xref: /linux/include/net/sock.h (revision 515c0ead788f4118a91b3ae55fe51f95543553ec)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Definitions for the AF_INET socket handler.
8  *
9  * Version:	@(#)sock.h	1.0.4	05/13/93
10  *
11  * Authors:	Ross Biro
12  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13  *		Corey Minyard <wf-rch!minyard@relay.EU.net>
14  *		Florian La Roche <flla@stud.uni-sb.de>
15  *
16  * Fixes:
17  *		Alan Cox	:	Volatiles in skbuff pointers. See
18  *					skbuff comments. May be overdone,
19  *					better to prove they can be removed
20  *					than the reverse.
21  *		Alan Cox	:	Added a zapped field for tcp to note
22  *					a socket is reset and must stay shut up
23  *		Alan Cox	:	New fields for options
24  *	Pauline Middelink	:	identd support
25  *		Alan Cox	:	Eliminate low level recv/recvfrom
26  *		David S. Miller	:	New socket lookup architecture.
27  *              Steve Whitehouse:       Default routines for sock_ops
28  *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
29  *              			protinfo be just a void pointer, as the
30  *              			protocol specific parts were moved to
31  *              			respective headers and ipv4/v6, etc now
32  *              			use private slabcaches for its socks
33  *              Pedro Hortas	:	New flags field for socket options
34  */
35 #ifndef _SOCK_H
36 #define _SOCK_H
37 
38 #include <linux/hardirq.h>
39 #include <linux/kernel.h>
40 #include <linux/list.h>
41 #include <linux/list_nulls.h>
42 #include <linux/timer.h>
43 #include <linux/cache.h>
44 #include <linux/bitops.h>
45 #include <linux/lockdep.h>
46 #include <linux/netdevice.h>
47 #include <linux/skbuff.h>	/* struct sk_buff */
48 #include <linux/mm.h>
49 #include <linux/security.h>
50 #include <linux/slab.h>
51 #include <linux/uaccess.h>
52 #include <linux/page_counter.h>
53 #include <linux/memcontrol.h>
54 #include <linux/static_key.h>
55 #include <linux/sched.h>
56 #include <linux/wait.h>
57 #include <linux/cgroup-defs.h>
58 #include <linux/rbtree.h>
59 #include <linux/rculist_nulls.h>
60 #include <linux/poll.h>
61 #include <linux/sockptr.h>
62 #include <linux/indirect_call_wrapper.h>
63 #include <linux/atomic.h>
64 #include <linux/refcount.h>
65 #include <linux/llist.h>
66 #include <net/dst.h>
67 #include <net/checksum.h>
68 #include <net/tcp_states.h>
69 #include <linux/net_tstamp.h>
70 #include <net/l3mdev.h>
71 #include <uapi/linux/socket.h>
72 
73 /*
74  * This structure really needs to be cleaned up.
75  * Most of it is for TCP, and not used by any of
76  * the other protocols.
77  */
78 
79 /* This is the per-socket lock.  The spinlock provides a synchronization
80  * between user contexts and software interrupt processing, whereas the
81  * mini-semaphore synchronizes multiple users amongst themselves.
82  */
83 typedef struct {
84 	spinlock_t		slock;
85 	int			owned;
86 	wait_queue_head_t	wq;
87 	/*
88 	 * We express the mutex-alike socket_lock semantics
89 	 * to the lock validator by explicitly managing
90 	 * the slock as a lock variant (in addition to
91 	 * the slock itself):
92 	 */
93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
94 	struct lockdep_map dep_map;
95 #endif
96 } socket_lock_t;
97 
98 struct sock;
99 struct proto;
100 struct net;
101 
102 typedef __u32 __bitwise __portpair;
103 typedef __u64 __bitwise __addrpair;
104 
105 /**
106  *	struct sock_common - minimal network layer representation of sockets
107  *	@skc_daddr: Foreign IPv4 addr
108  *	@skc_rcv_saddr: Bound local IPv4 addr
109  *	@skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
110  *	@skc_hash: hash value used with various protocol lookup tables
111  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
112  *	@skc_dport: placeholder for inet_dport/tw_dport
113  *	@skc_num: placeholder for inet_num/tw_num
114  *	@skc_portpair: __u32 union of @skc_dport & @skc_num
115  *	@skc_family: network address family
116  *	@skc_state: Connection state
117  *	@skc_reuse: %SO_REUSEADDR setting
118  *	@skc_reuseport: %SO_REUSEPORT setting
119  *	@skc_ipv6only: socket is IPV6 only
120  *	@skc_net_refcnt: socket is using net ref counting
121  *	@skc_bound_dev_if: bound device index if != 0
122  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
123  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
124  *	@skc_prot: protocol handlers inside a network family
125  *	@skc_net: reference to the network namespace of this socket
126  *	@skc_v6_daddr: IPV6 destination address
127  *	@skc_v6_rcv_saddr: IPV6 source address
128  *	@skc_cookie: socket's cookie value
129  *	@skc_node: main hash linkage for various protocol lookup tables
130  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
131  *	@skc_tx_queue_mapping: tx queue number for this connection
132  *	@skc_rx_queue_mapping: rx queue number for this connection
133  *	@skc_flags: place holder for sk_flags
134  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
135  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
136  *	@skc_listener: connection request listener socket (aka rsk_listener)
137  *		[union with @skc_flags]
138  *	@skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
139  *		[union with @skc_flags]
140  *	@skc_incoming_cpu: record/match cpu processing incoming packets
141  *	@skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
142  *		[union with @skc_incoming_cpu]
143  *	@skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
144  *		[union with @skc_incoming_cpu]
145  *	@skc_refcnt: reference count
146  *
147  *	This is the minimal network layer representation of sockets, the header
148  *	for struct sock and struct inet_timewait_sock.
149  */
150 struct sock_common {
151 	union {
152 		__addrpair	skc_addrpair;
153 		struct {
154 			__be32	skc_daddr;
155 			__be32	skc_rcv_saddr;
156 		};
157 	};
158 	union  {
159 		unsigned int	skc_hash;
160 		__u16		skc_u16hashes[2];
161 	};
162 	/* skc_dport && skc_num must be grouped as well */
163 	union {
164 		__portpair	skc_portpair;
165 		struct {
166 			__be16	skc_dport;
167 			__u16	skc_num;
168 		};
169 	};
170 
171 	unsigned short		skc_family;
172 	volatile unsigned char	skc_state;
173 	unsigned char		skc_reuse:4;
174 	unsigned char		skc_reuseport:1;
175 	unsigned char		skc_ipv6only:1;
176 	unsigned char		skc_net_refcnt:1;
177 	int			skc_bound_dev_if;
178 	union {
179 		struct hlist_node	skc_bind_node;
180 		struct hlist_node	skc_portaddr_node;
181 	};
182 	struct proto		*skc_prot;
183 	possible_net_t		skc_net;
184 
185 #if IS_ENABLED(CONFIG_IPV6)
186 	struct in6_addr		skc_v6_daddr;
187 	struct in6_addr		skc_v6_rcv_saddr;
188 #endif
189 
190 	atomic64_t		skc_cookie;
191 
192 	/* following fields are padding to force
193 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
194 	 * assuming IPV6 is enabled. We use this padding differently
195 	 * for different kind of 'sockets'
196 	 */
197 	union {
198 		unsigned long	skc_flags;
199 		struct sock	*skc_listener; /* request_sock */
200 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
201 	};
202 	/*
203 	 * fields between dontcopy_begin/dontcopy_end
204 	 * are not copied in sock_copy()
205 	 */
206 	/* private: */
207 	int			skc_dontcopy_begin[0];
208 	/* public: */
209 	union {
210 		struct hlist_node	skc_node;
211 		struct hlist_nulls_node skc_nulls_node;
212 	};
213 	unsigned short		skc_tx_queue_mapping;
214 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
215 	unsigned short		skc_rx_queue_mapping;
216 #endif
217 	union {
218 		int		skc_incoming_cpu;
219 		u32		skc_rcv_wnd;
220 		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
221 	};
222 
223 	refcount_t		skc_refcnt;
224 	/* private: */
225 	int                     skc_dontcopy_end[0];
226 	union {
227 		u32		skc_rxhash;
228 		u32		skc_window_clamp;
229 		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
230 	};
231 	/* public: */
232 };
233 
234 struct bpf_local_storage;
235 struct sk_filter;
236 
237 /**
238   *	struct sock - network layer representation of sockets
239   *	@__sk_common: shared layout with inet_timewait_sock
240   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
241   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
242   *	@sk_lock:	synchronizer
243   *	@sk_kern_sock: True if sock is using kernel lock classes
244   *	@sk_rcvbuf: size of receive buffer in bytes
245   *	@sk_wq: sock wait queue and async head
246   *	@sk_rx_dst: receive input route used by early demux
247   *	@sk_rx_dst_ifindex: ifindex for @sk_rx_dst
248   *	@sk_rx_dst_cookie: cookie for @sk_rx_dst
249   *	@sk_dst_cache: destination cache
250   *	@sk_dst_pending_confirm: need to confirm neighbour
251   *	@sk_policy: flow policy
252   *	@sk_receive_queue: incoming packets
253   *	@sk_wmem_alloc: transmit queue bytes committed
254   *	@sk_tsq_flags: TCP Small Queues flags
255   *	@sk_write_queue: Packet sending queue
256   *	@sk_omem_alloc: "o" is "option" or "other"
257   *	@sk_wmem_queued: persistent queue size
258   *	@sk_forward_alloc: space allocated forward
259   *	@sk_reserved_mem: space reserved and non-reclaimable for the socket
260   *	@sk_napi_id: id of the last napi context to receive data for sk
261   *	@sk_ll_usec: usecs to busypoll when there is no data
262   *	@sk_allocation: allocation mode
263   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
264   *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
265   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
266   *	@sk_sndbuf: size of send buffer in bytes
267   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
268   *	@sk_no_check_rx: allow zero checksum in RX packets
269   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
270   *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
271   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
272   *	@sk_gso_max_size: Maximum GSO segment size to build
273   *	@sk_gso_max_segs: Maximum number of GSO segments
274   *	@sk_pacing_shift: scaling factor for TCP Small Queues
275   *	@sk_lingertime: %SO_LINGER l_linger setting
276   *	@sk_backlog: always used with the per-socket spinlock held
277   *	@sk_callback_lock: used with the callbacks in the end of this struct
278   *	@sk_error_queue: rarely used
279   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
280   *			  IPV6_ADDRFORM for instance)
281   *	@sk_err: last error
282   *	@sk_err_soft: errors that don't cause failure but are the cause of a
283   *		      persistent failure not just 'timed out'
284   *	@sk_drops: raw/udp drops counter
285   *	@sk_drop_counters: optional pointer to numa_drop_counters
286   *	@sk_ack_backlog: current listen backlog
287   *	@sk_max_ack_backlog: listen backlog set in listen()
288   *	@sk_uid: user id of owner
289   *	@sk_ino: inode number (zero if orphaned)
290   *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
291   *	@sk_busy_poll_budget: napi processing budget when busypolling
292   *	@sk_priority: %SO_PRIORITY setting
293   *	@sk_type: socket type (%SOCK_STREAM, etc)
294   *	@sk_protocol: which protocol this socket belongs in this network family
295   *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
296   *	@sk_peer_pid: &struct pid for this socket's peer
297   *	@sk_peer_cred: %SO_PEERCRED setting
298   *	@sk_rcvlowat: %SO_RCVLOWAT setting
299   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
300   *	@sk_sndtimeo: %SO_SNDTIMEO setting
301   *	@sk_txhash: computed flow hash for use on transmit
302   *	@sk_txrehash: enable TX hash rethink
303   *	@sk_filter: socket filtering instructions
304   *	@sk_timer: sock cleanup timer
305   *	@sk_stamp: time stamp of last packet received
306   *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
307   *	@sk_tsflags: SO_TIMESTAMPING flags
308   *	@sk_bpf_cb_flags: used in bpf_setsockopt()
309   *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
310   *			   Sockets that can be used under memory reclaim should
311   *			   set this to false.
312   *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
313   *	              for timestamping
314   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
315   *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
316   *	@sk_socket: Identd and reporting IO signals
317   *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
318   *	@sk_frag: cached page frag
319   *	@sk_peek_off: current peek_offset value
320   *	@sk_send_head: front of stuff to transmit
321   *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
322   *	@sk_security: used by security modules
323   *	@sk_mark: generic packet mark
324   *	@sk_cgrp_data: cgroup data for this cgroup
325   *	@sk_memcg: this socket's memory cgroup association
326   *	@sk_write_pending: a write to stream socket waits to start
327   *	@sk_disconnects: number of disconnect operations performed on this sock
328   *	@sk_state_change: callback to indicate change in the state of the sock
329   *	@sk_data_ready: callback to indicate there is data to be processed
330   *	@sk_write_space: callback to indicate there is bf sending space available
331   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
332   *	@sk_backlog_rcv: callback to process the backlog
333   *	@sk_validate_xmit_skb: ptr to an optional validate function
334   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
335   *	@sk_reuseport_cb: reuseport group container
336   *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
337   *	@sk_rcu: used during RCU grace period
338   *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
339   *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
340   *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
341   *	@sk_txtime_unused: unused txtime flags
342   *	@sk_scm_recv_flags: all flags used by scm_recv()
343   *	@sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS
344   *	@sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY
345   *	@sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD
346   *	@sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS
347   *	@sk_scm_unused: unused flags for scm_recv()
348   *	@ns_tracker: tracker for netns reference
349   *	@sk_user_frags: xarray of pages the user is holding a reference on.
350   *	@sk_owner: reference to the real owner of the socket that calls
351   *		   sock_lock_init_class_and_name().
352   */
353 struct sock {
354 	/*
355 	 * Now struct inet_timewait_sock also uses sock_common, so please just
356 	 * don't add nothing before this first member (__sk_common) --acme
357 	 */
358 	struct sock_common	__sk_common;
359 #define sk_node			__sk_common.skc_node
360 #define sk_nulls_node		__sk_common.skc_nulls_node
361 #define sk_refcnt		__sk_common.skc_refcnt
362 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
363 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
364 #define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
365 #endif
366 
367 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
368 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
369 #define sk_hash			__sk_common.skc_hash
370 #define sk_portpair		__sk_common.skc_portpair
371 #define sk_num			__sk_common.skc_num
372 #define sk_dport		__sk_common.skc_dport
373 #define sk_addrpair		__sk_common.skc_addrpair
374 #define sk_daddr		__sk_common.skc_daddr
375 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
376 #define sk_family		__sk_common.skc_family
377 #define sk_state		__sk_common.skc_state
378 #define sk_reuse		__sk_common.skc_reuse
379 #define sk_reuseport		__sk_common.skc_reuseport
380 #define sk_ipv6only		__sk_common.skc_ipv6only
381 #define sk_net_refcnt		__sk_common.skc_net_refcnt
382 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
383 #define sk_bind_node		__sk_common.skc_bind_node
384 #define sk_prot			__sk_common.skc_prot
385 #define sk_net			__sk_common.skc_net
386 #define sk_v6_daddr		__sk_common.skc_v6_daddr
387 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
388 #define sk_cookie		__sk_common.skc_cookie
389 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
390 #define sk_flags		__sk_common.skc_flags
391 #define sk_rxhash		__sk_common.skc_rxhash
392 
393 	__cacheline_group_begin(sock_write_rx);
394 
395 	atomic_t		sk_drops;
396 	__s32			sk_peek_off;
397 	struct sk_buff_head	sk_error_queue;
398 	struct sk_buff_head	sk_receive_queue;
399 	/*
400 	 * The backlog queue is special, it is always used with
401 	 * the per-socket spinlock held and requires low latency
402 	 * access. Therefore we special case it's implementation.
403 	 * Note : rmem_alloc is in this structure to fill a hole
404 	 * on 64bit arches, not because its logically part of
405 	 * backlog.
406 	 */
407 	struct {
408 		atomic_t	rmem_alloc;
409 		int		len;
410 		struct sk_buff	*head;
411 		struct sk_buff	*tail;
412 	} sk_backlog;
413 #define sk_rmem_alloc sk_backlog.rmem_alloc
414 
415 	__cacheline_group_end(sock_write_rx);
416 
417 	__cacheline_group_begin(sock_read_rx);
418 	/* early demux fields */
419 	struct dst_entry __rcu	*sk_rx_dst;
420 	int			sk_rx_dst_ifindex;
421 	u32			sk_rx_dst_cookie;
422 
423 #ifdef CONFIG_NET_RX_BUSY_POLL
424 	unsigned int		sk_ll_usec;
425 	unsigned int		sk_napi_id;
426 	u16			sk_busy_poll_budget;
427 	u8			sk_prefer_busy_poll;
428 #endif
429 	u8			sk_userlocks;
430 	int			sk_rcvbuf;
431 
432 	struct sk_filter __rcu	*sk_filter;
433 	union {
434 		struct socket_wq __rcu	*sk_wq;
435 		/* private: */
436 		struct socket_wq	*sk_wq_raw;
437 		/* public: */
438 	};
439 
440 	void			(*sk_data_ready)(struct sock *sk);
441 	long			sk_rcvtimeo;
442 	int			sk_rcvlowat;
443 	__cacheline_group_end(sock_read_rx);
444 
445 	__cacheline_group_begin(sock_read_rxtx);
446 	int			sk_err;
447 	struct socket		*sk_socket;
448 #ifdef CONFIG_MEMCG
449 	struct mem_cgroup	*sk_memcg;
450 #endif
451 #ifdef CONFIG_XFRM
452 	struct xfrm_policy __rcu *sk_policy[2];
453 #endif
454 	struct numa_drop_counters *sk_drop_counters;
455 	__cacheline_group_end(sock_read_rxtx);
456 
457 	__cacheline_group_begin(sock_write_rxtx);
458 	socket_lock_t		sk_lock;
459 	u32			sk_reserved_mem;
460 	int			sk_forward_alloc;
461 	u32			sk_tsflags;
462 	__cacheline_group_end(sock_write_rxtx);
463 
464 	__cacheline_group_begin(sock_write_tx);
465 	int			sk_write_pending;
466 	atomic_t		sk_omem_alloc;
467 	int			sk_sndbuf;
468 
469 	int			sk_wmem_queued;
470 	refcount_t		sk_wmem_alloc;
471 	unsigned long		sk_tsq_flags;
472 	union {
473 		struct sk_buff	*sk_send_head;
474 		struct rb_root	tcp_rtx_queue;
475 	};
476 	struct sk_buff_head	sk_write_queue;
477 	u32			sk_dst_pending_confirm;
478 	u32			sk_pacing_status; /* see enum sk_pacing */
479 	struct page_frag	sk_frag;
480 	struct timer_list	sk_timer;
481 
482 	unsigned long		sk_pacing_rate; /* bytes per second */
483 	atomic_t		sk_zckey;
484 	atomic_t		sk_tskey;
485 	__cacheline_group_end(sock_write_tx);
486 
487 	__cacheline_group_begin(sock_read_tx);
488 	unsigned long		sk_max_pacing_rate;
489 	long			sk_sndtimeo;
490 	u32			sk_priority;
491 	u32			sk_mark;
492 	struct dst_entry __rcu	*sk_dst_cache;
493 	netdev_features_t	sk_route_caps;
494 #ifdef CONFIG_SOCK_VALIDATE_XMIT
495 	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
496 							struct net_device *dev,
497 							struct sk_buff *skb);
498 #endif
499 	u16			sk_gso_type;
500 	u16			sk_gso_max_segs;
501 	unsigned int		sk_gso_max_size;
502 	gfp_t			sk_allocation;
503 	u32			sk_txhash;
504 	u8			sk_pacing_shift;
505 	bool			sk_use_task_frag;
506 	__cacheline_group_end(sock_read_tx);
507 
508 	/*
509 	 * Because of non atomicity rules, all
510 	 * changes are protected by socket lock.
511 	 */
512 	u8			sk_gso_disabled : 1,
513 				sk_kern_sock : 1,
514 				sk_no_check_tx : 1,
515 				sk_no_check_rx : 1;
516 	u8			sk_shutdown;
517 	u16			sk_type;
518 	u16			sk_protocol;
519 	unsigned long	        sk_lingertime;
520 	struct proto		*sk_prot_creator;
521 	rwlock_t		sk_callback_lock;
522 	int			sk_err_soft;
523 	u32			sk_ack_backlog;
524 	u32			sk_max_ack_backlog;
525 	kuid_t			sk_uid;
526 	unsigned long		sk_ino;
527 	spinlock_t		sk_peer_lock;
528 	int			sk_bind_phc;
529 	struct pid		*sk_peer_pid;
530 	const struct cred	*sk_peer_cred;
531 
532 	ktime_t			sk_stamp;
533 #if BITS_PER_LONG==32
534 	seqlock_t		sk_stamp_seq;
535 #endif
536 	int			sk_disconnects;
537 
538 	union {
539 		u8		sk_txrehash;
540 		u8		sk_scm_recv_flags;
541 		struct {
542 			u8	sk_scm_credentials : 1,
543 				sk_scm_security : 1,
544 				sk_scm_pidfd : 1,
545 				sk_scm_rights : 1,
546 				sk_scm_unused : 4;
547 		};
548 	};
549 	u8			sk_clockid;
550 	u8			sk_txtime_deadline_mode : 1,
551 				sk_txtime_report_errors : 1,
552 				sk_txtime_unused : 6;
553 #define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG))
554 	u8			sk_bpf_cb_flags;
555 
556 	void			*sk_user_data;
557 #ifdef CONFIG_SECURITY
558 	void			*sk_security;
559 #endif
560 	struct sock_cgroup_data	sk_cgrp_data;
561 	void			(*sk_state_change)(struct sock *sk);
562 	void			(*sk_write_space)(struct sock *sk);
563 	void			(*sk_error_report)(struct sock *sk);
564 	int			(*sk_backlog_rcv)(struct sock *sk,
565 						  struct sk_buff *skb);
566 	void                    (*sk_destruct)(struct sock *sk);
567 	struct sock_reuseport __rcu	*sk_reuseport_cb;
568 #ifdef CONFIG_BPF_SYSCALL
569 	struct bpf_local_storage __rcu	*sk_bpf_storage;
570 #endif
571 	struct rcu_head		sk_rcu;
572 	netns_tracker		ns_tracker;
573 	struct xarray		sk_user_frags;
574 
575 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
576 	struct module		*sk_owner;
577 #endif
578 };
579 
580 struct sock_bh_locked {
581 	struct sock *sock;
582 	local_lock_t bh_lock;
583 };
584 
585 enum sk_pacing {
586 	SK_PACING_NONE		= 0,
587 	SK_PACING_NEEDED	= 1,
588 	SK_PACING_FQ		= 2,
589 };
590 
591 /* flag bits in sk_user_data
592  *
593  * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
594  *   not be suitable for copying when cloning the socket. For instance,
595  *   it can point to a reference counted object. sk_user_data bottom
596  *   bit is set if pointer must not be copied.
597  *
598  * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
599  *   managed/owned by a BPF reuseport array. This bit should be set
600  *   when sk_user_data's sk is added to the bpf's reuseport_array.
601  *
602  * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
603  *   sk_user_data points to psock type. This bit should be set
604  *   when sk_user_data is assigned to a psock object.
605  */
606 #define SK_USER_DATA_NOCOPY	1UL
607 #define SK_USER_DATA_BPF	2UL
608 #define SK_USER_DATA_PSOCK	4UL
609 #define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
610 				  SK_USER_DATA_PSOCK)
611 
612 /**
613  * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
614  * @sk: socket
615  */
616 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
617 {
618 	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
619 }
620 
621 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
622 
623 /**
624  * __locked_read_sk_user_data_with_flags - return the pointer
625  * only if argument flags all has been set in sk_user_data. Otherwise
626  * return NULL
627  *
628  * @sk: socket
629  * @flags: flag bits
630  *
631  * The caller must be holding sk->sk_callback_lock.
632  */
633 static inline void *
634 __locked_read_sk_user_data_with_flags(const struct sock *sk,
635 				      uintptr_t flags)
636 {
637 	uintptr_t sk_user_data =
638 		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
639 						 lockdep_is_held(&sk->sk_callback_lock));
640 
641 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
642 
643 	if ((sk_user_data & flags) == flags)
644 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
645 	return NULL;
646 }
647 
648 /**
649  * __rcu_dereference_sk_user_data_with_flags - return the pointer
650  * only if argument flags all has been set in sk_user_data. Otherwise
651  * return NULL
652  *
653  * @sk: socket
654  * @flags: flag bits
655  */
656 static inline void *
657 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
658 					  uintptr_t flags)
659 {
660 	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
661 
662 	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
663 
664 	if ((sk_user_data & flags) == flags)
665 		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
666 	return NULL;
667 }
668 
669 #define rcu_dereference_sk_user_data(sk)				\
670 	__rcu_dereference_sk_user_data_with_flags(sk, 0)
671 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
672 ({									\
673 	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
674 		  __tmp2 = (uintptr_t)(flags);				\
675 	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
676 	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
677 	rcu_assign_pointer(__sk_user_data((sk)),			\
678 			   __tmp1 | __tmp2);				\
679 })
680 #define rcu_assign_sk_user_data(sk, ptr)				\
681 	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
682 
683 static inline
684 struct net *sock_net(const struct sock *sk)
685 {
686 	return read_pnet(&sk->sk_net);
687 }
688 
689 static inline
690 void sock_net_set(struct sock *sk, struct net *net)
691 {
692 	write_pnet(&sk->sk_net, net);
693 }
694 
695 /*
696  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
697  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
698  * on a socket means that the socket will reuse everybody else's port
699  * without looking at the other's sk_reuse value.
700  */
701 
702 #define SK_NO_REUSE	0
703 #define SK_CAN_REUSE	1
704 #define SK_FORCE_REUSE	2
705 
706 int sk_set_peek_off(struct sock *sk, int val);
707 
708 static inline int sk_peek_offset(const struct sock *sk, int flags)
709 {
710 	if (unlikely(flags & MSG_PEEK)) {
711 		return READ_ONCE(sk->sk_peek_off);
712 	}
713 
714 	return 0;
715 }
716 
717 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
718 {
719 	s32 off = READ_ONCE(sk->sk_peek_off);
720 
721 	if (unlikely(off >= 0)) {
722 		off = max_t(s32, off - val, 0);
723 		WRITE_ONCE(sk->sk_peek_off, off);
724 	}
725 }
726 
727 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
728 {
729 	sk_peek_offset_bwd(sk, -val);
730 }
731 
732 /*
733  * Hashed lists helper routines
734  */
735 static inline struct sock *sk_entry(const struct hlist_node *node)
736 {
737 	return hlist_entry(node, struct sock, sk_node);
738 }
739 
740 static inline struct sock *__sk_head(const struct hlist_head *head)
741 {
742 	return hlist_entry(head->first, struct sock, sk_node);
743 }
744 
745 static inline struct sock *sk_head(const struct hlist_head *head)
746 {
747 	return hlist_empty(head) ? NULL : __sk_head(head);
748 }
749 
750 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
751 {
752 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
753 }
754 
755 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
756 {
757 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
758 }
759 
760 static inline struct sock *sk_next(const struct sock *sk)
761 {
762 	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
763 }
764 
765 static inline struct sock *sk_nulls_next(const struct sock *sk)
766 {
767 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
768 		hlist_nulls_entry(sk->sk_nulls_node.next,
769 				  struct sock, sk_nulls_node) :
770 		NULL;
771 }
772 
773 static inline bool sk_unhashed(const struct sock *sk)
774 {
775 	return hlist_unhashed(&sk->sk_node);
776 }
777 
778 static inline bool sk_hashed(const struct sock *sk)
779 {
780 	return !sk_unhashed(sk);
781 }
782 
783 static inline void sk_node_init(struct hlist_node *node)
784 {
785 	node->pprev = NULL;
786 }
787 
788 static inline void __sk_del_node(struct sock *sk)
789 {
790 	__hlist_del(&sk->sk_node);
791 }
792 
793 /* NB: equivalent to hlist_del_init_rcu */
794 static inline bool __sk_del_node_init(struct sock *sk)
795 {
796 	if (sk_hashed(sk)) {
797 		__sk_del_node(sk);
798 		sk_node_init(&sk->sk_node);
799 		return true;
800 	}
801 	return false;
802 }
803 
804 /* Grab socket reference count. This operation is valid only
805    when sk is ALREADY grabbed f.e. it is found in hash table
806    or a list and the lookup is made under lock preventing hash table
807    modifications.
808  */
809 
810 static __always_inline void sock_hold(struct sock *sk)
811 {
812 	refcount_inc(&sk->sk_refcnt);
813 }
814 
815 /* Ungrab socket in the context, which assumes that socket refcnt
816    cannot hit zero, f.e. it is true in context of any socketcall.
817  */
818 static __always_inline void __sock_put(struct sock *sk)
819 {
820 	refcount_dec(&sk->sk_refcnt);
821 }
822 
823 static inline bool sk_del_node_init(struct sock *sk)
824 {
825 	bool rc = __sk_del_node_init(sk);
826 
827 	if (rc) {
828 		/* paranoid for a while -acme */
829 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
830 		__sock_put(sk);
831 	}
832 	return rc;
833 }
834 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
835 
836 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
837 {
838 	if (sk_hashed(sk)) {
839 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
840 		return true;
841 	}
842 	return false;
843 }
844 
845 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
846 {
847 	bool rc = __sk_nulls_del_node_init_rcu(sk);
848 
849 	if (rc) {
850 		/* paranoid for a while -acme */
851 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
852 		__sock_put(sk);
853 	}
854 	return rc;
855 }
856 
857 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
858 {
859 	hlist_add_head(&sk->sk_node, list);
860 }
861 
862 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
863 {
864 	sock_hold(sk);
865 	__sk_add_node(sk, list);
866 }
867 
868 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
869 {
870 	sock_hold(sk);
871 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
872 	    sk->sk_family == AF_INET6)
873 		hlist_add_tail_rcu(&sk->sk_node, list);
874 	else
875 		hlist_add_head_rcu(&sk->sk_node, list);
876 }
877 
878 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
879 {
880 	sock_hold(sk);
881 	hlist_add_tail_rcu(&sk->sk_node, list);
882 }
883 
884 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
885 {
886 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
887 }
888 
889 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
890 {
891 	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
892 }
893 
894 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
895 {
896 	sock_hold(sk);
897 	__sk_nulls_add_node_rcu(sk, list);
898 }
899 
900 static inline void __sk_del_bind_node(struct sock *sk)
901 {
902 	__hlist_del(&sk->sk_bind_node);
903 }
904 
905 static inline void sk_add_bind_node(struct sock *sk,
906 					struct hlist_head *list)
907 {
908 	hlist_add_head(&sk->sk_bind_node, list);
909 }
910 
911 #define sk_for_each(__sk, list) \
912 	hlist_for_each_entry(__sk, list, sk_node)
913 #define sk_for_each_rcu(__sk, list) \
914 	hlist_for_each_entry_rcu(__sk, list, sk_node)
915 #define sk_nulls_for_each(__sk, node, list) \
916 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
917 #define sk_nulls_for_each_rcu(__sk, node, list) \
918 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
919 #define sk_for_each_from(__sk) \
920 	hlist_for_each_entry_from(__sk, sk_node)
921 #define sk_nulls_for_each_from(__sk, node) \
922 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
923 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
924 #define sk_for_each_safe(__sk, tmp, list) \
925 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
926 #define sk_for_each_bound(__sk, list) \
927 	hlist_for_each_entry(__sk, list, sk_bind_node)
928 #define sk_for_each_bound_safe(__sk, tmp, list) \
929 	hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node)
930 
931 /**
932  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
933  * @tpos:	the type * to use as a loop cursor.
934  * @pos:	the &struct hlist_node to use as a loop cursor.
935  * @head:	the head for your list.
936  * @offset:	offset of hlist_node within the struct.
937  *
938  */
939 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
940 	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
941 	     pos != NULL &&						       \
942 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
943 	     pos = rcu_dereference(hlist_next_rcu(pos)))
944 
945 static inline struct user_namespace *sk_user_ns(const struct sock *sk)
946 {
947 	/* Careful only use this in a context where these parameters
948 	 * can not change and must all be valid, such as recvmsg from
949 	 * userspace.
950 	 */
951 	return sk->sk_socket->file->f_cred->user_ns;
952 }
953 
954 /* Sock flags */
955 enum sock_flags {
956 	SOCK_DEAD,
957 	SOCK_DONE,
958 	SOCK_URGINLINE,
959 	SOCK_KEEPOPEN,
960 	SOCK_LINGER,
961 	SOCK_DESTROY,
962 	SOCK_BROADCAST,
963 	SOCK_TIMESTAMP,
964 	SOCK_ZAPPED,
965 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
966 	SOCK_DBG, /* %SO_DEBUG setting */
967 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
968 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
969 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
970 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
971 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
972 	SOCK_FASYNC, /* fasync() active */
973 	SOCK_RXQ_OVFL,
974 	SOCK_ZEROCOPY, /* buffers from userspace */
975 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
976 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
977 		     * Will use last 4 bytes of packet sent from
978 		     * user-space instead.
979 		     */
980 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
981 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
982 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
983 	SOCK_TXTIME,
984 	SOCK_XDP, /* XDP is attached */
985 	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
986 	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
987 	SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */
988 	SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */
989 };
990 
991 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
992 /*
993  * The highest bit of sk_tsflags is reserved for kernel-internal
994  * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that
995  * SOF_TIMESTAMPING* values do not reach this reserved area
996  */
997 #define SOCKCM_FLAG_TS_OPT_ID	BIT(31)
998 
999 static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
1000 {
1001 	nsk->sk_flags = osk->sk_flags;
1002 }
1003 
1004 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
1005 {
1006 	__set_bit(flag, &sk->sk_flags);
1007 }
1008 
1009 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
1010 {
1011 	__clear_bit(flag, &sk->sk_flags);
1012 }
1013 
1014 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
1015 				     int valbool)
1016 {
1017 	if (valbool)
1018 		sock_set_flag(sk, bit);
1019 	else
1020 		sock_reset_flag(sk, bit);
1021 }
1022 
1023 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
1024 {
1025 	return test_bit(flag, &sk->sk_flags);
1026 }
1027 
1028 #ifdef CONFIG_NET
1029 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
1030 static inline int sk_memalloc_socks(void)
1031 {
1032 	return static_branch_unlikely(&memalloc_socks_key);
1033 }
1034 
1035 void __receive_sock(struct file *file);
1036 #else
1037 
1038 static inline int sk_memalloc_socks(void)
1039 {
1040 	return 0;
1041 }
1042 
1043 static inline void __receive_sock(struct file *file)
1044 { }
1045 #endif
1046 
1047 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1048 {
1049 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1050 }
1051 
1052 static inline void sk_acceptq_removed(struct sock *sk)
1053 {
1054 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1055 }
1056 
1057 static inline void sk_acceptq_added(struct sock *sk)
1058 {
1059 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1060 }
1061 
1062 /* Note: If you think the test should be:
1063  *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1064  * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1065  */
1066 static inline bool sk_acceptq_is_full(const struct sock *sk)
1067 {
1068 	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1069 }
1070 
1071 /*
1072  * Compute minimal free write space needed to queue new packets.
1073  */
1074 static inline int sk_stream_min_wspace(const struct sock *sk)
1075 {
1076 	return READ_ONCE(sk->sk_wmem_queued) >> 1;
1077 }
1078 
1079 static inline int sk_stream_wspace(const struct sock *sk)
1080 {
1081 	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1082 }
1083 
1084 static inline void sk_wmem_queued_add(struct sock *sk, int val)
1085 {
1086 	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1087 }
1088 
1089 static inline void sk_forward_alloc_add(struct sock *sk, int val)
1090 {
1091 	/* Paired with lockless reads of sk->sk_forward_alloc */
1092 	WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1093 }
1094 
1095 void sk_stream_write_space(struct sock *sk);
1096 
1097 /* OOB backlog add */
1098 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1099 {
1100 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
1101 	skb_dst_force(skb);
1102 
1103 	if (!sk->sk_backlog.tail)
1104 		WRITE_ONCE(sk->sk_backlog.head, skb);
1105 	else
1106 		sk->sk_backlog.tail->next = skb;
1107 
1108 	WRITE_ONCE(sk->sk_backlog.tail, skb);
1109 	skb->next = NULL;
1110 }
1111 
1112 /*
1113  * Take into account size of receive queue and backlog queue
1114  * Do not take into account this skb truesize,
1115  * to allow even a single big packet to come.
1116  */
1117 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1118 {
1119 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1120 
1121 	return qsize > limit;
1122 }
1123 
1124 /* The per-socket spinlock must be held here. */
1125 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1126 					      unsigned int limit)
1127 {
1128 	if (sk_rcvqueues_full(sk, limit))
1129 		return -ENOBUFS;
1130 
1131 	/*
1132 	 * If the skb was allocated from pfmemalloc reserves, only
1133 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1134 	 * helping free memory
1135 	 */
1136 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1137 		return -ENOMEM;
1138 
1139 	__sk_add_backlog(sk, skb);
1140 	sk->sk_backlog.len += skb->truesize;
1141 	return 0;
1142 }
1143 
1144 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1145 
1146 INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1147 INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1148 
1149 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1150 {
1151 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1152 		return __sk_backlog_rcv(sk, skb);
1153 
1154 	return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1155 				  tcp_v6_do_rcv,
1156 				  tcp_v4_do_rcv,
1157 				  sk, skb);
1158 }
1159 
1160 static inline void sk_incoming_cpu_update(struct sock *sk)
1161 {
1162 	int cpu = raw_smp_processor_id();
1163 
1164 	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1165 		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1166 }
1167 
1168 
1169 static inline void sock_rps_save_rxhash(struct sock *sk,
1170 					const struct sk_buff *skb)
1171 {
1172 #ifdef CONFIG_RPS
1173 	/* The following WRITE_ONCE() is paired with the READ_ONCE()
1174 	 * here, and another one in sock_rps_record_flow().
1175 	 */
1176 	if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1177 		WRITE_ONCE(sk->sk_rxhash, skb->hash);
1178 #endif
1179 }
1180 
1181 static inline void sock_rps_reset_rxhash(struct sock *sk)
1182 {
1183 #ifdef CONFIG_RPS
1184 	/* Paired with READ_ONCE() in sock_rps_record_flow() */
1185 	WRITE_ONCE(sk->sk_rxhash, 0);
1186 #endif
1187 }
1188 
1189 #define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1190 	({	int __rc, __dis = __sk->sk_disconnects;			\
1191 		release_sock(__sk);					\
1192 		__rc = __condition;					\
1193 		if (!__rc) {						\
1194 			*(__timeo) = wait_woken(__wait,			\
1195 						TASK_INTERRUPTIBLE,	\
1196 						*(__timeo));		\
1197 		}							\
1198 		sched_annotate_sleep();					\
1199 		lock_sock(__sk);					\
1200 		__rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1201 		__rc;							\
1202 	})
1203 
1204 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1205 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1206 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1207 int sk_stream_error(struct sock *sk, int flags, int err);
1208 void sk_stream_kill_queues(struct sock *sk);
1209 void sk_set_memalloc(struct sock *sk);
1210 void sk_clear_memalloc(struct sock *sk);
1211 
1212 void __sk_flush_backlog(struct sock *sk);
1213 
1214 static inline bool sk_flush_backlog(struct sock *sk)
1215 {
1216 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1217 		__sk_flush_backlog(sk);
1218 		return true;
1219 	}
1220 	return false;
1221 }
1222 
1223 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1224 
1225 struct request_sock_ops;
1226 struct timewait_sock_ops;
1227 struct inet_hashinfo;
1228 struct raw_hashinfo;
1229 struct smc_hashinfo;
1230 struct module;
1231 struct sk_psock;
1232 
1233 /*
1234  * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1235  * un-modified. Special care is taken when initializing object to zero.
1236  */
1237 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1238 {
1239 	if (offsetof(struct sock, sk_node.next) != 0)
1240 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1241 	memset(&sk->sk_node.pprev, 0,
1242 	       size - offsetof(struct sock, sk_node.pprev));
1243 }
1244 
1245 struct proto_accept_arg {
1246 	int flags;
1247 	int err;
1248 	int is_empty;
1249 	bool kern;
1250 };
1251 
1252 /* Networking protocol blocks we attach to sockets.
1253  * socket layer -> transport layer interface
1254  */
1255 struct proto {
1256 	void			(*close)(struct sock *sk,
1257 					long timeout);
1258 	int			(*pre_connect)(struct sock *sk,
1259 					struct sockaddr *uaddr,
1260 					int addr_len);
1261 	int			(*connect)(struct sock *sk,
1262 					struct sockaddr *uaddr,
1263 					int addr_len);
1264 	int			(*disconnect)(struct sock *sk, int flags);
1265 
1266 	struct sock *		(*accept)(struct sock *sk,
1267 					  struct proto_accept_arg *arg);
1268 
1269 	int			(*ioctl)(struct sock *sk, int cmd,
1270 					 int *karg);
1271 	int			(*init)(struct sock *sk);
1272 	void			(*destroy)(struct sock *sk);
1273 	void			(*shutdown)(struct sock *sk, int how);
1274 	int			(*setsockopt)(struct sock *sk, int level,
1275 					int optname, sockptr_t optval,
1276 					unsigned int optlen);
1277 	int			(*getsockopt)(struct sock *sk, int level,
1278 					int optname, char __user *optval,
1279 					int __user *option);
1280 	void			(*keepalive)(struct sock *sk, int valbool);
1281 #ifdef CONFIG_COMPAT
1282 	int			(*compat_ioctl)(struct sock *sk,
1283 					unsigned int cmd, unsigned long arg);
1284 #endif
1285 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1286 					   size_t len);
1287 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1288 					   size_t len, int flags, int *addr_len);
1289 	void			(*splice_eof)(struct socket *sock);
1290 	int			(*bind)(struct sock *sk,
1291 					struct sockaddr *addr, int addr_len);
1292 	int			(*bind_add)(struct sock *sk,
1293 					struct sockaddr *addr, int addr_len);
1294 
1295 	int			(*backlog_rcv) (struct sock *sk,
1296 						struct sk_buff *skb);
1297 	bool			(*bpf_bypass_getsockopt)(int level,
1298 							 int optname);
1299 
1300 	void		(*release_cb)(struct sock *sk);
1301 
1302 	/* Keeping track of sk's, looking them up, and port selection methods. */
1303 	int			(*hash)(struct sock *sk);
1304 	void			(*unhash)(struct sock *sk);
1305 	void			(*rehash)(struct sock *sk);
1306 	int			(*get_port)(struct sock *sk, unsigned short snum);
1307 	void			(*put_port)(struct sock *sk);
1308 #ifdef CONFIG_BPF_SYSCALL
1309 	int			(*psock_update_sk_prot)(struct sock *sk,
1310 							struct sk_psock *psock,
1311 							bool restore);
1312 #endif
1313 
1314 	/* Keeping track of sockets in use */
1315 #ifdef CONFIG_PROC_FS
1316 	unsigned int		inuse_idx;
1317 #endif
1318 
1319 	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1320 	bool			(*sock_is_readable)(struct sock *sk);
1321 	/* Memory pressure */
1322 	void			(*enter_memory_pressure)(struct sock *sk);
1323 	void			(*leave_memory_pressure)(struct sock *sk);
1324 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1325 	int  __percpu		*per_cpu_fw_alloc;
1326 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1327 
1328 	/*
1329 	 * Pressure flag: try to collapse.
1330 	 * Technical note: it is used by multiple contexts non atomically.
1331 	 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1332 	 * All the __sk_mem_schedule() is of this nature: accounting
1333 	 * is strict, actions are advisory and have some latency.
1334 	 */
1335 	unsigned long		*memory_pressure;
1336 	long			*sysctl_mem;
1337 
1338 	int			*sysctl_wmem;
1339 	int			*sysctl_rmem;
1340 	u32			sysctl_wmem_offset;
1341 	u32			sysctl_rmem_offset;
1342 
1343 	int			max_header;
1344 	bool			no_autobind;
1345 
1346 	struct kmem_cache	*slab;
1347 	unsigned int		obj_size;
1348 	unsigned int		ipv6_pinfo_offset;
1349 	slab_flags_t		slab_flags;
1350 	unsigned int		useroffset;	/* Usercopy region offset */
1351 	unsigned int		usersize;	/* Usercopy region size */
1352 
1353 	struct request_sock_ops	*rsk_prot;
1354 	struct timewait_sock_ops *twsk_prot;
1355 
1356 	union {
1357 		struct inet_hashinfo	*hashinfo;
1358 		struct udp_table	*udp_table;
1359 		struct raw_hashinfo	*raw_hash;
1360 		struct smc_hashinfo	*smc_hash;
1361 	} h;
1362 
1363 	struct module		*owner;
1364 
1365 	char			name[32];
1366 
1367 	struct list_head	node;
1368 	int			(*diag_destroy)(struct sock *sk, int err);
1369 } __randomize_layout;
1370 
1371 int proto_register(struct proto *prot, int alloc_slab);
1372 void proto_unregister(struct proto *prot);
1373 int sock_load_diag_module(int family, int protocol);
1374 
1375 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1376 
1377 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1378 {
1379 	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1380 		return false;
1381 
1382 	return sk->sk_prot->stream_memory_free ?
1383 		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1384 				     tcp_stream_memory_free, sk, wake) : true;
1385 }
1386 
1387 static inline bool sk_stream_memory_free(const struct sock *sk)
1388 {
1389 	return __sk_stream_memory_free(sk, 0);
1390 }
1391 
1392 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1393 {
1394 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1395 	       __sk_stream_memory_free(sk, wake);
1396 }
1397 
1398 static inline bool sk_stream_is_writeable(const struct sock *sk)
1399 {
1400 	return __sk_stream_is_writeable(sk, 0);
1401 }
1402 
1403 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1404 					    struct cgroup *ancestor)
1405 {
1406 #ifdef CONFIG_SOCK_CGROUP_DATA
1407 	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1408 				    ancestor);
1409 #else
1410 	return -ENOTSUPP;
1411 #endif
1412 }
1413 
1414 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1415 
1416 static inline void sk_sockets_allocated_dec(struct sock *sk)
1417 {
1418 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1419 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1420 }
1421 
1422 static inline void sk_sockets_allocated_inc(struct sock *sk)
1423 {
1424 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1425 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1426 }
1427 
1428 static inline u64
1429 sk_sockets_allocated_read_positive(struct sock *sk)
1430 {
1431 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1432 }
1433 
1434 static inline int
1435 proto_sockets_allocated_sum_positive(struct proto *prot)
1436 {
1437 	return percpu_counter_sum_positive(prot->sockets_allocated);
1438 }
1439 
1440 #ifdef CONFIG_PROC_FS
1441 #define PROTO_INUSE_NR	64	/* should be enough for the first time */
1442 struct prot_inuse {
1443 	int all;
1444 	int val[PROTO_INUSE_NR];
1445 };
1446 
1447 static inline void sock_prot_inuse_add(const struct net *net,
1448 				       const struct proto *prot, int val)
1449 {
1450 	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1451 }
1452 
1453 static inline void sock_inuse_add(const struct net *net, int val)
1454 {
1455 	this_cpu_add(net->core.prot_inuse->all, val);
1456 }
1457 
1458 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1459 int sock_inuse_get(struct net *net);
1460 #else
1461 static inline void sock_prot_inuse_add(const struct net *net,
1462 				       const struct proto *prot, int val)
1463 {
1464 }
1465 
1466 static inline void sock_inuse_add(const struct net *net, int val)
1467 {
1468 }
1469 #endif
1470 
1471 
1472 /* With per-bucket locks this operation is not-atomic, so that
1473  * this version is not worse.
1474  */
1475 static inline int __sk_prot_rehash(struct sock *sk)
1476 {
1477 	sk->sk_prot->unhash(sk);
1478 	return sk->sk_prot->hash(sk);
1479 }
1480 
1481 /* About 10 seconds */
1482 #define SOCK_DESTROY_TIME (10*HZ)
1483 
1484 /* Sockets 0-1023 can't be bound to unless you are superuser */
1485 #define PROT_SOCK	1024
1486 
1487 #define SHUTDOWN_MASK	3
1488 #define RCV_SHUTDOWN	1
1489 #define SEND_SHUTDOWN	2
1490 
1491 #define SOCK_BINDADDR_LOCK	4
1492 #define SOCK_BINDPORT_LOCK	8
1493 
1494 struct socket_alloc {
1495 	struct socket socket;
1496 	struct inode vfs_inode;
1497 };
1498 
1499 static inline struct socket *SOCKET_I(struct inode *inode)
1500 {
1501 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1502 }
1503 
1504 static inline struct inode *SOCK_INODE(struct socket *socket)
1505 {
1506 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1507 }
1508 
1509 /*
1510  * Functions for memory accounting
1511  */
1512 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1513 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1514 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1515 void __sk_mem_reclaim(struct sock *sk, int amount);
1516 
1517 #define SK_MEM_SEND	0
1518 #define SK_MEM_RECV	1
1519 
1520 /* sysctl_mem values are in pages */
1521 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1522 {
1523 	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1524 }
1525 
1526 static inline int sk_mem_pages(int amt)
1527 {
1528 	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1529 }
1530 
1531 static inline bool sk_has_account(struct sock *sk)
1532 {
1533 	/* return true if protocol supports memory accounting */
1534 	return !!sk->sk_prot->memory_allocated;
1535 }
1536 
1537 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1538 {
1539 	int delta;
1540 
1541 	if (!sk_has_account(sk))
1542 		return true;
1543 	delta = size - sk->sk_forward_alloc;
1544 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1545 }
1546 
1547 static inline bool
1548 __sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc)
1549 {
1550 	int delta;
1551 
1552 	if (!sk_has_account(sk))
1553 		return true;
1554 	delta = size - sk->sk_forward_alloc;
1555 	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1556 	       pfmemalloc;
1557 }
1558 
1559 static inline bool
1560 sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size)
1561 {
1562 	return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb));
1563 }
1564 
1565 static inline int sk_unused_reserved_mem(const struct sock *sk)
1566 {
1567 	int unused_mem;
1568 
1569 	if (likely(!sk->sk_reserved_mem))
1570 		return 0;
1571 
1572 	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1573 			atomic_read(&sk->sk_rmem_alloc);
1574 
1575 	return unused_mem > 0 ? unused_mem : 0;
1576 }
1577 
1578 static inline void sk_mem_reclaim(struct sock *sk)
1579 {
1580 	int reclaimable;
1581 
1582 	if (!sk_has_account(sk))
1583 		return;
1584 
1585 	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1586 
1587 	if (reclaimable >= (int)PAGE_SIZE)
1588 		__sk_mem_reclaim(sk, reclaimable);
1589 }
1590 
1591 static inline void sk_mem_reclaim_final(struct sock *sk)
1592 {
1593 	sk->sk_reserved_mem = 0;
1594 	sk_mem_reclaim(sk);
1595 }
1596 
1597 static inline void sk_mem_charge(struct sock *sk, int size)
1598 {
1599 	if (!sk_has_account(sk))
1600 		return;
1601 	sk_forward_alloc_add(sk, -size);
1602 }
1603 
1604 static inline void sk_mem_uncharge(struct sock *sk, int size)
1605 {
1606 	if (!sk_has_account(sk))
1607 		return;
1608 	sk_forward_alloc_add(sk, size);
1609 	sk_mem_reclaim(sk);
1610 }
1611 
1612 #if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
1613 static inline void sk_owner_set(struct sock *sk, struct module *owner)
1614 {
1615 	__module_get(owner);
1616 	sk->sk_owner = owner;
1617 }
1618 
1619 static inline void sk_owner_clear(struct sock *sk)
1620 {
1621 	sk->sk_owner = NULL;
1622 }
1623 
1624 static inline void sk_owner_put(struct sock *sk)
1625 {
1626 	module_put(sk->sk_owner);
1627 }
1628 #else
1629 static inline void sk_owner_set(struct sock *sk, struct module *owner)
1630 {
1631 }
1632 
1633 static inline void sk_owner_clear(struct sock *sk)
1634 {
1635 }
1636 
1637 static inline void sk_owner_put(struct sock *sk)
1638 {
1639 }
1640 #endif
1641 /*
1642  * Macro so as to not evaluate some arguments when
1643  * lockdep is not enabled.
1644  *
1645  * Mark both the sk_lock and the sk_lock.slock as a
1646  * per-address-family lock class.
1647  */
1648 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1649 do {									\
1650 	sk_owner_set(sk, THIS_MODULE);					\
1651 	sk->sk_lock.owned = 0;						\
1652 	init_waitqueue_head(&sk->sk_lock.wq);				\
1653 	spin_lock_init(&(sk)->sk_lock.slock);				\
1654 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1655 				   sizeof((sk)->sk_lock));		\
1656 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1657 				   (skey), (sname));			\
1658 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1659 } while (0)
1660 
1661 static inline bool lockdep_sock_is_held(const struct sock *sk)
1662 {
1663 	return lockdep_is_held(&sk->sk_lock) ||
1664 	       lockdep_is_held(&sk->sk_lock.slock);
1665 }
1666 
1667 void lock_sock_nested(struct sock *sk, int subclass);
1668 
1669 static inline void lock_sock(struct sock *sk)
1670 {
1671 	lock_sock_nested(sk, 0);
1672 }
1673 
1674 void __lock_sock(struct sock *sk);
1675 void __release_sock(struct sock *sk);
1676 void release_sock(struct sock *sk);
1677 
1678 /* BH context may only use the following locking interface. */
1679 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1680 #define bh_lock_sock_nested(__sk) \
1681 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1682 				SINGLE_DEPTH_NESTING)
1683 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1684 
1685 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1686 
1687 /**
1688  * lock_sock_fast - fast version of lock_sock
1689  * @sk: socket
1690  *
1691  * This version should be used for very small section, where process won't block
1692  * return false if fast path is taken:
1693  *
1694  *   sk_lock.slock locked, owned = 0, BH disabled
1695  *
1696  * return true if slow path is taken:
1697  *
1698  *   sk_lock.slock unlocked, owned = 1, BH enabled
1699  */
1700 static inline bool lock_sock_fast(struct sock *sk)
1701 {
1702 	/* The sk_lock has mutex_lock() semantics here. */
1703 	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1704 
1705 	return __lock_sock_fast(sk);
1706 }
1707 
1708 /* fast socket lock variant for caller already holding a [different] socket lock */
1709 static inline bool lock_sock_fast_nested(struct sock *sk)
1710 {
1711 	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1712 
1713 	return __lock_sock_fast(sk);
1714 }
1715 
1716 /**
1717  * unlock_sock_fast - complement of lock_sock_fast
1718  * @sk: socket
1719  * @slow: slow mode
1720  *
1721  * fast unlock socket for user context.
1722  * If slow mode is on, we call regular release_sock()
1723  */
1724 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1725 	__releases(&sk->sk_lock.slock)
1726 {
1727 	if (slow) {
1728 		release_sock(sk);
1729 		__release(&sk->sk_lock.slock);
1730 	} else {
1731 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1732 		spin_unlock_bh(&sk->sk_lock.slock);
1733 	}
1734 }
1735 
1736 void sockopt_lock_sock(struct sock *sk);
1737 void sockopt_release_sock(struct sock *sk);
1738 bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1739 bool sockopt_capable(int cap);
1740 
1741 /* Used by processes to "lock" a socket state, so that
1742  * interrupts and bottom half handlers won't change it
1743  * from under us. It essentially blocks any incoming
1744  * packets, so that we won't get any new data or any
1745  * packets that change the state of the socket.
1746  *
1747  * While locked, BH processing will add new packets to
1748  * the backlog queue.  This queue is processed by the
1749  * owner of the socket lock right before it is released.
1750  *
1751  * Since ~2.3.5 it is also exclusive sleep lock serializing
1752  * accesses from user process context.
1753  */
1754 
1755 static inline void sock_owned_by_me(const struct sock *sk)
1756 {
1757 #ifdef CONFIG_LOCKDEP
1758 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1759 #endif
1760 }
1761 
1762 static inline void sock_not_owned_by_me(const struct sock *sk)
1763 {
1764 #ifdef CONFIG_LOCKDEP
1765 	WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1766 #endif
1767 }
1768 
1769 static inline bool sock_owned_by_user(const struct sock *sk)
1770 {
1771 	sock_owned_by_me(sk);
1772 	return sk->sk_lock.owned;
1773 }
1774 
1775 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1776 {
1777 	return sk->sk_lock.owned;
1778 }
1779 
1780 static inline void sock_release_ownership(struct sock *sk)
1781 {
1782 	DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1783 	sk->sk_lock.owned = 0;
1784 
1785 	/* The sk_lock has mutex_unlock() semantics: */
1786 	mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1787 }
1788 
1789 /* no reclassification while locks are held */
1790 static inline bool sock_allow_reclassification(const struct sock *csk)
1791 {
1792 	struct sock *sk = (struct sock *)csk;
1793 
1794 	return !sock_owned_by_user_nocheck(sk) &&
1795 		!spin_is_locked(&sk->sk_lock.slock);
1796 }
1797 
1798 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1799 		      struct proto *prot, int kern);
1800 void sk_free(struct sock *sk);
1801 void sk_net_refcnt_upgrade(struct sock *sk);
1802 void sk_destruct(struct sock *sk);
1803 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1804 
1805 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1806 			     gfp_t priority);
1807 void __sock_wfree(struct sk_buff *skb);
1808 void sock_wfree(struct sk_buff *skb);
1809 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1810 			     gfp_t priority);
1811 void skb_orphan_partial(struct sk_buff *skb);
1812 void sock_rfree(struct sk_buff *skb);
1813 void sock_efree(struct sk_buff *skb);
1814 #ifdef CONFIG_INET
1815 void sock_edemux(struct sk_buff *skb);
1816 void sock_pfree(struct sk_buff *skb);
1817 
1818 static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk)
1819 {
1820 	skb_orphan(skb);
1821 	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1822 		skb->sk = sk;
1823 		skb->destructor = sock_edemux;
1824 	}
1825 }
1826 #else
1827 #define sock_edemux sock_efree
1828 #endif
1829 
1830 int sk_setsockopt(struct sock *sk, int level, int optname,
1831 		  sockptr_t optval, unsigned int optlen);
1832 int sock_setsockopt(struct socket *sock, int level, int op,
1833 		    sockptr_t optval, unsigned int optlen);
1834 int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1835 		       int optname, sockptr_t optval, int optlen);
1836 int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1837 		       int optname, sockptr_t optval, sockptr_t optlen);
1838 
1839 int sk_getsockopt(struct sock *sk, int level, int optname,
1840 		  sockptr_t optval, sockptr_t optlen);
1841 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1842 		   bool timeval, bool time32);
1843 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1844 				     unsigned long data_len, int noblock,
1845 				     int *errcode, int max_page_order);
1846 
1847 static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1848 						  unsigned long size,
1849 						  int noblock, int *errcode)
1850 {
1851 	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1852 }
1853 
1854 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1855 void *sock_kmemdup(struct sock *sk, const void *src,
1856 		   int size, gfp_t priority);
1857 void sock_kfree_s(struct sock *sk, void *mem, int size);
1858 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1859 void sk_send_sigurg(struct sock *sk);
1860 
1861 static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1862 {
1863 	if (sk->sk_socket)
1864 		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1865 	WRITE_ONCE(sk->sk_prot, proto);
1866 }
1867 
1868 struct sockcm_cookie {
1869 	u64 transmit_time;
1870 	u32 mark;
1871 	u32 tsflags;
1872 	u32 ts_opt_id;
1873 	u32 priority;
1874 	u32 dmabuf_id;
1875 };
1876 
1877 static inline void sockcm_init(struct sockcm_cookie *sockc,
1878 			       const struct sock *sk)
1879 {
1880 	*sockc = (struct sockcm_cookie) {
1881 		.mark = READ_ONCE(sk->sk_mark),
1882 		.tsflags = READ_ONCE(sk->sk_tsflags),
1883 		.priority = READ_ONCE(sk->sk_priority),
1884 	};
1885 }
1886 
1887 int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1888 		     struct sockcm_cookie *sockc);
1889 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1890 		   struct sockcm_cookie *sockc);
1891 
1892 /*
1893  * Functions to fill in entries in struct proto_ops when a protocol
1894  * does not implement a particular function.
1895  */
1896 int sock_no_bind(struct socket *, struct sockaddr *, int);
1897 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1898 int sock_no_socketpair(struct socket *, struct socket *);
1899 int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *);
1900 int sock_no_getname(struct socket *, struct sockaddr *, int);
1901 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1902 int sock_no_listen(struct socket *, int);
1903 int sock_no_shutdown(struct socket *, int);
1904 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1905 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1906 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1907 int sock_no_mmap(struct file *file, struct socket *sock,
1908 		 struct vm_area_struct *vma);
1909 
1910 /*
1911  * Functions to fill in entries in struct proto_ops when a protocol
1912  * uses the inet style.
1913  */
1914 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1915 				  char __user *optval, int __user *optlen);
1916 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1917 			int flags);
1918 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1919 			   sockptr_t optval, unsigned int optlen);
1920 
1921 void sk_common_release(struct sock *sk);
1922 
1923 /*
1924  *	Default socket callbacks and setup code
1925  */
1926 
1927 /* Initialise core socket variables using an explicit uid. */
1928 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1929 
1930 /* Initialise core socket variables.
1931  * Assumes struct socket *sock is embedded in a struct socket_alloc.
1932  */
1933 void sock_init_data(struct socket *sock, struct sock *sk);
1934 
1935 /*
1936  * Socket reference counting postulates.
1937  *
1938  * * Each user of socket SHOULD hold a reference count.
1939  * * Each access point to socket (an hash table bucket, reference from a list,
1940  *   running timer, skb in flight MUST hold a reference count.
1941  * * When reference count hits 0, it means it will never increase back.
1942  * * When reference count hits 0, it means that no references from
1943  *   outside exist to this socket and current process on current CPU
1944  *   is last user and may/should destroy this socket.
1945  * * sk_free is called from any context: process, BH, IRQ. When
1946  *   it is called, socket has no references from outside -> sk_free
1947  *   may release descendant resources allocated by the socket, but
1948  *   to the time when it is called, socket is NOT referenced by any
1949  *   hash tables, lists etc.
1950  * * Packets, delivered from outside (from network or from another process)
1951  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1952  *   when they sit in queue. Otherwise, packets will leak to hole, when
1953  *   socket is looked up by one cpu and unhasing is made by another CPU.
1954  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1955  *   (leak to backlog). Packet socket does all the processing inside
1956  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1957  *   use separate SMP lock, so that they are prone too.
1958  */
1959 
1960 /* Ungrab socket and destroy it, if it was the last reference. */
1961 static inline void sock_put(struct sock *sk)
1962 {
1963 	if (refcount_dec_and_test(&sk->sk_refcnt))
1964 		sk_free(sk);
1965 }
1966 /* Generic version of sock_put(), dealing with all sockets
1967  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1968  */
1969 void sock_gen_put(struct sock *sk);
1970 
1971 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1972 		     unsigned int trim_cap, bool refcounted);
1973 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1974 				 const int nested)
1975 {
1976 	return __sk_receive_skb(sk, skb, nested, 1, true);
1977 }
1978 
1979 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1980 {
1981 	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1982 	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1983 		return;
1984 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1985 	 * other WRITE_ONCE() because socket lock might be not held.
1986 	 */
1987 	WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1988 }
1989 
1990 #define NO_QUEUE_MAPPING	USHRT_MAX
1991 
1992 static inline void sk_tx_queue_clear(struct sock *sk)
1993 {
1994 	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1995 	 * other WRITE_ONCE() because socket lock might be not held.
1996 	 */
1997 	WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1998 }
1999 
2000 static inline int sk_tx_queue_get(const struct sock *sk)
2001 {
2002 	if (sk) {
2003 		/* Paired with WRITE_ONCE() in sk_tx_queue_clear()
2004 		 * and sk_tx_queue_set().
2005 		 */
2006 		int val = READ_ONCE(sk->sk_tx_queue_mapping);
2007 
2008 		if (val != NO_QUEUE_MAPPING)
2009 			return val;
2010 	}
2011 	return -1;
2012 }
2013 
2014 static inline void __sk_rx_queue_set(struct sock *sk,
2015 				     const struct sk_buff *skb,
2016 				     bool force_set)
2017 {
2018 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2019 	if (skb_rx_queue_recorded(skb)) {
2020 		u16 rx_queue = skb_get_rx_queue(skb);
2021 
2022 		if (force_set ||
2023 		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2024 			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2025 	}
2026 #endif
2027 }
2028 
2029 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2030 {
2031 	__sk_rx_queue_set(sk, skb, true);
2032 }
2033 
2034 static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2035 {
2036 	__sk_rx_queue_set(sk, skb, false);
2037 }
2038 
2039 static inline void sk_rx_queue_clear(struct sock *sk)
2040 {
2041 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2042 	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2043 #endif
2044 }
2045 
2046 static inline int sk_rx_queue_get(const struct sock *sk)
2047 {
2048 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2049 	if (sk) {
2050 		int res = READ_ONCE(sk->sk_rx_queue_mapping);
2051 
2052 		if (res != NO_QUEUE_MAPPING)
2053 			return res;
2054 	}
2055 #endif
2056 
2057 	return -1;
2058 }
2059 
2060 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2061 {
2062 	sk->sk_socket = sock;
2063 	if (sock) {
2064 		WRITE_ONCE(sk->sk_uid, SOCK_INODE(sock)->i_uid);
2065 		WRITE_ONCE(sk->sk_ino, SOCK_INODE(sock)->i_ino);
2066 	}
2067 }
2068 
2069 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2070 {
2071 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2072 	return &rcu_dereference_raw(sk->sk_wq)->wait;
2073 }
2074 /* Detach socket from process context.
2075  * Announce socket dead, detach it from wait queue and inode.
2076  * Note that parent inode held reference count on this struct sock,
2077  * we do not release it in this function, because protocol
2078  * probably wants some additional cleanups or even continuing
2079  * to work with this socket (TCP).
2080  */
2081 static inline void sock_orphan(struct sock *sk)
2082 {
2083 	write_lock_bh(&sk->sk_callback_lock);
2084 	sock_set_flag(sk, SOCK_DEAD);
2085 	sk_set_socket(sk, NULL);
2086 	sk->sk_wq  = NULL;
2087 	/* Note: sk_uid is unchanged. */
2088 	WRITE_ONCE(sk->sk_ino, 0);
2089 	write_unlock_bh(&sk->sk_callback_lock);
2090 }
2091 
2092 static inline void sock_graft(struct sock *sk, struct socket *parent)
2093 {
2094 	WARN_ON(parent->sk);
2095 	write_lock_bh(&sk->sk_callback_lock);
2096 	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2097 	parent->sk = sk;
2098 	sk_set_socket(sk, parent);
2099 	security_sock_graft(sk, parent);
2100 	write_unlock_bh(&sk->sk_callback_lock);
2101 }
2102 
2103 static inline unsigned long sock_i_ino(const struct sock *sk)
2104 {
2105 	/* Paired with WRITE_ONCE() in sock_graft() and sock_orphan() */
2106 	return READ_ONCE(sk->sk_ino);
2107 }
2108 
2109 static inline kuid_t sk_uid(const struct sock *sk)
2110 {
2111 	/* Paired with WRITE_ONCE() in sockfs_setattr() */
2112 	return READ_ONCE(sk->sk_uid);
2113 }
2114 
2115 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2116 {
2117 	return sk ? sk_uid(sk) : make_kuid(net->user_ns, 0);
2118 }
2119 
2120 static inline u32 net_tx_rndhash(void)
2121 {
2122 	u32 v = get_random_u32();
2123 
2124 	return v ?: 1;
2125 }
2126 
2127 static inline void sk_set_txhash(struct sock *sk)
2128 {
2129 	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2130 	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2131 }
2132 
2133 static inline bool sk_rethink_txhash(struct sock *sk)
2134 {
2135 	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2136 		sk_set_txhash(sk);
2137 		return true;
2138 	}
2139 	return false;
2140 }
2141 
2142 static inline struct dst_entry *
2143 __sk_dst_get(const struct sock *sk)
2144 {
2145 	return rcu_dereference_check(sk->sk_dst_cache,
2146 				     lockdep_sock_is_held(sk));
2147 }
2148 
2149 static inline struct dst_entry *
2150 sk_dst_get(const struct sock *sk)
2151 {
2152 	struct dst_entry *dst;
2153 
2154 	rcu_read_lock();
2155 	dst = rcu_dereference(sk->sk_dst_cache);
2156 	if (dst && !rcuref_get(&dst->__rcuref))
2157 		dst = NULL;
2158 	rcu_read_unlock();
2159 	return dst;
2160 }
2161 
2162 static inline void __dst_negative_advice(struct sock *sk)
2163 {
2164 	struct dst_entry *dst = __sk_dst_get(sk);
2165 
2166 	if (dst && dst->ops->negative_advice)
2167 		dst->ops->negative_advice(sk, dst);
2168 }
2169 
2170 static inline void dst_negative_advice(struct sock *sk)
2171 {
2172 	sk_rethink_txhash(sk);
2173 	__dst_negative_advice(sk);
2174 }
2175 
2176 static inline void
2177 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2178 {
2179 	struct dst_entry *old_dst;
2180 
2181 	sk_tx_queue_clear(sk);
2182 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2183 	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2184 					    lockdep_sock_is_held(sk));
2185 	rcu_assign_pointer(sk->sk_dst_cache, dst);
2186 	dst_release(old_dst);
2187 }
2188 
2189 static inline void
2190 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2191 {
2192 	struct dst_entry *old_dst;
2193 
2194 	sk_tx_queue_clear(sk);
2195 	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2196 	old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst)));
2197 	dst_release(old_dst);
2198 }
2199 
2200 static inline void
2201 __sk_dst_reset(struct sock *sk)
2202 {
2203 	__sk_dst_set(sk, NULL);
2204 }
2205 
2206 static inline void
2207 sk_dst_reset(struct sock *sk)
2208 {
2209 	sk_dst_set(sk, NULL);
2210 }
2211 
2212 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2213 
2214 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2215 
2216 static inline void sk_dst_confirm(struct sock *sk)
2217 {
2218 	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2219 		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2220 }
2221 
2222 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2223 {
2224 	if (skb_get_dst_pending_confirm(skb)) {
2225 		struct sock *sk = skb->sk;
2226 
2227 		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2228 			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2229 		neigh_confirm(n);
2230 	}
2231 }
2232 
2233 bool sk_mc_loop(const struct sock *sk);
2234 
2235 static inline bool sk_can_gso(const struct sock *sk)
2236 {
2237 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2238 }
2239 
2240 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2241 
2242 static inline void sk_gso_disable(struct sock *sk)
2243 {
2244 	sk->sk_gso_disabled = 1;
2245 	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2246 }
2247 
2248 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2249 					   struct iov_iter *from, char *to,
2250 					   int copy, int offset)
2251 {
2252 	if (skb->ip_summed == CHECKSUM_NONE) {
2253 		__wsum csum = 0;
2254 		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2255 			return -EFAULT;
2256 		skb->csum = csum_block_add(skb->csum, csum, offset);
2257 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2258 		if (!copy_from_iter_full_nocache(to, copy, from))
2259 			return -EFAULT;
2260 	} else if (!copy_from_iter_full(to, copy, from))
2261 		return -EFAULT;
2262 
2263 	return 0;
2264 }
2265 
2266 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2267 				       struct iov_iter *from, int copy)
2268 {
2269 	int err, offset = skb->len;
2270 
2271 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2272 				       copy, offset);
2273 	if (err)
2274 		__skb_trim(skb, offset);
2275 
2276 	return err;
2277 }
2278 
2279 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2280 					   struct sk_buff *skb,
2281 					   struct page *page,
2282 					   int off, int copy)
2283 {
2284 	int err;
2285 
2286 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2287 				       copy, skb->len);
2288 	if (err)
2289 		return err;
2290 
2291 	skb_len_add(skb, copy);
2292 	sk_wmem_queued_add(sk, copy);
2293 	sk_mem_charge(sk, copy);
2294 	return 0;
2295 }
2296 
2297 /**
2298  * sk_wmem_alloc_get - returns write allocations
2299  * @sk: socket
2300  *
2301  * Return: sk_wmem_alloc minus initial offset of one
2302  */
2303 static inline int sk_wmem_alloc_get(const struct sock *sk)
2304 {
2305 	return refcount_read(&sk->sk_wmem_alloc) - 1;
2306 }
2307 
2308 /**
2309  * sk_rmem_alloc_get - returns read allocations
2310  * @sk: socket
2311  *
2312  * Return: sk_rmem_alloc
2313  */
2314 static inline int sk_rmem_alloc_get(const struct sock *sk)
2315 {
2316 	return atomic_read(&sk->sk_rmem_alloc);
2317 }
2318 
2319 /**
2320  * sk_has_allocations - check if allocations are outstanding
2321  * @sk: socket
2322  *
2323  * Return: true if socket has write or read allocations
2324  */
2325 static inline bool sk_has_allocations(const struct sock *sk)
2326 {
2327 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2328 }
2329 
2330 /**
2331  * skwq_has_sleeper - check if there are any waiting processes
2332  * @wq: struct socket_wq
2333  *
2334  * Return: true if socket_wq has waiting processes
2335  *
2336  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2337  * barrier call. They were added due to the race found within the tcp code.
2338  *
2339  * Consider following tcp code paths::
2340  *
2341  *   CPU1                CPU2
2342  *   sys_select          receive packet
2343  *   ...                 ...
2344  *   __add_wait_queue    update tp->rcv_nxt
2345  *   ...                 ...
2346  *   tp->rcv_nxt check   sock_def_readable
2347  *   ...                 {
2348  *   schedule               rcu_read_lock();
2349  *                          wq = rcu_dereference(sk->sk_wq);
2350  *                          if (wq && waitqueue_active(&wq->wait))
2351  *                              wake_up_interruptible(&wq->wait)
2352  *                          ...
2353  *                       }
2354  *
2355  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2356  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2357  * could then endup calling schedule and sleep forever if there are no more
2358  * data on the socket.
2359  *
2360  */
2361 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2362 {
2363 	return wq && wq_has_sleeper(&wq->wait);
2364 }
2365 
2366 /**
2367  * sock_poll_wait - wrapper for the poll_wait call.
2368  * @filp:           file
2369  * @sock:           socket to wait on
2370  * @p:              poll_table
2371  *
2372  * See the comments in the wq_has_sleeper function.
2373  */
2374 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2375 				  poll_table *p)
2376 {
2377 	/* Provides a barrier we need to be sure we are in sync
2378 	 * with the socket flags modification.
2379 	 *
2380 	 * This memory barrier is paired in the wq_has_sleeper.
2381 	 */
2382 	poll_wait(filp, &sock->wq.wait, p);
2383 }
2384 
2385 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2386 {
2387 	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2388 	u32 txhash = READ_ONCE(sk->sk_txhash);
2389 
2390 	if (txhash) {
2391 		skb->l4_hash = 1;
2392 		skb->hash = txhash;
2393 	}
2394 }
2395 
2396 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2397 
2398 /*
2399  *	Queue a received datagram if it will fit. Stream and sequenced
2400  *	protocols can't normally use this as they need to fit buffers in
2401  *	and play with them.
2402  *
2403  *	Inlined as it's very short and called for pretty much every
2404  *	packet ever received.
2405  */
2406 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2407 {
2408 	skb_orphan(skb);
2409 	skb->sk = sk;
2410 	skb->destructor = sock_rfree;
2411 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2412 	sk_mem_charge(sk, skb->truesize);
2413 }
2414 
2415 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2416 {
2417 	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2418 		skb_orphan(skb);
2419 		skb->destructor = sock_efree;
2420 		skb->sk = sk;
2421 		return true;
2422 	}
2423 	return false;
2424 }
2425 
2426 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2427 {
2428 	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2429 	if (skb) {
2430 		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2431 			skb_set_owner_r(skb, sk);
2432 			return skb;
2433 		}
2434 		__kfree_skb(skb);
2435 	}
2436 	return NULL;
2437 }
2438 
2439 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2440 {
2441 	if (skb->destructor != sock_wfree) {
2442 		skb_orphan(skb);
2443 		return;
2444 	}
2445 	skb->slow_gro = 1;
2446 }
2447 
2448 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2449 		    unsigned long expires);
2450 
2451 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2452 
2453 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2454 
2455 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2456 			struct sk_buff *skb, unsigned int flags,
2457 			void (*destructor)(struct sock *sk,
2458 					   struct sk_buff *skb));
2459 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2460 
2461 int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2462 			      enum skb_drop_reason *reason);
2463 
2464 static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2465 {
2466 	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2467 }
2468 
2469 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2470 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2471 
2472 /*
2473  *	Recover an error report and clear atomically
2474  */
2475 
2476 static inline int sock_error(struct sock *sk)
2477 {
2478 	int err;
2479 
2480 	/* Avoid an atomic operation for the common case.
2481 	 * This is racy since another cpu/thread can change sk_err under us.
2482 	 */
2483 	if (likely(data_race(!sk->sk_err)))
2484 		return 0;
2485 
2486 	err = xchg(&sk->sk_err, 0);
2487 	return -err;
2488 }
2489 
2490 void sk_error_report(struct sock *sk);
2491 
2492 static inline unsigned long sock_wspace(struct sock *sk)
2493 {
2494 	int amt = 0;
2495 
2496 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2497 		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2498 		if (amt < 0)
2499 			amt = 0;
2500 	}
2501 	return amt;
2502 }
2503 
2504 /* Note:
2505  *  We use sk->sk_wq_raw, from contexts knowing this
2506  *  pointer is not NULL and cannot disappear/change.
2507  */
2508 static inline void sk_set_bit(int nr, struct sock *sk)
2509 {
2510 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2511 	    !sock_flag(sk, SOCK_FASYNC))
2512 		return;
2513 
2514 	set_bit(nr, &sk->sk_wq_raw->flags);
2515 }
2516 
2517 static inline void sk_clear_bit(int nr, struct sock *sk)
2518 {
2519 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2520 	    !sock_flag(sk, SOCK_FASYNC))
2521 		return;
2522 
2523 	clear_bit(nr, &sk->sk_wq_raw->flags);
2524 }
2525 
2526 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2527 {
2528 	if (sock_flag(sk, SOCK_FASYNC)) {
2529 		rcu_read_lock();
2530 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2531 		rcu_read_unlock();
2532 	}
2533 }
2534 
2535 static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band)
2536 {
2537 	if (unlikely(sock_flag(sk, SOCK_FASYNC)))
2538 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2539 }
2540 
2541 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2542  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2543  * Note: for send buffers, TCP works better if we can build two skbs at
2544  * minimum.
2545  */
2546 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2547 
2548 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2549 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2550 
2551 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2552 {
2553 	u32 val;
2554 
2555 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2556 		return;
2557 
2558 	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2559 	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2560 
2561 	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2562 }
2563 
2564 /**
2565  * sk_page_frag - return an appropriate page_frag
2566  * @sk: socket
2567  *
2568  * Use the per task page_frag instead of the per socket one for
2569  * optimization when we know that we're in process context and own
2570  * everything that's associated with %current.
2571  *
2572  * Both direct reclaim and page faults can nest inside other
2573  * socket operations and end up recursing into sk_page_frag()
2574  * while it's already in use: explicitly avoid task page_frag
2575  * when users disable sk_use_task_frag.
2576  *
2577  * Return: a per task page_frag if context allows that,
2578  * otherwise a per socket one.
2579  */
2580 static inline struct page_frag *sk_page_frag(struct sock *sk)
2581 {
2582 	if (sk->sk_use_task_frag)
2583 		return &current->task_frag;
2584 
2585 	return &sk->sk_frag;
2586 }
2587 
2588 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2589 
2590 /*
2591  *	Default write policy as shown to user space via poll/select/SIGIO
2592  */
2593 static inline bool sock_writeable(const struct sock *sk)
2594 {
2595 	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2596 }
2597 
2598 static inline gfp_t gfp_any(void)
2599 {
2600 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2601 }
2602 
2603 static inline gfp_t gfp_memcg_charge(void)
2604 {
2605 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2606 }
2607 
2608 #ifdef CONFIG_MEMCG
2609 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2610 {
2611 	return sk->sk_memcg;
2612 }
2613 
2614 static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2615 {
2616 	return mem_cgroup_sockets_enabled && mem_cgroup_from_sk(sk);
2617 }
2618 
2619 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2620 {
2621 	struct mem_cgroup *memcg = mem_cgroup_from_sk(sk);
2622 
2623 #ifdef CONFIG_MEMCG_V1
2624 	if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2625 		return !!memcg->tcpmem_pressure;
2626 #endif /* CONFIG_MEMCG_V1 */
2627 
2628 	do {
2629 		if (time_before64(get_jiffies_64(), mem_cgroup_get_socket_pressure(memcg)))
2630 			return true;
2631 	} while ((memcg = parent_mem_cgroup(memcg)));
2632 
2633 	return false;
2634 }
2635 #else
2636 static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2637 {
2638 	return NULL;
2639 }
2640 
2641 static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2642 {
2643 	return false;
2644 }
2645 
2646 static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2647 {
2648 	return false;
2649 }
2650 #endif
2651 
2652 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2653 {
2654 	return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo);
2655 }
2656 
2657 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2658 {
2659 	return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo);
2660 }
2661 
2662 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2663 {
2664 	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2665 
2666 	return v ?: 1;
2667 }
2668 
2669 /* Alas, with timeout socket operations are not restartable.
2670  * Compare this to poll().
2671  */
2672 static inline int sock_intr_errno(long timeo)
2673 {
2674 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2675 }
2676 
2677 struct sock_skb_cb {
2678 	u32 dropcount;
2679 };
2680 
2681 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2682  * using skb->cb[] would keep using it directly and utilize its
2683  * alignment guarantee.
2684  */
2685 #define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \
2686 			    sizeof(struct sock_skb_cb))
2687 
2688 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2689 			    SOCK_SKB_CB_OFFSET))
2690 
2691 #define sock_skb_cb_check_size(size) \
2692 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2693 
2694 static inline void sk_drops_add(struct sock *sk, int segs)
2695 {
2696 	struct numa_drop_counters *ndc = sk->sk_drop_counters;
2697 
2698 	if (ndc)
2699 		numa_drop_add(ndc, segs);
2700 	else
2701 		atomic_add(segs, &sk->sk_drops);
2702 }
2703 
2704 static inline void sk_drops_inc(struct sock *sk)
2705 {
2706 	sk_drops_add(sk, 1);
2707 }
2708 
2709 static inline int sk_drops_read(const struct sock *sk)
2710 {
2711 	const struct numa_drop_counters *ndc = sk->sk_drop_counters;
2712 
2713 	if (ndc) {
2714 		DEBUG_NET_WARN_ON_ONCE(atomic_read(&sk->sk_drops));
2715 		return numa_drop_read(ndc);
2716 	}
2717 	return atomic_read(&sk->sk_drops);
2718 }
2719 
2720 static inline void sk_drops_reset(struct sock *sk)
2721 {
2722 	struct numa_drop_counters *ndc = sk->sk_drop_counters;
2723 
2724 	if (ndc)
2725 		numa_drop_reset(ndc);
2726 	atomic_set(&sk->sk_drops, 0);
2727 }
2728 
2729 static inline void
2730 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2731 {
2732 	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2733 						sk_drops_read(sk) : 0;
2734 }
2735 
2736 static inline void sk_drops_skbadd(struct sock *sk, const struct sk_buff *skb)
2737 {
2738 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2739 
2740 	sk_drops_add(sk, segs);
2741 }
2742 
2743 static inline ktime_t sock_read_timestamp(struct sock *sk)
2744 {
2745 #if BITS_PER_LONG==32
2746 	unsigned int seq;
2747 	ktime_t kt;
2748 
2749 	do {
2750 		seq = read_seqbegin(&sk->sk_stamp_seq);
2751 		kt = sk->sk_stamp;
2752 	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2753 
2754 	return kt;
2755 #else
2756 	return READ_ONCE(sk->sk_stamp);
2757 #endif
2758 }
2759 
2760 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2761 {
2762 #if BITS_PER_LONG==32
2763 	write_seqlock(&sk->sk_stamp_seq);
2764 	sk->sk_stamp = kt;
2765 	write_sequnlock(&sk->sk_stamp_seq);
2766 #else
2767 	WRITE_ONCE(sk->sk_stamp, kt);
2768 #endif
2769 }
2770 
2771 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2772 			   struct sk_buff *skb);
2773 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2774 			     struct sk_buff *skb);
2775 
2776 bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk);
2777 int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk,
2778 			 struct timespec64 *ts);
2779 
2780 static inline void
2781 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2782 {
2783 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2784 	u32 tsflags = READ_ONCE(sk->sk_tsflags);
2785 	ktime_t kt = skb->tstamp;
2786 	/*
2787 	 * generate control messages if
2788 	 * - receive time stamping in software requested
2789 	 * - software time stamp available and wanted
2790 	 * - hardware time stamps available and wanted
2791 	 */
2792 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2793 	    (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2794 	    (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2795 	    (hwtstamps->hwtstamp &&
2796 	     (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2797 		__sock_recv_timestamp(msg, sk, skb);
2798 	else
2799 		sock_write_timestamp(sk, kt);
2800 
2801 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2802 		__sock_recv_wifi_status(msg, sk, skb);
2803 }
2804 
2805 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2806 		       struct sk_buff *skb);
2807 
2808 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2809 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2810 				   struct sk_buff *skb)
2811 {
2812 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2813 			   (1UL << SOCK_RCVTSTAMP)			| \
2814 			   (1UL << SOCK_RCVMARK)			| \
2815 			   (1UL << SOCK_RCVPRIORITY)			| \
2816 			   (1UL << SOCK_TIMESTAMPING_ANY))
2817 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2818 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2819 
2820 	if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS)
2821 		__sock_recv_cmsgs(msg, sk, skb);
2822 	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2823 		sock_write_timestamp(sk, skb->tstamp);
2824 	else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2825 		sock_write_timestamp(sk, 0);
2826 }
2827 
2828 void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags);
2829 
2830 /**
2831  * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2832  * @sk:		socket sending this packet
2833  * @sockc:	pointer to socket cmsg cookie to get timestamping info
2834  * @tx_flags:	completed with instructions for time stamping
2835  * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2836  *
2837  * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2838  */
2839 static inline void _sock_tx_timestamp(struct sock *sk,
2840 				      const struct sockcm_cookie *sockc,
2841 				      __u8 *tx_flags, __u32 *tskey)
2842 {
2843 	__u32 tsflags = sockc->tsflags;
2844 
2845 	if (unlikely(tsflags)) {
2846 		__sock_tx_timestamp(tsflags, tx_flags);
2847 		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2848 		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) {
2849 			if (tsflags & SOCKCM_FLAG_TS_OPT_ID)
2850 				*tskey = sockc->ts_opt_id;
2851 			else
2852 				*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2853 		}
2854 	}
2855 }
2856 
2857 static inline void sock_tx_timestamp(struct sock *sk,
2858 				     const struct sockcm_cookie *sockc,
2859 				     __u8 *tx_flags)
2860 {
2861 	_sock_tx_timestamp(sk, sockc, tx_flags, NULL);
2862 }
2863 
2864 static inline void skb_setup_tx_timestamp(struct sk_buff *skb,
2865 					  const struct sockcm_cookie *sockc)
2866 {
2867 	_sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags,
2868 			   &skb_shinfo(skb)->tskey);
2869 }
2870 
2871 static inline bool sk_is_inet(const struct sock *sk)
2872 {
2873 	int family = READ_ONCE(sk->sk_family);
2874 
2875 	return family == AF_INET || family == AF_INET6;
2876 }
2877 
2878 static inline bool sk_is_tcp(const struct sock *sk)
2879 {
2880 	return sk_is_inet(sk) &&
2881 	       sk->sk_type == SOCK_STREAM &&
2882 	       sk->sk_protocol == IPPROTO_TCP;
2883 }
2884 
2885 static inline bool sk_is_udp(const struct sock *sk)
2886 {
2887 	return sk_is_inet(sk) &&
2888 	       sk->sk_type == SOCK_DGRAM &&
2889 	       sk->sk_protocol == IPPROTO_UDP;
2890 }
2891 
2892 static inline bool sk_is_unix(const struct sock *sk)
2893 {
2894 	return sk->sk_family == AF_UNIX;
2895 }
2896 
2897 static inline bool sk_is_stream_unix(const struct sock *sk)
2898 {
2899 	return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM;
2900 }
2901 
2902 static inline bool sk_is_vsock(const struct sock *sk)
2903 {
2904 	return sk->sk_family == AF_VSOCK;
2905 }
2906 
2907 static inline bool sk_may_scm_recv(const struct sock *sk)
2908 {
2909 	return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) ||
2910 		sk->sk_family == AF_NETLINK ||
2911 		(IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH);
2912 }
2913 
2914 /**
2915  * sk_eat_skb - Release a skb if it is no longer needed
2916  * @sk: socket to eat this skb from
2917  * @skb: socket buffer to eat
2918  *
2919  * This routine must be called with interrupts disabled or with the socket
2920  * locked so that the sk_buff queue operation is ok.
2921 */
2922 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2923 {
2924 	__skb_unlink(skb, &sk->sk_receive_queue);
2925 	__kfree_skb(skb);
2926 }
2927 
2928 static inline bool
2929 skb_sk_is_prefetched(struct sk_buff *skb)
2930 {
2931 #ifdef CONFIG_INET
2932 	return skb->destructor == sock_pfree;
2933 #else
2934 	return false;
2935 #endif /* CONFIG_INET */
2936 }
2937 
2938 /* This helper checks if a socket is a full socket,
2939  * ie _not_ a timewait or request socket.
2940  */
2941 static inline bool sk_fullsock(const struct sock *sk)
2942 {
2943 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2944 }
2945 
2946 static inline bool
2947 sk_is_refcounted(struct sock *sk)
2948 {
2949 	/* Only full sockets have sk->sk_flags. */
2950 	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2951 }
2952 
2953 static inline bool
2954 sk_requests_wifi_status(struct sock *sk)
2955 {
2956 	return sk && sk_fullsock(sk) && sock_flag(sk, SOCK_WIFI_STATUS);
2957 }
2958 
2959 /* Checks if this SKB belongs to an HW offloaded socket
2960  * and whether any SW fallbacks are required based on dev.
2961  * Check decrypted mark in case skb_orphan() cleared socket.
2962  */
2963 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2964 						   struct net_device *dev)
2965 {
2966 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2967 	struct sock *sk = skb->sk;
2968 
2969 	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2970 		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2971 	} else if (unlikely(skb_is_decrypted(skb))) {
2972 		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2973 		kfree_skb(skb);
2974 		skb = NULL;
2975 	}
2976 #endif
2977 
2978 	return skb;
2979 }
2980 
2981 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2982  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2983  */
2984 static inline bool sk_listener(const struct sock *sk)
2985 {
2986 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2987 }
2988 
2989 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT
2990  * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE)
2991  * TCP RST and ACK can be attached to TIME_WAIT.
2992  */
2993 static inline bool sk_listener_or_tw(const struct sock *sk)
2994 {
2995 	return (1 << READ_ONCE(sk->sk_state)) &
2996 	       (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT);
2997 }
2998 
2999 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
3000 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
3001 		       int type);
3002 
3003 bool sk_ns_capable(const struct sock *sk,
3004 		   struct user_namespace *user_ns, int cap);
3005 bool sk_capable(const struct sock *sk, int cap);
3006 bool sk_net_capable(const struct sock *sk, int cap);
3007 
3008 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
3009 
3010 /* Take into consideration the size of the struct sk_buff overhead in the
3011  * determination of these values, since that is non-constant across
3012  * platforms.  This makes socket queueing behavior and performance
3013  * not depend upon such differences.
3014  */
3015 #define _SK_MEM_PACKETS		256
3016 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
3017 #define SK_WMEM_DEFAULT		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3018 #define SK_RMEM_DEFAULT		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3019 
3020 extern __u32 sysctl_wmem_max;
3021 extern __u32 sysctl_rmem_max;
3022 
3023 extern __u32 sysctl_wmem_default;
3024 extern __u32 sysctl_rmem_default;
3025 
3026 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
3027 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3028 
3029 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
3030 {
3031 	/* Does this proto have per netns sysctl_wmem ? */
3032 	if (proto->sysctl_wmem_offset)
3033 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
3034 
3035 	return READ_ONCE(*proto->sysctl_wmem);
3036 }
3037 
3038 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
3039 {
3040 	/* Does this proto have per netns sysctl_rmem ? */
3041 	if (proto->sysctl_rmem_offset)
3042 		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
3043 
3044 	return READ_ONCE(*proto->sysctl_rmem);
3045 }
3046 
3047 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
3048  * Some wifi drivers need to tweak it to get more chunks.
3049  * They can use this helper from their ndo_start_xmit()
3050  */
3051 static inline void sk_pacing_shift_update(struct sock *sk, int val)
3052 {
3053 	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
3054 		return;
3055 	WRITE_ONCE(sk->sk_pacing_shift, val);
3056 }
3057 
3058 /* if a socket is bound to a device, check that the given device
3059  * index is either the same or that the socket is bound to an L3
3060  * master device and the given device index is also enslaved to
3061  * that L3 master
3062  */
3063 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
3064 {
3065 	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
3066 	int mdif;
3067 
3068 	if (!bound_dev_if || bound_dev_if == dif)
3069 		return true;
3070 
3071 	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
3072 	if (mdif && mdif == bound_dev_if)
3073 		return true;
3074 
3075 	return false;
3076 }
3077 
3078 void sock_def_readable(struct sock *sk);
3079 
3080 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
3081 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
3082 int sock_set_timestamping(struct sock *sk, int optname,
3083 			  struct so_timestamping timestamping);
3084 
3085 #if defined(CONFIG_CGROUP_BPF)
3086 void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op);
3087 #else
3088 static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
3089 {
3090 }
3091 #endif
3092 void sock_no_linger(struct sock *sk);
3093 void sock_set_keepalive(struct sock *sk);
3094 void sock_set_priority(struct sock *sk, u32 priority);
3095 void sock_set_rcvbuf(struct sock *sk, int val);
3096 void sock_set_mark(struct sock *sk, u32 val);
3097 void sock_set_reuseaddr(struct sock *sk);
3098 void sock_set_reuseport(struct sock *sk);
3099 void sock_set_sndtimeo(struct sock *sk, s64 secs);
3100 
3101 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
3102 
3103 int sock_get_timeout(long timeo, void *optval, bool old_timeval);
3104 int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
3105 			   sockptr_t optval, int optlen, bool old_timeval);
3106 
3107 int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
3108 		     void __user *arg, void *karg, size_t size);
3109 int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
3110 static inline bool sk_is_readable(struct sock *sk)
3111 {
3112 	const struct proto *prot = READ_ONCE(sk->sk_prot);
3113 
3114 	if (prot->sock_is_readable)
3115 		return prot->sock_is_readable(sk);
3116 
3117 	return false;
3118 }
3119 #endif	/* _SOCK_H */
3120