xref: /linux/include/net/sock.h (revision f197902cd21ae833850679b216bb62c0d056bbb3)
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   *	@psp_assoc: PSP association, if socket is PSP-secured
253   *	@sk_receive_queue: incoming packets
254   *	@sk_wmem_alloc: transmit queue bytes committed
255   *	@sk_tsq_flags: TCP Small Queues flags
256   *	@sk_write_queue: Packet sending queue
257   *	@sk_omem_alloc: "o" is "option" or "other"
258   *	@sk_wmem_queued: persistent queue size
259   *	@sk_forward_alloc: space allocated forward
260   *	@sk_reserved_mem: space reserved and non-reclaimable for the socket
261   *	@sk_napi_id: id of the last napi context to receive data for sk
262   *	@sk_ll_usec: usecs to busypoll when there is no data
263   *	@sk_allocation: allocation mode
264   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
265   *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
266   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
267   *	@sk_sndbuf: size of send buffer in bytes
268   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
269   *	@sk_no_check_rx: allow zero checksum in RX packets
270   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
271   *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
272   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
273   *	@sk_gso_max_size: Maximum GSO segment size to build
274   *	@sk_gso_max_segs: Maximum number of GSO segments
275   *	@sk_pacing_shift: scaling factor for TCP Small Queues
276   *	@sk_lingertime: %SO_LINGER l_linger setting
277   *	@sk_backlog: always used with the per-socket spinlock held
278   *	@sk_callback_lock: used with the callbacks in the end of this struct
279   *	@sk_error_queue: rarely used
280   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
281   *			  IPV6_ADDRFORM for instance)
282   *	@sk_err: last error
283   *	@sk_err_soft: errors that don't cause failure but are the cause of a
284   *		      persistent failure not just 'timed out'
285   *	@sk_drops: raw/udp drops counter
286   *	@sk_drop_counters: optional pointer to numa_drop_counters
287   *	@sk_ack_backlog: current listen backlog
288   *	@sk_max_ack_backlog: listen backlog set in listen()
289   *	@sk_uid: user id of owner
290   *	@sk_ino: inode number (zero if orphaned)
291   *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
292   *	@sk_busy_poll_budget: napi processing budget when busypolling
293   *	@sk_priority: %SO_PRIORITY setting
294   *	@sk_type: socket type (%SOCK_STREAM, etc)
295   *	@sk_protocol: which protocol this socket belongs in this network family
296   *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
297   *	@sk_peer_pid: &struct pid for this socket's peer
298   *	@sk_peer_cred: %SO_PEERCRED setting
299   *	@sk_rcvlowat: %SO_RCVLOWAT setting
300   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
301   *	@sk_sndtimeo: %SO_SNDTIMEO setting
302   *	@sk_txhash: computed flow hash for use on transmit
303   *	@sk_txrehash: enable TX hash rethink
304   *	@sk_filter: socket filtering instructions
305   *	@sk_timer: sock cleanup timer
306   *	@sk_stamp: time stamp of last packet received
307   *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
308   *	@sk_tsflags: SO_TIMESTAMPING flags
309   *	@sk_bpf_cb_flags: used in bpf_setsockopt()
310   *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
311   *			   Sockets that can be used under memory reclaim should
312   *			   set this to false.
313   *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
314   *	              for timestamping
315   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
316   *	@sk_tx_queue_mapping_jiffies: time in jiffies of last @sk_tx_queue_mapping refresh.
317   *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
318   *	@sk_socket: Identd and reporting IO signals
319   *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
320   *	@sk_frag: cached page frag
321   *	@sk_peek_off: current peek_offset value
322   *	@sk_send_head: front of stuff to transmit
323   *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
324   *	@sk_security: used by security modules
325   *	@sk_mark: generic packet mark
326   *	@sk_cgrp_data: cgroup data for this cgroup
327   *	@sk_memcg: this socket's memory cgroup association
328   *	@sk_write_pending: a write to stream socket waits to start
329   *	@sk_disconnects: number of disconnect operations performed on this sock
330   *	@sk_state_change: callback to indicate change in the state of the sock
331   *	@sk_data_ready: callback to indicate there is data to be processed
332   *	@sk_write_space: callback to indicate there is bf sending space available
333   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
334   *	@sk_backlog_rcv: callback to process the backlog
335   *	@sk_validate_xmit_skb: ptr to an optional validate function
336   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
337   *	@sk_reuseport_cb: reuseport group container
338   *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
339   *	@sk_rcu: used during RCU grace period
340   *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
341   *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
342   *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
343   *	@sk_txtime_unused: unused txtime flags
344   *	@sk_scm_recv_flags: all flags used by scm_recv()
345   *	@sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS
346   *	@sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY
347   *	@sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD
348   *	@sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS
349   *	@sk_scm_unused: unused flags for scm_recv()
350   *	@ns_tracker: tracker for netns reference
351   *	@sk_user_frags: xarray of pages the user is holding a reference on.
352   *	@sk_owner: reference to the real owner of the socket that calls
353   *		   sock_lock_init_class_and_name().
354   */
355 struct sock {
356 	/*
357 	 * Now struct inet_timewait_sock also uses sock_common, so please just
358 	 * don't add nothing before this first member (__sk_common) --acme
359 	 */
360 	struct sock_common	__sk_common;
361 #define sk_node			__sk_common.skc_node
362 #define sk_nulls_node		__sk_common.skc_nulls_node
363 #define sk_refcnt		__sk_common.skc_refcnt
364 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
365 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
366 #define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
367 #endif
368 
369 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
370 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
371 #define sk_hash			__sk_common.skc_hash
372 #define sk_portpair		__sk_common.skc_portpair
373 #define sk_num			__sk_common.skc_num
374 #define sk_dport		__sk_common.skc_dport
375 #define sk_addrpair		__sk_common.skc_addrpair
376 #define sk_daddr		__sk_common.skc_daddr
377 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
378 #define sk_family		__sk_common.skc_family
379 #define sk_state		__sk_common.skc_state
380 #define sk_reuse		__sk_common.skc_reuse
381 #define sk_reuseport		__sk_common.skc_reuseport
382 #define sk_ipv6only		__sk_common.skc_ipv6only
383 #define sk_net_refcnt		__sk_common.skc_net_refcnt
384 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
385 #define sk_bind_node		__sk_common.skc_bind_node
386 #define sk_prot			__sk_common.skc_prot
387 #define sk_net			__sk_common.skc_net
388 #define sk_v6_daddr		__sk_common.skc_v6_daddr
389 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
390 #define sk_cookie		__sk_common.skc_cookie
391 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
392 #define sk_flags		__sk_common.skc_flags
393 #define sk_rxhash		__sk_common.skc_rxhash
394 
395 	__cacheline_group_begin(sock_write_rx);
396 
397 	atomic_t		sk_drops;
398 	__s32			sk_peek_off;
399 	struct sk_buff_head	sk_error_queue;
400 	struct sk_buff_head	sk_receive_queue;
401 	/*
402 	 * The backlog queue is special, it is always used with
403 	 * the per-socket spinlock held and requires low latency
404 	 * access. Therefore we special case it's implementation.
405 	 * Note : rmem_alloc is in this structure to fill a hole
406 	 * on 64bit arches, not because its logically part of
407 	 * backlog.
408 	 */
409 	struct {
410 		atomic_t	rmem_alloc;
411 		int		len;
412 		struct sk_buff	*head;
413 		struct sk_buff	*tail;
414 	} sk_backlog;
415 #define sk_rmem_alloc sk_backlog.rmem_alloc
416 
417 	__cacheline_group_end(sock_write_rx);
418 
419 	__cacheline_group_begin(sock_read_rx);
420 	/* early demux fields */
421 	struct dst_entry __rcu	*sk_rx_dst;
422 	int			sk_rx_dst_ifindex;
423 	u32			sk_rx_dst_cookie;
424 
425 #ifdef CONFIG_NET_RX_BUSY_POLL
426 	unsigned int		sk_ll_usec;
427 	unsigned int		sk_napi_id;
428 	u16			sk_busy_poll_budget;
429 	u8			sk_prefer_busy_poll;
430 #endif
431 	u8			sk_userlocks;
432 	int			sk_rcvbuf;
433 
434 	struct sk_filter __rcu	*sk_filter;
435 	union {
436 		struct socket_wq __rcu	*sk_wq;
437 		/* private: */
438 		struct socket_wq	*sk_wq_raw;
439 		/* public: */
440 	};
441 
442 	void			(*sk_data_ready)(struct sock *sk);
443 	long			sk_rcvtimeo;
444 	int			sk_rcvlowat;
445 	__cacheline_group_end(sock_read_rx);
446 
447 	__cacheline_group_begin(sock_read_rxtx);
448 	int			sk_err;
449 	struct socket		*sk_socket;
450 #ifdef CONFIG_MEMCG
451 	struct mem_cgroup	*sk_memcg;
452 #endif
453 #ifdef CONFIG_XFRM
454 	struct xfrm_policy __rcu *sk_policy[2];
455 #endif
456 #if IS_ENABLED(CONFIG_INET_PSP)
457 	struct psp_assoc __rcu	*psp_assoc;
458 #endif
459 	__cacheline_group_end(sock_read_rxtx);
460 
461 	__cacheline_group_begin(sock_write_rxtx);
462 	socket_lock_t		sk_lock;
463 	u32			sk_reserved_mem;
464 	int			sk_forward_alloc;
465 	u32			sk_tsflags;
466 	__cacheline_group_end(sock_write_rxtx);
467 
468 	__cacheline_group_begin(sock_write_tx);
469 	int			sk_write_pending;
470 	atomic_t		sk_omem_alloc;
471 	int			sk_err_soft;
472 
473 	int			sk_wmem_queued;
474 	refcount_t		sk_wmem_alloc;
475 	unsigned long		sk_tsq_flags;
476 	union {
477 		struct sk_buff	*sk_send_head;
478 		struct rb_root	tcp_rtx_queue;
479 	};
480 	struct sk_buff_head	sk_write_queue;
481 	u32			sk_dst_pending_confirm;
482 	u32			sk_pacing_status; /* see enum sk_pacing */
483 	struct page_frag	sk_frag;
484 	struct timer_list	sk_timer;
485 
486 	unsigned long		sk_pacing_rate; /* bytes per second */
487 	atomic_t		sk_zckey;
488 	atomic_t		sk_tskey;
489 	unsigned long		sk_tx_queue_mapping_jiffies;
490 	__cacheline_group_end(sock_write_tx);
491 
492 	__cacheline_group_begin(sock_read_tx);
493 	unsigned long		sk_max_pacing_rate;
494 	long			sk_sndtimeo;
495 	u32			sk_priority;
496 	u32			sk_mark;
497 	kuid_t			sk_uid;
498 	u16			sk_protocol;
499 	u16			sk_type;
500 	struct dst_entry __rcu	*sk_dst_cache;
501 	netdev_features_t	sk_route_caps;
502 #ifdef CONFIG_SOCK_VALIDATE_XMIT
503 	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
504 							struct net_device *dev,
505 							struct sk_buff *skb);
506 #endif
507 	u16			sk_gso_type;
508 	u16			sk_gso_max_segs;
509 	unsigned int		sk_gso_max_size;
510 	gfp_t			sk_allocation;
511 	u32			sk_txhash;
512 	int			sk_sndbuf;
513 	u8			sk_pacing_shift;
514 	bool			sk_use_task_frag;
515 	__cacheline_group_end(sock_read_tx);
516 
517 	/*
518 	 * Because of non atomicity rules, all
519 	 * changes are protected by socket lock.
520 	 */
521 	u8			sk_gso_disabled : 1,
522 				sk_kern_sock : 1,
523 				sk_no_check_tx : 1,
524 				sk_no_check_rx : 1;
525 	u8			sk_shutdown;
526 	unsigned long	        sk_lingertime;
527 	struct proto		*sk_prot_creator;
528 	rwlock_t		sk_callback_lock;
529 	u32			sk_ack_backlog;
530 	u32			sk_max_ack_backlog;
531 	unsigned long		sk_ino;
532 	spinlock_t		sk_peer_lock;
533 	int			sk_bind_phc;
534 	struct pid		*sk_peer_pid;
535 	const struct cred	*sk_peer_cred;
536 
537 	ktime_t			sk_stamp;
538 #if BITS_PER_LONG==32
539 	seqlock_t		sk_stamp_seq;
540 #endif
541 	int			sk_disconnects;
542 
543 	union {
544 		u8		sk_txrehash;
545 		u8		sk_scm_recv_flags;
546 		struct {
547 			u8	sk_scm_credentials : 1,
548 				sk_scm_security : 1,
549 				sk_scm_pidfd : 1,
550 				sk_scm_rights : 1,
551 				sk_scm_unused : 4;
552 		};
553 	};
554 	u8			sk_clockid;
555 	u8			sk_txtime_deadline_mode : 1,
556 				sk_txtime_report_errors : 1,
557 				sk_txtime_unused : 6;
558 #define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG))
559 	u8			sk_bpf_cb_flags;
560 
561 	void			*sk_user_data;
562 #ifdef CONFIG_SECURITY
563 	void			*sk_security;
564 #endif
565 	struct sock_cgroup_data	sk_cgrp_data;
566 	void			(*sk_state_change)(struct sock *sk);
567 	void			(*sk_write_space)(struct sock *sk);
568 	void			(*sk_error_report)(struct sock *sk);
569 	int			(*sk_backlog_rcv)(struct sock *sk,
570 						  struct sk_buff *skb);
571 	void                    (*sk_destruct)(struct sock *sk);
572 	struct sock_reuseport __rcu	*sk_reuseport_cb;
573 #ifdef CONFIG_BPF_SYSCALL
574 	struct bpf_local_storage __rcu	*sk_bpf_storage;
575 #endif
576 	struct numa_drop_counters *sk_drop_counters;
577 	struct rcu_head		sk_rcu;
578 	netns_tracker		ns_tracker;
579 	struct xarray		sk_user_frags;
580 
581 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
582 	struct module		*sk_owner;
583 #endif
584 };
585 
586 struct sock_bh_locked {
587 	struct sock *sock;
588 	local_lock_t bh_lock;
589 };
590 
591 enum sk_pacing {
592 	SK_PACING_NONE		= 0,
593 	SK_PACING_NEEDED	= 1,
594 	SK_PACING_FQ		= 2,
595 };
596 
597 /* flag bits in sk_user_data
598  *
599  * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
600  *   not be suitable for copying when cloning the socket. For instance,
601  *   it can point to a reference counted object. sk_user_data bottom
602  *   bit is set if pointer must not be copied.
603  *
604  * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
605  *   managed/owned by a BPF reuseport array. This bit should be set
606  *   when sk_user_data's sk is added to the bpf's reuseport_array.
607  *
608  * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
609  *   sk_user_data points to psock type. This bit should be set
610  *   when sk_user_data is assigned to a psock object.
611  */
612 #define SK_USER_DATA_NOCOPY	1UL
613 #define SK_USER_DATA_BPF	2UL
614 #define SK_USER_DATA_PSOCK	4UL
615 #define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
616 				  SK_USER_DATA_PSOCK)
617 
618 /**
619  * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
620  * @sk: socket
621  */
622 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
623 {
624 	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
625 }
626 
627 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
628 
629 /**
630  * __locked_read_sk_user_data_with_flags - return the pointer
631  * only if argument flags all has been set in sk_user_data. Otherwise
632  * return NULL
633  *
634  * @sk: socket
635  * @flags: flag bits
636  *
637  * The caller must be holding sk->sk_callback_lock.
638  */
639 static inline void *
640 __locked_read_sk_user_data_with_flags(const struct sock *sk,
641 				      uintptr_t flags)
642 {
643 	uintptr_t sk_user_data =
644 		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
645 						 lockdep_is_held(&sk->sk_callback_lock));
646 
647 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
648 
649 	if ((sk_user_data & flags) == flags)
650 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
651 	return NULL;
652 }
653 
654 /**
655  * __rcu_dereference_sk_user_data_with_flags - return the pointer
656  * only if argument flags all has been set in sk_user_data. Otherwise
657  * return NULL
658  *
659  * @sk: socket
660  * @flags: flag bits
661  */
662 static inline void *
663 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
664 					  uintptr_t flags)
665 {
666 	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
667 
668 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
669 
670 	if ((sk_user_data & flags) == flags)
671 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
672 	return NULL;
673 }
674 
675 #define rcu_dereference_sk_user_data(sk)				\
676 	__rcu_dereference_sk_user_data_with_flags(sk, 0)
677 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
678 ({									\
679 	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
680 		  __tmp2 = (uintptr_t)(flags);				\
681 	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
682 	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
683 	rcu_assign_pointer(__sk_user_data((sk)),			\
684 			   __tmp1 | __tmp2);				\
685 })
686 #define rcu_assign_sk_user_data(sk, ptr)				\
687 	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
688 
689 static inline
690 struct net *sock_net(const struct sock *sk)
691 {
692 	return read_pnet(&sk->sk_net);
693 }
694 
695 static inline
696 void sock_net_set(struct sock *sk, struct net *net)
697 {
698 	write_pnet(&sk->sk_net, net);
699 }
700 
701 /*
702  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
703  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
704  * on a socket means that the socket will reuse everybody else's port
705  * without looking at the other's sk_reuse value.
706  */
707 
708 #define SK_NO_REUSE	0
709 #define SK_CAN_REUSE	1
710 #define SK_FORCE_REUSE	2
711 
712 int sk_set_peek_off(struct sock *sk, int val);
713 
714 static inline int sk_peek_offset(const struct sock *sk, int flags)
715 {
716 	if (unlikely(flags & MSG_PEEK)) {
717 		return READ_ONCE(sk->sk_peek_off);
718 	}
719 
720 	return 0;
721 }
722 
723 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
724 {
725 	s32 off = READ_ONCE(sk->sk_peek_off);
726 
727 	if (unlikely(off >= 0)) {
728 		off = max_t(s32, off - val, 0);
729 		WRITE_ONCE(sk->sk_peek_off, off);
730 	}
731 }
732 
733 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
734 {
735 	sk_peek_offset_bwd(sk, -val);
736 }
737 
738 /*
739  * Hashed lists helper routines
740  */
741 static inline struct sock *sk_entry(const struct hlist_node *node)
742 {
743 	return hlist_entry(node, struct sock, sk_node);
744 }
745 
746 static inline struct sock *__sk_head(const struct hlist_head *head)
747 {
748 	return hlist_entry(head->first, struct sock, sk_node);
749 }
750 
751 static inline struct sock *sk_head(const struct hlist_head *head)
752 {
753 	return hlist_empty(head) ? NULL : __sk_head(head);
754 }
755 
756 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
757 {
758 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
759 }
760 
761 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
762 {
763 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
764 }
765 
766 static inline struct sock *sk_next(const struct sock *sk)
767 {
768 	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
769 }
770 
771 static inline struct sock *sk_nulls_next(const struct sock *sk)
772 {
773 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
774 		hlist_nulls_entry(sk->sk_nulls_node.next,
775 				  struct sock, sk_nulls_node) :
776 		NULL;
777 }
778 
779 static inline bool sk_unhashed(const struct sock *sk)
780 {
781 	return hlist_unhashed(&sk->sk_node);
782 }
783 
784 static inline bool sk_hashed(const struct sock *sk)
785 {
786 	return !sk_unhashed(sk);
787 }
788 
789 static inline void sk_node_init(struct hlist_node *node)
790 {
791 	node->pprev = NULL;
792 }
793 
794 static inline void __sk_del_node(struct sock *sk)
795 {
796 	__hlist_del(&sk->sk_node);
797 }
798 
799 /* NB: equivalent to hlist_del_init_rcu */
800 static inline bool __sk_del_node_init(struct sock *sk)
801 {
802 	if (sk_hashed(sk)) {
803 		__sk_del_node(sk);
804 		sk_node_init(&sk->sk_node);
805 		return true;
806 	}
807 	return false;
808 }
809 
810 /* Grab socket reference count. This operation is valid only
811    when sk is ALREADY grabbed f.e. it is found in hash table
812    or a list and the lookup is made under lock preventing hash table
813    modifications.
814  */
815 
816 static __always_inline void sock_hold(struct sock *sk)
817 {
818 	refcount_inc(&sk->sk_refcnt);
819 }
820 
821 /* Ungrab socket in the context, which assumes that socket refcnt
822    cannot hit zero, f.e. it is true in context of any socketcall.
823  */
824 static __always_inline void __sock_put(struct sock *sk)
825 {
826 	refcount_dec(&sk->sk_refcnt);
827 }
828 
829 static inline bool sk_del_node_init(struct sock *sk)
830 {
831 	bool rc = __sk_del_node_init(sk);
832 
833 	if (rc)
834 		__sock_put(sk);
835 
836 	return rc;
837 }
838 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
839 
840 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
841 {
842 	if (sk_hashed(sk)) {
843 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
844 		return true;
845 	}
846 	return false;
847 }
848 
849 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
850 {
851 	bool rc = __sk_nulls_del_node_init_rcu(sk);
852 
853 	if (rc)
854 		__sock_put(sk);
855 
856 	return rc;
857 }
858 
859 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
860 {
861 	hlist_add_head(&sk->sk_node, list);
862 }
863 
864 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
865 {
866 	sock_hold(sk);
867 	__sk_add_node(sk, list);
868 }
869 
870 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
871 {
872 	sock_hold(sk);
873 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
874 	    sk->sk_family == AF_INET6)
875 		hlist_add_tail_rcu(&sk->sk_node, list);
876 	else
877 		hlist_add_head_rcu(&sk->sk_node, list);
878 }
879 
880 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
881 {
882 	sock_hold(sk);
883 	hlist_add_tail_rcu(&sk->sk_node, list);
884 }
885 
886 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
887 {
888 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
889 }
890 
891 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
892 {
893 	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
894 }
895 
896 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
897 {
898 	sock_hold(sk);
899 	__sk_nulls_add_node_rcu(sk, list);
900 }
901 
902 static inline void __sk_del_bind_node(struct sock *sk)
903 {
904 	__hlist_del(&sk->sk_bind_node);
905 }
906 
907 static inline void sk_add_bind_node(struct sock *sk,
908 					struct hlist_head *list)
909 {
910 	hlist_add_head(&sk->sk_bind_node, list);
911 }
912 
913 #define sk_for_each(__sk, list) \
914 	hlist_for_each_entry(__sk, list, sk_node)
915 #define sk_for_each_rcu(__sk, list) \
916 	hlist_for_each_entry_rcu(__sk, list, sk_node)
917 #define sk_nulls_for_each(__sk, node, list) \
918 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
919 #define sk_nulls_for_each_rcu(__sk, node, list) \
920 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
921 #define sk_for_each_from(__sk) \
922 	hlist_for_each_entry_from(__sk, sk_node)
923 #define sk_nulls_for_each_from(__sk, node) \
924 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
925 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
926 #define sk_for_each_safe(__sk, tmp, list) \
927 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
928 #define sk_for_each_bound(__sk, list) \
929 	hlist_for_each_entry(__sk, list, sk_bind_node)
930 #define sk_for_each_bound_safe(__sk, tmp, list) \
931 	hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node)
932 
933 /**
934  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
935  * @tpos:	the type * to use as a loop cursor.
936  * @pos:	the &struct hlist_node to use as a loop cursor.
937  * @head:	the head for your list.
938  * @offset:	offset of hlist_node within the struct.
939  *
940  */
941 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
942 	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
943 	     pos != NULL &&						       \
944 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
945 	     pos = rcu_dereference(hlist_next_rcu(pos)))
946 
947 static inline struct user_namespace *sk_user_ns(const struct sock *sk)
948 {
949 	/* Careful only use this in a context where these parameters
950 	 * can not change and must all be valid, such as recvmsg from
951 	 * userspace.
952 	 */
953 	return sk->sk_socket->file->f_cred->user_ns;
954 }
955 
956 /* Sock flags */
957 enum sock_flags {
958 	SOCK_DEAD,
959 	SOCK_DONE,
960 	SOCK_URGINLINE,
961 	SOCK_KEEPOPEN,
962 	SOCK_LINGER,
963 	SOCK_DESTROY,
964 	SOCK_BROADCAST,
965 	SOCK_TIMESTAMP,
966 	SOCK_ZAPPED,
967 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
968 	SOCK_DBG, /* %SO_DEBUG setting */
969 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
970 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
971 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
972 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
973 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
974 	SOCK_FASYNC, /* fasync() active */
975 	SOCK_RXQ_OVFL,
976 	SOCK_ZEROCOPY, /* buffers from userspace */
977 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
978 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
979 		     * Will use last 4 bytes of packet sent from
980 		     * user-space instead.
981 		     */
982 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
983 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
984 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
985 	SOCK_TXTIME,
986 	SOCK_XDP, /* XDP is attached */
987 	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
988 	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
989 	SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */
990 	SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */
991 };
992 
993 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
994 /*
995  * The highest bit of sk_tsflags is reserved for kernel-internal
996  * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that
997  * SOF_TIMESTAMPING* values do not reach this reserved area
998  */
999 #define SOCKCM_FLAG_TS_OPT_ID	BIT(31)
1000 
1001 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
1002 {
1003 	nsk->sk_flags = osk->sk_flags;
1004 }
1005 
1006 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
1007 {
1008 	__set_bit(flag, &sk->sk_flags);
1009 }
1010 
1011 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
1012 {
1013 	__clear_bit(flag, &sk->sk_flags);
1014 }
1015 
1016 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
1017 				     int valbool)
1018 {
1019 	if (valbool)
1020 		sock_set_flag(sk, bit);
1021 	else
1022 		sock_reset_flag(sk, bit);
1023 }
1024 
1025 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
1026 {
1027 	return test_bit(flag, &sk->sk_flags);
1028 }
1029 
1030 #ifdef CONFIG_NET
1031 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
1032 static inline int sk_memalloc_socks(void)
1033 {
1034 	return static_branch_unlikely(&memalloc_socks_key);
1035 }
1036 
1037 void __receive_sock(struct file *file);
1038 #else
1039 
1040 static inline int sk_memalloc_socks(void)
1041 {
1042 	return 0;
1043 }
1044 
1045 static inline void __receive_sock(struct file *file)
1046 { }
1047 #endif
1048 
1049 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1050 {
1051 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1052 }
1053 
1054 static inline void sk_acceptq_removed(struct sock *sk)
1055 {
1056 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1057 }
1058 
1059 static inline void sk_acceptq_added(struct sock *sk)
1060 {
1061 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1062 }
1063 
1064 /* Note: If you think the test should be:
1065  *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1066  * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1067  */
1068 static inline bool sk_acceptq_is_full(const struct sock *sk)
1069 {
1070 	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1071 }
1072 
1073 /*
1074  * Compute minimal free write space needed to queue new packets.
1075  */
1076 static inline int sk_stream_min_wspace(const struct sock *sk)
1077 {
1078 	return READ_ONCE(sk->sk_wmem_queued) >> 1;
1079 }
1080 
1081 static inline int sk_stream_wspace(const struct sock *sk)
1082 {
1083 	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1084 }
1085 
1086 static inline void sk_wmem_queued_add(struct sock *sk, int val)
1087 {
1088 	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1089 }
1090 
1091 static inline void sk_forward_alloc_add(struct sock *sk, int val)
1092 {
1093 	/* Paired with lockless reads of sk->sk_forward_alloc */
1094 	WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1095 }
1096 
1097 void sk_stream_write_space(struct sock *sk);
1098 
1099 /* OOB backlog add */
1100 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1101 {
1102 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
1103 	skb_dst_force(skb);
1104 
1105 	if (!sk->sk_backlog.tail)
1106 		WRITE_ONCE(sk->sk_backlog.head, skb);
1107 	else
1108 		sk->sk_backlog.tail->next = skb;
1109 
1110 	WRITE_ONCE(sk->sk_backlog.tail, skb);
1111 	skb->next = NULL;
1112 }
1113 
1114 /*
1115  * Take into account size of receive queue and backlog queue
1116  * Do not take into account this skb truesize,
1117  * to allow even a single big packet to come.
1118  */
1119 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1120 {
1121 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1122 
1123 	return qsize > limit;
1124 }
1125 
1126 /* The per-socket spinlock must be held here. */
1127 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1128 					      unsigned int limit)
1129 {
1130 	if (sk_rcvqueues_full(sk, limit))
1131 		return -ENOBUFS;
1132 
1133 	/*
1134 	 * If the skb was allocated from pfmemalloc reserves, only
1135 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1136 	 * helping free memory
1137 	 */
1138 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1139 		return -ENOMEM;
1140 
1141 	__sk_add_backlog(sk, skb);
1142 	sk->sk_backlog.len += skb->truesize;
1143 	return 0;
1144 }
1145 
1146 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1147 
1148 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1149 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1150 
1151 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1152 {
1153 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1154 		return __sk_backlog_rcv(sk, skb);
1155 
1156 	return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1157 				  tcp_v6_do_rcv,
1158 				  tcp_v4_do_rcv,
1159 				  sk, skb);
1160 }
1161 
1162 static inline void sk_incoming_cpu_update(struct sock *sk)
1163 {
1164 	int cpu = raw_smp_processor_id();
1165 
1166 	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1167 		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1168 }
1169 
1170 
1171 static inline void sock_rps_save_rxhash(struct sock *sk,
1172 					const struct sk_buff *skb)
1173 {
1174 #ifdef CONFIG_RPS
1175 	/* The following WRITE_ONCE() is paired with the READ_ONCE()
1176 	 * here, and another one in sock_rps_record_flow().
1177 	 */
1178 	if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1179 		WRITE_ONCE(sk->sk_rxhash, skb->hash);
1180 #endif
1181 }
1182 
1183 static inline void sock_rps_reset_rxhash(struct sock *sk)
1184 {
1185 #ifdef CONFIG_RPS
1186 	/* Paired with READ_ONCE() in sock_rps_record_flow() */
1187 	WRITE_ONCE(sk->sk_rxhash, 0);
1188 #endif
1189 }
1190 
1191 #define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1192 	({	int __rc, __dis = __sk->sk_disconnects;			\
1193 		release_sock(__sk);					\
1194 		__rc = __condition;					\
1195 		if (!__rc) {						\
1196 			*(__timeo) = wait_woken(__wait,			\
1197 						TASK_INTERRUPTIBLE,	\
1198 						*(__timeo));		\
1199 		}							\
1200 		sched_annotate_sleep();					\
1201 		lock_sock(__sk);					\
1202 		__rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1203 		__rc;							\
1204 	})
1205 
1206 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1207 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1208 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1209 int sk_stream_error(struct sock *sk, int flags, int err);
1210 void sk_stream_kill_queues(struct sock *sk);
1211 void sk_set_memalloc(struct sock *sk);
1212 void sk_clear_memalloc(struct sock *sk);
1213 
1214 void __sk_flush_backlog(struct sock *sk);
1215 
1216 static inline bool sk_flush_backlog(struct sock *sk)
1217 {
1218 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1219 		__sk_flush_backlog(sk);
1220 		return true;
1221 	}
1222 	return false;
1223 }
1224 
1225 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1226 
1227 struct request_sock_ops;
1228 struct timewait_sock_ops;
1229 struct inet_hashinfo;
1230 struct raw_hashinfo;
1231 struct smc_hashinfo;
1232 struct module;
1233 struct sk_psock;
1234 
1235 /*
1236  * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1237  * un-modified. Special care is taken when initializing object to zero.
1238  */
1239 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1240 {
1241 	if (offsetof(struct sock, sk_node.next) != 0)
1242 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1243 	memset(&sk->sk_node.pprev, 0,
1244 	       size - offsetof(struct sock, sk_node.pprev));
1245 }
1246 
1247 struct proto_accept_arg {
1248 	int flags;
1249 	int err;
1250 	int is_empty;
1251 	bool kern;
1252 };
1253 
1254 /* Networking protocol blocks we attach to sockets.
1255  * socket layer -> transport layer interface
1256  */
1257 struct proto {
1258 	void			(*close)(struct sock *sk,
1259 					long timeout);
1260 	int			(*pre_connect)(struct sock *sk,
1261 					struct sockaddr *uaddr,
1262 					int addr_len);
1263 	int			(*connect)(struct sock *sk,
1264 					struct sockaddr *uaddr,
1265 					int addr_len);
1266 	int			(*disconnect)(struct sock *sk, int flags);
1267 
1268 	struct sock *		(*accept)(struct sock *sk,
1269 					  struct proto_accept_arg *arg);
1270 
1271 	int			(*ioctl)(struct sock *sk, int cmd,
1272 					 int *karg);
1273 	int			(*init)(struct sock *sk);
1274 	void			(*destroy)(struct sock *sk);
1275 	void			(*shutdown)(struct sock *sk, int how);
1276 	int			(*setsockopt)(struct sock *sk, int level,
1277 					int optname, sockptr_t optval,
1278 					unsigned int optlen);
1279 	int			(*getsockopt)(struct sock *sk, int level,
1280 					int optname, char __user *optval,
1281 					int __user *option);
1282 	void			(*keepalive)(struct sock *sk, int valbool);
1283 #ifdef CONFIG_COMPAT
1284 	int			(*compat_ioctl)(struct sock *sk,
1285 					unsigned int cmd, unsigned long arg);
1286 #endif
1287 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1288 					   size_t len);
1289 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1290 					   size_t len, int flags, int *addr_len);
1291 	void			(*splice_eof)(struct socket *sock);
1292 	int			(*bind)(struct sock *sk,
1293 					struct sockaddr *addr, int addr_len);
1294 	int			(*bind_add)(struct sock *sk,
1295 					struct sockaddr *addr, int addr_len);
1296 
1297 	int			(*backlog_rcv) (struct sock *sk,
1298 						struct sk_buff *skb);
1299 	bool			(*bpf_bypass_getsockopt)(int level,
1300 							 int optname);
1301 
1302 	void		(*release_cb)(struct sock *sk);
1303 
1304 	/* Keeping track of sk's, looking them up, and port selection methods. */
1305 	int			(*hash)(struct sock *sk);
1306 	void			(*unhash)(struct sock *sk);
1307 	void			(*rehash)(struct sock *sk);
1308 	int			(*get_port)(struct sock *sk, unsigned short snum);
1309 	void			(*put_port)(struct sock *sk);
1310 #ifdef CONFIG_BPF_SYSCALL
1311 	int			(*psock_update_sk_prot)(struct sock *sk,
1312 							struct sk_psock *psock,
1313 							bool restore);
1314 #endif
1315 
1316 	/* Keeping track of sockets in use */
1317 #ifdef CONFIG_PROC_FS
1318 	unsigned int		inuse_idx;
1319 #endif
1320 
1321 	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1322 	bool			(*sock_is_readable)(struct sock *sk);
1323 	/* Memory pressure */
1324 	void			(*enter_memory_pressure)(struct sock *sk);
1325 	void			(*leave_memory_pressure)(struct sock *sk);
1326 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1327 	int  __percpu		*per_cpu_fw_alloc;
1328 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1329 
1330 	/*
1331 	 * Pressure flag: try to collapse.
1332 	 * Technical note: it is used by multiple contexts non atomically.
1333 	 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1334 	 * All the __sk_mem_schedule() is of this nature: accounting
1335 	 * is strict, actions are advisory and have some latency.
1336 	 */
1337 	unsigned long		*memory_pressure;
1338 	long			*sysctl_mem;
1339 
1340 	int			*sysctl_wmem;
1341 	int			*sysctl_rmem;
1342 	u32			sysctl_wmem_offset;
1343 	u32			sysctl_rmem_offset;
1344 
1345 	int			max_header;
1346 	bool			no_autobind;
1347 
1348 	struct kmem_cache	*slab;
1349 	unsigned int		obj_size;
1350 	unsigned int		ipv6_pinfo_offset;
1351 	slab_flags_t		slab_flags;
1352 	unsigned int		useroffset;	/* Usercopy region offset */
1353 	unsigned int		usersize;	/* Usercopy region size */
1354 
1355 	struct request_sock_ops	*rsk_prot;
1356 	struct timewait_sock_ops *twsk_prot;
1357 
1358 	union {
1359 		struct inet_hashinfo	*hashinfo;
1360 		struct udp_table	*udp_table;
1361 		struct raw_hashinfo	*raw_hash;
1362 		struct smc_hashinfo	*smc_hash;
1363 	} h;
1364 
1365 	struct module		*owner;
1366 
1367 	char			name[32];
1368 
1369 	struct list_head	node;
1370 	int			(*diag_destroy)(struct sock *sk, int err);
1371 } __randomize_layout;
1372 
1373 int proto_register(struct proto *prot, int alloc_slab);
1374 void proto_unregister(struct proto *prot);
1375 int sock_load_diag_module(int family, int protocol);
1376 
1377 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1378 
1379 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1380 {
1381 	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1382 		return false;
1383 
1384 	return sk->sk_prot->stream_memory_free ?
1385 		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1386 				     tcp_stream_memory_free, sk, wake) : true;
1387 }
1388 
1389 static inline bool sk_stream_memory_free(const struct sock *sk)
1390 {
1391 	return __sk_stream_memory_free(sk, 0);
1392 }
1393 
1394 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1395 {
1396 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1397 	       __sk_stream_memory_free(sk, wake);
1398 }
1399 
1400 static inline bool sk_stream_is_writeable(const struct sock *sk)
1401 {
1402 	return __sk_stream_is_writeable(sk, 0);
1403 }
1404 
1405 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1406 					    struct cgroup *ancestor)
1407 {
1408 #ifdef CONFIG_SOCK_CGROUP_DATA
1409 	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1410 				    ancestor);
1411 #else
1412 	return -ENOTSUPP;
1413 #endif
1414 }
1415 
1416 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1417 
1418 static inline void sk_sockets_allocated_dec(struct sock *sk)
1419 {
1420 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1421 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1422 }
1423 
1424 static inline void sk_sockets_allocated_inc(struct sock *sk)
1425 {
1426 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1427 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1428 }
1429 
1430 static inline u64
1431 sk_sockets_allocated_read_positive(struct sock *sk)
1432 {
1433 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1434 }
1435 
1436 static inline int
1437 proto_sockets_allocated_sum_positive(struct proto *prot)
1438 {
1439 	return percpu_counter_sum_positive(prot->sockets_allocated);
1440 }
1441 
1442 #ifdef CONFIG_PROC_FS
1443 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
1444 struct prot_inuse {
1445 	int all;
1446 	int val[PROTO_INUSE_NR];
1447 };
1448 
1449 static inline void sock_prot_inuse_add(const struct net *net,
1450 				       const struct proto *prot, int val)
1451 {
1452 	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1453 }
1454 
1455 static inline void sock_inuse_add(const struct net *net, int val)
1456 {
1457 	this_cpu_add(net->core.prot_inuse->all, val);
1458 }
1459 
1460 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1461 int sock_inuse_get(struct net *net);
1462 #else
1463 static inline void sock_prot_inuse_add(const struct net *net,
1464 				       const struct proto *prot, int val)
1465 {
1466 }
1467 
1468 static inline void sock_inuse_add(const struct net *net, int val)
1469 {
1470 }
1471 #endif
1472 
1473 
1474 /* With per-bucket locks this operation is not-atomic, so that
1475  * this version is not worse.
1476  */
1477 static inline int __sk_prot_rehash(struct sock *sk)
1478 {
1479 	sk->sk_prot->unhash(sk);
1480 	return sk->sk_prot->hash(sk);
1481 }
1482 
1483 /* About 10 seconds */
1484 #define SOCK_DESTROY_TIME (10*HZ)
1485 
1486 /* Sockets 0-1023 can't be bound to unless you are superuser */
1487 #define PROT_SOCK	1024
1488 
1489 #define SHUTDOWN_MASK	3
1490 #define RCV_SHUTDOWN	1
1491 #define SEND_SHUTDOWN	2
1492 
1493 #define SOCK_BINDADDR_LOCK	4
1494 #define SOCK_BINDPORT_LOCK	8
1495 /**
1496  * define SOCK_CONNECT_BIND - &sock->sk_userlocks flag for auto-bind at connect() time
1497  */
1498 #define SOCK_CONNECT_BIND	16
1499 
1500 struct socket_alloc {
1501 	struct socket socket;
1502 	struct inode vfs_inode;
1503 };
1504 
1505 static inline struct socket *SOCKET_I(struct inode *inode)
1506 {
1507 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1508 }
1509 
1510 static inline struct inode *SOCK_INODE(struct socket *socket)
1511 {
1512 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1513 }
1514 
1515 /*
1516  * Functions for memory accounting
1517  */
1518 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1519 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1520 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1521 void __sk_mem_reclaim(struct sock *sk, int amount);
1522 
1523 #define SK_MEM_SEND	0
1524 #define SK_MEM_RECV	1
1525 
1526 /* sysctl_mem values are in pages */
1527 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1528 {
1529 	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1530 }
1531 
1532 static inline int sk_mem_pages(int amt)
1533 {
1534 	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1535 }
1536 
1537 static inline bool sk_has_account(struct sock *sk)
1538 {
1539 	/* return true if protocol supports memory accounting */
1540 	return !!sk->sk_prot->memory_allocated;
1541 }
1542 
1543 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1544 {
1545 	int delta;
1546 
1547 	if (!sk_has_account(sk))
1548 		return true;
1549 	delta = size - sk->sk_forward_alloc;
1550 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1551 }
1552 
1553 static inline bool
1554 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc)
1555 {
1556 	int delta;
1557 
1558 	if (!sk_has_account(sk))
1559 		return true;
1560 	delta = size - sk->sk_forward_alloc;
1561 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1562 	       pfmemalloc;
1563 }
1564 
1565 static inline bool
1566 sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size)
1567 {
1568 	return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb));
1569 }
1570 
1571 static inline int sk_unused_reserved_mem(const struct sock *sk)
1572 {
1573 	int unused_mem;
1574 
1575 	if (likely(!sk->sk_reserved_mem))
1576 		return 0;
1577 
1578 	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1579 			atomic_read(&sk->sk_rmem_alloc);
1580 
1581 	return unused_mem > 0 ? unused_mem : 0;
1582 }
1583 
1584 static inline void sk_mem_reclaim(struct sock *sk)
1585 {
1586 	int reclaimable;
1587 
1588 	if (!sk_has_account(sk))
1589 		return;
1590 
1591 	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1592 
1593 	if (reclaimable >= (int)PAGE_SIZE)
1594 		__sk_mem_reclaim(sk, reclaimable);
1595 }
1596 
1597 static inline void sk_mem_reclaim_final(struct sock *sk)
1598 {
1599 	sk->sk_reserved_mem = 0;
1600 	sk_mem_reclaim(sk);
1601 }
1602 
1603 static inline void sk_mem_charge(struct sock *sk, int size)
1604 {
1605 	if (!sk_has_account(sk))
1606 		return;
1607 	sk_forward_alloc_add(sk, -size);
1608 }
1609 
1610 static inline void sk_mem_uncharge(struct sock *sk, int size)
1611 {
1612 	if (!sk_has_account(sk))
1613 		return;
1614 	sk_forward_alloc_add(sk, size);
1615 	sk_mem_reclaim(sk);
1616 }
1617 
1618 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
1619 static inline void sk_owner_set(struct sock *sk, struct module *owner)
1620 {
1621 	__module_get(owner);
1622 	sk->sk_owner = owner;
1623 }
1624 
1625 static inline void sk_owner_clear(struct sock *sk)
1626 {
1627 	sk->sk_owner = NULL;
1628 }
1629 
1630 static inline void sk_owner_put(struct sock *sk)
1631 {
1632 	module_put(sk->sk_owner);
1633 }
1634 #else
1635 static inline void sk_owner_set(struct sock *sk, struct module *owner)
1636 {
1637 }
1638 
1639 static inline void sk_owner_clear(struct sock *sk)
1640 {
1641 }
1642 
1643 static inline void sk_owner_put(struct sock *sk)
1644 {
1645 }
1646 #endif
1647 /*
1648  * Macro so as to not evaluate some arguments when
1649  * lockdep is not enabled.
1650  *
1651  * Mark both the sk_lock and the sk_lock.slock as a
1652  * per-address-family lock class.
1653  */
1654 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1655 do {									\
1656 	sk_owner_set(sk, THIS_MODULE);					\
1657 	sk->sk_lock.owned = 0;						\
1658 	init_waitqueue_head(&sk->sk_lock.wq);				\
1659 	spin_lock_init(&(sk)->sk_lock.slock);				\
1660 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1661 				   sizeof((sk)->sk_lock));		\
1662 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1663 				   (skey), (sname));			\
1664 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1665 } while (0)
1666 
1667 static inline bool lockdep_sock_is_held(const struct sock *sk)
1668 {
1669 	return lockdep_is_held(&sk->sk_lock) ||
1670 	       lockdep_is_held(&sk->sk_lock.slock);
1671 }
1672 
1673 void lock_sock_nested(struct sock *sk, int subclass);
1674 
1675 static inline void lock_sock(struct sock *sk)
1676 {
1677 	lock_sock_nested(sk, 0);
1678 }
1679 
1680 void __lock_sock(struct sock *sk);
1681 void __release_sock(struct sock *sk);
1682 void release_sock(struct sock *sk);
1683 
1684 /* BH context may only use the following locking interface. */
1685 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1686 #define bh_lock_sock_nested(__sk) \
1687 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1688 				SINGLE_DEPTH_NESTING)
1689 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1690 
1691 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1692 
1693 /**
1694  * lock_sock_fast - fast version of lock_sock
1695  * @sk: socket
1696  *
1697  * This version should be used for very small section, where process won't block
1698  * return false if fast path is taken:
1699  *
1700  *   sk_lock.slock locked, owned = 0, BH disabled
1701  *
1702  * return true if slow path is taken:
1703  *
1704  *   sk_lock.slock unlocked, owned = 1, BH enabled
1705  */
1706 static inline bool lock_sock_fast(struct sock *sk)
1707 {
1708 	/* The sk_lock has mutex_lock() semantics here. */
1709 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1710 
1711 	return __lock_sock_fast(sk);
1712 }
1713 
1714 /* fast socket lock variant for caller already holding a [different] socket lock */
1715 static inline bool lock_sock_fast_nested(struct sock *sk)
1716 {
1717 	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1718 
1719 	return __lock_sock_fast(sk);
1720 }
1721 
1722 /**
1723  * unlock_sock_fast - complement of lock_sock_fast
1724  * @sk: socket
1725  * @slow: slow mode
1726  *
1727  * fast unlock socket for user context.
1728  * If slow mode is on, we call regular release_sock()
1729  */
1730 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1731 	__releases(&sk->sk_lock.slock)
1732 {
1733 	if (slow) {
1734 		release_sock(sk);
1735 		__release(&sk->sk_lock.slock);
1736 	} else {
1737 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1738 		spin_unlock_bh(&sk->sk_lock.slock);
1739 	}
1740 }
1741 
1742 void sockopt_lock_sock(struct sock *sk);
1743 void sockopt_release_sock(struct sock *sk);
1744 bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1745 bool sockopt_capable(int cap);
1746 
1747 /* Used by processes to "lock" a socket state, so that
1748  * interrupts and bottom half handlers won't change it
1749  * from under us. It essentially blocks any incoming
1750  * packets, so that we won't get any new data or any
1751  * packets that change the state of the socket.
1752  *
1753  * While locked, BH processing will add new packets to
1754  * the backlog queue.  This queue is processed by the
1755  * owner of the socket lock right before it is released.
1756  *
1757  * Since ~2.3.5 it is also exclusive sleep lock serializing
1758  * accesses from user process context.
1759  */
1760 
1761 static inline void sock_owned_by_me(const struct sock *sk)
1762 {
1763 #ifdef CONFIG_LOCKDEP
1764 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1765 #endif
1766 }
1767 
1768 static inline void sock_not_owned_by_me(const struct sock *sk)
1769 {
1770 #ifdef CONFIG_LOCKDEP
1771 	WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1772 #endif
1773 }
1774 
1775 static inline bool sock_owned_by_user(const struct sock *sk)
1776 {
1777 	sock_owned_by_me(sk);
1778 	return sk->sk_lock.owned;
1779 }
1780 
1781 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1782 {
1783 	return sk->sk_lock.owned;
1784 }
1785 
1786 static inline void sock_release_ownership(struct sock *sk)
1787 {
1788 	DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1789 	sk->sk_lock.owned = 0;
1790 
1791 	/* The sk_lock has mutex_unlock() semantics: */
1792 	mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1793 }
1794 
1795 /* no reclassification while locks are held */
1796 static inline bool sock_allow_reclassification(const struct sock *csk)
1797 {
1798 	struct sock *sk = (struct sock *)csk;
1799 
1800 	return !sock_owned_by_user_nocheck(sk) &&
1801 		!spin_is_locked(&sk->sk_lock.slock);
1802 }
1803 
1804 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1805 		      struct proto *prot, int kern);
1806 void sk_free(struct sock *sk);
1807 void sk_net_refcnt_upgrade(struct sock *sk);
1808 void sk_destruct(struct sock *sk);
1809 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1810 
1811 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1812 			     gfp_t priority);
1813 void __sock_wfree(struct sk_buff *skb);
1814 void sock_wfree(struct sk_buff *skb);
1815 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1816 			     gfp_t priority);
1817 void skb_orphan_partial(struct sk_buff *skb);
1818 void sock_rfree(struct sk_buff *skb);
1819 void sock_efree(struct sk_buff *skb);
1820 #ifdef CONFIG_INET
1821 void sock_edemux(struct sk_buff *skb);
1822 void sock_pfree(struct sk_buff *skb);
1823 
1824 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk)
1825 {
1826 	skb_orphan(skb);
1827 	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1828 		skb->sk = sk;
1829 		skb->destructor = sock_edemux;
1830 	}
1831 }
1832 #else
1833 #define sock_edemux sock_efree
1834 #endif
1835 
1836 int sk_setsockopt(struct sock *sk, int level, int optname,
1837 		  sockptr_t optval, unsigned int optlen);
1838 int sock_setsockopt(struct socket *sock, int level, int op,
1839 		    sockptr_t optval, unsigned int optlen);
1840 int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1841 		       int optname, sockptr_t optval, int optlen);
1842 int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1843 		       int optname, sockptr_t optval, sockptr_t optlen);
1844 
1845 int sk_getsockopt(struct sock *sk, int level, int optname,
1846 		  sockptr_t optval, sockptr_t optlen);
1847 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1848 		   bool timeval, bool time32);
1849 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1850 				     unsigned long data_len, int noblock,
1851 				     int *errcode, int max_page_order);
1852 
1853 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1854 						  unsigned long size,
1855 						  int noblock, int *errcode)
1856 {
1857 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1858 }
1859 
1860 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1861 void *sock_kmemdup(struct sock *sk, const void *src,
1862 		   int size, gfp_t priority);
1863 void sock_kfree_s(struct sock *sk, void *mem, int size);
1864 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1865 void sk_send_sigurg(struct sock *sk);
1866 
1867 static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1868 {
1869 	if (sk->sk_socket)
1870 		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1871 	WRITE_ONCE(sk->sk_prot, proto);
1872 }
1873 
1874 struct sockcm_cookie {
1875 	u64 transmit_time;
1876 	u32 mark;
1877 	u32 tsflags;
1878 	u32 ts_opt_id;
1879 	u32 priority;
1880 	u32 dmabuf_id;
1881 };
1882 
1883 static inline void sockcm_init(struct sockcm_cookie *sockc,
1884 			       const struct sock *sk)
1885 {
1886 	*sockc = (struct sockcm_cookie) {
1887 		.mark = READ_ONCE(sk->sk_mark),
1888 		.tsflags = READ_ONCE(sk->sk_tsflags),
1889 		.priority = READ_ONCE(sk->sk_priority),
1890 	};
1891 }
1892 
1893 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1894 		     struct sockcm_cookie *sockc);
1895 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1896 		   struct sockcm_cookie *sockc);
1897 
1898 /*
1899  * Functions to fill in entries in struct proto_ops when a protocol
1900  * does not implement a particular function.
1901  */
1902 int sock_no_bind(struct socket *, struct sockaddr *, int);
1903 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1904 int sock_no_socketpair(struct socket *, struct socket *);
1905 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *);
1906 int sock_no_getname(struct socket *, struct sockaddr *, int);
1907 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1908 int sock_no_listen(struct socket *, int);
1909 int sock_no_shutdown(struct socket *, int);
1910 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1911 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1912 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1913 int sock_no_mmap(struct file *file, struct socket *sock,
1914 		 struct vm_area_struct *vma);
1915 
1916 /*
1917  * Functions to fill in entries in struct proto_ops when a protocol
1918  * uses the inet style.
1919  */
1920 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1921 				  char __user *optval, int __user *optlen);
1922 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1923 			int flags);
1924 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1925 			   sockptr_t optval, unsigned int optlen);
1926 
1927 void sk_common_release(struct sock *sk);
1928 
1929 /*
1930  *	Default socket callbacks and setup code
1931  */
1932 
1933 /* Initialise core socket variables using an explicit uid. */
1934 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1935 
1936 /* Initialise core socket variables.
1937  * Assumes struct socket *sock is embedded in a struct socket_alloc.
1938  */
1939 void sock_init_data(struct socket *sock, struct sock *sk);
1940 
1941 /*
1942  * Socket reference counting postulates.
1943  *
1944  * * Each user of socket SHOULD hold a reference count.
1945  * * Each access point to socket (an hash table bucket, reference from a list,
1946  *   running timer, skb in flight MUST hold a reference count.
1947  * * When reference count hits 0, it means it will never increase back.
1948  * * When reference count hits 0, it means that no references from
1949  *   outside exist to this socket and current process on current CPU
1950  *   is last user and may/should destroy this socket.
1951  * * sk_free is called from any context: process, BH, IRQ. When
1952  *   it is called, socket has no references from outside -> sk_free
1953  *   may release descendant resources allocated by the socket, but
1954  *   to the time when it is called, socket is NOT referenced by any
1955  *   hash tables, lists etc.
1956  * * Packets, delivered from outside (from network or from another process)
1957  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1958  *   when they sit in queue. Otherwise, packets will leak to hole, when
1959  *   socket is looked up by one cpu and unhasing is made by another CPU.
1960  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1961  *   (leak to backlog). Packet socket does all the processing inside
1962  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1963  *   use separate SMP lock, so that they are prone too.
1964  */
1965 
1966 /* Ungrab socket and destroy it, if it was the last reference. */
1967 static inline void sock_put(struct sock *sk)
1968 {
1969 	if (refcount_dec_and_test(&sk->sk_refcnt))
1970 		sk_free(sk);
1971 }
1972 /* Generic version of sock_put(), dealing with all sockets
1973  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1974  */
1975 void sock_gen_put(struct sock *sk);
1976 
1977 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1978 		     unsigned int trim_cap, bool refcounted);
1979 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1980 				 const int nested)
1981 {
1982 	return __sk_receive_skb(sk, skb, nested, 1, true);
1983 }
1984 
1985 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1986 {
1987 	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1988 	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1989 		return;
1990 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1991 	 * other WRITE_ONCE() because socket lock might be not held.
1992 	 */
1993 	if (READ_ONCE(sk->sk_tx_queue_mapping) != tx_queue) {
1994 		WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1995 		WRITE_ONCE(sk->sk_tx_queue_mapping_jiffies, jiffies);
1996 		return;
1997 	}
1998 
1999 	/* Refresh sk_tx_queue_mapping_jiffies if too old. */
2000 	if (time_is_before_jiffies(READ_ONCE(sk->sk_tx_queue_mapping_jiffies) + HZ))
2001 		WRITE_ONCE(sk->sk_tx_queue_mapping_jiffies, jiffies);
2002 }
2003 
2004 #define NO_QUEUE_MAPPING	USHRT_MAX
2005 
2006 static inline void sk_tx_queue_clear(struct sock *sk)
2007 {
2008 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
2009 	 * other WRITE_ONCE() because socket lock might be not held.
2010 	 */
2011 	WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
2012 }
2013 
2014 int sk_tx_queue_get(const struct sock *sk);
2015 
2016 static inline void __sk_rx_queue_set(struct sock *sk,
2017 				     const struct sk_buff *skb,
2018 				     bool force_set)
2019 {
2020 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2021 	if (skb_rx_queue_recorded(skb)) {
2022 		u16 rx_queue = skb_get_rx_queue(skb);
2023 
2024 		if (force_set ||
2025 		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2026 			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2027 	}
2028 #endif
2029 }
2030 
2031 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2032 {
2033 	__sk_rx_queue_set(sk, skb, true);
2034 }
2035 
2036 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2037 {
2038 	__sk_rx_queue_set(sk, skb, false);
2039 }
2040 
2041 static inline void sk_rx_queue_clear(struct sock *sk)
2042 {
2043 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2044 	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2045 #endif
2046 }
2047 
2048 static inline int sk_rx_queue_get(const struct sock *sk)
2049 {
2050 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2051 	if (sk) {
2052 		int res = READ_ONCE(sk->sk_rx_queue_mapping);
2053 
2054 		if (res != NO_QUEUE_MAPPING)
2055 			return res;
2056 	}
2057 #endif
2058 
2059 	return -1;
2060 }
2061 
2062 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2063 {
2064 	sk->sk_socket = sock;
2065 	if (sock) {
2066 		WRITE_ONCE(sk->sk_uid, SOCK_INODE(sock)->i_uid);
2067 		WRITE_ONCE(sk->sk_ino, SOCK_INODE(sock)->i_ino);
2068 	} else {
2069 		/* Note: sk_uid is unchanged. */
2070 		WRITE_ONCE(sk->sk_ino, 0);
2071 	}
2072 }
2073 
2074 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2075 {
2076 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2077 	return &rcu_dereference_raw(sk->sk_wq)->wait;
2078 }
2079 /* Detach socket from process context.
2080  * Announce socket dead, detach it from wait queue and inode.
2081  * Note that parent inode held reference count on this struct sock,
2082  * we do not release it in this function, because protocol
2083  * probably wants some additional cleanups or even continuing
2084  * to work with this socket (TCP).
2085  */
2086 static inline void sock_orphan(struct sock *sk)
2087 {
2088 	write_lock_bh(&sk->sk_callback_lock);
2089 	sock_set_flag(sk, SOCK_DEAD);
2090 	sk_set_socket(sk, NULL);
2091 	sk->sk_wq  = NULL;
2092 	write_unlock_bh(&sk->sk_callback_lock);
2093 }
2094 
2095 static inline void sock_graft(struct sock *sk, struct socket *parent)
2096 {
2097 	WARN_ON(parent->sk);
2098 	write_lock_bh(&sk->sk_callback_lock);
2099 	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2100 	parent->sk = sk;
2101 	sk_set_socket(sk, parent);
2102 	security_sock_graft(sk, parent);
2103 	write_unlock_bh(&sk->sk_callback_lock);
2104 }
2105 
2106 static inline unsigned long sock_i_ino(const struct sock *sk)
2107 {
2108 	/* Paired with WRITE_ONCE() in sock_graft() and sock_orphan() */
2109 	return READ_ONCE(sk->sk_ino);
2110 }
2111 
2112 static inline kuid_t sk_uid(const struct sock *sk)
2113 {
2114 	/* Paired with WRITE_ONCE() in sockfs_setattr() */
2115 	return READ_ONCE(sk->sk_uid);
2116 }
2117 
2118 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2119 {
2120 	return sk ? sk_uid(sk) : make_kuid(net->user_ns, 0);
2121 }
2122 
2123 static inline u32 net_tx_rndhash(void)
2124 {
2125 	u32 v = get_random_u32();
2126 
2127 	return v ?: 1;
2128 }
2129 
2130 static inline void sk_set_txhash(struct sock *sk)
2131 {
2132 	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2133 	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2134 }
2135 
2136 static inline bool sk_rethink_txhash(struct sock *sk)
2137 {
2138 	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2139 		sk_set_txhash(sk);
2140 		return true;
2141 	}
2142 	return false;
2143 }
2144 
2145 static inline struct dst_entry *
2146 __sk_dst_get(const struct sock *sk)
2147 {
2148 	return rcu_dereference_check(sk->sk_dst_cache,
2149 				     lockdep_sock_is_held(sk));
2150 }
2151 
2152 static inline struct dst_entry *
2153 sk_dst_get(const struct sock *sk)
2154 {
2155 	struct dst_entry *dst;
2156 
2157 	rcu_read_lock();
2158 	dst = rcu_dereference(sk->sk_dst_cache);
2159 	if (dst && !rcuref_get(&dst->__rcuref))
2160 		dst = NULL;
2161 	rcu_read_unlock();
2162 	return dst;
2163 }
2164 
2165 static inline void __dst_negative_advice(struct sock *sk)
2166 {
2167 	struct dst_entry *dst = __sk_dst_get(sk);
2168 
2169 	if (dst && dst->ops->negative_advice)
2170 		dst->ops->negative_advice(sk, dst);
2171 }
2172 
2173 static inline void dst_negative_advice(struct sock *sk)
2174 {
2175 	sk_rethink_txhash(sk);
2176 	__dst_negative_advice(sk);
2177 }
2178 
2179 static inline void
2180 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2181 {
2182 	struct dst_entry *old_dst;
2183 
2184 	sk_tx_queue_clear(sk);
2185 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2186 	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2187 					    lockdep_sock_is_held(sk));
2188 	rcu_assign_pointer(sk->sk_dst_cache, dst);
2189 	dst_release(old_dst);
2190 }
2191 
2192 static inline void
2193 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2194 {
2195 	struct dst_entry *old_dst;
2196 
2197 	sk_tx_queue_clear(sk);
2198 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2199 	old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst)));
2200 	dst_release(old_dst);
2201 }
2202 
2203 static inline void
2204 __sk_dst_reset(struct sock *sk)
2205 {
2206 	__sk_dst_set(sk, NULL);
2207 }
2208 
2209 static inline void
2210 sk_dst_reset(struct sock *sk)
2211 {
2212 	sk_dst_set(sk, NULL);
2213 }
2214 
2215 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2216 
2217 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2218 
2219 static inline void sk_dst_confirm(struct sock *sk)
2220 {
2221 	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2222 		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2223 }
2224 
2225 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2226 {
2227 	if (skb_get_dst_pending_confirm(skb)) {
2228 		struct sock *sk = skb->sk;
2229 
2230 		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2231 			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2232 		neigh_confirm(n);
2233 	}
2234 }
2235 
2236 bool sk_mc_loop(const struct sock *sk);
2237 
2238 static inline bool sk_can_gso(const struct sock *sk)
2239 {
2240 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2241 }
2242 
2243 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2244 
2245 static inline void sk_gso_disable(struct sock *sk)
2246 {
2247 	sk->sk_gso_disabled = 1;
2248 	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2249 }
2250 
2251 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2252 					   struct iov_iter *from, char *to,
2253 					   int copy, int offset)
2254 {
2255 	if (skb->ip_summed == CHECKSUM_NONE) {
2256 		__wsum csum = 0;
2257 		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2258 			return -EFAULT;
2259 		skb->csum = csum_block_add(skb->csum, csum, offset);
2260 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2261 		if (!copy_from_iter_full_nocache(to, copy, from))
2262 			return -EFAULT;
2263 	} else if (!copy_from_iter_full(to, copy, from))
2264 		return -EFAULT;
2265 
2266 	return 0;
2267 }
2268 
2269 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2270 				       struct iov_iter *from, int copy)
2271 {
2272 	int err, offset = skb->len;
2273 
2274 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2275 				       copy, offset);
2276 	if (err)
2277 		__skb_trim(skb, offset);
2278 
2279 	return err;
2280 }
2281 
2282 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2283 					   struct sk_buff *skb,
2284 					   struct page *page,
2285 					   int off, int copy)
2286 {
2287 	int err;
2288 
2289 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2290 				       copy, skb->len);
2291 	if (err)
2292 		return err;
2293 
2294 	skb_len_add(skb, copy);
2295 	sk_wmem_queued_add(sk, copy);
2296 	sk_mem_charge(sk, copy);
2297 	return 0;
2298 }
2299 
2300 #define SK_WMEM_ALLOC_BIAS 1
2301 /**
2302  * sk_wmem_alloc_get - returns write allocations
2303  * @sk: socket
2304  *
2305  * Return: sk_wmem_alloc minus initial offset of one
2306  */
2307 static inline int sk_wmem_alloc_get(const struct sock *sk)
2308 {
2309 	return refcount_read(&sk->sk_wmem_alloc) - SK_WMEM_ALLOC_BIAS;
2310 }
2311 
2312 /**
2313  * sk_rmem_alloc_get - returns read allocations
2314  * @sk: socket
2315  *
2316  * Return: sk_rmem_alloc
2317  */
2318 static inline int sk_rmem_alloc_get(const struct sock *sk)
2319 {
2320 	return atomic_read(&sk->sk_rmem_alloc);
2321 }
2322 
2323 /**
2324  * sk_has_allocations - check if allocations are outstanding
2325  * @sk: socket
2326  *
2327  * Return: true if socket has write or read allocations
2328  */
2329 static inline bool sk_has_allocations(const struct sock *sk)
2330 {
2331 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2332 }
2333 
2334 /**
2335  * skwq_has_sleeper - check if there are any waiting processes
2336  * @wq: struct socket_wq
2337  *
2338  * Return: true if socket_wq has waiting processes
2339  *
2340  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2341  * barrier call. They were added due to the race found within the tcp code.
2342  *
2343  * Consider following tcp code paths::
2344  *
2345  *   CPU1                CPU2
2346  *   sys_select          receive packet
2347  *   ...                 ...
2348  *   __add_wait_queue    update tp->rcv_nxt
2349  *   ...                 ...
2350  *   tp->rcv_nxt check   sock_def_readable
2351  *   ...                 {
2352  *   schedule               rcu_read_lock();
2353  *                          wq = rcu_dereference(sk->sk_wq);
2354  *                          if (wq && waitqueue_active(&wq->wait))
2355  *                              wake_up_interruptible(&wq->wait)
2356  *                          ...
2357  *                       }
2358  *
2359  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2360  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2361  * could then endup calling schedule and sleep forever if there are no more
2362  * data on the socket.
2363  *
2364  */
2365 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2366 {
2367 	return wq && wq_has_sleeper(&wq->wait);
2368 }
2369 
2370 /**
2371  * sock_poll_wait - wrapper for the poll_wait call.
2372  * @filp:           file
2373  * @sock:           socket to wait on
2374  * @p:              poll_table
2375  *
2376  * See the comments in the wq_has_sleeper function.
2377  */
2378 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2379 				  poll_table *p)
2380 {
2381 	/* Provides a barrier we need to be sure we are in sync
2382 	 * with the socket flags modification.
2383 	 *
2384 	 * This memory barrier is paired in the wq_has_sleeper.
2385 	 */
2386 	poll_wait(filp, &sock->wq.wait, p);
2387 }
2388 
2389 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2390 {
2391 	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2392 	u32 txhash = READ_ONCE(sk->sk_txhash);
2393 
2394 	if (txhash) {
2395 		skb->l4_hash = 1;
2396 		skb->hash = txhash;
2397 	}
2398 }
2399 
2400 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2401 
2402 /*
2403  *	Queue a received datagram if it will fit. Stream and sequenced
2404  *	protocols can't normally use this as they need to fit buffers in
2405  *	and play with them.
2406  *
2407  *	Inlined as it's very short and called for pretty much every
2408  *	packet ever received.
2409  */
2410 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2411 {
2412 	skb_orphan(skb);
2413 	skb->sk = sk;
2414 	skb->destructor = sock_rfree;
2415 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2416 	sk_mem_charge(sk, skb->truesize);
2417 }
2418 
2419 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2420 {
2421 	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2422 		skb_orphan(skb);
2423 		skb->destructor = sock_efree;
2424 		skb->sk = sk;
2425 		return true;
2426 	}
2427 	return false;
2428 }
2429 
2430 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2431 {
2432 	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2433 	if (skb) {
2434 		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2435 			skb_set_owner_r(skb, sk);
2436 			return skb;
2437 		}
2438 		__kfree_skb(skb);
2439 	}
2440 	return NULL;
2441 }
2442 
2443 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2444 {
2445 	if (skb->destructor != sock_wfree) {
2446 		skb_orphan(skb);
2447 		return;
2448 	}
2449 	skb->slow_gro = 1;
2450 }
2451 
2452 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2453 		    unsigned long expires);
2454 
2455 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2456 
2457 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2458 
2459 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2460 			struct sk_buff *skb, unsigned int flags,
2461 			void (*destructor)(struct sock *sk,
2462 					   struct sk_buff *skb));
2463 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2464 
2465 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2466 			      enum skb_drop_reason *reason);
2467 
2468 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2469 {
2470 	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2471 }
2472 
2473 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2474 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2475 
2476 /*
2477  *	Recover an error report and clear atomically
2478  */
2479 
2480 static inline int sock_error(struct sock *sk)
2481 {
2482 	int err;
2483 
2484 	/* Avoid an atomic operation for the common case.
2485 	 * This is racy since another cpu/thread can change sk_err under us.
2486 	 */
2487 	if (likely(data_race(!sk->sk_err)))
2488 		return 0;
2489 
2490 	err = xchg(&sk->sk_err, 0);
2491 	return -err;
2492 }
2493 
2494 void sk_error_report(struct sock *sk);
2495 
2496 static inline unsigned long sock_wspace(struct sock *sk)
2497 {
2498 	int amt = 0;
2499 
2500 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2501 		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2502 		if (amt < 0)
2503 			amt = 0;
2504 	}
2505 	return amt;
2506 }
2507 
2508 /* Note:
2509  *  We use sk->sk_wq_raw, from contexts knowing this
2510  *  pointer is not NULL and cannot disappear/change.
2511  */
2512 static inline void sk_set_bit(int nr, struct sock *sk)
2513 {
2514 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2515 	    !sock_flag(sk, SOCK_FASYNC))
2516 		return;
2517 
2518 	set_bit(nr, &sk->sk_wq_raw->flags);
2519 }
2520 
2521 static inline void sk_clear_bit(int nr, struct sock *sk)
2522 {
2523 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2524 	    !sock_flag(sk, SOCK_FASYNC))
2525 		return;
2526 
2527 	clear_bit(nr, &sk->sk_wq_raw->flags);
2528 }
2529 
2530 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2531 {
2532 	if (sock_flag(sk, SOCK_FASYNC)) {
2533 		rcu_read_lock();
2534 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2535 		rcu_read_unlock();
2536 	}
2537 }
2538 
2539 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band)
2540 {
2541 	if (unlikely(sock_flag(sk, SOCK_FASYNC)))
2542 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2543 }
2544 
2545 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2546  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2547  * Note: for send buffers, TCP works better if we can build two skbs at
2548  * minimum.
2549  */
2550 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2551 
2552 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2553 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2554 
2555 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2556 {
2557 	u32 val;
2558 
2559 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2560 		return;
2561 
2562 	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2563 	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2564 
2565 	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2566 }
2567 
2568 /**
2569  * sk_page_frag - return an appropriate page_frag
2570  * @sk: socket
2571  *
2572  * Use the per task page_frag instead of the per socket one for
2573  * optimization when we know that we're in process context and own
2574  * everything that's associated with %current.
2575  *
2576  * Both direct reclaim and page faults can nest inside other
2577  * socket operations and end up recursing into sk_page_frag()
2578  * while it's already in use: explicitly avoid task page_frag
2579  * when users disable sk_use_task_frag.
2580  *
2581  * Return: a per task page_frag if context allows that,
2582  * otherwise a per socket one.
2583  */
2584 static inline struct page_frag *sk_page_frag(struct sock *sk)
2585 {
2586 	if (sk->sk_use_task_frag)
2587 		return &current->task_frag;
2588 
2589 	return &sk->sk_frag;
2590 }
2591 
2592 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2593 
2594 /*
2595  *	Default write policy as shown to user space via poll/select/SIGIO
2596  */
2597 static inline bool sock_writeable(const struct sock *sk)
2598 {
2599 	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2600 }
2601 
2602 static inline gfp_t gfp_any(void)
2603 {
2604 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2605 }
2606 
2607 static inline gfp_t gfp_memcg_charge(void)
2608 {
2609 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2610 }
2611 
2612 #ifdef CONFIG_MEMCG
2613 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2614 {
2615 	return sk->sk_memcg;
2616 }
2617 
2618 static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2619 {
2620 	return mem_cgroup_sockets_enabled && mem_cgroup_from_sk(sk);
2621 }
2622 
2623 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2624 {
2625 	struct mem_cgroup *memcg = mem_cgroup_from_sk(sk);
2626 
2627 #ifdef CONFIG_MEMCG_V1
2628 	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2629 		return !!memcg->tcpmem_pressure;
2630 #endif /* CONFIG_MEMCG_V1 */
2631 
2632 	do {
2633 		if (time_before64(get_jiffies_64(), mem_cgroup_get_socket_pressure(memcg)))
2634 			return true;
2635 	} while ((memcg = parent_mem_cgroup(memcg)));
2636 
2637 	return false;
2638 }
2639 #else
2640 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2641 {
2642 	return NULL;
2643 }
2644 
2645 static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2646 {
2647 	return false;
2648 }
2649 
2650 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2651 {
2652 	return false;
2653 }
2654 #endif
2655 
2656 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2657 {
2658 	return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo);
2659 }
2660 
2661 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2662 {
2663 	return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo);
2664 }
2665 
2666 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2667 {
2668 	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2669 
2670 	return v ?: 1;
2671 }
2672 
2673 /* Alas, with timeout socket operations are not restartable.
2674  * Compare this to poll().
2675  */
2676 static inline int sock_intr_errno(long timeo)
2677 {
2678 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2679 }
2680 
2681 struct sock_skb_cb {
2682 	u32 dropcount;
2683 };
2684 
2685 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2686  * using skb->cb[] would keep using it directly and utilize its
2687  * alignment guarantee.
2688  */
2689 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \
2690 			    sizeof(struct sock_skb_cb))
2691 
2692 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2693 			    SOCK_SKB_CB_OFFSET))
2694 
2695 #define sock_skb_cb_check_size(size) \
2696 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2697 
2698 static inline void sk_drops_add(struct sock *sk, int segs)
2699 {
2700 	struct numa_drop_counters *ndc = sk->sk_drop_counters;
2701 
2702 	if (ndc)
2703 		numa_drop_add(ndc, segs);
2704 	else
2705 		atomic_add(segs, &sk->sk_drops);
2706 }
2707 
2708 static inline void sk_drops_inc(struct sock *sk)
2709 {
2710 	sk_drops_add(sk, 1);
2711 }
2712 
2713 static inline int sk_drops_read(const struct sock *sk)
2714 {
2715 	const struct numa_drop_counters *ndc = sk->sk_drop_counters;
2716 
2717 	if (ndc) {
2718 		DEBUG_NET_WARN_ON_ONCE(atomic_read(&sk->sk_drops));
2719 		return numa_drop_read(ndc);
2720 	}
2721 	return atomic_read(&sk->sk_drops);
2722 }
2723 
2724 static inline void sk_drops_reset(struct sock *sk)
2725 {
2726 	struct numa_drop_counters *ndc = sk->sk_drop_counters;
2727 
2728 	if (ndc)
2729 		numa_drop_reset(ndc);
2730 	atomic_set(&sk->sk_drops, 0);
2731 }
2732 
2733 static inline void
2734 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2735 {
2736 	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2737 						sk_drops_read(sk) : 0;
2738 }
2739 
2740 static inline void sk_drops_skbadd(struct sock *sk, const struct sk_buff *skb)
2741 {
2742 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2743 
2744 	sk_drops_add(sk, segs);
2745 }
2746 
2747 static inline ktime_t sock_read_timestamp(struct sock *sk)
2748 {
2749 #if BITS_PER_LONG==32
2750 	unsigned int seq;
2751 	ktime_t kt;
2752 
2753 	do {
2754 		seq = read_seqbegin(&sk->sk_stamp_seq);
2755 		kt = sk->sk_stamp;
2756 	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2757 
2758 	return kt;
2759 #else
2760 	return READ_ONCE(sk->sk_stamp);
2761 #endif
2762 }
2763 
2764 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2765 {
2766 #if BITS_PER_LONG==32
2767 	write_seqlock(&sk->sk_stamp_seq);
2768 	sk->sk_stamp = kt;
2769 	write_sequnlock(&sk->sk_stamp_seq);
2770 #else
2771 	WRITE_ONCE(sk->sk_stamp, kt);
2772 #endif
2773 }
2774 
2775 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2776 			   struct sk_buff *skb);
2777 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2778 			     struct sk_buff *skb);
2779 
2780 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk);
2781 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk,
2782 			 struct timespec64 *ts);
2783 
2784 static inline void
2785 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2786 {
2787 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2788 	u32 tsflags = READ_ONCE(sk->sk_tsflags);
2789 	ktime_t kt = skb->tstamp;
2790 	/*
2791 	 * generate control messages if
2792 	 * - receive time stamping in software requested
2793 	 * - software time stamp available and wanted
2794 	 * - hardware time stamps available and wanted
2795 	 */
2796 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2797 	    (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2798 	    (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2799 	    (hwtstamps->hwtstamp &&
2800 	     (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2801 		__sock_recv_timestamp(msg, sk, skb);
2802 	else
2803 		sock_write_timestamp(sk, kt);
2804 
2805 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2806 		__sock_recv_wifi_status(msg, sk, skb);
2807 }
2808 
2809 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2810 		       struct sk_buff *skb);
2811 
2812 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2813 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2814 				   struct sk_buff *skb)
2815 {
2816 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2817 			   (1UL << SOCK_RCVTSTAMP)			| \
2818 			   (1UL << SOCK_RCVMARK)			| \
2819 			   (1UL << SOCK_RCVPRIORITY)			| \
2820 			   (1UL << SOCK_TIMESTAMPING_ANY))
2821 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2822 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2823 
2824 	if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS)
2825 		__sock_recv_cmsgs(msg, sk, skb);
2826 	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2827 		sock_write_timestamp(sk, skb->tstamp);
2828 	else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2829 		sock_write_timestamp(sk, 0);
2830 }
2831 
2832 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags);
2833 
2834 /**
2835  * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2836  * @sk:		socket sending this packet
2837  * @sockc:	pointer to socket cmsg cookie to get timestamping info
2838  * @tx_flags:	completed with instructions for time stamping
2839  * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2840  *
2841  * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2842  */
2843 static inline void _sock_tx_timestamp(struct sock *sk,
2844 				      const struct sockcm_cookie *sockc,
2845 				      __u8 *tx_flags, __u32 *tskey)
2846 {
2847 	__u32 tsflags = sockc->tsflags;
2848 
2849 	if (unlikely(tsflags)) {
2850 		__sock_tx_timestamp(tsflags, tx_flags);
2851 		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2852 		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) {
2853 			if (tsflags & SOCKCM_FLAG_TS_OPT_ID)
2854 				*tskey = sockc->ts_opt_id;
2855 			else
2856 				*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2857 		}
2858 	}
2859 }
2860 
2861 static inline void sock_tx_timestamp(struct sock *sk,
2862 				     const struct sockcm_cookie *sockc,
2863 				     __u8 *tx_flags)
2864 {
2865 	_sock_tx_timestamp(sk, sockc, tx_flags, NULL);
2866 }
2867 
2868 static inline void skb_setup_tx_timestamp(struct sk_buff *skb,
2869 					  const struct sockcm_cookie *sockc)
2870 {
2871 	_sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags,
2872 			   &skb_shinfo(skb)->tskey);
2873 }
2874 
2875 static inline bool sk_is_inet(const struct sock *sk)
2876 {
2877 	int family = READ_ONCE(sk->sk_family);
2878 
2879 	return family == AF_INET || family == AF_INET6;
2880 }
2881 
2882 static inline bool sk_is_tcp(const struct sock *sk)
2883 {
2884 	return sk_is_inet(sk) &&
2885 	       sk->sk_type == SOCK_STREAM &&
2886 	       sk->sk_protocol == IPPROTO_TCP;
2887 }
2888 
2889 static inline bool sk_is_udp(const struct sock *sk)
2890 {
2891 	return sk_is_inet(sk) &&
2892 	       sk->sk_type == SOCK_DGRAM &&
2893 	       sk->sk_protocol == IPPROTO_UDP;
2894 }
2895 
2896 static inline bool sk_is_unix(const struct sock *sk)
2897 {
2898 	return sk->sk_family == AF_UNIX;
2899 }
2900 
2901 static inline bool sk_is_stream_unix(const struct sock *sk)
2902 {
2903 	return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM;
2904 }
2905 
2906 static inline bool sk_is_vsock(const struct sock *sk)
2907 {
2908 	return sk->sk_family == AF_VSOCK;
2909 }
2910 
2911 static inline bool sk_may_scm_recv(const struct sock *sk)
2912 {
2913 	return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) ||
2914 		sk->sk_family == AF_NETLINK ||
2915 		(IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH);
2916 }
2917 
2918 /**
2919  * sk_eat_skb - Release a skb if it is no longer needed
2920  * @sk: socket to eat this skb from
2921  * @skb: socket buffer to eat
2922  *
2923  * This routine must be called with interrupts disabled or with the socket
2924  * locked so that the sk_buff queue operation is ok.
2925 */
2926 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2927 {
2928 	__skb_unlink(skb, &sk->sk_receive_queue);
2929 	__kfree_skb(skb);
2930 }
2931 
2932 static inline bool
2933 skb_sk_is_prefetched(struct sk_buff *skb)
2934 {
2935 #ifdef CONFIG_INET
2936 	return skb->destructor == sock_pfree;
2937 #else
2938 	return false;
2939 #endif /* CONFIG_INET */
2940 }
2941 
2942 /* This helper checks if a socket is a full socket,
2943  * ie _not_ a timewait or request socket.
2944  */
2945 static inline bool sk_fullsock(const struct sock *sk)
2946 {
2947 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2948 }
2949 
2950 static inline bool
2951 sk_is_refcounted(struct sock *sk)
2952 {
2953 	/* Only full sockets have sk->sk_flags. */
2954 	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2955 }
2956 
2957 static inline bool
2958 sk_requests_wifi_status(struct sock *sk)
2959 {
2960 	return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS);
2961 }
2962 
2963 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2964  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2965  */
2966 static inline bool sk_listener(const struct sock *sk)
2967 {
2968 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2969 }
2970 
2971 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT
2972  * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE)
2973  * TCP RST and ACK can be attached to TIME_WAIT.
2974  */
2975 static inline bool sk_listener_or_tw(const struct sock *sk)
2976 {
2977 	return (1 << READ_ONCE(sk->sk_state)) &
2978 	       (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT);
2979 }
2980 
2981 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2982 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2983 		       int type);
2984 
2985 bool sk_ns_capable(const struct sock *sk,
2986 		   struct user_namespace *user_ns, int cap);
2987 bool sk_capable(const struct sock *sk, int cap);
2988 bool sk_net_capable(const struct sock *sk, int cap);
2989 
2990 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2991 
2992 /* Take into consideration the size of the struct sk_buff overhead in the
2993  * determination of these values, since that is non-constant across
2994  * platforms.  This makes socket queueing behavior and performance
2995  * not depend upon such differences.
2996  */
2997 #define _SK_MEM_PACKETS		256
2998 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2999 #define SK_WMEM_DEFAULT		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3000 #define SK_RMEM_DEFAULT		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3001 
3002 extern __u32 sysctl_wmem_max;
3003 extern __u32 sysctl_rmem_max;
3004 
3005 extern __u32 sysctl_wmem_default;
3006 extern __u32 sysctl_rmem_default;
3007 
3008 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
3009 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3010 
3011 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
3012 {
3013 	/* Does this proto have per netns sysctl_wmem ? */
3014 	if (proto->sysctl_wmem_offset)
3015 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
3016 
3017 	return READ_ONCE(*proto->sysctl_wmem);
3018 }
3019 
3020 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
3021 {
3022 	/* Does this proto have per netns sysctl_rmem ? */
3023 	if (proto->sysctl_rmem_offset)
3024 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
3025 
3026 	return READ_ONCE(*proto->sysctl_rmem);
3027 }
3028 
3029 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
3030  * Some wifi drivers need to tweak it to get more chunks.
3031  * They can use this helper from their ndo_start_xmit()
3032  */
3033 static inline void sk_pacing_shift_update(struct sock *sk, int val)
3034 {
3035 	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
3036 		return;
3037 	WRITE_ONCE(sk->sk_pacing_shift, val);
3038 }
3039 
3040 /* if a socket is bound to a device, check that the given device
3041  * index is either the same or that the socket is bound to an L3
3042  * master device and the given device index is also enslaved to
3043  * that L3 master
3044  */
3045 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
3046 {
3047 	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
3048 	int mdif;
3049 
3050 	if (!bound_dev_if || bound_dev_if == dif)
3051 		return true;
3052 
3053 	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
3054 	if (mdif && mdif == bound_dev_if)
3055 		return true;
3056 
3057 	return false;
3058 }
3059 
3060 void sock_def_readable(struct sock *sk);
3061 
3062 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
3063 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
3064 int sock_set_timestamping(struct sock *sk, int optname,
3065 			  struct so_timestamping timestamping);
3066 
3067 #if defined(CONFIG_CGROUP_BPF)
3068 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op);
3069 #else
3070 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
3071 {
3072 }
3073 #endif
3074 void sock_no_linger(struct sock *sk);
3075 void sock_set_keepalive(struct sock *sk);
3076 void sock_set_priority(struct sock *sk, u32 priority);
3077 void sock_set_rcvbuf(struct sock *sk, int val);
3078 void sock_set_mark(struct sock *sk, u32 val);
3079 void sock_set_reuseaddr(struct sock *sk);
3080 void sock_set_reuseport(struct sock *sk);
3081 void sock_set_sndtimeo(struct sock *sk, s64 secs);
3082 
3083 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
3084 
3085 int sock_get_timeout(long timeo, void *optval, bool old_timeval);
3086 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
3087 			   sockptr_t optval, int optlen, bool old_timeval);
3088 
3089 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
3090 		     void __user *arg, void *karg, size_t size);
3091 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
3092 static inline bool sk_is_readable(struct sock *sk)
3093 {
3094 	const struct proto *prot = READ_ONCE(sk->sk_prot);
3095 
3096 	if (prot->sock_is_readable)
3097 		return prot->sock_is_readable(sk);
3098 
3099 	return false;
3100 }
3101 #endif	/* _SOCK_H */
3102