xref: /linux/include/net/sock.h (revision fc3a2810412c163b5df1b377d332e048860f45db)
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 struct socket_drop_counters {
106 	atomic_t	drops0 ____cacheline_aligned_in_smp;
107 	atomic_t	drops1 ____cacheline_aligned_in_smp;
108 };
109 
110 /**
111  *	struct sock_common - minimal network layer representation of sockets
112  *	@skc_daddr: Foreign IPv4 addr
113  *	@skc_rcv_saddr: Bound local IPv4 addr
114  *	@skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
115  *	@skc_hash: hash value used with various protocol lookup tables
116  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
117  *	@skc_dport: placeholder for inet_dport/tw_dport
118  *	@skc_num: placeholder for inet_num/tw_num
119  *	@skc_portpair: __u32 union of @skc_dport & @skc_num
120  *	@skc_family: network address family
121  *	@skc_state: Connection state
122  *	@skc_reuse: %SO_REUSEADDR setting
123  *	@skc_reuseport: %SO_REUSEPORT setting
124  *	@skc_ipv6only: socket is IPV6 only
125  *	@skc_net_refcnt: socket is using net ref counting
126  *	@skc_bound_dev_if: bound device index if != 0
127  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
128  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
129  *	@skc_prot: protocol handlers inside a network family
130  *	@skc_net: reference to the network namespace of this socket
131  *	@skc_v6_daddr: IPV6 destination address
132  *	@skc_v6_rcv_saddr: IPV6 source address
133  *	@skc_cookie: socket's cookie value
134  *	@skc_node: main hash linkage for various protocol lookup tables
135  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
136  *	@skc_tx_queue_mapping: tx queue number for this connection
137  *	@skc_rx_queue_mapping: rx queue number for this connection
138  *	@skc_flags: place holder for sk_flags
139  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
140  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
141  *	@skc_listener: connection request listener socket (aka rsk_listener)
142  *		[union with @skc_flags]
143  *	@skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
144  *		[union with @skc_flags]
145  *	@skc_incoming_cpu: record/match cpu processing incoming packets
146  *	@skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
147  *		[union with @skc_incoming_cpu]
148  *	@skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
149  *		[union with @skc_incoming_cpu]
150  *	@skc_refcnt: reference count
151  *
152  *	This is the minimal network layer representation of sockets, the header
153  *	for struct sock and struct inet_timewait_sock.
154  */
155 struct sock_common {
156 	union {
157 		__addrpair	skc_addrpair;
158 		struct {
159 			__be32	skc_daddr;
160 			__be32	skc_rcv_saddr;
161 		};
162 	};
163 	union  {
164 		unsigned int	skc_hash;
165 		__u16		skc_u16hashes[2];
166 	};
167 	/* skc_dport && skc_num must be grouped as well */
168 	union {
169 		__portpair	skc_portpair;
170 		struct {
171 			__be16	skc_dport;
172 			__u16	skc_num;
173 		};
174 	};
175 
176 	unsigned short		skc_family;
177 	volatile unsigned char	skc_state;
178 	unsigned char		skc_reuse:4;
179 	unsigned char		skc_reuseport:1;
180 	unsigned char		skc_ipv6only:1;
181 	unsigned char		skc_net_refcnt:1;
182 	int			skc_bound_dev_if;
183 	union {
184 		struct hlist_node	skc_bind_node;
185 		struct hlist_node	skc_portaddr_node;
186 	};
187 	struct proto		*skc_prot;
188 	possible_net_t		skc_net;
189 
190 #if IS_ENABLED(CONFIG_IPV6)
191 	struct in6_addr		skc_v6_daddr;
192 	struct in6_addr		skc_v6_rcv_saddr;
193 #endif
194 
195 	atomic64_t		skc_cookie;
196 
197 	/* following fields are padding to force
198 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
199 	 * assuming IPV6 is enabled. We use this padding differently
200 	 * for different kind of 'sockets'
201 	 */
202 	union {
203 		unsigned long	skc_flags;
204 		struct sock	*skc_listener; /* request_sock */
205 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
206 	};
207 	/*
208 	 * fields between dontcopy_begin/dontcopy_end
209 	 * are not copied in sock_copy()
210 	 */
211 	/* private: */
212 	int			skc_dontcopy_begin[0];
213 	/* public: */
214 	union {
215 		struct hlist_node	skc_node;
216 		struct hlist_nulls_node skc_nulls_node;
217 	};
218 	unsigned short		skc_tx_queue_mapping;
219 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
220 	unsigned short		skc_rx_queue_mapping;
221 #endif
222 	union {
223 		int		skc_incoming_cpu;
224 		u32		skc_rcv_wnd;
225 		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
226 	};
227 
228 	refcount_t		skc_refcnt;
229 	/* private: */
230 	int                     skc_dontcopy_end[0];
231 	union {
232 		u32		skc_rxhash;
233 		u32		skc_window_clamp;
234 		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
235 	};
236 	/* public: */
237 };
238 
239 struct bpf_local_storage;
240 struct sk_filter;
241 
242 /**
243   *	struct sock - network layer representation of sockets
244   *	@__sk_common: shared layout with inet_timewait_sock
245   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
246   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
247   *	@sk_lock:	synchronizer
248   *	@sk_kern_sock: True if sock is using kernel lock classes
249   *	@sk_rcvbuf: size of receive buffer in bytes
250   *	@sk_wq: sock wait queue and async head
251   *	@sk_rx_dst: receive input route used by early demux
252   *	@sk_rx_dst_ifindex: ifindex for @sk_rx_dst
253   *	@sk_rx_dst_cookie: cookie for @sk_rx_dst
254   *	@sk_dst_cache: destination cache
255   *	@sk_dst_pending_confirm: need to confirm neighbour
256   *	@sk_policy: flow policy
257   *	@sk_receive_queue: incoming packets
258   *	@sk_wmem_alloc: transmit queue bytes committed
259   *	@sk_tsq_flags: TCP Small Queues flags
260   *	@sk_write_queue: Packet sending queue
261   *	@sk_omem_alloc: "o" is "option" or "other"
262   *	@sk_wmem_queued: persistent queue size
263   *	@sk_forward_alloc: space allocated forward
264   *	@sk_reserved_mem: space reserved and non-reclaimable for the socket
265   *	@sk_napi_id: id of the last napi context to receive data for sk
266   *	@sk_ll_usec: usecs to busypoll when there is no data
267   *	@sk_allocation: allocation mode
268   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
269   *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
270   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
271   *	@sk_sndbuf: size of send buffer in bytes
272   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
273   *	@sk_no_check_rx: allow zero checksum in RX packets
274   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
275   *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
276   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
277   *	@sk_gso_max_size: Maximum GSO segment size to build
278   *	@sk_gso_max_segs: Maximum number of GSO segments
279   *	@sk_pacing_shift: scaling factor for TCP Small Queues
280   *	@sk_lingertime: %SO_LINGER l_linger setting
281   *	@sk_backlog: always used with the per-socket spinlock held
282   *	@sk_callback_lock: used with the callbacks in the end of this struct
283   *	@sk_error_queue: rarely used
284   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
285   *			  IPV6_ADDRFORM for instance)
286   *	@sk_err: last error
287   *	@sk_err_soft: errors that don't cause failure but are the cause of a
288   *		      persistent failure not just 'timed out'
289   *	@sk_drops: raw/udp drops counter
290   *	@sk_drop_counters: optional pointer to socket_drop_counters
291   *	@sk_ack_backlog: current listen backlog
292   *	@sk_max_ack_backlog: listen backlog set in listen()
293   *	@sk_uid: user id of owner
294   *	@sk_ino: inode number (zero if orphaned)
295   *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
296   *	@sk_busy_poll_budget: napi processing budget when busypolling
297   *	@sk_priority: %SO_PRIORITY setting
298   *	@sk_type: socket type (%SOCK_STREAM, etc)
299   *	@sk_protocol: which protocol this socket belongs in this network family
300   *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
301   *	@sk_peer_pid: &struct pid for this socket's peer
302   *	@sk_peer_cred: %SO_PEERCRED setting
303   *	@sk_rcvlowat: %SO_RCVLOWAT setting
304   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
305   *	@sk_sndtimeo: %SO_SNDTIMEO setting
306   *	@sk_txhash: computed flow hash for use on transmit
307   *	@sk_txrehash: enable TX hash rethink
308   *	@sk_filter: socket filtering instructions
309   *	@sk_timer: sock cleanup timer
310   *	@sk_stamp: time stamp of last packet received
311   *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
312   *	@sk_tsflags: SO_TIMESTAMPING flags
313   *	@sk_bpf_cb_flags: used in bpf_setsockopt()
314   *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
315   *			   Sockets that can be used under memory reclaim should
316   *			   set this to false.
317   *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
318   *	              for timestamping
319   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
320   *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
321   *	@sk_socket: Identd and reporting IO signals
322   *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
323   *	@sk_frag: cached page frag
324   *	@sk_peek_off: current peek_offset value
325   *	@sk_send_head: front of stuff to transmit
326   *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
327   *	@sk_security: used by security modules
328   *	@sk_mark: generic packet mark
329   *	@sk_cgrp_data: cgroup data for this cgroup
330   *	@sk_memcg: this socket's memory cgroup association
331   *	@sk_write_pending: a write to stream socket waits to start
332   *	@sk_disconnects: number of disconnect operations performed on this sock
333   *	@sk_state_change: callback to indicate change in the state of the sock
334   *	@sk_data_ready: callback to indicate there is data to be processed
335   *	@sk_write_space: callback to indicate there is bf sending space available
336   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
337   *	@sk_backlog_rcv: callback to process the backlog
338   *	@sk_validate_xmit_skb: ptr to an optional validate function
339   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
340   *	@sk_reuseport_cb: reuseport group container
341   *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
342   *	@sk_rcu: used during RCU grace period
343   *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
344   *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
345   *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
346   *	@sk_txtime_unused: unused txtime flags
347   *	@sk_scm_recv_flags: all flags used by scm_recv()
348   *	@sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS
349   *	@sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY
350   *	@sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD
351   *	@sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS
352   *	@sk_scm_unused: unused flags for scm_recv()
353   *	@ns_tracker: tracker for netns reference
354   *	@sk_user_frags: xarray of pages the user is holding a reference on.
355   *	@sk_owner: reference to the real owner of the socket that calls
356   *		   sock_lock_init_class_and_name().
357   */
358 struct sock {
359 	/*
360 	 * Now struct inet_timewait_sock also uses sock_common, so please just
361 	 * don't add nothing before this first member (__sk_common) --acme
362 	 */
363 	struct sock_common	__sk_common;
364 #define sk_node			__sk_common.skc_node
365 #define sk_nulls_node		__sk_common.skc_nulls_node
366 #define sk_refcnt		__sk_common.skc_refcnt
367 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
368 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
369 #define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
370 #endif
371 
372 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
373 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
374 #define sk_hash			__sk_common.skc_hash
375 #define sk_portpair		__sk_common.skc_portpair
376 #define sk_num			__sk_common.skc_num
377 #define sk_dport		__sk_common.skc_dport
378 #define sk_addrpair		__sk_common.skc_addrpair
379 #define sk_daddr		__sk_common.skc_daddr
380 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
381 #define sk_family		__sk_common.skc_family
382 #define sk_state		__sk_common.skc_state
383 #define sk_reuse		__sk_common.skc_reuse
384 #define sk_reuseport		__sk_common.skc_reuseport
385 #define sk_ipv6only		__sk_common.skc_ipv6only
386 #define sk_net_refcnt		__sk_common.skc_net_refcnt
387 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
388 #define sk_bind_node		__sk_common.skc_bind_node
389 #define sk_prot			__sk_common.skc_prot
390 #define sk_net			__sk_common.skc_net
391 #define sk_v6_daddr		__sk_common.skc_v6_daddr
392 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
393 #define sk_cookie		__sk_common.skc_cookie
394 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
395 #define sk_flags		__sk_common.skc_flags
396 #define sk_rxhash		__sk_common.skc_rxhash
397 
398 	__cacheline_group_begin(sock_write_rx);
399 
400 	atomic_t		sk_drops;
401 	__s32			sk_peek_off;
402 	struct sk_buff_head	sk_error_queue;
403 	struct sk_buff_head	sk_receive_queue;
404 	/*
405 	 * The backlog queue is special, it is always used with
406 	 * the per-socket spinlock held and requires low latency
407 	 * access. Therefore we special case it's implementation.
408 	 * Note : rmem_alloc is in this structure to fill a hole
409 	 * on 64bit arches, not because its logically part of
410 	 * backlog.
411 	 */
412 	struct {
413 		atomic_t	rmem_alloc;
414 		int		len;
415 		struct sk_buff	*head;
416 		struct sk_buff	*tail;
417 	} sk_backlog;
418 #define sk_rmem_alloc sk_backlog.rmem_alloc
419 
420 	__cacheline_group_end(sock_write_rx);
421 
422 	__cacheline_group_begin(sock_read_rx);
423 	/* early demux fields */
424 	struct dst_entry __rcu	*sk_rx_dst;
425 	int			sk_rx_dst_ifindex;
426 	u32			sk_rx_dst_cookie;
427 
428 #ifdef CONFIG_NET_RX_BUSY_POLL
429 	unsigned int		sk_ll_usec;
430 	unsigned int		sk_napi_id;
431 	u16			sk_busy_poll_budget;
432 	u8			sk_prefer_busy_poll;
433 #endif
434 	u8			sk_userlocks;
435 	int			sk_rcvbuf;
436 
437 	struct sk_filter __rcu	*sk_filter;
438 	union {
439 		struct socket_wq __rcu	*sk_wq;
440 		/* private: */
441 		struct socket_wq	*sk_wq_raw;
442 		/* public: */
443 	};
444 
445 	void			(*sk_data_ready)(struct sock *sk);
446 	long			sk_rcvtimeo;
447 	int			sk_rcvlowat;
448 	__cacheline_group_end(sock_read_rx);
449 
450 	__cacheline_group_begin(sock_read_rxtx);
451 	int			sk_err;
452 	struct socket		*sk_socket;
453 #ifdef CONFIG_MEMCG
454 	struct mem_cgroup	*sk_memcg;
455 #endif
456 #ifdef CONFIG_XFRM
457 	struct xfrm_policy __rcu *sk_policy[2];
458 #endif
459 	struct socket_drop_counters *sk_drop_counters;
460 	__cacheline_group_end(sock_read_rxtx);
461 
462 	__cacheline_group_begin(sock_write_rxtx);
463 	socket_lock_t		sk_lock;
464 	u32			sk_reserved_mem;
465 	int			sk_forward_alloc;
466 	u32			sk_tsflags;
467 	__cacheline_group_end(sock_write_rxtx);
468 
469 	__cacheline_group_begin(sock_write_tx);
470 	int			sk_write_pending;
471 	atomic_t		sk_omem_alloc;
472 	int			sk_sndbuf;
473 
474 	int			sk_wmem_queued;
475 	refcount_t		sk_wmem_alloc;
476 	unsigned long		sk_tsq_flags;
477 	union {
478 		struct sk_buff	*sk_send_head;
479 		struct rb_root	tcp_rtx_queue;
480 	};
481 	struct sk_buff_head	sk_write_queue;
482 	u32			sk_dst_pending_confirm;
483 	u32			sk_pacing_status; /* see enum sk_pacing */
484 	struct page_frag	sk_frag;
485 	struct timer_list	sk_timer;
486 
487 	unsigned long		sk_pacing_rate; /* bytes per second */
488 	atomic_t		sk_zckey;
489 	atomic_t		sk_tskey;
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 	struct dst_entry __rcu	*sk_dst_cache;
498 	netdev_features_t	sk_route_caps;
499 #ifdef CONFIG_SOCK_VALIDATE_XMIT
500 	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
501 							struct net_device *dev,
502 							struct sk_buff *skb);
503 #endif
504 	u16			sk_gso_type;
505 	u16			sk_gso_max_segs;
506 	unsigned int		sk_gso_max_size;
507 	gfp_t			sk_allocation;
508 	u32			sk_txhash;
509 	u8			sk_pacing_shift;
510 	bool			sk_use_task_frag;
511 	__cacheline_group_end(sock_read_tx);
512 
513 	/*
514 	 * Because of non atomicity rules, all
515 	 * changes are protected by socket lock.
516 	 */
517 	u8			sk_gso_disabled : 1,
518 				sk_kern_sock : 1,
519 				sk_no_check_tx : 1,
520 				sk_no_check_rx : 1;
521 	u8			sk_shutdown;
522 	u16			sk_type;
523 	u16			sk_protocol;
524 	unsigned long	        sk_lingertime;
525 	struct proto		*sk_prot_creator;
526 	rwlock_t		sk_callback_lock;
527 	int			sk_err_soft;
528 	u32			sk_ack_backlog;
529 	u32			sk_max_ack_backlog;
530 	kuid_t			sk_uid;
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 rcu_head		sk_rcu;
577 	netns_tracker		ns_tracker;
578 	struct xarray		sk_user_frags;
579 
580 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
581 	struct module		*sk_owner;
582 #endif
583 };
584 
585 struct sock_bh_locked {
586 	struct sock *sock;
587 	local_lock_t bh_lock;
588 };
589 
590 enum sk_pacing {
591 	SK_PACING_NONE		= 0,
592 	SK_PACING_NEEDED	= 1,
593 	SK_PACING_FQ		= 2,
594 };
595 
596 /* flag bits in sk_user_data
597  *
598  * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
599  *   not be suitable for copying when cloning the socket. For instance,
600  *   it can point to a reference counted object. sk_user_data bottom
601  *   bit is set if pointer must not be copied.
602  *
603  * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
604  *   managed/owned by a BPF reuseport array. This bit should be set
605  *   when sk_user_data's sk is added to the bpf's reuseport_array.
606  *
607  * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
608  *   sk_user_data points to psock type. This bit should be set
609  *   when sk_user_data is assigned to a psock object.
610  */
611 #define SK_USER_DATA_NOCOPY	1UL
612 #define SK_USER_DATA_BPF	2UL
613 #define SK_USER_DATA_PSOCK	4UL
614 #define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
615 				  SK_USER_DATA_PSOCK)
616 
617 /**
618  * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
619  * @sk: socket
620  */
621 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
622 {
623 	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
624 }
625 
626 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
627 
628 /**
629  * __locked_read_sk_user_data_with_flags - return the pointer
630  * only if argument flags all has been set in sk_user_data. Otherwise
631  * return NULL
632  *
633  * @sk: socket
634  * @flags: flag bits
635  *
636  * The caller must be holding sk->sk_callback_lock.
637  */
638 static inline void *
639 __locked_read_sk_user_data_with_flags(const struct sock *sk,
640 				      uintptr_t flags)
641 {
642 	uintptr_t sk_user_data =
643 		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
644 						 lockdep_is_held(&sk->sk_callback_lock));
645 
646 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
647 
648 	if ((sk_user_data & flags) == flags)
649 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
650 	return NULL;
651 }
652 
653 /**
654  * __rcu_dereference_sk_user_data_with_flags - return the pointer
655  * only if argument flags all has been set in sk_user_data. Otherwise
656  * return NULL
657  *
658  * @sk: socket
659  * @flags: flag bits
660  */
661 static inline void *
662 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
663 					  uintptr_t flags)
664 {
665 	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
666 
667 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
668 
669 	if ((sk_user_data & flags) == flags)
670 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
671 	return NULL;
672 }
673 
674 #define rcu_dereference_sk_user_data(sk)				\
675 	__rcu_dereference_sk_user_data_with_flags(sk, 0)
676 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
677 ({									\
678 	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
679 		  __tmp2 = (uintptr_t)(flags);				\
680 	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
681 	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
682 	rcu_assign_pointer(__sk_user_data((sk)),			\
683 			   __tmp1 | __tmp2);				\
684 })
685 #define rcu_assign_sk_user_data(sk, ptr)				\
686 	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
687 
688 static inline
689 struct net *sock_net(const struct sock *sk)
690 {
691 	return read_pnet(&sk->sk_net);
692 }
693 
694 static inline
695 void sock_net_set(struct sock *sk, struct net *net)
696 {
697 	write_pnet(&sk->sk_net, net);
698 }
699 
700 /*
701  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
702  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
703  * on a socket means that the socket will reuse everybody else's port
704  * without looking at the other's sk_reuse value.
705  */
706 
707 #define SK_NO_REUSE	0
708 #define SK_CAN_REUSE	1
709 #define SK_FORCE_REUSE	2
710 
711 int sk_set_peek_off(struct sock *sk, int val);
712 
713 static inline int sk_peek_offset(const struct sock *sk, int flags)
714 {
715 	if (unlikely(flags & MSG_PEEK)) {
716 		return READ_ONCE(sk->sk_peek_off);
717 	}
718 
719 	return 0;
720 }
721 
722 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
723 {
724 	s32 off = READ_ONCE(sk->sk_peek_off);
725 
726 	if (unlikely(off >= 0)) {
727 		off = max_t(s32, off - val, 0);
728 		WRITE_ONCE(sk->sk_peek_off, off);
729 	}
730 }
731 
732 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
733 {
734 	sk_peek_offset_bwd(sk, -val);
735 }
736 
737 /*
738  * Hashed lists helper routines
739  */
740 static inline struct sock *sk_entry(const struct hlist_node *node)
741 {
742 	return hlist_entry(node, struct sock, sk_node);
743 }
744 
745 static inline struct sock *__sk_head(const struct hlist_head *head)
746 {
747 	return hlist_entry(head->first, struct sock, sk_node);
748 }
749 
750 static inline struct sock *sk_head(const struct hlist_head *head)
751 {
752 	return hlist_empty(head) ? NULL : __sk_head(head);
753 }
754 
755 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
756 {
757 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
758 }
759 
760 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
761 {
762 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
763 }
764 
765 static inline struct sock *sk_next(const struct sock *sk)
766 {
767 	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
768 }
769 
770 static inline struct sock *sk_nulls_next(const struct sock *sk)
771 {
772 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
773 		hlist_nulls_entry(sk->sk_nulls_node.next,
774 				  struct sock, sk_nulls_node) :
775 		NULL;
776 }
777 
778 static inline bool sk_unhashed(const struct sock *sk)
779 {
780 	return hlist_unhashed(&sk->sk_node);
781 }
782 
783 static inline bool sk_hashed(const struct sock *sk)
784 {
785 	return !sk_unhashed(sk);
786 }
787 
788 static inline void sk_node_init(struct hlist_node *node)
789 {
790 	node->pprev = NULL;
791 }
792 
793 static inline void __sk_del_node(struct sock *sk)
794 {
795 	__hlist_del(&sk->sk_node);
796 }
797 
798 /* NB: equivalent to hlist_del_init_rcu */
799 static inline bool __sk_del_node_init(struct sock *sk)
800 {
801 	if (sk_hashed(sk)) {
802 		__sk_del_node(sk);
803 		sk_node_init(&sk->sk_node);
804 		return true;
805 	}
806 	return false;
807 }
808 
809 /* Grab socket reference count. This operation is valid only
810    when sk is ALREADY grabbed f.e. it is found in hash table
811    or a list and the lookup is made under lock preventing hash table
812    modifications.
813  */
814 
815 static __always_inline void sock_hold(struct sock *sk)
816 {
817 	refcount_inc(&sk->sk_refcnt);
818 }
819 
820 /* Ungrab socket in the context, which assumes that socket refcnt
821    cannot hit zero, f.e. it is true in context of any socketcall.
822  */
823 static __always_inline void __sock_put(struct sock *sk)
824 {
825 	refcount_dec(&sk->sk_refcnt);
826 }
827 
828 static inline bool sk_del_node_init(struct sock *sk)
829 {
830 	bool rc = __sk_del_node_init(sk);
831 
832 	if (rc) {
833 		/* paranoid for a while -acme */
834 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
835 		__sock_put(sk);
836 	}
837 	return rc;
838 }
839 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
840 
841 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
842 {
843 	if (sk_hashed(sk)) {
844 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
845 		return true;
846 	}
847 	return false;
848 }
849 
850 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
851 {
852 	bool rc = __sk_nulls_del_node_init_rcu(sk);
853 
854 	if (rc) {
855 		/* paranoid for a while -acme */
856 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
857 		__sock_put(sk);
858 	}
859 	return rc;
860 }
861 
862 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
863 {
864 	hlist_add_head(&sk->sk_node, list);
865 }
866 
867 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
868 {
869 	sock_hold(sk);
870 	__sk_add_node(sk, list);
871 }
872 
873 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
874 {
875 	sock_hold(sk);
876 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
877 	    sk->sk_family == AF_INET6)
878 		hlist_add_tail_rcu(&sk->sk_node, list);
879 	else
880 		hlist_add_head_rcu(&sk->sk_node, list);
881 }
882 
883 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
884 {
885 	sock_hold(sk);
886 	hlist_add_tail_rcu(&sk->sk_node, list);
887 }
888 
889 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
890 {
891 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
892 }
893 
894 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
895 {
896 	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
897 }
898 
899 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
900 {
901 	sock_hold(sk);
902 	__sk_nulls_add_node_rcu(sk, list);
903 }
904 
905 static inline void __sk_del_bind_node(struct sock *sk)
906 {
907 	__hlist_del(&sk->sk_bind_node);
908 }
909 
910 static inline void sk_add_bind_node(struct sock *sk,
911 					struct hlist_head *list)
912 {
913 	hlist_add_head(&sk->sk_bind_node, list);
914 }
915 
916 #define sk_for_each(__sk, list) \
917 	hlist_for_each_entry(__sk, list, sk_node)
918 #define sk_for_each_rcu(__sk, list) \
919 	hlist_for_each_entry_rcu(__sk, list, sk_node)
920 #define sk_nulls_for_each(__sk, node, list) \
921 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
922 #define sk_nulls_for_each_rcu(__sk, node, list) \
923 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
924 #define sk_for_each_from(__sk) \
925 	hlist_for_each_entry_from(__sk, sk_node)
926 #define sk_nulls_for_each_from(__sk, node) \
927 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
928 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
929 #define sk_for_each_safe(__sk, tmp, list) \
930 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
931 #define sk_for_each_bound(__sk, list) \
932 	hlist_for_each_entry(__sk, list, sk_bind_node)
933 #define sk_for_each_bound_safe(__sk, tmp, list) \
934 	hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node)
935 
936 /**
937  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
938  * @tpos:	the type * to use as a loop cursor.
939  * @pos:	the &struct hlist_node to use as a loop cursor.
940  * @head:	the head for your list.
941  * @offset:	offset of hlist_node within the struct.
942  *
943  */
944 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
945 	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
946 	     pos != NULL &&						       \
947 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
948 	     pos = rcu_dereference(hlist_next_rcu(pos)))
949 
950 static inline struct user_namespace *sk_user_ns(const struct sock *sk)
951 {
952 	/* Careful only use this in a context where these parameters
953 	 * can not change and must all be valid, such as recvmsg from
954 	 * userspace.
955 	 */
956 	return sk->sk_socket->file->f_cred->user_ns;
957 }
958 
959 /* Sock flags */
960 enum sock_flags {
961 	SOCK_DEAD,
962 	SOCK_DONE,
963 	SOCK_URGINLINE,
964 	SOCK_KEEPOPEN,
965 	SOCK_LINGER,
966 	SOCK_DESTROY,
967 	SOCK_BROADCAST,
968 	SOCK_TIMESTAMP,
969 	SOCK_ZAPPED,
970 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
971 	SOCK_DBG, /* %SO_DEBUG setting */
972 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
973 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
974 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
975 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
976 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
977 	SOCK_FASYNC, /* fasync() active */
978 	SOCK_RXQ_OVFL,
979 	SOCK_ZEROCOPY, /* buffers from userspace */
980 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
981 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
982 		     * Will use last 4 bytes of packet sent from
983 		     * user-space instead.
984 		     */
985 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
986 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
987 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
988 	SOCK_TXTIME,
989 	SOCK_XDP, /* XDP is attached */
990 	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
991 	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
992 	SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */
993 	SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */
994 };
995 
996 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
997 /*
998  * The highest bit of sk_tsflags is reserved for kernel-internal
999  * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that
1000  * SOF_TIMESTAMPING* values do not reach this reserved area
1001  */
1002 #define SOCKCM_FLAG_TS_OPT_ID	BIT(31)
1003 
1004 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
1005 {
1006 	nsk->sk_flags = osk->sk_flags;
1007 }
1008 
1009 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
1010 {
1011 	__set_bit(flag, &sk->sk_flags);
1012 }
1013 
1014 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
1015 {
1016 	__clear_bit(flag, &sk->sk_flags);
1017 }
1018 
1019 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
1020 				     int valbool)
1021 {
1022 	if (valbool)
1023 		sock_set_flag(sk, bit);
1024 	else
1025 		sock_reset_flag(sk, bit);
1026 }
1027 
1028 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
1029 {
1030 	return test_bit(flag, &sk->sk_flags);
1031 }
1032 
1033 #ifdef CONFIG_NET
1034 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
1035 static inline int sk_memalloc_socks(void)
1036 {
1037 	return static_branch_unlikely(&memalloc_socks_key);
1038 }
1039 
1040 void __receive_sock(struct file *file);
1041 #else
1042 
1043 static inline int sk_memalloc_socks(void)
1044 {
1045 	return 0;
1046 }
1047 
1048 static inline void __receive_sock(struct file *file)
1049 { }
1050 #endif
1051 
1052 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1053 {
1054 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1055 }
1056 
1057 static inline void sk_acceptq_removed(struct sock *sk)
1058 {
1059 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1060 }
1061 
1062 static inline void sk_acceptq_added(struct sock *sk)
1063 {
1064 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1065 }
1066 
1067 /* Note: If you think the test should be:
1068  *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1069  * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1070  */
1071 static inline bool sk_acceptq_is_full(const struct sock *sk)
1072 {
1073 	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1074 }
1075 
1076 /*
1077  * Compute minimal free write space needed to queue new packets.
1078  */
1079 static inline int sk_stream_min_wspace(const struct sock *sk)
1080 {
1081 	return READ_ONCE(sk->sk_wmem_queued) >> 1;
1082 }
1083 
1084 static inline int sk_stream_wspace(const struct sock *sk)
1085 {
1086 	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1087 }
1088 
1089 static inline void sk_wmem_queued_add(struct sock *sk, int val)
1090 {
1091 	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1092 }
1093 
1094 static inline void sk_forward_alloc_add(struct sock *sk, int val)
1095 {
1096 	/* Paired with lockless reads of sk->sk_forward_alloc */
1097 	WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1098 }
1099 
1100 void sk_stream_write_space(struct sock *sk);
1101 
1102 /* OOB backlog add */
1103 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1104 {
1105 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
1106 	skb_dst_force(skb);
1107 
1108 	if (!sk->sk_backlog.tail)
1109 		WRITE_ONCE(sk->sk_backlog.head, skb);
1110 	else
1111 		sk->sk_backlog.tail->next = skb;
1112 
1113 	WRITE_ONCE(sk->sk_backlog.tail, skb);
1114 	skb->next = NULL;
1115 }
1116 
1117 /*
1118  * Take into account size of receive queue and backlog queue
1119  * Do not take into account this skb truesize,
1120  * to allow even a single big packet to come.
1121  */
1122 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1123 {
1124 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1125 
1126 	return qsize > limit;
1127 }
1128 
1129 /* The per-socket spinlock must be held here. */
1130 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1131 					      unsigned int limit)
1132 {
1133 	if (sk_rcvqueues_full(sk, limit))
1134 		return -ENOBUFS;
1135 
1136 	/*
1137 	 * If the skb was allocated from pfmemalloc reserves, only
1138 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1139 	 * helping free memory
1140 	 */
1141 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1142 		return -ENOMEM;
1143 
1144 	__sk_add_backlog(sk, skb);
1145 	sk->sk_backlog.len += skb->truesize;
1146 	return 0;
1147 }
1148 
1149 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1150 
1151 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1152 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1153 
1154 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1155 {
1156 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1157 		return __sk_backlog_rcv(sk, skb);
1158 
1159 	return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1160 				  tcp_v6_do_rcv,
1161 				  tcp_v4_do_rcv,
1162 				  sk, skb);
1163 }
1164 
1165 static inline void sk_incoming_cpu_update(struct sock *sk)
1166 {
1167 	int cpu = raw_smp_processor_id();
1168 
1169 	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1170 		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1171 }
1172 
1173 
1174 static inline void sock_rps_save_rxhash(struct sock *sk,
1175 					const struct sk_buff *skb)
1176 {
1177 #ifdef CONFIG_RPS
1178 	/* The following WRITE_ONCE() is paired with the READ_ONCE()
1179 	 * here, and another one in sock_rps_record_flow().
1180 	 */
1181 	if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1182 		WRITE_ONCE(sk->sk_rxhash, skb->hash);
1183 #endif
1184 }
1185 
1186 static inline void sock_rps_reset_rxhash(struct sock *sk)
1187 {
1188 #ifdef CONFIG_RPS
1189 	/* Paired with READ_ONCE() in sock_rps_record_flow() */
1190 	WRITE_ONCE(sk->sk_rxhash, 0);
1191 #endif
1192 }
1193 
1194 #define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1195 	({	int __rc, __dis = __sk->sk_disconnects;			\
1196 		release_sock(__sk);					\
1197 		__rc = __condition;					\
1198 		if (!__rc) {						\
1199 			*(__timeo) = wait_woken(__wait,			\
1200 						TASK_INTERRUPTIBLE,	\
1201 						*(__timeo));		\
1202 		}							\
1203 		sched_annotate_sleep();					\
1204 		lock_sock(__sk);					\
1205 		__rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1206 		__rc;							\
1207 	})
1208 
1209 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1210 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1211 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1212 int sk_stream_error(struct sock *sk, int flags, int err);
1213 void sk_stream_kill_queues(struct sock *sk);
1214 void sk_set_memalloc(struct sock *sk);
1215 void sk_clear_memalloc(struct sock *sk);
1216 
1217 void __sk_flush_backlog(struct sock *sk);
1218 
1219 static inline bool sk_flush_backlog(struct sock *sk)
1220 {
1221 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1222 		__sk_flush_backlog(sk);
1223 		return true;
1224 	}
1225 	return false;
1226 }
1227 
1228 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1229 
1230 struct request_sock_ops;
1231 struct timewait_sock_ops;
1232 struct inet_hashinfo;
1233 struct raw_hashinfo;
1234 struct smc_hashinfo;
1235 struct module;
1236 struct sk_psock;
1237 
1238 /*
1239  * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1240  * un-modified. Special care is taken when initializing object to zero.
1241  */
1242 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1243 {
1244 	if (offsetof(struct sock, sk_node.next) != 0)
1245 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1246 	memset(&sk->sk_node.pprev, 0,
1247 	       size - offsetof(struct sock, sk_node.pprev));
1248 }
1249 
1250 struct proto_accept_arg {
1251 	int flags;
1252 	int err;
1253 	int is_empty;
1254 	bool kern;
1255 };
1256 
1257 /* Networking protocol blocks we attach to sockets.
1258  * socket layer -> transport layer interface
1259  */
1260 struct proto {
1261 	void			(*close)(struct sock *sk,
1262 					long timeout);
1263 	int			(*pre_connect)(struct sock *sk,
1264 					struct sockaddr *uaddr,
1265 					int addr_len);
1266 	int			(*connect)(struct sock *sk,
1267 					struct sockaddr *uaddr,
1268 					int addr_len);
1269 	int			(*disconnect)(struct sock *sk, int flags);
1270 
1271 	struct sock *		(*accept)(struct sock *sk,
1272 					  struct proto_accept_arg *arg);
1273 
1274 	int			(*ioctl)(struct sock *sk, int cmd,
1275 					 int *karg);
1276 	int			(*init)(struct sock *sk);
1277 	void			(*destroy)(struct sock *sk);
1278 	void			(*shutdown)(struct sock *sk, int how);
1279 	int			(*setsockopt)(struct sock *sk, int level,
1280 					int optname, sockptr_t optval,
1281 					unsigned int optlen);
1282 	int			(*getsockopt)(struct sock *sk, int level,
1283 					int optname, char __user *optval,
1284 					int __user *option);
1285 	void			(*keepalive)(struct sock *sk, int valbool);
1286 #ifdef CONFIG_COMPAT
1287 	int			(*compat_ioctl)(struct sock *sk,
1288 					unsigned int cmd, unsigned long arg);
1289 #endif
1290 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1291 					   size_t len);
1292 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1293 					   size_t len, int flags, int *addr_len);
1294 	void			(*splice_eof)(struct socket *sock);
1295 	int			(*bind)(struct sock *sk,
1296 					struct sockaddr *addr, int addr_len);
1297 	int			(*bind_add)(struct sock *sk,
1298 					struct sockaddr *addr, int addr_len);
1299 
1300 	int			(*backlog_rcv) (struct sock *sk,
1301 						struct sk_buff *skb);
1302 	bool			(*bpf_bypass_getsockopt)(int level,
1303 							 int optname);
1304 
1305 	void		(*release_cb)(struct sock *sk);
1306 
1307 	/* Keeping track of sk's, looking them up, and port selection methods. */
1308 	int			(*hash)(struct sock *sk);
1309 	void			(*unhash)(struct sock *sk);
1310 	void			(*rehash)(struct sock *sk);
1311 	int			(*get_port)(struct sock *sk, unsigned short snum);
1312 	void			(*put_port)(struct sock *sk);
1313 #ifdef CONFIG_BPF_SYSCALL
1314 	int			(*psock_update_sk_prot)(struct sock *sk,
1315 							struct sk_psock *psock,
1316 							bool restore);
1317 #endif
1318 
1319 	/* Keeping track of sockets in use */
1320 #ifdef CONFIG_PROC_FS
1321 	unsigned int		inuse_idx;
1322 #endif
1323 
1324 	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1325 	bool			(*sock_is_readable)(struct sock *sk);
1326 	/* Memory pressure */
1327 	void			(*enter_memory_pressure)(struct sock *sk);
1328 	void			(*leave_memory_pressure)(struct sock *sk);
1329 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1330 	int  __percpu		*per_cpu_fw_alloc;
1331 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1332 
1333 	/*
1334 	 * Pressure flag: try to collapse.
1335 	 * Technical note: it is used by multiple contexts non atomically.
1336 	 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1337 	 * All the __sk_mem_schedule() is of this nature: accounting
1338 	 * is strict, actions are advisory and have some latency.
1339 	 */
1340 	unsigned long		*memory_pressure;
1341 	long			*sysctl_mem;
1342 
1343 	int			*sysctl_wmem;
1344 	int			*sysctl_rmem;
1345 	u32			sysctl_wmem_offset;
1346 	u32			sysctl_rmem_offset;
1347 
1348 	int			max_header;
1349 	bool			no_autobind;
1350 
1351 	struct kmem_cache	*slab;
1352 	unsigned int		obj_size;
1353 	unsigned int		ipv6_pinfo_offset;
1354 	slab_flags_t		slab_flags;
1355 	unsigned int		useroffset;	/* Usercopy region offset */
1356 	unsigned int		usersize;	/* Usercopy region size */
1357 
1358 	struct request_sock_ops	*rsk_prot;
1359 	struct timewait_sock_ops *twsk_prot;
1360 
1361 	union {
1362 		struct inet_hashinfo	*hashinfo;
1363 		struct udp_table	*udp_table;
1364 		struct raw_hashinfo	*raw_hash;
1365 		struct smc_hashinfo	*smc_hash;
1366 	} h;
1367 
1368 	struct module		*owner;
1369 
1370 	char			name[32];
1371 
1372 	struct list_head	node;
1373 	int			(*diag_destroy)(struct sock *sk, int err);
1374 } __randomize_layout;
1375 
1376 int proto_register(struct proto *prot, int alloc_slab);
1377 void proto_unregister(struct proto *prot);
1378 int sock_load_diag_module(int family, int protocol);
1379 
1380 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1381 
1382 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1383 {
1384 	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1385 		return false;
1386 
1387 	return sk->sk_prot->stream_memory_free ?
1388 		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1389 				     tcp_stream_memory_free, sk, wake) : true;
1390 }
1391 
1392 static inline bool sk_stream_memory_free(const struct sock *sk)
1393 {
1394 	return __sk_stream_memory_free(sk, 0);
1395 }
1396 
1397 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1398 {
1399 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1400 	       __sk_stream_memory_free(sk, wake);
1401 }
1402 
1403 static inline bool sk_stream_is_writeable(const struct sock *sk)
1404 {
1405 	return __sk_stream_is_writeable(sk, 0);
1406 }
1407 
1408 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1409 					    struct cgroup *ancestor)
1410 {
1411 #ifdef CONFIG_SOCK_CGROUP_DATA
1412 	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1413 				    ancestor);
1414 #else
1415 	return -ENOTSUPP;
1416 #endif
1417 }
1418 
1419 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1420 
1421 static inline void sk_sockets_allocated_dec(struct sock *sk)
1422 {
1423 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1424 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1425 }
1426 
1427 static inline void sk_sockets_allocated_inc(struct sock *sk)
1428 {
1429 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1430 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1431 }
1432 
1433 static inline u64
1434 sk_sockets_allocated_read_positive(struct sock *sk)
1435 {
1436 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1437 }
1438 
1439 static inline int
1440 proto_sockets_allocated_sum_positive(struct proto *prot)
1441 {
1442 	return percpu_counter_sum_positive(prot->sockets_allocated);
1443 }
1444 
1445 #ifdef CONFIG_PROC_FS
1446 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
1447 struct prot_inuse {
1448 	int all;
1449 	int val[PROTO_INUSE_NR];
1450 };
1451 
1452 static inline void sock_prot_inuse_add(const struct net *net,
1453 				       const struct proto *prot, int val)
1454 {
1455 	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1456 }
1457 
1458 static inline void sock_inuse_add(const struct net *net, int val)
1459 {
1460 	this_cpu_add(net->core.prot_inuse->all, val);
1461 }
1462 
1463 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1464 int sock_inuse_get(struct net *net);
1465 #else
1466 static inline void sock_prot_inuse_add(const struct net *net,
1467 				       const struct proto *prot, int val)
1468 {
1469 }
1470 
1471 static inline void sock_inuse_add(const struct net *net, int val)
1472 {
1473 }
1474 #endif
1475 
1476 
1477 /* With per-bucket locks this operation is not-atomic, so that
1478  * this version is not worse.
1479  */
1480 static inline int __sk_prot_rehash(struct sock *sk)
1481 {
1482 	sk->sk_prot->unhash(sk);
1483 	return sk->sk_prot->hash(sk);
1484 }
1485 
1486 /* About 10 seconds */
1487 #define SOCK_DESTROY_TIME (10*HZ)
1488 
1489 /* Sockets 0-1023 can't be bound to unless you are superuser */
1490 #define PROT_SOCK	1024
1491 
1492 #define SHUTDOWN_MASK	3
1493 #define RCV_SHUTDOWN	1
1494 #define SEND_SHUTDOWN	2
1495 
1496 #define SOCK_BINDADDR_LOCK	4
1497 #define SOCK_BINDPORT_LOCK	8
1498 
1499 struct socket_alloc {
1500 	struct socket socket;
1501 	struct inode vfs_inode;
1502 };
1503 
1504 static inline struct socket *SOCKET_I(struct inode *inode)
1505 {
1506 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1507 }
1508 
1509 static inline struct inode *SOCK_INODE(struct socket *socket)
1510 {
1511 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1512 }
1513 
1514 /*
1515  * Functions for memory accounting
1516  */
1517 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1518 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1519 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1520 void __sk_mem_reclaim(struct sock *sk, int amount);
1521 
1522 #define SK_MEM_SEND	0
1523 #define SK_MEM_RECV	1
1524 
1525 /* sysctl_mem values are in pages */
1526 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1527 {
1528 	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1529 }
1530 
1531 static inline int sk_mem_pages(int amt)
1532 {
1533 	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1534 }
1535 
1536 static inline bool sk_has_account(struct sock *sk)
1537 {
1538 	/* return true if protocol supports memory accounting */
1539 	return !!sk->sk_prot->memory_allocated;
1540 }
1541 
1542 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1543 {
1544 	int delta;
1545 
1546 	if (!sk_has_account(sk))
1547 		return true;
1548 	delta = size - sk->sk_forward_alloc;
1549 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1550 }
1551 
1552 static inline bool
1553 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc)
1554 {
1555 	int delta;
1556 
1557 	if (!sk_has_account(sk))
1558 		return true;
1559 	delta = size - sk->sk_forward_alloc;
1560 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1561 	       pfmemalloc;
1562 }
1563 
1564 static inline bool
1565 sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size)
1566 {
1567 	return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb));
1568 }
1569 
1570 static inline int sk_unused_reserved_mem(const struct sock *sk)
1571 {
1572 	int unused_mem;
1573 
1574 	if (likely(!sk->sk_reserved_mem))
1575 		return 0;
1576 
1577 	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1578 			atomic_read(&sk->sk_rmem_alloc);
1579 
1580 	return unused_mem > 0 ? unused_mem : 0;
1581 }
1582 
1583 static inline void sk_mem_reclaim(struct sock *sk)
1584 {
1585 	int reclaimable;
1586 
1587 	if (!sk_has_account(sk))
1588 		return;
1589 
1590 	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1591 
1592 	if (reclaimable >= (int)PAGE_SIZE)
1593 		__sk_mem_reclaim(sk, reclaimable);
1594 }
1595 
1596 static inline void sk_mem_reclaim_final(struct sock *sk)
1597 {
1598 	sk->sk_reserved_mem = 0;
1599 	sk_mem_reclaim(sk);
1600 }
1601 
1602 static inline void sk_mem_charge(struct sock *sk, int size)
1603 {
1604 	if (!sk_has_account(sk))
1605 		return;
1606 	sk_forward_alloc_add(sk, -size);
1607 }
1608 
1609 static inline void sk_mem_uncharge(struct sock *sk, int size)
1610 {
1611 	if (!sk_has_account(sk))
1612 		return;
1613 	sk_forward_alloc_add(sk, size);
1614 	sk_mem_reclaim(sk);
1615 }
1616 
1617 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
1618 static inline void sk_owner_set(struct sock *sk, struct module *owner)
1619 {
1620 	__module_get(owner);
1621 	sk->sk_owner = owner;
1622 }
1623 
1624 static inline void sk_owner_clear(struct sock *sk)
1625 {
1626 	sk->sk_owner = NULL;
1627 }
1628 
1629 static inline void sk_owner_put(struct sock *sk)
1630 {
1631 	module_put(sk->sk_owner);
1632 }
1633 #else
1634 static inline void sk_owner_set(struct sock *sk, struct module *owner)
1635 {
1636 }
1637 
1638 static inline void sk_owner_clear(struct sock *sk)
1639 {
1640 }
1641 
1642 static inline void sk_owner_put(struct sock *sk)
1643 {
1644 }
1645 #endif
1646 /*
1647  * Macro so as to not evaluate some arguments when
1648  * lockdep is not enabled.
1649  *
1650  * Mark both the sk_lock and the sk_lock.slock as a
1651  * per-address-family lock class.
1652  */
1653 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1654 do {									\
1655 	sk_owner_set(sk, THIS_MODULE);					\
1656 	sk->sk_lock.owned = 0;						\
1657 	init_waitqueue_head(&sk->sk_lock.wq);				\
1658 	spin_lock_init(&(sk)->sk_lock.slock);				\
1659 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1660 				   sizeof((sk)->sk_lock));		\
1661 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1662 				   (skey), (sname));			\
1663 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1664 } while (0)
1665 
1666 static inline bool lockdep_sock_is_held(const struct sock *sk)
1667 {
1668 	return lockdep_is_held(&sk->sk_lock) ||
1669 	       lockdep_is_held(&sk->sk_lock.slock);
1670 }
1671 
1672 void lock_sock_nested(struct sock *sk, int subclass);
1673 
1674 static inline void lock_sock(struct sock *sk)
1675 {
1676 	lock_sock_nested(sk, 0);
1677 }
1678 
1679 void __lock_sock(struct sock *sk);
1680 void __release_sock(struct sock *sk);
1681 void release_sock(struct sock *sk);
1682 
1683 /* BH context may only use the following locking interface. */
1684 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1685 #define bh_lock_sock_nested(__sk) \
1686 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1687 				SINGLE_DEPTH_NESTING)
1688 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1689 
1690 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1691 
1692 /**
1693  * lock_sock_fast - fast version of lock_sock
1694  * @sk: socket
1695  *
1696  * This version should be used for very small section, where process won't block
1697  * return false if fast path is taken:
1698  *
1699  *   sk_lock.slock locked, owned = 0, BH disabled
1700  *
1701  * return true if slow path is taken:
1702  *
1703  *   sk_lock.slock unlocked, owned = 1, BH enabled
1704  */
1705 static inline bool lock_sock_fast(struct sock *sk)
1706 {
1707 	/* The sk_lock has mutex_lock() semantics here. */
1708 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1709 
1710 	return __lock_sock_fast(sk);
1711 }
1712 
1713 /* fast socket lock variant for caller already holding a [different] socket lock */
1714 static inline bool lock_sock_fast_nested(struct sock *sk)
1715 {
1716 	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1717 
1718 	return __lock_sock_fast(sk);
1719 }
1720 
1721 /**
1722  * unlock_sock_fast - complement of lock_sock_fast
1723  * @sk: socket
1724  * @slow: slow mode
1725  *
1726  * fast unlock socket for user context.
1727  * If slow mode is on, we call regular release_sock()
1728  */
1729 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1730 	__releases(&sk->sk_lock.slock)
1731 {
1732 	if (slow) {
1733 		release_sock(sk);
1734 		__release(&sk->sk_lock.slock);
1735 	} else {
1736 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1737 		spin_unlock_bh(&sk->sk_lock.slock);
1738 	}
1739 }
1740 
1741 void sockopt_lock_sock(struct sock *sk);
1742 void sockopt_release_sock(struct sock *sk);
1743 bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1744 bool sockopt_capable(int cap);
1745 
1746 /* Used by processes to "lock" a socket state, so that
1747  * interrupts and bottom half handlers won't change it
1748  * from under us. It essentially blocks any incoming
1749  * packets, so that we won't get any new data or any
1750  * packets that change the state of the socket.
1751  *
1752  * While locked, BH processing will add new packets to
1753  * the backlog queue.  This queue is processed by the
1754  * owner of the socket lock right before it is released.
1755  *
1756  * Since ~2.3.5 it is also exclusive sleep lock serializing
1757  * accesses from user process context.
1758  */
1759 
1760 static inline void sock_owned_by_me(const struct sock *sk)
1761 {
1762 #ifdef CONFIG_LOCKDEP
1763 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1764 #endif
1765 }
1766 
1767 static inline void sock_not_owned_by_me(const struct sock *sk)
1768 {
1769 #ifdef CONFIG_LOCKDEP
1770 	WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1771 #endif
1772 }
1773 
1774 static inline bool sock_owned_by_user(const struct sock *sk)
1775 {
1776 	sock_owned_by_me(sk);
1777 	return sk->sk_lock.owned;
1778 }
1779 
1780 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1781 {
1782 	return sk->sk_lock.owned;
1783 }
1784 
1785 static inline void sock_release_ownership(struct sock *sk)
1786 {
1787 	DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1788 	sk->sk_lock.owned = 0;
1789 
1790 	/* The sk_lock has mutex_unlock() semantics: */
1791 	mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1792 }
1793 
1794 /* no reclassification while locks are held */
1795 static inline bool sock_allow_reclassification(const struct sock *csk)
1796 {
1797 	struct sock *sk = (struct sock *)csk;
1798 
1799 	return !sock_owned_by_user_nocheck(sk) &&
1800 		!spin_is_locked(&sk->sk_lock.slock);
1801 }
1802 
1803 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1804 		      struct proto *prot, int kern);
1805 void sk_free(struct sock *sk);
1806 void sk_net_refcnt_upgrade(struct sock *sk);
1807 void sk_destruct(struct sock *sk);
1808 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1809 
1810 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1811 			     gfp_t priority);
1812 void __sock_wfree(struct sk_buff *skb);
1813 void sock_wfree(struct sk_buff *skb);
1814 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1815 			     gfp_t priority);
1816 void skb_orphan_partial(struct sk_buff *skb);
1817 void sock_rfree(struct sk_buff *skb);
1818 void sock_efree(struct sk_buff *skb);
1819 #ifdef CONFIG_INET
1820 void sock_edemux(struct sk_buff *skb);
1821 void sock_pfree(struct sk_buff *skb);
1822 
1823 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk)
1824 {
1825 	skb_orphan(skb);
1826 	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1827 		skb->sk = sk;
1828 		skb->destructor = sock_edemux;
1829 	}
1830 }
1831 #else
1832 #define sock_edemux sock_efree
1833 #endif
1834 
1835 int sk_setsockopt(struct sock *sk, int level, int optname,
1836 		  sockptr_t optval, unsigned int optlen);
1837 int sock_setsockopt(struct socket *sock, int level, int op,
1838 		    sockptr_t optval, unsigned int optlen);
1839 int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1840 		       int optname, sockptr_t optval, int optlen);
1841 int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1842 		       int optname, sockptr_t optval, sockptr_t optlen);
1843 
1844 int sk_getsockopt(struct sock *sk, int level, int optname,
1845 		  sockptr_t optval, sockptr_t optlen);
1846 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1847 		   bool timeval, bool time32);
1848 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1849 				     unsigned long data_len, int noblock,
1850 				     int *errcode, int max_page_order);
1851 
1852 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1853 						  unsigned long size,
1854 						  int noblock, int *errcode)
1855 {
1856 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1857 }
1858 
1859 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1860 void *sock_kmemdup(struct sock *sk, const void *src,
1861 		   int size, gfp_t priority);
1862 void sock_kfree_s(struct sock *sk, void *mem, int size);
1863 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1864 void sk_send_sigurg(struct sock *sk);
1865 
1866 static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1867 {
1868 	if (sk->sk_socket)
1869 		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1870 	WRITE_ONCE(sk->sk_prot, proto);
1871 }
1872 
1873 struct sockcm_cookie {
1874 	u64 transmit_time;
1875 	u32 mark;
1876 	u32 tsflags;
1877 	u32 ts_opt_id;
1878 	u32 priority;
1879 	u32 dmabuf_id;
1880 };
1881 
1882 static inline void sockcm_init(struct sockcm_cookie *sockc,
1883 			       const struct sock *sk)
1884 {
1885 	*sockc = (struct sockcm_cookie) {
1886 		.mark = READ_ONCE(sk->sk_mark),
1887 		.tsflags = READ_ONCE(sk->sk_tsflags),
1888 		.priority = READ_ONCE(sk->sk_priority),
1889 	};
1890 }
1891 
1892 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1893 		     struct sockcm_cookie *sockc);
1894 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1895 		   struct sockcm_cookie *sockc);
1896 
1897 /*
1898  * Functions to fill in entries in struct proto_ops when a protocol
1899  * does not implement a particular function.
1900  */
1901 int sock_no_bind(struct socket *, struct sockaddr *, int);
1902 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1903 int sock_no_socketpair(struct socket *, struct socket *);
1904 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *);
1905 int sock_no_getname(struct socket *, struct sockaddr *, int);
1906 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1907 int sock_no_listen(struct socket *, int);
1908 int sock_no_shutdown(struct socket *, int);
1909 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1910 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1911 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1912 int sock_no_mmap(struct file *file, struct socket *sock,
1913 		 struct vm_area_struct *vma);
1914 
1915 /*
1916  * Functions to fill in entries in struct proto_ops when a protocol
1917  * uses the inet style.
1918  */
1919 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1920 				  char __user *optval, int __user *optlen);
1921 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1922 			int flags);
1923 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1924 			   sockptr_t optval, unsigned int optlen);
1925 
1926 void sk_common_release(struct sock *sk);
1927 
1928 /*
1929  *	Default socket callbacks and setup code
1930  */
1931 
1932 /* Initialise core socket variables using an explicit uid. */
1933 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1934 
1935 /* Initialise core socket variables.
1936  * Assumes struct socket *sock is embedded in a struct socket_alloc.
1937  */
1938 void sock_init_data(struct socket *sock, struct sock *sk);
1939 
1940 /*
1941  * Socket reference counting postulates.
1942  *
1943  * * Each user of socket SHOULD hold a reference count.
1944  * * Each access point to socket (an hash table bucket, reference from a list,
1945  *   running timer, skb in flight MUST hold a reference count.
1946  * * When reference count hits 0, it means it will never increase back.
1947  * * When reference count hits 0, it means that no references from
1948  *   outside exist to this socket and current process on current CPU
1949  *   is last user and may/should destroy this socket.
1950  * * sk_free is called from any context: process, BH, IRQ. When
1951  *   it is called, socket has no references from outside -> sk_free
1952  *   may release descendant resources allocated by the socket, but
1953  *   to the time when it is called, socket is NOT referenced by any
1954  *   hash tables, lists etc.
1955  * * Packets, delivered from outside (from network or from another process)
1956  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1957  *   when they sit in queue. Otherwise, packets will leak to hole, when
1958  *   socket is looked up by one cpu and unhasing is made by another CPU.
1959  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1960  *   (leak to backlog). Packet socket does all the processing inside
1961  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1962  *   use separate SMP lock, so that they are prone too.
1963  */
1964 
1965 /* Ungrab socket and destroy it, if it was the last reference. */
1966 static inline void sock_put(struct sock *sk)
1967 {
1968 	if (refcount_dec_and_test(&sk->sk_refcnt))
1969 		sk_free(sk);
1970 }
1971 /* Generic version of sock_put(), dealing with all sockets
1972  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1973  */
1974 void sock_gen_put(struct sock *sk);
1975 
1976 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1977 		     unsigned int trim_cap, bool refcounted);
1978 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1979 				 const int nested)
1980 {
1981 	return __sk_receive_skb(sk, skb, nested, 1, true);
1982 }
1983 
1984 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1985 {
1986 	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1987 	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1988 		return;
1989 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1990 	 * other WRITE_ONCE() because socket lock might be not held.
1991 	 */
1992 	WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1993 }
1994 
1995 #define NO_QUEUE_MAPPING	USHRT_MAX
1996 
1997 static inline void sk_tx_queue_clear(struct sock *sk)
1998 {
1999 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
2000 	 * other WRITE_ONCE() because socket lock might be not held.
2001 	 */
2002 	WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
2003 }
2004 
2005 static inline int sk_tx_queue_get(const struct sock *sk)
2006 {
2007 	if (sk) {
2008 		/* Paired with WRITE_ONCE() in sk_tx_queue_clear()
2009 		 * and sk_tx_queue_set().
2010 		 */
2011 		int val = READ_ONCE(sk->sk_tx_queue_mapping);
2012 
2013 		if (val != NO_QUEUE_MAPPING)
2014 			return val;
2015 	}
2016 	return -1;
2017 }
2018 
2019 static inline void __sk_rx_queue_set(struct sock *sk,
2020 				     const struct sk_buff *skb,
2021 				     bool force_set)
2022 {
2023 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2024 	if (skb_rx_queue_recorded(skb)) {
2025 		u16 rx_queue = skb_get_rx_queue(skb);
2026 
2027 		if (force_set ||
2028 		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2029 			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2030 	}
2031 #endif
2032 }
2033 
2034 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2035 {
2036 	__sk_rx_queue_set(sk, skb, true);
2037 }
2038 
2039 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2040 {
2041 	__sk_rx_queue_set(sk, skb, false);
2042 }
2043 
2044 static inline void sk_rx_queue_clear(struct sock *sk)
2045 {
2046 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2047 	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2048 #endif
2049 }
2050 
2051 static inline int sk_rx_queue_get(const struct sock *sk)
2052 {
2053 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2054 	if (sk) {
2055 		int res = READ_ONCE(sk->sk_rx_queue_mapping);
2056 
2057 		if (res != NO_QUEUE_MAPPING)
2058 			return res;
2059 	}
2060 #endif
2061 
2062 	return -1;
2063 }
2064 
2065 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2066 {
2067 	sk->sk_socket = sock;
2068 	if (sock) {
2069 		WRITE_ONCE(sk->sk_uid, SOCK_INODE(sock)->i_uid);
2070 		WRITE_ONCE(sk->sk_ino, SOCK_INODE(sock)->i_ino);
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 	/* Note: sk_uid is unchanged. */
2093 	WRITE_ONCE(sk->sk_ino, 0);
2094 	write_unlock_bh(&sk->sk_callback_lock);
2095 }
2096 
2097 static inline void sock_graft(struct sock *sk, struct socket *parent)
2098 {
2099 	WARN_ON(parent->sk);
2100 	write_lock_bh(&sk->sk_callback_lock);
2101 	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2102 	parent->sk = sk;
2103 	sk_set_socket(sk, parent);
2104 	security_sock_graft(sk, parent);
2105 	write_unlock_bh(&sk->sk_callback_lock);
2106 }
2107 
2108 static inline unsigned long sock_i_ino(const struct sock *sk)
2109 {
2110 	/* Paired with WRITE_ONCE() in sock_graft() and sock_orphan() */
2111 	return READ_ONCE(sk->sk_ino);
2112 }
2113 
2114 static inline kuid_t sk_uid(const struct sock *sk)
2115 {
2116 	/* Paired with WRITE_ONCE() in sockfs_setattr() */
2117 	return READ_ONCE(sk->sk_uid);
2118 }
2119 
2120 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2121 {
2122 	return sk ? sk_uid(sk) : make_kuid(net->user_ns, 0);
2123 }
2124 
2125 static inline u32 net_tx_rndhash(void)
2126 {
2127 	u32 v = get_random_u32();
2128 
2129 	return v ?: 1;
2130 }
2131 
2132 static inline void sk_set_txhash(struct sock *sk)
2133 {
2134 	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2135 	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2136 }
2137 
2138 static inline bool sk_rethink_txhash(struct sock *sk)
2139 {
2140 	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2141 		sk_set_txhash(sk);
2142 		return true;
2143 	}
2144 	return false;
2145 }
2146 
2147 static inline struct dst_entry *
2148 __sk_dst_get(const struct sock *sk)
2149 {
2150 	return rcu_dereference_check(sk->sk_dst_cache,
2151 				     lockdep_sock_is_held(sk));
2152 }
2153 
2154 static inline struct dst_entry *
2155 sk_dst_get(const struct sock *sk)
2156 {
2157 	struct dst_entry *dst;
2158 
2159 	rcu_read_lock();
2160 	dst = rcu_dereference(sk->sk_dst_cache);
2161 	if (dst && !rcuref_get(&dst->__rcuref))
2162 		dst = NULL;
2163 	rcu_read_unlock();
2164 	return dst;
2165 }
2166 
2167 static inline void __dst_negative_advice(struct sock *sk)
2168 {
2169 	struct dst_entry *dst = __sk_dst_get(sk);
2170 
2171 	if (dst && dst->ops->negative_advice)
2172 		dst->ops->negative_advice(sk, dst);
2173 }
2174 
2175 static inline void dst_negative_advice(struct sock *sk)
2176 {
2177 	sk_rethink_txhash(sk);
2178 	__dst_negative_advice(sk);
2179 }
2180 
2181 static inline void
2182 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2183 {
2184 	struct dst_entry *old_dst;
2185 
2186 	sk_tx_queue_clear(sk);
2187 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2188 	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2189 					    lockdep_sock_is_held(sk));
2190 	rcu_assign_pointer(sk->sk_dst_cache, dst);
2191 	dst_release(old_dst);
2192 }
2193 
2194 static inline void
2195 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2196 {
2197 	struct dst_entry *old_dst;
2198 
2199 	sk_tx_queue_clear(sk);
2200 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2201 	old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst)));
2202 	dst_release(old_dst);
2203 }
2204 
2205 static inline void
2206 __sk_dst_reset(struct sock *sk)
2207 {
2208 	__sk_dst_set(sk, NULL);
2209 }
2210 
2211 static inline void
2212 sk_dst_reset(struct sock *sk)
2213 {
2214 	sk_dst_set(sk, NULL);
2215 }
2216 
2217 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2218 
2219 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2220 
2221 static inline void sk_dst_confirm(struct sock *sk)
2222 {
2223 	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2224 		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2225 }
2226 
2227 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2228 {
2229 	if (skb_get_dst_pending_confirm(skb)) {
2230 		struct sock *sk = skb->sk;
2231 
2232 		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2233 			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2234 		neigh_confirm(n);
2235 	}
2236 }
2237 
2238 bool sk_mc_loop(const struct sock *sk);
2239 
2240 static inline bool sk_can_gso(const struct sock *sk)
2241 {
2242 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2243 }
2244 
2245 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2246 
2247 static inline void sk_gso_disable(struct sock *sk)
2248 {
2249 	sk->sk_gso_disabled = 1;
2250 	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2251 }
2252 
2253 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2254 					   struct iov_iter *from, char *to,
2255 					   int copy, int offset)
2256 {
2257 	if (skb->ip_summed == CHECKSUM_NONE) {
2258 		__wsum csum = 0;
2259 		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2260 			return -EFAULT;
2261 		skb->csum = csum_block_add(skb->csum, csum, offset);
2262 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2263 		if (!copy_from_iter_full_nocache(to, copy, from))
2264 			return -EFAULT;
2265 	} else if (!copy_from_iter_full(to, copy, from))
2266 		return -EFAULT;
2267 
2268 	return 0;
2269 }
2270 
2271 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2272 				       struct iov_iter *from, int copy)
2273 {
2274 	int err, offset = skb->len;
2275 
2276 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2277 				       copy, offset);
2278 	if (err)
2279 		__skb_trim(skb, offset);
2280 
2281 	return err;
2282 }
2283 
2284 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2285 					   struct sk_buff *skb,
2286 					   struct page *page,
2287 					   int off, int copy)
2288 {
2289 	int err;
2290 
2291 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2292 				       copy, skb->len);
2293 	if (err)
2294 		return err;
2295 
2296 	skb_len_add(skb, copy);
2297 	sk_wmem_queued_add(sk, copy);
2298 	sk_mem_charge(sk, copy);
2299 	return 0;
2300 }
2301 
2302 /**
2303  * sk_wmem_alloc_get - returns write allocations
2304  * @sk: socket
2305  *
2306  * Return: sk_wmem_alloc minus initial offset of one
2307  */
2308 static inline int sk_wmem_alloc_get(const struct sock *sk)
2309 {
2310 	return refcount_read(&sk->sk_wmem_alloc) - 1;
2311 }
2312 
2313 /**
2314  * sk_rmem_alloc_get - returns read allocations
2315  * @sk: socket
2316  *
2317  * Return: sk_rmem_alloc
2318  */
2319 static inline int sk_rmem_alloc_get(const struct sock *sk)
2320 {
2321 	return atomic_read(&sk->sk_rmem_alloc);
2322 }
2323 
2324 /**
2325  * sk_has_allocations - check if allocations are outstanding
2326  * @sk: socket
2327  *
2328  * Return: true if socket has write or read allocations
2329  */
2330 static inline bool sk_has_allocations(const struct sock *sk)
2331 {
2332 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2333 }
2334 
2335 /**
2336  * skwq_has_sleeper - check if there are any waiting processes
2337  * @wq: struct socket_wq
2338  *
2339  * Return: true if socket_wq has waiting processes
2340  *
2341  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2342  * barrier call. They were added due to the race found within the tcp code.
2343  *
2344  * Consider following tcp code paths::
2345  *
2346  *   CPU1                CPU2
2347  *   sys_select          receive packet
2348  *   ...                 ...
2349  *   __add_wait_queue    update tp->rcv_nxt
2350  *   ...                 ...
2351  *   tp->rcv_nxt check   sock_def_readable
2352  *   ...                 {
2353  *   schedule               rcu_read_lock();
2354  *                          wq = rcu_dereference(sk->sk_wq);
2355  *                          if (wq && waitqueue_active(&wq->wait))
2356  *                              wake_up_interruptible(&wq->wait)
2357  *                          ...
2358  *                       }
2359  *
2360  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2361  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2362  * could then endup calling schedule and sleep forever if there are no more
2363  * data on the socket.
2364  *
2365  */
2366 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2367 {
2368 	return wq && wq_has_sleeper(&wq->wait);
2369 }
2370 
2371 /**
2372  * sock_poll_wait - wrapper for the poll_wait call.
2373  * @filp:           file
2374  * @sock:           socket to wait on
2375  * @p:              poll_table
2376  *
2377  * See the comments in the wq_has_sleeper function.
2378  */
2379 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2380 				  poll_table *p)
2381 {
2382 	/* Provides a barrier we need to be sure we are in sync
2383 	 * with the socket flags modification.
2384 	 *
2385 	 * This memory barrier is paired in the wq_has_sleeper.
2386 	 */
2387 	poll_wait(filp, &sock->wq.wait, p);
2388 }
2389 
2390 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2391 {
2392 	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2393 	u32 txhash = READ_ONCE(sk->sk_txhash);
2394 
2395 	if (txhash) {
2396 		skb->l4_hash = 1;
2397 		skb->hash = txhash;
2398 	}
2399 }
2400 
2401 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2402 
2403 /*
2404  *	Queue a received datagram if it will fit. Stream and sequenced
2405  *	protocols can't normally use this as they need to fit buffers in
2406  *	and play with them.
2407  *
2408  *	Inlined as it's very short and called for pretty much every
2409  *	packet ever received.
2410  */
2411 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2412 {
2413 	skb_orphan(skb);
2414 	skb->sk = sk;
2415 	skb->destructor = sock_rfree;
2416 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2417 	sk_mem_charge(sk, skb->truesize);
2418 }
2419 
2420 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2421 {
2422 	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2423 		skb_orphan(skb);
2424 		skb->destructor = sock_efree;
2425 		skb->sk = sk;
2426 		return true;
2427 	}
2428 	return false;
2429 }
2430 
2431 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2432 {
2433 	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2434 	if (skb) {
2435 		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2436 			skb_set_owner_r(skb, sk);
2437 			return skb;
2438 		}
2439 		__kfree_skb(skb);
2440 	}
2441 	return NULL;
2442 }
2443 
2444 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2445 {
2446 	if (skb->destructor != sock_wfree) {
2447 		skb_orphan(skb);
2448 		return;
2449 	}
2450 	skb->slow_gro = 1;
2451 }
2452 
2453 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2454 		    unsigned long expires);
2455 
2456 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2457 
2458 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2459 
2460 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2461 			struct sk_buff *skb, unsigned int flags,
2462 			void (*destructor)(struct sock *sk,
2463 					   struct sk_buff *skb));
2464 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2465 
2466 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2467 			      enum skb_drop_reason *reason);
2468 
2469 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2470 {
2471 	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2472 }
2473 
2474 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2475 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2476 
2477 /*
2478  *	Recover an error report and clear atomically
2479  */
2480 
2481 static inline int sock_error(struct sock *sk)
2482 {
2483 	int err;
2484 
2485 	/* Avoid an atomic operation for the common case.
2486 	 * This is racy since another cpu/thread can change sk_err under us.
2487 	 */
2488 	if (likely(data_race(!sk->sk_err)))
2489 		return 0;
2490 
2491 	err = xchg(&sk->sk_err, 0);
2492 	return -err;
2493 }
2494 
2495 void sk_error_report(struct sock *sk);
2496 
2497 static inline unsigned long sock_wspace(struct sock *sk)
2498 {
2499 	int amt = 0;
2500 
2501 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2502 		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2503 		if (amt < 0)
2504 			amt = 0;
2505 	}
2506 	return amt;
2507 }
2508 
2509 /* Note:
2510  *  We use sk->sk_wq_raw, from contexts knowing this
2511  *  pointer is not NULL and cannot disappear/change.
2512  */
2513 static inline void sk_set_bit(int nr, struct sock *sk)
2514 {
2515 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2516 	    !sock_flag(sk, SOCK_FASYNC))
2517 		return;
2518 
2519 	set_bit(nr, &sk->sk_wq_raw->flags);
2520 }
2521 
2522 static inline void sk_clear_bit(int nr, struct sock *sk)
2523 {
2524 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2525 	    !sock_flag(sk, SOCK_FASYNC))
2526 		return;
2527 
2528 	clear_bit(nr, &sk->sk_wq_raw->flags);
2529 }
2530 
2531 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2532 {
2533 	if (sock_flag(sk, SOCK_FASYNC)) {
2534 		rcu_read_lock();
2535 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2536 		rcu_read_unlock();
2537 	}
2538 }
2539 
2540 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band)
2541 {
2542 	if (unlikely(sock_flag(sk, SOCK_FASYNC)))
2543 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2544 }
2545 
2546 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2547  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2548  * Note: for send buffers, TCP works better if we can build two skbs at
2549  * minimum.
2550  */
2551 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2552 
2553 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2554 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2555 
2556 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2557 {
2558 	u32 val;
2559 
2560 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2561 		return;
2562 
2563 	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2564 	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2565 
2566 	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2567 }
2568 
2569 /**
2570  * sk_page_frag - return an appropriate page_frag
2571  * @sk: socket
2572  *
2573  * Use the per task page_frag instead of the per socket one for
2574  * optimization when we know that we're in process context and own
2575  * everything that's associated with %current.
2576  *
2577  * Both direct reclaim and page faults can nest inside other
2578  * socket operations and end up recursing into sk_page_frag()
2579  * while it's already in use: explicitly avoid task page_frag
2580  * when users disable sk_use_task_frag.
2581  *
2582  * Return: a per task page_frag if context allows that,
2583  * otherwise a per socket one.
2584  */
2585 static inline struct page_frag *sk_page_frag(struct sock *sk)
2586 {
2587 	if (sk->sk_use_task_frag)
2588 		return &current->task_frag;
2589 
2590 	return &sk->sk_frag;
2591 }
2592 
2593 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2594 
2595 /*
2596  *	Default write policy as shown to user space via poll/select/SIGIO
2597  */
2598 static inline bool sock_writeable(const struct sock *sk)
2599 {
2600 	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2601 }
2602 
2603 static inline gfp_t gfp_any(void)
2604 {
2605 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2606 }
2607 
2608 static inline gfp_t gfp_memcg_charge(void)
2609 {
2610 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2611 }
2612 
2613 #ifdef CONFIG_MEMCG
2614 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2615 {
2616 	return sk->sk_memcg;
2617 }
2618 
2619 static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2620 {
2621 	return mem_cgroup_sockets_enabled && mem_cgroup_from_sk(sk);
2622 }
2623 
2624 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2625 {
2626 	struct mem_cgroup *memcg = mem_cgroup_from_sk(sk);
2627 
2628 #ifdef CONFIG_MEMCG_V1
2629 	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2630 		return !!memcg->tcpmem_pressure;
2631 #endif /* CONFIG_MEMCG_V1 */
2632 
2633 	do {
2634 		if (time_before64(get_jiffies_64(), mem_cgroup_get_socket_pressure(memcg)))
2635 			return true;
2636 	} while ((memcg = parent_mem_cgroup(memcg)));
2637 
2638 	return false;
2639 }
2640 #else
2641 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2642 {
2643 	return NULL;
2644 }
2645 
2646 static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2647 {
2648 	return false;
2649 }
2650 
2651 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2652 {
2653 	return false;
2654 }
2655 #endif
2656 
2657 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2658 {
2659 	return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo);
2660 }
2661 
2662 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2663 {
2664 	return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo);
2665 }
2666 
2667 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2668 {
2669 	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2670 
2671 	return v ?: 1;
2672 }
2673 
2674 /* Alas, with timeout socket operations are not restartable.
2675  * Compare this to poll().
2676  */
2677 static inline int sock_intr_errno(long timeo)
2678 {
2679 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2680 }
2681 
2682 struct sock_skb_cb {
2683 	u32 dropcount;
2684 };
2685 
2686 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2687  * using skb->cb[] would keep using it directly and utilize its
2688  * alignment guarantee.
2689  */
2690 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \
2691 			    sizeof(struct sock_skb_cb))
2692 
2693 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2694 			    SOCK_SKB_CB_OFFSET))
2695 
2696 #define sock_skb_cb_check_size(size) \
2697 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2698 
2699 static inline void sk_drops_add(struct sock *sk, int segs)
2700 {
2701 	struct socket_drop_counters *sdc = sk->sk_drop_counters;
2702 
2703 	if (sdc) {
2704 		int n = numa_node_id() % 2;
2705 
2706 		if (n)
2707 			atomic_add(segs, &sdc->drops1);
2708 		else
2709 			atomic_add(segs, &sdc->drops0);
2710 	} else {
2711 		atomic_add(segs, &sk->sk_drops);
2712 	}
2713 }
2714 
2715 static inline void sk_drops_inc(struct sock *sk)
2716 {
2717 	sk_drops_add(sk, 1);
2718 }
2719 
2720 static inline int sk_drops_read(const struct sock *sk)
2721 {
2722 	const struct socket_drop_counters *sdc = sk->sk_drop_counters;
2723 
2724 	if (sdc) {
2725 		DEBUG_NET_WARN_ON_ONCE(atomic_read(&sk->sk_drops));
2726 		return atomic_read(&sdc->drops0) + atomic_read(&sdc->drops1);
2727 	}
2728 	return atomic_read(&sk->sk_drops);
2729 }
2730 
2731 static inline void sk_drops_reset(struct sock *sk)
2732 {
2733 	struct socket_drop_counters *sdc = sk->sk_drop_counters;
2734 
2735 	if (sdc) {
2736 		atomic_set(&sdc->drops0, 0);
2737 		atomic_set(&sdc->drops1, 0);
2738 	}
2739 	atomic_set(&sk->sk_drops, 0);
2740 }
2741 
2742 static inline void
2743 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2744 {
2745 	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2746 						sk_drops_read(sk) : 0;
2747 }
2748 
2749 static inline void sk_drops_skbadd(struct sock *sk, const struct sk_buff *skb)
2750 {
2751 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2752 
2753 	sk_drops_add(sk, segs);
2754 }
2755 
2756 static inline ktime_t sock_read_timestamp(struct sock *sk)
2757 {
2758 #if BITS_PER_LONG==32
2759 	unsigned int seq;
2760 	ktime_t kt;
2761 
2762 	do {
2763 		seq = read_seqbegin(&sk->sk_stamp_seq);
2764 		kt = sk->sk_stamp;
2765 	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2766 
2767 	return kt;
2768 #else
2769 	return READ_ONCE(sk->sk_stamp);
2770 #endif
2771 }
2772 
2773 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2774 {
2775 #if BITS_PER_LONG==32
2776 	write_seqlock(&sk->sk_stamp_seq);
2777 	sk->sk_stamp = kt;
2778 	write_sequnlock(&sk->sk_stamp_seq);
2779 #else
2780 	WRITE_ONCE(sk->sk_stamp, kt);
2781 #endif
2782 }
2783 
2784 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2785 			   struct sk_buff *skb);
2786 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2787 			     struct sk_buff *skb);
2788 
2789 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk);
2790 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk,
2791 			 struct timespec64 *ts);
2792 
2793 static inline void
2794 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2795 {
2796 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2797 	u32 tsflags = READ_ONCE(sk->sk_tsflags);
2798 	ktime_t kt = skb->tstamp;
2799 	/*
2800 	 * generate control messages if
2801 	 * - receive time stamping in software requested
2802 	 * - software time stamp available and wanted
2803 	 * - hardware time stamps available and wanted
2804 	 */
2805 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2806 	    (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2807 	    (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2808 	    (hwtstamps->hwtstamp &&
2809 	     (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2810 		__sock_recv_timestamp(msg, sk, skb);
2811 	else
2812 		sock_write_timestamp(sk, kt);
2813 
2814 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2815 		__sock_recv_wifi_status(msg, sk, skb);
2816 }
2817 
2818 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2819 		       struct sk_buff *skb);
2820 
2821 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2822 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2823 				   struct sk_buff *skb)
2824 {
2825 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2826 			   (1UL << SOCK_RCVTSTAMP)			| \
2827 			   (1UL << SOCK_RCVMARK)			| \
2828 			   (1UL << SOCK_RCVPRIORITY)			| \
2829 			   (1UL << SOCK_TIMESTAMPING_ANY))
2830 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2831 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2832 
2833 	if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS)
2834 		__sock_recv_cmsgs(msg, sk, skb);
2835 	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2836 		sock_write_timestamp(sk, skb->tstamp);
2837 	else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2838 		sock_write_timestamp(sk, 0);
2839 }
2840 
2841 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags);
2842 
2843 /**
2844  * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2845  * @sk:		socket sending this packet
2846  * @sockc:	pointer to socket cmsg cookie to get timestamping info
2847  * @tx_flags:	completed with instructions for time stamping
2848  * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2849  *
2850  * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2851  */
2852 static inline void _sock_tx_timestamp(struct sock *sk,
2853 				      const struct sockcm_cookie *sockc,
2854 				      __u8 *tx_flags, __u32 *tskey)
2855 {
2856 	__u32 tsflags = sockc->tsflags;
2857 
2858 	if (unlikely(tsflags)) {
2859 		__sock_tx_timestamp(tsflags, tx_flags);
2860 		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2861 		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) {
2862 			if (tsflags & SOCKCM_FLAG_TS_OPT_ID)
2863 				*tskey = sockc->ts_opt_id;
2864 			else
2865 				*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2866 		}
2867 	}
2868 }
2869 
2870 static inline void sock_tx_timestamp(struct sock *sk,
2871 				     const struct sockcm_cookie *sockc,
2872 				     __u8 *tx_flags)
2873 {
2874 	_sock_tx_timestamp(sk, sockc, tx_flags, NULL);
2875 }
2876 
2877 static inline void skb_setup_tx_timestamp(struct sk_buff *skb,
2878 					  const struct sockcm_cookie *sockc)
2879 {
2880 	_sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags,
2881 			   &skb_shinfo(skb)->tskey);
2882 }
2883 
2884 static inline bool sk_is_inet(const struct sock *sk)
2885 {
2886 	int family = READ_ONCE(sk->sk_family);
2887 
2888 	return family == AF_INET || family == AF_INET6;
2889 }
2890 
2891 static inline bool sk_is_tcp(const struct sock *sk)
2892 {
2893 	return sk_is_inet(sk) &&
2894 	       sk->sk_type == SOCK_STREAM &&
2895 	       sk->sk_protocol == IPPROTO_TCP;
2896 }
2897 
2898 static inline bool sk_is_udp(const struct sock *sk)
2899 {
2900 	return sk_is_inet(sk) &&
2901 	       sk->sk_type == SOCK_DGRAM &&
2902 	       sk->sk_protocol == IPPROTO_UDP;
2903 }
2904 
2905 static inline bool sk_is_unix(const struct sock *sk)
2906 {
2907 	return sk->sk_family == AF_UNIX;
2908 }
2909 
2910 static inline bool sk_is_stream_unix(const struct sock *sk)
2911 {
2912 	return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM;
2913 }
2914 
2915 static inline bool sk_is_vsock(const struct sock *sk)
2916 {
2917 	return sk->sk_family == AF_VSOCK;
2918 }
2919 
2920 static inline bool sk_may_scm_recv(const struct sock *sk)
2921 {
2922 	return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) ||
2923 		sk->sk_family == AF_NETLINK ||
2924 		(IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH);
2925 }
2926 
2927 /**
2928  * sk_eat_skb - Release a skb if it is no longer needed
2929  * @sk: socket to eat this skb from
2930  * @skb: socket buffer to eat
2931  *
2932  * This routine must be called with interrupts disabled or with the socket
2933  * locked so that the sk_buff queue operation is ok.
2934 */
2935 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2936 {
2937 	__skb_unlink(skb, &sk->sk_receive_queue);
2938 	__kfree_skb(skb);
2939 }
2940 
2941 static inline bool
2942 skb_sk_is_prefetched(struct sk_buff *skb)
2943 {
2944 #ifdef CONFIG_INET
2945 	return skb->destructor == sock_pfree;
2946 #else
2947 	return false;
2948 #endif /* CONFIG_INET */
2949 }
2950 
2951 /* This helper checks if a socket is a full socket,
2952  * ie _not_ a timewait or request socket.
2953  */
2954 static inline bool sk_fullsock(const struct sock *sk)
2955 {
2956 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2957 }
2958 
2959 static inline bool
2960 sk_is_refcounted(struct sock *sk)
2961 {
2962 	/* Only full sockets have sk->sk_flags. */
2963 	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2964 }
2965 
2966 static inline bool
2967 sk_requests_wifi_status(struct sock *sk)
2968 {
2969 	return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS);
2970 }
2971 
2972 /* Checks if this SKB belongs to an HW offloaded socket
2973  * and whether any SW fallbacks are required based on dev.
2974  * Check decrypted mark in case skb_orphan() cleared socket.
2975  */
2976 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2977 						   struct net_device *dev)
2978 {
2979 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2980 	struct sock *sk = skb->sk;
2981 
2982 	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2983 		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2984 	} else if (unlikely(skb_is_decrypted(skb))) {
2985 		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2986 		kfree_skb(skb);
2987 		skb = NULL;
2988 	}
2989 #endif
2990 
2991 	return skb;
2992 }
2993 
2994 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2995  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2996  */
2997 static inline bool sk_listener(const struct sock *sk)
2998 {
2999 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
3000 }
3001 
3002 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT
3003  * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE)
3004  * TCP RST and ACK can be attached to TIME_WAIT.
3005  */
3006 static inline bool sk_listener_or_tw(const struct sock *sk)
3007 {
3008 	return (1 << READ_ONCE(sk->sk_state)) &
3009 	       (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT);
3010 }
3011 
3012 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
3013 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
3014 		       int type);
3015 
3016 bool sk_ns_capable(const struct sock *sk,
3017 		   struct user_namespace *user_ns, int cap);
3018 bool sk_capable(const struct sock *sk, int cap);
3019 bool sk_net_capable(const struct sock *sk, int cap);
3020 
3021 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
3022 
3023 /* Take into consideration the size of the struct sk_buff overhead in the
3024  * determination of these values, since that is non-constant across
3025  * platforms.  This makes socket queueing behavior and performance
3026  * not depend upon such differences.
3027  */
3028 #define _SK_MEM_PACKETS		256
3029 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
3030 #define SK_WMEM_DEFAULT		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3031 #define SK_RMEM_DEFAULT		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3032 
3033 extern __u32 sysctl_wmem_max;
3034 extern __u32 sysctl_rmem_max;
3035 
3036 extern __u32 sysctl_wmem_default;
3037 extern __u32 sysctl_rmem_default;
3038 
3039 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
3040 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3041 
3042 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
3043 {
3044 	/* Does this proto have per netns sysctl_wmem ? */
3045 	if (proto->sysctl_wmem_offset)
3046 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
3047 
3048 	return READ_ONCE(*proto->sysctl_wmem);
3049 }
3050 
3051 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
3052 {
3053 	/* Does this proto have per netns sysctl_rmem ? */
3054 	if (proto->sysctl_rmem_offset)
3055 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
3056 
3057 	return READ_ONCE(*proto->sysctl_rmem);
3058 }
3059 
3060 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
3061  * Some wifi drivers need to tweak it to get more chunks.
3062  * They can use this helper from their ndo_start_xmit()
3063  */
3064 static inline void sk_pacing_shift_update(struct sock *sk, int val)
3065 {
3066 	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
3067 		return;
3068 	WRITE_ONCE(sk->sk_pacing_shift, val);
3069 }
3070 
3071 /* if a socket is bound to a device, check that the given device
3072  * index is either the same or that the socket is bound to an L3
3073  * master device and the given device index is also enslaved to
3074  * that L3 master
3075  */
3076 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
3077 {
3078 	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
3079 	int mdif;
3080 
3081 	if (!bound_dev_if || bound_dev_if == dif)
3082 		return true;
3083 
3084 	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
3085 	if (mdif && mdif == bound_dev_if)
3086 		return true;
3087 
3088 	return false;
3089 }
3090 
3091 void sock_def_readable(struct sock *sk);
3092 
3093 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
3094 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
3095 int sock_set_timestamping(struct sock *sk, int optname,
3096 			  struct so_timestamping timestamping);
3097 
3098 #if defined(CONFIG_CGROUP_BPF)
3099 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op);
3100 #else
3101 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
3102 {
3103 }
3104 #endif
3105 void sock_no_linger(struct sock *sk);
3106 void sock_set_keepalive(struct sock *sk);
3107 void sock_set_priority(struct sock *sk, u32 priority);
3108 void sock_set_rcvbuf(struct sock *sk, int val);
3109 void sock_set_mark(struct sock *sk, u32 val);
3110 void sock_set_reuseaddr(struct sock *sk);
3111 void sock_set_reuseport(struct sock *sk);
3112 void sock_set_sndtimeo(struct sock *sk, s64 secs);
3113 
3114 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
3115 
3116 int sock_get_timeout(long timeo, void *optval, bool old_timeval);
3117 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
3118 			   sockptr_t optval, int optlen, bool old_timeval);
3119 
3120 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
3121 		     void __user *arg, void *karg, size_t size);
3122 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
3123 static inline bool sk_is_readable(struct sock *sk)
3124 {
3125 	const struct proto *prot = READ_ONCE(sk->sk_prot);
3126 
3127 	if (prot->sock_is_readable)
3128 		return prot->sock_is_readable(sk);
3129 
3130 	return false;
3131 }
3132 #endif	/* _SOCK_H */
3133