xref: /kernel/linux/linux-6.6/include/net/sock.h (revision 62306a36)
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
4 *		operating system.  INET is implemented using the  BSD Socket
5 *		interface as the means of communication with the user level.
6 *
7 *		Definitions for the AF_INET socket handler.
8 *
9 * Version:	@(#)sock.h	1.0.4	05/13/93
10 *
11 * Authors:	Ross Biro
12 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
14 *		Florian La Roche <flla@stud.uni-sb.de>
15 *
16 * Fixes:
17 *		Alan Cox	:	Volatiles in skbuff pointers. See
18 *					skbuff comments. May be overdone,
19 *					better to prove they can be removed
20 *					than the reverse.
21 *		Alan Cox	:	Added a zapped field for tcp to note
22 *					a socket is reset and must stay shut up
23 *		Alan Cox	:	New fields for options
24 *	Pauline Middelink	:	identd support
25 *		Alan Cox	:	Eliminate low level recv/recvfrom
26 *		David S. Miller	:	New socket lookup architecture.
27 *              Steve Whitehouse:       Default routines for sock_ops
28 *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
29 *              			protinfo be just a void pointer, as the
30 *              			protocol specific parts were moved to
31 *              			respective headers and ipv4/v6, etc now
32 *              			use private slabcaches for its socks
33 *              Pedro Hortas	:	New flags field for socket options
34 */
35#ifndef _SOCK_H
36#define _SOCK_H
37
38#include <linux/hardirq.h>
39#include <linux/kernel.h>
40#include <linux/list.h>
41#include <linux/list_nulls.h>
42#include <linux/timer.h>
43#include <linux/cache.h>
44#include <linux/bitops.h>
45#include <linux/lockdep.h>
46#include <linux/netdevice.h>
47#include <linux/skbuff.h>	/* struct sk_buff */
48#include <linux/mm.h>
49#include <linux/security.h>
50#include <linux/slab.h>
51#include <linux/uaccess.h>
52#include <linux/page_counter.h>
53#include <linux/memcontrol.h>
54#include <linux/static_key.h>
55#include <linux/sched.h>
56#include <linux/wait.h>
57#include <linux/cgroup-defs.h>
58#include <linux/rbtree.h>
59#include <linux/rculist_nulls.h>
60#include <linux/poll.h>
61#include <linux/sockptr.h>
62#include <linux/indirect_call_wrapper.h>
63#include <linux/atomic.h>
64#include <linux/refcount.h>
65#include <linux/llist.h>
66#include <net/dst.h>
67#include <net/checksum.h>
68#include <net/tcp_states.h>
69#include <linux/net_tstamp.h>
70#include <net/l3mdev.h>
71#include <uapi/linux/socket.h>
72
73/*
74 * This structure really needs to be cleaned up.
75 * Most of it is for TCP, and not used by any of
76 * the other protocols.
77 */
78
79/* 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 */
86static inline __printf(2, 3)
87void 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 */
96typedef 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
111struct sock;
112struct proto;
113struct net;
114
115typedef __u32 __bitwise __portpair;
116typedef __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 */
163struct sock_common {
164	union {
165		__addrpair	skc_addrpair;
166		struct {
167			__be32	skc_daddr;
168			__be32	skc_rcv_saddr;
169		};
170	};
171	union  {
172		unsigned int	skc_hash;
173		__u16		skc_u16hashes[2];
174	};
175	/* skc_dport && skc_num must be grouped as well */
176	union {
177		__portpair	skc_portpair;
178		struct {
179			__be16	skc_dport;
180			__u16	skc_num;
181		};
182	};
183
184	unsigned short		skc_family;
185	volatile unsigned char	skc_state;
186	unsigned char		skc_reuse:4;
187	unsigned char		skc_reuseport:1;
188	unsigned char		skc_ipv6only:1;
189	unsigned char		skc_net_refcnt:1;
190	int			skc_bound_dev_if;
191	union {
192		struct hlist_node	skc_bind_node;
193		struct hlist_node	skc_portaddr_node;
194	};
195	struct proto		*skc_prot;
196	possible_net_t		skc_net;
197
198#if IS_ENABLED(CONFIG_IPV6)
199	struct in6_addr		skc_v6_daddr;
200	struct in6_addr		skc_v6_rcv_saddr;
201#endif
202
203	atomic64_t		skc_cookie;
204
205	/* following fields are padding to force
206	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
207	 * assuming IPV6 is enabled. We use this padding differently
208	 * for different kind of 'sockets'
209	 */
210	union {
211		unsigned long	skc_flags;
212		struct sock	*skc_listener; /* request_sock */
213		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
214	};
215	/*
216	 * fields between dontcopy_begin/dontcopy_end
217	 * are not copied in sock_copy()
218	 */
219	/* private: */
220	int			skc_dontcopy_begin[0];
221	/* public: */
222	union {
223		struct hlist_node	skc_node;
224		struct hlist_nulls_node skc_nulls_node;
225	};
226	unsigned short		skc_tx_queue_mapping;
227#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
228	unsigned short		skc_rx_queue_mapping;
229#endif
230	union {
231		int		skc_incoming_cpu;
232		u32		skc_rcv_wnd;
233		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
234	};
235
236	refcount_t		skc_refcnt;
237	/* private: */
238	int                     skc_dontcopy_end[0];
239	union {
240		u32		skc_rxhash;
241		u32		skc_window_clamp;
242		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
243	};
244	/* public: */
245};
246
247struct bpf_local_storage;
248struct sk_filter;
249
250/**
251  *	struct sock - network layer representation of sockets
252  *	@__sk_common: shared layout with inet_timewait_sock
253  *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
254  *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
255  *	@sk_lock:	synchronizer
256  *	@sk_kern_sock: True if sock is using kernel lock classes
257  *	@sk_rcvbuf: size of receive buffer in bytes
258  *	@sk_wq: sock wait queue and async head
259  *	@sk_rx_dst: receive input route used by early demux
260  *	@sk_rx_dst_ifindex: ifindex for @sk_rx_dst
261  *	@sk_rx_dst_cookie: cookie for @sk_rx_dst
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_reserved_mem: space reserved and non-reclaimable for the socket
273  *	@sk_napi_id: id of the last napi context to receive data for sk
274  *	@sk_ll_usec: usecs to busypoll when there is no data
275  *	@sk_allocation: allocation mode
276  *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
277  *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
278  *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
279  *	@sk_sndbuf: size of send buffer in bytes
280  *	@__sk_flags_offset: empty field used to determine location of bitfield
281  *	@sk_padding: unused element for alignment
282  *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
283  *	@sk_no_check_rx: allow zero checksum in RX packets
284  *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
285  *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
286  *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
287  *	@sk_gso_max_size: Maximum GSO segment size to build
288  *	@sk_gso_max_segs: Maximum number of GSO segments
289  *	@sk_pacing_shift: scaling factor for TCP Small Queues
290  *	@sk_lingertime: %SO_LINGER l_linger setting
291  *	@sk_backlog: always used with the per-socket spinlock held
292  *	@sk_callback_lock: used with the callbacks in the end of this struct
293  *	@sk_error_queue: rarely used
294  *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
295  *			  IPV6_ADDRFORM for instance)
296  *	@sk_err: last error
297  *	@sk_err_soft: errors that don't cause failure but are the cause of a
298  *		      persistent failure not just 'timed out'
299  *	@sk_drops: raw/udp drops counter
300  *	@sk_ack_backlog: current listen backlog
301  *	@sk_max_ack_backlog: listen backlog set in listen()
302  *	@sk_uid: user id of owner
303  *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
304  *	@sk_busy_poll_budget: napi processing budget when busypolling
305  *	@sk_priority: %SO_PRIORITY setting
306  *	@sk_type: socket type (%SOCK_STREAM, etc)
307  *	@sk_protocol: which protocol this socket belongs in this network family
308  *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
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_txrehash: enable TX hash rethink
316  *	@sk_filter: socket filtering instructions
317  *	@sk_timer: sock cleanup timer
318  *	@sk_stamp: time stamp of last packet received
319  *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
320  *	@sk_tsflags: SO_TIMESTAMPING flags
321  *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
322  *			   Sockets that can be used under memory reclaim should
323  *			   set this to false.
324  *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
325  *	              for timestamping
326  *	@sk_tskey: counter to disambiguate concurrent tstamp requests
327  *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
328  *	@sk_socket: Identd and reporting IO signals
329  *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
330  *	@sk_frag: cached page frag
331  *	@sk_peek_off: current peek_offset value
332  *	@sk_send_head: front of stuff to transmit
333  *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
334  *	@sk_security: used by security modules
335  *	@sk_mark: generic packet mark
336  *	@sk_cgrp_data: cgroup data for this cgroup
337  *	@sk_memcg: this socket's memory cgroup association
338  *	@sk_write_pending: a write to stream socket waits to start
339  *	@sk_disconnects: number of disconnect operations performed on this sock
340  *	@sk_state_change: callback to indicate change in the state of the sock
341  *	@sk_data_ready: callback to indicate there is data to be processed
342  *	@sk_write_space: callback to indicate there is bf sending space available
343  *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
344  *	@sk_backlog_rcv: callback to process the backlog
345  *	@sk_validate_xmit_skb: ptr to an optional validate function
346  *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
347  *	@sk_reuseport_cb: reuseport group container
348  *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
349  *	@sk_rcu: used during RCU grace period
350  *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
351  *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
352  *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
353  *	@sk_txtime_unused: unused txtime flags
354  *	@ns_tracker: tracker for netns reference
355  *	@sk_bind2_node: bind node in the bhash2 table
356  */
357struct sock {
358	/*
359	 * Now struct inet_timewait_sock also uses sock_common, so please just
360	 * don't add nothing before this first member (__sk_common) --acme
361	 */
362	struct sock_common	__sk_common;
363#define sk_node			__sk_common.skc_node
364#define sk_nulls_node		__sk_common.skc_nulls_node
365#define sk_refcnt		__sk_common.skc_refcnt
366#define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
367#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
368#define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
369#endif
370
371#define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
372#define sk_dontcopy_end		__sk_common.skc_dontcopy_end
373#define sk_hash			__sk_common.skc_hash
374#define sk_portpair		__sk_common.skc_portpair
375#define sk_num			__sk_common.skc_num
376#define sk_dport		__sk_common.skc_dport
377#define sk_addrpair		__sk_common.skc_addrpair
378#define sk_daddr		__sk_common.skc_daddr
379#define sk_rcv_saddr		__sk_common.skc_rcv_saddr
380#define sk_family		__sk_common.skc_family
381#define sk_state		__sk_common.skc_state
382#define sk_reuse		__sk_common.skc_reuse
383#define sk_reuseport		__sk_common.skc_reuseport
384#define sk_ipv6only		__sk_common.skc_ipv6only
385#define sk_net_refcnt		__sk_common.skc_net_refcnt
386#define sk_bound_dev_if		__sk_common.skc_bound_dev_if
387#define sk_bind_node		__sk_common.skc_bind_node
388#define sk_prot			__sk_common.skc_prot
389#define sk_net			__sk_common.skc_net
390#define sk_v6_daddr		__sk_common.skc_v6_daddr
391#define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
392#define sk_cookie		__sk_common.skc_cookie
393#define sk_incoming_cpu		__sk_common.skc_incoming_cpu
394#define sk_flags		__sk_common.skc_flags
395#define sk_rxhash		__sk_common.skc_rxhash
396
397	/* early demux fields */
398	struct dst_entry __rcu	*sk_rx_dst;
399	int			sk_rx_dst_ifindex;
400	u32			sk_rx_dst_cookie;
401
402	socket_lock_t		sk_lock;
403	atomic_t		sk_drops;
404	int			sk_rcvlowat;
405	struct sk_buff_head	sk_error_queue;
406	struct sk_buff_head	sk_receive_queue;
407	/*
408	 * The backlog queue is special, it is always used with
409	 * the per-socket spinlock held and requires low latency
410	 * access. Therefore we special case it's implementation.
411	 * Note : rmem_alloc is in this structure to fill a hole
412	 * on 64bit arches, not because its logically part of
413	 * backlog.
414	 */
415	struct {
416		atomic_t	rmem_alloc;
417		int		len;
418		struct sk_buff	*head;
419		struct sk_buff	*tail;
420	} sk_backlog;
421
422#define sk_rmem_alloc sk_backlog.rmem_alloc
423
424	int			sk_forward_alloc;
425	u32			sk_reserved_mem;
426#ifdef CONFIG_NET_RX_BUSY_POLL
427	unsigned int		sk_ll_usec;
428	/* ===== mostly read cache line ===== */
429	unsigned int		sk_napi_id;
430#endif
431	int			sk_rcvbuf;
432	int			sk_disconnects;
433
434	struct sk_filter __rcu	*sk_filter;
435	union {
436		struct socket_wq __rcu	*sk_wq;
437		/* private: */
438		struct socket_wq	*sk_wq_raw;
439		/* public: */
440	};
441#ifdef CONFIG_XFRM
442	struct xfrm_policy __rcu *sk_policy[2];
443#endif
444
445	struct dst_entry __rcu	*sk_dst_cache;
446	atomic_t		sk_omem_alloc;
447	int			sk_sndbuf;
448
449	/* ===== cache line for TX ===== */
450	int			sk_wmem_queued;
451	refcount_t		sk_wmem_alloc;
452	unsigned long		sk_tsq_flags;
453	union {
454		struct sk_buff	*sk_send_head;
455		struct rb_root	tcp_rtx_queue;
456	};
457	struct sk_buff_head	sk_write_queue;
458	__s32			sk_peek_off;
459	int			sk_write_pending;
460	__u32			sk_dst_pending_confirm;
461	u32			sk_pacing_status; /* see enum sk_pacing */
462	long			sk_sndtimeo;
463	struct timer_list	sk_timer;
464	__u32			sk_priority;
465	__u32			sk_mark;
466	unsigned long		sk_pacing_rate; /* bytes per second */
467	unsigned long		sk_max_pacing_rate;
468	struct page_frag	sk_frag;
469	netdev_features_t	sk_route_caps;
470	int			sk_gso_type;
471	unsigned int		sk_gso_max_size;
472	gfp_t			sk_allocation;
473	__u32			sk_txhash;
474
475	/*
476	 * Because of non atomicity rules, all
477	 * changes are protected by socket lock.
478	 */
479	u8			sk_gso_disabled : 1,
480				sk_kern_sock : 1,
481				sk_no_check_tx : 1,
482				sk_no_check_rx : 1,
483				sk_userlocks : 4;
484	u8			sk_pacing_shift;
485	u16			sk_type;
486	u16			sk_protocol;
487	u16			sk_gso_max_segs;
488	unsigned long	        sk_lingertime;
489	struct proto		*sk_prot_creator;
490	rwlock_t		sk_callback_lock;
491	int			sk_err,
492				sk_err_soft;
493	u32			sk_ack_backlog;
494	u32			sk_max_ack_backlog;
495	kuid_t			sk_uid;
496	u8			sk_txrehash;
497#ifdef CONFIG_NET_RX_BUSY_POLL
498	u8			sk_prefer_busy_poll;
499	u16			sk_busy_poll_budget;
500#endif
501	spinlock_t		sk_peer_lock;
502	int			sk_bind_phc;
503	struct pid		*sk_peer_pid;
504	const struct cred	*sk_peer_cred;
505
506	long			sk_rcvtimeo;
507	ktime_t			sk_stamp;
508#if BITS_PER_LONG==32
509	seqlock_t		sk_stamp_seq;
510#endif
511	atomic_t		sk_tskey;
512	atomic_t		sk_zckey;
513	u32			sk_tsflags;
514	u8			sk_shutdown;
515
516	u8			sk_clockid;
517	u8			sk_txtime_deadline_mode : 1,
518				sk_txtime_report_errors : 1,
519				sk_txtime_unused : 6;
520	bool			sk_use_task_frag;
521
522	struct socket		*sk_socket;
523	void			*sk_user_data;
524#ifdef CONFIG_SECURITY
525	void			*sk_security;
526#endif
527	struct sock_cgroup_data	sk_cgrp_data;
528	struct mem_cgroup	*sk_memcg;
529	void			(*sk_state_change)(struct sock *sk);
530	void			(*sk_data_ready)(struct sock *sk);
531	void			(*sk_write_space)(struct sock *sk);
532	void			(*sk_error_report)(struct sock *sk);
533	int			(*sk_backlog_rcv)(struct sock *sk,
534						  struct sk_buff *skb);
535#ifdef CONFIG_SOCK_VALIDATE_XMIT
536	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
537							struct net_device *dev,
538							struct sk_buff *skb);
539#endif
540	void                    (*sk_destruct)(struct sock *sk);
541	struct sock_reuseport __rcu	*sk_reuseport_cb;
542#ifdef CONFIG_BPF_SYSCALL
543	struct bpf_local_storage __rcu	*sk_bpf_storage;
544#endif
545	struct rcu_head		sk_rcu;
546	netns_tracker		ns_tracker;
547	struct hlist_node	sk_bind2_node;
548};
549
550enum sk_pacing {
551	SK_PACING_NONE		= 0,
552	SK_PACING_NEEDED	= 1,
553	SK_PACING_FQ		= 2,
554};
555
556/* flag bits in sk_user_data
557 *
558 * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
559 *   not be suitable for copying when cloning the socket. For instance,
560 *   it can point to a reference counted object. sk_user_data bottom
561 *   bit is set if pointer must not be copied.
562 *
563 * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
564 *   managed/owned by a BPF reuseport array. This bit should be set
565 *   when sk_user_data's sk is added to the bpf's reuseport_array.
566 *
567 * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
568 *   sk_user_data points to psock type. This bit should be set
569 *   when sk_user_data is assigned to a psock object.
570 */
571#define SK_USER_DATA_NOCOPY	1UL
572#define SK_USER_DATA_BPF	2UL
573#define SK_USER_DATA_PSOCK	4UL
574#define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
575				  SK_USER_DATA_PSOCK)
576
577/**
578 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
579 * @sk: socket
580 */
581static inline bool sk_user_data_is_nocopy(const struct sock *sk)
582{
583	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
584}
585
586#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
587
588/**
589 * __locked_read_sk_user_data_with_flags - return the pointer
590 * only if argument flags all has been set in sk_user_data. Otherwise
591 * return NULL
592 *
593 * @sk: socket
594 * @flags: flag bits
595 *
596 * The caller must be holding sk->sk_callback_lock.
597 */
598static inline void *
599__locked_read_sk_user_data_with_flags(const struct sock *sk,
600				      uintptr_t flags)
601{
602	uintptr_t sk_user_data =
603		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
604						 lockdep_is_held(&sk->sk_callback_lock));
605
606	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
607
608	if ((sk_user_data & flags) == flags)
609		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
610	return NULL;
611}
612
613/**
614 * __rcu_dereference_sk_user_data_with_flags - return the pointer
615 * only if argument flags all has been set in sk_user_data. Otherwise
616 * return NULL
617 *
618 * @sk: socket
619 * @flags: flag bits
620 */
621static inline void *
622__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
623					  uintptr_t flags)
624{
625	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
626
627	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
628
629	if ((sk_user_data & flags) == flags)
630		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
631	return NULL;
632}
633
634#define rcu_dereference_sk_user_data(sk)				\
635	__rcu_dereference_sk_user_data_with_flags(sk, 0)
636#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
637({									\
638	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
639		  __tmp2 = (uintptr_t)(flags);				\
640	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
641	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
642	rcu_assign_pointer(__sk_user_data((sk)),			\
643			   __tmp1 | __tmp2);				\
644})
645#define rcu_assign_sk_user_data(sk, ptr)				\
646	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
647
648static inline
649struct net *sock_net(const struct sock *sk)
650{
651	return read_pnet(&sk->sk_net);
652}
653
654static inline
655void sock_net_set(struct sock *sk, struct net *net)
656{
657	write_pnet(&sk->sk_net, net);
658}
659
660/*
661 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
662 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
663 * on a socket means that the socket will reuse everybody else's port
664 * without looking at the other's sk_reuse value.
665 */
666
667#define SK_NO_REUSE	0
668#define SK_CAN_REUSE	1
669#define SK_FORCE_REUSE	2
670
671int sk_set_peek_off(struct sock *sk, int val);
672
673static inline int sk_peek_offset(const struct sock *sk, int flags)
674{
675	if (unlikely(flags & MSG_PEEK)) {
676		return READ_ONCE(sk->sk_peek_off);
677	}
678
679	return 0;
680}
681
682static inline void sk_peek_offset_bwd(struct sock *sk, int val)
683{
684	s32 off = READ_ONCE(sk->sk_peek_off);
685
686	if (unlikely(off >= 0)) {
687		off = max_t(s32, off - val, 0);
688		WRITE_ONCE(sk->sk_peek_off, off);
689	}
690}
691
692static inline void sk_peek_offset_fwd(struct sock *sk, int val)
693{
694	sk_peek_offset_bwd(sk, -val);
695}
696
697/*
698 * Hashed lists helper routines
699 */
700static inline struct sock *sk_entry(const struct hlist_node *node)
701{
702	return hlist_entry(node, struct sock, sk_node);
703}
704
705static inline struct sock *__sk_head(const struct hlist_head *head)
706{
707	return hlist_entry(head->first, struct sock, sk_node);
708}
709
710static inline struct sock *sk_head(const struct hlist_head *head)
711{
712	return hlist_empty(head) ? NULL : __sk_head(head);
713}
714
715static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
716{
717	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
718}
719
720static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
721{
722	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
723}
724
725static inline struct sock *sk_next(const struct sock *sk)
726{
727	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
728}
729
730static inline struct sock *sk_nulls_next(const struct sock *sk)
731{
732	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
733		hlist_nulls_entry(sk->sk_nulls_node.next,
734				  struct sock, sk_nulls_node) :
735		NULL;
736}
737
738static inline bool sk_unhashed(const struct sock *sk)
739{
740	return hlist_unhashed(&sk->sk_node);
741}
742
743static inline bool sk_hashed(const struct sock *sk)
744{
745	return !sk_unhashed(sk);
746}
747
748static inline void sk_node_init(struct hlist_node *node)
749{
750	node->pprev = NULL;
751}
752
753static inline void __sk_del_node(struct sock *sk)
754{
755	__hlist_del(&sk->sk_node);
756}
757
758/* NB: equivalent to hlist_del_init_rcu */
759static inline bool __sk_del_node_init(struct sock *sk)
760{
761	if (sk_hashed(sk)) {
762		__sk_del_node(sk);
763		sk_node_init(&sk->sk_node);
764		return true;
765	}
766	return false;
767}
768
769/* Grab socket reference count. This operation is valid only
770   when sk is ALREADY grabbed f.e. it is found in hash table
771   or a list and the lookup is made under lock preventing hash table
772   modifications.
773 */
774
775static __always_inline void sock_hold(struct sock *sk)
776{
777	refcount_inc(&sk->sk_refcnt);
778}
779
780/* Ungrab socket in the context, which assumes that socket refcnt
781   cannot hit zero, f.e. it is true in context of any socketcall.
782 */
783static __always_inline void __sock_put(struct sock *sk)
784{
785	refcount_dec(&sk->sk_refcnt);
786}
787
788static inline bool sk_del_node_init(struct sock *sk)
789{
790	bool rc = __sk_del_node_init(sk);
791
792	if (rc) {
793		/* paranoid for a while -acme */
794		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
795		__sock_put(sk);
796	}
797	return rc;
798}
799#define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
800
801static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
802{
803	if (sk_hashed(sk)) {
804		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
805		return true;
806	}
807	return false;
808}
809
810static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
811{
812	bool rc = __sk_nulls_del_node_init_rcu(sk);
813
814	if (rc) {
815		/* paranoid for a while -acme */
816		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
817		__sock_put(sk);
818	}
819	return rc;
820}
821
822static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
823{
824	hlist_add_head(&sk->sk_node, list);
825}
826
827static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
828{
829	sock_hold(sk);
830	__sk_add_node(sk, list);
831}
832
833static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
834{
835	sock_hold(sk);
836	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
837	    sk->sk_family == AF_INET6)
838		hlist_add_tail_rcu(&sk->sk_node, list);
839	else
840		hlist_add_head_rcu(&sk->sk_node, list);
841}
842
843static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
844{
845	sock_hold(sk);
846	hlist_add_tail_rcu(&sk->sk_node, list);
847}
848
849static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
850{
851	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
852}
853
854static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
855{
856	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
857}
858
859static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
860{
861	sock_hold(sk);
862	__sk_nulls_add_node_rcu(sk, list);
863}
864
865static inline void __sk_del_bind_node(struct sock *sk)
866{
867	__hlist_del(&sk->sk_bind_node);
868}
869
870static inline void sk_add_bind_node(struct sock *sk,
871					struct hlist_head *list)
872{
873	hlist_add_head(&sk->sk_bind_node, list);
874}
875
876static inline void __sk_del_bind2_node(struct sock *sk)
877{
878	__hlist_del(&sk->sk_bind2_node);
879}
880
881static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list)
882{
883	hlist_add_head(&sk->sk_bind2_node, list);
884}
885
886#define sk_for_each(__sk, list) \
887	hlist_for_each_entry(__sk, list, sk_node)
888#define sk_for_each_rcu(__sk, list) \
889	hlist_for_each_entry_rcu(__sk, list, sk_node)
890#define sk_nulls_for_each(__sk, node, list) \
891	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
892#define sk_nulls_for_each_rcu(__sk, node, list) \
893	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
894#define sk_for_each_from(__sk) \
895	hlist_for_each_entry_from(__sk, sk_node)
896#define sk_nulls_for_each_from(__sk, node) \
897	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
898		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
899#define sk_for_each_safe(__sk, tmp, list) \
900	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
901#define sk_for_each_bound(__sk, list) \
902	hlist_for_each_entry(__sk, list, sk_bind_node)
903#define sk_for_each_bound_bhash2(__sk, list) \
904	hlist_for_each_entry(__sk, list, sk_bind2_node)
905
906/**
907 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
908 * @tpos:	the type * to use as a loop cursor.
909 * @pos:	the &struct hlist_node to use as a loop cursor.
910 * @head:	the head for your list.
911 * @offset:	offset of hlist_node within the struct.
912 *
913 */
914#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
915	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
916	     pos != NULL &&						       \
917		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
918	     pos = rcu_dereference(hlist_next_rcu(pos)))
919
920static inline struct user_namespace *sk_user_ns(const struct sock *sk)
921{
922	/* Careful only use this in a context where these parameters
923	 * can not change and must all be valid, such as recvmsg from
924	 * userspace.
925	 */
926	return sk->sk_socket->file->f_cred->user_ns;
927}
928
929/* Sock flags */
930enum sock_flags {
931	SOCK_DEAD,
932	SOCK_DONE,
933	SOCK_URGINLINE,
934	SOCK_KEEPOPEN,
935	SOCK_LINGER,
936	SOCK_DESTROY,
937	SOCK_BROADCAST,
938	SOCK_TIMESTAMP,
939	SOCK_ZAPPED,
940	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
941	SOCK_DBG, /* %SO_DEBUG setting */
942	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
943	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
944	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
945	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
946	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
947	SOCK_FASYNC, /* fasync() active */
948	SOCK_RXQ_OVFL,
949	SOCK_ZEROCOPY, /* buffers from userspace */
950	SOCK_WIFI_STATUS, /* push wifi status to userspace */
951	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
952		     * Will use last 4 bytes of packet sent from
953		     * user-space instead.
954		     */
955	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
956	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
957	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
958	SOCK_TXTIME,
959	SOCK_XDP, /* XDP is attached */
960	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
961	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
962};
963
964#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
965
966static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
967{
968	nsk->sk_flags = osk->sk_flags;
969}
970
971static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
972{
973	__set_bit(flag, &sk->sk_flags);
974}
975
976static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
977{
978	__clear_bit(flag, &sk->sk_flags);
979}
980
981static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
982				     int valbool)
983{
984	if (valbool)
985		sock_set_flag(sk, bit);
986	else
987		sock_reset_flag(sk, bit);
988}
989
990static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
991{
992	return test_bit(flag, &sk->sk_flags);
993}
994
995#ifdef CONFIG_NET
996DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
997static inline int sk_memalloc_socks(void)
998{
999	return static_branch_unlikely(&memalloc_socks_key);
1000}
1001
1002void __receive_sock(struct file *file);
1003#else
1004
1005static inline int sk_memalloc_socks(void)
1006{
1007	return 0;
1008}
1009
1010static inline void __receive_sock(struct file *file)
1011{ }
1012#endif
1013
1014static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1015{
1016	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1017}
1018
1019static inline void sk_acceptq_removed(struct sock *sk)
1020{
1021	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1022}
1023
1024static inline void sk_acceptq_added(struct sock *sk)
1025{
1026	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1027}
1028
1029/* Note: If you think the test should be:
1030 *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1031 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1032 */
1033static inline bool sk_acceptq_is_full(const struct sock *sk)
1034{
1035	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1036}
1037
1038/*
1039 * Compute minimal free write space needed to queue new packets.
1040 */
1041static inline int sk_stream_min_wspace(const struct sock *sk)
1042{
1043	return READ_ONCE(sk->sk_wmem_queued) >> 1;
1044}
1045
1046static inline int sk_stream_wspace(const struct sock *sk)
1047{
1048	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1049}
1050
1051static inline void sk_wmem_queued_add(struct sock *sk, int val)
1052{
1053	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1054}
1055
1056static inline void sk_forward_alloc_add(struct sock *sk, int val)
1057{
1058	/* Paired with lockless reads of sk->sk_forward_alloc */
1059	WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1060}
1061
1062void sk_stream_write_space(struct sock *sk);
1063
1064/* OOB backlog add */
1065static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1066{
1067	/* dont let skb dst not refcounted, we are going to leave rcu lock */
1068	skb_dst_force(skb);
1069
1070	if (!sk->sk_backlog.tail)
1071		WRITE_ONCE(sk->sk_backlog.head, skb);
1072	else
1073		sk->sk_backlog.tail->next = skb;
1074
1075	WRITE_ONCE(sk->sk_backlog.tail, skb);
1076	skb->next = NULL;
1077}
1078
1079/*
1080 * Take into account size of receive queue and backlog queue
1081 * Do not take into account this skb truesize,
1082 * to allow even a single big packet to come.
1083 */
1084static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1085{
1086	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1087
1088	return qsize > limit;
1089}
1090
1091/* The per-socket spinlock must be held here. */
1092static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1093					      unsigned int limit)
1094{
1095	if (sk_rcvqueues_full(sk, limit))
1096		return -ENOBUFS;
1097
1098	/*
1099	 * If the skb was allocated from pfmemalloc reserves, only
1100	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1101	 * helping free memory
1102	 */
1103	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1104		return -ENOMEM;
1105
1106	__sk_add_backlog(sk, skb);
1107	sk->sk_backlog.len += skb->truesize;
1108	return 0;
1109}
1110
1111int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1112
1113INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1114INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1115
1116static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1117{
1118	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1119		return __sk_backlog_rcv(sk, skb);
1120
1121	return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1122				  tcp_v6_do_rcv,
1123				  tcp_v4_do_rcv,
1124				  sk, skb);
1125}
1126
1127static inline void sk_incoming_cpu_update(struct sock *sk)
1128{
1129	int cpu = raw_smp_processor_id();
1130
1131	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1132		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1133}
1134
1135static inline void sock_rps_record_flow_hash(__u32 hash)
1136{
1137#ifdef CONFIG_RPS
1138	struct rps_sock_flow_table *sock_flow_table;
1139
1140	rcu_read_lock();
1141	sock_flow_table = rcu_dereference(rps_sock_flow_table);
1142	rps_record_sock_flow(sock_flow_table, hash);
1143	rcu_read_unlock();
1144#endif
1145}
1146
1147static inline void sock_rps_record_flow(const struct sock *sk)
1148{
1149#ifdef CONFIG_RPS
1150	if (static_branch_unlikely(&rfs_needed)) {
1151		/* Reading sk->sk_rxhash might incur an expensive cache line
1152		 * miss.
1153		 *
1154		 * TCP_ESTABLISHED does cover almost all states where RFS
1155		 * might be useful, and is cheaper [1] than testing :
1156		 *	IPv4: inet_sk(sk)->inet_daddr
1157		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1158		 * OR	an additional socket flag
1159		 * [1] : sk_state and sk_prot are in the same cache line.
1160		 */
1161		if (sk->sk_state == TCP_ESTABLISHED) {
1162			/* This READ_ONCE() is paired with the WRITE_ONCE()
1163			 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
1164			 */
1165			sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1166		}
1167	}
1168#endif
1169}
1170
1171static inline void sock_rps_save_rxhash(struct sock *sk,
1172					const struct sk_buff *skb)
1173{
1174#ifdef CONFIG_RPS
1175	/* The following WRITE_ONCE() is paired with the READ_ONCE()
1176	 * here, and another one in sock_rps_record_flow().
1177	 */
1178	if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1179		WRITE_ONCE(sk->sk_rxhash, skb->hash);
1180#endif
1181}
1182
1183static inline void sock_rps_reset_rxhash(struct sock *sk)
1184{
1185#ifdef CONFIG_RPS
1186	/* Paired with READ_ONCE() in sock_rps_record_flow() */
1187	WRITE_ONCE(sk->sk_rxhash, 0);
1188#endif
1189}
1190
1191#define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1192	({	int __rc, __dis = __sk->sk_disconnects;			\
1193		release_sock(__sk);					\
1194		__rc = __condition;					\
1195		if (!__rc) {						\
1196			*(__timeo) = wait_woken(__wait,			\
1197						TASK_INTERRUPTIBLE,	\
1198						*(__timeo));		\
1199		}							\
1200		sched_annotate_sleep();					\
1201		lock_sock(__sk);					\
1202		__rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1203		__rc;							\
1204	})
1205
1206int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1207int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1208void sk_stream_wait_close(struct sock *sk, long timeo_p);
1209int sk_stream_error(struct sock *sk, int flags, int err);
1210void sk_stream_kill_queues(struct sock *sk);
1211void sk_set_memalloc(struct sock *sk);
1212void sk_clear_memalloc(struct sock *sk);
1213
1214void __sk_flush_backlog(struct sock *sk);
1215
1216static inline bool sk_flush_backlog(struct sock *sk)
1217{
1218	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1219		__sk_flush_backlog(sk);
1220		return true;
1221	}
1222	return false;
1223}
1224
1225int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1226
1227struct request_sock_ops;
1228struct timewait_sock_ops;
1229struct inet_hashinfo;
1230struct raw_hashinfo;
1231struct smc_hashinfo;
1232struct module;
1233struct sk_psock;
1234
1235/*
1236 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1237 * un-modified. Special care is taken when initializing object to zero.
1238 */
1239static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1240{
1241	if (offsetof(struct sock, sk_node.next) != 0)
1242		memset(sk, 0, offsetof(struct sock, sk_node.next));
1243	memset(&sk->sk_node.pprev, 0,
1244	       size - offsetof(struct sock, sk_node.pprev));
1245}
1246
1247/* Networking protocol blocks we attach to sockets.
1248 * socket layer -> transport layer interface
1249 */
1250struct proto {
1251	void			(*close)(struct sock *sk,
1252					long timeout);
1253	int			(*pre_connect)(struct sock *sk,
1254					struct sockaddr *uaddr,
1255					int addr_len);
1256	int			(*connect)(struct sock *sk,
1257					struct sockaddr *uaddr,
1258					int addr_len);
1259	int			(*disconnect)(struct sock *sk, int flags);
1260
1261	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1262					  bool kern);
1263
1264	int			(*ioctl)(struct sock *sk, int cmd,
1265					 int *karg);
1266	int			(*init)(struct sock *sk);
1267	void			(*destroy)(struct sock *sk);
1268	void			(*shutdown)(struct sock *sk, int how);
1269	int			(*setsockopt)(struct sock *sk, int level,
1270					int optname, sockptr_t optval,
1271					unsigned int optlen);
1272	int			(*getsockopt)(struct sock *sk, int level,
1273					int optname, char __user *optval,
1274					int __user *option);
1275	void			(*keepalive)(struct sock *sk, int valbool);
1276#ifdef CONFIG_COMPAT
1277	int			(*compat_ioctl)(struct sock *sk,
1278					unsigned int cmd, unsigned long arg);
1279#endif
1280	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1281					   size_t len);
1282	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1283					   size_t len, int flags, int *addr_len);
1284	void			(*splice_eof)(struct socket *sock);
1285	int			(*bind)(struct sock *sk,
1286					struct sockaddr *addr, int addr_len);
1287	int			(*bind_add)(struct sock *sk,
1288					struct sockaddr *addr, int addr_len);
1289
1290	int			(*backlog_rcv) (struct sock *sk,
1291						struct sk_buff *skb);
1292	bool			(*bpf_bypass_getsockopt)(int level,
1293							 int optname);
1294
1295	void		(*release_cb)(struct sock *sk);
1296
1297	/* Keeping track of sk's, looking them up, and port selection methods. */
1298	int			(*hash)(struct sock *sk);
1299	void			(*unhash)(struct sock *sk);
1300	void			(*rehash)(struct sock *sk);
1301	int			(*get_port)(struct sock *sk, unsigned short snum);
1302	void			(*put_port)(struct sock *sk);
1303#ifdef CONFIG_BPF_SYSCALL
1304	int			(*psock_update_sk_prot)(struct sock *sk,
1305							struct sk_psock *psock,
1306							bool restore);
1307#endif
1308
1309	/* Keeping track of sockets in use */
1310#ifdef CONFIG_PROC_FS
1311	unsigned int		inuse_idx;
1312#endif
1313
1314#if IS_ENABLED(CONFIG_MPTCP)
1315	int			(*forward_alloc_get)(const struct sock *sk);
1316#endif
1317
1318	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1319	bool			(*sock_is_readable)(struct sock *sk);
1320	/* Memory pressure */
1321	void			(*enter_memory_pressure)(struct sock *sk);
1322	void			(*leave_memory_pressure)(struct sock *sk);
1323	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1324	int  __percpu		*per_cpu_fw_alloc;
1325	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1326
1327	/*
1328	 * Pressure flag: try to collapse.
1329	 * Technical note: it is used by multiple contexts non atomically.
1330	 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1331	 * All the __sk_mem_schedule() is of this nature: accounting
1332	 * is strict, actions are advisory and have some latency.
1333	 */
1334	unsigned long		*memory_pressure;
1335	long			*sysctl_mem;
1336
1337	int			*sysctl_wmem;
1338	int			*sysctl_rmem;
1339	u32			sysctl_wmem_offset;
1340	u32			sysctl_rmem_offset;
1341
1342	int			max_header;
1343	bool			no_autobind;
1344
1345	struct kmem_cache	*slab;
1346	unsigned int		obj_size;
1347	unsigned int		ipv6_pinfo_offset;
1348	slab_flags_t		slab_flags;
1349	unsigned int		useroffset;	/* Usercopy region offset */
1350	unsigned int		usersize;	/* Usercopy region size */
1351
1352	unsigned int __percpu	*orphan_count;
1353
1354	struct request_sock_ops	*rsk_prot;
1355	struct timewait_sock_ops *twsk_prot;
1356
1357	union {
1358		struct inet_hashinfo	*hashinfo;
1359		struct udp_table	*udp_table;
1360		struct raw_hashinfo	*raw_hash;
1361		struct smc_hashinfo	*smc_hash;
1362	} h;
1363
1364	struct module		*owner;
1365
1366	char			name[32];
1367
1368	struct list_head	node;
1369	int			(*diag_destroy)(struct sock *sk, int err);
1370} __randomize_layout;
1371
1372int proto_register(struct proto *prot, int alloc_slab);
1373void proto_unregister(struct proto *prot);
1374int sock_load_diag_module(int family, int protocol);
1375
1376INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1377
1378static inline int sk_forward_alloc_get(const struct sock *sk)
1379{
1380#if IS_ENABLED(CONFIG_MPTCP)
1381	if (sk->sk_prot->forward_alloc_get)
1382		return sk->sk_prot->forward_alloc_get(sk);
1383#endif
1384	return READ_ONCE(sk->sk_forward_alloc);
1385}
1386
1387static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1388{
1389	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1390		return false;
1391
1392	return sk->sk_prot->stream_memory_free ?
1393		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1394				     tcp_stream_memory_free, sk, wake) : true;
1395}
1396
1397static inline bool sk_stream_memory_free(const struct sock *sk)
1398{
1399	return __sk_stream_memory_free(sk, 0);
1400}
1401
1402static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1403{
1404	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1405	       __sk_stream_memory_free(sk, wake);
1406}
1407
1408static inline bool sk_stream_is_writeable(const struct sock *sk)
1409{
1410	return __sk_stream_is_writeable(sk, 0);
1411}
1412
1413static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1414					    struct cgroup *ancestor)
1415{
1416#ifdef CONFIG_SOCK_CGROUP_DATA
1417	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1418				    ancestor);
1419#else
1420	return -ENOTSUPP;
1421#endif
1422}
1423
1424static inline bool sk_has_memory_pressure(const struct sock *sk)
1425{
1426	return sk->sk_prot->memory_pressure != NULL;
1427}
1428
1429static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1430{
1431	return sk->sk_prot->memory_pressure &&
1432		!!READ_ONCE(*sk->sk_prot->memory_pressure);
1433}
1434
1435static inline bool sk_under_memory_pressure(const struct sock *sk)
1436{
1437	if (!sk->sk_prot->memory_pressure)
1438		return false;
1439
1440	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1441	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1442		return true;
1443
1444	return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1445}
1446
1447static inline long
1448proto_memory_allocated(const struct proto *prot)
1449{
1450	return max(0L, atomic_long_read(prot->memory_allocated));
1451}
1452
1453static inline long
1454sk_memory_allocated(const struct sock *sk)
1455{
1456	return proto_memory_allocated(sk->sk_prot);
1457}
1458
1459/* 1 MB per cpu, in page units */
1460#define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT))
1461
1462static inline void
1463sk_memory_allocated_add(struct sock *sk, int amt)
1464{
1465	int local_reserve;
1466
1467	preempt_disable();
1468	local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1469	if (local_reserve >= SK_MEMORY_PCPU_RESERVE) {
1470		__this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1471		atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1472	}
1473	preempt_enable();
1474}
1475
1476static inline void
1477sk_memory_allocated_sub(struct sock *sk, int amt)
1478{
1479	int local_reserve;
1480
1481	preempt_disable();
1482	local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1483	if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) {
1484		__this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1485		atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1486	}
1487	preempt_enable();
1488}
1489
1490#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1491
1492static inline void sk_sockets_allocated_dec(struct sock *sk)
1493{
1494	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1495				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1496}
1497
1498static inline void sk_sockets_allocated_inc(struct sock *sk)
1499{
1500	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1501				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1502}
1503
1504static inline u64
1505sk_sockets_allocated_read_positive(struct sock *sk)
1506{
1507	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1508}
1509
1510static inline int
1511proto_sockets_allocated_sum_positive(struct proto *prot)
1512{
1513	return percpu_counter_sum_positive(prot->sockets_allocated);
1514}
1515
1516static inline bool
1517proto_memory_pressure(struct proto *prot)
1518{
1519	if (!prot->memory_pressure)
1520		return false;
1521	return !!READ_ONCE(*prot->memory_pressure);
1522}
1523
1524
1525#ifdef CONFIG_PROC_FS
1526#define PROTO_INUSE_NR	64	/* should be enough for the first time */
1527struct prot_inuse {
1528	int all;
1529	int val[PROTO_INUSE_NR];
1530};
1531
1532static inline void sock_prot_inuse_add(const struct net *net,
1533				       const struct proto *prot, int val)
1534{
1535	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1536}
1537
1538static inline void sock_inuse_add(const struct net *net, int val)
1539{
1540	this_cpu_add(net->core.prot_inuse->all, val);
1541}
1542
1543int sock_prot_inuse_get(struct net *net, struct proto *proto);
1544int sock_inuse_get(struct net *net);
1545#else
1546static inline void sock_prot_inuse_add(const struct net *net,
1547				       const struct proto *prot, int val)
1548{
1549}
1550
1551static inline void sock_inuse_add(const struct net *net, int val)
1552{
1553}
1554#endif
1555
1556
1557/* With per-bucket locks this operation is not-atomic, so that
1558 * this version is not worse.
1559 */
1560static inline int __sk_prot_rehash(struct sock *sk)
1561{
1562	sk->sk_prot->unhash(sk);
1563	return sk->sk_prot->hash(sk);
1564}
1565
1566/* About 10 seconds */
1567#define SOCK_DESTROY_TIME (10*HZ)
1568
1569/* Sockets 0-1023 can't be bound to unless you are superuser */
1570#define PROT_SOCK	1024
1571
1572#define SHUTDOWN_MASK	3
1573#define RCV_SHUTDOWN	1
1574#define SEND_SHUTDOWN	2
1575
1576#define SOCK_BINDADDR_LOCK	4
1577#define SOCK_BINDPORT_LOCK	8
1578
1579struct socket_alloc {
1580	struct socket socket;
1581	struct inode vfs_inode;
1582};
1583
1584static inline struct socket *SOCKET_I(struct inode *inode)
1585{
1586	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1587}
1588
1589static inline struct inode *SOCK_INODE(struct socket *socket)
1590{
1591	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1592}
1593
1594/*
1595 * Functions for memory accounting
1596 */
1597int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1598int __sk_mem_schedule(struct sock *sk, int size, int kind);
1599void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1600void __sk_mem_reclaim(struct sock *sk, int amount);
1601
1602#define SK_MEM_SEND	0
1603#define SK_MEM_RECV	1
1604
1605/* sysctl_mem values are in pages */
1606static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1607{
1608	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1609}
1610
1611static inline int sk_mem_pages(int amt)
1612{
1613	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1614}
1615
1616static inline bool sk_has_account(struct sock *sk)
1617{
1618	/* return true if protocol supports memory accounting */
1619	return !!sk->sk_prot->memory_allocated;
1620}
1621
1622static inline bool sk_wmem_schedule(struct sock *sk, int size)
1623{
1624	int delta;
1625
1626	if (!sk_has_account(sk))
1627		return true;
1628	delta = size - sk->sk_forward_alloc;
1629	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1630}
1631
1632static inline bool
1633sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1634{
1635	int delta;
1636
1637	if (!sk_has_account(sk))
1638		return true;
1639	delta = size - sk->sk_forward_alloc;
1640	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1641		skb_pfmemalloc(skb);
1642}
1643
1644static inline int sk_unused_reserved_mem(const struct sock *sk)
1645{
1646	int unused_mem;
1647
1648	if (likely(!sk->sk_reserved_mem))
1649		return 0;
1650
1651	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1652			atomic_read(&sk->sk_rmem_alloc);
1653
1654	return unused_mem > 0 ? unused_mem : 0;
1655}
1656
1657static inline void sk_mem_reclaim(struct sock *sk)
1658{
1659	int reclaimable;
1660
1661	if (!sk_has_account(sk))
1662		return;
1663
1664	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1665
1666	if (reclaimable >= (int)PAGE_SIZE)
1667		__sk_mem_reclaim(sk, reclaimable);
1668}
1669
1670static inline void sk_mem_reclaim_final(struct sock *sk)
1671{
1672	sk->sk_reserved_mem = 0;
1673	sk_mem_reclaim(sk);
1674}
1675
1676static inline void sk_mem_charge(struct sock *sk, int size)
1677{
1678	if (!sk_has_account(sk))
1679		return;
1680	sk_forward_alloc_add(sk, -size);
1681}
1682
1683static inline void sk_mem_uncharge(struct sock *sk, int size)
1684{
1685	if (!sk_has_account(sk))
1686		return;
1687	sk_forward_alloc_add(sk, size);
1688	sk_mem_reclaim(sk);
1689}
1690
1691/*
1692 * Macro so as to not evaluate some arguments when
1693 * lockdep is not enabled.
1694 *
1695 * Mark both the sk_lock and the sk_lock.slock as a
1696 * per-address-family lock class.
1697 */
1698#define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1699do {									\
1700	sk->sk_lock.owned = 0;						\
1701	init_waitqueue_head(&sk->sk_lock.wq);				\
1702	spin_lock_init(&(sk)->sk_lock.slock);				\
1703	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1704			sizeof((sk)->sk_lock));				\
1705	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1706				(skey), (sname));				\
1707	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1708} while (0)
1709
1710static inline bool lockdep_sock_is_held(const struct sock *sk)
1711{
1712	return lockdep_is_held(&sk->sk_lock) ||
1713	       lockdep_is_held(&sk->sk_lock.slock);
1714}
1715
1716void lock_sock_nested(struct sock *sk, int subclass);
1717
1718static inline void lock_sock(struct sock *sk)
1719{
1720	lock_sock_nested(sk, 0);
1721}
1722
1723void __lock_sock(struct sock *sk);
1724void __release_sock(struct sock *sk);
1725void release_sock(struct sock *sk);
1726
1727/* BH context may only use the following locking interface. */
1728#define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1729#define bh_lock_sock_nested(__sk) \
1730				spin_lock_nested(&((__sk)->sk_lock.slock), \
1731				SINGLE_DEPTH_NESTING)
1732#define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1733
1734bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1735
1736/**
1737 * lock_sock_fast - fast version of lock_sock
1738 * @sk: socket
1739 *
1740 * This version should be used for very small section, where process wont block
1741 * return false if fast path is taken:
1742 *
1743 *   sk_lock.slock locked, owned = 0, BH disabled
1744 *
1745 * return true if slow path is taken:
1746 *
1747 *   sk_lock.slock unlocked, owned = 1, BH enabled
1748 */
1749static inline bool lock_sock_fast(struct sock *sk)
1750{
1751	/* The sk_lock has mutex_lock() semantics here. */
1752	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1753
1754	return __lock_sock_fast(sk);
1755}
1756
1757/* fast socket lock variant for caller already holding a [different] socket lock */
1758static inline bool lock_sock_fast_nested(struct sock *sk)
1759{
1760	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1761
1762	return __lock_sock_fast(sk);
1763}
1764
1765/**
1766 * unlock_sock_fast - complement of lock_sock_fast
1767 * @sk: socket
1768 * @slow: slow mode
1769 *
1770 * fast unlock socket for user context.
1771 * If slow mode is on, we call regular release_sock()
1772 */
1773static inline void unlock_sock_fast(struct sock *sk, bool slow)
1774	__releases(&sk->sk_lock.slock)
1775{
1776	if (slow) {
1777		release_sock(sk);
1778		__release(&sk->sk_lock.slock);
1779	} else {
1780		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1781		spin_unlock_bh(&sk->sk_lock.slock);
1782	}
1783}
1784
1785void sockopt_lock_sock(struct sock *sk);
1786void sockopt_release_sock(struct sock *sk);
1787bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1788bool sockopt_capable(int cap);
1789
1790/* Used by processes to "lock" a socket state, so that
1791 * interrupts and bottom half handlers won't change it
1792 * from under us. It essentially blocks any incoming
1793 * packets, so that we won't get any new data or any
1794 * packets that change the state of the socket.
1795 *
1796 * While locked, BH processing will add new packets to
1797 * the backlog queue.  This queue is processed by the
1798 * owner of the socket lock right before it is released.
1799 *
1800 * Since ~2.3.5 it is also exclusive sleep lock serializing
1801 * accesses from user process context.
1802 */
1803
1804static inline void sock_owned_by_me(const struct sock *sk)
1805{
1806#ifdef CONFIG_LOCKDEP
1807	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1808#endif
1809}
1810
1811static inline bool sock_owned_by_user(const struct sock *sk)
1812{
1813	sock_owned_by_me(sk);
1814	return sk->sk_lock.owned;
1815}
1816
1817static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1818{
1819	return sk->sk_lock.owned;
1820}
1821
1822static inline void sock_release_ownership(struct sock *sk)
1823{
1824	if (sock_owned_by_user_nocheck(sk)) {
1825		sk->sk_lock.owned = 0;
1826
1827		/* The sk_lock has mutex_unlock() semantics: */
1828		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1829	}
1830}
1831
1832/* no reclassification while locks are held */
1833static inline bool sock_allow_reclassification(const struct sock *csk)
1834{
1835	struct sock *sk = (struct sock *)csk;
1836
1837	return !sock_owned_by_user_nocheck(sk) &&
1838		!spin_is_locked(&sk->sk_lock.slock);
1839}
1840
1841struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1842		      struct proto *prot, int kern);
1843void sk_free(struct sock *sk);
1844void sk_destruct(struct sock *sk);
1845struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1846void sk_free_unlock_clone(struct sock *sk);
1847
1848struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1849			     gfp_t priority);
1850void __sock_wfree(struct sk_buff *skb);
1851void sock_wfree(struct sk_buff *skb);
1852struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1853			     gfp_t priority);
1854void skb_orphan_partial(struct sk_buff *skb);
1855void sock_rfree(struct sk_buff *skb);
1856void sock_efree(struct sk_buff *skb);
1857#ifdef CONFIG_INET
1858void sock_edemux(struct sk_buff *skb);
1859void sock_pfree(struct sk_buff *skb);
1860#else
1861#define sock_edemux sock_efree
1862#endif
1863
1864int sk_setsockopt(struct sock *sk, int level, int optname,
1865		  sockptr_t optval, unsigned int optlen);
1866int sock_setsockopt(struct socket *sock, int level, int op,
1867		    sockptr_t optval, unsigned int optlen);
1868
1869int sk_getsockopt(struct sock *sk, int level, int optname,
1870		  sockptr_t optval, sockptr_t optlen);
1871int sock_getsockopt(struct socket *sock, int level, int op,
1872		    char __user *optval, int __user *optlen);
1873int sock_gettstamp(struct socket *sock, void __user *userstamp,
1874		   bool timeval, bool time32);
1875struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1876				     unsigned long data_len, int noblock,
1877				     int *errcode, int max_page_order);
1878
1879static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1880						  unsigned long size,
1881						  int noblock, int *errcode)
1882{
1883	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1884}
1885
1886void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1887void sock_kfree_s(struct sock *sk, void *mem, int size);
1888void sock_kzfree_s(struct sock *sk, void *mem, int size);
1889void sk_send_sigurg(struct sock *sk);
1890
1891static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1892{
1893	if (sk->sk_socket)
1894		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1895	WRITE_ONCE(sk->sk_prot, proto);
1896}
1897
1898struct sockcm_cookie {
1899	u64 transmit_time;
1900	u32 mark;
1901	u32 tsflags;
1902};
1903
1904static inline void sockcm_init(struct sockcm_cookie *sockc,
1905			       const struct sock *sk)
1906{
1907	*sockc = (struct sockcm_cookie) {
1908		.tsflags = READ_ONCE(sk->sk_tsflags)
1909	};
1910}
1911
1912int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1913		     struct sockcm_cookie *sockc);
1914int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1915		   struct sockcm_cookie *sockc);
1916
1917/*
1918 * Functions to fill in entries in struct proto_ops when a protocol
1919 * does not implement a particular function.
1920 */
1921int sock_no_bind(struct socket *, struct sockaddr *, int);
1922int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1923int sock_no_socketpair(struct socket *, struct socket *);
1924int sock_no_accept(struct socket *, struct socket *, int, bool);
1925int sock_no_getname(struct socket *, struct sockaddr *, int);
1926int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1927int sock_no_listen(struct socket *, int);
1928int sock_no_shutdown(struct socket *, int);
1929int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1930int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1931int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1932int sock_no_mmap(struct file *file, struct socket *sock,
1933		 struct vm_area_struct *vma);
1934
1935/*
1936 * Functions to fill in entries in struct proto_ops when a protocol
1937 * uses the inet style.
1938 */
1939int sock_common_getsockopt(struct socket *sock, int level, int optname,
1940				  char __user *optval, int __user *optlen);
1941int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1942			int flags);
1943int sock_common_setsockopt(struct socket *sock, int level, int optname,
1944			   sockptr_t optval, unsigned int optlen);
1945
1946void sk_common_release(struct sock *sk);
1947
1948/*
1949 *	Default socket callbacks and setup code
1950 */
1951
1952/* Initialise core socket variables using an explicit uid. */
1953void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1954
1955/* Initialise core socket variables.
1956 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1957 */
1958void sock_init_data(struct socket *sock, struct sock *sk);
1959
1960/*
1961 * Socket reference counting postulates.
1962 *
1963 * * Each user of socket SHOULD hold a reference count.
1964 * * Each access point to socket (an hash table bucket, reference from a list,
1965 *   running timer, skb in flight MUST hold a reference count.
1966 * * When reference count hits 0, it means it will never increase back.
1967 * * When reference count hits 0, it means that no references from
1968 *   outside exist to this socket and current process on current CPU
1969 *   is last user and may/should destroy this socket.
1970 * * sk_free is called from any context: process, BH, IRQ. When
1971 *   it is called, socket has no references from outside -> sk_free
1972 *   may release descendant resources allocated by the socket, but
1973 *   to the time when it is called, socket is NOT referenced by any
1974 *   hash tables, lists etc.
1975 * * Packets, delivered from outside (from network or from another process)
1976 *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1977 *   when they sit in queue. Otherwise, packets will leak to hole, when
1978 *   socket is looked up by one cpu and unhasing is made by another CPU.
1979 *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1980 *   (leak to backlog). Packet socket does all the processing inside
1981 *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1982 *   use separate SMP lock, so that they are prone too.
1983 */
1984
1985/* Ungrab socket and destroy it, if it was the last reference. */
1986static inline void sock_put(struct sock *sk)
1987{
1988	if (refcount_dec_and_test(&sk->sk_refcnt))
1989		sk_free(sk);
1990}
1991/* Generic version of sock_put(), dealing with all sockets
1992 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1993 */
1994void sock_gen_put(struct sock *sk);
1995
1996int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1997		     unsigned int trim_cap, bool refcounted);
1998static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1999				 const int nested)
2000{
2001	return __sk_receive_skb(sk, skb, nested, 1, true);
2002}
2003
2004static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
2005{
2006	/* sk_tx_queue_mapping accept only upto a 16-bit value */
2007	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
2008		return;
2009	/* Paired with READ_ONCE() in sk_tx_queue_get() and
2010	 * other WRITE_ONCE() because socket lock might be not held.
2011	 */
2012	WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
2013}
2014
2015#define NO_QUEUE_MAPPING	USHRT_MAX
2016
2017static inline void sk_tx_queue_clear(struct sock *sk)
2018{
2019	/* Paired with READ_ONCE() in sk_tx_queue_get() and
2020	 * other WRITE_ONCE() because socket lock might be not held.
2021	 */
2022	WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
2023}
2024
2025static inline int sk_tx_queue_get(const struct sock *sk)
2026{
2027	if (sk) {
2028		/* Paired with WRITE_ONCE() in sk_tx_queue_clear()
2029		 * and sk_tx_queue_set().
2030		 */
2031		int val = READ_ONCE(sk->sk_tx_queue_mapping);
2032
2033		if (val != NO_QUEUE_MAPPING)
2034			return val;
2035	}
2036	return -1;
2037}
2038
2039static inline void __sk_rx_queue_set(struct sock *sk,
2040				     const struct sk_buff *skb,
2041				     bool force_set)
2042{
2043#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2044	if (skb_rx_queue_recorded(skb)) {
2045		u16 rx_queue = skb_get_rx_queue(skb);
2046
2047		if (force_set ||
2048		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2049			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2050	}
2051#endif
2052}
2053
2054static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2055{
2056	__sk_rx_queue_set(sk, skb, true);
2057}
2058
2059static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2060{
2061	__sk_rx_queue_set(sk, skb, false);
2062}
2063
2064static inline void sk_rx_queue_clear(struct sock *sk)
2065{
2066#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2067	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2068#endif
2069}
2070
2071static inline int sk_rx_queue_get(const struct sock *sk)
2072{
2073#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2074	if (sk) {
2075		int res = READ_ONCE(sk->sk_rx_queue_mapping);
2076
2077		if (res != NO_QUEUE_MAPPING)
2078			return res;
2079	}
2080#endif
2081
2082	return -1;
2083}
2084
2085static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2086{
2087	sk->sk_socket = sock;
2088}
2089
2090static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2091{
2092	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2093	return &rcu_dereference_raw(sk->sk_wq)->wait;
2094}
2095/* Detach socket from process context.
2096 * Announce socket dead, detach it from wait queue and inode.
2097 * Note that parent inode held reference count on this struct sock,
2098 * we do not release it in this function, because protocol
2099 * probably wants some additional cleanups or even continuing
2100 * to work with this socket (TCP).
2101 */
2102static inline void sock_orphan(struct sock *sk)
2103{
2104	write_lock_bh(&sk->sk_callback_lock);
2105	sock_set_flag(sk, SOCK_DEAD);
2106	sk_set_socket(sk, NULL);
2107	sk->sk_wq  = NULL;
2108	write_unlock_bh(&sk->sk_callback_lock);
2109}
2110
2111static inline void sock_graft(struct sock *sk, struct socket *parent)
2112{
2113	WARN_ON(parent->sk);
2114	write_lock_bh(&sk->sk_callback_lock);
2115	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2116	parent->sk = sk;
2117	sk_set_socket(sk, parent);
2118	sk->sk_uid = SOCK_INODE(parent)->i_uid;
2119	security_sock_graft(sk, parent);
2120	write_unlock_bh(&sk->sk_callback_lock);
2121}
2122
2123kuid_t sock_i_uid(struct sock *sk);
2124unsigned long __sock_i_ino(struct sock *sk);
2125unsigned long sock_i_ino(struct sock *sk);
2126
2127static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2128{
2129	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2130}
2131
2132static inline u32 net_tx_rndhash(void)
2133{
2134	u32 v = get_random_u32();
2135
2136	return v ?: 1;
2137}
2138
2139static inline void sk_set_txhash(struct sock *sk)
2140{
2141	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2142	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2143}
2144
2145static inline bool sk_rethink_txhash(struct sock *sk)
2146{
2147	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2148		sk_set_txhash(sk);
2149		return true;
2150	}
2151	return false;
2152}
2153
2154static inline struct dst_entry *
2155__sk_dst_get(const struct sock *sk)
2156{
2157	return rcu_dereference_check(sk->sk_dst_cache,
2158				     lockdep_sock_is_held(sk));
2159}
2160
2161static inline struct dst_entry *
2162sk_dst_get(const struct sock *sk)
2163{
2164	struct dst_entry *dst;
2165
2166	rcu_read_lock();
2167	dst = rcu_dereference(sk->sk_dst_cache);
2168	if (dst && !rcuref_get(&dst->__rcuref))
2169		dst = NULL;
2170	rcu_read_unlock();
2171	return dst;
2172}
2173
2174static inline void __dst_negative_advice(struct sock *sk)
2175{
2176	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2177
2178	if (dst && dst->ops->negative_advice) {
2179		ndst = dst->ops->negative_advice(dst);
2180
2181		if (ndst != dst) {
2182			rcu_assign_pointer(sk->sk_dst_cache, ndst);
2183			sk_tx_queue_clear(sk);
2184			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2185		}
2186	}
2187}
2188
2189static inline void dst_negative_advice(struct sock *sk)
2190{
2191	sk_rethink_txhash(sk);
2192	__dst_negative_advice(sk);
2193}
2194
2195static inline void
2196__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2197{
2198	struct dst_entry *old_dst;
2199
2200	sk_tx_queue_clear(sk);
2201	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2202	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2203					    lockdep_sock_is_held(sk));
2204	rcu_assign_pointer(sk->sk_dst_cache, dst);
2205	dst_release(old_dst);
2206}
2207
2208static inline void
2209sk_dst_set(struct sock *sk, struct dst_entry *dst)
2210{
2211	struct dst_entry *old_dst;
2212
2213	sk_tx_queue_clear(sk);
2214	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2215	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2216	dst_release(old_dst);
2217}
2218
2219static inline void
2220__sk_dst_reset(struct sock *sk)
2221{
2222	__sk_dst_set(sk, NULL);
2223}
2224
2225static inline void
2226sk_dst_reset(struct sock *sk)
2227{
2228	sk_dst_set(sk, NULL);
2229}
2230
2231struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2232
2233struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2234
2235static inline void sk_dst_confirm(struct sock *sk)
2236{
2237	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2238		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2239}
2240
2241static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2242{
2243	if (skb_get_dst_pending_confirm(skb)) {
2244		struct sock *sk = skb->sk;
2245
2246		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2247			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2248		neigh_confirm(n);
2249	}
2250}
2251
2252bool sk_mc_loop(struct sock *sk);
2253
2254static inline bool sk_can_gso(const struct sock *sk)
2255{
2256	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2257}
2258
2259void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2260
2261static inline void sk_gso_disable(struct sock *sk)
2262{
2263	sk->sk_gso_disabled = 1;
2264	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2265}
2266
2267static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2268					   struct iov_iter *from, char *to,
2269					   int copy, int offset)
2270{
2271	if (skb->ip_summed == CHECKSUM_NONE) {
2272		__wsum csum = 0;
2273		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2274			return -EFAULT;
2275		skb->csum = csum_block_add(skb->csum, csum, offset);
2276	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2277		if (!copy_from_iter_full_nocache(to, copy, from))
2278			return -EFAULT;
2279	} else if (!copy_from_iter_full(to, copy, from))
2280		return -EFAULT;
2281
2282	return 0;
2283}
2284
2285static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2286				       struct iov_iter *from, int copy)
2287{
2288	int err, offset = skb->len;
2289
2290	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2291				       copy, offset);
2292	if (err)
2293		__skb_trim(skb, offset);
2294
2295	return err;
2296}
2297
2298static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2299					   struct sk_buff *skb,
2300					   struct page *page,
2301					   int off, int copy)
2302{
2303	int err;
2304
2305	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2306				       copy, skb->len);
2307	if (err)
2308		return err;
2309
2310	skb_len_add(skb, copy);
2311	sk_wmem_queued_add(sk, copy);
2312	sk_mem_charge(sk, copy);
2313	return 0;
2314}
2315
2316/**
2317 * sk_wmem_alloc_get - returns write allocations
2318 * @sk: socket
2319 *
2320 * Return: sk_wmem_alloc minus initial offset of one
2321 */
2322static inline int sk_wmem_alloc_get(const struct sock *sk)
2323{
2324	return refcount_read(&sk->sk_wmem_alloc) - 1;
2325}
2326
2327/**
2328 * sk_rmem_alloc_get - returns read allocations
2329 * @sk: socket
2330 *
2331 * Return: sk_rmem_alloc
2332 */
2333static inline int sk_rmem_alloc_get(const struct sock *sk)
2334{
2335	return atomic_read(&sk->sk_rmem_alloc);
2336}
2337
2338/**
2339 * sk_has_allocations - check if allocations are outstanding
2340 * @sk: socket
2341 *
2342 * Return: true if socket has write or read allocations
2343 */
2344static inline bool sk_has_allocations(const struct sock *sk)
2345{
2346	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2347}
2348
2349/**
2350 * skwq_has_sleeper - check if there are any waiting processes
2351 * @wq: struct socket_wq
2352 *
2353 * Return: true if socket_wq has waiting processes
2354 *
2355 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2356 * barrier call. They were added due to the race found within the tcp code.
2357 *
2358 * Consider following tcp code paths::
2359 *
2360 *   CPU1                CPU2
2361 *   sys_select          receive packet
2362 *   ...                 ...
2363 *   __add_wait_queue    update tp->rcv_nxt
2364 *   ...                 ...
2365 *   tp->rcv_nxt check   sock_def_readable
2366 *   ...                 {
2367 *   schedule               rcu_read_lock();
2368 *                          wq = rcu_dereference(sk->sk_wq);
2369 *                          if (wq && waitqueue_active(&wq->wait))
2370 *                              wake_up_interruptible(&wq->wait)
2371 *                          ...
2372 *                       }
2373 *
2374 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2375 * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2376 * could then endup calling schedule and sleep forever if there are no more
2377 * data on the socket.
2378 *
2379 */
2380static inline bool skwq_has_sleeper(struct socket_wq *wq)
2381{
2382	return wq && wq_has_sleeper(&wq->wait);
2383}
2384
2385/**
2386 * sock_poll_wait - place memory barrier behind the poll_wait call.
2387 * @filp:           file
2388 * @sock:           socket to wait on
2389 * @p:              poll_table
2390 *
2391 * See the comments in the wq_has_sleeper function.
2392 */
2393static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2394				  poll_table *p)
2395{
2396	if (!poll_does_not_wait(p)) {
2397		poll_wait(filp, &sock->wq.wait, p);
2398		/* We need to be sure we are in sync with the
2399		 * socket flags modification.
2400		 *
2401		 * This memory barrier is paired in the wq_has_sleeper.
2402		 */
2403		smp_mb();
2404	}
2405}
2406
2407static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2408{
2409	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2410	u32 txhash = READ_ONCE(sk->sk_txhash);
2411
2412	if (txhash) {
2413		skb->l4_hash = 1;
2414		skb->hash = txhash;
2415	}
2416}
2417
2418void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2419
2420/*
2421 *	Queue a received datagram if it will fit. Stream and sequenced
2422 *	protocols can't normally use this as they need to fit buffers in
2423 *	and play with them.
2424 *
2425 *	Inlined as it's very short and called for pretty much every
2426 *	packet ever received.
2427 */
2428static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2429{
2430	skb_orphan(skb);
2431	skb->sk = sk;
2432	skb->destructor = sock_rfree;
2433	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2434	sk_mem_charge(sk, skb->truesize);
2435}
2436
2437static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2438{
2439	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2440		skb_orphan(skb);
2441		skb->destructor = sock_efree;
2442		skb->sk = sk;
2443		return true;
2444	}
2445	return false;
2446}
2447
2448static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2449{
2450	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2451	if (skb) {
2452		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2453			skb_set_owner_r(skb, sk);
2454			return skb;
2455		}
2456		__kfree_skb(skb);
2457	}
2458	return NULL;
2459}
2460
2461static inline void skb_prepare_for_gro(struct sk_buff *skb)
2462{
2463	if (skb->destructor != sock_wfree) {
2464		skb_orphan(skb);
2465		return;
2466	}
2467	skb->slow_gro = 1;
2468}
2469
2470void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2471		    unsigned long expires);
2472
2473void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2474
2475void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2476
2477int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2478			struct sk_buff *skb, unsigned int flags,
2479			void (*destructor)(struct sock *sk,
2480					   struct sk_buff *skb));
2481int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2482
2483int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2484			      enum skb_drop_reason *reason);
2485
2486static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2487{
2488	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2489}
2490
2491int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2492struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2493
2494/*
2495 *	Recover an error report and clear atomically
2496 */
2497
2498static inline int sock_error(struct sock *sk)
2499{
2500	int err;
2501
2502	/* Avoid an atomic operation for the common case.
2503	 * This is racy since another cpu/thread can change sk_err under us.
2504	 */
2505	if (likely(data_race(!sk->sk_err)))
2506		return 0;
2507
2508	err = xchg(&sk->sk_err, 0);
2509	return -err;
2510}
2511
2512void sk_error_report(struct sock *sk);
2513
2514static inline unsigned long sock_wspace(struct sock *sk)
2515{
2516	int amt = 0;
2517
2518	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2519		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2520		if (amt < 0)
2521			amt = 0;
2522	}
2523	return amt;
2524}
2525
2526/* Note:
2527 *  We use sk->sk_wq_raw, from contexts knowing this
2528 *  pointer is not NULL and cannot disappear/change.
2529 */
2530static inline void sk_set_bit(int nr, struct sock *sk)
2531{
2532	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2533	    !sock_flag(sk, SOCK_FASYNC))
2534		return;
2535
2536	set_bit(nr, &sk->sk_wq_raw->flags);
2537}
2538
2539static inline void sk_clear_bit(int nr, struct sock *sk)
2540{
2541	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2542	    !sock_flag(sk, SOCK_FASYNC))
2543		return;
2544
2545	clear_bit(nr, &sk->sk_wq_raw->flags);
2546}
2547
2548static inline void sk_wake_async(const struct sock *sk, int how, int band)
2549{
2550	if (sock_flag(sk, SOCK_FASYNC)) {
2551		rcu_read_lock();
2552		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2553		rcu_read_unlock();
2554	}
2555}
2556
2557/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2558 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2559 * Note: for send buffers, TCP works better if we can build two skbs at
2560 * minimum.
2561 */
2562#define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2563
2564#define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2565#define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2566
2567static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2568{
2569	u32 val;
2570
2571	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2572		return;
2573
2574	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2575	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2576
2577	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2578}
2579
2580/**
2581 * sk_page_frag - return an appropriate page_frag
2582 * @sk: socket
2583 *
2584 * Use the per task page_frag instead of the per socket one for
2585 * optimization when we know that we're in process context and own
2586 * everything that's associated with %current.
2587 *
2588 * Both direct reclaim and page faults can nest inside other
2589 * socket operations and end up recursing into sk_page_frag()
2590 * while it's already in use: explicitly avoid task page_frag
2591 * when users disable sk_use_task_frag.
2592 *
2593 * Return: a per task page_frag if context allows that,
2594 * otherwise a per socket one.
2595 */
2596static inline struct page_frag *sk_page_frag(struct sock *sk)
2597{
2598	if (sk->sk_use_task_frag)
2599		return &current->task_frag;
2600
2601	return &sk->sk_frag;
2602}
2603
2604bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2605
2606/*
2607 *	Default write policy as shown to user space via poll/select/SIGIO
2608 */
2609static inline bool sock_writeable(const struct sock *sk)
2610{
2611	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2612}
2613
2614static inline gfp_t gfp_any(void)
2615{
2616	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2617}
2618
2619static inline gfp_t gfp_memcg_charge(void)
2620{
2621	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2622}
2623
2624static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2625{
2626	return noblock ? 0 : sk->sk_rcvtimeo;
2627}
2628
2629static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2630{
2631	return noblock ? 0 : sk->sk_sndtimeo;
2632}
2633
2634static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2635{
2636	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2637
2638	return v ?: 1;
2639}
2640
2641/* Alas, with timeout socket operations are not restartable.
2642 * Compare this to poll().
2643 */
2644static inline int sock_intr_errno(long timeo)
2645{
2646	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2647}
2648
2649struct sock_skb_cb {
2650	u32 dropcount;
2651};
2652
2653/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2654 * using skb->cb[] would keep using it directly and utilize its
2655 * alignement guarantee.
2656 */
2657#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2658			    sizeof(struct sock_skb_cb)))
2659
2660#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2661			    SOCK_SKB_CB_OFFSET))
2662
2663#define sock_skb_cb_check_size(size) \
2664	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2665
2666static inline void
2667sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2668{
2669	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2670						atomic_read(&sk->sk_drops) : 0;
2671}
2672
2673static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2674{
2675	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2676
2677	atomic_add(segs, &sk->sk_drops);
2678}
2679
2680static inline ktime_t sock_read_timestamp(struct sock *sk)
2681{
2682#if BITS_PER_LONG==32
2683	unsigned int seq;
2684	ktime_t kt;
2685
2686	do {
2687		seq = read_seqbegin(&sk->sk_stamp_seq);
2688		kt = sk->sk_stamp;
2689	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2690
2691	return kt;
2692#else
2693	return READ_ONCE(sk->sk_stamp);
2694#endif
2695}
2696
2697static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2698{
2699#if BITS_PER_LONG==32
2700	write_seqlock(&sk->sk_stamp_seq);
2701	sk->sk_stamp = kt;
2702	write_sequnlock(&sk->sk_stamp_seq);
2703#else
2704	WRITE_ONCE(sk->sk_stamp, kt);
2705#endif
2706}
2707
2708void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2709			   struct sk_buff *skb);
2710void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2711			     struct sk_buff *skb);
2712
2713static inline void
2714sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2715{
2716	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2717	u32 tsflags = READ_ONCE(sk->sk_tsflags);
2718	ktime_t kt = skb->tstamp;
2719	/*
2720	 * generate control messages if
2721	 * - receive time stamping in software requested
2722	 * - software time stamp available and wanted
2723	 * - hardware time stamps available and wanted
2724	 */
2725	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2726	    (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2727	    (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2728	    (hwtstamps->hwtstamp &&
2729	     (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2730		__sock_recv_timestamp(msg, sk, skb);
2731	else
2732		sock_write_timestamp(sk, kt);
2733
2734	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2735		__sock_recv_wifi_status(msg, sk, skb);
2736}
2737
2738void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2739		       struct sk_buff *skb);
2740
2741#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2742static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2743				   struct sk_buff *skb)
2744{
2745#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2746			   (1UL << SOCK_RCVTSTAMP)			| \
2747			   (1UL << SOCK_RCVMARK))
2748#define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2749			   SOF_TIMESTAMPING_RAW_HARDWARE)
2750
2751	if (sk->sk_flags & FLAGS_RECV_CMSGS ||
2752	    READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY)
2753		__sock_recv_cmsgs(msg, sk, skb);
2754	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2755		sock_write_timestamp(sk, skb->tstamp);
2756	else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2757		sock_write_timestamp(sk, 0);
2758}
2759
2760void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2761
2762/**
2763 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2764 * @sk:		socket sending this packet
2765 * @tsflags:	timestamping flags to use
2766 * @tx_flags:	completed with instructions for time stamping
2767 * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2768 *
2769 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2770 */
2771static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2772				      __u8 *tx_flags, __u32 *tskey)
2773{
2774	if (unlikely(tsflags)) {
2775		__sock_tx_timestamp(tsflags, tx_flags);
2776		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2777		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2778			*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2779	}
2780	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2781		*tx_flags |= SKBTX_WIFI_STATUS;
2782}
2783
2784static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2785				     __u8 *tx_flags)
2786{
2787	_sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2788}
2789
2790static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2791{
2792	_sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2793			   &skb_shinfo(skb)->tskey);
2794}
2795
2796static inline bool sk_is_inet(const struct sock *sk)
2797{
2798	int family = READ_ONCE(sk->sk_family);
2799
2800	return family == AF_INET || family == AF_INET6;
2801}
2802
2803static inline bool sk_is_tcp(const struct sock *sk)
2804{
2805	return sk_is_inet(sk) &&
2806	       sk->sk_type == SOCK_STREAM &&
2807	       sk->sk_protocol == IPPROTO_TCP;
2808}
2809
2810static inline bool sk_is_udp(const struct sock *sk)
2811{
2812	return sk_is_inet(sk) &&
2813	       sk->sk_type == SOCK_DGRAM &&
2814	       sk->sk_protocol == IPPROTO_UDP;
2815}
2816
2817static inline bool sk_is_stream_unix(const struct sock *sk)
2818{
2819	return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM;
2820}
2821
2822/**
2823 * sk_eat_skb - Release a skb if it is no longer needed
2824 * @sk: socket to eat this skb from
2825 * @skb: socket buffer to eat
2826 *
2827 * This routine must be called with interrupts disabled or with the socket
2828 * locked so that the sk_buff queue operation is ok.
2829*/
2830static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2831{
2832	__skb_unlink(skb, &sk->sk_receive_queue);
2833	__kfree_skb(skb);
2834}
2835
2836static inline bool
2837skb_sk_is_prefetched(struct sk_buff *skb)
2838{
2839#ifdef CONFIG_INET
2840	return skb->destructor == sock_pfree;
2841#else
2842	return false;
2843#endif /* CONFIG_INET */
2844}
2845
2846/* This helper checks if a socket is a full socket,
2847 * ie _not_ a timewait or request socket.
2848 */
2849static inline bool sk_fullsock(const struct sock *sk)
2850{
2851	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2852}
2853
2854static inline bool
2855sk_is_refcounted(struct sock *sk)
2856{
2857	/* Only full sockets have sk->sk_flags. */
2858	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2859}
2860
2861/**
2862 * skb_steal_sock - steal a socket from an sk_buff
2863 * @skb: sk_buff to steal the socket from
2864 * @refcounted: is set to true if the socket is reference-counted
2865 * @prefetched: is set to true if the socket was assigned from bpf
2866 */
2867static inline struct sock *
2868skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched)
2869{
2870	if (skb->sk) {
2871		struct sock *sk = skb->sk;
2872
2873		*refcounted = true;
2874		*prefetched = skb_sk_is_prefetched(skb);
2875		if (*prefetched)
2876			*refcounted = sk_is_refcounted(sk);
2877		skb->destructor = NULL;
2878		skb->sk = NULL;
2879		return sk;
2880	}
2881	*prefetched = false;
2882	*refcounted = false;
2883	return NULL;
2884}
2885
2886/* Checks if this SKB belongs to an HW offloaded socket
2887 * and whether any SW fallbacks are required based on dev.
2888 * Check decrypted mark in case skb_orphan() cleared socket.
2889 */
2890static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2891						   struct net_device *dev)
2892{
2893#ifdef CONFIG_SOCK_VALIDATE_XMIT
2894	struct sock *sk = skb->sk;
2895
2896	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2897		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2898#ifdef CONFIG_TLS_DEVICE
2899	} else if (unlikely(skb->decrypted)) {
2900		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2901		kfree_skb(skb);
2902		skb = NULL;
2903#endif
2904	}
2905#endif
2906
2907	return skb;
2908}
2909
2910/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2911 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2912 */
2913static inline bool sk_listener(const struct sock *sk)
2914{
2915	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2916}
2917
2918void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2919int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2920		       int type);
2921
2922bool sk_ns_capable(const struct sock *sk,
2923		   struct user_namespace *user_ns, int cap);
2924bool sk_capable(const struct sock *sk, int cap);
2925bool sk_net_capable(const struct sock *sk, int cap);
2926
2927void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2928
2929/* Take into consideration the size of the struct sk_buff overhead in the
2930 * determination of these values, since that is non-constant across
2931 * platforms.  This makes socket queueing behavior and performance
2932 * not depend upon such differences.
2933 */
2934#define _SK_MEM_PACKETS		256
2935#define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2936#define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2937#define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2938
2939extern __u32 sysctl_wmem_max;
2940extern __u32 sysctl_rmem_max;
2941
2942extern int sysctl_tstamp_allow_data;
2943extern int sysctl_optmem_max;
2944
2945extern __u32 sysctl_wmem_default;
2946extern __u32 sysctl_rmem_default;
2947
2948#define SKB_FRAG_PAGE_ORDER	get_order(32768)
2949DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2950
2951static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2952{
2953	/* Does this proto have per netns sysctl_wmem ? */
2954	if (proto->sysctl_wmem_offset)
2955		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2956
2957	return READ_ONCE(*proto->sysctl_wmem);
2958}
2959
2960static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2961{
2962	/* Does this proto have per netns sysctl_rmem ? */
2963	if (proto->sysctl_rmem_offset)
2964		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2965
2966	return READ_ONCE(*proto->sysctl_rmem);
2967}
2968
2969/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2970 * Some wifi drivers need to tweak it to get more chunks.
2971 * They can use this helper from their ndo_start_xmit()
2972 */
2973static inline void sk_pacing_shift_update(struct sock *sk, int val)
2974{
2975	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2976		return;
2977	WRITE_ONCE(sk->sk_pacing_shift, val);
2978}
2979
2980/* if a socket is bound to a device, check that the given device
2981 * index is either the same or that the socket is bound to an L3
2982 * master device and the given device index is also enslaved to
2983 * that L3 master
2984 */
2985static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2986{
2987	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2988	int mdif;
2989
2990	if (!bound_dev_if || bound_dev_if == dif)
2991		return true;
2992
2993	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2994	if (mdif && mdif == bound_dev_if)
2995		return true;
2996
2997	return false;
2998}
2999
3000void sock_def_readable(struct sock *sk);
3001
3002int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
3003void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
3004int sock_set_timestamping(struct sock *sk, int optname,
3005			  struct so_timestamping timestamping);
3006
3007void sock_enable_timestamps(struct sock *sk);
3008void sock_no_linger(struct sock *sk);
3009void sock_set_keepalive(struct sock *sk);
3010void sock_set_priority(struct sock *sk, u32 priority);
3011void sock_set_rcvbuf(struct sock *sk, int val);
3012void sock_set_mark(struct sock *sk, u32 val);
3013void sock_set_reuseaddr(struct sock *sk);
3014void sock_set_reuseport(struct sock *sk);
3015void sock_set_sndtimeo(struct sock *sk, s64 secs);
3016
3017int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
3018
3019int sock_get_timeout(long timeo, void *optval, bool old_timeval);
3020int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
3021			   sockptr_t optval, int optlen, bool old_timeval);
3022
3023int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
3024		     void __user *arg, void *karg, size_t size);
3025int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
3026static inline bool sk_is_readable(struct sock *sk)
3027{
3028	if (sk->sk_prot->sock_is_readable)
3029		return sk->sk_prot->sock_is_readable(sk);
3030	return false;
3031}
3032#endif	/* _SOCK_H */
3033