begin(sock_read_rxtx);
435 	int			sk_err;
436 	struct socket		*sk_socket;
437 	struct mem_cgroup	*sk_memcg;
438 #ifdef CONFIG_XFRM
439 	struct xfrm_policy __rcu *sk_policy[2];
440 #endif
441 	__cacheline_group_end(sock_read_rxtx);
442 
443 	__cacheline_group_begin(sock_write_rxtx);
444 	socket_lock_t		sk_lock;
445 	u32			sk_reserved_mem;
446 	int			sk_forward_alloc;
447 	u32			sk_tsflags;
448 	__cacheline_group_end(sock_write_rxtx);
449 
450 	__cacheline_group_begin(sock_write_tx);
451 	int			sk_write_pending;
452 	atomic_t		sk_omem_alloc;
453 	int			sk_sndbuf;
454 
455 	int			sk_wmem_queued;
456 	refcount_t		sk_wmem_alloc;
457 	unsigned long		sk_tsq_flags;
458 	union {
459 		struct sk_buff	*sk_send_head;
460 		struct rb_root	tcp_rtx_queue;
461 	};
462 	struct sk_buff_head	sk_write_queue;
463 	u32			sk_dst_pending_confirm;
464 	u32			sk_pacing_status; /* see enum sk_pacing */
465 	struct page_frag	sk_frag;
466 	struct timer_list	sk_timer;
467 
468 	unsigned long		sk_pacing_rate; /* bytes per second */
469 	atomic_t		sk_zckey;
470 	atomic_t		sk_tskey;
471 	__cacheline_group_end(sock_write_tx);
472 
473 	__cacheline_group_begin(sock_read_tx);
474 	unsigned long		sk_max_pacing_rate;
475 	long			sk_sndtimeo;
476 	u32			sk_priority;
477 	u32			sk_mark;
478 	struct dst_entry __rcu	*sk_dst_cache;
479 	netdev_features_t	sk_route_caps;
480 #ifdef CONFIG_SOCK_VALIDATE_XMIT
481 	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
482 							struct net_device *dev,
483 							struct sk_buff *skb);
484 #endif
485 	u16			sk_gso_type;
486 	u16			sk_gso_max_segs;
487 	unsigned int		sk_gso_max_size;
488 	gfp_t			sk_allocation;
489 	u32			sk_txhash;
490 	u8			sk_pacing_shift;
491 	bool			sk_use_task_frag;
492 	__cacheline_group_end(sock_read_tx);
493 
494 	/*
495 	 * Because of non atomicity rules, all
496 	 * changes are protected by socket lock.
497 	 */
498 	u8			sk_gso_disabled : 1,
499 				sk_kern_sock : 1,
500 				sk_no_check_tx : 1,
501 				sk_no_check_rx : 1;
502 	u8			sk_shutdown;
503 	u16			sk_type;
504 	u16			sk_protocol;
505 	unsigned long	        sk_lingertime;
506 	struct proto		*sk_prot_creator;
507 	rwlock_t		sk_callback_lock;
508 	int			sk_err_soft;
509 	u32			sk_ack_backlog;
510 	u32			sk_max_ack_backlog;
511 	kuid_t			sk_uid;
512 	spinlock_t		sk_peer_lock;
513 	int			sk_bind_phc;
514 	struct pid		*sk_peer_pid;
515 	const struct cred	*sk_peer_cred;
516 
517 	ktime_t			sk_stamp;
518 #if BITS_PER_LONG==32
519 	seqlock_t		sk_stamp_seq;
520 #endif
521 	int			sk_disconnects;
522 
523 	u8			sk_txrehash;
524 	u8			sk_clockid;
525 	u8			sk_txtime_deadline_mode : 1,
526 				sk_txtime_report_errors : 1,
527 				sk_txtime_unused : 6;
528 
529 	void			*sk_user_data;
530 #ifdef CONFIG_SECURITY
531 	void			*sk_security;
532 #endif
533 	struct sock_cgroup_data	sk_cgrp_data;
534 	void			(*sk_state_change)(struct sock *sk);
535 	void			(*sk_write_space)(struct sock *sk);
536 	void			(*sk_error_report)(struct sock *sk);
537 	int			(*sk_backlog_rcv)(struct sock *sk,
538 						  struct sk_buff *skb);
539 	void                    (*sk_destruct)(struct sock *sk);
540 	struct sock_reuseport __rcu	*sk_reuseport_cb;
541 #ifdef CONFIG_BPF_SYSCALL
542 	struct bpf_local_storage __rcu	*sk_bpf_storage;
543 #endif
544 	struct rcu_head		sk_rcu;
545 	netns_tracker		ns_tracker;
546 	struct xarray		sk_user_frags;
547 };
548 
549 struct sock_bh_locked {
550 	struct sock *sock;
551 	local_lock_t bh_lock;
552 };
553 
554 enum sk_pacing {
555 	SK_PACING_NONE		= 0,
556 	SK_PACING_NEEDED	= 1,
557 	SK_PACING_FQ		= 2,
558 };
559 
560 /* flag bits in sk_user_data
561  *
562  * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
563  *   not be suitable for copying when cloning the socket. For instance,
564  *   it can point to a reference counted object. sk_user_data bottom
565  *   bit is set if pointer must not be copied.
566  *
567  * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
568  *   managed/owned by a BPF reuseport array. This bit should be set
569  *   when sk_user_data's sk is added to the bpf's reuseport_array.
570  *
571  * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
572  *   sk_user_data points to psock type. This bit should be set
573  *   when sk_user_data is assigned to a psock object.
574  */
575 #define SK_USER_DATA_NOCOPY	1UL
576 #define SK_USER_DATA_BPF	2UL
577 #define SK_USER_DATA_PSOCK	4UL
578 #define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
579 				  SK_USER_DATA_PSOCK)
580 
581 /**
582  * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
583  * @sk: socket
584  */
585 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
586 {
587 	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
588 }
589 
590 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
591 
592 /**
593  * __locked_read_sk_user_data_with_flags - return the pointer
594  * only if argument flags all has been set in sk_user_data. Otherwise
595  * return NULL
596  *
597  * @sk: socket
598  * @flags: flag bits
599  *
600  * The caller must be holding sk->sk_callback_lock.
601  */
602 static inline void *
603 __locked_read_sk_user_data_with_flags(const struct sock *sk,
604 				      uintptr_t flags)
605 {
606 	uintptr_t sk_user_data =
607 		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
608 						 lockdep_is_held(&sk->sk_callback_lock));
609 
610 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
611 
612 	if ((sk_user_data & flags) == flags)
613 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
614 	return NULL;
615 }
616 
617 /**
618  * __rcu_dereference_sk_user_data_with_flags - return the pointer
619  * only if argument flags all has been set in sk_user_data. Otherwise
620  * return NULL
621  *
622  * @sk: socket
623  * @flags: flag bits
624  */
625 static inline void *
626 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
627 					  uintptr_t flags)
628 {
629 	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
630 
631 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
632 
633 	if ((sk_user_data & flags) == flags)
634 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
635 	return NULL;
636 }
637 
638 #define rcu_dereference_sk_user_data(sk)				\
639 	__rcu_dereference_sk_user_data_with_flags(sk, 0)
640 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
641 ({									\
642 	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
643 		  __tmp2 = (uintptr_t)(flags);				\
644 	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
645 	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
646 	rcu_assign_pointer(__sk_user_data((sk)),			\
647 			   __tmp1 | __tmp2);				\
648 })
649 #define rcu_assign_sk_user_data(sk, ptr)				\
650 	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
651 
652 static inline
653 struct net *sock_net(const struct sock *sk)
654 {
655 	return read_pnet(&sk->sk_net);
656 }
657 
658 static inline
659 void sock_net_set(struct sock *sk, struct net *net)
660 {
661 	write_pnet(&sk->sk_net, net);
662 }
663 
664 /*
665  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
666  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
667  * on a socket means that the socket will reuse everybody else's port
668  * without looking at the other's sk_reuse value.
669  */
670 
671 #define SK_NO_REUSE	0
672 #define SK_CAN_REUSE	1
673 #define SK_FORCE_REUSE	2
674 
675 int sk_set_peek_off(struct sock *sk, int val);
676 
677 static inline int sk_peek_offset(const struct sock *sk, int flags)
678 {
679 	if (unlikely(flags & MSG_PEEK)) {
680 		return READ_ONCE(sk->sk_peek_off);
681 	}
682 
683 	return 0;
684 }
685 
686 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
687 {
688 	s32 off = READ_ONCE(sk->sk_peek_off);
689 
690 	if (unlikely(off >= 0)) {
691 		off = max_t(s32, off - val, 0);
692 		WRITE_ONCE(sk->sk_peek_off, off);
693 	}
694 }
695 
696 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
697 {
698 	sk_peek_offset_bwd(sk, -val);
699 }
700 
701 /*
702  * Hashed lists helper routines
703  */
704 static inline struct sock *sk_entry(const struct hlist_node *node)
705 {
706 	return hlist_entry(node, struct sock, sk_node);
707 }
708 
709 static inline struct sock *__sk_head(const struct hlist_head *head)
710 {
711 	return hlist_entry(head->first, struct sock, sk_node);
712 }
713 
714 static inline struct sock *sk_head(const struct hlist_head *head)
715 {
716 	return hlist_empty(head) ? NULL : __sk_head(head);
717 }
718 
719 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
720 {
721 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
722 }
723 
724 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
725 {
726 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
727 }
728 
729 static inline struct sock *sk_next(const struct sock *sk)
730 {
731 	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
732 }
733 
734 static inline struct sock *sk_nulls_next(const struct sock *sk)
735 {
736 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
737 		hlist_nulls_entry(sk->sk_nulls_node.next,
738 				  struct sock, sk_nulls_node) :
739 		NULL;
740 }
741 
742 static inline bool sk_unhashed(const struct sock *sk)
743 {
744 	return hlist_unhashed(&sk->sk_node);
745 }
746 
747 static inline bool sk_hashed(const struct sock *sk)
748 {
749 	return !sk_unhashed(sk);
750 }
751 
752 static inline void sk_node_init(struct hlist_node *node)
753 {
754 	node->pprev = NULL;
755 }
756 
757 static inline void __sk_del_node(struct sock *sk)
758 {
759 	__hlist_del(&sk->sk_node);
760 }
761 
762 /* NB: equivalent to hlist_del_init_rcu */
763 static inline bool __sk_del_node_init(struct sock *sk)
764 {
765 	if (sk_hashed(sk)) {
766 		__sk_del_node(sk);
767 		sk_node_init(&sk->sk_node);
768 		return true;
769 	}
770 	return false;
771 }
772 
773 /* Grab socket reference count. This operation is valid only
774    when sk is ALREADY grabbed f.e. it is found in hash table
775    or a list and the lookup is made under lock preventing hash table
776    modifications.
777  */
778 
779 static __always_inline void sock_hold(struct sock *sk)
780 {
781 	refcount_inc(&sk->sk_refcnt);
782 }
783 
784 /* Ungrab socket in the context, which assumes that socket refcnt
785    cannot hit zero, f.e. it is true in context of any socketcall.
786  */
787 static __always_inline void __sock_put(struct sock *sk)
788 {
789 	refcount_dec(&sk->sk_refcnt);
790 }
791 
792 static inline bool sk_del_node_init(struct sock *sk)
793 {
794 	bool rc = __sk_del_node_init(sk);
795 
796 	if (rc) {
797 		/* paranoid for a while -acme */
798 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
799 		__sock_put(sk);
800 	}
801 	return rc;
802 }
803 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
804 
805 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
806 {
807 	if (sk_hashed(sk)) {
808 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
809 		return true;
810 	}
811 	return false;
812 }
813 
814 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
815 {
816 	bool rc = __sk_nulls_del_node_init_rcu(sk);
817 
818 	if (rc) {
819 		/* paranoid for a while -acme */
820 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
821 		__sock_put(sk);
822 	}
823 	return rc;
824 }
825 
826 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
827 {
828 	hlist_add_head(&sk->sk_node, list);
829 }
830 
831 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
832 {
833 	sock_hold(sk);
834 	__sk_add_node(sk, list);
835 }
836 
837 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
838 {
839 	sock_hold(sk);
840 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
841 	    sk->sk_family == AF_INET6)
842 		hlist_add_tail_rcu(&sk->sk_node, list);
843 	else
844 		hlist_add_head_rcu(&sk->sk_node, list);
845 }
846 
847 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
848 {
849 	sock_hold(sk);
850 	hlist_add_tail_rcu(&sk->sk_node, list);
851 }
852 
853 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
854 {
855 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
856 }
857 
858 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
859 {
860 	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
861 }
862 
863 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
864 {
865 	sock_hold(sk);
866 	__sk_nulls_add_node_rcu(sk, list);
867 }
868 
869 static inline void __sk_del_bind_node(struct sock *sk)
870 {
871 	__hlist_del(&sk->sk_bind_node);
872 }
873 
874 static inline void sk_add_bind_node(struct sock *sk,
875 					struct hlist_head *list)
876 {
877 	hlist_add_head(&sk->sk_bind_node, list);
878 }
879 
880 #define sk_for_each(__sk, list) \
881 	hlist_for_each_entry(__sk, list, sk_node)
882 #define sk_for_each_rcu(__sk, list) \
883 	hlist_for_each_entry_rcu(__sk, list, sk_node)
884 #define sk_nulls_for_each(__sk, node, list) \
885 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
886 #define sk_nulls_for_each_rcu(__sk, node, list) \
887 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
888 #define sk_for_each_from(__sk) \
889 	hlist_for_each_entry_from(__sk, sk_node)
890 #define sk_nulls_for_each_from(__sk, node) \
891 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
892 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
893 #define sk_for_each_safe(__sk, tmp, list) \
894 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
895 #define sk_for_each_bound(__sk, list) \
896 	hlist_for_each_entry(__sk, list, sk_bind_node)
897 
898 /**
899  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
900  * @tpos:	the type * to use as a loop cursor.
901  * @pos:	the &struct hlist_node to use as a loop cursor.
902  * @head:	the head for your list.
903  * @offset:	offset of hlist_node within the struct.
904  *
905  */
906 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
907 	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
908 	     pos != NULL &&						       \
909 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
910 	     pos = rcu_dereference(hlist_next_rcu(pos)))
911 
912 static inline struct user_namespace *sk_user_ns(const struct sock *sk)
913 {
914 	/* Careful only use this in a context where these parameters
915 	 * can not change and must all be valid, such as recvmsg from
916 	 * userspace.
917 	 */
918 	return sk->sk_socket->file->f_cred->user_ns;
919 }
920 
921 /* Sock flags */
922 enum sock_flags {
923 	SOCK_DEAD,
924 	SOCK_DONE,
925 	SOCK_URGINLINE,
926 	SOCK_KEEPOPEN,
927 	SOCK_LINGER,
928 	SOCK_DESTROY,
929 	SOCK_BROADCAST,
930 	SOCK_TIMESTAMP,
931 	SOCK_ZAPPED,
932 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
933 	SOCK_DBG, /* %SO_DEBUG setting */
934 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
935 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
936 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
937 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
938 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
939 	SOCK_FASYNC, /* fasync() active */
940 	SOCK_RXQ_OVFL,
941 	SOCK_ZEROCOPY, /* buffers from userspace */
942 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
943 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
944 		     * Will use last 4 bytes of packet sent from
945 		     * user-space instead.
946 		     */
947 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
948 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
949 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
950 	SOCK_TXTIME,
951 	SOCK_XDP, /* XDP is attached */
952 	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
953 	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
954 };
955 
956 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
957 /*
958  * The highest bit of sk_tsflags is reserved for kernel-internal
959  * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that
960  * SOF_TIMESTAMPING* values do not reach this reserved area
961  */
962 #define SOCKCM_FLAG_TS_OPT_ID	BIT(31)
963 
964 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
965 {
966 	nsk->sk_flags = osk->sk_flags;
967 }
968 
969 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
970 {
971 	__set_bit(flag, &sk->sk_flags);
972 }
973 
974 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
975 {
976 	__clear_bit(flag, &sk->sk_flags);
977 }
978 
979 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
980 				     int valbool)
981 {
982 	if (valbool)
983 		sock_set_flag(sk, bit);
984 	else
985 		sock_reset_flag(sk, bit);
986 }
987 
988 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
989 {
990 	return test_bit(flag, &sk->sk_flags);
991 }
992 
993 #ifdef CONFIG_NET
994 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
995 static inline int sk_memalloc_socks(void)
996 {
997 	return static_branch_unlikely(&memalloc_socks_key);
998 }
999 
1000 void __receive_sock(struct file *file);
1001 #else
1002 
1003 static inline int sk_memalloc_socks(void)
1004 {
1005 	return 0;
1006 }
1007 
1008 static inline void __receive_sock(struct file *file)
1009 { }
1010 #endif
1011 
1012 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1013 {
1014 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1015 }
1016 
1017 static inline void sk_acceptq_removed(struct sock *sk)
1018 {
1019 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1020 }
1021 
1022 static inline void sk_acceptq_added(struct sock *sk)
1023 {
1024 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1025 }
1026 
1027 /* Note: If you think the test should be:
1028  *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1029  * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1030  */
1031 static inline bool sk_acceptq_is_full(const struct sock *sk)
1032 {
1033 	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1034 }
1035 
1036 /*
1037  * Compute minimal free write space needed to queue new packets.
1038  */
1039 static inline int sk_stream_min_wspace(const struct sock *sk)
1040 {
1041 	return READ_ONCE(sk->sk_wmem_queued) >> 1;
1042 }
1043 
1044 static inline int sk_stream_wspace(const struct sock *sk)
1045 {
1046 	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1047 }
1048 
1049 static inline void sk_wmem_queued_add(struct sock *sk, int val)
1050 {
1051 	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1052 }
1053 
1054 static inline void sk_forward_alloc_add(struct sock *sk, int val)
1055 {
1056 	/* Paired with lockless reads of sk->sk_forward_alloc */
1057 	WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1058 }
1059 
1060 void sk_stream_write_space(struct sock *sk);
1061 
1062 /* OOB backlog add */
1063 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1064 {
1065 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
1066 	skb_dst_force(skb);
1067 
1068 	if (!sk->sk_backlog.tail)
1069 		WRITE_ONCE(sk->sk_backlog.head, skb);
1070 	else
1071 		sk->sk_backlog.tail->next = skb;
1072 
1073 	WRITE_ONCE(sk->sk_backlog.tail, skb);
1074 	skb->next = NULL;
1075 }
1076 
1077 /*
1078  * Take into account size of receive queue and backlog queue
1079  * Do not take into account this skb truesize,
1080  * to allow even a single big packet to come.
1081  */
1082 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1083 {
1084 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1085 
1086 	return qsize > limit;
1087 }
1088 
1089 /* The per-socket spinlock must be held here. */
1090 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1091 					      unsigned int limit)
1092 {
1093 	if (sk_rcvqueues_full(sk, limit))
1094 		return -ENOBUFS;
1095 
1096 	/*
1097 	 * If the skb was allocated from pfmemalloc reserves, only
1098 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1099 	 * helping free memory
1100 	 */
1101 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1102 		return -ENOMEM;
1103 
1104 	__sk_add_backlog(sk, skb);
1105 	sk->sk_backlog.len += skb->truesize;
1106 	return 0;
1107 }
1108 
1109 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1110 
1111 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1112 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1113 
1114 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1115 {
1116 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1117 		return __sk_backlog_rcv(sk, skb);
1118 
1119 	return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1120 				  tcp_v6_do_rcv,
1121 				  tcp_v4_do_rcv,
1122 				  sk, skb);
1123 }
1124 
1125 static inline void sk_incoming_cpu_update(struct sock *sk)
1126 {
1127 	int cpu = raw_smp_processor_id();
1128 
1129 	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1130 		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1131 }
1132 
1133 
1134 static inline void sock_rps_save_rxhash(struct sock *sk,
1135 					const struct sk_buff *skb)
1136 {
1137 #ifdef CONFIG_RPS
1138 	/* The following WRITE_ONCE() is paired with the READ_ONCE()
1139 	 * here, and another one in sock_rps_record_flow().
1140 	 */
1141 	if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1142 		WRITE_ONCE(sk->sk_rxhash, skb->hash);
1143 #endif
1144 }
1145 
1146 static inline void sock_rps_reset_rxhash(struct sock *sk)
1147 {
1148 #ifdef CONFIG_RPS
1149 	/* Paired with READ_ONCE() in sock_rps_record_flow() */
1150 	WRITE_ONCE(sk->sk_rxhash, 0);
1151 #endif
1152 }
1153 
1154 #define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1155 	({	int __rc, __dis = __sk->sk_disconnects;			\
1156 		release_sock(__sk);					\
1157 		__rc = __condition;					\
1158 		if (!__rc) {						\
1159 			*(__timeo) = wait_woken(__wait,			\
1160 						TASK_INTERRUPTIBLE,	\
1161 						*(__timeo));		\
1162 		}							\
1163 		sched_annotate_sleep();					\
1164 		lock_sock(__sk);					\
1165 		__rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1166 		__rc;							\
1167 	})
1168 
1169 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1170 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1171 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1172 int sk_stream_error(struct sock *sk, int flags, int err);
1173 void sk_stream_kill_queues(struct sock *sk);
1174 void sk_set_memalloc(struct sock *sk);
1175 void sk_clear_memalloc(struct sock *sk);
1176 
1177 void __sk_flush_backlog(struct sock *sk);
1178 
1179 static inline bool sk_flush_backlog(struct sock *sk)
1180 {
1181 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1182 		__sk_flush_backlog(sk);
1183 		return true;
1184 	}
1185 	return false;
1186 }
1187 
1188 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1189 
1190 struct request_sock_ops;
1191 struct timewait_sock_ops;
1192 struct inet_hashinfo;
1193 struct raw_hashinfo;
1194 struct smc_hashinfo;
1195 struct module;
1196 struct sk_psock;
1197 
1198 /*
1199  * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1200  * un-modified. Special care is taken when initializing object to zero.
1201  */
1202 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1203 {
1204 	if (offsetof(struct sock, sk_node.next) != 0)
1205 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1206 	memset(&sk->sk_node.pprev, 0,
1207 	       size - offsetof(struct sock, sk_node.pprev));
1208 }
1209 
1210 struct proto_accept_arg {
1211 	int flags;
1212 	int err;
1213 	int is_empty;
1214 	bool kern;
1215 };
1216 
1217 /* Networking protocol blocks we attach to sockets.
1218  * socket layer -> transport layer interface
1219  */
1220 struct proto {
1221 	void			(*close)(struct sock *sk,
1222 					long timeout);
1223 	int			(*pre_connect)(struct sock *sk,
1224 					struct sockaddr *uaddr,
1225 					int addr_len);
1226 	int			(*connect)(struct sock *sk,
1227 					struct sockaddr *uaddr,
1228 					int addr_len);
1229 	int			(*disconnect)(struct sock *sk, int flags);
1230 
1231 	struct sock *		(*accept)(struct sock *sk,
1232 					  struct proto_accept_arg *arg);
1233 
1234 	int			(*ioctl)(struct sock *sk, int cmd,
1235 					 int *karg);
1236 	int			(*init)(struct sock *sk);
1237 	void			(*destroy)(struct sock *sk);
1238 	void			(*shutdown)(struct sock *sk, int how);
1239 	int			(*setsockopt)(struct sock *sk, int level,
1240 					int optname, sockptr_t optval,
1241 					unsigned int optlen);
1242 	int			(*getsockopt)(struct sock *sk, int level,
1243 					int optname, char __user *optval,
1244 					int __user *option);
1245 	void			(*keepalive)(struct sock *sk, int valbool);
1246 #ifdef CONFIG_COMPAT
1247 	int			(*compat_ioctl)(struct sock *sk,
1248 					unsigned int cmd, unsigned long arg);
1249 #endif
1250 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1251 					   size_t len);
1252 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1253 					   size_t len, int flags, int *addr_len);
1254 	void			(*splice_eof)(struct socket *sock);
1255 	int			(*bind)(struct sock *sk,
1256 					struct sockaddr *addr, int addr_len);
1257 	int			(*bind_add)(struct sock *sk,
1258 					struct sockaddr *addr, int addr_len);
1259 
1260 	int			(*backlog_rcv) (struct sock *sk,
1261 						struct sk_buff *skb);
1262 	bool			(*bpf_bypass_getsockopt)(int level,
1263 							 int optname);
1264 
1265 	void		(*release_cb)(struct sock *sk);
1266 
1267 	/* Keeping track of sk's, looking them up, and port selection methods. */
1268 	int			(*hash)(struct sock *sk);
1269 	void			(*unhash)(struct sock *sk);
1270 	void			(*rehash)(struct sock *sk);
1271 	int			(*get_port)(struct sock *sk, unsigned short snum);
1272 	void			(*put_port)(struct sock *sk);
1273 #ifdef CONFIG_BPF_SYSCALL
1274 	int			(*psock_update_sk_prot)(struct sock *sk,
1275 							struct sk_psock *psock,
1276 							bool restore);
1277 #endif
1278 
1279 	/* Keeping track of sockets in use */
1280 #ifdef CONFIG_PROC_FS
1281 	unsigned int		inuse_idx;
1282 #endif
1283 
1284 #if IS_ENABLED(CONFIG_MPTCP)
1285 	int			(*forward_alloc_get)(const struct sock *sk);
1286 #endif
1287 
1288 	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1289 	bool			(*sock_is_readable)(struct sock *sk);
1290 	/* Memory pressure */
1291 	void			(*enter_memory_pressure)(struct sock *sk);
1292 	void			(*leave_memory_pressure)(struct sock *sk);
1293 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1294 	int  __percpu		*per_cpu_fw_alloc;
1295 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1296 
1297 	/*
1298 	 * Pressure flag: try to collapse.
1299 	 * Technical note: it is used by multiple contexts non atomically.
1300 	 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1301 	 * All the __sk_mem_schedule() is of this nature: accounting
1302 	 * is strict, actions are advisory and have some latency.
1303 	 */
1304 	unsigned long		*memory_pressure;
1305 	long			*sysctl_mem;
1306 
1307 	int			*sysctl_wmem;
1308 	int			*sysctl_rmem;
1309 	u32			sysctl_wmem_offset;
1310 	u32			sysctl_rmem_offset;
1311 
1312 	int			max_header;
1313 	bool			no_autobind;
1314 
1315 	struct kmem_cache	*slab;
1316 	unsigned int		obj_size;
1317 	unsigned int		ipv6_pinfo_offset;
1318 	slab_flags_t		slab_flags;
1319 	unsigned int		useroffset;	/* Usercopy region offset */
1320 	unsigned int		usersize;	/* Usercopy region size */
1321 
1322 	unsigned int __percpu	*orphan_count;
1323 
1324 	struct request_sock_ops	*rsk_prot;
1325 	struct timewait_sock_ops *twsk_prot;
1326 
1327 	union {
1328 		struct inet_hashinfo	*hashinfo;
1329 		struct udp_table	*udp_table;
1330 		struct raw_hashinfo	*raw_hash;
1331 		struct smc_hashinfo	*smc_hash;
1332 	} h;
1333 
1334 	struct module		*owner;
1335 
1336 	char			name[32];
1337 
1338 	struct list_head	node;
1339 	int			(*diag_destroy)(struct sock *sk, int err);
1340 } __randomize_layout;
1341 
1342 int proto_register(struct proto *prot, int alloc_slab);
1343 void proto_unregister(struct proto *prot);
1344 int sock_load_diag_module(int family, int protocol);
1345 
1346 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1347 
1348 static inline int sk_forward_alloc_get(const struct sock *sk)
1349 {
1350 #if IS_ENABLED(CONFIG_MPTCP)
1351 	if (sk->sk_prot->forward_alloc_get)
1352 		return sk->sk_prot->forward_alloc_get(sk);
1353 #endif
1354 	return READ_ONCE(sk->sk_forward_alloc);
1355 }
1356 
1357 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1358 {
1359 	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1360 		return false;
1361 
1362 	return sk->sk_prot->stream_memory_free ?
1363 		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1364 				     tcp_stream_memory_free, sk, wake) : true;
1365 }
1366 
1367 static inline bool sk_stream_memory_free(const struct sock *sk)
1368 {
1369 	return __sk_stream_memory_free(sk, 0);
1370 }
1371 
1372 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1373 {
1374 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1375 	       __sk_stream_memory_free(sk, wake);
1376 }
1377 
1378 static inline bool sk_stream_is_writeable(const struct sock *sk)
1379 {
1380 	return __sk_stream_is_writeable(sk, 0);
1381 }
1382 
1383 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1384 					    struct cgroup *ancestor)
1385 {
1386 #ifdef CONFIG_SOCK_CGROUP_DATA
1387 	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1388 				    ancestor);
1389 #else
1390 	return -ENOTSUPP;
1391 #endif
1392 }
1393 
1394 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1395 
1396 static inline void sk_sockets_allocated_dec(struct sock *sk)
1397 {
1398 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1399 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1400 }
1401 
1402 static inline void sk_sockets_allocated_inc(struct sock *sk)
1403 {
1404 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1405 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1406 }
1407 
1408 static inline u64
1409 sk_sockets_allocated_read_positive(struct sock *sk)
1410 {
1411 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1412 }
1413 
1414 static inline int
1415 proto_sockets_allocated_sum_positive(struct proto *prot)
1416 {
1417 	return percpu_counter_sum_positive(prot->sockets_allocated);
1418 }
1419 
1420 #ifdef CONFIG_PROC_FS
1421 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
1422 struct prot_inuse {
1423 	int all;
1424 	int val[PROTO_INUSE_NR];
1425 };
1426 
1427 static inline void sock_prot_inuse_add(const struct net *net,
1428 				       const struct proto *prot, int val)
1429 {
1430 	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1431 }
1432 
1433 static inline void sock_inuse_add(const struct net *net, int val)
1434 {
1435 	this_cpu_add(net->core.prot_inuse->all, val);
1436 }
1437 
1438 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1439 int sock_inuse_get(struct net *net);
1440 #else
1441 static inline void sock_prot_inuse_add(const struct net *net,
1442 				       const struct proto *prot, int val)
1443 {
1444 }
1445 
1446 static inline void sock_inuse_add(const struct net *net, int val)
1447 {
1448 }
1449 #endif
1450 
1451 
1452 /* With per-bucket locks this operation is not-atomic, so that
1453  * this version is not worse.
1454  */
1455 static inline int __sk_prot_rehash(struct sock *sk)
1456 {
1457 	sk->sk_prot->unhash(sk);
1458 	return sk->sk_prot->hash(sk);
1459 }
1460 
1461 /* About 10 seconds */
1462 #define SOCK_DESTROY_TIME (10*HZ)
1463 
1464 /* Sockets 0-1023 can't be bound to unless you are superuser */
1465 #define PROT_SOCK	1024
1466 
1467 #define SHUTDOWN_MASK	3
1468 #define RCV_SHUTDOWN	1
1469 #define SEND_SHUTDOWN	2
1470 
1471 #define SOCK_BINDADDR_LOCK	4
1472 #define SOCK_BINDPORT_LOCK	8
1473 
1474 struct socket_alloc {
1475 	struct socket socket;
1476 	struct inode vfs_inode;
1477 };
1478 
1479 static inline struct socket *SOCKET_I(struct inode *inode)
1480 {
1481 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1482 }
1483 
1484 static inline struct inode *SOCK_INODE(struct socket *socket)
1485 {
1486 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1487 }
1488 
1489 /*
1490  * Functions for memory accounting
1491  */
1492 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1493 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1494 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1495 void __sk_mem_reclaim(struct sock *sk, int amount);
1496 
1497 #define SK_MEM_SEND	0
1498 #define SK_MEM_RECV	1
1499 
1500 /* sysctl_mem values are in pages */
1501 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1502 {
1503 	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1504 }
1505 
1506 static inline int sk_mem_pages(int amt)
1507 {
1508 	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1509 }
1510 
1511 static inline bool sk_has_account(struct sock *sk)
1512 {
1513 	/* return true if protocol supports memory accounting */
1514 	return !!sk->sk_prot->memory_allocated;
1515 }
1516 
1517 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1518 {
1519 	int delta;
1520 
1521 	if (!sk_has_account(sk))
1522 		return true;
1523 	delta = size - sk->sk_forward_alloc;
1524 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1525 }
1526 
1527 static inline bool
1528 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1529 {
1530 	int delta;
1531 
1532 	if (!sk_has_account(sk))
1533 		return true;
1534 	delta = size - sk->sk_forward_alloc;
1535 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1536 		skb_pfmemalloc(skb);
1537 }
1538 
1539 static inline int sk_unused_reserved_mem(const struct sock *sk)
1540 {
1541 	int unused_mem;
1542 
1543 	if (likely(!sk->sk_reserved_mem))
1544 		return 0;
1545 
1546 	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1547 			atomic_read(&sk->sk_rmem_alloc);
1548 
1549 	return unused_mem > 0 ? unused_mem : 0;
1550 }
1551 
1552 static inline void sk_mem_reclaim(struct sock *sk)
1553 {
1554 	int reclaimable;
1555 
1556 	if (!sk_has_account(sk))
1557 		return;
1558 
1559 	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1560 
1561 	if (reclaimable >= (int)PAGE_SIZE)
1562 		__sk_mem_reclaim(sk, reclaimable);
1563 }
1564 
1565 static inline void sk_mem_reclaim_final(struct sock *sk)
1566 {
1567 	sk->sk_reserved_mem = 0;
1568 	sk_mem_reclaim(sk);
1569 }
1570 
1571 static inline void sk_mem_charge(struct sock *sk, int size)
1572 {
1573 	if (!sk_has_account(sk))
1574 		return;
1575 	sk_forward_alloc_add(sk, -size);
1576 }
1577 
1578 static inline void sk_mem_uncharge(struct sock *sk, int size)
1579 {
1580 	if (!sk_has_account(sk))
1581 		return;
1582 	sk_forward_alloc_add(sk, size);
1583 	sk_mem_reclaim(sk);
1584 }
1585 
1586 /*
1587  * Macro so as to not evaluate some arguments when
1588  * lockdep is not enabled.
1589  *
1590  * Mark both the sk_lock and the sk_lock.slock as a
1591  * per-address-family lock class.
1592  */
1593 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1594 do {									\
1595 	sk->sk_lock.owned = 0;						\
1596 	init_waitqueue_head(&sk->sk_lock.wq);				\
1597 	spin_lock_init(&(sk)->sk_lock.slock);				\
1598 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1599 			sizeof((sk)->sk_lock));				\
1600 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1601 				(skey), (sname));				\
1602 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1603 } while (0)
1604 
1605 static inline bool lockdep_sock_is_held(const struct sock *sk)
1606 {
1607 	return lockdep_is_held(&sk->sk_lock) ||
1608 	       lockdep_is_held(&sk->sk_lock.slock);
1609 }
1610 
1611 void lock_sock_nested(struct sock *sk, int subclass);
1612 
1613 static inline void lock_sock(struct sock *sk)
1614 {
1615 	lock_sock_nested(sk, 0);
1616 }
1617 
1618 void __lock_sock(struct sock *sk);
1619 void __release_sock(struct sock *sk);
1620 void release_sock(struct sock *sk);
1621 
1622 /* BH context may only use the following locking interface. */
1623 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1624 #define bh_lock_sock_nested(__sk) \
1625 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1626 				SINGLE_DEPTH_NESTING)
1627 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1628 
1629 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1630 
1631 /**
1632  * lock_sock_fast - fast version of lock_sock
1633  * @sk: socket
1634  *
1635  * This version should be used for very small section, where process won't block
1636  * return false if fast path is taken:
1637  *
1638  *   sk_lock.slock locked, owned = 0, BH disabled
1639  *
1640  * return true if slow path is taken:
1641  *
1642  *   sk_lock.slock unlocked, owned = 1, BH enabled
1643  */
1644 static inline bool lock_sock_fast(struct sock *sk)
1645 {
1646 	/* The sk_lock has mutex_lock() semantics here. */
1647 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1648 
1649 	return __lock_sock_fast(sk);
1650 }
1651 
1652 /* fast socket lock variant for caller already holding a [different] socket lock */
1653 static inline bool lock_sock_fast_nested(struct sock *sk)
1654 {
1655 	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1656 
1657 	return __lock_sock_fast(sk);
1658 }
1659 
1660 /**
1661  * unlock_sock_fast - complement of lock_sock_fast
1662  * @sk: socket
1663  * @slow: slow mode
1664  *
1665  * fast unlock socket for user context.
1666  * If slow mode is on, we call regular release_sock()
1667  */
1668 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1669 	__releases(&sk->sk_lock.slock)
1670 {
1671 	if (slow) {
1672 		release_sock(sk);
1673 		__release(&sk->sk_lock.slock);
1674 	} else {
1675 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1676 		spin_unlock_bh(&sk->sk_lock.slock);
1677 	}
1678 }
1679 
1680 void sockopt_lock_sock(struct sock *sk);
1681 void sockopt_release_sock(struct sock *sk);
1682 bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1683 bool sockopt_capable(int cap);
1684 
1685 /* Used by processes to "lock" a socket state, so that
1686  * interrupts and bottom half handlers won't change it
1687  * from under us. It essentially blocks any incoming
1688  * packets, so that we won't get any new data or any
1689  * packets that change the state of the socket.
1690  *
1691  * While locked, BH processing will add new packets to
1692  * the backlog queue.  This queue is processed by the
1693  * owner of the socket lock right before it is released.
1694  *
1695  * Since ~2.3.5 it is also exclusive sleep lock serializing
1696  * accesses from user process context.
1697  */
1698 
1699 static inline void sock_owned_by_me(const struct sock *sk)
1700 {
1701 #ifdef CONFIG_LOCKDEP
1702 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1703 #endif
1704 }
1705 
1706 static inline void sock_not_owned_by_me(const struct sock *sk)
1707 {
1708 #ifdef CONFIG_LOCKDEP
1709 	WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1710 #endif
1711 }
1712 
1713 static inline bool sock_owned_by_user(const struct sock *sk)
1714 {
1715 	sock_owned_by_me(sk);
1716 	return sk->sk_lock.owned;
1717 }
1718 
1719 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1720 {
1721 	return sk->sk_lock.owned;
1722 }
1723 
1724 static inline void sock_release_ownership(struct sock *sk)
1725 {
1726 	DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1727 	sk->sk_lock.owned = 0;
1728 
1729 	/* The sk_lock has mutex_unlock() semantics: */
1730 	mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1731 }
1732 
1733 /* no reclassification while locks are held */
1734 static inline bool sock_allow_reclassification(const struct sock *csk)
1735 {
1736 	struct sock *sk = (struct sock *)csk;
1737 
1738 	return !sock_owned_by_user_nocheck(sk) &&
1739 		!spin_is_locked(&sk->sk_lock.slock);
1740 }
1741 
1742 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1743 		      struct proto *prot, int kern);
1744 void sk_free(struct sock *sk);
1745 void sk_destruct(struct sock *sk);
1746 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1747 void sk_free_unlock_clone(struct sock *sk);
1748 
1749 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1750 			     gfp_t priority);
1751 void __sock_wfree(struct sk_buff *skb);
1752 void sock_wfree(struct sk_buff *skb);
1753 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1754 			     gfp_t priority);
1755 void skb_orphan_partial(struct sk_buff *skb);
1756 void sock_rfree(struct sk_buff *skb);
1757 void sock_efree(struct sk_buff *skb);
1758 #ifdef CONFIG_INET
1759 void sock_edemux(struct sk_buff *skb);
1760 void sock_pfree(struct sk_buff *skb);
1761 #else
1762 #define sock_edemux sock_efree
1763 #endif
1764 
1765 int sk_setsockopt(struct sock *sk, int level, int optname,
1766 		  sockptr_t optval, unsigned int optlen);
1767 int sock_setsockopt(struct socket *sock, int level, int op,
1768 		    sockptr_t optval, unsigned int optlen);
1769 int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1770 		       int optname, sockptr_t optval, int optlen);
1771 int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1772 		       int optname, sockptr_t optval, sockptr_t optlen);
1773 
1774 int sk_getsockopt(struct sock *sk, int level, int optname,
1775 		  sockptr_t optval, sockptr_t optlen);
1776 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1777 		   bool timeval, bool time32);
1778 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1779 				     unsigned long data_len, int noblock,
1780 				     int *errcode, int max_page_order);
1781 
1782 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1783 						  unsigned long size,
1784 						  int noblock, int *errcode)
1785 {
1786 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1787 }
1788 
1789 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1790 void sock_kfree_s(struct sock *sk, void *mem, int size);
1791 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1792 void sk_send_sigurg(struct sock *sk);
1793 
1794 static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1795 {
1796 	if (sk->sk_socket)
1797 		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1798 	WRITE_ONCE(sk->sk_prot, proto);
1799 }
1800 
1801 struct sockcm_cookie {
1802 	u64 transmit_time;
1803 	u32 mark;
1804 	u32 tsflags;
1805 	u32 ts_opt_id;
1806 };
1807 
1808 static inline void sockcm_init(struct sockcm_cookie *sockc,
1809 			       const struct sock *sk)
1810 {
1811 	*sockc = (struct sockcm_cookie) {
1812 		.tsflags = READ_ONCE(sk->sk_tsflags)
1813 	};
1814 }
1815 
1816 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1817 		     struct sockcm_cookie *sockc);
1818 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1819 		   struct sockcm_cookie *sockc);
1820 
1821 /*
1822  * Functions to fill in entries in struct proto_ops when a protocol
1823  * does not implement a particular function.
1824  */
1825 int sock_no_bind(struct socket *, struct sockaddr *, int);
1826 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1827 int sock_no_socketpair(struct socket *, struct socket *);
1828 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *);
1829 int sock_no_getname(struct socket *, struct sockaddr *, int);
1830 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1831 int sock_no_listen(struct socket *, int);
1832 int sock_no_shutdown(struct socket *, int);
1833 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1834 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1835 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1836 int sock_no_mmap(struct file *file, struct socket *sock,
1837 		 struct vm_area_struct *vma);
1838 
1839 /*
1840  * Functions to fill in entries in struct proto_ops when a protocol
1841  * uses the inet style.
1842  */
1843 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1844 				  char __user *optval, int __user *optlen);
1845 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1846 			int flags);
1847 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1848 			   sockptr_t optval, unsigned int optlen);
1849 
1850 void sk_common_release(struct sock *sk);
1851 
1852 /*
1853  *	Default socket callbacks and setup code
1854  */
1855 
1856 /* Initialise core socket variables using an explicit uid. */
1857 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1858 
1859 /* Initialise core socket variables.
1860  * Assumes struct socket *sock is embedded in a struct socket_alloc.
1861  */
1862 void sock_init_data(struct socket *sock, struct sock *sk);
1863 
1864 /*
1865  * Socket reference counting postulates.
1866  *
1867  * * Each user of socket SHOULD hold a reference count.
1868  * * Each access point to socket (an hash table bucket, reference from a list,
1869  *   running timer, skb in flight MUST hold a reference count.
1870  * * When reference count hits 0, it means it will never increase back.
1871  * * When reference count hits 0, it means that no references from
1872  *   outside exist to this socket and current process on current CPU
1873  *   is last user and may/should destroy this socket.
1874  * * sk_free is called from any context: process, BH, IRQ. When
1875  *   it is called, socket has no references from outside -> sk_free
1876  *   may release descendant resources allocated by the socket, but
1877  *   to the time when it is called, socket is NOT referenced by any
1878  *   hash tables, lists etc.
1879  * * Packets, delivered from outside (from network or from another process)
1880  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1881  *   when they sit in queue. Otherwise, packets will leak to hole, when
1882  *   socket is looked up by one cpu and unhasing is made by another CPU.
1883  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1884  *   (leak to backlog). Packet socket does all the processing inside
1885  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1886  *   use separate SMP lock, so that they are prone too.
1887  */
1888 
1889 /* Ungrab socket and destroy it, if it was the last reference. */
1890 static inline void sock_put(struct sock *sk)
1891 {
1892 	if (refcount_dec_and_test(&sk->sk_refcnt))
1893 		sk_free(sk);
1894 }
1895 /* Generic version of sock_put(), dealing with all sockets
1896  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1897  */
1898 void sock_gen_put(struct sock *sk);
1899 
1900 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1901 		     unsigned int trim_cap, bool refcounted);
1902 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1903 				 const int nested)
1904 {
1905 	return __sk_receive_skb(sk, skb, nested, 1, true);
1906 }
1907 
1908 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1909 {
1910 	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1911 	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1912 		return;
1913 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1914 	 * other WRITE_ONCE() because socket lock might be not held.
1915 	 */
1916 	WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1917 }
1918 
1919 #define NO_QUEUE_MAPPING	USHRT_MAX
1920 
1921 static inline void sk_tx_queue_clear(struct sock *sk)
1922 {
1923 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1924 	 * other WRITE_ONCE() because socket lock might be not held.
1925 	 */
1926 	WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1927 }
1928 
1929 static inline int sk_tx_queue_get(const struct sock *sk)
1930 {
1931 	if (sk) {
1932 		/* Paired with WRITE_ONCE() in sk_tx_queue_clear()
1933 		 * and sk_tx_queue_set().
1934 		 */
1935 		int val = READ_ONCE(sk->sk_tx_queue_mapping);
1936 
1937 		if (val != NO_QUEUE_MAPPING)
1938 			return val;
1939 	}
1940 	return -1;
1941 }
1942 
1943 static inline void __sk_rx_queue_set(struct sock *sk,
1944 				     const struct sk_buff *skb,
1945 				     bool force_set)
1946 {
1947 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1948 	if (skb_rx_queue_recorded(skb)) {
1949 		u16 rx_queue = skb_get_rx_queue(skb);
1950 
1951 		if (force_set ||
1952 		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
1953 			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
1954 	}
1955 #endif
1956 }
1957 
1958 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1959 {
1960 	__sk_rx_queue_set(sk, skb, true);
1961 }
1962 
1963 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
1964 {
1965 	__sk_rx_queue_set(sk, skb, false);
1966 }
1967 
1968 static inline void sk_rx_queue_clear(struct sock *sk)
1969 {
1970 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1971 	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
1972 #endif
1973 }
1974 
1975 static inline int sk_rx_queue_get(const struct sock *sk)
1976 {
1977 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1978 	if (sk) {
1979 		int res = READ_ONCE(sk->sk_rx_queue_mapping);
1980 
1981 		if (res != NO_QUEUE_MAPPING)
1982 			return res;
1983 	}
1984 #endif
1985 
1986 	return -1;
1987 }
1988 
1989 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1990 {
1991 	sk->sk_socket = sock;
1992 }
1993 
1994 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1995 {
1996 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1997 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1998 }
1999 /* Detach socket from process context.
2000  * Announce socket dead, detach it from wait queue and inode.
2001  * Note that parent inode held reference count on this struct sock,
2002  * we do not release it in this function, because protocol
2003  * probably wants some additional cleanups or even continuing
2004  * to work with this socket (TCP).
2005  */
2006 static inline void sock_orphan(struct sock *sk)
2007 {
2008 	write_lock_bh(&sk->sk_callback_lock);
2009 	sock_set_flag(sk, SOCK_DEAD);
2010 	sk_set_socket(sk, NULL);
2011 	sk->sk_wq  = NULL;
2012 	write_unlock_bh(&sk->sk_callback_lock);
2013 }
2014 
2015 static inline void sock_graft(struct sock *sk, struct socket *parent)
2016 {
2017 	WARN_ON(parent->sk);
2018 	write_lock_bh(&sk->sk_callback_lock);
2019 	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2020 	parent->sk = sk;
2021 	sk_set_socket(sk, parent);
2022 	sk->sk_uid = SOCK_INODE(parent)->i_uid;
2023 	security_sock_graft(sk, parent);
2024 	write_unlock_bh(&sk->sk_callback_lock);
2025 }
2026 
2027 kuid_t sock_i_uid(struct sock *sk);
2028 unsigned long __sock_i_ino(struct sock *sk);
2029 unsigned long sock_i_ino(struct sock *sk);
2030 
2031 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2032 {
2033 	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2034 }
2035 
2036 static inline u32 net_tx_rndhash(void)
2037 {
2038 	u32 v = get_random_u32();
2039 
2040 	return v ?: 1;
2041 }
2042 
2043 static inline void sk_set_txhash(struct sock *sk)
2044 {
2045 	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2046 	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2047 }
2048 
2049 static inline bool sk_rethink_txhash(struct sock *sk)
2050 {
2051 	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2052 		sk_set_txhash(sk);
2053 		return true;
2054 	}
2055 	return false;
2056 }
2057 
2058 static inline struct dst_entry *
2059 __sk_dst_get(const struct sock *sk)
2060 {
2061 	return rcu_dereference_check(sk->sk_dst_cache,
2062 				     lockdep_sock_is_held(sk));
2063 }
2064 
2065 static inline struct dst_entry *
2066 sk_dst_get(const struct sock *sk)
2067 {
2068 	struct dst_entry *dst;
2069 
2070 	rcu_read_lock();
2071 	dst = rcu_dereference(sk->sk_dst_cache);
2072 	if (dst && !rcuref_get(&dst->__rcuref))
2073 		dst = NULL;
2074 	rcu_read_unlock();
2075 	return dst;
2076 }
2077 
2078 static inline void __dst_negative_advice(struct sock *sk)
2079 {
2080 	struct dst_entry *dst = __sk_dst_get(sk);
2081 
2082 	if (dst && dst->ops->negative_advice)
2083 		dst->ops->negative_advice(sk, dst);
2084 }
2085 
2086 static inline void dst_negative_advice(struct sock *sk)
2087 {
2088 	sk_rethink_txhash(sk);
2089 	__dst_negative_advice(sk);
2090 }
2091 
2092 static inline void
2093 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2094 {
2095 	struct dst_entry *old_dst;
2096 
2097 	sk_tx_queue_clear(sk);
2098 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2099 	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2100 					    lockdep_sock_is_held(sk));
2101 	rcu_assign_pointer(sk->sk_dst_cache, dst);
2102 	dst_release(old_dst);
2103 }
2104 
2105 static inline void
2106 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2107 {
2108 	struct dst_entry *old_dst;
2109 
2110 	sk_tx_queue_clear(sk);
2111 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2112 	old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst)));
2113 	dst_release(old_dst);
2114 }
2115 
2116 static inline void
2117 __sk_dst_reset(struct sock *sk)
2118 {
2119 	__sk_dst_set(sk, NULL);
2120 }
2121 
2122 static inline void
2123 sk_dst_reset(struct sock *sk)
2124 {
2125 	sk_dst_set(sk, NULL);
2126 }
2127 
2128 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2129 
2130 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2131 
2132 static inline void sk_dst_confirm(struct sock *sk)
2133 {
2134 	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2135 		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2136 }
2137 
2138 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2139 {
2140 	if (skb_get_dst_pending_confirm(skb)) {
2141 		struct sock *sk = skb->sk;
2142 
2143 		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2144 			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2145 		neigh_confirm(n);
2146 	}
2147 }
2148 
2149 bool sk_mc_loop(const struct sock *sk);
2150 
2151 static inline bool sk_can_gso(const struct sock *sk)
2152 {
2153 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2154 }
2155 
2156 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2157 
2158 static inline void sk_gso_disable(struct sock *sk)
2159 {
2160 	sk->sk_gso_disabled = 1;
2161 	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2162 }
2163 
2164 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2165 					   struct iov_iter *from, char *to,
2166 					   int copy, int offset)
2167 {
2168 	if (skb->ip_summed == CHECKSUM_NONE) {
2169 		__wsum csum = 0;
2170 		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2171 			return -EFAULT;
2172 		skb->csum = csum_block_add(skb->csum, csum, offset);
2173 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2174 		if (!copy_from_iter_full_nocache(to, copy, from))
2175 			return -EFAULT;
2176 	} else if (!copy_from_iter_full(to, copy, from))
2177 		return -EFAULT;
2178 
2179 	return 0;
2180 }
2181 
2182 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2183 				       struct iov_iter *from, int copy)
2184 {
2185 	int err, offset = skb->len;
2186 
2187 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2188 				       copy, offset);
2189 	if (err)
2190 		__skb_trim(skb, offset);
2191 
2192 	return err;
2193 }
2194 
2195 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2196 					   struct sk_buff *skb,
2197 					   struct page *page,
2198 					   int off, int copy)
2199 {
2200 	int err;
2201 
2202 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2203 				       copy, skb->len);
2204 	if (err)
2205 		return err;
2206 
2207 	skb_len_add(skb, copy);
2208 	sk_wmem_queued_add(sk, copy);
2209 	sk_mem_charge(sk, copy);
2210 	return 0;
2211 }
2212 
2213 /**
2214  * sk_wmem_alloc_get - returns write allocations
2215  * @sk: socket
2216  *
2217  * Return: sk_wmem_alloc minus initial offset of one
2218  */
2219 static inline int sk_wmem_alloc_get(const struct sock *sk)
2220 {
2221 	return refcount_read(&sk->sk_wmem_alloc) - 1;
2222 }
2223 
2224 /**
2225  * sk_rmem_alloc_get - returns read allocations
2226  * @sk: socket
2227  *
2228  * Return: sk_rmem_alloc
2229  */
2230 static inline int sk_rmem_alloc_get(const struct sock *sk)
2231 {
2232 	return atomic_read(&sk->sk_rmem_alloc);
2233 }
2234 
2235 /**
2236  * sk_has_allocations - check if allocations are outstanding
2237  * @sk: socket
2238  *
2239  * Return: true if socket has write or read allocations
2240  */
2241 static inline bool sk_has_allocations(const struct sock *sk)
2242 {
2243 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2244 }
2245 
2246 /**
2247  * skwq_has_sleeper - check if there are any waiting processes
2248  * @wq: struct socket_wq
2249  *
2250  * Return: true if socket_wq has waiting processes
2251  *
2252  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2253  * barrier call. They were added due to the race found within the tcp code.
2254  *
2255  * Consider following tcp code paths::
2256  *
2257  *   CPU1                CPU2
2258  *   sys_select          receive packet
2259  *   ...                 ...
2260  *   __add_wait_queue    update tp->rcv_nxt
2261  *   ...                 ...
2262  *   tp->rcv_nxt check   sock_def_readable
2263  *   ...                 {
2264  *   schedule               rcu_read_lock();
2265  *                          wq = rcu_dereference(sk->sk_wq);
2266  *                          if (wq && waitqueue_active(&wq->wait))
2267  *                              wake_up_interruptible(&wq->wait)
2268  *                          ...
2269  *                       }
2270  *
2271  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2272  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2273  * could then endup calling schedule and sleep forever if there are no more
2274  * data on the socket.
2275  *
2276  */
2277 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2278 {
2279 	return wq && wq_has_sleeper(&wq->wait);
2280 }
2281 
2282 /**
2283  * sock_poll_wait - place memory barrier behind the poll_wait call.
2284  * @filp:           file
2285  * @sock:           socket to wait on
2286  * @p:              poll_table
2287  *
2288  * See the comments in the wq_has_sleeper function.
2289  */
2290 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2291 				  poll_table *p)
2292 {
2293 	if (!poll_does_not_wait(p)) {
2294 		poll_wait(filp, &sock->wq.wait, p);
2295 		/* We need to be sure we are in sync with the
2296 		 * socket flags modification.
2297 		 *
2298 		 * This memory barrier is paired in the wq_has_sleeper.
2299 		 */
2300 		smp_mb();
2301 	}
2302 }
2303 
2304 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2305 {
2306 	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2307 	u32 txhash = READ_ONCE(sk->sk_txhash);
2308 
2309 	if (txhash) {
2310 		skb->l4_hash = 1;
2311 		skb->hash = txhash;
2312 	}
2313 }
2314 
2315 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2316 
2317 /*
2318  *	Queue a received datagram if it will fit. Stream and sequenced
2319  *	protocols can't normally use this as they need to fit buffers in
2320  *	and play with them.
2321  *
2322  *	Inlined as it's very short and called for pretty much every
2323  *	packet ever received.
2324  */
2325 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2326 {
2327 	skb_orphan(skb);
2328 	skb->sk = sk;
2329 	skb->destructor = sock_rfree;
2330 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2331 	sk_mem_charge(sk, skb->truesize);
2332 }
2333 
2334 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2335 {
2336 	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2337 		skb_orphan(skb);
2338 		skb->destructor = sock_efree;
2339 		skb->sk = sk;
2340 		return true;
2341 	}
2342 	return false;
2343 }
2344 
2345 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2346 {
2347 	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2348 	if (skb) {
2349 		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2350 			skb_set_owner_r(skb, sk);
2351 			return skb;
2352 		}
2353 		__kfree_skb(skb);
2354 	}
2355 	return NULL;
2356 }
2357 
2358 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2359 {
2360 	if (skb->destructor != sock_wfree) {
2361 		skb_orphan(skb);
2362 		return;
2363 	}
2364 	skb->slow_gro = 1;
2365 }
2366 
2367 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2368 		    unsigned long expires);
2369 
2370 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2371 
2372 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2373 
2374 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2375 			struct sk_buff *skb, unsigned int flags,
2376 			void (*destructor)(struct sock *sk,
2377 					   struct sk_buff *skb));
2378 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2379 
2380 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2381 			      enum skb_drop_reason *reason);
2382 
2383 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2384 {
2385 	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2386 }
2387 
2388 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2389 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2390 
2391 /*
2392  *	Recover an error report and clear atomically
2393  */
2394 
2395 static inline int sock_error(struct sock *sk)
2396 {
2397 	int err;
2398 
2399 	/* Avoid an atomic operation for the common case.
2400 	 * This is racy since another cpu/thread can change sk_err under us.
2401 	 */
2402 	if (likely(data_race(!sk->sk_err)))
2403 		return 0;
2404 
2405 	err = xchg(&sk->sk_err, 0);
2406 	return -err;
2407 }
2408 
2409 void sk_error_report(struct sock *sk);
2410 
2411 static inline unsigned long sock_wspace(struct sock *sk)
2412 {
2413 	int amt = 0;
2414 
2415 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2416 		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2417 		if (amt < 0)
2418 			amt = 0;
2419 	}
2420 	return amt;
2421 }
2422 
2423 /* Note:
2424  *  We use sk->sk_wq_raw, from contexts knowing this
2425  *  pointer is not NULL and cannot disappear/change.
2426  */
2427 static inline void sk_set_bit(int nr, struct sock *sk)
2428 {
2429 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2430 	    !sock_flag(sk, SOCK_FASYNC))
2431 		return;
2432 
2433 	set_bit(nr, &sk->sk_wq_raw->flags);
2434 }
2435 
2436 static inline void sk_clear_bit(int nr, struct sock *sk)
2437 {
2438 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2439 	    !sock_flag(sk, SOCK_FASYNC))
2440 		return;
2441 
2442 	clear_bit(nr, &sk->sk_wq_raw->flags);
2443 }
2444 
2445 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2446 {
2447 	if (sock_flag(sk, SOCK_FASYNC)) {
2448 		rcu_read_lock();
2449 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2450 		rcu_read_unlock();
2451 	}
2452 }
2453 
2454 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band)
2455 {
2456 	if (unlikely(sock_flag(sk, SOCK_FASYNC)))
2457 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2458 }
2459 
2460 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2461  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2462  * Note: for send buffers, TCP works better if we can build two skbs at
2463  * minimum.
2464  */
2465 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2466 
2467 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2468 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2469 
2470 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2471 {
2472 	u32 val;
2473 
2474 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2475 		return;
2476 
2477 	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2478 	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2479 
2480 	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2481 }
2482 
2483 /**
2484  * sk_page_frag - return an appropriate page_frag
2485  * @sk: socket
2486  *
2487  * Use the per task page_frag instead of the per socket one for
2488  * optimization when we know that we're in process context and own
2489  * everything that's associated with %current.
2490  *
2491  * Both direct reclaim and page faults can nest inside other
2492  * socket operations and end up recursing into sk_page_frag()
2493  * while it's already in use: explicitly avoid task page_frag
2494  * when users disable sk_use_task_frag.
2495  *
2496  * Return: a per task page_frag if context allows that,
2497  * otherwise a per socket one.
2498  */
2499 static inline struct page_frag *sk_page_frag(struct sock *sk)
2500 {
2501 	if (sk->sk_use_task_frag)
2502 		return &current->task_frag;
2503 
2504 	return &sk->sk_frag;
2505 }
2506 
2507 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2508 
2509 /*
2510  *	Default write policy as shown to user space via poll/select/SIGIO
2511  */
2512 static inline bool sock_writeable(const struct sock *sk)
2513 {
2514 	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2515 }
2516 
2517 static inline gfp_t gfp_any(void)
2518 {
2519 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2520 }
2521 
2522 static inline gfp_t gfp_memcg_charge(void)
2523 {
2524 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2525 }
2526 
2527 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2528 {
2529 	return noblock ? 0 : sk->sk_rcvtimeo;
2530 }
2531 
2532 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2533 {
2534 	return noblock ? 0 : sk->sk_sndtimeo;
2535 }
2536 
2537 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2538 {
2539 	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2540 
2541 	return v ?: 1;
2542 }
2543 
2544 /* Alas, with timeout socket operations are not restartable.
2545  * Compare this to poll().
2546  */
2547 static inline int sock_intr_errno(long timeo)
2548 {
2549 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2550 }
2551 
2552 struct sock_skb_cb {
2553 	u32 dropcount;
2554 };
2555 
2556 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2557  * using skb->cb[] would keep using it directly and utilize its
2558  * alignment guarantee.
2559  */
2560 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2561 			    sizeof(struct sock_skb_cb)))
2562 
2563 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2564 			    SOCK_SKB_CB_OFFSET))
2565 
2566 #define sock_skb_cb_check_size(size) \
2567 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2568 
2569 static inline void
2570 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2571 {
2572 	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2573 						atomic_read(&sk->sk_drops) : 0;
2574 }
2575 
2576 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2577 {
2578 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2579 
2580 	atomic_add(segs, &sk->sk_drops);
2581 }
2582 
2583 static inline ktime_t sock_read_timestamp(struct sock *sk)
2584 {
2585 #if BITS_PER_LONG==32
2586 	unsigned int seq;
2587 	ktime_t kt;
2588 
2589 	do {
2590 		seq = read_seqbegin(&sk->sk_stamp_seq);
2591 		kt = sk->sk_stamp;
2592 	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2593 
2594 	return kt;
2595 #else
2596 	return READ_ONCE(sk->sk_stamp);
2597 #endif
2598 }
2599 
2600 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2601 {
2602 #if BITS_PER_LONG==32
2603 	write_seqlock(&sk->sk_stamp_seq);
2604 	sk->sk_stamp = kt;
2605 	write_sequnlock(&sk->sk_stamp_seq);
2606 #else
2607 	WRITE_ONCE(sk->sk_stamp, kt);
2608 #endif
2609 }
2610 
2611 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2612 			   struct sk_buff *skb);
2613 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2614 			     struct sk_buff *skb);
2615 
2616 static inline void
2617 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2618 {
2619 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2620 	u32 tsflags = READ_ONCE(sk->sk_tsflags);
2621 	ktime_t kt = skb->tstamp;
2622 	/*
2623 	 * generate control messages if
2624 	 * - receive time stamping in software requested
2625 	 * - software time stamp available and wanted
2626 	 * - hardware time stamps available and wanted
2627 	 */
2628 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2629 	    (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2630 	    (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2631 	    (hwtstamps->hwtstamp &&
2632 	     (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2633 		__sock_recv_timestamp(msg, sk, skb);
2634 	else
2635 		sock_write_timestamp(sk, kt);
2636 
2637 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2638 		__sock_recv_wifi_status(msg, sk, skb);
2639 }
2640 
2641 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2642 		       struct sk_buff *skb);
2643 
2644 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2645 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2646 				   struct sk_buff *skb)
2647 {
2648 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2649 			   (1UL << SOCK_RCVTSTAMP)			| \
2650 			   (1UL << SOCK_RCVMARK))
2651 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2652 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2653 
2654 	if (sk->sk_flags & FLAGS_RECV_CMSGS ||
2655 	    READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY)
2656 		__sock_recv_cmsgs(msg, sk, skb);
2657 	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2658 		sock_write_timestamp(sk, skb->tstamp);
2659 	else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2660 		sock_write_timestamp(sk, 0);
2661 }
2662 
2663 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags);
2664 
2665 /**
2666  * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2667  * @sk:		socket sending this packet
2668  * @sockc:	pointer to socket cmsg cookie to get timestamping info
2669  * @tx_flags:	completed with instructions for time stamping
2670  * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2671  *
2672  * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2673  */
2674 static inline void _sock_tx_timestamp(struct sock *sk,
2675 				      const struct sockcm_cookie *sockc,
2676 				      __u8 *tx_flags, __u32 *tskey)
2677 {
2678 	__u32 tsflags = sockc->tsflags;
2679 
2680 	if (unlikely(tsflags)) {
2681 		__sock_tx_timestamp(tsflags, tx_flags);
2682 		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2683 		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) {
2684 			if (tsflags & SOCKCM_FLAG_TS_OPT_ID)
2685 				*tskey = sockc->ts_opt_id;
2686 			else
2687 				*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2688 		}
2689 	}
2690 	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2691 		*tx_flags |= SKBTX_WIFI_STATUS;
2692 }
2693 
2694 static inline void sock_tx_timestamp(struct sock *sk,
2695 				     const struct sockcm_cookie *sockc,
2696 				     __u8 *tx_flags)
2697 {
2698 	_sock_tx_timestamp(sk, sockc, tx_flags, NULL);
2699 }
2700 
2701 static inline void skb_setup_tx_timestamp(struct sk_buff *skb,
2702 					  const struct sockcm_cookie *sockc)
2703 {
2704 	_sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags,
2705 			   &skb_shinfo(skb)->tskey);
2706 }
2707 
2708 static inline bool sk_is_inet(const struct sock *sk)
2709 {
2710 	int family = READ_ONCE(sk->sk_family);
2711 
2712 	return family == AF_INET || family == AF_INET6;
2713 }
2714 
2715 static inline bool sk_is_tcp(const struct sock *sk)
2716 {
2717 	return sk_is_inet(sk) &&
2718 	       sk->sk_type == SOCK_STREAM &&
2719 	       sk->sk_protocol == IPPROTO_TCP;
2720 }
2721 
2722 static inline bool sk_is_udp(const struct sock *sk)
2723 {
2724 	return sk_is_inet(sk) &&
2725 	       sk->sk_type == SOCK_DGRAM &&
2726 	       sk->sk_protocol == IPPROTO_UDP;
2727 }
2728 
2729 static inline bool sk_is_stream_unix(const struct sock *sk)
2730 {
2731 	return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM;
2732 }
2733 
2734 /**
2735  * sk_eat_skb - Release a skb if it is no longer needed
2736  * @sk: socket to eat this skb from
2737  * @skb: socket buffer to eat
2738  *
2739  * This routine must be called with interrupts disabled or with the socket
2740  * locked so that the sk_buff queue operation is ok.
2741 */
2742 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2743 {
2744 	__skb_unlink(skb, &sk->sk_receive_queue);
2745 	__kfree_skb(skb);
2746 }
2747 
2748 static inline bool
2749 skb_sk_is_prefetched(struct sk_buff *skb)
2750 {
2751 #ifdef CONFIG_INET
2752 	return skb->destructor == sock_pfree;
2753 #else
2754 	return false;
2755 #endif /* CONFIG_INET */
2756 }
2757 
2758 /* This helper checks if a socket is a full socket,
2759  * ie _not_ a timewait or request socket.
2760  */
2761 static inline bool sk_fullsock(const struct sock *sk)
2762 {
2763 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2764 }
2765 
2766 static inline bool
2767 sk_is_refcounted(struct sock *sk)
2768 {
2769 	/* Only full sockets have sk->sk_flags. */
2770 	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2771 }
2772 
2773 /* Checks if this SKB belongs to an HW offloaded socket
2774  * and whether any SW fallbacks are required based on dev.
2775  * Check decrypted mark in case skb_orphan() cleared socket.
2776  */
2777 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2778 						   struct net_device *dev)
2779 {
2780 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2781 	struct sock *sk = skb->sk;
2782 
2783 	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2784 		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2785 	} else if (unlikely(skb_is_decrypted(skb))) {
2786 		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2787 		kfree_skb(skb);
2788 		skb = NULL;
2789 	}
2790 #endif
2791 
2792 	return skb;
2793 }
2794 
2795 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2796  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2797  */
2798 static inline bool sk_listener(const struct sock *sk)
2799 {
2800 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2801 }
2802 
2803 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2804 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2805 		       int type);
2806 
2807 bool sk_ns_capable(const struct sock *sk,
2808 		   struct user_namespace *user_ns, int cap);
2809 bool sk_capable(const struct sock *sk, int cap);
2810 bool sk_net_capable(const struct sock *sk, int cap);
2811 
2812 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2813 
2814 /* Take into consideration the size of the struct sk_buff overhead in the
2815  * determination of these values, since that is non-constant across
2816  * platforms.  This makes socket queueing behavior and performance
2817  * not depend upon such differences.
2818  */
2819 #define _SK_MEM_PACKETS		256
2820 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2821 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2822 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2823 
2824 extern __u32 sysctl_wmem_max;
2825 extern __u32 sysctl_rmem_max;
2826 
2827 extern __u32 sysctl_wmem_default;
2828 extern __u32 sysctl_rmem_default;
2829 
2830 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
2831 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2832 
2833 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2834 {
2835 	/* Does this proto have per netns sysctl_wmem ? */
2836 	if (proto->sysctl_wmem_offset)
2837 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2838 
2839 	return READ_ONCE(*proto->sysctl_wmem);
2840 }
2841 
2842 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2843 {
2844 	/* Does this proto have per netns sysctl_rmem ? */
2845 	if (proto->sysctl_rmem_offset)
2846 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2847 
2848 	return READ_ONCE(*proto->sysctl_rmem);
2849 }
2850 
2851 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2852  * Some wifi drivers need to tweak it to get more chunks.
2853  * They can use this helper from their ndo_start_xmit()
2854  */
2855 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2856 {
2857 	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2858 		return;
2859 	WRITE_ONCE(sk->sk_pacing_shift, val);
2860 }
2861 
2862 /* if a socket is bound to a device, check that the given device
2863  * index is either the same or that the socket is bound to an L3
2864  * master device and the given device index is also enslaved to
2865  * that L3 master
2866  */
2867 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2868 {
2869 	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2870 	int mdif;
2871 
2872 	if (!bound_dev_if || bound_dev_if == dif)
2873 		return true;
2874 
2875 	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2876 	if (mdif && mdif == bound_dev_if)
2877 		return true;
2878 
2879 	return false;
2880 }
2881 
2882 void sock_def_readable(struct sock *sk);
2883 
2884 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2885 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2886 int sock_set_timestamping(struct sock *sk, int optname,
2887 			  struct so_timestamping timestamping);
2888 
2889 void sock_enable_timestamps(struct sock *sk);
2890 void sock_no_linger(struct sock *sk);
2891 void sock_set_keepalive(struct sock *sk);
2892 void sock_set_priority(struct sock *sk, u32 priority);
2893 void sock_set_rcvbuf(struct sock *sk, int val);
2894 void sock_set_mark(struct sock *sk, u32 val);
2895 void sock_set_reuseaddr(struct sock *sk);
2896 void sock_set_reuseport(struct sock *sk);
2897 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2898 
2899 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2900 
2901 int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2902 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2903 			   sockptr_t optval, int optlen, bool old_timeval);
2904 
2905 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
2906 		     void __user *arg, void *karg, size_t size);
2907 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
2908 static inline bool sk_is_readable(struct sock *sk)
2909 {
2910 	if (sk->sk_prot->sock_is_readable)
2911 		return sk->sk_prot->sock_is_readable(sk);
2912 	return false;
2913 }
2914 #endif	/* _SOCK_H */
2915