xref: /linux/include/net/sock.h (revision f04ed7d277e842af9934b71b529341d1ba31a9c1)
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/filter.h>
60 #include <linux/rculist_nulls.h>
61 #include <linux/poll.h>
62 #include <linux/sockptr.h>
63 #include <linux/indirect_call_wrapper.h>
64 #include <linux/atomic.h>
65 #include <linux/refcount.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 /* Define this to get the SOCK_DBG debugging facility. */
80 #define SOCK_DEBUGGING
81 #ifdef SOCK_DEBUGGING
82 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
83 					printk(KERN_DEBUG msg); } while (0)
84 #else
85 /* Validate arguments and do nothing */
86 static inline __printf(2, 3)
87 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
88 {
89 }
90 #endif
91 
92 /* This is the per-socket lock.  The spinlock provides a synchronization
93  * between user contexts and software interrupt processing, whereas the
94  * mini-semaphore synchronizes multiple users amongst themselves.
95  */
96 typedef struct {
97 	spinlock_t		slock;
98 	int			owned;
99 	wait_queue_head_t	wq;
100 	/*
101 	 * We express the mutex-alike socket_lock semantics
102 	 * to the lock validator by explicitly managing
103 	 * the slock as a lock variant (in addition to
104 	 * the slock itself):
105 	 */
106 #ifdef CONFIG_DEBUG_LOCK_ALLOC
107 	struct lockdep_map dep_map;
108 #endif
109 } socket_lock_t;
110 
111 struct sock;
112 struct proto;
113 struct net;
114 
115 typedef __u32 __bitwise __portpair;
116 typedef __u64 __bitwise __addrpair;
117 
118 /**
119  *	struct sock_common - minimal network layer representation of sockets
120  *	@skc_daddr: Foreign IPv4 addr
121  *	@skc_rcv_saddr: Bound local IPv4 addr
122  *	@skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
123  *	@skc_hash: hash value used with various protocol lookup tables
124  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
125  *	@skc_dport: placeholder for inet_dport/tw_dport
126  *	@skc_num: placeholder for inet_num/tw_num
127  *	@skc_portpair: __u32 union of @skc_dport & @skc_num
128  *	@skc_family: network address family
129  *	@skc_state: Connection state
130  *	@skc_reuse: %SO_REUSEADDR setting
131  *	@skc_reuseport: %SO_REUSEPORT setting
132  *	@skc_ipv6only: socket is IPV6 only
133  *	@skc_net_refcnt: socket is using net ref counting
134  *	@skc_bound_dev_if: bound device index if != 0
135  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
136  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137  *	@skc_prot: protocol handlers inside a network family
138  *	@skc_net: reference to the network namespace of this socket
139  *	@skc_v6_daddr: IPV6 destination address
140  *	@skc_v6_rcv_saddr: IPV6 source address
141  *	@skc_cookie: socket's cookie value
142  *	@skc_node: main hash linkage for various protocol lookup tables
143  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
144  *	@skc_tx_queue_mapping: tx queue number for this connection
145  *	@skc_rx_queue_mapping: rx queue number for this connection
146  *	@skc_flags: place holder for sk_flags
147  *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
148  *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
149  *	@skc_listener: connection request listener socket (aka rsk_listener)
150  *		[union with @skc_flags]
151  *	@skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
152  *		[union with @skc_flags]
153  *	@skc_incoming_cpu: record/match cpu processing incoming packets
154  *	@skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
155  *		[union with @skc_incoming_cpu]
156  *	@skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
157  *		[union with @skc_incoming_cpu]
158  *	@skc_refcnt: reference count
159  *
160  *	This is the minimal network layer representation of sockets, the header
161  *	for struct sock and struct inet_timewait_sock.
162  */
163 struct sock_common {
164 	/* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
165 	 * address on 64bit arches : cf INET_MATCH()
166 	 */
167 	union {
168 		__addrpair	skc_addrpair;
169 		struct {
170 			__be32	skc_daddr;
171 			__be32	skc_rcv_saddr;
172 		};
173 	};
174 	union  {
175 		unsigned int	skc_hash;
176 		__u16		skc_u16hashes[2];
177 	};
178 	/* skc_dport && skc_num must be grouped as well */
179 	union {
180 		__portpair	skc_portpair;
181 		struct {
182 			__be16	skc_dport;
183 			__u16	skc_num;
184 		};
185 	};
186 
187 	unsigned short		skc_family;
188 	volatile unsigned char	skc_state;
189 	unsigned char		skc_reuse:4;
190 	unsigned char		skc_reuseport:1;
191 	unsigned char		skc_ipv6only:1;
192 	unsigned char		skc_net_refcnt:1;
193 	int			skc_bound_dev_if;
194 	union {
195 		struct hlist_node	skc_bind_node;
196 		struct hlist_node	skc_portaddr_node;
197 	};
198 	struct proto		*skc_prot;
199 	possible_net_t		skc_net;
200 
201 #if IS_ENABLED(CONFIG_IPV6)
202 	struct in6_addr		skc_v6_daddr;
203 	struct in6_addr		skc_v6_rcv_saddr;
204 #endif
205 
206 	atomic64_t		skc_cookie;
207 
208 	/* following fields are padding to force
209 	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
210 	 * assuming IPV6 is enabled. We use this padding differently
211 	 * for different kind of 'sockets'
212 	 */
213 	union {
214 		unsigned long	skc_flags;
215 		struct sock	*skc_listener; /* request_sock */
216 		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
217 	};
218 	/*
219 	 * fields between dontcopy_begin/dontcopy_end
220 	 * are not copied in sock_copy()
221 	 */
222 	/* private: */
223 	int			skc_dontcopy_begin[0];
224 	/* public: */
225 	union {
226 		struct hlist_node	skc_node;
227 		struct hlist_nulls_node skc_nulls_node;
228 	};
229 	unsigned short		skc_tx_queue_mapping;
230 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
231 	unsigned short		skc_rx_queue_mapping;
232 #endif
233 	union {
234 		int		skc_incoming_cpu;
235 		u32		skc_rcv_wnd;
236 		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
237 	};
238 
239 	refcount_t		skc_refcnt;
240 	/* private: */
241 	int                     skc_dontcopy_end[0];
242 	union {
243 		u32		skc_rxhash;
244 		u32		skc_window_clamp;
245 		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
246 	};
247 	/* public: */
248 };
249 
250 struct bpf_local_storage;
251 
252 /**
253   *	struct sock - network layer representation of sockets
254   *	@__sk_common: shared layout with inet_timewait_sock
255   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
256   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
257   *	@sk_lock:	synchronizer
258   *	@sk_kern_sock: True if sock is using kernel lock classes
259   *	@sk_rcvbuf: size of receive buffer in bytes
260   *	@sk_wq: sock wait queue and async head
261   *	@sk_rx_dst: receive input route used by early demux
262   *	@sk_dst_cache: destination cache
263   *	@sk_dst_pending_confirm: need to confirm neighbour
264   *	@sk_policy: flow policy
265   *	@sk_receive_queue: incoming packets
266   *	@sk_wmem_alloc: transmit queue bytes committed
267   *	@sk_tsq_flags: TCP Small Queues flags
268   *	@sk_write_queue: Packet sending queue
269   *	@sk_omem_alloc: "o" is "option" or "other"
270   *	@sk_wmem_queued: persistent queue size
271   *	@sk_forward_alloc: space allocated forward
272   *	@sk_napi_id: id of the last napi context to receive data for sk
273   *	@sk_ll_usec: usecs to busypoll when there is no data
274   *	@sk_allocation: allocation mode
275   *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
276   *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
277   *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
278   *	@sk_sndbuf: size of send buffer in bytes
279   *	@__sk_flags_offset: empty field used to determine location of bitfield
280   *	@sk_padding: unused element for alignment
281   *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
282   *	@sk_no_check_rx: allow zero checksum in RX packets
283   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
284   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
285   *	@sk_route_forced_caps: static, forced route capabilities
286   *		(set in tcp_init_sock())
287   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
288   *	@sk_gso_max_size: Maximum GSO segment size to build
289   *	@sk_gso_max_segs: Maximum number of GSO segments
290   *	@sk_pacing_shift: scaling factor for TCP Small Queues
291   *	@sk_lingertime: %SO_LINGER l_linger setting
292   *	@sk_backlog: always used with the per-socket spinlock held
293   *	@sk_callback_lock: used with the callbacks in the end of this struct
294   *	@sk_error_queue: rarely used
295   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
296   *			  IPV6_ADDRFORM for instance)
297   *	@sk_err: last error
298   *	@sk_err_soft: errors that don't cause failure but are the cause of a
299   *		      persistent failure not just 'timed out'
300   *	@sk_drops: raw/udp drops counter
301   *	@sk_ack_backlog: current listen backlog
302   *	@sk_max_ack_backlog: listen backlog set in listen()
303   *	@sk_uid: user id of owner
304   *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
305   *	@sk_busy_poll_budget: napi processing budget when busypolling
306   *	@sk_priority: %SO_PRIORITY setting
307   *	@sk_type: socket type (%SOCK_STREAM, etc)
308   *	@sk_protocol: which protocol this socket belongs in this network family
309   *	@sk_peer_pid: &struct pid for this socket's peer
310   *	@sk_peer_cred: %SO_PEERCRED setting
311   *	@sk_rcvlowat: %SO_RCVLOWAT setting
312   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
313   *	@sk_sndtimeo: %SO_SNDTIMEO setting
314   *	@sk_txhash: computed flow hash for use on transmit
315   *	@sk_filter: socket filtering instructions
316   *	@sk_timer: sock cleanup timer
317   *	@sk_stamp: time stamp of last packet received
318   *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
319   *	@sk_tsflags: SO_TIMESTAMPING flags
320   *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
321   *	              for timestamping
322   *	@sk_tskey: counter to disambiguate concurrent tstamp requests
323   *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
324   *	@sk_socket: Identd and reporting IO signals
325   *	@sk_user_data: RPC layer private data
326   *	@sk_frag: cached page frag
327   *	@sk_peek_off: current peek_offset value
328   *	@sk_send_head: front of stuff to transmit
329   *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
330   *	@sk_security: used by security modules
331   *	@sk_mark: generic packet mark
332   *	@sk_cgrp_data: cgroup data for this cgroup
333   *	@sk_memcg: this socket's memory cgroup association
334   *	@sk_write_pending: a write to stream socket waits to start
335   *	@sk_state_change: callback to indicate change in the state of the sock
336   *	@sk_data_ready: callback to indicate there is data to be processed
337   *	@sk_write_space: callback to indicate there is bf sending space available
338   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
339   *	@sk_backlog_rcv: callback to process the backlog
340   *	@sk_validate_xmit_skb: ptr to an optional validate function
341   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
342   *	@sk_reuseport_cb: reuseport group container
343   *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
344   *	@sk_rcu: used during RCU grace period
345   *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
346   *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
347   *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
348   *	@sk_txtime_unused: unused txtime flags
349   */
350 struct sock {
351 	/*
352 	 * Now struct inet_timewait_sock also uses sock_common, so please just
353 	 * don't add nothing before this first member (__sk_common) --acme
354 	 */
355 	struct sock_common	__sk_common;
356 #define sk_node			__sk_common.skc_node
357 #define sk_nulls_node		__sk_common.skc_nulls_node
358 #define sk_refcnt		__sk_common.skc_refcnt
359 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
360 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
361 #define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
362 #endif
363 
364 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
365 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
366 #define sk_hash			__sk_common.skc_hash
367 #define sk_portpair		__sk_common.skc_portpair
368 #define sk_num			__sk_common.skc_num
369 #define sk_dport		__sk_common.skc_dport
370 #define sk_addrpair		__sk_common.skc_addrpair
371 #define sk_daddr		__sk_common.skc_daddr
372 #define sk_rcv_saddr		__sk_common.skc_rcv_saddr
373 #define sk_family		__sk_common.skc_family
374 #define sk_state		__sk_common.skc_state
375 #define sk_reuse		__sk_common.skc_reuse
376 #define sk_reuseport		__sk_common.skc_reuseport
377 #define sk_ipv6only		__sk_common.skc_ipv6only
378 #define sk_net_refcnt		__sk_common.skc_net_refcnt
379 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
380 #define sk_bind_node		__sk_common.skc_bind_node
381 #define sk_prot			__sk_common.skc_prot
382 #define sk_net			__sk_common.skc_net
383 #define sk_v6_daddr		__sk_common.skc_v6_daddr
384 #define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
385 #define sk_cookie		__sk_common.skc_cookie
386 #define sk_incoming_cpu		__sk_common.skc_incoming_cpu
387 #define sk_flags		__sk_common.skc_flags
388 #define sk_rxhash		__sk_common.skc_rxhash
389 
390 	socket_lock_t		sk_lock;
391 	atomic_t		sk_drops;
392 	int			sk_rcvlowat;
393 	struct sk_buff_head	sk_error_queue;
394 	struct sk_buff_head	sk_receive_queue;
395 	/*
396 	 * The backlog queue is special, it is always used with
397 	 * the per-socket spinlock held and requires low latency
398 	 * access. Therefore we special case it's implementation.
399 	 * Note : rmem_alloc is in this structure to fill a hole
400 	 * on 64bit arches, not because its logically part of
401 	 * backlog.
402 	 */
403 	struct {
404 		atomic_t	rmem_alloc;
405 		int		len;
406 		struct sk_buff	*head;
407 		struct sk_buff	*tail;
408 	} sk_backlog;
409 #define sk_rmem_alloc sk_backlog.rmem_alloc
410 
411 	int			sk_forward_alloc;
412 #ifdef CONFIG_NET_RX_BUSY_POLL
413 	unsigned int		sk_ll_usec;
414 	/* ===== mostly read cache line ===== */
415 	unsigned int		sk_napi_id;
416 #endif
417 	int			sk_rcvbuf;
418 
419 	struct sk_filter __rcu	*sk_filter;
420 	union {
421 		struct socket_wq __rcu	*sk_wq;
422 		/* private: */
423 		struct socket_wq	*sk_wq_raw;
424 		/* public: */
425 	};
426 #ifdef CONFIG_XFRM
427 	struct xfrm_policy __rcu *sk_policy[2];
428 #endif
429 	struct dst_entry	*sk_rx_dst;
430 	struct dst_entry __rcu	*sk_dst_cache;
431 	atomic_t		sk_omem_alloc;
432 	int			sk_sndbuf;
433 
434 	/* ===== cache line for TX ===== */
435 	int			sk_wmem_queued;
436 	refcount_t		sk_wmem_alloc;
437 	unsigned long		sk_tsq_flags;
438 	union {
439 		struct sk_buff	*sk_send_head;
440 		struct rb_root	tcp_rtx_queue;
441 	};
442 	struct sk_buff_head	sk_write_queue;
443 	__s32			sk_peek_off;
444 	int			sk_write_pending;
445 	__u32			sk_dst_pending_confirm;
446 	u32			sk_pacing_status; /* see enum sk_pacing */
447 	long			sk_sndtimeo;
448 	struct timer_list	sk_timer;
449 	__u32			sk_priority;
450 	__u32			sk_mark;
451 	unsigned long		sk_pacing_rate; /* bytes per second */
452 	unsigned long		sk_max_pacing_rate;
453 	struct page_frag	sk_frag;
454 	netdev_features_t	sk_route_caps;
455 	netdev_features_t	sk_route_nocaps;
456 	netdev_features_t	sk_route_forced_caps;
457 	int			sk_gso_type;
458 	unsigned int		sk_gso_max_size;
459 	gfp_t			sk_allocation;
460 	__u32			sk_txhash;
461 
462 	/*
463 	 * Because of non atomicity rules, all
464 	 * changes are protected by socket lock.
465 	 */
466 	u8			sk_padding : 1,
467 				sk_kern_sock : 1,
468 				sk_no_check_tx : 1,
469 				sk_no_check_rx : 1,
470 				sk_userlocks : 4;
471 	u8			sk_pacing_shift;
472 	u16			sk_type;
473 	u16			sk_protocol;
474 	u16			sk_gso_max_segs;
475 	unsigned long	        sk_lingertime;
476 	struct proto		*sk_prot_creator;
477 	rwlock_t		sk_callback_lock;
478 	int			sk_err,
479 				sk_err_soft;
480 	u32			sk_ack_backlog;
481 	u32			sk_max_ack_backlog;
482 	kuid_t			sk_uid;
483 #ifdef CONFIG_NET_RX_BUSY_POLL
484 	u8			sk_prefer_busy_poll;
485 	u16			sk_busy_poll_budget;
486 #endif
487 	struct pid		*sk_peer_pid;
488 	const struct cred	*sk_peer_cred;
489 	long			sk_rcvtimeo;
490 	ktime_t			sk_stamp;
491 #if BITS_PER_LONG==32
492 	seqlock_t		sk_stamp_seq;
493 #endif
494 	u16			sk_tsflags;
495 	int			sk_bind_phc;
496 	u8			sk_shutdown;
497 	u32			sk_tskey;
498 	atomic_t		sk_zckey;
499 
500 	u8			sk_clockid;
501 	u8			sk_txtime_deadline_mode : 1,
502 				sk_txtime_report_errors : 1,
503 				sk_txtime_unused : 6;
504 
505 	struct socket		*sk_socket;
506 	void			*sk_user_data;
507 #ifdef CONFIG_SECURITY
508 	void			*sk_security;
509 #endif
510 	struct sock_cgroup_data	sk_cgrp_data;
511 	struct mem_cgroup	*sk_memcg;
512 	void			(*sk_state_change)(struct sock *sk);
513 	void			(*sk_data_ready)(struct sock *sk);
514 	void			(*sk_write_space)(struct sock *sk);
515 	void			(*sk_error_report)(struct sock *sk);
516 	int			(*sk_backlog_rcv)(struct sock *sk,
517 						  struct sk_buff *skb);
518 #ifdef CONFIG_SOCK_VALIDATE_XMIT
519 	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
520 							struct net_device *dev,
521 							struct sk_buff *skb);
522 #endif
523 	void                    (*sk_destruct)(struct sock *sk);
524 	struct sock_reuseport __rcu	*sk_reuseport_cb;
525 #ifdef CONFIG_BPF_SYSCALL
526 	struct bpf_local_storage __rcu	*sk_bpf_storage;
527 #endif
528 	struct rcu_head		sk_rcu;
529 };
530 
531 enum sk_pacing {
532 	SK_PACING_NONE		= 0,
533 	SK_PACING_NEEDED	= 1,
534 	SK_PACING_FQ		= 2,
535 };
536 
537 /* Pointer stored in sk_user_data might not be suitable for copying
538  * when cloning the socket. For instance, it can point to a reference
539  * counted object. sk_user_data bottom bit is set if pointer must not
540  * be copied.
541  */
542 #define SK_USER_DATA_NOCOPY	1UL
543 #define SK_USER_DATA_BPF	2UL	/* Managed by BPF */
544 #define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF)
545 
546 /**
547  * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
548  * @sk: socket
549  */
550 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
551 {
552 	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
553 }
554 
555 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
556 
557 #define rcu_dereference_sk_user_data(sk)				\
558 ({									\
559 	void *__tmp = rcu_dereference(__sk_user_data((sk)));		\
560 	(void *)((uintptr_t)__tmp & SK_USER_DATA_PTRMASK);		\
561 })
562 #define rcu_assign_sk_user_data(sk, ptr)				\
563 ({									\
564 	uintptr_t __tmp = (uintptr_t)(ptr);				\
565 	WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK);			\
566 	rcu_assign_pointer(__sk_user_data((sk)), __tmp);		\
567 })
568 #define rcu_assign_sk_user_data_nocopy(sk, ptr)				\
569 ({									\
570 	uintptr_t __tmp = (uintptr_t)(ptr);				\
571 	WARN_ON_ONCE(__tmp & ~SK_USER_DATA_PTRMASK);			\
572 	rcu_assign_pointer(__sk_user_data((sk)),			\
573 			   __tmp | SK_USER_DATA_NOCOPY);		\
574 })
575 
576 /*
577  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
578  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
579  * on a socket means that the socket will reuse everybody else's port
580  * without looking at the other's sk_reuse value.
581  */
582 
583 #define SK_NO_REUSE	0
584 #define SK_CAN_REUSE	1
585 #define SK_FORCE_REUSE	2
586 
587 int sk_set_peek_off(struct sock *sk, int val);
588 
589 static inline int sk_peek_offset(struct sock *sk, int flags)
590 {
591 	if (unlikely(flags & MSG_PEEK)) {
592 		return READ_ONCE(sk->sk_peek_off);
593 	}
594 
595 	return 0;
596 }
597 
598 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
599 {
600 	s32 off = READ_ONCE(sk->sk_peek_off);
601 
602 	if (unlikely(off >= 0)) {
603 		off = max_t(s32, off - val, 0);
604 		WRITE_ONCE(sk->sk_peek_off, off);
605 	}
606 }
607 
608 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
609 {
610 	sk_peek_offset_bwd(sk, -val);
611 }
612 
613 /*
614  * Hashed lists helper routines
615  */
616 static inline struct sock *sk_entry(const struct hlist_node *node)
617 {
618 	return hlist_entry(node, struct sock, sk_node);
619 }
620 
621 static inline struct sock *__sk_head(const struct hlist_head *head)
622 {
623 	return hlist_entry(head->first, struct sock, sk_node);
624 }
625 
626 static inline struct sock *sk_head(const struct hlist_head *head)
627 {
628 	return hlist_empty(head) ? NULL : __sk_head(head);
629 }
630 
631 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
632 {
633 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
634 }
635 
636 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
637 {
638 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
639 }
640 
641 static inline struct sock *sk_next(const struct sock *sk)
642 {
643 	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
644 }
645 
646 static inline struct sock *sk_nulls_next(const struct sock *sk)
647 {
648 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
649 		hlist_nulls_entry(sk->sk_nulls_node.next,
650 				  struct sock, sk_nulls_node) :
651 		NULL;
652 }
653 
654 static inline bool sk_unhashed(const struct sock *sk)
655 {
656 	return hlist_unhashed(&sk->sk_node);
657 }
658 
659 static inline bool sk_hashed(const struct sock *sk)
660 {
661 	return !sk_unhashed(sk);
662 }
663 
664 static inline void sk_node_init(struct hlist_node *node)
665 {
666 	node->pprev = NULL;
667 }
668 
669 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
670 {
671 	node->pprev = NULL;
672 }
673 
674 static inline void __sk_del_node(struct sock *sk)
675 {
676 	__hlist_del(&sk->sk_node);
677 }
678 
679 /* NB: equivalent to hlist_del_init_rcu */
680 static inline bool __sk_del_node_init(struct sock *sk)
681 {
682 	if (sk_hashed(sk)) {
683 		__sk_del_node(sk);
684 		sk_node_init(&sk->sk_node);
685 		return true;
686 	}
687 	return false;
688 }
689 
690 /* Grab socket reference count. This operation is valid only
691    when sk is ALREADY grabbed f.e. it is found in hash table
692    or a list and the lookup is made under lock preventing hash table
693    modifications.
694  */
695 
696 static __always_inline void sock_hold(struct sock *sk)
697 {
698 	refcount_inc(&sk->sk_refcnt);
699 }
700 
701 /* Ungrab socket in the context, which assumes that socket refcnt
702    cannot hit zero, f.e. it is true in context of any socketcall.
703  */
704 static __always_inline void __sock_put(struct sock *sk)
705 {
706 	refcount_dec(&sk->sk_refcnt);
707 }
708 
709 static inline bool sk_del_node_init(struct sock *sk)
710 {
711 	bool rc = __sk_del_node_init(sk);
712 
713 	if (rc) {
714 		/* paranoid for a while -acme */
715 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
716 		__sock_put(sk);
717 	}
718 	return rc;
719 }
720 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
721 
722 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
723 {
724 	if (sk_hashed(sk)) {
725 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
726 		return true;
727 	}
728 	return false;
729 }
730 
731 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
732 {
733 	bool rc = __sk_nulls_del_node_init_rcu(sk);
734 
735 	if (rc) {
736 		/* paranoid for a while -acme */
737 		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
738 		__sock_put(sk);
739 	}
740 	return rc;
741 }
742 
743 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
744 {
745 	hlist_add_head(&sk->sk_node, list);
746 }
747 
748 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
749 {
750 	sock_hold(sk);
751 	__sk_add_node(sk, list);
752 }
753 
754 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
755 {
756 	sock_hold(sk);
757 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
758 	    sk->sk_family == AF_INET6)
759 		hlist_add_tail_rcu(&sk->sk_node, list);
760 	else
761 		hlist_add_head_rcu(&sk->sk_node, list);
762 }
763 
764 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
765 {
766 	sock_hold(sk);
767 	hlist_add_tail_rcu(&sk->sk_node, list);
768 }
769 
770 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
771 {
772 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
773 }
774 
775 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
776 {
777 	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
778 }
779 
780 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
781 {
782 	sock_hold(sk);
783 	__sk_nulls_add_node_rcu(sk, list);
784 }
785 
786 static inline void __sk_del_bind_node(struct sock *sk)
787 {
788 	__hlist_del(&sk->sk_bind_node);
789 }
790 
791 static inline void sk_add_bind_node(struct sock *sk,
792 					struct hlist_head *list)
793 {
794 	hlist_add_head(&sk->sk_bind_node, list);
795 }
796 
797 #define sk_for_each(__sk, list) \
798 	hlist_for_each_entry(__sk, list, sk_node)
799 #define sk_for_each_rcu(__sk, list) \
800 	hlist_for_each_entry_rcu(__sk, list, sk_node)
801 #define sk_nulls_for_each(__sk, node, list) \
802 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
803 #define sk_nulls_for_each_rcu(__sk, node, list) \
804 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
805 #define sk_for_each_from(__sk) \
806 	hlist_for_each_entry_from(__sk, sk_node)
807 #define sk_nulls_for_each_from(__sk, node) \
808 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
809 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
810 #define sk_for_each_safe(__sk, tmp, list) \
811 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
812 #define sk_for_each_bound(__sk, list) \
813 	hlist_for_each_entry(__sk, list, sk_bind_node)
814 
815 /**
816  * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
817  * @tpos:	the type * to use as a loop cursor.
818  * @pos:	the &struct hlist_node to use as a loop cursor.
819  * @head:	the head for your list.
820  * @offset:	offset of hlist_node within the struct.
821  *
822  */
823 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
824 	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
825 	     pos != NULL &&						       \
826 		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
827 	     pos = rcu_dereference(hlist_next_rcu(pos)))
828 
829 static inline struct user_namespace *sk_user_ns(struct sock *sk)
830 {
831 	/* Careful only use this in a context where these parameters
832 	 * can not change and must all be valid, such as recvmsg from
833 	 * userspace.
834 	 */
835 	return sk->sk_socket->file->f_cred->user_ns;
836 }
837 
838 /* Sock flags */
839 enum sock_flags {
840 	SOCK_DEAD,
841 	SOCK_DONE,
842 	SOCK_URGINLINE,
843 	SOCK_KEEPOPEN,
844 	SOCK_LINGER,
845 	SOCK_DESTROY,
846 	SOCK_BROADCAST,
847 	SOCK_TIMESTAMP,
848 	SOCK_ZAPPED,
849 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
850 	SOCK_DBG, /* %SO_DEBUG setting */
851 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
852 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
853 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
854 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
855 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
856 	SOCK_FASYNC, /* fasync() active */
857 	SOCK_RXQ_OVFL,
858 	SOCK_ZEROCOPY, /* buffers from userspace */
859 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
860 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
861 		     * Will use last 4 bytes of packet sent from
862 		     * user-space instead.
863 		     */
864 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
865 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
866 	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
867 	SOCK_TXTIME,
868 	SOCK_XDP, /* XDP is attached */
869 	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
870 };
871 
872 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
873 
874 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
875 {
876 	nsk->sk_flags = osk->sk_flags;
877 }
878 
879 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
880 {
881 	__set_bit(flag, &sk->sk_flags);
882 }
883 
884 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
885 {
886 	__clear_bit(flag, &sk->sk_flags);
887 }
888 
889 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
890 				     int valbool)
891 {
892 	if (valbool)
893 		sock_set_flag(sk, bit);
894 	else
895 		sock_reset_flag(sk, bit);
896 }
897 
898 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
899 {
900 	return test_bit(flag, &sk->sk_flags);
901 }
902 
903 #ifdef CONFIG_NET
904 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
905 static inline int sk_memalloc_socks(void)
906 {
907 	return static_branch_unlikely(&memalloc_socks_key);
908 }
909 
910 void __receive_sock(struct file *file);
911 #else
912 
913 static inline int sk_memalloc_socks(void)
914 {
915 	return 0;
916 }
917 
918 static inline void __receive_sock(struct file *file)
919 { }
920 #endif
921 
922 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
923 {
924 	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
925 }
926 
927 static inline void sk_acceptq_removed(struct sock *sk)
928 {
929 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
930 }
931 
932 static inline void sk_acceptq_added(struct sock *sk)
933 {
934 	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
935 }
936 
937 /* Note: If you think the test should be:
938  *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
939  * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
940  */
941 static inline bool sk_acceptq_is_full(const struct sock *sk)
942 {
943 	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
944 }
945 
946 /*
947  * Compute minimal free write space needed to queue new packets.
948  */
949 static inline int sk_stream_min_wspace(const struct sock *sk)
950 {
951 	return READ_ONCE(sk->sk_wmem_queued) >> 1;
952 }
953 
954 static inline int sk_stream_wspace(const struct sock *sk)
955 {
956 	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
957 }
958 
959 static inline void sk_wmem_queued_add(struct sock *sk, int val)
960 {
961 	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
962 }
963 
964 void sk_stream_write_space(struct sock *sk);
965 
966 /* OOB backlog add */
967 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
968 {
969 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
970 	skb_dst_force(skb);
971 
972 	if (!sk->sk_backlog.tail)
973 		WRITE_ONCE(sk->sk_backlog.head, skb);
974 	else
975 		sk->sk_backlog.tail->next = skb;
976 
977 	WRITE_ONCE(sk->sk_backlog.tail, skb);
978 	skb->next = NULL;
979 }
980 
981 /*
982  * Take into account size of receive queue and backlog queue
983  * Do not take into account this skb truesize,
984  * to allow even a single big packet to come.
985  */
986 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
987 {
988 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
989 
990 	return qsize > limit;
991 }
992 
993 /* The per-socket spinlock must be held here. */
994 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
995 					      unsigned int limit)
996 {
997 	if (sk_rcvqueues_full(sk, limit))
998 		return -ENOBUFS;
999 
1000 	/*
1001 	 * If the skb was allocated from pfmemalloc reserves, only
1002 	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1003 	 * helping free memory
1004 	 */
1005 	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1006 		return -ENOMEM;
1007 
1008 	__sk_add_backlog(sk, skb);
1009 	sk->sk_backlog.len += skb->truesize;
1010 	return 0;
1011 }
1012 
1013 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1014 
1015 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1016 {
1017 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1018 		return __sk_backlog_rcv(sk, skb);
1019 
1020 	return sk->sk_backlog_rcv(sk, skb);
1021 }
1022 
1023 static inline void sk_incoming_cpu_update(struct sock *sk)
1024 {
1025 	int cpu = raw_smp_processor_id();
1026 
1027 	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1028 		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1029 }
1030 
1031 static inline void sock_rps_record_flow_hash(__u32 hash)
1032 {
1033 #ifdef CONFIG_RPS
1034 	struct rps_sock_flow_table *sock_flow_table;
1035 
1036 	rcu_read_lock();
1037 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
1038 	rps_record_sock_flow(sock_flow_table, hash);
1039 	rcu_read_unlock();
1040 #endif
1041 }
1042 
1043 static inline void sock_rps_record_flow(const struct sock *sk)
1044 {
1045 #ifdef CONFIG_RPS
1046 	if (static_branch_unlikely(&rfs_needed)) {
1047 		/* Reading sk->sk_rxhash might incur an expensive cache line
1048 		 * miss.
1049 		 *
1050 		 * TCP_ESTABLISHED does cover almost all states where RFS
1051 		 * might be useful, and is cheaper [1] than testing :
1052 		 *	IPv4: inet_sk(sk)->inet_daddr
1053 		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1054 		 * OR	an additional socket flag
1055 		 * [1] : sk_state and sk_prot are in the same cache line.
1056 		 */
1057 		if (sk->sk_state == TCP_ESTABLISHED)
1058 			sock_rps_record_flow_hash(sk->sk_rxhash);
1059 	}
1060 #endif
1061 }
1062 
1063 static inline void sock_rps_save_rxhash(struct sock *sk,
1064 					const struct sk_buff *skb)
1065 {
1066 #ifdef CONFIG_RPS
1067 	if (unlikely(sk->sk_rxhash != skb->hash))
1068 		sk->sk_rxhash = skb->hash;
1069 #endif
1070 }
1071 
1072 static inline void sock_rps_reset_rxhash(struct sock *sk)
1073 {
1074 #ifdef CONFIG_RPS
1075 	sk->sk_rxhash = 0;
1076 #endif
1077 }
1078 
1079 #define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1080 	({	int __rc;						\
1081 		release_sock(__sk);					\
1082 		__rc = __condition;					\
1083 		if (!__rc) {						\
1084 			*(__timeo) = wait_woken(__wait,			\
1085 						TASK_INTERRUPTIBLE,	\
1086 						*(__timeo));		\
1087 		}							\
1088 		sched_annotate_sleep();					\
1089 		lock_sock(__sk);					\
1090 		__rc = __condition;					\
1091 		__rc;							\
1092 	})
1093 
1094 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1095 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1096 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1097 int sk_stream_error(struct sock *sk, int flags, int err);
1098 void sk_stream_kill_queues(struct sock *sk);
1099 void sk_set_memalloc(struct sock *sk);
1100 void sk_clear_memalloc(struct sock *sk);
1101 
1102 void __sk_flush_backlog(struct sock *sk);
1103 
1104 static inline bool sk_flush_backlog(struct sock *sk)
1105 {
1106 	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1107 		__sk_flush_backlog(sk);
1108 		return true;
1109 	}
1110 	return false;
1111 }
1112 
1113 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1114 
1115 struct request_sock_ops;
1116 struct timewait_sock_ops;
1117 struct inet_hashinfo;
1118 struct raw_hashinfo;
1119 struct smc_hashinfo;
1120 struct module;
1121 struct sk_psock;
1122 
1123 /*
1124  * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1125  * un-modified. Special care is taken when initializing object to zero.
1126  */
1127 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1128 {
1129 	if (offsetof(struct sock, sk_node.next) != 0)
1130 		memset(sk, 0, offsetof(struct sock, sk_node.next));
1131 	memset(&sk->sk_node.pprev, 0,
1132 	       size - offsetof(struct sock, sk_node.pprev));
1133 }
1134 
1135 /* Networking protocol blocks we attach to sockets.
1136  * socket layer -> transport layer interface
1137  */
1138 struct proto {
1139 	void			(*close)(struct sock *sk,
1140 					long timeout);
1141 	int			(*pre_connect)(struct sock *sk,
1142 					struct sockaddr *uaddr,
1143 					int addr_len);
1144 	int			(*connect)(struct sock *sk,
1145 					struct sockaddr *uaddr,
1146 					int addr_len);
1147 	int			(*disconnect)(struct sock *sk, int flags);
1148 
1149 	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1150 					  bool kern);
1151 
1152 	int			(*ioctl)(struct sock *sk, int cmd,
1153 					 unsigned long arg);
1154 	int			(*init)(struct sock *sk);
1155 	void			(*destroy)(struct sock *sk);
1156 	void			(*shutdown)(struct sock *sk, int how);
1157 	int			(*setsockopt)(struct sock *sk, int level,
1158 					int optname, sockptr_t optval,
1159 					unsigned int optlen);
1160 	int			(*getsockopt)(struct sock *sk, int level,
1161 					int optname, char __user *optval,
1162 					int __user *option);
1163 	void			(*keepalive)(struct sock *sk, int valbool);
1164 #ifdef CONFIG_COMPAT
1165 	int			(*compat_ioctl)(struct sock *sk,
1166 					unsigned int cmd, unsigned long arg);
1167 #endif
1168 	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1169 					   size_t len);
1170 	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1171 					   size_t len, int noblock, int flags,
1172 					   int *addr_len);
1173 	int			(*sendpage)(struct sock *sk, struct page *page,
1174 					int offset, size_t size, int flags);
1175 	int			(*bind)(struct sock *sk,
1176 					struct sockaddr *addr, int addr_len);
1177 	int			(*bind_add)(struct sock *sk,
1178 					struct sockaddr *addr, int addr_len);
1179 
1180 	int			(*backlog_rcv) (struct sock *sk,
1181 						struct sk_buff *skb);
1182 	bool			(*bpf_bypass_getsockopt)(int level,
1183 							 int optname);
1184 
1185 	void		(*release_cb)(struct sock *sk);
1186 
1187 	/* Keeping track of sk's, looking them up, and port selection methods. */
1188 	int			(*hash)(struct sock *sk);
1189 	void			(*unhash)(struct sock *sk);
1190 	void			(*rehash)(struct sock *sk);
1191 	int			(*get_port)(struct sock *sk, unsigned short snum);
1192 #ifdef CONFIG_BPF_SYSCALL
1193 	int			(*psock_update_sk_prot)(struct sock *sk,
1194 							struct sk_psock *psock,
1195 							bool restore);
1196 #endif
1197 
1198 	/* Keeping track of sockets in use */
1199 #ifdef CONFIG_PROC_FS
1200 	unsigned int		inuse_idx;
1201 #endif
1202 
1203 	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1204 	bool			(*stream_memory_read)(const struct sock *sk);
1205 	/* Memory pressure */
1206 	void			(*enter_memory_pressure)(struct sock *sk);
1207 	void			(*leave_memory_pressure)(struct sock *sk);
1208 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1209 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1210 	/*
1211 	 * Pressure flag: try to collapse.
1212 	 * Technical note: it is used by multiple contexts non atomically.
1213 	 * All the __sk_mem_schedule() is of this nature: accounting
1214 	 * is strict, actions are advisory and have some latency.
1215 	 */
1216 	unsigned long		*memory_pressure;
1217 	long			*sysctl_mem;
1218 
1219 	int			*sysctl_wmem;
1220 	int			*sysctl_rmem;
1221 	u32			sysctl_wmem_offset;
1222 	u32			sysctl_rmem_offset;
1223 
1224 	int			max_header;
1225 	bool			no_autobind;
1226 
1227 	struct kmem_cache	*slab;
1228 	unsigned int		obj_size;
1229 	slab_flags_t		slab_flags;
1230 	unsigned int		useroffset;	/* Usercopy region offset */
1231 	unsigned int		usersize;	/* Usercopy region size */
1232 
1233 	struct percpu_counter	*orphan_count;
1234 
1235 	struct request_sock_ops	*rsk_prot;
1236 	struct timewait_sock_ops *twsk_prot;
1237 
1238 	union {
1239 		struct inet_hashinfo	*hashinfo;
1240 		struct udp_table	*udp_table;
1241 		struct raw_hashinfo	*raw_hash;
1242 		struct smc_hashinfo	*smc_hash;
1243 	} h;
1244 
1245 	struct module		*owner;
1246 
1247 	char			name[32];
1248 
1249 	struct list_head	node;
1250 #ifdef SOCK_REFCNT_DEBUG
1251 	atomic_t		socks;
1252 #endif
1253 	int			(*diag_destroy)(struct sock *sk, int err);
1254 } __randomize_layout;
1255 
1256 int proto_register(struct proto *prot, int alloc_slab);
1257 void proto_unregister(struct proto *prot);
1258 int sock_load_diag_module(int family, int protocol);
1259 
1260 #ifdef SOCK_REFCNT_DEBUG
1261 static inline void sk_refcnt_debug_inc(struct sock *sk)
1262 {
1263 	atomic_inc(&sk->sk_prot->socks);
1264 }
1265 
1266 static inline void sk_refcnt_debug_dec(struct sock *sk)
1267 {
1268 	atomic_dec(&sk->sk_prot->socks);
1269 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1270 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1271 }
1272 
1273 static inline void sk_refcnt_debug_release(const struct sock *sk)
1274 {
1275 	if (refcount_read(&sk->sk_refcnt) != 1)
1276 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1277 		       sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1278 }
1279 #else /* SOCK_REFCNT_DEBUG */
1280 #define sk_refcnt_debug_inc(sk) do { } while (0)
1281 #define sk_refcnt_debug_dec(sk) do { } while (0)
1282 #define sk_refcnt_debug_release(sk) do { } while (0)
1283 #endif /* SOCK_REFCNT_DEBUG */
1284 
1285 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1286 
1287 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1288 {
1289 	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1290 		return false;
1291 
1292 #ifdef CONFIG_INET
1293 	return sk->sk_prot->stream_memory_free ?
1294 		INDIRECT_CALL_1(sk->sk_prot->stream_memory_free,
1295 			        tcp_stream_memory_free,
1296 				sk, wake) : true;
1297 #else
1298 	return sk->sk_prot->stream_memory_free ?
1299 		sk->sk_prot->stream_memory_free(sk, wake) : true;
1300 #endif
1301 }
1302 
1303 static inline bool sk_stream_memory_free(const struct sock *sk)
1304 {
1305 	return __sk_stream_memory_free(sk, 0);
1306 }
1307 
1308 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1309 {
1310 	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1311 	       __sk_stream_memory_free(sk, wake);
1312 }
1313 
1314 static inline bool sk_stream_is_writeable(const struct sock *sk)
1315 {
1316 	return __sk_stream_is_writeable(sk, 0);
1317 }
1318 
1319 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1320 					    struct cgroup *ancestor)
1321 {
1322 #ifdef CONFIG_SOCK_CGROUP_DATA
1323 	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1324 				    ancestor);
1325 #else
1326 	return -ENOTSUPP;
1327 #endif
1328 }
1329 
1330 static inline bool sk_has_memory_pressure(const struct sock *sk)
1331 {
1332 	return sk->sk_prot->memory_pressure != NULL;
1333 }
1334 
1335 static inline bool sk_under_memory_pressure(const struct sock *sk)
1336 {
1337 	if (!sk->sk_prot->memory_pressure)
1338 		return false;
1339 
1340 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1341 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1342 		return true;
1343 
1344 	return !!*sk->sk_prot->memory_pressure;
1345 }
1346 
1347 static inline long
1348 sk_memory_allocated(const struct sock *sk)
1349 {
1350 	return atomic_long_read(sk->sk_prot->memory_allocated);
1351 }
1352 
1353 static inline long
1354 sk_memory_allocated_add(struct sock *sk, int amt)
1355 {
1356 	return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1357 }
1358 
1359 static inline void
1360 sk_memory_allocated_sub(struct sock *sk, int amt)
1361 {
1362 	atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1363 }
1364 
1365 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1366 
1367 static inline void sk_sockets_allocated_dec(struct sock *sk)
1368 {
1369 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1370 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1371 }
1372 
1373 static inline void sk_sockets_allocated_inc(struct sock *sk)
1374 {
1375 	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1376 				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1377 }
1378 
1379 static inline u64
1380 sk_sockets_allocated_read_positive(struct sock *sk)
1381 {
1382 	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1383 }
1384 
1385 static inline int
1386 proto_sockets_allocated_sum_positive(struct proto *prot)
1387 {
1388 	return percpu_counter_sum_positive(prot->sockets_allocated);
1389 }
1390 
1391 static inline long
1392 proto_memory_allocated(struct proto *prot)
1393 {
1394 	return atomic_long_read(prot->memory_allocated);
1395 }
1396 
1397 static inline bool
1398 proto_memory_pressure(struct proto *prot)
1399 {
1400 	if (!prot->memory_pressure)
1401 		return false;
1402 	return !!*prot->memory_pressure;
1403 }
1404 
1405 
1406 #ifdef CONFIG_PROC_FS
1407 /* Called with local bh disabled */
1408 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1409 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1410 int sock_inuse_get(struct net *net);
1411 #else
1412 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1413 		int inc)
1414 {
1415 }
1416 #endif
1417 
1418 
1419 /* With per-bucket locks this operation is not-atomic, so that
1420  * this version is not worse.
1421  */
1422 static inline int __sk_prot_rehash(struct sock *sk)
1423 {
1424 	sk->sk_prot->unhash(sk);
1425 	return sk->sk_prot->hash(sk);
1426 }
1427 
1428 /* About 10 seconds */
1429 #define SOCK_DESTROY_TIME (10*HZ)
1430 
1431 /* Sockets 0-1023 can't be bound to unless you are superuser */
1432 #define PROT_SOCK	1024
1433 
1434 #define SHUTDOWN_MASK	3
1435 #define RCV_SHUTDOWN	1
1436 #define SEND_SHUTDOWN	2
1437 
1438 #define SOCK_BINDADDR_LOCK	4
1439 #define SOCK_BINDPORT_LOCK	8
1440 
1441 struct socket_alloc {
1442 	struct socket socket;
1443 	struct inode vfs_inode;
1444 };
1445 
1446 static inline struct socket *SOCKET_I(struct inode *inode)
1447 {
1448 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1449 }
1450 
1451 static inline struct inode *SOCK_INODE(struct socket *socket)
1452 {
1453 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1454 }
1455 
1456 /*
1457  * Functions for memory accounting
1458  */
1459 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1460 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1461 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1462 void __sk_mem_reclaim(struct sock *sk, int amount);
1463 
1464 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1465  * do not necessarily have 16x time more memory than 4KB ones.
1466  */
1467 #define SK_MEM_QUANTUM 4096
1468 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1469 #define SK_MEM_SEND	0
1470 #define SK_MEM_RECV	1
1471 
1472 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1473 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1474 {
1475 	long val = sk->sk_prot->sysctl_mem[index];
1476 
1477 #if PAGE_SIZE > SK_MEM_QUANTUM
1478 	val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1479 #elif PAGE_SIZE < SK_MEM_QUANTUM
1480 	val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1481 #endif
1482 	return val;
1483 }
1484 
1485 static inline int sk_mem_pages(int amt)
1486 {
1487 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1488 }
1489 
1490 static inline bool sk_has_account(struct sock *sk)
1491 {
1492 	/* return true if protocol supports memory accounting */
1493 	return !!sk->sk_prot->memory_allocated;
1494 }
1495 
1496 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1497 {
1498 	if (!sk_has_account(sk))
1499 		return true;
1500 	return size <= sk->sk_forward_alloc ||
1501 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1502 }
1503 
1504 static inline bool
1505 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1506 {
1507 	if (!sk_has_account(sk))
1508 		return true;
1509 	return size <= sk->sk_forward_alloc ||
1510 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1511 		skb_pfmemalloc(skb);
1512 }
1513 
1514 static inline void sk_mem_reclaim(struct sock *sk)
1515 {
1516 	if (!sk_has_account(sk))
1517 		return;
1518 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1519 		__sk_mem_reclaim(sk, sk->sk_forward_alloc);
1520 }
1521 
1522 static inline void sk_mem_reclaim_partial(struct sock *sk)
1523 {
1524 	if (!sk_has_account(sk))
1525 		return;
1526 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1527 		__sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1528 }
1529 
1530 static inline void sk_mem_charge(struct sock *sk, int size)
1531 {
1532 	if (!sk_has_account(sk))
1533 		return;
1534 	sk->sk_forward_alloc -= size;
1535 }
1536 
1537 static inline void sk_mem_uncharge(struct sock *sk, int size)
1538 {
1539 	if (!sk_has_account(sk))
1540 		return;
1541 	sk->sk_forward_alloc += size;
1542 
1543 	/* Avoid a possible overflow.
1544 	 * TCP send queues can make this happen, if sk_mem_reclaim()
1545 	 * is not called and more than 2 GBytes are released at once.
1546 	 *
1547 	 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1548 	 * no need to hold that much forward allocation anyway.
1549 	 */
1550 	if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1551 		__sk_mem_reclaim(sk, 1 << 20);
1552 }
1553 
1554 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1555 {
1556 	sk_wmem_queued_add(sk, -skb->truesize);
1557 	sk_mem_uncharge(sk, skb->truesize);
1558 	__kfree_skb(skb);
1559 }
1560 
1561 static inline void sock_release_ownership(struct sock *sk)
1562 {
1563 	if (sk->sk_lock.owned) {
1564 		sk->sk_lock.owned = 0;
1565 
1566 		/* The sk_lock has mutex_unlock() semantics: */
1567 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1568 	}
1569 }
1570 
1571 /*
1572  * Macro so as to not evaluate some arguments when
1573  * lockdep is not enabled.
1574  *
1575  * Mark both the sk_lock and the sk_lock.slock as a
1576  * per-address-family lock class.
1577  */
1578 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1579 do {									\
1580 	sk->sk_lock.owned = 0;						\
1581 	init_waitqueue_head(&sk->sk_lock.wq);				\
1582 	spin_lock_init(&(sk)->sk_lock.slock);				\
1583 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1584 			sizeof((sk)->sk_lock));				\
1585 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1586 				(skey), (sname));				\
1587 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1588 } while (0)
1589 
1590 static inline bool lockdep_sock_is_held(const struct sock *sk)
1591 {
1592 	return lockdep_is_held(&sk->sk_lock) ||
1593 	       lockdep_is_held(&sk->sk_lock.slock);
1594 }
1595 
1596 void lock_sock_nested(struct sock *sk, int subclass);
1597 
1598 static inline void lock_sock(struct sock *sk)
1599 {
1600 	lock_sock_nested(sk, 0);
1601 }
1602 
1603 void __lock_sock(struct sock *sk);
1604 void __release_sock(struct sock *sk);
1605 void release_sock(struct sock *sk);
1606 
1607 /* BH context may only use the following locking interface. */
1608 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1609 #define bh_lock_sock_nested(__sk) \
1610 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1611 				SINGLE_DEPTH_NESTING)
1612 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1613 
1614 bool lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1615 
1616 /**
1617  * unlock_sock_fast - complement of lock_sock_fast
1618  * @sk: socket
1619  * @slow: slow mode
1620  *
1621  * fast unlock socket for user context.
1622  * If slow mode is on, we call regular release_sock()
1623  */
1624 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1625 	__releases(&sk->sk_lock.slock)
1626 {
1627 	if (slow) {
1628 		release_sock(sk);
1629 		__release(&sk->sk_lock.slock);
1630 	} else {
1631 		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1632 		spin_unlock_bh(&sk->sk_lock.slock);
1633 	}
1634 }
1635 
1636 /* Used by processes to "lock" a socket state, so that
1637  * interrupts and bottom half handlers won't change it
1638  * from under us. It essentially blocks any incoming
1639  * packets, so that we won't get any new data or any
1640  * packets that change the state of the socket.
1641  *
1642  * While locked, BH processing will add new packets to
1643  * the backlog queue.  This queue is processed by the
1644  * owner of the socket lock right before it is released.
1645  *
1646  * Since ~2.3.5 it is also exclusive sleep lock serializing
1647  * accesses from user process context.
1648  */
1649 
1650 static inline void sock_owned_by_me(const struct sock *sk)
1651 {
1652 #ifdef CONFIG_LOCKDEP
1653 	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1654 #endif
1655 }
1656 
1657 static inline bool sock_owned_by_user(const struct sock *sk)
1658 {
1659 	sock_owned_by_me(sk);
1660 	return sk->sk_lock.owned;
1661 }
1662 
1663 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1664 {
1665 	return sk->sk_lock.owned;
1666 }
1667 
1668 /* no reclassification while locks are held */
1669 static inline bool sock_allow_reclassification(const struct sock *csk)
1670 {
1671 	struct sock *sk = (struct sock *)csk;
1672 
1673 	return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1674 }
1675 
1676 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1677 		      struct proto *prot, int kern);
1678 void sk_free(struct sock *sk);
1679 void sk_destruct(struct sock *sk);
1680 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1681 void sk_free_unlock_clone(struct sock *sk);
1682 
1683 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1684 			     gfp_t priority);
1685 void __sock_wfree(struct sk_buff *skb);
1686 void sock_wfree(struct sk_buff *skb);
1687 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1688 			     gfp_t priority);
1689 void skb_orphan_partial(struct sk_buff *skb);
1690 void sock_rfree(struct sk_buff *skb);
1691 void sock_efree(struct sk_buff *skb);
1692 #ifdef CONFIG_INET
1693 void sock_edemux(struct sk_buff *skb);
1694 void sock_pfree(struct sk_buff *skb);
1695 #else
1696 #define sock_edemux sock_efree
1697 #endif
1698 
1699 int sock_setsockopt(struct socket *sock, int level, int op,
1700 		    sockptr_t optval, unsigned int optlen);
1701 
1702 int sock_getsockopt(struct socket *sock, int level, int op,
1703 		    char __user *optval, int __user *optlen);
1704 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1705 		   bool timeval, bool time32);
1706 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1707 				    int noblock, int *errcode);
1708 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1709 				     unsigned long data_len, int noblock,
1710 				     int *errcode, int max_page_order);
1711 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1712 void sock_kfree_s(struct sock *sk, void *mem, int size);
1713 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1714 void sk_send_sigurg(struct sock *sk);
1715 
1716 struct sockcm_cookie {
1717 	u64 transmit_time;
1718 	u32 mark;
1719 	u16 tsflags;
1720 };
1721 
1722 static inline void sockcm_init(struct sockcm_cookie *sockc,
1723 			       const struct sock *sk)
1724 {
1725 	*sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1726 }
1727 
1728 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1729 		     struct sockcm_cookie *sockc);
1730 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1731 		   struct sockcm_cookie *sockc);
1732 
1733 /*
1734  * Functions to fill in entries in struct proto_ops when a protocol
1735  * does not implement a particular function.
1736  */
1737 int sock_no_bind(struct socket *, struct sockaddr *, int);
1738 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1739 int sock_no_socketpair(struct socket *, struct socket *);
1740 int sock_no_accept(struct socket *, struct socket *, int, bool);
1741 int sock_no_getname(struct socket *, struct sockaddr *, int);
1742 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1743 int sock_no_listen(struct socket *, int);
1744 int sock_no_shutdown(struct socket *, int);
1745 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1746 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1747 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1748 int sock_no_mmap(struct file *file, struct socket *sock,
1749 		 struct vm_area_struct *vma);
1750 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1751 			 size_t size, int flags);
1752 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1753 				int offset, size_t size, int flags);
1754 
1755 /*
1756  * Functions to fill in entries in struct proto_ops when a protocol
1757  * uses the inet style.
1758  */
1759 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1760 				  char __user *optval, int __user *optlen);
1761 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1762 			int flags);
1763 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1764 			   sockptr_t optval, unsigned int optlen);
1765 
1766 void sk_common_release(struct sock *sk);
1767 
1768 /*
1769  *	Default socket callbacks and setup code
1770  */
1771 
1772 /* Initialise core socket variables */
1773 void sock_init_data(struct socket *sock, struct sock *sk);
1774 
1775 /*
1776  * Socket reference counting postulates.
1777  *
1778  * * Each user of socket SHOULD hold a reference count.
1779  * * Each access point to socket (an hash table bucket, reference from a list,
1780  *   running timer, skb in flight MUST hold a reference count.
1781  * * When reference count hits 0, it means it will never increase back.
1782  * * When reference count hits 0, it means that no references from
1783  *   outside exist to this socket and current process on current CPU
1784  *   is last user and may/should destroy this socket.
1785  * * sk_free is called from any context: process, BH, IRQ. When
1786  *   it is called, socket has no references from outside -> sk_free
1787  *   may release descendant resources allocated by the socket, but
1788  *   to the time when it is called, socket is NOT referenced by any
1789  *   hash tables, lists etc.
1790  * * Packets, delivered from outside (from network or from another process)
1791  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1792  *   when they sit in queue. Otherwise, packets will leak to hole, when
1793  *   socket is looked up by one cpu and unhasing is made by another CPU.
1794  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1795  *   (leak to backlog). Packet socket does all the processing inside
1796  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1797  *   use separate SMP lock, so that they are prone too.
1798  */
1799 
1800 /* Ungrab socket and destroy it, if it was the last reference. */
1801 static inline void sock_put(struct sock *sk)
1802 {
1803 	if (refcount_dec_and_test(&sk->sk_refcnt))
1804 		sk_free(sk);
1805 }
1806 /* Generic version of sock_put(), dealing with all sockets
1807  * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1808  */
1809 void sock_gen_put(struct sock *sk);
1810 
1811 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1812 		     unsigned int trim_cap, bool refcounted);
1813 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1814 				 const int nested)
1815 {
1816 	return __sk_receive_skb(sk, skb, nested, 1, true);
1817 }
1818 
1819 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1820 {
1821 	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1822 	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1823 		return;
1824 	sk->sk_tx_queue_mapping = tx_queue;
1825 }
1826 
1827 #define NO_QUEUE_MAPPING	USHRT_MAX
1828 
1829 static inline void sk_tx_queue_clear(struct sock *sk)
1830 {
1831 	sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1832 }
1833 
1834 static inline int sk_tx_queue_get(const struct sock *sk)
1835 {
1836 	if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1837 		return sk->sk_tx_queue_mapping;
1838 
1839 	return -1;
1840 }
1841 
1842 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1843 {
1844 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1845 	if (skb_rx_queue_recorded(skb)) {
1846 		u16 rx_queue = skb_get_rx_queue(skb);
1847 
1848 		if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1849 			return;
1850 
1851 		sk->sk_rx_queue_mapping = rx_queue;
1852 	}
1853 #endif
1854 }
1855 
1856 static inline void sk_rx_queue_clear(struct sock *sk)
1857 {
1858 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1859 	sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1860 #endif
1861 }
1862 
1863 static inline int sk_rx_queue_get(const struct sock *sk)
1864 {
1865 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1866 	if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1867 		return sk->sk_rx_queue_mapping;
1868 #endif
1869 
1870 	return -1;
1871 }
1872 
1873 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1874 {
1875 	sk->sk_socket = sock;
1876 }
1877 
1878 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1879 {
1880 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1881 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1882 }
1883 /* Detach socket from process context.
1884  * Announce socket dead, detach it from wait queue and inode.
1885  * Note that parent inode held reference count on this struct sock,
1886  * we do not release it in this function, because protocol
1887  * probably wants some additional cleanups or even continuing
1888  * to work with this socket (TCP).
1889  */
1890 static inline void sock_orphan(struct sock *sk)
1891 {
1892 	write_lock_bh(&sk->sk_callback_lock);
1893 	sock_set_flag(sk, SOCK_DEAD);
1894 	sk_set_socket(sk, NULL);
1895 	sk->sk_wq  = NULL;
1896 	write_unlock_bh(&sk->sk_callback_lock);
1897 }
1898 
1899 static inline void sock_graft(struct sock *sk, struct socket *parent)
1900 {
1901 	WARN_ON(parent->sk);
1902 	write_lock_bh(&sk->sk_callback_lock);
1903 	rcu_assign_pointer(sk->sk_wq, &parent->wq);
1904 	parent->sk = sk;
1905 	sk_set_socket(sk, parent);
1906 	sk->sk_uid = SOCK_INODE(parent)->i_uid;
1907 	security_sock_graft(sk, parent);
1908 	write_unlock_bh(&sk->sk_callback_lock);
1909 }
1910 
1911 kuid_t sock_i_uid(struct sock *sk);
1912 unsigned long sock_i_ino(struct sock *sk);
1913 
1914 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1915 {
1916 	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1917 }
1918 
1919 static inline u32 net_tx_rndhash(void)
1920 {
1921 	u32 v = prandom_u32();
1922 
1923 	return v ?: 1;
1924 }
1925 
1926 static inline void sk_set_txhash(struct sock *sk)
1927 {
1928 	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
1929 	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
1930 }
1931 
1932 static inline bool sk_rethink_txhash(struct sock *sk)
1933 {
1934 	if (sk->sk_txhash) {
1935 		sk_set_txhash(sk);
1936 		return true;
1937 	}
1938 	return false;
1939 }
1940 
1941 static inline struct dst_entry *
1942 __sk_dst_get(struct sock *sk)
1943 {
1944 	return rcu_dereference_check(sk->sk_dst_cache,
1945 				     lockdep_sock_is_held(sk));
1946 }
1947 
1948 static inline struct dst_entry *
1949 sk_dst_get(struct sock *sk)
1950 {
1951 	struct dst_entry *dst;
1952 
1953 	rcu_read_lock();
1954 	dst = rcu_dereference(sk->sk_dst_cache);
1955 	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1956 		dst = NULL;
1957 	rcu_read_unlock();
1958 	return dst;
1959 }
1960 
1961 static inline void __dst_negative_advice(struct sock *sk)
1962 {
1963 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1964 
1965 	if (dst && dst->ops->negative_advice) {
1966 		ndst = dst->ops->negative_advice(dst);
1967 
1968 		if (ndst != dst) {
1969 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1970 			sk_tx_queue_clear(sk);
1971 			sk->sk_dst_pending_confirm = 0;
1972 		}
1973 	}
1974 }
1975 
1976 static inline void dst_negative_advice(struct sock *sk)
1977 {
1978 	sk_rethink_txhash(sk);
1979 	__dst_negative_advice(sk);
1980 }
1981 
1982 static inline void
1983 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1984 {
1985 	struct dst_entry *old_dst;
1986 
1987 	sk_tx_queue_clear(sk);
1988 	sk->sk_dst_pending_confirm = 0;
1989 	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1990 					    lockdep_sock_is_held(sk));
1991 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1992 	dst_release(old_dst);
1993 }
1994 
1995 static inline void
1996 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1997 {
1998 	struct dst_entry *old_dst;
1999 
2000 	sk_tx_queue_clear(sk);
2001 	sk->sk_dst_pending_confirm = 0;
2002 	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2003 	dst_release(old_dst);
2004 }
2005 
2006 static inline void
2007 __sk_dst_reset(struct sock *sk)
2008 {
2009 	__sk_dst_set(sk, NULL);
2010 }
2011 
2012 static inline void
2013 sk_dst_reset(struct sock *sk)
2014 {
2015 	sk_dst_set(sk, NULL);
2016 }
2017 
2018 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2019 
2020 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2021 
2022 static inline void sk_dst_confirm(struct sock *sk)
2023 {
2024 	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2025 		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2026 }
2027 
2028 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2029 {
2030 	if (skb_get_dst_pending_confirm(skb)) {
2031 		struct sock *sk = skb->sk;
2032 		unsigned long now = jiffies;
2033 
2034 		/* avoid dirtying neighbour */
2035 		if (READ_ONCE(n->confirmed) != now)
2036 			WRITE_ONCE(n->confirmed, now);
2037 		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2038 			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2039 	}
2040 }
2041 
2042 bool sk_mc_loop(struct sock *sk);
2043 
2044 static inline bool sk_can_gso(const struct sock *sk)
2045 {
2046 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2047 }
2048 
2049 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2050 
2051 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2052 {
2053 	sk->sk_route_nocaps |= flags;
2054 	sk->sk_route_caps &= ~flags;
2055 }
2056 
2057 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2058 					   struct iov_iter *from, char *to,
2059 					   int copy, int offset)
2060 {
2061 	if (skb->ip_summed == CHECKSUM_NONE) {
2062 		__wsum csum = 0;
2063 		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2064 			return -EFAULT;
2065 		skb->csum = csum_block_add(skb->csum, csum, offset);
2066 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2067 		if (!copy_from_iter_full_nocache(to, copy, from))
2068 			return -EFAULT;
2069 	} else if (!copy_from_iter_full(to, copy, from))
2070 		return -EFAULT;
2071 
2072 	return 0;
2073 }
2074 
2075 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2076 				       struct iov_iter *from, int copy)
2077 {
2078 	int err, offset = skb->len;
2079 
2080 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2081 				       copy, offset);
2082 	if (err)
2083 		__skb_trim(skb, offset);
2084 
2085 	return err;
2086 }
2087 
2088 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2089 					   struct sk_buff *skb,
2090 					   struct page *page,
2091 					   int off, int copy)
2092 {
2093 	int err;
2094 
2095 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2096 				       copy, skb->len);
2097 	if (err)
2098 		return err;
2099 
2100 	skb->len	     += copy;
2101 	skb->data_len	     += copy;
2102 	skb->truesize	     += copy;
2103 	sk_wmem_queued_add(sk, copy);
2104 	sk_mem_charge(sk, copy);
2105 	return 0;
2106 }
2107 
2108 /**
2109  * sk_wmem_alloc_get - returns write allocations
2110  * @sk: socket
2111  *
2112  * Return: sk_wmem_alloc minus initial offset of one
2113  */
2114 static inline int sk_wmem_alloc_get(const struct sock *sk)
2115 {
2116 	return refcount_read(&sk->sk_wmem_alloc) - 1;
2117 }
2118 
2119 /**
2120  * sk_rmem_alloc_get - returns read allocations
2121  * @sk: socket
2122  *
2123  * Return: sk_rmem_alloc
2124  */
2125 static inline int sk_rmem_alloc_get(const struct sock *sk)
2126 {
2127 	return atomic_read(&sk->sk_rmem_alloc);
2128 }
2129 
2130 /**
2131  * sk_has_allocations - check if allocations are outstanding
2132  * @sk: socket
2133  *
2134  * Return: true if socket has write or read allocations
2135  */
2136 static inline bool sk_has_allocations(const struct sock *sk)
2137 {
2138 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2139 }
2140 
2141 /**
2142  * skwq_has_sleeper - check if there are any waiting processes
2143  * @wq: struct socket_wq
2144  *
2145  * Return: true if socket_wq has waiting processes
2146  *
2147  * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2148  * barrier call. They were added due to the race found within the tcp code.
2149  *
2150  * Consider following tcp code paths::
2151  *
2152  *   CPU1                CPU2
2153  *   sys_select          receive packet
2154  *   ...                 ...
2155  *   __add_wait_queue    update tp->rcv_nxt
2156  *   ...                 ...
2157  *   tp->rcv_nxt check   sock_def_readable
2158  *   ...                 {
2159  *   schedule               rcu_read_lock();
2160  *                          wq = rcu_dereference(sk->sk_wq);
2161  *                          if (wq && waitqueue_active(&wq->wait))
2162  *                              wake_up_interruptible(&wq->wait)
2163  *                          ...
2164  *                       }
2165  *
2166  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2167  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2168  * could then endup calling schedule and sleep forever if there are no more
2169  * data on the socket.
2170  *
2171  */
2172 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2173 {
2174 	return wq && wq_has_sleeper(&wq->wait);
2175 }
2176 
2177 /**
2178  * sock_poll_wait - place memory barrier behind the poll_wait call.
2179  * @filp:           file
2180  * @sock:           socket to wait on
2181  * @p:              poll_table
2182  *
2183  * See the comments in the wq_has_sleeper function.
2184  */
2185 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2186 				  poll_table *p)
2187 {
2188 	if (!poll_does_not_wait(p)) {
2189 		poll_wait(filp, &sock->wq.wait, p);
2190 		/* We need to be sure we are in sync with the
2191 		 * socket flags modification.
2192 		 *
2193 		 * This memory barrier is paired in the wq_has_sleeper.
2194 		 */
2195 		smp_mb();
2196 	}
2197 }
2198 
2199 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2200 {
2201 	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2202 	u32 txhash = READ_ONCE(sk->sk_txhash);
2203 
2204 	if (txhash) {
2205 		skb->l4_hash = 1;
2206 		skb->hash = txhash;
2207 	}
2208 }
2209 
2210 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2211 
2212 /*
2213  *	Queue a received datagram if it will fit. Stream and sequenced
2214  *	protocols can't normally use this as they need to fit buffers in
2215  *	and play with them.
2216  *
2217  *	Inlined as it's very short and called for pretty much every
2218  *	packet ever received.
2219  */
2220 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2221 {
2222 	skb_orphan(skb);
2223 	skb->sk = sk;
2224 	skb->destructor = sock_rfree;
2225 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2226 	sk_mem_charge(sk, skb->truesize);
2227 }
2228 
2229 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2230 {
2231 	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2232 		skb_orphan(skb);
2233 		skb->destructor = sock_efree;
2234 		skb->sk = sk;
2235 		return true;
2236 	}
2237 	return false;
2238 }
2239 
2240 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2241 {
2242 	if (skb->destructor != sock_wfree) {
2243 		skb_orphan(skb);
2244 		return;
2245 	}
2246 	skb->slow_gro = 1;
2247 }
2248 
2249 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2250 		    unsigned long expires);
2251 
2252 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2253 
2254 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2255 
2256 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2257 			struct sk_buff *skb, unsigned int flags,
2258 			void (*destructor)(struct sock *sk,
2259 					   struct sk_buff *skb));
2260 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2261 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2262 
2263 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2264 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2265 
2266 /*
2267  *	Recover an error report and clear atomically
2268  */
2269 
2270 static inline int sock_error(struct sock *sk)
2271 {
2272 	int err;
2273 
2274 	/* Avoid an atomic operation for the common case.
2275 	 * This is racy since another cpu/thread can change sk_err under us.
2276 	 */
2277 	if (likely(data_race(!sk->sk_err)))
2278 		return 0;
2279 
2280 	err = xchg(&sk->sk_err, 0);
2281 	return -err;
2282 }
2283 
2284 void sk_error_report(struct sock *sk);
2285 
2286 static inline unsigned long sock_wspace(struct sock *sk)
2287 {
2288 	int amt = 0;
2289 
2290 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2291 		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2292 		if (amt < 0)
2293 			amt = 0;
2294 	}
2295 	return amt;
2296 }
2297 
2298 /* Note:
2299  *  We use sk->sk_wq_raw, from contexts knowing this
2300  *  pointer is not NULL and cannot disappear/change.
2301  */
2302 static inline void sk_set_bit(int nr, struct sock *sk)
2303 {
2304 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2305 	    !sock_flag(sk, SOCK_FASYNC))
2306 		return;
2307 
2308 	set_bit(nr, &sk->sk_wq_raw->flags);
2309 }
2310 
2311 static inline void sk_clear_bit(int nr, struct sock *sk)
2312 {
2313 	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2314 	    !sock_flag(sk, SOCK_FASYNC))
2315 		return;
2316 
2317 	clear_bit(nr, &sk->sk_wq_raw->flags);
2318 }
2319 
2320 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2321 {
2322 	if (sock_flag(sk, SOCK_FASYNC)) {
2323 		rcu_read_lock();
2324 		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2325 		rcu_read_unlock();
2326 	}
2327 }
2328 
2329 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2330  * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2331  * Note: for send buffers, TCP works better if we can build two skbs at
2332  * minimum.
2333  */
2334 #define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2335 
2336 #define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2337 #define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2338 
2339 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2340 {
2341 	u32 val;
2342 
2343 	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2344 		return;
2345 
2346 	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2347 
2348 	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2349 }
2350 
2351 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2352 				    bool force_schedule);
2353 
2354 /**
2355  * sk_page_frag - return an appropriate page_frag
2356  * @sk: socket
2357  *
2358  * Use the per task page_frag instead of the per socket one for
2359  * optimization when we know that we're in the normal context and owns
2360  * everything that's associated with %current.
2361  *
2362  * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2363  * inside other socket operations and end up recursing into sk_page_frag()
2364  * while it's already in use.
2365  *
2366  * Return: a per task page_frag if context allows that,
2367  * otherwise a per socket one.
2368  */
2369 static inline struct page_frag *sk_page_frag(struct sock *sk)
2370 {
2371 	if (gfpflags_normal_context(sk->sk_allocation))
2372 		return &current->task_frag;
2373 
2374 	return &sk->sk_frag;
2375 }
2376 
2377 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2378 
2379 /*
2380  *	Default write policy as shown to user space via poll/select/SIGIO
2381  */
2382 static inline bool sock_writeable(const struct sock *sk)
2383 {
2384 	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2385 }
2386 
2387 static inline gfp_t gfp_any(void)
2388 {
2389 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2390 }
2391 
2392 static inline gfp_t gfp_memcg_charge(void)
2393 {
2394 	return in_softirq() ? GFP_NOWAIT : GFP_KERNEL;
2395 }
2396 
2397 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2398 {
2399 	return noblock ? 0 : sk->sk_rcvtimeo;
2400 }
2401 
2402 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2403 {
2404 	return noblock ? 0 : sk->sk_sndtimeo;
2405 }
2406 
2407 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2408 {
2409 	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2410 
2411 	return v ?: 1;
2412 }
2413 
2414 /* Alas, with timeout socket operations are not restartable.
2415  * Compare this to poll().
2416  */
2417 static inline int sock_intr_errno(long timeo)
2418 {
2419 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2420 }
2421 
2422 struct sock_skb_cb {
2423 	u32 dropcount;
2424 };
2425 
2426 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2427  * using skb->cb[] would keep using it directly and utilize its
2428  * alignement guarantee.
2429  */
2430 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2431 			    sizeof(struct sock_skb_cb)))
2432 
2433 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2434 			    SOCK_SKB_CB_OFFSET))
2435 
2436 #define sock_skb_cb_check_size(size) \
2437 	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2438 
2439 static inline void
2440 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2441 {
2442 	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2443 						atomic_read(&sk->sk_drops) : 0;
2444 }
2445 
2446 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2447 {
2448 	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2449 
2450 	atomic_add(segs, &sk->sk_drops);
2451 }
2452 
2453 static inline ktime_t sock_read_timestamp(struct sock *sk)
2454 {
2455 #if BITS_PER_LONG==32
2456 	unsigned int seq;
2457 	ktime_t kt;
2458 
2459 	do {
2460 		seq = read_seqbegin(&sk->sk_stamp_seq);
2461 		kt = sk->sk_stamp;
2462 	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2463 
2464 	return kt;
2465 #else
2466 	return READ_ONCE(sk->sk_stamp);
2467 #endif
2468 }
2469 
2470 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2471 {
2472 #if BITS_PER_LONG==32
2473 	write_seqlock(&sk->sk_stamp_seq);
2474 	sk->sk_stamp = kt;
2475 	write_sequnlock(&sk->sk_stamp_seq);
2476 #else
2477 	WRITE_ONCE(sk->sk_stamp, kt);
2478 #endif
2479 }
2480 
2481 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2482 			   struct sk_buff *skb);
2483 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2484 			     struct sk_buff *skb);
2485 
2486 static inline void
2487 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2488 {
2489 	ktime_t kt = skb->tstamp;
2490 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2491 
2492 	/*
2493 	 * generate control messages if
2494 	 * - receive time stamping in software requested
2495 	 * - software time stamp available and wanted
2496 	 * - hardware time stamps available and wanted
2497 	 */
2498 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2499 	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2500 	    (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2501 	    (hwtstamps->hwtstamp &&
2502 	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2503 		__sock_recv_timestamp(msg, sk, skb);
2504 	else
2505 		sock_write_timestamp(sk, kt);
2506 
2507 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2508 		__sock_recv_wifi_status(msg, sk, skb);
2509 }
2510 
2511 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2512 			      struct sk_buff *skb);
2513 
2514 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2515 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2516 					  struct sk_buff *skb)
2517 {
2518 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2519 			   (1UL << SOCK_RCVTSTAMP))
2520 #define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2521 			   SOF_TIMESTAMPING_RAW_HARDWARE)
2522 
2523 	if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2524 		__sock_recv_ts_and_drops(msg, sk, skb);
2525 	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2526 		sock_write_timestamp(sk, skb->tstamp);
2527 	else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2528 		sock_write_timestamp(sk, 0);
2529 }
2530 
2531 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2532 
2533 /**
2534  * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2535  * @sk:		socket sending this packet
2536  * @tsflags:	timestamping flags to use
2537  * @tx_flags:	completed with instructions for time stamping
2538  * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2539  *
2540  * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2541  */
2542 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2543 				      __u8 *tx_flags, __u32 *tskey)
2544 {
2545 	if (unlikely(tsflags)) {
2546 		__sock_tx_timestamp(tsflags, tx_flags);
2547 		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2548 		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2549 			*tskey = sk->sk_tskey++;
2550 	}
2551 	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2552 		*tx_flags |= SKBTX_WIFI_STATUS;
2553 }
2554 
2555 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2556 				     __u8 *tx_flags)
2557 {
2558 	_sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2559 }
2560 
2561 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2562 {
2563 	_sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2564 			   &skb_shinfo(skb)->tskey);
2565 }
2566 
2567 /**
2568  * sk_eat_skb - Release a skb if it is no longer needed
2569  * @sk: socket to eat this skb from
2570  * @skb: socket buffer to eat
2571  *
2572  * This routine must be called with interrupts disabled or with the socket
2573  * locked so that the sk_buff queue operation is ok.
2574 */
2575 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2576 {
2577 	__skb_unlink(skb, &sk->sk_receive_queue);
2578 	__kfree_skb(skb);
2579 }
2580 
2581 static inline
2582 struct net *sock_net(const struct sock *sk)
2583 {
2584 	return read_pnet(&sk->sk_net);
2585 }
2586 
2587 static inline
2588 void sock_net_set(struct sock *sk, struct net *net)
2589 {
2590 	write_pnet(&sk->sk_net, net);
2591 }
2592 
2593 static inline bool
2594 skb_sk_is_prefetched(struct sk_buff *skb)
2595 {
2596 #ifdef CONFIG_INET
2597 	return skb->destructor == sock_pfree;
2598 #else
2599 	return false;
2600 #endif /* CONFIG_INET */
2601 }
2602 
2603 /* This helper checks if a socket is a full socket,
2604  * ie _not_ a timewait or request socket.
2605  */
2606 static inline bool sk_fullsock(const struct sock *sk)
2607 {
2608 	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2609 }
2610 
2611 static inline bool
2612 sk_is_refcounted(struct sock *sk)
2613 {
2614 	/* Only full sockets have sk->sk_flags. */
2615 	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2616 }
2617 
2618 /**
2619  * skb_steal_sock - steal a socket from an sk_buff
2620  * @skb: sk_buff to steal the socket from
2621  * @refcounted: is set to true if the socket is reference-counted
2622  */
2623 static inline struct sock *
2624 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2625 {
2626 	if (skb->sk) {
2627 		struct sock *sk = skb->sk;
2628 
2629 		*refcounted = true;
2630 		if (skb_sk_is_prefetched(skb))
2631 			*refcounted = sk_is_refcounted(sk);
2632 		skb->destructor = NULL;
2633 		skb->sk = NULL;
2634 		return sk;
2635 	}
2636 	*refcounted = false;
2637 	return NULL;
2638 }
2639 
2640 /* Checks if this SKB belongs to an HW offloaded socket
2641  * and whether any SW fallbacks are required based on dev.
2642  * Check decrypted mark in case skb_orphan() cleared socket.
2643  */
2644 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2645 						   struct net_device *dev)
2646 {
2647 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2648 	struct sock *sk = skb->sk;
2649 
2650 	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2651 		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2652 #ifdef CONFIG_TLS_DEVICE
2653 	} else if (unlikely(skb->decrypted)) {
2654 		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2655 		kfree_skb(skb);
2656 		skb = NULL;
2657 #endif
2658 	}
2659 #endif
2660 
2661 	return skb;
2662 }
2663 
2664 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2665  * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2666  */
2667 static inline bool sk_listener(const struct sock *sk)
2668 {
2669 	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2670 }
2671 
2672 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2673 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2674 		       int type);
2675 
2676 bool sk_ns_capable(const struct sock *sk,
2677 		   struct user_namespace *user_ns, int cap);
2678 bool sk_capable(const struct sock *sk, int cap);
2679 bool sk_net_capable(const struct sock *sk, int cap);
2680 
2681 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2682 
2683 /* Take into consideration the size of the struct sk_buff overhead in the
2684  * determination of these values, since that is non-constant across
2685  * platforms.  This makes socket queueing behavior and performance
2686  * not depend upon such differences.
2687  */
2688 #define _SK_MEM_PACKETS		256
2689 #define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2690 #define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2691 #define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2692 
2693 extern __u32 sysctl_wmem_max;
2694 extern __u32 sysctl_rmem_max;
2695 
2696 extern int sysctl_tstamp_allow_data;
2697 extern int sysctl_optmem_max;
2698 
2699 extern __u32 sysctl_wmem_default;
2700 extern __u32 sysctl_rmem_default;
2701 
2702 #define SKB_FRAG_PAGE_ORDER	get_order(32768)
2703 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2704 
2705 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2706 {
2707 	/* Does this proto have per netns sysctl_wmem ? */
2708 	if (proto->sysctl_wmem_offset)
2709 		return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2710 
2711 	return *proto->sysctl_wmem;
2712 }
2713 
2714 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2715 {
2716 	/* Does this proto have per netns sysctl_rmem ? */
2717 	if (proto->sysctl_rmem_offset)
2718 		return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2719 
2720 	return *proto->sysctl_rmem;
2721 }
2722 
2723 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2724  * Some wifi drivers need to tweak it to get more chunks.
2725  * They can use this helper from their ndo_start_xmit()
2726  */
2727 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2728 {
2729 	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2730 		return;
2731 	WRITE_ONCE(sk->sk_pacing_shift, val);
2732 }
2733 
2734 /* if a socket is bound to a device, check that the given device
2735  * index is either the same or that the socket is bound to an L3
2736  * master device and the given device index is also enslaved to
2737  * that L3 master
2738  */
2739 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2740 {
2741 	int mdif;
2742 
2743 	if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2744 		return true;
2745 
2746 	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2747 	if (mdif && mdif == sk->sk_bound_dev_if)
2748 		return true;
2749 
2750 	return false;
2751 }
2752 
2753 void sock_def_readable(struct sock *sk);
2754 
2755 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2756 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2757 int sock_set_timestamping(struct sock *sk, int optname,
2758 			  struct so_timestamping timestamping);
2759 
2760 void sock_enable_timestamps(struct sock *sk);
2761 void sock_no_linger(struct sock *sk);
2762 void sock_set_keepalive(struct sock *sk);
2763 void sock_set_priority(struct sock *sk, u32 priority);
2764 void sock_set_rcvbuf(struct sock *sk, int val);
2765 void sock_set_mark(struct sock *sk, u32 val);
2766 void sock_set_reuseaddr(struct sock *sk);
2767 void sock_set_reuseport(struct sock *sk);
2768 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2769 
2770 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2771 
2772 #endif	/* _SOCK_H */
2773