Linux Audio

Check our new training course

Loading...
v6.13.7
   1/* SPDX-License-Identifier: GPL-2.0-or-later */
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Definitions for the AF_INET socket handler.
   8 *
   9 * Version:	@(#)sock.h	1.0.4	05/13/93
  10 *
  11 * Authors:	Ross Biro
  12 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  13 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
  14 *		Florian La Roche <flla@stud.uni-sb.de>
  15 *
  16 * Fixes:
  17 *		Alan Cox	:	Volatiles in skbuff pointers. See
  18 *					skbuff comments. May be overdone,
  19 *					better to prove they can be removed
  20 *					than the reverse.
  21 *		Alan Cox	:	Added a zapped field for tcp to note
  22 *					a socket is reset and must stay shut up
  23 *		Alan Cox	:	New fields for options
  24 *	Pauline Middelink	:	identd support
  25 *		Alan Cox	:	Eliminate low level recv/recvfrom
  26 *		David S. Miller	:	New socket lookup architecture.
  27 *              Steve Whitehouse:       Default routines for sock_ops
  28 *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
  29 *              			protinfo be just a void pointer, as the
  30 *              			protocol specific parts were moved to
  31 *              			respective headers and ipv4/v6, etc now
  32 *              			use private slabcaches for its socks
  33 *              Pedro Hortas	:	New flags field for socket options
  34 */
  35#ifndef _SOCK_H
  36#define _SOCK_H
  37
  38#include <linux/hardirq.h>
  39#include <linux/kernel.h>
  40#include <linux/list.h>
  41#include <linux/list_nulls.h>
  42#include <linux/timer.h>
  43#include <linux/cache.h>
  44#include <linux/bitops.h>
  45#include <linux/lockdep.h>
  46#include <linux/netdevice.h>
  47#include <linux/skbuff.h>	/* struct sk_buff */
  48#include <linux/mm.h>
  49#include <linux/security.h>
  50#include <linux/slab.h>
  51#include <linux/uaccess.h>
  52#include <linux/page_counter.h>
  53#include <linux/memcontrol.h>
  54#include <linux/static_key.h>
  55#include <linux/sched.h>
  56#include <linux/wait.h>
  57#include <linux/cgroup-defs.h>
  58#include <linux/rbtree.h>
  59#include <linux/rculist_nulls.h>
  60#include <linux/poll.h>
  61#include <linux/sockptr.h>
  62#include <linux/indirect_call_wrapper.h>
  63#include <linux/atomic.h>
  64#include <linux/refcount.h>
  65#include <linux/llist.h>
  66#include <net/dst.h>
  67#include <net/checksum.h>
  68#include <net/tcp_states.h>
  69#include <linux/net_tstamp.h>
  70#include <net/l3mdev.h>
  71#include <uapi/linux/socket.h>
  72
  73/*
  74 * This structure really needs to be cleaned up.
  75 * Most of it is for TCP, and not used by any of
  76 * the other protocols.
  77 */
  78
  79/* This is the per-socket lock.  The spinlock provides a synchronization
  80 * between user contexts and software interrupt processing, whereas the
  81 * mini-semaphore synchronizes multiple users amongst themselves.
  82 */
  83typedef struct {
  84	spinlock_t		slock;
  85	int			owned;
  86	wait_queue_head_t	wq;
  87	/*
  88	 * We express the mutex-alike socket_lock semantics
  89	 * to the lock validator by explicitly managing
  90	 * the slock as a lock variant (in addition to
  91	 * the slock itself):
  92	 */
  93#ifdef CONFIG_DEBUG_LOCK_ALLOC
  94	struct lockdep_map dep_map;
  95#endif
  96} socket_lock_t;
  97
  98struct sock;
  99struct proto;
 100struct net;
 101
 102typedef __u32 __bitwise __portpair;
 103typedef __u64 __bitwise __addrpair;
 104
 105/**
 106 *	struct sock_common - minimal network layer representation of sockets
 107 *	@skc_daddr: Foreign IPv4 addr
 108 *	@skc_rcv_saddr: Bound local IPv4 addr
 109 *	@skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
 110 *	@skc_hash: hash value used with various protocol lookup tables
 111 *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
 112 *	@skc_dport: placeholder for inet_dport/tw_dport
 113 *	@skc_num: placeholder for inet_num/tw_num
 114 *	@skc_portpair: __u32 union of @skc_dport & @skc_num
 115 *	@skc_family: network address family
 116 *	@skc_state: Connection state
 117 *	@skc_reuse: %SO_REUSEADDR setting
 118 *	@skc_reuseport: %SO_REUSEPORT setting
 119 *	@skc_ipv6only: socket is IPV6 only
 120 *	@skc_net_refcnt: socket is using net ref counting
 121 *	@skc_bound_dev_if: bound device index if != 0
 122 *	@skc_bind_node: bind hash linkage for various protocol lookup tables
 123 *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
 124 *	@skc_prot: protocol handlers inside a network family
 125 *	@skc_net: reference to the network namespace of this socket
 126 *	@skc_v6_daddr: IPV6 destination address
 127 *	@skc_v6_rcv_saddr: IPV6 source address
 128 *	@skc_cookie: socket's cookie value
 129 *	@skc_node: main hash linkage for various protocol lookup tables
 130 *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
 131 *	@skc_tx_queue_mapping: tx queue number for this connection
 132 *	@skc_rx_queue_mapping: rx queue number for this connection
 133 *	@skc_flags: place holder for sk_flags
 134 *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
 135 *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
 136 *	@skc_listener: connection request listener socket (aka rsk_listener)
 137 *		[union with @skc_flags]
 138 *	@skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
 139 *		[union with @skc_flags]
 140 *	@skc_incoming_cpu: record/match cpu processing incoming packets
 141 *	@skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
 142 *		[union with @skc_incoming_cpu]
 143 *	@skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
 144 *		[union with @skc_incoming_cpu]
 145 *	@skc_refcnt: reference count
 146 *
 147 *	This is the minimal network layer representation of sockets, the header
 148 *	for struct sock and struct inet_timewait_sock.
 149 */
 150struct sock_common {
 151	union {
 152		__addrpair	skc_addrpair;
 153		struct {
 154			__be32	skc_daddr;
 155			__be32	skc_rcv_saddr;
 156		};
 157	};
 158	union  {
 159		unsigned int	skc_hash;
 160		__u16		skc_u16hashes[2];
 161	};
 162	/* skc_dport && skc_num must be grouped as well */
 163	union {
 164		__portpair	skc_portpair;
 165		struct {
 166			__be16	skc_dport;
 167			__u16	skc_num;
 168		};
 169	};
 170
 171	unsigned short		skc_family;
 172	volatile unsigned char	skc_state;
 173	unsigned char		skc_reuse:4;
 174	unsigned char		skc_reuseport:1;
 175	unsigned char		skc_ipv6only:1;
 176	unsigned char		skc_net_refcnt:1;
 177	int			skc_bound_dev_if;
 178	union {
 179		struct hlist_node	skc_bind_node;
 180		struct hlist_node	skc_portaddr_node;
 181	};
 182	struct proto		*skc_prot;
 183	possible_net_t		skc_net;
 184
 185#if IS_ENABLED(CONFIG_IPV6)
 186	struct in6_addr		skc_v6_daddr;
 187	struct in6_addr		skc_v6_rcv_saddr;
 188#endif
 189
 190	atomic64_t		skc_cookie;
 191
 192	/* following fields are padding to force
 193	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
 194	 * assuming IPV6 is enabled. We use this padding differently
 195	 * for different kind of 'sockets'
 196	 */
 197	union {
 198		unsigned long	skc_flags;
 199		struct sock	*skc_listener; /* request_sock */
 200		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
 201	};
 202	/*
 203	 * fields between dontcopy_begin/dontcopy_end
 204	 * are not copied in sock_copy()
 205	 */
 206	/* private: */
 207	int			skc_dontcopy_begin[0];
 208	/* public: */
 209	union {
 210		struct hlist_node	skc_node;
 211		struct hlist_nulls_node skc_nulls_node;
 212	};
 213	unsigned short		skc_tx_queue_mapping;
 214#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
 215	unsigned short		skc_rx_queue_mapping;
 216#endif
 217	union {
 218		int		skc_incoming_cpu;
 219		u32		skc_rcv_wnd;
 220		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
 221	};
 222
 223	refcount_t		skc_refcnt;
 224	/* private: */
 225	int                     skc_dontcopy_end[0];
 226	union {
 227		u32		skc_rxhash;
 228		u32		skc_window_clamp;
 229		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
 230	};
 231	/* public: */
 232};
 233
 234struct bpf_local_storage;
 235struct sk_filter;
 236
 237/**
 238  *	struct sock - network layer representation of sockets
 239  *	@__sk_common: shared layout with inet_timewait_sock
 240  *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
 241  *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
 242  *	@sk_lock:	synchronizer
 243  *	@sk_kern_sock: True if sock is using kernel lock classes
 244  *	@sk_rcvbuf: size of receive buffer in bytes
 245  *	@sk_wq: sock wait queue and async head
 246  *	@sk_rx_dst: receive input route used by early demux
 247  *	@sk_rx_dst_ifindex: ifindex for @sk_rx_dst
 248  *	@sk_rx_dst_cookie: cookie for @sk_rx_dst
 249  *	@sk_dst_cache: destination cache
 250  *	@sk_dst_pending_confirm: need to confirm neighbour
 251  *	@sk_policy: flow policy
 252  *	@sk_receive_queue: incoming packets
 253  *	@sk_wmem_alloc: transmit queue bytes committed
 254  *	@sk_tsq_flags: TCP Small Queues flags
 255  *	@sk_write_queue: Packet sending queue
 256  *	@sk_omem_alloc: "o" is "option" or "other"
 257  *	@sk_wmem_queued: persistent queue size
 258  *	@sk_forward_alloc: space allocated forward
 259  *	@sk_reserved_mem: space reserved and non-reclaimable for the socket
 260  *	@sk_napi_id: id of the last napi context to receive data for sk
 261  *	@sk_ll_usec: usecs to busypoll when there is no data
 262  *	@sk_allocation: allocation mode
 263  *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
 264  *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
 265  *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
 266  *	@sk_sndbuf: size of send buffer in bytes
 267  *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
 268  *	@sk_no_check_rx: allow zero checksum in RX packets
 269  *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
 270  *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
 271  *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
 272  *	@sk_gso_max_size: Maximum GSO segment size to build
 273  *	@sk_gso_max_segs: Maximum number of GSO segments
 274  *	@sk_pacing_shift: scaling factor for TCP Small Queues
 275  *	@sk_lingertime: %SO_LINGER l_linger setting
 276  *	@sk_backlog: always used with the per-socket spinlock held
 277  *	@sk_callback_lock: used with the callbacks in the end of this struct
 278  *	@sk_error_queue: rarely used
 279  *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
 280  *			  IPV6_ADDRFORM for instance)
 281  *	@sk_err: last error
 282  *	@sk_err_soft: errors that don't cause failure but are the cause of a
 283  *		      persistent failure not just 'timed out'
 284  *	@sk_drops: raw/udp drops counter
 285  *	@sk_ack_backlog: current listen backlog
 286  *	@sk_max_ack_backlog: listen backlog set in listen()
 287  *	@sk_uid: user id of owner
 288  *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
 289  *	@sk_busy_poll_budget: napi processing budget when busypolling
 290  *	@sk_priority: %SO_PRIORITY setting
 291  *	@sk_type: socket type (%SOCK_STREAM, etc)
 292  *	@sk_protocol: which protocol this socket belongs in this network family
 293  *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
 294  *	@sk_peer_pid: &struct pid for this socket's peer
 295  *	@sk_peer_cred: %SO_PEERCRED setting
 296  *	@sk_rcvlowat: %SO_RCVLOWAT setting
 297  *	@sk_rcvtimeo: %SO_RCVTIMEO setting
 298  *	@sk_sndtimeo: %SO_SNDTIMEO setting
 299  *	@sk_txhash: computed flow hash for use on transmit
 300  *	@sk_txrehash: enable TX hash rethink
 301  *	@sk_filter: socket filtering instructions
 302  *	@sk_timer: sock cleanup timer
 303  *	@sk_stamp: time stamp of last packet received
 304  *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
 305  *	@sk_tsflags: SO_TIMESTAMPING flags
 306  *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
 307  *			   Sockets that can be used under memory reclaim should
 308  *			   set this to false.
 309  *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
 310  *	              for timestamping
 311  *	@sk_tskey: counter to disambiguate concurrent tstamp requests
 312  *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
 313  *	@sk_socket: Identd and reporting IO signals
 314  *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
 315  *	@sk_frag: cached page frag
 316  *	@sk_peek_off: current peek_offset value
 317  *	@sk_send_head: front of stuff to transmit
 318  *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
 319  *	@sk_security: used by security modules
 320  *	@sk_mark: generic packet mark
 321  *	@sk_cgrp_data: cgroup data for this cgroup
 322  *	@sk_memcg: this socket's memory cgroup association
 323  *	@sk_write_pending: a write to stream socket waits to start
 324  *	@sk_disconnects: number of disconnect operations performed on this sock
 325  *	@sk_state_change: callback to indicate change in the state of the sock
 326  *	@sk_data_ready: callback to indicate there is data to be processed
 327  *	@sk_write_space: callback to indicate there is bf sending space available
 328  *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
 329  *	@sk_backlog_rcv: callback to process the backlog
 330  *	@sk_validate_xmit_skb: ptr to an optional validate function
 331  *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
 332  *	@sk_reuseport_cb: reuseport group container
 333  *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
 334  *	@sk_rcu: used during RCU grace period
 335  *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
 336  *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
 337  *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
 338  *	@sk_txtime_unused: unused txtime flags
 339  *	@ns_tracker: tracker for netns reference
 340  *	@sk_user_frags: xarray of pages the user is holding a reference on.
 341  */
 342struct sock {
 343	/*
 344	 * Now struct inet_timewait_sock also uses sock_common, so please just
 345	 * don't add nothing before this first member (__sk_common) --acme
 346	 */
 347	struct sock_common	__sk_common;
 348#define sk_node			__sk_common.skc_node
 349#define sk_nulls_node		__sk_common.skc_nulls_node
 350#define sk_refcnt		__sk_common.skc_refcnt
 351#define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
 352#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
 353#define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
 354#endif
 355
 356#define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
 357#define sk_dontcopy_end		__sk_common.skc_dontcopy_end
 358#define sk_hash			__sk_common.skc_hash
 359#define sk_portpair		__sk_common.skc_portpair
 360#define sk_num			__sk_common.skc_num
 361#define sk_dport		__sk_common.skc_dport
 362#define sk_addrpair		__sk_common.skc_addrpair
 363#define sk_daddr		__sk_common.skc_daddr
 364#define sk_rcv_saddr		__sk_common.skc_rcv_saddr
 365#define sk_family		__sk_common.skc_family
 366#define sk_state		__sk_common.skc_state
 367#define sk_reuse		__sk_common.skc_reuse
 368#define sk_reuseport		__sk_common.skc_reuseport
 369#define sk_ipv6only		__sk_common.skc_ipv6only
 370#define sk_net_refcnt		__sk_common.skc_net_refcnt
 371#define sk_bound_dev_if		__sk_common.skc_bound_dev_if
 372#define sk_bind_node		__sk_common.skc_bind_node
 373#define sk_prot			__sk_common.skc_prot
 374#define sk_net			__sk_common.skc_net
 375#define sk_v6_daddr		__sk_common.skc_v6_daddr
 376#define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
 377#define sk_cookie		__sk_common.skc_cookie
 378#define sk_incoming_cpu		__sk_common.skc_incoming_cpu
 379#define sk_flags		__sk_common.skc_flags
 380#define sk_rxhash		__sk_common.skc_rxhash
 381
 382	__cacheline_group_begin(sock_write_rx);
 
 
 
 383
 
 384	atomic_t		sk_drops;
 385	__s32			sk_peek_off;
 386	struct sk_buff_head	sk_error_queue;
 387	struct sk_buff_head	sk_receive_queue;
 388	/*
 389	 * The backlog queue is special, it is always used with
 390	 * the per-socket spinlock held and requires low latency
 391	 * access. Therefore we special case it's implementation.
 392	 * Note : rmem_alloc is in this structure to fill a hole
 393	 * on 64bit arches, not because its logically part of
 394	 * backlog.
 395	 */
 396	struct {
 397		atomic_t	rmem_alloc;
 398		int		len;
 399		struct sk_buff	*head;
 400		struct sk_buff	*tail;
 401	} sk_backlog;
 402#define sk_rmem_alloc sk_backlog.rmem_alloc
 403
 404	__cacheline_group_end(sock_write_rx);
 405
 406	__cacheline_group_begin(sock_read_rx);
 407	/* early demux fields */
 408	struct dst_entry __rcu	*sk_rx_dst;
 409	int			sk_rx_dst_ifindex;
 410	u32			sk_rx_dst_cookie;
 411
 
 
 412#ifdef CONFIG_NET_RX_BUSY_POLL
 413	unsigned int		sk_ll_usec;
 
 414	unsigned int		sk_napi_id;
 415	u16			sk_busy_poll_budget;
 416	u8			sk_prefer_busy_poll;
 417#endif
 418	u8			sk_userlocks;
 419	int			sk_rcvbuf;
 
 420
 421	struct sk_filter __rcu	*sk_filter;
 422	union {
 423		struct socket_wq __rcu	*sk_wq;
 424		/* private: */
 425		struct socket_wq	*sk_wq_raw;
 426		/* public: */
 427	};
 428
 429	void			(*sk_data_ready)(struct sock *sk);
 430	long			sk_rcvtimeo;
 431	int			sk_rcvlowat;
 432	__cacheline_group_end(sock_read_rx);
 433
 434	__cacheline_group_begin(sock_read_rxtx);
 435	int			sk_err;
 436	struct socket		*sk_socket;
 437	struct mem_cgroup	*sk_memcg;
 438#ifdef CONFIG_XFRM
 439	struct xfrm_policy __rcu *sk_policy[2];
 440#endif
 441	__cacheline_group_end(sock_read_rxtx);
 442
 443	__cacheline_group_begin(sock_write_rxtx);
 444	socket_lock_t		sk_lock;
 445	u32			sk_reserved_mem;
 446	int			sk_forward_alloc;
 447	u32			sk_tsflags;
 448	__cacheline_group_end(sock_write_rxtx);
 449
 450	__cacheline_group_begin(sock_write_tx);
 451	int			sk_write_pending;
 452	atomic_t		sk_omem_alloc;
 453	int			sk_sndbuf;
 454
 
 455	int			sk_wmem_queued;
 456	refcount_t		sk_wmem_alloc;
 457	unsigned long		sk_tsq_flags;
 458	union {
 459		struct sk_buff	*sk_send_head;
 460		struct rb_root	tcp_rtx_queue;
 461	};
 462	struct sk_buff_head	sk_write_queue;
 463	u32			sk_dst_pending_confirm;
 
 
 464	u32			sk_pacing_status; /* see enum sk_pacing */
 465	struct page_frag	sk_frag;
 466	struct timer_list	sk_timer;
 467
 
 468	unsigned long		sk_pacing_rate; /* bytes per second */
 469	atomic_t		sk_zckey;
 470	atomic_t		sk_tskey;
 471	__cacheline_group_end(sock_write_tx);
 472
 473	__cacheline_group_begin(sock_read_tx);
 474	unsigned long		sk_max_pacing_rate;
 475	long			sk_sndtimeo;
 476	u32			sk_priority;
 477	u32			sk_mark;
 478	struct dst_entry __rcu	*sk_dst_cache;
 479	netdev_features_t	sk_route_caps;
 480#ifdef CONFIG_SOCK_VALIDATE_XMIT
 481	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
 482							struct net_device *dev,
 483							struct sk_buff *skb);
 484#endif
 485	u16			sk_gso_type;
 486	u16			sk_gso_max_segs;
 487	unsigned int		sk_gso_max_size;
 488	gfp_t			sk_allocation;
 489	u32			sk_txhash;
 490	u8			sk_pacing_shift;
 491	bool			sk_use_task_frag;
 492	__cacheline_group_end(sock_read_tx);
 493
 494	/*
 495	 * Because of non atomicity rules, all
 496	 * changes are protected by socket lock.
 497	 */
 498	u8			sk_gso_disabled : 1,
 499				sk_kern_sock : 1,
 500				sk_no_check_tx : 1,
 501				sk_no_check_rx : 1;
 502	u8			sk_shutdown;
 
 503	u16			sk_type;
 504	u16			sk_protocol;
 
 505	unsigned long	        sk_lingertime;
 506	struct proto		*sk_prot_creator;
 507	rwlock_t		sk_callback_lock;
 508	int			sk_err_soft;
 
 509	u32			sk_ack_backlog;
 510	u32			sk_max_ack_backlog;
 511	kuid_t			sk_uid;
 
 
 
 
 
 512	spinlock_t		sk_peer_lock;
 513	int			sk_bind_phc;
 514	struct pid		*sk_peer_pid;
 515	const struct cred	*sk_peer_cred;
 516
 
 517	ktime_t			sk_stamp;
 518#if BITS_PER_LONG==32
 519	seqlock_t		sk_stamp_seq;
 520#endif
 521	int			sk_disconnects;
 
 
 
 522
 523	u8			sk_txrehash;
 524	u8			sk_clockid;
 525	u8			sk_txtime_deadline_mode : 1,
 526				sk_txtime_report_errors : 1,
 527				sk_txtime_unused : 6;
 
 528
 
 529	void			*sk_user_data;
 530#ifdef CONFIG_SECURITY
 531	void			*sk_security;
 532#endif
 533	struct sock_cgroup_data	sk_cgrp_data;
 
 534	void			(*sk_state_change)(struct sock *sk);
 
 535	void			(*sk_write_space)(struct sock *sk);
 536	void			(*sk_error_report)(struct sock *sk);
 537	int			(*sk_backlog_rcv)(struct sock *sk,
 538						  struct sk_buff *skb);
 
 
 
 
 
 539	void                    (*sk_destruct)(struct sock *sk);
 540	struct sock_reuseport __rcu	*sk_reuseport_cb;
 541#ifdef CONFIG_BPF_SYSCALL
 542	struct bpf_local_storage __rcu	*sk_bpf_storage;
 543#endif
 544	struct rcu_head		sk_rcu;
 545	netns_tracker		ns_tracker;
 546	struct xarray		sk_user_frags;
 547};
 548
 549struct sock_bh_locked {
 550	struct sock *sock;
 551	local_lock_t bh_lock;
 552};
 553
 554enum sk_pacing {
 555	SK_PACING_NONE		= 0,
 556	SK_PACING_NEEDED	= 1,
 557	SK_PACING_FQ		= 2,
 558};
 559
 560/* flag bits in sk_user_data
 561 *
 562 * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
 563 *   not be suitable for copying when cloning the socket. For instance,
 564 *   it can point to a reference counted object. sk_user_data bottom
 565 *   bit is set if pointer must not be copied.
 566 *
 567 * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
 568 *   managed/owned by a BPF reuseport array. This bit should be set
 569 *   when sk_user_data's sk is added to the bpf's reuseport_array.
 570 *
 571 * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
 572 *   sk_user_data points to psock type. This bit should be set
 573 *   when sk_user_data is assigned to a psock object.
 574 */
 575#define SK_USER_DATA_NOCOPY	1UL
 576#define SK_USER_DATA_BPF	2UL
 577#define SK_USER_DATA_PSOCK	4UL
 578#define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
 579				  SK_USER_DATA_PSOCK)
 580
 581/**
 582 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
 583 * @sk: socket
 584 */
 585static inline bool sk_user_data_is_nocopy(const struct sock *sk)
 586{
 587	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
 588}
 589
 590#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
 591
 592/**
 593 * __locked_read_sk_user_data_with_flags - return the pointer
 594 * only if argument flags all has been set in sk_user_data. Otherwise
 595 * return NULL
 596 *
 597 * @sk: socket
 598 * @flags: flag bits
 599 *
 600 * The caller must be holding sk->sk_callback_lock.
 601 */
 602static inline void *
 603__locked_read_sk_user_data_with_flags(const struct sock *sk,
 604				      uintptr_t flags)
 605{
 606	uintptr_t sk_user_data =
 607		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
 608						 lockdep_is_held(&sk->sk_callback_lock));
 609
 610	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
 611
 612	if ((sk_user_data & flags) == flags)
 613		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
 614	return NULL;
 615}
 616
 617/**
 618 * __rcu_dereference_sk_user_data_with_flags - return the pointer
 619 * only if argument flags all has been set in sk_user_data. Otherwise
 620 * return NULL
 621 *
 622 * @sk: socket
 623 * @flags: flag bits
 624 */
 625static inline void *
 626__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
 627					  uintptr_t flags)
 628{
 629	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
 630
 631	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
 632
 633	if ((sk_user_data & flags) == flags)
 634		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
 635	return NULL;
 636}
 637
 638#define rcu_dereference_sk_user_data(sk)				\
 639	__rcu_dereference_sk_user_data_with_flags(sk, 0)
 640#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
 641({									\
 642	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
 643		  __tmp2 = (uintptr_t)(flags);				\
 644	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
 645	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
 646	rcu_assign_pointer(__sk_user_data((sk)),			\
 647			   __tmp1 | __tmp2);				\
 648})
 649#define rcu_assign_sk_user_data(sk, ptr)				\
 650	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
 651
 652static inline
 653struct net *sock_net(const struct sock *sk)
 654{
 655	return read_pnet(&sk->sk_net);
 656}
 657
 658static inline
 659void sock_net_set(struct sock *sk, struct net *net)
 660{
 661	write_pnet(&sk->sk_net, net);
 662}
 663
 664/*
 665 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
 666 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
 667 * on a socket means that the socket will reuse everybody else's port
 668 * without looking at the other's sk_reuse value.
 669 */
 670
 671#define SK_NO_REUSE	0
 672#define SK_CAN_REUSE	1
 673#define SK_FORCE_REUSE	2
 674
 675int sk_set_peek_off(struct sock *sk, int val);
 676
 677static inline int sk_peek_offset(const struct sock *sk, int flags)
 678{
 679	if (unlikely(flags & MSG_PEEK)) {
 680		return READ_ONCE(sk->sk_peek_off);
 681	}
 682
 683	return 0;
 684}
 685
 686static inline void sk_peek_offset_bwd(struct sock *sk, int val)
 687{
 688	s32 off = READ_ONCE(sk->sk_peek_off);
 689
 690	if (unlikely(off >= 0)) {
 691		off = max_t(s32, off - val, 0);
 692		WRITE_ONCE(sk->sk_peek_off, off);
 693	}
 694}
 695
 696static inline void sk_peek_offset_fwd(struct sock *sk, int val)
 697{
 698	sk_peek_offset_bwd(sk, -val);
 699}
 700
 701/*
 702 * Hashed lists helper routines
 703 */
 704static inline struct sock *sk_entry(const struct hlist_node *node)
 705{
 706	return hlist_entry(node, struct sock, sk_node);
 707}
 708
 709static inline struct sock *__sk_head(const struct hlist_head *head)
 710{
 711	return hlist_entry(head->first, struct sock, sk_node);
 712}
 713
 714static inline struct sock *sk_head(const struct hlist_head *head)
 715{
 716	return hlist_empty(head) ? NULL : __sk_head(head);
 717}
 718
 719static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
 720{
 721	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
 722}
 723
 724static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
 725{
 726	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
 727}
 728
 729static inline struct sock *sk_next(const struct sock *sk)
 730{
 731	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
 732}
 733
 734static inline struct sock *sk_nulls_next(const struct sock *sk)
 735{
 736	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
 737		hlist_nulls_entry(sk->sk_nulls_node.next,
 738				  struct sock, sk_nulls_node) :
 739		NULL;
 740}
 741
 742static inline bool sk_unhashed(const struct sock *sk)
 743{
 744	return hlist_unhashed(&sk->sk_node);
 745}
 746
 747static inline bool sk_hashed(const struct sock *sk)
 748{
 749	return !sk_unhashed(sk);
 750}
 751
 752static inline void sk_node_init(struct hlist_node *node)
 753{
 754	node->pprev = NULL;
 755}
 756
 757static inline void __sk_del_node(struct sock *sk)
 758{
 759	__hlist_del(&sk->sk_node);
 760}
 761
 762/* NB: equivalent to hlist_del_init_rcu */
 763static inline bool __sk_del_node_init(struct sock *sk)
 764{
 765	if (sk_hashed(sk)) {
 766		__sk_del_node(sk);
 767		sk_node_init(&sk->sk_node);
 768		return true;
 769	}
 770	return false;
 771}
 772
 773/* Grab socket reference count. This operation is valid only
 774   when sk is ALREADY grabbed f.e. it is found in hash table
 775   or a list and the lookup is made under lock preventing hash table
 776   modifications.
 777 */
 778
 779static __always_inline void sock_hold(struct sock *sk)
 780{
 781	refcount_inc(&sk->sk_refcnt);
 782}
 783
 784/* Ungrab socket in the context, which assumes that socket refcnt
 785   cannot hit zero, f.e. it is true in context of any socketcall.
 786 */
 787static __always_inline void __sock_put(struct sock *sk)
 788{
 789	refcount_dec(&sk->sk_refcnt);
 790}
 791
 792static inline bool sk_del_node_init(struct sock *sk)
 793{
 794	bool rc = __sk_del_node_init(sk);
 795
 796	if (rc) {
 797		/* paranoid for a while -acme */
 798		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 799		__sock_put(sk);
 800	}
 801	return rc;
 802}
 803#define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
 804
 805static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
 806{
 807	if (sk_hashed(sk)) {
 808		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
 809		return true;
 810	}
 811	return false;
 812}
 813
 814static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
 815{
 816	bool rc = __sk_nulls_del_node_init_rcu(sk);
 817
 818	if (rc) {
 819		/* paranoid for a while -acme */
 820		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 821		__sock_put(sk);
 822	}
 823	return rc;
 824}
 825
 826static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
 827{
 828	hlist_add_head(&sk->sk_node, list);
 829}
 830
 831static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
 832{
 833	sock_hold(sk);
 834	__sk_add_node(sk, list);
 835}
 836
 837static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
 838{
 839	sock_hold(sk);
 840	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 841	    sk->sk_family == AF_INET6)
 842		hlist_add_tail_rcu(&sk->sk_node, list);
 843	else
 844		hlist_add_head_rcu(&sk->sk_node, list);
 845}
 846
 847static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
 848{
 849	sock_hold(sk);
 850	hlist_add_tail_rcu(&sk->sk_node, list);
 851}
 852
 853static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 854{
 855	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
 856}
 857
 858static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
 859{
 860	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
 861}
 862
 863static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 864{
 865	sock_hold(sk);
 866	__sk_nulls_add_node_rcu(sk, list);
 867}
 868
 869static inline void __sk_del_bind_node(struct sock *sk)
 870{
 871	__hlist_del(&sk->sk_bind_node);
 872}
 873
 874static inline void sk_add_bind_node(struct sock *sk,
 875					struct hlist_head *list)
 876{
 877	hlist_add_head(&sk->sk_bind_node, list);
 878}
 879
 880#define sk_for_each(__sk, list) \
 881	hlist_for_each_entry(__sk, list, sk_node)
 882#define sk_for_each_rcu(__sk, list) \
 883	hlist_for_each_entry_rcu(__sk, list, sk_node)
 884#define sk_nulls_for_each(__sk, node, list) \
 885	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
 886#define sk_nulls_for_each_rcu(__sk, node, list) \
 887	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
 888#define sk_for_each_from(__sk) \
 889	hlist_for_each_entry_from(__sk, sk_node)
 890#define sk_nulls_for_each_from(__sk, node) \
 891	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
 892		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
 893#define sk_for_each_safe(__sk, tmp, list) \
 894	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
 895#define sk_for_each_bound(__sk, list) \
 896	hlist_for_each_entry(__sk, list, sk_bind_node)
 897#define sk_for_each_bound_safe(__sk, tmp, list) \
 898	hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node)
 899
 900/**
 901 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
 902 * @tpos:	the type * to use as a loop cursor.
 903 * @pos:	the &struct hlist_node to use as a loop cursor.
 904 * @head:	the head for your list.
 905 * @offset:	offset of hlist_node within the struct.
 906 *
 907 */
 908#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
 909	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
 910	     pos != NULL &&						       \
 911		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
 912	     pos = rcu_dereference(hlist_next_rcu(pos)))
 913
 914static inline struct user_namespace *sk_user_ns(const struct sock *sk)
 915{
 916	/* Careful only use this in a context where these parameters
 917	 * can not change and must all be valid, such as recvmsg from
 918	 * userspace.
 919	 */
 920	return sk->sk_socket->file->f_cred->user_ns;
 921}
 922
 923/* Sock flags */
 924enum sock_flags {
 925	SOCK_DEAD,
 926	SOCK_DONE,
 927	SOCK_URGINLINE,
 928	SOCK_KEEPOPEN,
 929	SOCK_LINGER,
 930	SOCK_DESTROY,
 931	SOCK_BROADCAST,
 932	SOCK_TIMESTAMP,
 933	SOCK_ZAPPED,
 934	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
 935	SOCK_DBG, /* %SO_DEBUG setting */
 936	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
 937	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
 938	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
 939	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
 940	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
 941	SOCK_FASYNC, /* fasync() active */
 942	SOCK_RXQ_OVFL,
 943	SOCK_ZEROCOPY, /* buffers from userspace */
 944	SOCK_WIFI_STATUS, /* push wifi status to userspace */
 945	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
 946		     * Will use last 4 bytes of packet sent from
 947		     * user-space instead.
 948		     */
 949	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
 950	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
 951	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
 952	SOCK_TXTIME,
 953	SOCK_XDP, /* XDP is attached */
 954	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
 955	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
 956};
 957
 958#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
 959/*
 960 * The highest bit of sk_tsflags is reserved for kernel-internal
 961 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that
 962 * SOF_TIMESTAMPING* values do not reach this reserved area
 963 */
 964#define SOCKCM_FLAG_TS_OPT_ID	BIT(31)
 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
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1135
1136static inline void sock_rps_save_rxhash(struct sock *sk,
1137					const struct sk_buff *skb)
1138{
1139#ifdef CONFIG_RPS
1140	/* The following WRITE_ONCE() is paired with the READ_ONCE()
1141	 * here, and another one in sock_rps_record_flow().
1142	 */
1143	if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1144		WRITE_ONCE(sk->sk_rxhash, skb->hash);
1145#endif
1146}
1147
1148static inline void sock_rps_reset_rxhash(struct sock *sk)
1149{
1150#ifdef CONFIG_RPS
1151	/* Paired with READ_ONCE() in sock_rps_record_flow() */
1152	WRITE_ONCE(sk->sk_rxhash, 0);
1153#endif
1154}
1155
1156#define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1157	({	int __rc, __dis = __sk->sk_disconnects;			\
1158		release_sock(__sk);					\
1159		__rc = __condition;					\
1160		if (!__rc) {						\
1161			*(__timeo) = wait_woken(__wait,			\
1162						TASK_INTERRUPTIBLE,	\
1163						*(__timeo));		\
1164		}							\
1165		sched_annotate_sleep();					\
1166		lock_sock(__sk);					\
1167		__rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1168		__rc;							\
1169	})
1170
1171int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1172int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1173void sk_stream_wait_close(struct sock *sk, long timeo_p);
1174int sk_stream_error(struct sock *sk, int flags, int err);
1175void sk_stream_kill_queues(struct sock *sk);
1176void sk_set_memalloc(struct sock *sk);
1177void sk_clear_memalloc(struct sock *sk);
1178
1179void __sk_flush_backlog(struct sock *sk);
1180
1181static inline bool sk_flush_backlog(struct sock *sk)
1182{
1183	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1184		__sk_flush_backlog(sk);
1185		return true;
1186	}
1187	return false;
1188}
1189
1190int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1191
1192struct request_sock_ops;
1193struct timewait_sock_ops;
1194struct inet_hashinfo;
1195struct raw_hashinfo;
1196struct smc_hashinfo;
1197struct module;
1198struct sk_psock;
1199
1200/*
1201 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1202 * un-modified. Special care is taken when initializing object to zero.
1203 */
1204static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1205{
1206	if (offsetof(struct sock, sk_node.next) != 0)
1207		memset(sk, 0, offsetof(struct sock, sk_node.next));
1208	memset(&sk->sk_node.pprev, 0,
1209	       size - offsetof(struct sock, sk_node.pprev));
1210}
1211
1212struct proto_accept_arg {
1213	int flags;
1214	int err;
1215	int is_empty;
1216	bool kern;
1217};
1218
1219/* Networking protocol blocks we attach to sockets.
1220 * socket layer -> transport layer interface
1221 */
1222struct proto {
1223	void			(*close)(struct sock *sk,
1224					long timeout);
1225	int			(*pre_connect)(struct sock *sk,
1226					struct sockaddr *uaddr,
1227					int addr_len);
1228	int			(*connect)(struct sock *sk,
1229					struct sockaddr *uaddr,
1230					int addr_len);
1231	int			(*disconnect)(struct sock *sk, int flags);
1232
1233	struct sock *		(*accept)(struct sock *sk,
1234					  struct proto_accept_arg *arg);
1235
1236	int			(*ioctl)(struct sock *sk, int cmd,
1237					 int *karg);
1238	int			(*init)(struct sock *sk);
1239	void			(*destroy)(struct sock *sk);
1240	void			(*shutdown)(struct sock *sk, int how);
1241	int			(*setsockopt)(struct sock *sk, int level,
1242					int optname, sockptr_t optval,
1243					unsigned int optlen);
1244	int			(*getsockopt)(struct sock *sk, int level,
1245					int optname, char __user *optval,
1246					int __user *option);
1247	void			(*keepalive)(struct sock *sk, int valbool);
1248#ifdef CONFIG_COMPAT
1249	int			(*compat_ioctl)(struct sock *sk,
1250					unsigned int cmd, unsigned long arg);
1251#endif
1252	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1253					   size_t len);
1254	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1255					   size_t len, int flags, int *addr_len);
1256	void			(*splice_eof)(struct socket *sock);
1257	int			(*bind)(struct sock *sk,
1258					struct sockaddr *addr, int addr_len);
1259	int			(*bind_add)(struct sock *sk,
1260					struct sockaddr *addr, int addr_len);
1261
1262	int			(*backlog_rcv) (struct sock *sk,
1263						struct sk_buff *skb);
1264	bool			(*bpf_bypass_getsockopt)(int level,
1265							 int optname);
1266
1267	void		(*release_cb)(struct sock *sk);
1268
1269	/* Keeping track of sk's, looking them up, and port selection methods. */
1270	int			(*hash)(struct sock *sk);
1271	void			(*unhash)(struct sock *sk);
1272	void			(*rehash)(struct sock *sk);
1273	int			(*get_port)(struct sock *sk, unsigned short snum);
1274	void			(*put_port)(struct sock *sk);
1275#ifdef CONFIG_BPF_SYSCALL
1276	int			(*psock_update_sk_prot)(struct sock *sk,
1277							struct sk_psock *psock,
1278							bool restore);
1279#endif
1280
1281	/* Keeping track of sockets in use */
1282#ifdef CONFIG_PROC_FS
1283	unsigned int		inuse_idx;
1284#endif
1285
1286#if IS_ENABLED(CONFIG_MPTCP)
1287	int			(*forward_alloc_get)(const struct sock *sk);
1288#endif
1289
1290	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1291	bool			(*sock_is_readable)(struct sock *sk);
1292	/* Memory pressure */
1293	void			(*enter_memory_pressure)(struct sock *sk);
1294	void			(*leave_memory_pressure)(struct sock *sk);
1295	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1296	int  __percpu		*per_cpu_fw_alloc;
1297	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1298
1299	/*
1300	 * Pressure flag: try to collapse.
1301	 * Technical note: it is used by multiple contexts non atomically.
1302	 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1303	 * All the __sk_mem_schedule() is of this nature: accounting
1304	 * is strict, actions are advisory and have some latency.
1305	 */
1306	unsigned long		*memory_pressure;
1307	long			*sysctl_mem;
1308
1309	int			*sysctl_wmem;
1310	int			*sysctl_rmem;
1311	u32			sysctl_wmem_offset;
1312	u32			sysctl_rmem_offset;
1313
1314	int			max_header;
1315	bool			no_autobind;
1316
1317	struct kmem_cache	*slab;
1318	unsigned int		obj_size;
1319	unsigned int		ipv6_pinfo_offset;
1320	slab_flags_t		slab_flags;
1321	unsigned int		useroffset;	/* Usercopy region offset */
1322	unsigned int		usersize;	/* Usercopy region size */
1323
1324	unsigned int __percpu	*orphan_count;
1325
1326	struct request_sock_ops	*rsk_prot;
1327	struct timewait_sock_ops *twsk_prot;
1328
1329	union {
1330		struct inet_hashinfo	*hashinfo;
1331		struct udp_table	*udp_table;
1332		struct raw_hashinfo	*raw_hash;
1333		struct smc_hashinfo	*smc_hash;
1334	} h;
1335
1336	struct module		*owner;
1337
1338	char			name[32];
1339
1340	struct list_head	node;
1341	int			(*diag_destroy)(struct sock *sk, int err);
1342} __randomize_layout;
1343
1344int proto_register(struct proto *prot, int alloc_slab);
1345void proto_unregister(struct proto *prot);
1346int sock_load_diag_module(int family, int protocol);
1347
1348INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1349
1350static inline int sk_forward_alloc_get(const struct sock *sk)
1351{
1352#if IS_ENABLED(CONFIG_MPTCP)
1353	if (sk->sk_prot->forward_alloc_get)
1354		return sk->sk_prot->forward_alloc_get(sk);
1355#endif
1356	return READ_ONCE(sk->sk_forward_alloc);
1357}
1358
1359static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1360{
1361	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1362		return false;
1363
1364	return sk->sk_prot->stream_memory_free ?
1365		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1366				     tcp_stream_memory_free, sk, wake) : true;
1367}
1368
1369static inline bool sk_stream_memory_free(const struct sock *sk)
1370{
1371	return __sk_stream_memory_free(sk, 0);
1372}
1373
1374static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1375{
1376	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1377	       __sk_stream_memory_free(sk, wake);
1378}
1379
1380static inline bool sk_stream_is_writeable(const struct sock *sk)
1381{
1382	return __sk_stream_is_writeable(sk, 0);
1383}
1384
1385static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1386					    struct cgroup *ancestor)
1387{
1388#ifdef CONFIG_SOCK_CGROUP_DATA
1389	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1390				    ancestor);
1391#else
1392	return -ENOTSUPP;
1393#endif
1394}
1395
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1396#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1397
1398static inline void sk_sockets_allocated_dec(struct sock *sk)
1399{
1400	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1401				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1402}
1403
1404static inline void sk_sockets_allocated_inc(struct sock *sk)
1405{
1406	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1407				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1408}
1409
1410static inline u64
1411sk_sockets_allocated_read_positive(struct sock *sk)
1412{
1413	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1414}
1415
1416static inline int
1417proto_sockets_allocated_sum_positive(struct proto *prot)
1418{
1419	return percpu_counter_sum_positive(prot->sockets_allocated);
1420}
1421
 
 
 
 
 
 
 
 
 
1422#ifdef CONFIG_PROC_FS
1423#define PROTO_INUSE_NR	64	/* should be enough for the first time */
1424struct prot_inuse {
1425	int all;
1426	int val[PROTO_INUSE_NR];
1427};
1428
1429static inline void sock_prot_inuse_add(const struct net *net,
1430				       const struct proto *prot, int val)
1431{
1432	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1433}
1434
1435static inline void sock_inuse_add(const struct net *net, int val)
1436{
1437	this_cpu_add(net->core.prot_inuse->all, val);
1438}
1439
1440int sock_prot_inuse_get(struct net *net, struct proto *proto);
1441int sock_inuse_get(struct net *net);
1442#else
1443static inline void sock_prot_inuse_add(const struct net *net,
1444				       const struct proto *prot, int val)
1445{
1446}
1447
1448static inline void sock_inuse_add(const struct net *net, int val)
1449{
1450}
1451#endif
1452
1453
1454/* With per-bucket locks this operation is not-atomic, so that
1455 * this version is not worse.
1456 */
1457static inline int __sk_prot_rehash(struct sock *sk)
1458{
1459	sk->sk_prot->unhash(sk);
1460	return sk->sk_prot->hash(sk);
1461}
1462
1463/* About 10 seconds */
1464#define SOCK_DESTROY_TIME (10*HZ)
1465
1466/* Sockets 0-1023 can't be bound to unless you are superuser */
1467#define PROT_SOCK	1024
1468
1469#define SHUTDOWN_MASK	3
1470#define RCV_SHUTDOWN	1
1471#define SEND_SHUTDOWN	2
1472
1473#define SOCK_BINDADDR_LOCK	4
1474#define SOCK_BINDPORT_LOCK	8
1475
1476struct socket_alloc {
1477	struct socket socket;
1478	struct inode vfs_inode;
1479};
1480
1481static inline struct socket *SOCKET_I(struct inode *inode)
1482{
1483	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1484}
1485
1486static inline struct inode *SOCK_INODE(struct socket *socket)
1487{
1488	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1489}
1490
1491/*
1492 * Functions for memory accounting
1493 */
1494int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1495int __sk_mem_schedule(struct sock *sk, int size, int kind);
1496void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1497void __sk_mem_reclaim(struct sock *sk, int amount);
1498
1499#define SK_MEM_SEND	0
1500#define SK_MEM_RECV	1
1501
1502/* sysctl_mem values are in pages */
1503static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1504{
1505	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1506}
1507
1508static inline int sk_mem_pages(int amt)
1509{
1510	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1511}
1512
1513static inline bool sk_has_account(struct sock *sk)
1514{
1515	/* return true if protocol supports memory accounting */
1516	return !!sk->sk_prot->memory_allocated;
1517}
1518
1519static inline bool sk_wmem_schedule(struct sock *sk, int size)
1520{
1521	int delta;
1522
1523	if (!sk_has_account(sk))
1524		return true;
1525	delta = size - sk->sk_forward_alloc;
1526	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1527}
1528
1529static inline bool
1530__sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc)
1531{
1532	int delta;
1533
1534	if (!sk_has_account(sk))
1535		return true;
1536	delta = size - sk->sk_forward_alloc;
1537	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1538	       pfmemalloc;
1539}
1540
1541static inline bool
1542sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1543{
1544	return __sk_rmem_schedule(sk, size, skb_pfmemalloc(skb));
1545}
1546
1547static inline int sk_unused_reserved_mem(const struct sock *sk)
1548{
1549	int unused_mem;
1550
1551	if (likely(!sk->sk_reserved_mem))
1552		return 0;
1553
1554	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1555			atomic_read(&sk->sk_rmem_alloc);
1556
1557	return unused_mem > 0 ? unused_mem : 0;
1558}
1559
1560static inline void sk_mem_reclaim(struct sock *sk)
1561{
1562	int reclaimable;
1563
1564	if (!sk_has_account(sk))
1565		return;
1566
1567	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1568
1569	if (reclaimable >= (int)PAGE_SIZE)
1570		__sk_mem_reclaim(sk, reclaimable);
1571}
1572
1573static inline void sk_mem_reclaim_final(struct sock *sk)
1574{
1575	sk->sk_reserved_mem = 0;
1576	sk_mem_reclaim(sk);
1577}
1578
1579static inline void sk_mem_charge(struct sock *sk, int size)
1580{
1581	if (!sk_has_account(sk))
1582		return;
1583	sk_forward_alloc_add(sk, -size);
1584}
1585
1586static inline void sk_mem_uncharge(struct sock *sk, int size)
1587{
1588	if (!sk_has_account(sk))
1589		return;
1590	sk_forward_alloc_add(sk, size);
1591	sk_mem_reclaim(sk);
1592}
1593
1594/*
1595 * Macro so as to not evaluate some arguments when
1596 * lockdep is not enabled.
1597 *
1598 * Mark both the sk_lock and the sk_lock.slock as a
1599 * per-address-family lock class.
1600 */
1601#define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1602do {									\
1603	sk->sk_lock.owned = 0;						\
1604	init_waitqueue_head(&sk->sk_lock.wq);				\
1605	spin_lock_init(&(sk)->sk_lock.slock);				\
1606	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1607			sizeof((sk)->sk_lock));				\
1608	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1609				(skey), (sname));				\
1610	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1611} while (0)
1612
1613static inline bool lockdep_sock_is_held(const struct sock *sk)
1614{
1615	return lockdep_is_held(&sk->sk_lock) ||
1616	       lockdep_is_held(&sk->sk_lock.slock);
1617}
1618
1619void lock_sock_nested(struct sock *sk, int subclass);
1620
1621static inline void lock_sock(struct sock *sk)
1622{
1623	lock_sock_nested(sk, 0);
1624}
1625
1626void __lock_sock(struct sock *sk);
1627void __release_sock(struct sock *sk);
1628void release_sock(struct sock *sk);
1629
1630/* BH context may only use the following locking interface. */
1631#define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1632#define bh_lock_sock_nested(__sk) \
1633				spin_lock_nested(&((__sk)->sk_lock.slock), \
1634				SINGLE_DEPTH_NESTING)
1635#define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1636
1637bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1638
1639/**
1640 * lock_sock_fast - fast version of lock_sock
1641 * @sk: socket
1642 *
1643 * This version should be used for very small section, where process won't block
1644 * return false if fast path is taken:
1645 *
1646 *   sk_lock.slock locked, owned = 0, BH disabled
1647 *
1648 * return true if slow path is taken:
1649 *
1650 *   sk_lock.slock unlocked, owned = 1, BH enabled
1651 */
1652static inline bool lock_sock_fast(struct sock *sk)
1653{
1654	/* The sk_lock has mutex_lock() semantics here. */
1655	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1656
1657	return __lock_sock_fast(sk);
1658}
1659
1660/* fast socket lock variant for caller already holding a [different] socket lock */
1661static inline bool lock_sock_fast_nested(struct sock *sk)
1662{
1663	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1664
1665	return __lock_sock_fast(sk);
1666}
1667
1668/**
1669 * unlock_sock_fast - complement of lock_sock_fast
1670 * @sk: socket
1671 * @slow: slow mode
1672 *
1673 * fast unlock socket for user context.
1674 * If slow mode is on, we call regular release_sock()
1675 */
1676static inline void unlock_sock_fast(struct sock *sk, bool slow)
1677	__releases(&sk->sk_lock.slock)
1678{
1679	if (slow) {
1680		release_sock(sk);
1681		__release(&sk->sk_lock.slock);
1682	} else {
1683		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1684		spin_unlock_bh(&sk->sk_lock.slock);
1685	}
1686}
1687
1688void sockopt_lock_sock(struct sock *sk);
1689void sockopt_release_sock(struct sock *sk);
1690bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1691bool sockopt_capable(int cap);
1692
1693/* Used by processes to "lock" a socket state, so that
1694 * interrupts and bottom half handlers won't change it
1695 * from under us. It essentially blocks any incoming
1696 * packets, so that we won't get any new data or any
1697 * packets that change the state of the socket.
1698 *
1699 * While locked, BH processing will add new packets to
1700 * the backlog queue.  This queue is processed by the
1701 * owner of the socket lock right before it is released.
1702 *
1703 * Since ~2.3.5 it is also exclusive sleep lock serializing
1704 * accesses from user process context.
1705 */
1706
1707static inline void sock_owned_by_me(const struct sock *sk)
1708{
1709#ifdef CONFIG_LOCKDEP
1710	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1711#endif
1712}
1713
1714static inline void sock_not_owned_by_me(const struct sock *sk)
1715{
1716#ifdef CONFIG_LOCKDEP
1717	WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1718#endif
1719}
1720
1721static inline bool sock_owned_by_user(const struct sock *sk)
1722{
1723	sock_owned_by_me(sk);
1724	return sk->sk_lock.owned;
1725}
1726
1727static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1728{
1729	return sk->sk_lock.owned;
1730}
1731
1732static inline void sock_release_ownership(struct sock *sk)
1733{
1734	DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1735	sk->sk_lock.owned = 0;
1736
1737	/* The sk_lock has mutex_unlock() semantics: */
1738	mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1739}
1740
1741/* no reclassification while locks are held */
1742static inline bool sock_allow_reclassification(const struct sock *csk)
1743{
1744	struct sock *sk = (struct sock *)csk;
1745
1746	return !sock_owned_by_user_nocheck(sk) &&
1747		!spin_is_locked(&sk->sk_lock.slock);
1748}
1749
1750struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1751		      struct proto *prot, int kern);
1752void sk_free(struct sock *sk);
1753void sk_net_refcnt_upgrade(struct sock *sk);
1754void sk_destruct(struct sock *sk);
1755struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1756void sk_free_unlock_clone(struct sock *sk);
1757
1758struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1759			     gfp_t priority);
1760void __sock_wfree(struct sk_buff *skb);
1761void sock_wfree(struct sk_buff *skb);
1762struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1763			     gfp_t priority);
1764void skb_orphan_partial(struct sk_buff *skb);
1765void sock_rfree(struct sk_buff *skb);
1766void sock_efree(struct sk_buff *skb);
1767#ifdef CONFIG_INET
1768void sock_edemux(struct sk_buff *skb);
1769void sock_pfree(struct sk_buff *skb);
1770
1771static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk)
1772{
1773	skb_orphan(skb);
1774	if (refcount_inc_not_zero(&sk->sk_refcnt)) {
1775		skb->sk = sk;
1776		skb->destructor = sock_edemux;
1777	}
1778}
1779#else
1780#define sock_edemux sock_efree
1781#endif
1782
1783int sk_setsockopt(struct sock *sk, int level, int optname,
1784		  sockptr_t optval, unsigned int optlen);
1785int sock_setsockopt(struct socket *sock, int level, int op,
1786		    sockptr_t optval, unsigned int optlen);
1787int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1788		       int optname, sockptr_t optval, int optlen);
1789int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1790		       int optname, sockptr_t optval, sockptr_t optlen);
1791
1792int sk_getsockopt(struct sock *sk, int level, int optname,
1793		  sockptr_t optval, sockptr_t optlen);
1794int sock_gettstamp(struct socket *sock, void __user *userstamp,
1795		   bool timeval, bool time32);
1796struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1797				     unsigned long data_len, int noblock,
1798				     int *errcode, int max_page_order);
1799
1800static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1801						  unsigned long size,
1802						  int noblock, int *errcode)
1803{
1804	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1805}
1806
1807void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1808void sock_kfree_s(struct sock *sk, void *mem, int size);
1809void sock_kzfree_s(struct sock *sk, void *mem, int size);
1810void sk_send_sigurg(struct sock *sk);
1811
1812static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1813{
1814	if (sk->sk_socket)
1815		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1816	WRITE_ONCE(sk->sk_prot, proto);
1817}
1818
1819struct sockcm_cookie {
1820	u64 transmit_time;
1821	u32 mark;
1822	u32 tsflags;
1823	u32 ts_opt_id;
1824};
1825
1826static inline void sockcm_init(struct sockcm_cookie *sockc,
1827			       const struct sock *sk)
1828{
1829	*sockc = (struct sockcm_cookie) {
1830		.tsflags = READ_ONCE(sk->sk_tsflags)
1831	};
1832}
1833
1834int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1835		     struct sockcm_cookie *sockc);
1836int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1837		   struct sockcm_cookie *sockc);
1838
1839/*
1840 * Functions to fill in entries in struct proto_ops when a protocol
1841 * does not implement a particular function.
1842 */
1843int sock_no_bind(struct socket *, struct sockaddr *, int);
1844int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1845int sock_no_socketpair(struct socket *, struct socket *);
1846int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *);
1847int sock_no_getname(struct socket *, struct sockaddr *, int);
1848int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1849int sock_no_listen(struct socket *, int);
1850int sock_no_shutdown(struct socket *, int);
1851int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1852int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1853int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1854int sock_no_mmap(struct file *file, struct socket *sock,
1855		 struct vm_area_struct *vma);
1856
1857/*
1858 * Functions to fill in entries in struct proto_ops when a protocol
1859 * uses the inet style.
1860 */
1861int sock_common_getsockopt(struct socket *sock, int level, int optname,
1862				  char __user *optval, int __user *optlen);
1863int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1864			int flags);
1865int sock_common_setsockopt(struct socket *sock, int level, int optname,
1866			   sockptr_t optval, unsigned int optlen);
1867
1868void sk_common_release(struct sock *sk);
1869
1870/*
1871 *	Default socket callbacks and setup code
1872 */
1873
1874/* Initialise core socket variables using an explicit uid. */
1875void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1876
1877/* Initialise core socket variables.
1878 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1879 */
1880void sock_init_data(struct socket *sock, struct sock *sk);
1881
1882/*
1883 * Socket reference counting postulates.
1884 *
1885 * * Each user of socket SHOULD hold a reference count.
1886 * * Each access point to socket (an hash table bucket, reference from a list,
1887 *   running timer, skb in flight MUST hold a reference count.
1888 * * When reference count hits 0, it means it will never increase back.
1889 * * When reference count hits 0, it means that no references from
1890 *   outside exist to this socket and current process on current CPU
1891 *   is last user and may/should destroy this socket.
1892 * * sk_free is called from any context: process, BH, IRQ. When
1893 *   it is called, socket has no references from outside -> sk_free
1894 *   may release descendant resources allocated by the socket, but
1895 *   to the time when it is called, socket is NOT referenced by any
1896 *   hash tables, lists etc.
1897 * * Packets, delivered from outside (from network or from another process)
1898 *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1899 *   when they sit in queue. Otherwise, packets will leak to hole, when
1900 *   socket is looked up by one cpu and unhasing is made by another CPU.
1901 *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1902 *   (leak to backlog). Packet socket does all the processing inside
1903 *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1904 *   use separate SMP lock, so that they are prone too.
1905 */
1906
1907/* Ungrab socket and destroy it, if it was the last reference. */
1908static inline void sock_put(struct sock *sk)
1909{
1910	if (refcount_dec_and_test(&sk->sk_refcnt))
1911		sk_free(sk);
1912}
1913/* Generic version of sock_put(), dealing with all sockets
1914 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1915 */
1916void sock_gen_put(struct sock *sk);
1917
1918int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1919		     unsigned int trim_cap, bool refcounted);
1920static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1921				 const int nested)
1922{
1923	return __sk_receive_skb(sk, skb, nested, 1, true);
1924}
1925
1926static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1927{
1928	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1929	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1930		return;
1931	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1932	 * other WRITE_ONCE() because socket lock might be not held.
1933	 */
1934	WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1935}
1936
1937#define NO_QUEUE_MAPPING	USHRT_MAX
1938
1939static inline void sk_tx_queue_clear(struct sock *sk)
1940{
1941	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1942	 * other WRITE_ONCE() because socket lock might be not held.
1943	 */
1944	WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1945}
1946
1947static inline int sk_tx_queue_get(const struct sock *sk)
1948{
1949	if (sk) {
1950		/* Paired with WRITE_ONCE() in sk_tx_queue_clear()
1951		 * and sk_tx_queue_set().
1952		 */
1953		int val = READ_ONCE(sk->sk_tx_queue_mapping);
1954
1955		if (val != NO_QUEUE_MAPPING)
1956			return val;
1957	}
1958	return -1;
1959}
1960
1961static inline void __sk_rx_queue_set(struct sock *sk,
1962				     const struct sk_buff *skb,
1963				     bool force_set)
1964{
1965#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1966	if (skb_rx_queue_recorded(skb)) {
1967		u16 rx_queue = skb_get_rx_queue(skb);
1968
1969		if (force_set ||
1970		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
1971			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
1972	}
1973#endif
1974}
1975
1976static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1977{
1978	__sk_rx_queue_set(sk, skb, true);
1979}
1980
1981static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
1982{
1983	__sk_rx_queue_set(sk, skb, false);
1984}
1985
1986static inline void sk_rx_queue_clear(struct sock *sk)
1987{
1988#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1989	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
1990#endif
1991}
1992
1993static inline int sk_rx_queue_get(const struct sock *sk)
1994{
1995#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1996	if (sk) {
1997		int res = READ_ONCE(sk->sk_rx_queue_mapping);
1998
1999		if (res != NO_QUEUE_MAPPING)
2000			return res;
2001	}
2002#endif
2003
2004	return -1;
2005}
2006
2007static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2008{
2009	sk->sk_socket = sock;
2010}
2011
2012static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2013{
2014	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2015	return &rcu_dereference_raw(sk->sk_wq)->wait;
2016}
2017/* Detach socket from process context.
2018 * Announce socket dead, detach it from wait queue and inode.
2019 * Note that parent inode held reference count on this struct sock,
2020 * we do not release it in this function, because protocol
2021 * probably wants some additional cleanups or even continuing
2022 * to work with this socket (TCP).
2023 */
2024static inline void sock_orphan(struct sock *sk)
2025{
2026	write_lock_bh(&sk->sk_callback_lock);
2027	sock_set_flag(sk, SOCK_DEAD);
2028	sk_set_socket(sk, NULL);
2029	sk->sk_wq  = NULL;
2030	write_unlock_bh(&sk->sk_callback_lock);
2031}
2032
2033static inline void sock_graft(struct sock *sk, struct socket *parent)
2034{
2035	WARN_ON(parent->sk);
2036	write_lock_bh(&sk->sk_callback_lock);
2037	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2038	parent->sk = sk;
2039	sk_set_socket(sk, parent);
2040	sk->sk_uid = SOCK_INODE(parent)->i_uid;
2041	security_sock_graft(sk, parent);
2042	write_unlock_bh(&sk->sk_callback_lock);
2043}
2044
2045kuid_t sock_i_uid(struct sock *sk);
2046unsigned long __sock_i_ino(struct sock *sk);
2047unsigned long sock_i_ino(struct sock *sk);
2048
2049static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2050{
2051	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2052}
2053
2054static inline u32 net_tx_rndhash(void)
2055{
2056	u32 v = get_random_u32();
2057
2058	return v ?: 1;
2059}
2060
2061static inline void sk_set_txhash(struct sock *sk)
2062{
2063	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2064	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2065}
2066
2067static inline bool sk_rethink_txhash(struct sock *sk)
2068{
2069	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2070		sk_set_txhash(sk);
2071		return true;
2072	}
2073	return false;
2074}
2075
2076static inline struct dst_entry *
2077__sk_dst_get(const struct sock *sk)
2078{
2079	return rcu_dereference_check(sk->sk_dst_cache,
2080				     lockdep_sock_is_held(sk));
2081}
2082
2083static inline struct dst_entry *
2084sk_dst_get(const struct sock *sk)
2085{
2086	struct dst_entry *dst;
2087
2088	rcu_read_lock();
2089	dst = rcu_dereference(sk->sk_dst_cache);
2090	if (dst && !rcuref_get(&dst->__rcuref))
2091		dst = NULL;
2092	rcu_read_unlock();
2093	return dst;
2094}
2095
2096static inline void __dst_negative_advice(struct sock *sk)
2097{
2098	struct dst_entry *dst = __sk_dst_get(sk);
2099
2100	if (dst && dst->ops->negative_advice)
2101		dst->ops->negative_advice(sk, dst);
 
 
 
 
 
 
 
2102}
2103
2104static inline void dst_negative_advice(struct sock *sk)
2105{
2106	sk_rethink_txhash(sk);
2107	__dst_negative_advice(sk);
2108}
2109
2110static inline void
2111__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2112{
2113	struct dst_entry *old_dst;
2114
2115	sk_tx_queue_clear(sk);
2116	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2117	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2118					    lockdep_sock_is_held(sk));
2119	rcu_assign_pointer(sk->sk_dst_cache, dst);
2120	dst_release(old_dst);
2121}
2122
2123static inline void
2124sk_dst_set(struct sock *sk, struct dst_entry *dst)
2125{
2126	struct dst_entry *old_dst;
2127
2128	sk_tx_queue_clear(sk);
2129	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2130	old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst)));
2131	dst_release(old_dst);
2132}
2133
2134static inline void
2135__sk_dst_reset(struct sock *sk)
2136{
2137	__sk_dst_set(sk, NULL);
2138}
2139
2140static inline void
2141sk_dst_reset(struct sock *sk)
2142{
2143	sk_dst_set(sk, NULL);
2144}
2145
2146struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2147
2148struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2149
2150static inline void sk_dst_confirm(struct sock *sk)
2151{
2152	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2153		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2154}
2155
2156static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2157{
2158	if (skb_get_dst_pending_confirm(skb)) {
2159		struct sock *sk = skb->sk;
2160
2161		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2162			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2163		neigh_confirm(n);
2164	}
2165}
2166
2167bool sk_mc_loop(const struct sock *sk);
2168
2169static inline bool sk_can_gso(const struct sock *sk)
2170{
2171	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2172}
2173
2174void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2175
2176static inline void sk_gso_disable(struct sock *sk)
2177{
2178	sk->sk_gso_disabled = 1;
2179	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2180}
2181
2182static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2183					   struct iov_iter *from, char *to,
2184					   int copy, int offset)
2185{
2186	if (skb->ip_summed == CHECKSUM_NONE) {
2187		__wsum csum = 0;
2188		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2189			return -EFAULT;
2190		skb->csum = csum_block_add(skb->csum, csum, offset);
2191	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2192		if (!copy_from_iter_full_nocache(to, copy, from))
2193			return -EFAULT;
2194	} else if (!copy_from_iter_full(to, copy, from))
2195		return -EFAULT;
2196
2197	return 0;
2198}
2199
2200static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2201				       struct iov_iter *from, int copy)
2202{
2203	int err, offset = skb->len;
2204
2205	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2206				       copy, offset);
2207	if (err)
2208		__skb_trim(skb, offset);
2209
2210	return err;
2211}
2212
2213static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2214					   struct sk_buff *skb,
2215					   struct page *page,
2216					   int off, int copy)
2217{
2218	int err;
2219
2220	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2221				       copy, skb->len);
2222	if (err)
2223		return err;
2224
2225	skb_len_add(skb, copy);
2226	sk_wmem_queued_add(sk, copy);
2227	sk_mem_charge(sk, copy);
2228	return 0;
2229}
2230
2231/**
2232 * sk_wmem_alloc_get - returns write allocations
2233 * @sk: socket
2234 *
2235 * Return: sk_wmem_alloc minus initial offset of one
2236 */
2237static inline int sk_wmem_alloc_get(const struct sock *sk)
2238{
2239	return refcount_read(&sk->sk_wmem_alloc) - 1;
2240}
2241
2242/**
2243 * sk_rmem_alloc_get - returns read allocations
2244 * @sk: socket
2245 *
2246 * Return: sk_rmem_alloc
2247 */
2248static inline int sk_rmem_alloc_get(const struct sock *sk)
2249{
2250	return atomic_read(&sk->sk_rmem_alloc);
2251}
2252
2253/**
2254 * sk_has_allocations - check if allocations are outstanding
2255 * @sk: socket
2256 *
2257 * Return: true if socket has write or read allocations
2258 */
2259static inline bool sk_has_allocations(const struct sock *sk)
2260{
2261	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2262}
2263
2264/**
2265 * skwq_has_sleeper - check if there are any waiting processes
2266 * @wq: struct socket_wq
2267 *
2268 * Return: true if socket_wq has waiting processes
2269 *
2270 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2271 * barrier call. They were added due to the race found within the tcp code.
2272 *
2273 * Consider following tcp code paths::
2274 *
2275 *   CPU1                CPU2
2276 *   sys_select          receive packet
2277 *   ...                 ...
2278 *   __add_wait_queue    update tp->rcv_nxt
2279 *   ...                 ...
2280 *   tp->rcv_nxt check   sock_def_readable
2281 *   ...                 {
2282 *   schedule               rcu_read_lock();
2283 *                          wq = rcu_dereference(sk->sk_wq);
2284 *                          if (wq && waitqueue_active(&wq->wait))
2285 *                              wake_up_interruptible(&wq->wait)
2286 *                          ...
2287 *                       }
2288 *
2289 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2290 * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2291 * could then endup calling schedule and sleep forever if there are no more
2292 * data on the socket.
2293 *
2294 */
2295static inline bool skwq_has_sleeper(struct socket_wq *wq)
2296{
2297	return wq && wq_has_sleeper(&wq->wait);
2298}
2299
2300/**
2301 * sock_poll_wait - wrapper for the poll_wait call.
2302 * @filp:           file
2303 * @sock:           socket to wait on
2304 * @p:              poll_table
2305 *
2306 * See the comments in the wq_has_sleeper function.
2307 */
2308static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2309				  poll_table *p)
2310{
2311	/* Provides a barrier we need to be sure we are in sync
2312	 * with the socket flags modification.
2313	 *
2314	 * This memory barrier is paired in the wq_has_sleeper.
2315	 */
2316	poll_wait(filp, &sock->wq.wait, p);
 
 
 
2317}
2318
2319static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2320{
2321	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2322	u32 txhash = READ_ONCE(sk->sk_txhash);
2323
2324	if (txhash) {
2325		skb->l4_hash = 1;
2326		skb->hash = txhash;
2327	}
2328}
2329
2330void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2331
2332/*
2333 *	Queue a received datagram if it will fit. Stream and sequenced
2334 *	protocols can't normally use this as they need to fit buffers in
2335 *	and play with them.
2336 *
2337 *	Inlined as it's very short and called for pretty much every
2338 *	packet ever received.
2339 */
2340static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2341{
2342	skb_orphan(skb);
2343	skb->sk = sk;
2344	skb->destructor = sock_rfree;
2345	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2346	sk_mem_charge(sk, skb->truesize);
2347}
2348
2349static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2350{
2351	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2352		skb_orphan(skb);
2353		skb->destructor = sock_efree;
2354		skb->sk = sk;
2355		return true;
2356	}
2357	return false;
2358}
2359
2360static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2361{
2362	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2363	if (skb) {
2364		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2365			skb_set_owner_r(skb, sk);
2366			return skb;
2367		}
2368		__kfree_skb(skb);
2369	}
2370	return NULL;
2371}
2372
2373static inline void skb_prepare_for_gro(struct sk_buff *skb)
2374{
2375	if (skb->destructor != sock_wfree) {
2376		skb_orphan(skb);
2377		return;
2378	}
2379	skb->slow_gro = 1;
2380}
2381
2382void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2383		    unsigned long expires);
2384
2385void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2386
2387void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2388
2389int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2390			struct sk_buff *skb, unsigned int flags,
2391			void (*destructor)(struct sock *sk,
2392					   struct sk_buff *skb));
2393int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2394
2395int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2396			      enum skb_drop_reason *reason);
2397
2398static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2399{
2400	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2401}
2402
2403int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2404struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2405
2406/*
2407 *	Recover an error report and clear atomically
2408 */
2409
2410static inline int sock_error(struct sock *sk)
2411{
2412	int err;
2413
2414	/* Avoid an atomic operation for the common case.
2415	 * This is racy since another cpu/thread can change sk_err under us.
2416	 */
2417	if (likely(data_race(!sk->sk_err)))
2418		return 0;
2419
2420	err = xchg(&sk->sk_err, 0);
2421	return -err;
2422}
2423
2424void sk_error_report(struct sock *sk);
2425
2426static inline unsigned long sock_wspace(struct sock *sk)
2427{
2428	int amt = 0;
2429
2430	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2431		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2432		if (amt < 0)
2433			amt = 0;
2434	}
2435	return amt;
2436}
2437
2438/* Note:
2439 *  We use sk->sk_wq_raw, from contexts knowing this
2440 *  pointer is not NULL and cannot disappear/change.
2441 */
2442static inline void sk_set_bit(int nr, struct sock *sk)
2443{
2444	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2445	    !sock_flag(sk, SOCK_FASYNC))
2446		return;
2447
2448	set_bit(nr, &sk->sk_wq_raw->flags);
2449}
2450
2451static inline void sk_clear_bit(int nr, struct sock *sk)
2452{
2453	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2454	    !sock_flag(sk, SOCK_FASYNC))
2455		return;
2456
2457	clear_bit(nr, &sk->sk_wq_raw->flags);
2458}
2459
2460static inline void sk_wake_async(const struct sock *sk, int how, int band)
2461{
2462	if (sock_flag(sk, SOCK_FASYNC)) {
2463		rcu_read_lock();
2464		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2465		rcu_read_unlock();
2466	}
2467}
2468
2469static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band)
2470{
2471	if (unlikely(sock_flag(sk, SOCK_FASYNC)))
2472		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2473}
2474
2475/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2476 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2477 * Note: for send buffers, TCP works better if we can build two skbs at
2478 * minimum.
2479 */
2480#define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2481
2482#define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2483#define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2484
2485static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2486{
2487	u32 val;
2488
2489	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2490		return;
2491
2492	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2493	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2494
2495	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2496}
2497
2498/**
2499 * sk_page_frag - return an appropriate page_frag
2500 * @sk: socket
2501 *
2502 * Use the per task page_frag instead of the per socket one for
2503 * optimization when we know that we're in process context and own
2504 * everything that's associated with %current.
2505 *
2506 * Both direct reclaim and page faults can nest inside other
2507 * socket operations and end up recursing into sk_page_frag()
2508 * while it's already in use: explicitly avoid task page_frag
2509 * when users disable sk_use_task_frag.
2510 *
2511 * Return: a per task page_frag if context allows that,
2512 * otherwise a per socket one.
2513 */
2514static inline struct page_frag *sk_page_frag(struct sock *sk)
2515{
2516	if (sk->sk_use_task_frag)
2517		return &current->task_frag;
2518
2519	return &sk->sk_frag;
2520}
2521
2522bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2523
2524/*
2525 *	Default write policy as shown to user space via poll/select/SIGIO
2526 */
2527static inline bool sock_writeable(const struct sock *sk)
2528{
2529	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2530}
2531
2532static inline gfp_t gfp_any(void)
2533{
2534	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2535}
2536
2537static inline gfp_t gfp_memcg_charge(void)
2538{
2539	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2540}
2541
2542static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2543{
2544	return noblock ? 0 : sk->sk_rcvtimeo;
2545}
2546
2547static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2548{
2549	return noblock ? 0 : sk->sk_sndtimeo;
2550}
2551
2552static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2553{
2554	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2555
2556	return v ?: 1;
2557}
2558
2559/* Alas, with timeout socket operations are not restartable.
2560 * Compare this to poll().
2561 */
2562static inline int sock_intr_errno(long timeo)
2563{
2564	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2565}
2566
2567struct sock_skb_cb {
2568	u32 dropcount;
2569};
2570
2571/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2572 * using skb->cb[] would keep using it directly and utilize its
2573 * alignment guarantee.
2574 */
2575#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2576			    sizeof(struct sock_skb_cb)))
2577
2578#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2579			    SOCK_SKB_CB_OFFSET))
2580
2581#define sock_skb_cb_check_size(size) \
2582	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2583
2584static inline void
2585sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2586{
2587	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2588						atomic_read(&sk->sk_drops) : 0;
2589}
2590
2591static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2592{
2593	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2594
2595	atomic_add(segs, &sk->sk_drops);
2596}
2597
2598static inline ktime_t sock_read_timestamp(struct sock *sk)
2599{
2600#if BITS_PER_LONG==32
2601	unsigned int seq;
2602	ktime_t kt;
2603
2604	do {
2605		seq = read_seqbegin(&sk->sk_stamp_seq);
2606		kt = sk->sk_stamp;
2607	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2608
2609	return kt;
2610#else
2611	return READ_ONCE(sk->sk_stamp);
2612#endif
2613}
2614
2615static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2616{
2617#if BITS_PER_LONG==32
2618	write_seqlock(&sk->sk_stamp_seq);
2619	sk->sk_stamp = kt;
2620	write_sequnlock(&sk->sk_stamp_seq);
2621#else
2622	WRITE_ONCE(sk->sk_stamp, kt);
2623#endif
2624}
2625
2626void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2627			   struct sk_buff *skb);
2628void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2629			     struct sk_buff *skb);
2630
2631static inline void
2632sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2633{
2634	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2635	u32 tsflags = READ_ONCE(sk->sk_tsflags);
2636	ktime_t kt = skb->tstamp;
2637	/*
2638	 * generate control messages if
2639	 * - receive time stamping in software requested
2640	 * - software time stamp available and wanted
2641	 * - hardware time stamps available and wanted
2642	 */
2643	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2644	    (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2645	    (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2646	    (hwtstamps->hwtstamp &&
2647	     (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2648		__sock_recv_timestamp(msg, sk, skb);
2649	else
2650		sock_write_timestamp(sk, kt);
2651
2652	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2653		__sock_recv_wifi_status(msg, sk, skb);
2654}
2655
2656void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2657		       struct sk_buff *skb);
2658
2659#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2660static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2661				   struct sk_buff *skb)
2662{
2663#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2664			   (1UL << SOCK_RCVTSTAMP)			| \
2665			   (1UL << SOCK_RCVMARK))
2666#define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2667			   SOF_TIMESTAMPING_RAW_HARDWARE)
2668
2669	if (sk->sk_flags & FLAGS_RECV_CMSGS ||
2670	    READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY)
2671		__sock_recv_cmsgs(msg, sk, skb);
2672	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2673		sock_write_timestamp(sk, skb->tstamp);
2674	else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2675		sock_write_timestamp(sk, 0);
2676}
2677
2678void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags);
2679
2680/**
2681 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2682 * @sk:		socket sending this packet
2683 * @sockc:	pointer to socket cmsg cookie to get timestamping info
2684 * @tx_flags:	completed with instructions for time stamping
2685 * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2686 *
2687 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2688 */
2689static inline void _sock_tx_timestamp(struct sock *sk,
2690				      const struct sockcm_cookie *sockc,
2691				      __u8 *tx_flags, __u32 *tskey)
2692{
2693	__u32 tsflags = sockc->tsflags;
2694
2695	if (unlikely(tsflags)) {
2696		__sock_tx_timestamp(tsflags, tx_flags);
2697		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2698		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) {
2699			if (tsflags & SOCKCM_FLAG_TS_OPT_ID)
2700				*tskey = sockc->ts_opt_id;
2701			else
2702				*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2703		}
2704	}
2705	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2706		*tx_flags |= SKBTX_WIFI_STATUS;
2707}
2708
2709static inline void sock_tx_timestamp(struct sock *sk,
2710				     const struct sockcm_cookie *sockc,
2711				     __u8 *tx_flags)
2712{
2713	_sock_tx_timestamp(sk, sockc, tx_flags, NULL);
2714}
2715
2716static inline void skb_setup_tx_timestamp(struct sk_buff *skb,
2717					  const struct sockcm_cookie *sockc)
2718{
2719	_sock_tx_timestamp(skb->sk, sockc, &skb_shinfo(skb)->tx_flags,
2720			   &skb_shinfo(skb)->tskey);
2721}
2722
2723static inline bool sk_is_inet(const struct sock *sk)
2724{
2725	int family = READ_ONCE(sk->sk_family);
2726
2727	return family == AF_INET || family == AF_INET6;
2728}
2729
2730static inline bool sk_is_tcp(const struct sock *sk)
2731{
2732	return sk_is_inet(sk) &&
2733	       sk->sk_type == SOCK_STREAM &&
2734	       sk->sk_protocol == IPPROTO_TCP;
2735}
2736
2737static inline bool sk_is_udp(const struct sock *sk)
2738{
2739	return sk_is_inet(sk) &&
2740	       sk->sk_type == SOCK_DGRAM &&
2741	       sk->sk_protocol == IPPROTO_UDP;
2742}
2743
2744static inline bool sk_is_stream_unix(const struct sock *sk)
2745{
2746	return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM;
2747}
2748
2749static inline bool sk_is_vsock(const struct sock *sk)
2750{
2751	return sk->sk_family == AF_VSOCK;
2752}
2753
2754/**
2755 * sk_eat_skb - Release a skb if it is no longer needed
2756 * @sk: socket to eat this skb from
2757 * @skb: socket buffer to eat
2758 *
2759 * This routine must be called with interrupts disabled or with the socket
2760 * locked so that the sk_buff queue operation is ok.
2761*/
2762static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2763{
2764	__skb_unlink(skb, &sk->sk_receive_queue);
2765	__kfree_skb(skb);
2766}
2767
2768static inline bool
2769skb_sk_is_prefetched(struct sk_buff *skb)
2770{
2771#ifdef CONFIG_INET
2772	return skb->destructor == sock_pfree;
2773#else
2774	return false;
2775#endif /* CONFIG_INET */
2776}
2777
2778/* This helper checks if a socket is a full socket,
2779 * ie _not_ a timewait or request socket.
2780 */
2781static inline bool sk_fullsock(const struct sock *sk)
2782{
2783	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2784}
2785
2786static inline bool
2787sk_is_refcounted(struct sock *sk)
2788{
2789	/* Only full sockets have sk->sk_flags. */
2790	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2791}
2792
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2793/* Checks if this SKB belongs to an HW offloaded socket
2794 * and whether any SW fallbacks are required based on dev.
2795 * Check decrypted mark in case skb_orphan() cleared socket.
2796 */
2797static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2798						   struct net_device *dev)
2799{
2800#ifdef CONFIG_SOCK_VALIDATE_XMIT
2801	struct sock *sk = skb->sk;
2802
2803	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2804		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2805	} else if (unlikely(skb_is_decrypted(skb))) {
 
2806		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2807		kfree_skb(skb);
2808		skb = NULL;
 
2809	}
2810#endif
2811
2812	return skb;
2813}
2814
2815/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2816 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2817 */
2818static inline bool sk_listener(const struct sock *sk)
2819{
2820	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2821}
2822
2823/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT
2824 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE)
2825 * TCP RST and ACK can be attached to TIME_WAIT.
2826 */
2827static inline bool sk_listener_or_tw(const struct sock *sk)
2828{
2829	return (1 << READ_ONCE(sk->sk_state)) &
2830	       (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT);
2831}
2832
2833void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2834int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2835		       int type);
2836
2837bool sk_ns_capable(const struct sock *sk,
2838		   struct user_namespace *user_ns, int cap);
2839bool sk_capable(const struct sock *sk, int cap);
2840bool sk_net_capable(const struct sock *sk, int cap);
2841
2842void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2843
2844/* Take into consideration the size of the struct sk_buff overhead in the
2845 * determination of these values, since that is non-constant across
2846 * platforms.  This makes socket queueing behavior and performance
2847 * not depend upon such differences.
2848 */
2849#define _SK_MEM_PACKETS		256
2850#define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2851#define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2852#define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2853
2854extern __u32 sysctl_wmem_max;
2855extern __u32 sysctl_rmem_max;
 
 
2856
2857extern __u32 sysctl_wmem_default;
2858extern __u32 sysctl_rmem_default;
2859
2860#define SKB_FRAG_PAGE_ORDER	get_order(32768)
2861DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2862
2863static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2864{
2865	/* Does this proto have per netns sysctl_wmem ? */
2866	if (proto->sysctl_wmem_offset)
2867		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2868
2869	return READ_ONCE(*proto->sysctl_wmem);
2870}
2871
2872static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2873{
2874	/* Does this proto have per netns sysctl_rmem ? */
2875	if (proto->sysctl_rmem_offset)
2876		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2877
2878	return READ_ONCE(*proto->sysctl_rmem);
2879}
2880
2881/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2882 * Some wifi drivers need to tweak it to get more chunks.
2883 * They can use this helper from their ndo_start_xmit()
2884 */
2885static inline void sk_pacing_shift_update(struct sock *sk, int val)
2886{
2887	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2888		return;
2889	WRITE_ONCE(sk->sk_pacing_shift, val);
2890}
2891
2892/* if a socket is bound to a device, check that the given device
2893 * index is either the same or that the socket is bound to an L3
2894 * master device and the given device index is also enslaved to
2895 * that L3 master
2896 */
2897static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2898{
2899	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2900	int mdif;
2901
2902	if (!bound_dev_if || bound_dev_if == dif)
2903		return true;
2904
2905	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2906	if (mdif && mdif == bound_dev_if)
2907		return true;
2908
2909	return false;
2910}
2911
2912void sock_def_readable(struct sock *sk);
2913
2914int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2915void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2916int sock_set_timestamping(struct sock *sk, int optname,
2917			  struct so_timestamping timestamping);
2918
2919void sock_enable_timestamps(struct sock *sk);
2920void sock_no_linger(struct sock *sk);
2921void sock_set_keepalive(struct sock *sk);
2922void sock_set_priority(struct sock *sk, u32 priority);
2923void sock_set_rcvbuf(struct sock *sk, int val);
2924void sock_set_mark(struct sock *sk, u32 val);
2925void sock_set_reuseaddr(struct sock *sk);
2926void sock_set_reuseport(struct sock *sk);
2927void sock_set_sndtimeo(struct sock *sk, s64 secs);
2928
2929int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2930
2931int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2932int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2933			   sockptr_t optval, int optlen, bool old_timeval);
2934
2935int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
2936		     void __user *arg, void *karg, size_t size);
2937int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
2938static inline bool sk_is_readable(struct sock *sk)
2939{
2940	if (sk->sk_prot->sock_is_readable)
2941		return sk->sk_prot->sock_is_readable(sk);
2942	return false;
2943}
2944#endif	/* _SOCK_H */
v6.8
   1/* SPDX-License-Identifier: GPL-2.0-or-later */
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Definitions for the AF_INET socket handler.
   8 *
   9 * Version:	@(#)sock.h	1.0.4	05/13/93
  10 *
  11 * Authors:	Ross Biro
  12 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  13 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
  14 *		Florian La Roche <flla@stud.uni-sb.de>
  15 *
  16 * Fixes:
  17 *		Alan Cox	:	Volatiles in skbuff pointers. See
  18 *					skbuff comments. May be overdone,
  19 *					better to prove they can be removed
  20 *					than the reverse.
  21 *		Alan Cox	:	Added a zapped field for tcp to note
  22 *					a socket is reset and must stay shut up
  23 *		Alan Cox	:	New fields for options
  24 *	Pauline Middelink	:	identd support
  25 *		Alan Cox	:	Eliminate low level recv/recvfrom
  26 *		David S. Miller	:	New socket lookup architecture.
  27 *              Steve Whitehouse:       Default routines for sock_ops
  28 *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
  29 *              			protinfo be just a void pointer, as the
  30 *              			protocol specific parts were moved to
  31 *              			respective headers and ipv4/v6, etc now
  32 *              			use private slabcaches for its socks
  33 *              Pedro Hortas	:	New flags field for socket options
  34 */
  35#ifndef _SOCK_H
  36#define _SOCK_H
  37
  38#include <linux/hardirq.h>
  39#include <linux/kernel.h>
  40#include <linux/list.h>
  41#include <linux/list_nulls.h>
  42#include <linux/timer.h>
  43#include <linux/cache.h>
  44#include <linux/bitops.h>
  45#include <linux/lockdep.h>
  46#include <linux/netdevice.h>
  47#include <linux/skbuff.h>	/* struct sk_buff */
  48#include <linux/mm.h>
  49#include <linux/security.h>
  50#include <linux/slab.h>
  51#include <linux/uaccess.h>
  52#include <linux/page_counter.h>
  53#include <linux/memcontrol.h>
  54#include <linux/static_key.h>
  55#include <linux/sched.h>
  56#include <linux/wait.h>
  57#include <linux/cgroup-defs.h>
  58#include <linux/rbtree.h>
  59#include <linux/rculist_nulls.h>
  60#include <linux/poll.h>
  61#include <linux/sockptr.h>
  62#include <linux/indirect_call_wrapper.h>
  63#include <linux/atomic.h>
  64#include <linux/refcount.h>
  65#include <linux/llist.h>
  66#include <net/dst.h>
  67#include <net/checksum.h>
  68#include <net/tcp_states.h>
  69#include <linux/net_tstamp.h>
  70#include <net/l3mdev.h>
  71#include <uapi/linux/socket.h>
  72
  73/*
  74 * This structure really needs to be cleaned up.
  75 * Most of it is for TCP, and not used by any of
  76 * the other protocols.
  77 */
  78
  79/* This is the per-socket lock.  The spinlock provides a synchronization
  80 * between user contexts and software interrupt processing, whereas the
  81 * mini-semaphore synchronizes multiple users amongst themselves.
  82 */
  83typedef struct {
  84	spinlock_t		slock;
  85	int			owned;
  86	wait_queue_head_t	wq;
  87	/*
  88	 * We express the mutex-alike socket_lock semantics
  89	 * to the lock validator by explicitly managing
  90	 * the slock as a lock variant (in addition to
  91	 * the slock itself):
  92	 */
  93#ifdef CONFIG_DEBUG_LOCK_ALLOC
  94	struct lockdep_map dep_map;
  95#endif
  96} socket_lock_t;
  97
  98struct sock;
  99struct proto;
 100struct net;
 101
 102typedef __u32 __bitwise __portpair;
 103typedef __u64 __bitwise __addrpair;
 104
 105/**
 106 *	struct sock_common - minimal network layer representation of sockets
 107 *	@skc_daddr: Foreign IPv4 addr
 108 *	@skc_rcv_saddr: Bound local IPv4 addr
 109 *	@skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
 110 *	@skc_hash: hash value used with various protocol lookup tables
 111 *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
 112 *	@skc_dport: placeholder for inet_dport/tw_dport
 113 *	@skc_num: placeholder for inet_num/tw_num
 114 *	@skc_portpair: __u32 union of @skc_dport & @skc_num
 115 *	@skc_family: network address family
 116 *	@skc_state: Connection state
 117 *	@skc_reuse: %SO_REUSEADDR setting
 118 *	@skc_reuseport: %SO_REUSEPORT setting
 119 *	@skc_ipv6only: socket is IPV6 only
 120 *	@skc_net_refcnt: socket is using net ref counting
 121 *	@skc_bound_dev_if: bound device index if != 0
 122 *	@skc_bind_node: bind hash linkage for various protocol lookup tables
 123 *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
 124 *	@skc_prot: protocol handlers inside a network family
 125 *	@skc_net: reference to the network namespace of this socket
 126 *	@skc_v6_daddr: IPV6 destination address
 127 *	@skc_v6_rcv_saddr: IPV6 source address
 128 *	@skc_cookie: socket's cookie value
 129 *	@skc_node: main hash linkage for various protocol lookup tables
 130 *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
 131 *	@skc_tx_queue_mapping: tx queue number for this connection
 132 *	@skc_rx_queue_mapping: rx queue number for this connection
 133 *	@skc_flags: place holder for sk_flags
 134 *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
 135 *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
 136 *	@skc_listener: connection request listener socket (aka rsk_listener)
 137 *		[union with @skc_flags]
 138 *	@skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
 139 *		[union with @skc_flags]
 140 *	@skc_incoming_cpu: record/match cpu processing incoming packets
 141 *	@skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
 142 *		[union with @skc_incoming_cpu]
 143 *	@skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
 144 *		[union with @skc_incoming_cpu]
 145 *	@skc_refcnt: reference count
 146 *
 147 *	This is the minimal network layer representation of sockets, the header
 148 *	for struct sock and struct inet_timewait_sock.
 149 */
 150struct sock_common {
 151	union {
 152		__addrpair	skc_addrpair;
 153		struct {
 154			__be32	skc_daddr;
 155			__be32	skc_rcv_saddr;
 156		};
 157	};
 158	union  {
 159		unsigned int	skc_hash;
 160		__u16		skc_u16hashes[2];
 161	};
 162	/* skc_dport && skc_num must be grouped as well */
 163	union {
 164		__portpair	skc_portpair;
 165		struct {
 166			__be16	skc_dport;
 167			__u16	skc_num;
 168		};
 169	};
 170
 171	unsigned short		skc_family;
 172	volatile unsigned char	skc_state;
 173	unsigned char		skc_reuse:4;
 174	unsigned char		skc_reuseport:1;
 175	unsigned char		skc_ipv6only:1;
 176	unsigned char		skc_net_refcnt:1;
 177	int			skc_bound_dev_if;
 178	union {
 179		struct hlist_node	skc_bind_node;
 180		struct hlist_node	skc_portaddr_node;
 181	};
 182	struct proto		*skc_prot;
 183	possible_net_t		skc_net;
 184
 185#if IS_ENABLED(CONFIG_IPV6)
 186	struct in6_addr		skc_v6_daddr;
 187	struct in6_addr		skc_v6_rcv_saddr;
 188#endif
 189
 190	atomic64_t		skc_cookie;
 191
 192	/* following fields are padding to force
 193	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
 194	 * assuming IPV6 is enabled. We use this padding differently
 195	 * for different kind of 'sockets'
 196	 */
 197	union {
 198		unsigned long	skc_flags;
 199		struct sock	*skc_listener; /* request_sock */
 200		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
 201	};
 202	/*
 203	 * fields between dontcopy_begin/dontcopy_end
 204	 * are not copied in sock_copy()
 205	 */
 206	/* private: */
 207	int			skc_dontcopy_begin[0];
 208	/* public: */
 209	union {
 210		struct hlist_node	skc_node;
 211		struct hlist_nulls_node skc_nulls_node;
 212	};
 213	unsigned short		skc_tx_queue_mapping;
 214#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
 215	unsigned short		skc_rx_queue_mapping;
 216#endif
 217	union {
 218		int		skc_incoming_cpu;
 219		u32		skc_rcv_wnd;
 220		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
 221	};
 222
 223	refcount_t		skc_refcnt;
 224	/* private: */
 225	int                     skc_dontcopy_end[0];
 226	union {
 227		u32		skc_rxhash;
 228		u32		skc_window_clamp;
 229		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
 230	};
 231	/* public: */
 232};
 233
 234struct bpf_local_storage;
 235struct sk_filter;
 236
 237/**
 238  *	struct sock - network layer representation of sockets
 239  *	@__sk_common: shared layout with inet_timewait_sock
 240  *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
 241  *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
 242  *	@sk_lock:	synchronizer
 243  *	@sk_kern_sock: True if sock is using kernel lock classes
 244  *	@sk_rcvbuf: size of receive buffer in bytes
 245  *	@sk_wq: sock wait queue and async head
 246  *	@sk_rx_dst: receive input route used by early demux
 247  *	@sk_rx_dst_ifindex: ifindex for @sk_rx_dst
 248  *	@sk_rx_dst_cookie: cookie for @sk_rx_dst
 249  *	@sk_dst_cache: destination cache
 250  *	@sk_dst_pending_confirm: need to confirm neighbour
 251  *	@sk_policy: flow policy
 252  *	@sk_receive_queue: incoming packets
 253  *	@sk_wmem_alloc: transmit queue bytes committed
 254  *	@sk_tsq_flags: TCP Small Queues flags
 255  *	@sk_write_queue: Packet sending queue
 256  *	@sk_omem_alloc: "o" is "option" or "other"
 257  *	@sk_wmem_queued: persistent queue size
 258  *	@sk_forward_alloc: space allocated forward
 259  *	@sk_reserved_mem: space reserved and non-reclaimable for the socket
 260  *	@sk_napi_id: id of the last napi context to receive data for sk
 261  *	@sk_ll_usec: usecs to busypoll when there is no data
 262  *	@sk_allocation: allocation mode
 263  *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
 264  *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
 265  *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
 266  *	@sk_sndbuf: size of send buffer in bytes
 267  *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
 268  *	@sk_no_check_rx: allow zero checksum in RX packets
 269  *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
 270  *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
 271  *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
 272  *	@sk_gso_max_size: Maximum GSO segment size to build
 273  *	@sk_gso_max_segs: Maximum number of GSO segments
 274  *	@sk_pacing_shift: scaling factor for TCP Small Queues
 275  *	@sk_lingertime: %SO_LINGER l_linger setting
 276  *	@sk_backlog: always used with the per-socket spinlock held
 277  *	@sk_callback_lock: used with the callbacks in the end of this struct
 278  *	@sk_error_queue: rarely used
 279  *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
 280  *			  IPV6_ADDRFORM for instance)
 281  *	@sk_err: last error
 282  *	@sk_err_soft: errors that don't cause failure but are the cause of a
 283  *		      persistent failure not just 'timed out'
 284  *	@sk_drops: raw/udp drops counter
 285  *	@sk_ack_backlog: current listen backlog
 286  *	@sk_max_ack_backlog: listen backlog set in listen()
 287  *	@sk_uid: user id of owner
 288  *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
 289  *	@sk_busy_poll_budget: napi processing budget when busypolling
 290  *	@sk_priority: %SO_PRIORITY setting
 291  *	@sk_type: socket type (%SOCK_STREAM, etc)
 292  *	@sk_protocol: which protocol this socket belongs in this network family
 293  *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
 294  *	@sk_peer_pid: &struct pid for this socket's peer
 295  *	@sk_peer_cred: %SO_PEERCRED setting
 296  *	@sk_rcvlowat: %SO_RCVLOWAT setting
 297  *	@sk_rcvtimeo: %SO_RCVTIMEO setting
 298  *	@sk_sndtimeo: %SO_SNDTIMEO setting
 299  *	@sk_txhash: computed flow hash for use on transmit
 300  *	@sk_txrehash: enable TX hash rethink
 301  *	@sk_filter: socket filtering instructions
 302  *	@sk_timer: sock cleanup timer
 303  *	@sk_stamp: time stamp of last packet received
 304  *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
 305  *	@sk_tsflags: SO_TIMESTAMPING flags
 306  *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
 307  *			   Sockets that can be used under memory reclaim should
 308  *			   set this to false.
 309  *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
 310  *	              for timestamping
 311  *	@sk_tskey: counter to disambiguate concurrent tstamp requests
 312  *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
 313  *	@sk_socket: Identd and reporting IO signals
 314  *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
 315  *	@sk_frag: cached page frag
 316  *	@sk_peek_off: current peek_offset value
 317  *	@sk_send_head: front of stuff to transmit
 318  *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
 319  *	@sk_security: used by security modules
 320  *	@sk_mark: generic packet mark
 321  *	@sk_cgrp_data: cgroup data for this cgroup
 322  *	@sk_memcg: this socket's memory cgroup association
 323  *	@sk_write_pending: a write to stream socket waits to start
 324  *	@sk_disconnects: number of disconnect operations performed on this sock
 325  *	@sk_state_change: callback to indicate change in the state of the sock
 326  *	@sk_data_ready: callback to indicate there is data to be processed
 327  *	@sk_write_space: callback to indicate there is bf sending space available
 328  *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
 329  *	@sk_backlog_rcv: callback to process the backlog
 330  *	@sk_validate_xmit_skb: ptr to an optional validate function
 331  *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
 332  *	@sk_reuseport_cb: reuseport group container
 333  *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
 334  *	@sk_rcu: used during RCU grace period
 335  *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
 336  *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
 337  *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
 338  *	@sk_txtime_unused: unused txtime flags
 339  *	@ns_tracker: tracker for netns reference
 
 340  */
 341struct sock {
 342	/*
 343	 * Now struct inet_timewait_sock also uses sock_common, so please just
 344	 * don't add nothing before this first member (__sk_common) --acme
 345	 */
 346	struct sock_common	__sk_common;
 347#define sk_node			__sk_common.skc_node
 348#define sk_nulls_node		__sk_common.skc_nulls_node
 349#define sk_refcnt		__sk_common.skc_refcnt
 350#define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
 351#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
 352#define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
 353#endif
 354
 355#define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
 356#define sk_dontcopy_end		__sk_common.skc_dontcopy_end
 357#define sk_hash			__sk_common.skc_hash
 358#define sk_portpair		__sk_common.skc_portpair
 359#define sk_num			__sk_common.skc_num
 360#define sk_dport		__sk_common.skc_dport
 361#define sk_addrpair		__sk_common.skc_addrpair
 362#define sk_daddr		__sk_common.skc_daddr
 363#define sk_rcv_saddr		__sk_common.skc_rcv_saddr
 364#define sk_family		__sk_common.skc_family
 365#define sk_state		__sk_common.skc_state
 366#define sk_reuse		__sk_common.skc_reuse
 367#define sk_reuseport		__sk_common.skc_reuseport
 368#define sk_ipv6only		__sk_common.skc_ipv6only
 369#define sk_net_refcnt		__sk_common.skc_net_refcnt
 370#define sk_bound_dev_if		__sk_common.skc_bound_dev_if
 371#define sk_bind_node		__sk_common.skc_bind_node
 372#define sk_prot			__sk_common.skc_prot
 373#define sk_net			__sk_common.skc_net
 374#define sk_v6_daddr		__sk_common.skc_v6_daddr
 375#define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
 376#define sk_cookie		__sk_common.skc_cookie
 377#define sk_incoming_cpu		__sk_common.skc_incoming_cpu
 378#define sk_flags		__sk_common.skc_flags
 379#define sk_rxhash		__sk_common.skc_rxhash
 380
 381	/* early demux fields */
 382	struct dst_entry __rcu	*sk_rx_dst;
 383	int			sk_rx_dst_ifindex;
 384	u32			sk_rx_dst_cookie;
 385
 386	socket_lock_t		sk_lock;
 387	atomic_t		sk_drops;
 388	int			sk_rcvlowat;
 389	struct sk_buff_head	sk_error_queue;
 390	struct sk_buff_head	sk_receive_queue;
 391	/*
 392	 * The backlog queue is special, it is always used with
 393	 * the per-socket spinlock held and requires low latency
 394	 * access. Therefore we special case it's implementation.
 395	 * Note : rmem_alloc is in this structure to fill a hole
 396	 * on 64bit arches, not because its logically part of
 397	 * backlog.
 398	 */
 399	struct {
 400		atomic_t	rmem_alloc;
 401		int		len;
 402		struct sk_buff	*head;
 403		struct sk_buff	*tail;
 404	} sk_backlog;
 
 
 
 405
 406#define sk_rmem_alloc sk_backlog.rmem_alloc
 
 
 
 
 407
 408	int			sk_forward_alloc;
 409	u32			sk_reserved_mem;
 410#ifdef CONFIG_NET_RX_BUSY_POLL
 411	unsigned int		sk_ll_usec;
 412	/* ===== mostly read cache line ===== */
 413	unsigned int		sk_napi_id;
 
 
 414#endif
 
 415	int			sk_rcvbuf;
 416	int			sk_disconnects;
 417
 418	struct sk_filter __rcu	*sk_filter;
 419	union {
 420		struct socket_wq __rcu	*sk_wq;
 421		/* private: */
 422		struct socket_wq	*sk_wq_raw;
 423		/* public: */
 424	};
 
 
 
 
 
 
 
 
 
 
 425#ifdef CONFIG_XFRM
 426	struct xfrm_policy __rcu *sk_policy[2];
 427#endif
 
 
 
 
 
 
 
 
 428
 429	struct dst_entry __rcu	*sk_dst_cache;
 
 430	atomic_t		sk_omem_alloc;
 431	int			sk_sndbuf;
 432
 433	/* ===== cache line for TX ===== */
 434	int			sk_wmem_queued;
 435	refcount_t		sk_wmem_alloc;
 436	unsigned long		sk_tsq_flags;
 437	union {
 438		struct sk_buff	*sk_send_head;
 439		struct rb_root	tcp_rtx_queue;
 440	};
 441	struct sk_buff_head	sk_write_queue;
 442	__s32			sk_peek_off;
 443	int			sk_write_pending;
 444	__u32			sk_dst_pending_confirm;
 445	u32			sk_pacing_status; /* see enum sk_pacing */
 446	long			sk_sndtimeo;
 447	struct timer_list	sk_timer;
 448	__u32			sk_priority;
 449	__u32			sk_mark;
 450	unsigned long		sk_pacing_rate; /* bytes per second */
 
 
 
 
 
 451	unsigned long		sk_max_pacing_rate;
 452	struct page_frag	sk_frag;
 
 
 
 453	netdev_features_t	sk_route_caps;
 454	int			sk_gso_type;
 
 
 
 
 
 
 455	unsigned int		sk_gso_max_size;
 456	gfp_t			sk_allocation;
 457	__u32			sk_txhash;
 
 
 
 458
 459	/*
 460	 * Because of non atomicity rules, all
 461	 * changes are protected by socket lock.
 462	 */
 463	u8			sk_gso_disabled : 1,
 464				sk_kern_sock : 1,
 465				sk_no_check_tx : 1,
 466				sk_no_check_rx : 1,
 467				sk_userlocks : 4;
 468	u8			sk_pacing_shift;
 469	u16			sk_type;
 470	u16			sk_protocol;
 471	u16			sk_gso_max_segs;
 472	unsigned long	        sk_lingertime;
 473	struct proto		*sk_prot_creator;
 474	rwlock_t		sk_callback_lock;
 475	int			sk_err,
 476				sk_err_soft;
 477	u32			sk_ack_backlog;
 478	u32			sk_max_ack_backlog;
 479	kuid_t			sk_uid;
 480	u8			sk_txrehash;
 481#ifdef CONFIG_NET_RX_BUSY_POLL
 482	u8			sk_prefer_busy_poll;
 483	u16			sk_busy_poll_budget;
 484#endif
 485	spinlock_t		sk_peer_lock;
 486	int			sk_bind_phc;
 487	struct pid		*sk_peer_pid;
 488	const struct cred	*sk_peer_cred;
 489
 490	long			sk_rcvtimeo;
 491	ktime_t			sk_stamp;
 492#if BITS_PER_LONG==32
 493	seqlock_t		sk_stamp_seq;
 494#endif
 495	atomic_t		sk_tskey;
 496	atomic_t		sk_zckey;
 497	u32			sk_tsflags;
 498	u8			sk_shutdown;
 499
 
 500	u8			sk_clockid;
 501	u8			sk_txtime_deadline_mode : 1,
 502				sk_txtime_report_errors : 1,
 503				sk_txtime_unused : 6;
 504	bool			sk_use_task_frag;
 505
 506	struct socket		*sk_socket;
 507	void			*sk_user_data;
 508#ifdef CONFIG_SECURITY
 509	void			*sk_security;
 510#endif
 511	struct sock_cgroup_data	sk_cgrp_data;
 512	struct mem_cgroup	*sk_memcg;
 513	void			(*sk_state_change)(struct sock *sk);
 514	void			(*sk_data_ready)(struct sock *sk);
 515	void			(*sk_write_space)(struct sock *sk);
 516	void			(*sk_error_report)(struct sock *sk);
 517	int			(*sk_backlog_rcv)(struct sock *sk,
 518						  struct sk_buff *skb);
 519#ifdef CONFIG_SOCK_VALIDATE_XMIT
 520	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
 521							struct net_device *dev,
 522							struct sk_buff *skb);
 523#endif
 524	void                    (*sk_destruct)(struct sock *sk);
 525	struct sock_reuseport __rcu	*sk_reuseport_cb;
 526#ifdef CONFIG_BPF_SYSCALL
 527	struct bpf_local_storage __rcu	*sk_bpf_storage;
 528#endif
 529	struct rcu_head		sk_rcu;
 530	netns_tracker		ns_tracker;
 
 
 
 
 
 
 531};
 532
 533enum sk_pacing {
 534	SK_PACING_NONE		= 0,
 535	SK_PACING_NEEDED	= 1,
 536	SK_PACING_FQ		= 2,
 537};
 538
 539/* flag bits in sk_user_data
 540 *
 541 * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
 542 *   not be suitable for copying when cloning the socket. For instance,
 543 *   it can point to a reference counted object. sk_user_data bottom
 544 *   bit is set if pointer must not be copied.
 545 *
 546 * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
 547 *   managed/owned by a BPF reuseport array. This bit should be set
 548 *   when sk_user_data's sk is added to the bpf's reuseport_array.
 549 *
 550 * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
 551 *   sk_user_data points to psock type. This bit should be set
 552 *   when sk_user_data is assigned to a psock object.
 553 */
 554#define SK_USER_DATA_NOCOPY	1UL
 555#define SK_USER_DATA_BPF	2UL
 556#define SK_USER_DATA_PSOCK	4UL
 557#define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
 558				  SK_USER_DATA_PSOCK)
 559
 560/**
 561 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
 562 * @sk: socket
 563 */
 564static inline bool sk_user_data_is_nocopy(const struct sock *sk)
 565{
 566	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
 567}
 568
 569#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
 570
 571/**
 572 * __locked_read_sk_user_data_with_flags - return the pointer
 573 * only if argument flags all has been set in sk_user_data. Otherwise
 574 * return NULL
 575 *
 576 * @sk: socket
 577 * @flags: flag bits
 578 *
 579 * The caller must be holding sk->sk_callback_lock.
 580 */
 581static inline void *
 582__locked_read_sk_user_data_with_flags(const struct sock *sk,
 583				      uintptr_t flags)
 584{
 585	uintptr_t sk_user_data =
 586		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
 587						 lockdep_is_held(&sk->sk_callback_lock));
 588
 589	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
 590
 591	if ((sk_user_data & flags) == flags)
 592		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
 593	return NULL;
 594}
 595
 596/**
 597 * __rcu_dereference_sk_user_data_with_flags - return the pointer
 598 * only if argument flags all has been set in sk_user_data. Otherwise
 599 * return NULL
 600 *
 601 * @sk: socket
 602 * @flags: flag bits
 603 */
 604static inline void *
 605__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
 606					  uintptr_t flags)
 607{
 608	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
 609
 610	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
 611
 612	if ((sk_user_data & flags) == flags)
 613		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
 614	return NULL;
 615}
 616
 617#define rcu_dereference_sk_user_data(sk)				\
 618	__rcu_dereference_sk_user_data_with_flags(sk, 0)
 619#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
 620({									\
 621	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
 622		  __tmp2 = (uintptr_t)(flags);				\
 623	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
 624	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
 625	rcu_assign_pointer(__sk_user_data((sk)),			\
 626			   __tmp1 | __tmp2);				\
 627})
 628#define rcu_assign_sk_user_data(sk, ptr)				\
 629	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
 630
 631static inline
 632struct net *sock_net(const struct sock *sk)
 633{
 634	return read_pnet(&sk->sk_net);
 635}
 636
 637static inline
 638void sock_net_set(struct sock *sk, struct net *net)
 639{
 640	write_pnet(&sk->sk_net, net);
 641}
 642
 643/*
 644 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
 645 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
 646 * on a socket means that the socket will reuse everybody else's port
 647 * without looking at the other's sk_reuse value.
 648 */
 649
 650#define SK_NO_REUSE	0
 651#define SK_CAN_REUSE	1
 652#define SK_FORCE_REUSE	2
 653
 654int sk_set_peek_off(struct sock *sk, int val);
 655
 656static inline int sk_peek_offset(const struct sock *sk, int flags)
 657{
 658	if (unlikely(flags & MSG_PEEK)) {
 659		return READ_ONCE(sk->sk_peek_off);
 660	}
 661
 662	return 0;
 663}
 664
 665static inline void sk_peek_offset_bwd(struct sock *sk, int val)
 666{
 667	s32 off = READ_ONCE(sk->sk_peek_off);
 668
 669	if (unlikely(off >= 0)) {
 670		off = max_t(s32, off - val, 0);
 671		WRITE_ONCE(sk->sk_peek_off, off);
 672	}
 673}
 674
 675static inline void sk_peek_offset_fwd(struct sock *sk, int val)
 676{
 677	sk_peek_offset_bwd(sk, -val);
 678}
 679
 680/*
 681 * Hashed lists helper routines
 682 */
 683static inline struct sock *sk_entry(const struct hlist_node *node)
 684{
 685	return hlist_entry(node, struct sock, sk_node);
 686}
 687
 688static inline struct sock *__sk_head(const struct hlist_head *head)
 689{
 690	return hlist_entry(head->first, struct sock, sk_node);
 691}
 692
 693static inline struct sock *sk_head(const struct hlist_head *head)
 694{
 695	return hlist_empty(head) ? NULL : __sk_head(head);
 696}
 697
 698static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
 699{
 700	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
 701}
 702
 703static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
 704{
 705	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
 706}
 707
 708static inline struct sock *sk_next(const struct sock *sk)
 709{
 710	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
 711}
 712
 713static inline struct sock *sk_nulls_next(const struct sock *sk)
 714{
 715	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
 716		hlist_nulls_entry(sk->sk_nulls_node.next,
 717				  struct sock, sk_nulls_node) :
 718		NULL;
 719}
 720
 721static inline bool sk_unhashed(const struct sock *sk)
 722{
 723	return hlist_unhashed(&sk->sk_node);
 724}
 725
 726static inline bool sk_hashed(const struct sock *sk)
 727{
 728	return !sk_unhashed(sk);
 729}
 730
 731static inline void sk_node_init(struct hlist_node *node)
 732{
 733	node->pprev = NULL;
 734}
 735
 736static inline void __sk_del_node(struct sock *sk)
 737{
 738	__hlist_del(&sk->sk_node);
 739}
 740
 741/* NB: equivalent to hlist_del_init_rcu */
 742static inline bool __sk_del_node_init(struct sock *sk)
 743{
 744	if (sk_hashed(sk)) {
 745		__sk_del_node(sk);
 746		sk_node_init(&sk->sk_node);
 747		return true;
 748	}
 749	return false;
 750}
 751
 752/* Grab socket reference count. This operation is valid only
 753   when sk is ALREADY grabbed f.e. it is found in hash table
 754   or a list and the lookup is made under lock preventing hash table
 755   modifications.
 756 */
 757
 758static __always_inline void sock_hold(struct sock *sk)
 759{
 760	refcount_inc(&sk->sk_refcnt);
 761}
 762
 763/* Ungrab socket in the context, which assumes that socket refcnt
 764   cannot hit zero, f.e. it is true in context of any socketcall.
 765 */
 766static __always_inline void __sock_put(struct sock *sk)
 767{
 768	refcount_dec(&sk->sk_refcnt);
 769}
 770
 771static inline bool sk_del_node_init(struct sock *sk)
 772{
 773	bool rc = __sk_del_node_init(sk);
 774
 775	if (rc) {
 776		/* paranoid for a while -acme */
 777		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 778		__sock_put(sk);
 779	}
 780	return rc;
 781}
 782#define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
 783
 784static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
 785{
 786	if (sk_hashed(sk)) {
 787		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
 788		return true;
 789	}
 790	return false;
 791}
 792
 793static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
 794{
 795	bool rc = __sk_nulls_del_node_init_rcu(sk);
 796
 797	if (rc) {
 798		/* paranoid for a while -acme */
 799		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 800		__sock_put(sk);
 801	}
 802	return rc;
 803}
 804
 805static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
 806{
 807	hlist_add_head(&sk->sk_node, list);
 808}
 809
 810static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
 811{
 812	sock_hold(sk);
 813	__sk_add_node(sk, list);
 814}
 815
 816static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
 817{
 818	sock_hold(sk);
 819	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 820	    sk->sk_family == AF_INET6)
 821		hlist_add_tail_rcu(&sk->sk_node, list);
 822	else
 823		hlist_add_head_rcu(&sk->sk_node, list);
 824}
 825
 826static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
 827{
 828	sock_hold(sk);
 829	hlist_add_tail_rcu(&sk->sk_node, list);
 830}
 831
 832static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 833{
 834	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
 835}
 836
 837static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
 838{
 839	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
 840}
 841
 842static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 843{
 844	sock_hold(sk);
 845	__sk_nulls_add_node_rcu(sk, list);
 846}
 847
 848static inline void __sk_del_bind_node(struct sock *sk)
 849{
 850	__hlist_del(&sk->sk_bind_node);
 851}
 852
 853static inline void sk_add_bind_node(struct sock *sk,
 854					struct hlist_head *list)
 855{
 856	hlist_add_head(&sk->sk_bind_node, list);
 857}
 858
 859#define sk_for_each(__sk, list) \
 860	hlist_for_each_entry(__sk, list, sk_node)
 861#define sk_for_each_rcu(__sk, list) \
 862	hlist_for_each_entry_rcu(__sk, list, sk_node)
 863#define sk_nulls_for_each(__sk, node, list) \
 864	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
 865#define sk_nulls_for_each_rcu(__sk, node, list) \
 866	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
 867#define sk_for_each_from(__sk) \
 868	hlist_for_each_entry_from(__sk, sk_node)
 869#define sk_nulls_for_each_from(__sk, node) \
 870	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
 871		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
 872#define sk_for_each_safe(__sk, tmp, list) \
 873	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
 874#define sk_for_each_bound(__sk, list) \
 875	hlist_for_each_entry(__sk, list, sk_bind_node)
 
 
 876
 877/**
 878 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
 879 * @tpos:	the type * to use as a loop cursor.
 880 * @pos:	the &struct hlist_node to use as a loop cursor.
 881 * @head:	the head for your list.
 882 * @offset:	offset of hlist_node within the struct.
 883 *
 884 */
 885#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
 886	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
 887	     pos != NULL &&						       \
 888		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
 889	     pos = rcu_dereference(hlist_next_rcu(pos)))
 890
 891static inline struct user_namespace *sk_user_ns(const struct sock *sk)
 892{
 893	/* Careful only use this in a context where these parameters
 894	 * can not change and must all be valid, such as recvmsg from
 895	 * userspace.
 896	 */
 897	return sk->sk_socket->file->f_cred->user_ns;
 898}
 899
 900/* Sock flags */
 901enum sock_flags {
 902	SOCK_DEAD,
 903	SOCK_DONE,
 904	SOCK_URGINLINE,
 905	SOCK_KEEPOPEN,
 906	SOCK_LINGER,
 907	SOCK_DESTROY,
 908	SOCK_BROADCAST,
 909	SOCK_TIMESTAMP,
 910	SOCK_ZAPPED,
 911	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
 912	SOCK_DBG, /* %SO_DEBUG setting */
 913	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
 914	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
 915	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
 916	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
 917	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
 918	SOCK_FASYNC, /* fasync() active */
 919	SOCK_RXQ_OVFL,
 920	SOCK_ZEROCOPY, /* buffers from userspace */
 921	SOCK_WIFI_STATUS, /* push wifi status to userspace */
 922	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
 923		     * Will use last 4 bytes of packet sent from
 924		     * user-space instead.
 925		     */
 926	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
 927	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
 928	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
 929	SOCK_TXTIME,
 930	SOCK_XDP, /* XDP is attached */
 931	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
 932	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
 933};
 934
 935#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
 
 
 
 
 
 
 936
 937static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
 938{
 939	nsk->sk_flags = osk->sk_flags;
 940}
 941
 942static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
 943{
 944	__set_bit(flag, &sk->sk_flags);
 945}
 946
 947static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
 948{
 949	__clear_bit(flag, &sk->sk_flags);
 950}
 951
 952static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
 953				     int valbool)
 954{
 955	if (valbool)
 956		sock_set_flag(sk, bit);
 957	else
 958		sock_reset_flag(sk, bit);
 959}
 960
 961static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
 962{
 963	return test_bit(flag, &sk->sk_flags);
 964}
 965
 966#ifdef CONFIG_NET
 967DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
 968static inline int sk_memalloc_socks(void)
 969{
 970	return static_branch_unlikely(&memalloc_socks_key);
 971}
 972
 973void __receive_sock(struct file *file);
 974#else
 975
 976static inline int sk_memalloc_socks(void)
 977{
 978	return 0;
 979}
 980
 981static inline void __receive_sock(struct file *file)
 982{ }
 983#endif
 984
 985static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
 986{
 987	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
 988}
 989
 990static inline void sk_acceptq_removed(struct sock *sk)
 991{
 992	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
 993}
 994
 995static inline void sk_acceptq_added(struct sock *sk)
 996{
 997	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
 998}
 999
1000/* Note: If you think the test should be:
1001 *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1002 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1003 */
1004static inline bool sk_acceptq_is_full(const struct sock *sk)
1005{
1006	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1007}
1008
1009/*
1010 * Compute minimal free write space needed to queue new packets.
1011 */
1012static inline int sk_stream_min_wspace(const struct sock *sk)
1013{
1014	return READ_ONCE(sk->sk_wmem_queued) >> 1;
1015}
1016
1017static inline int sk_stream_wspace(const struct sock *sk)
1018{
1019	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1020}
1021
1022static inline void sk_wmem_queued_add(struct sock *sk, int val)
1023{
1024	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1025}
1026
1027static inline void sk_forward_alloc_add(struct sock *sk, int val)
1028{
1029	/* Paired with lockless reads of sk->sk_forward_alloc */
1030	WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1031}
1032
1033void sk_stream_write_space(struct sock *sk);
1034
1035/* OOB backlog add */
1036static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1037{
1038	/* dont let skb dst not refcounted, we are going to leave rcu lock */
1039	skb_dst_force(skb);
1040
1041	if (!sk->sk_backlog.tail)
1042		WRITE_ONCE(sk->sk_backlog.head, skb);
1043	else
1044		sk->sk_backlog.tail->next = skb;
1045
1046	WRITE_ONCE(sk->sk_backlog.tail, skb);
1047	skb->next = NULL;
1048}
1049
1050/*
1051 * Take into account size of receive queue and backlog queue
1052 * Do not take into account this skb truesize,
1053 * to allow even a single big packet to come.
1054 */
1055static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1056{
1057	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1058
1059	return qsize > limit;
1060}
1061
1062/* The per-socket spinlock must be held here. */
1063static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1064					      unsigned int limit)
1065{
1066	if (sk_rcvqueues_full(sk, limit))
1067		return -ENOBUFS;
1068
1069	/*
1070	 * If the skb was allocated from pfmemalloc reserves, only
1071	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1072	 * helping free memory
1073	 */
1074	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1075		return -ENOMEM;
1076
1077	__sk_add_backlog(sk, skb);
1078	sk->sk_backlog.len += skb->truesize;
1079	return 0;
1080}
1081
1082int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1083
1084INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1085INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1086
1087static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1088{
1089	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1090		return __sk_backlog_rcv(sk, skb);
1091
1092	return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1093				  tcp_v6_do_rcv,
1094				  tcp_v4_do_rcv,
1095				  sk, skb);
1096}
1097
1098static inline void sk_incoming_cpu_update(struct sock *sk)
1099{
1100	int cpu = raw_smp_processor_id();
1101
1102	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1103		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1104}
1105
1106static inline void sock_rps_record_flow_hash(__u32 hash)
1107{
1108#ifdef CONFIG_RPS
1109	struct rps_sock_flow_table *sock_flow_table;
1110
1111	rcu_read_lock();
1112	sock_flow_table = rcu_dereference(rps_sock_flow_table);
1113	rps_record_sock_flow(sock_flow_table, hash);
1114	rcu_read_unlock();
1115#endif
1116}
1117
1118static inline void sock_rps_record_flow(const struct sock *sk)
1119{
1120#ifdef CONFIG_RPS
1121	if (static_branch_unlikely(&rfs_needed)) {
1122		/* Reading sk->sk_rxhash might incur an expensive cache line
1123		 * miss.
1124		 *
1125		 * TCP_ESTABLISHED does cover almost all states where RFS
1126		 * might be useful, and is cheaper [1] than testing :
1127		 *	IPv4: inet_sk(sk)->inet_daddr
1128		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1129		 * OR	an additional socket flag
1130		 * [1] : sk_state and sk_prot are in the same cache line.
1131		 */
1132		if (sk->sk_state == TCP_ESTABLISHED) {
1133			/* This READ_ONCE() is paired with the WRITE_ONCE()
1134			 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
1135			 */
1136			sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1137		}
1138	}
1139#endif
1140}
1141
1142static inline void sock_rps_save_rxhash(struct sock *sk,
1143					const struct sk_buff *skb)
1144{
1145#ifdef CONFIG_RPS
1146	/* The following WRITE_ONCE() is paired with the READ_ONCE()
1147	 * here, and another one in sock_rps_record_flow().
1148	 */
1149	if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1150		WRITE_ONCE(sk->sk_rxhash, skb->hash);
1151#endif
1152}
1153
1154static inline void sock_rps_reset_rxhash(struct sock *sk)
1155{
1156#ifdef CONFIG_RPS
1157	/* Paired with READ_ONCE() in sock_rps_record_flow() */
1158	WRITE_ONCE(sk->sk_rxhash, 0);
1159#endif
1160}
1161
1162#define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1163	({	int __rc, __dis = __sk->sk_disconnects;			\
1164		release_sock(__sk);					\
1165		__rc = __condition;					\
1166		if (!__rc) {						\
1167			*(__timeo) = wait_woken(__wait,			\
1168						TASK_INTERRUPTIBLE,	\
1169						*(__timeo));		\
1170		}							\
1171		sched_annotate_sleep();					\
1172		lock_sock(__sk);					\
1173		__rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1174		__rc;							\
1175	})
1176
1177int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1178int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1179void sk_stream_wait_close(struct sock *sk, long timeo_p);
1180int sk_stream_error(struct sock *sk, int flags, int err);
1181void sk_stream_kill_queues(struct sock *sk);
1182void sk_set_memalloc(struct sock *sk);
1183void sk_clear_memalloc(struct sock *sk);
1184
1185void __sk_flush_backlog(struct sock *sk);
1186
1187static inline bool sk_flush_backlog(struct sock *sk)
1188{
1189	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1190		__sk_flush_backlog(sk);
1191		return true;
1192	}
1193	return false;
1194}
1195
1196int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1197
1198struct request_sock_ops;
1199struct timewait_sock_ops;
1200struct inet_hashinfo;
1201struct raw_hashinfo;
1202struct smc_hashinfo;
1203struct module;
1204struct sk_psock;
1205
1206/*
1207 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1208 * un-modified. Special care is taken when initializing object to zero.
1209 */
1210static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1211{
1212	if (offsetof(struct sock, sk_node.next) != 0)
1213		memset(sk, 0, offsetof(struct sock, sk_node.next));
1214	memset(&sk->sk_node.pprev, 0,
1215	       size - offsetof(struct sock, sk_node.pprev));
1216}
1217
 
 
 
 
 
 
 
1218/* Networking protocol blocks we attach to sockets.
1219 * socket layer -> transport layer interface
1220 */
1221struct proto {
1222	void			(*close)(struct sock *sk,
1223					long timeout);
1224	int			(*pre_connect)(struct sock *sk,
1225					struct sockaddr *uaddr,
1226					int addr_len);
1227	int			(*connect)(struct sock *sk,
1228					struct sockaddr *uaddr,
1229					int addr_len);
1230	int			(*disconnect)(struct sock *sk, int flags);
1231
1232	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1233					  bool kern);
1234
1235	int			(*ioctl)(struct sock *sk, int cmd,
1236					 int *karg);
1237	int			(*init)(struct sock *sk);
1238	void			(*destroy)(struct sock *sk);
1239	void			(*shutdown)(struct sock *sk, int how);
1240	int			(*setsockopt)(struct sock *sk, int level,
1241					int optname, sockptr_t optval,
1242					unsigned int optlen);
1243	int			(*getsockopt)(struct sock *sk, int level,
1244					int optname, char __user *optval,
1245					int __user *option);
1246	void			(*keepalive)(struct sock *sk, int valbool);
1247#ifdef CONFIG_COMPAT
1248	int			(*compat_ioctl)(struct sock *sk,
1249					unsigned int cmd, unsigned long arg);
1250#endif
1251	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1252					   size_t len);
1253	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1254					   size_t len, int flags, int *addr_len);
1255	void			(*splice_eof)(struct socket *sock);
1256	int			(*bind)(struct sock *sk,
1257					struct sockaddr *addr, int addr_len);
1258	int			(*bind_add)(struct sock *sk,
1259					struct sockaddr *addr, int addr_len);
1260
1261	int			(*backlog_rcv) (struct sock *sk,
1262						struct sk_buff *skb);
1263	bool			(*bpf_bypass_getsockopt)(int level,
1264							 int optname);
1265
1266	void		(*release_cb)(struct sock *sk);
1267
1268	/* Keeping track of sk's, looking them up, and port selection methods. */
1269	int			(*hash)(struct sock *sk);
1270	void			(*unhash)(struct sock *sk);
1271	void			(*rehash)(struct sock *sk);
1272	int			(*get_port)(struct sock *sk, unsigned short snum);
1273	void			(*put_port)(struct sock *sk);
1274#ifdef CONFIG_BPF_SYSCALL
1275	int			(*psock_update_sk_prot)(struct sock *sk,
1276							struct sk_psock *psock,
1277							bool restore);
1278#endif
1279
1280	/* Keeping track of sockets in use */
1281#ifdef CONFIG_PROC_FS
1282	unsigned int		inuse_idx;
1283#endif
1284
1285#if IS_ENABLED(CONFIG_MPTCP)
1286	int			(*forward_alloc_get)(const struct sock *sk);
1287#endif
1288
1289	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1290	bool			(*sock_is_readable)(struct sock *sk);
1291	/* Memory pressure */
1292	void			(*enter_memory_pressure)(struct sock *sk);
1293	void			(*leave_memory_pressure)(struct sock *sk);
1294	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1295	int  __percpu		*per_cpu_fw_alloc;
1296	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1297
1298	/*
1299	 * Pressure flag: try to collapse.
1300	 * Technical note: it is used by multiple contexts non atomically.
1301	 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1302	 * All the __sk_mem_schedule() is of this nature: accounting
1303	 * is strict, actions are advisory and have some latency.
1304	 */
1305	unsigned long		*memory_pressure;
1306	long			*sysctl_mem;
1307
1308	int			*sysctl_wmem;
1309	int			*sysctl_rmem;
1310	u32			sysctl_wmem_offset;
1311	u32			sysctl_rmem_offset;
1312
1313	int			max_header;
1314	bool			no_autobind;
1315
1316	struct kmem_cache	*slab;
1317	unsigned int		obj_size;
1318	unsigned int		ipv6_pinfo_offset;
1319	slab_flags_t		slab_flags;
1320	unsigned int		useroffset;	/* Usercopy region offset */
1321	unsigned int		usersize;	/* Usercopy region size */
1322
1323	unsigned int __percpu	*orphan_count;
1324
1325	struct request_sock_ops	*rsk_prot;
1326	struct timewait_sock_ops *twsk_prot;
1327
1328	union {
1329		struct inet_hashinfo	*hashinfo;
1330		struct udp_table	*udp_table;
1331		struct raw_hashinfo	*raw_hash;
1332		struct smc_hashinfo	*smc_hash;
1333	} h;
1334
1335	struct module		*owner;
1336
1337	char			name[32];
1338
1339	struct list_head	node;
1340	int			(*diag_destroy)(struct sock *sk, int err);
1341} __randomize_layout;
1342
1343int proto_register(struct proto *prot, int alloc_slab);
1344void proto_unregister(struct proto *prot);
1345int sock_load_diag_module(int family, int protocol);
1346
1347INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1348
1349static inline int sk_forward_alloc_get(const struct sock *sk)
1350{
1351#if IS_ENABLED(CONFIG_MPTCP)
1352	if (sk->sk_prot->forward_alloc_get)
1353		return sk->sk_prot->forward_alloc_get(sk);
1354#endif
1355	return READ_ONCE(sk->sk_forward_alloc);
1356}
1357
1358static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1359{
1360	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1361		return false;
1362
1363	return sk->sk_prot->stream_memory_free ?
1364		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1365				     tcp_stream_memory_free, sk, wake) : true;
1366}
1367
1368static inline bool sk_stream_memory_free(const struct sock *sk)
1369{
1370	return __sk_stream_memory_free(sk, 0);
1371}
1372
1373static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1374{
1375	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1376	       __sk_stream_memory_free(sk, wake);
1377}
1378
1379static inline bool sk_stream_is_writeable(const struct sock *sk)
1380{
1381	return __sk_stream_is_writeable(sk, 0);
1382}
1383
1384static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1385					    struct cgroup *ancestor)
1386{
1387#ifdef CONFIG_SOCK_CGROUP_DATA
1388	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1389				    ancestor);
1390#else
1391	return -ENOTSUPP;
1392#endif
1393}
1394
1395static inline bool sk_has_memory_pressure(const struct sock *sk)
1396{
1397	return sk->sk_prot->memory_pressure != NULL;
1398}
1399
1400static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1401{
1402	return sk->sk_prot->memory_pressure &&
1403		!!READ_ONCE(*sk->sk_prot->memory_pressure);
1404}
1405
1406static inline bool sk_under_memory_pressure(const struct sock *sk)
1407{
1408	if (!sk->sk_prot->memory_pressure)
1409		return false;
1410
1411	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1412	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1413		return true;
1414
1415	return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1416}
1417
1418static inline long
1419proto_memory_allocated(const struct proto *prot)
1420{
1421	return max(0L, atomic_long_read(prot->memory_allocated));
1422}
1423
1424static inline long
1425sk_memory_allocated(const struct sock *sk)
1426{
1427	return proto_memory_allocated(sk->sk_prot);
1428}
1429
1430/* 1 MB per cpu, in page units */
1431#define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT))
1432
1433static inline void
1434sk_memory_allocated_add(struct sock *sk, int amt)
1435{
1436	int local_reserve;
1437
1438	preempt_disable();
1439	local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1440	if (local_reserve >= SK_MEMORY_PCPU_RESERVE) {
1441		__this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1442		atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1443	}
1444	preempt_enable();
1445}
1446
1447static inline void
1448sk_memory_allocated_sub(struct sock *sk, int amt)
1449{
1450	int local_reserve;
1451
1452	preempt_disable();
1453	local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1454	if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) {
1455		__this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1456		atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1457	}
1458	preempt_enable();
1459}
1460
1461#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1462
1463static inline void sk_sockets_allocated_dec(struct sock *sk)
1464{
1465	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1466				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1467}
1468
1469static inline void sk_sockets_allocated_inc(struct sock *sk)
1470{
1471	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1472				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1473}
1474
1475static inline u64
1476sk_sockets_allocated_read_positive(struct sock *sk)
1477{
1478	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1479}
1480
1481static inline int
1482proto_sockets_allocated_sum_positive(struct proto *prot)
1483{
1484	return percpu_counter_sum_positive(prot->sockets_allocated);
1485}
1486
1487static inline bool
1488proto_memory_pressure(struct proto *prot)
1489{
1490	if (!prot->memory_pressure)
1491		return false;
1492	return !!READ_ONCE(*prot->memory_pressure);
1493}
1494
1495
1496#ifdef CONFIG_PROC_FS
1497#define PROTO_INUSE_NR	64	/* should be enough for the first time */
1498struct prot_inuse {
1499	int all;
1500	int val[PROTO_INUSE_NR];
1501};
1502
1503static inline void sock_prot_inuse_add(const struct net *net,
1504				       const struct proto *prot, int val)
1505{
1506	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1507}
1508
1509static inline void sock_inuse_add(const struct net *net, int val)
1510{
1511	this_cpu_add(net->core.prot_inuse->all, val);
1512}
1513
1514int sock_prot_inuse_get(struct net *net, struct proto *proto);
1515int sock_inuse_get(struct net *net);
1516#else
1517static inline void sock_prot_inuse_add(const struct net *net,
1518				       const struct proto *prot, int val)
1519{
1520}
1521
1522static inline void sock_inuse_add(const struct net *net, int val)
1523{
1524}
1525#endif
1526
1527
1528/* With per-bucket locks this operation is not-atomic, so that
1529 * this version is not worse.
1530 */
1531static inline int __sk_prot_rehash(struct sock *sk)
1532{
1533	sk->sk_prot->unhash(sk);
1534	return sk->sk_prot->hash(sk);
1535}
1536
1537/* About 10 seconds */
1538#define SOCK_DESTROY_TIME (10*HZ)
1539
1540/* Sockets 0-1023 can't be bound to unless you are superuser */
1541#define PROT_SOCK	1024
1542
1543#define SHUTDOWN_MASK	3
1544#define RCV_SHUTDOWN	1
1545#define SEND_SHUTDOWN	2
1546
1547#define SOCK_BINDADDR_LOCK	4
1548#define SOCK_BINDPORT_LOCK	8
1549
1550struct socket_alloc {
1551	struct socket socket;
1552	struct inode vfs_inode;
1553};
1554
1555static inline struct socket *SOCKET_I(struct inode *inode)
1556{
1557	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1558}
1559
1560static inline struct inode *SOCK_INODE(struct socket *socket)
1561{
1562	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1563}
1564
1565/*
1566 * Functions for memory accounting
1567 */
1568int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1569int __sk_mem_schedule(struct sock *sk, int size, int kind);
1570void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1571void __sk_mem_reclaim(struct sock *sk, int amount);
1572
1573#define SK_MEM_SEND	0
1574#define SK_MEM_RECV	1
1575
1576/* sysctl_mem values are in pages */
1577static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1578{
1579	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1580}
1581
1582static inline int sk_mem_pages(int amt)
1583{
1584	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1585}
1586
1587static inline bool sk_has_account(struct sock *sk)
1588{
1589	/* return true if protocol supports memory accounting */
1590	return !!sk->sk_prot->memory_allocated;
1591}
1592
1593static inline bool sk_wmem_schedule(struct sock *sk, int size)
1594{
1595	int delta;
1596
1597	if (!sk_has_account(sk))
1598		return true;
1599	delta = size - sk->sk_forward_alloc;
1600	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1601}
1602
1603static inline bool
1604sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1605{
1606	int delta;
1607
1608	if (!sk_has_account(sk))
1609		return true;
1610	delta = size - sk->sk_forward_alloc;
1611	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1612		skb_pfmemalloc(skb);
 
 
 
 
 
 
1613}
1614
1615static inline int sk_unused_reserved_mem(const struct sock *sk)
1616{
1617	int unused_mem;
1618
1619	if (likely(!sk->sk_reserved_mem))
1620		return 0;
1621
1622	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1623			atomic_read(&sk->sk_rmem_alloc);
1624
1625	return unused_mem > 0 ? unused_mem : 0;
1626}
1627
1628static inline void sk_mem_reclaim(struct sock *sk)
1629{
1630	int reclaimable;
1631
1632	if (!sk_has_account(sk))
1633		return;
1634
1635	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1636
1637	if (reclaimable >= (int)PAGE_SIZE)
1638		__sk_mem_reclaim(sk, reclaimable);
1639}
1640
1641static inline void sk_mem_reclaim_final(struct sock *sk)
1642{
1643	sk->sk_reserved_mem = 0;
1644	sk_mem_reclaim(sk);
1645}
1646
1647static inline void sk_mem_charge(struct sock *sk, int size)
1648{
1649	if (!sk_has_account(sk))
1650		return;
1651	sk_forward_alloc_add(sk, -size);
1652}
1653
1654static inline void sk_mem_uncharge(struct sock *sk, int size)
1655{
1656	if (!sk_has_account(sk))
1657		return;
1658	sk_forward_alloc_add(sk, size);
1659	sk_mem_reclaim(sk);
1660}
1661
1662/*
1663 * Macro so as to not evaluate some arguments when
1664 * lockdep is not enabled.
1665 *
1666 * Mark both the sk_lock and the sk_lock.slock as a
1667 * per-address-family lock class.
1668 */
1669#define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1670do {									\
1671	sk->sk_lock.owned = 0;						\
1672	init_waitqueue_head(&sk->sk_lock.wq);				\
1673	spin_lock_init(&(sk)->sk_lock.slock);				\
1674	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1675			sizeof((sk)->sk_lock));				\
1676	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1677				(skey), (sname));				\
1678	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1679} while (0)
1680
1681static inline bool lockdep_sock_is_held(const struct sock *sk)
1682{
1683	return lockdep_is_held(&sk->sk_lock) ||
1684	       lockdep_is_held(&sk->sk_lock.slock);
1685}
1686
1687void lock_sock_nested(struct sock *sk, int subclass);
1688
1689static inline void lock_sock(struct sock *sk)
1690{
1691	lock_sock_nested(sk, 0);
1692}
1693
1694void __lock_sock(struct sock *sk);
1695void __release_sock(struct sock *sk);
1696void release_sock(struct sock *sk);
1697
1698/* BH context may only use the following locking interface. */
1699#define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1700#define bh_lock_sock_nested(__sk) \
1701				spin_lock_nested(&((__sk)->sk_lock.slock), \
1702				SINGLE_DEPTH_NESTING)
1703#define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1704
1705bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1706
1707/**
1708 * lock_sock_fast - fast version of lock_sock
1709 * @sk: socket
1710 *
1711 * This version should be used for very small section, where process wont block
1712 * return false if fast path is taken:
1713 *
1714 *   sk_lock.slock locked, owned = 0, BH disabled
1715 *
1716 * return true if slow path is taken:
1717 *
1718 *   sk_lock.slock unlocked, owned = 1, BH enabled
1719 */
1720static inline bool lock_sock_fast(struct sock *sk)
1721{
1722	/* The sk_lock has mutex_lock() semantics here. */
1723	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1724
1725	return __lock_sock_fast(sk);
1726}
1727
1728/* fast socket lock variant for caller already holding a [different] socket lock */
1729static inline bool lock_sock_fast_nested(struct sock *sk)
1730{
1731	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1732
1733	return __lock_sock_fast(sk);
1734}
1735
1736/**
1737 * unlock_sock_fast - complement of lock_sock_fast
1738 * @sk: socket
1739 * @slow: slow mode
1740 *
1741 * fast unlock socket for user context.
1742 * If slow mode is on, we call regular release_sock()
1743 */
1744static inline void unlock_sock_fast(struct sock *sk, bool slow)
1745	__releases(&sk->sk_lock.slock)
1746{
1747	if (slow) {
1748		release_sock(sk);
1749		__release(&sk->sk_lock.slock);
1750	} else {
1751		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1752		spin_unlock_bh(&sk->sk_lock.slock);
1753	}
1754}
1755
1756void sockopt_lock_sock(struct sock *sk);
1757void sockopt_release_sock(struct sock *sk);
1758bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1759bool sockopt_capable(int cap);
1760
1761/* Used by processes to "lock" a socket state, so that
1762 * interrupts and bottom half handlers won't change it
1763 * from under us. It essentially blocks any incoming
1764 * packets, so that we won't get any new data or any
1765 * packets that change the state of the socket.
1766 *
1767 * While locked, BH processing will add new packets to
1768 * the backlog queue.  This queue is processed by the
1769 * owner of the socket lock right before it is released.
1770 *
1771 * Since ~2.3.5 it is also exclusive sleep lock serializing
1772 * accesses from user process context.
1773 */
1774
1775static inline void sock_owned_by_me(const struct sock *sk)
1776{
1777#ifdef CONFIG_LOCKDEP
1778	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1779#endif
1780}
1781
 
 
 
 
 
 
 
1782static inline bool sock_owned_by_user(const struct sock *sk)
1783{
1784	sock_owned_by_me(sk);
1785	return sk->sk_lock.owned;
1786}
1787
1788static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1789{
1790	return sk->sk_lock.owned;
1791}
1792
1793static inline void sock_release_ownership(struct sock *sk)
1794{
1795	DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1796	sk->sk_lock.owned = 0;
1797
1798	/* The sk_lock has mutex_unlock() semantics: */
1799	mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1800}
1801
1802/* no reclassification while locks are held */
1803static inline bool sock_allow_reclassification(const struct sock *csk)
1804{
1805	struct sock *sk = (struct sock *)csk;
1806
1807	return !sock_owned_by_user_nocheck(sk) &&
1808		!spin_is_locked(&sk->sk_lock.slock);
1809}
1810
1811struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1812		      struct proto *prot, int kern);
1813void sk_free(struct sock *sk);
 
1814void sk_destruct(struct sock *sk);
1815struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1816void sk_free_unlock_clone(struct sock *sk);
1817
1818struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1819			     gfp_t priority);
1820void __sock_wfree(struct sk_buff *skb);
1821void sock_wfree(struct sk_buff *skb);
1822struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1823			     gfp_t priority);
1824void skb_orphan_partial(struct sk_buff *skb);
1825void sock_rfree(struct sk_buff *skb);
1826void sock_efree(struct sk_buff *skb);
1827#ifdef CONFIG_INET
1828void sock_edemux(struct sk_buff *skb);
1829void sock_pfree(struct sk_buff *skb);
 
 
 
 
 
 
 
 
 
1830#else
1831#define sock_edemux sock_efree
1832#endif
1833
1834int sk_setsockopt(struct sock *sk, int level, int optname,
1835		  sockptr_t optval, unsigned int optlen);
1836int sock_setsockopt(struct socket *sock, int level, int op,
1837		    sockptr_t optval, unsigned int optlen);
1838int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1839		       int optname, sockptr_t optval, int optlen);
1840int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1841		       int optname, sockptr_t optval, sockptr_t optlen);
1842
1843int sk_getsockopt(struct sock *sk, int level, int optname,
1844		  sockptr_t optval, sockptr_t optlen);
1845int sock_gettstamp(struct socket *sock, void __user *userstamp,
1846		   bool timeval, bool time32);
1847struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1848				     unsigned long data_len, int noblock,
1849				     int *errcode, int max_page_order);
1850
1851static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1852						  unsigned long size,
1853						  int noblock, int *errcode)
1854{
1855	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1856}
1857
1858void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1859void sock_kfree_s(struct sock *sk, void *mem, int size);
1860void sock_kzfree_s(struct sock *sk, void *mem, int size);
1861void sk_send_sigurg(struct sock *sk);
1862
1863static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1864{
1865	if (sk->sk_socket)
1866		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1867	WRITE_ONCE(sk->sk_prot, proto);
1868}
1869
1870struct sockcm_cookie {
1871	u64 transmit_time;
1872	u32 mark;
1873	u32 tsflags;
 
1874};
1875
1876static inline void sockcm_init(struct sockcm_cookie *sockc,
1877			       const struct sock *sk)
1878{
1879	*sockc = (struct sockcm_cookie) {
1880		.tsflags = READ_ONCE(sk->sk_tsflags)
1881	};
1882}
1883
1884int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1885		     struct sockcm_cookie *sockc);
1886int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1887		   struct sockcm_cookie *sockc);
1888
1889/*
1890 * Functions to fill in entries in struct proto_ops when a protocol
1891 * does not implement a particular function.
1892 */
1893int sock_no_bind(struct socket *, struct sockaddr *, int);
1894int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1895int sock_no_socketpair(struct socket *, struct socket *);
1896int sock_no_accept(struct socket *, struct socket *, int, bool);
1897int sock_no_getname(struct socket *, struct sockaddr *, int);
1898int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1899int sock_no_listen(struct socket *, int);
1900int sock_no_shutdown(struct socket *, int);
1901int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1902int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1903int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1904int sock_no_mmap(struct file *file, struct socket *sock,
1905		 struct vm_area_struct *vma);
1906
1907/*
1908 * Functions to fill in entries in struct proto_ops when a protocol
1909 * uses the inet style.
1910 */
1911int sock_common_getsockopt(struct socket *sock, int level, int optname,
1912				  char __user *optval, int __user *optlen);
1913int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1914			int flags);
1915int sock_common_setsockopt(struct socket *sock, int level, int optname,
1916			   sockptr_t optval, unsigned int optlen);
1917
1918void sk_common_release(struct sock *sk);
1919
1920/*
1921 *	Default socket callbacks and setup code
1922 */
1923
1924/* Initialise core socket variables using an explicit uid. */
1925void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1926
1927/* Initialise core socket variables.
1928 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1929 */
1930void sock_init_data(struct socket *sock, struct sock *sk);
1931
1932/*
1933 * Socket reference counting postulates.
1934 *
1935 * * Each user of socket SHOULD hold a reference count.
1936 * * Each access point to socket (an hash table bucket, reference from a list,
1937 *   running timer, skb in flight MUST hold a reference count.
1938 * * When reference count hits 0, it means it will never increase back.
1939 * * When reference count hits 0, it means that no references from
1940 *   outside exist to this socket and current process on current CPU
1941 *   is last user and may/should destroy this socket.
1942 * * sk_free is called from any context: process, BH, IRQ. When
1943 *   it is called, socket has no references from outside -> sk_free
1944 *   may release descendant resources allocated by the socket, but
1945 *   to the time when it is called, socket is NOT referenced by any
1946 *   hash tables, lists etc.
1947 * * Packets, delivered from outside (from network or from another process)
1948 *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1949 *   when they sit in queue. Otherwise, packets will leak to hole, when
1950 *   socket is looked up by one cpu and unhasing is made by another CPU.
1951 *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1952 *   (leak to backlog). Packet socket does all the processing inside
1953 *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1954 *   use separate SMP lock, so that they are prone too.
1955 */
1956
1957/* Ungrab socket and destroy it, if it was the last reference. */
1958static inline void sock_put(struct sock *sk)
1959{
1960	if (refcount_dec_and_test(&sk->sk_refcnt))
1961		sk_free(sk);
1962}
1963/* Generic version of sock_put(), dealing with all sockets
1964 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1965 */
1966void sock_gen_put(struct sock *sk);
1967
1968int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1969		     unsigned int trim_cap, bool refcounted);
1970static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1971				 const int nested)
1972{
1973	return __sk_receive_skb(sk, skb, nested, 1, true);
1974}
1975
1976static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1977{
1978	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1979	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1980		return;
1981	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1982	 * other WRITE_ONCE() because socket lock might be not held.
1983	 */
1984	WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1985}
1986
1987#define NO_QUEUE_MAPPING	USHRT_MAX
1988
1989static inline void sk_tx_queue_clear(struct sock *sk)
1990{
1991	/* Paired with READ_ONCE() in sk_tx_queue_get() and
1992	 * other WRITE_ONCE() because socket lock might be not held.
1993	 */
1994	WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1995}
1996
1997static inline int sk_tx_queue_get(const struct sock *sk)
1998{
1999	if (sk) {
2000		/* Paired with WRITE_ONCE() in sk_tx_queue_clear()
2001		 * and sk_tx_queue_set().
2002		 */
2003		int val = READ_ONCE(sk->sk_tx_queue_mapping);
2004
2005		if (val != NO_QUEUE_MAPPING)
2006			return val;
2007	}
2008	return -1;
2009}
2010
2011static inline void __sk_rx_queue_set(struct sock *sk,
2012				     const struct sk_buff *skb,
2013				     bool force_set)
2014{
2015#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2016	if (skb_rx_queue_recorded(skb)) {
2017		u16 rx_queue = skb_get_rx_queue(skb);
2018
2019		if (force_set ||
2020		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2021			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2022	}
2023#endif
2024}
2025
2026static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2027{
2028	__sk_rx_queue_set(sk, skb, true);
2029}
2030
2031static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2032{
2033	__sk_rx_queue_set(sk, skb, false);
2034}
2035
2036static inline void sk_rx_queue_clear(struct sock *sk)
2037{
2038#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2039	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2040#endif
2041}
2042
2043static inline int sk_rx_queue_get(const struct sock *sk)
2044{
2045#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2046	if (sk) {
2047		int res = READ_ONCE(sk->sk_rx_queue_mapping);
2048
2049		if (res != NO_QUEUE_MAPPING)
2050			return res;
2051	}
2052#endif
2053
2054	return -1;
2055}
2056
2057static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2058{
2059	sk->sk_socket = sock;
2060}
2061
2062static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2063{
2064	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2065	return &rcu_dereference_raw(sk->sk_wq)->wait;
2066}
2067/* Detach socket from process context.
2068 * Announce socket dead, detach it from wait queue and inode.
2069 * Note that parent inode held reference count on this struct sock,
2070 * we do not release it in this function, because protocol
2071 * probably wants some additional cleanups or even continuing
2072 * to work with this socket (TCP).
2073 */
2074static inline void sock_orphan(struct sock *sk)
2075{
2076	write_lock_bh(&sk->sk_callback_lock);
2077	sock_set_flag(sk, SOCK_DEAD);
2078	sk_set_socket(sk, NULL);
2079	sk->sk_wq  = NULL;
2080	write_unlock_bh(&sk->sk_callback_lock);
2081}
2082
2083static inline void sock_graft(struct sock *sk, struct socket *parent)
2084{
2085	WARN_ON(parent->sk);
2086	write_lock_bh(&sk->sk_callback_lock);
2087	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2088	parent->sk = sk;
2089	sk_set_socket(sk, parent);
2090	sk->sk_uid = SOCK_INODE(parent)->i_uid;
2091	security_sock_graft(sk, parent);
2092	write_unlock_bh(&sk->sk_callback_lock);
2093}
2094
2095kuid_t sock_i_uid(struct sock *sk);
2096unsigned long __sock_i_ino(struct sock *sk);
2097unsigned long sock_i_ino(struct sock *sk);
2098
2099static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2100{
2101	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2102}
2103
2104static inline u32 net_tx_rndhash(void)
2105{
2106	u32 v = get_random_u32();
2107
2108	return v ?: 1;
2109}
2110
2111static inline void sk_set_txhash(struct sock *sk)
2112{
2113	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2114	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2115}
2116
2117static inline bool sk_rethink_txhash(struct sock *sk)
2118{
2119	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2120		sk_set_txhash(sk);
2121		return true;
2122	}
2123	return false;
2124}
2125
2126static inline struct dst_entry *
2127__sk_dst_get(const struct sock *sk)
2128{
2129	return rcu_dereference_check(sk->sk_dst_cache,
2130				     lockdep_sock_is_held(sk));
2131}
2132
2133static inline struct dst_entry *
2134sk_dst_get(const struct sock *sk)
2135{
2136	struct dst_entry *dst;
2137
2138	rcu_read_lock();
2139	dst = rcu_dereference(sk->sk_dst_cache);
2140	if (dst && !rcuref_get(&dst->__rcuref))
2141		dst = NULL;
2142	rcu_read_unlock();
2143	return dst;
2144}
2145
2146static inline void __dst_negative_advice(struct sock *sk)
2147{
2148	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2149
2150	if (dst && dst->ops->negative_advice) {
2151		ndst = dst->ops->negative_advice(dst);
2152
2153		if (ndst != dst) {
2154			rcu_assign_pointer(sk->sk_dst_cache, ndst);
2155			sk_tx_queue_clear(sk);
2156			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2157		}
2158	}
2159}
2160
2161static inline void dst_negative_advice(struct sock *sk)
2162{
2163	sk_rethink_txhash(sk);
2164	__dst_negative_advice(sk);
2165}
2166
2167static inline void
2168__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2169{
2170	struct dst_entry *old_dst;
2171
2172	sk_tx_queue_clear(sk);
2173	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2174	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2175					    lockdep_sock_is_held(sk));
2176	rcu_assign_pointer(sk->sk_dst_cache, dst);
2177	dst_release(old_dst);
2178}
2179
2180static inline void
2181sk_dst_set(struct sock *sk, struct dst_entry *dst)
2182{
2183	struct dst_entry *old_dst;
2184
2185	sk_tx_queue_clear(sk);
2186	WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2187	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2188	dst_release(old_dst);
2189}
2190
2191static inline void
2192__sk_dst_reset(struct sock *sk)
2193{
2194	__sk_dst_set(sk, NULL);
2195}
2196
2197static inline void
2198sk_dst_reset(struct sock *sk)
2199{
2200	sk_dst_set(sk, NULL);
2201}
2202
2203struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2204
2205struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2206
2207static inline void sk_dst_confirm(struct sock *sk)
2208{
2209	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2210		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2211}
2212
2213static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2214{
2215	if (skb_get_dst_pending_confirm(skb)) {
2216		struct sock *sk = skb->sk;
2217
2218		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2219			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2220		neigh_confirm(n);
2221	}
2222}
2223
2224bool sk_mc_loop(const struct sock *sk);
2225
2226static inline bool sk_can_gso(const struct sock *sk)
2227{
2228	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2229}
2230
2231void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2232
2233static inline void sk_gso_disable(struct sock *sk)
2234{
2235	sk->sk_gso_disabled = 1;
2236	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2237}
2238
2239static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2240					   struct iov_iter *from, char *to,
2241					   int copy, int offset)
2242{
2243	if (skb->ip_summed == CHECKSUM_NONE) {
2244		__wsum csum = 0;
2245		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2246			return -EFAULT;
2247		skb->csum = csum_block_add(skb->csum, csum, offset);
2248	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2249		if (!copy_from_iter_full_nocache(to, copy, from))
2250			return -EFAULT;
2251	} else if (!copy_from_iter_full(to, copy, from))
2252		return -EFAULT;
2253
2254	return 0;
2255}
2256
2257static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2258				       struct iov_iter *from, int copy)
2259{
2260	int err, offset = skb->len;
2261
2262	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2263				       copy, offset);
2264	if (err)
2265		__skb_trim(skb, offset);
2266
2267	return err;
2268}
2269
2270static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2271					   struct sk_buff *skb,
2272					   struct page *page,
2273					   int off, int copy)
2274{
2275	int err;
2276
2277	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2278				       copy, skb->len);
2279	if (err)
2280		return err;
2281
2282	skb_len_add(skb, copy);
2283	sk_wmem_queued_add(sk, copy);
2284	sk_mem_charge(sk, copy);
2285	return 0;
2286}
2287
2288/**
2289 * sk_wmem_alloc_get - returns write allocations
2290 * @sk: socket
2291 *
2292 * Return: sk_wmem_alloc minus initial offset of one
2293 */
2294static inline int sk_wmem_alloc_get(const struct sock *sk)
2295{
2296	return refcount_read(&sk->sk_wmem_alloc) - 1;
2297}
2298
2299/**
2300 * sk_rmem_alloc_get - returns read allocations
2301 * @sk: socket
2302 *
2303 * Return: sk_rmem_alloc
2304 */
2305static inline int sk_rmem_alloc_get(const struct sock *sk)
2306{
2307	return atomic_read(&sk->sk_rmem_alloc);
2308}
2309
2310/**
2311 * sk_has_allocations - check if allocations are outstanding
2312 * @sk: socket
2313 *
2314 * Return: true if socket has write or read allocations
2315 */
2316static inline bool sk_has_allocations(const struct sock *sk)
2317{
2318	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2319}
2320
2321/**
2322 * skwq_has_sleeper - check if there are any waiting processes
2323 * @wq: struct socket_wq
2324 *
2325 * Return: true if socket_wq has waiting processes
2326 *
2327 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2328 * barrier call. They were added due to the race found within the tcp code.
2329 *
2330 * Consider following tcp code paths::
2331 *
2332 *   CPU1                CPU2
2333 *   sys_select          receive packet
2334 *   ...                 ...
2335 *   __add_wait_queue    update tp->rcv_nxt
2336 *   ...                 ...
2337 *   tp->rcv_nxt check   sock_def_readable
2338 *   ...                 {
2339 *   schedule               rcu_read_lock();
2340 *                          wq = rcu_dereference(sk->sk_wq);
2341 *                          if (wq && waitqueue_active(&wq->wait))
2342 *                              wake_up_interruptible(&wq->wait)
2343 *                          ...
2344 *                       }
2345 *
2346 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2347 * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2348 * could then endup calling schedule and sleep forever if there are no more
2349 * data on the socket.
2350 *
2351 */
2352static inline bool skwq_has_sleeper(struct socket_wq *wq)
2353{
2354	return wq && wq_has_sleeper(&wq->wait);
2355}
2356
2357/**
2358 * sock_poll_wait - place memory barrier behind the poll_wait call.
2359 * @filp:           file
2360 * @sock:           socket to wait on
2361 * @p:              poll_table
2362 *
2363 * See the comments in the wq_has_sleeper function.
2364 */
2365static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2366				  poll_table *p)
2367{
2368	if (!poll_does_not_wait(p)) {
2369		poll_wait(filp, &sock->wq.wait, p);
2370		/* We need to be sure we are in sync with the
2371		 * socket flags modification.
2372		 *
2373		 * This memory barrier is paired in the wq_has_sleeper.
2374		 */
2375		smp_mb();
2376	}
2377}
2378
2379static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2380{
2381	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2382	u32 txhash = READ_ONCE(sk->sk_txhash);
2383
2384	if (txhash) {
2385		skb->l4_hash = 1;
2386		skb->hash = txhash;
2387	}
2388}
2389
2390void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2391
2392/*
2393 *	Queue a received datagram if it will fit. Stream and sequenced
2394 *	protocols can't normally use this as they need to fit buffers in
2395 *	and play with them.
2396 *
2397 *	Inlined as it's very short and called for pretty much every
2398 *	packet ever received.
2399 */
2400static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2401{
2402	skb_orphan(skb);
2403	skb->sk = sk;
2404	skb->destructor = sock_rfree;
2405	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2406	sk_mem_charge(sk, skb->truesize);
2407}
2408
2409static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2410{
2411	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2412		skb_orphan(skb);
2413		skb->destructor = sock_efree;
2414		skb->sk = sk;
2415		return true;
2416	}
2417	return false;
2418}
2419
2420static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2421{
2422	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2423	if (skb) {
2424		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2425			skb_set_owner_r(skb, sk);
2426			return skb;
2427		}
2428		__kfree_skb(skb);
2429	}
2430	return NULL;
2431}
2432
2433static inline void skb_prepare_for_gro(struct sk_buff *skb)
2434{
2435	if (skb->destructor != sock_wfree) {
2436		skb_orphan(skb);
2437		return;
2438	}
2439	skb->slow_gro = 1;
2440}
2441
2442void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2443		    unsigned long expires);
2444
2445void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2446
2447void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2448
2449int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2450			struct sk_buff *skb, unsigned int flags,
2451			void (*destructor)(struct sock *sk,
2452					   struct sk_buff *skb));
2453int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2454
2455int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2456			      enum skb_drop_reason *reason);
2457
2458static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2459{
2460	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2461}
2462
2463int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2464struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2465
2466/*
2467 *	Recover an error report and clear atomically
2468 */
2469
2470static inline int sock_error(struct sock *sk)
2471{
2472	int err;
2473
2474	/* Avoid an atomic operation for the common case.
2475	 * This is racy since another cpu/thread can change sk_err under us.
2476	 */
2477	if (likely(data_race(!sk->sk_err)))
2478		return 0;
2479
2480	err = xchg(&sk->sk_err, 0);
2481	return -err;
2482}
2483
2484void sk_error_report(struct sock *sk);
2485
2486static inline unsigned long sock_wspace(struct sock *sk)
2487{
2488	int amt = 0;
2489
2490	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2491		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2492		if (amt < 0)
2493			amt = 0;
2494	}
2495	return amt;
2496}
2497
2498/* Note:
2499 *  We use sk->sk_wq_raw, from contexts knowing this
2500 *  pointer is not NULL and cannot disappear/change.
2501 */
2502static inline void sk_set_bit(int nr, struct sock *sk)
2503{
2504	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2505	    !sock_flag(sk, SOCK_FASYNC))
2506		return;
2507
2508	set_bit(nr, &sk->sk_wq_raw->flags);
2509}
2510
2511static inline void sk_clear_bit(int nr, struct sock *sk)
2512{
2513	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2514	    !sock_flag(sk, SOCK_FASYNC))
2515		return;
2516
2517	clear_bit(nr, &sk->sk_wq_raw->flags);
2518}
2519
2520static inline void sk_wake_async(const struct sock *sk, int how, int band)
2521{
2522	if (sock_flag(sk, SOCK_FASYNC)) {
2523		rcu_read_lock();
2524		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2525		rcu_read_unlock();
2526	}
2527}
2528
 
 
 
 
 
 
2529/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2530 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2531 * Note: for send buffers, TCP works better if we can build two skbs at
2532 * minimum.
2533 */
2534#define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2535
2536#define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2537#define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2538
2539static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2540{
2541	u32 val;
2542
2543	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2544		return;
2545
2546	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2547	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2548
2549	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2550}
2551
2552/**
2553 * sk_page_frag - return an appropriate page_frag
2554 * @sk: socket
2555 *
2556 * Use the per task page_frag instead of the per socket one for
2557 * optimization when we know that we're in process context and own
2558 * everything that's associated with %current.
2559 *
2560 * Both direct reclaim and page faults can nest inside other
2561 * socket operations and end up recursing into sk_page_frag()
2562 * while it's already in use: explicitly avoid task page_frag
2563 * when users disable sk_use_task_frag.
2564 *
2565 * Return: a per task page_frag if context allows that,
2566 * otherwise a per socket one.
2567 */
2568static inline struct page_frag *sk_page_frag(struct sock *sk)
2569{
2570	if (sk->sk_use_task_frag)
2571		return &current->task_frag;
2572
2573	return &sk->sk_frag;
2574}
2575
2576bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2577
2578/*
2579 *	Default write policy as shown to user space via poll/select/SIGIO
2580 */
2581static inline bool sock_writeable(const struct sock *sk)
2582{
2583	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2584}
2585
2586static inline gfp_t gfp_any(void)
2587{
2588	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2589}
2590
2591static inline gfp_t gfp_memcg_charge(void)
2592{
2593	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2594}
2595
2596static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2597{
2598	return noblock ? 0 : sk->sk_rcvtimeo;
2599}
2600
2601static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2602{
2603	return noblock ? 0 : sk->sk_sndtimeo;
2604}
2605
2606static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2607{
2608	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2609
2610	return v ?: 1;
2611}
2612
2613/* Alas, with timeout socket operations are not restartable.
2614 * Compare this to poll().
2615 */
2616static inline int sock_intr_errno(long timeo)
2617{
2618	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2619}
2620
2621struct sock_skb_cb {
2622	u32 dropcount;
2623};
2624
2625/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2626 * using skb->cb[] would keep using it directly and utilize its
2627 * alignement guarantee.
2628 */
2629#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2630			    sizeof(struct sock_skb_cb)))
2631
2632#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2633			    SOCK_SKB_CB_OFFSET))
2634
2635#define sock_skb_cb_check_size(size) \
2636	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2637
2638static inline void
2639sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2640{
2641	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2642						atomic_read(&sk->sk_drops) : 0;
2643}
2644
2645static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2646{
2647	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2648
2649	atomic_add(segs, &sk->sk_drops);
2650}
2651
2652static inline ktime_t sock_read_timestamp(struct sock *sk)
2653{
2654#if BITS_PER_LONG==32
2655	unsigned int seq;
2656	ktime_t kt;
2657
2658	do {
2659		seq = read_seqbegin(&sk->sk_stamp_seq);
2660		kt = sk->sk_stamp;
2661	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2662
2663	return kt;
2664#else
2665	return READ_ONCE(sk->sk_stamp);
2666#endif
2667}
2668
2669static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2670{
2671#if BITS_PER_LONG==32
2672	write_seqlock(&sk->sk_stamp_seq);
2673	sk->sk_stamp = kt;
2674	write_sequnlock(&sk->sk_stamp_seq);
2675#else
2676	WRITE_ONCE(sk->sk_stamp, kt);
2677#endif
2678}
2679
2680void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2681			   struct sk_buff *skb);
2682void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2683			     struct sk_buff *skb);
2684
2685static inline void
2686sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2687{
2688	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2689	u32 tsflags = READ_ONCE(sk->sk_tsflags);
2690	ktime_t kt = skb->tstamp;
2691	/*
2692	 * generate control messages if
2693	 * - receive time stamping in software requested
2694	 * - software time stamp available and wanted
2695	 * - hardware time stamps available and wanted
2696	 */
2697	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2698	    (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2699	    (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2700	    (hwtstamps->hwtstamp &&
2701	     (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2702		__sock_recv_timestamp(msg, sk, skb);
2703	else
2704		sock_write_timestamp(sk, kt);
2705
2706	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2707		__sock_recv_wifi_status(msg, sk, skb);
2708}
2709
2710void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2711		       struct sk_buff *skb);
2712
2713#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2714static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2715				   struct sk_buff *skb)
2716{
2717#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2718			   (1UL << SOCK_RCVTSTAMP)			| \
2719			   (1UL << SOCK_RCVMARK))
2720#define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2721			   SOF_TIMESTAMPING_RAW_HARDWARE)
2722
2723	if (sk->sk_flags & FLAGS_RECV_CMSGS ||
2724	    READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY)
2725		__sock_recv_cmsgs(msg, sk, skb);
2726	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2727		sock_write_timestamp(sk, skb->tstamp);
2728	else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2729		sock_write_timestamp(sk, 0);
2730}
2731
2732void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2733
2734/**
2735 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2736 * @sk:		socket sending this packet
2737 * @tsflags:	timestamping flags to use
2738 * @tx_flags:	completed with instructions for time stamping
2739 * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2740 *
2741 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2742 */
2743static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
 
2744				      __u8 *tx_flags, __u32 *tskey)
2745{
 
 
2746	if (unlikely(tsflags)) {
2747		__sock_tx_timestamp(tsflags, tx_flags);
2748		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2749		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2750			*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
 
 
 
 
2751	}
2752	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2753		*tx_flags |= SKBTX_WIFI_STATUS;
2754}
2755
2756static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
 
2757				     __u8 *tx_flags)
2758{
2759	_sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2760}
2761
2762static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
 
2763{
2764	_sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2765			   &skb_shinfo(skb)->tskey);
2766}
2767
2768static inline bool sk_is_inet(const struct sock *sk)
2769{
2770	int family = READ_ONCE(sk->sk_family);
2771
2772	return family == AF_INET || family == AF_INET6;
2773}
2774
2775static inline bool sk_is_tcp(const struct sock *sk)
2776{
2777	return sk_is_inet(sk) &&
2778	       sk->sk_type == SOCK_STREAM &&
2779	       sk->sk_protocol == IPPROTO_TCP;
2780}
2781
2782static inline bool sk_is_udp(const struct sock *sk)
2783{
2784	return sk_is_inet(sk) &&
2785	       sk->sk_type == SOCK_DGRAM &&
2786	       sk->sk_protocol == IPPROTO_UDP;
2787}
2788
2789static inline bool sk_is_stream_unix(const struct sock *sk)
2790{
2791	return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM;
2792}
2793
 
 
 
 
 
2794/**
2795 * sk_eat_skb - Release a skb if it is no longer needed
2796 * @sk: socket to eat this skb from
2797 * @skb: socket buffer to eat
2798 *
2799 * This routine must be called with interrupts disabled or with the socket
2800 * locked so that the sk_buff queue operation is ok.
2801*/
2802static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2803{
2804	__skb_unlink(skb, &sk->sk_receive_queue);
2805	__kfree_skb(skb);
2806}
2807
2808static inline bool
2809skb_sk_is_prefetched(struct sk_buff *skb)
2810{
2811#ifdef CONFIG_INET
2812	return skb->destructor == sock_pfree;
2813#else
2814	return false;
2815#endif /* CONFIG_INET */
2816}
2817
2818/* This helper checks if a socket is a full socket,
2819 * ie _not_ a timewait or request socket.
2820 */
2821static inline bool sk_fullsock(const struct sock *sk)
2822{
2823	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2824}
2825
2826static inline bool
2827sk_is_refcounted(struct sock *sk)
2828{
2829	/* Only full sockets have sk->sk_flags. */
2830	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2831}
2832
2833/**
2834 * skb_steal_sock - steal a socket from an sk_buff
2835 * @skb: sk_buff to steal the socket from
2836 * @refcounted: is set to true if the socket is reference-counted
2837 * @prefetched: is set to true if the socket was assigned from bpf
2838 */
2839static inline struct sock *
2840skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched)
2841{
2842	if (skb->sk) {
2843		struct sock *sk = skb->sk;
2844
2845		*refcounted = true;
2846		*prefetched = skb_sk_is_prefetched(skb);
2847		if (*prefetched)
2848			*refcounted = sk_is_refcounted(sk);
2849		skb->destructor = NULL;
2850		skb->sk = NULL;
2851		return sk;
2852	}
2853	*prefetched = false;
2854	*refcounted = false;
2855	return NULL;
2856}
2857
2858/* Checks if this SKB belongs to an HW offloaded socket
2859 * and whether any SW fallbacks are required based on dev.
2860 * Check decrypted mark in case skb_orphan() cleared socket.
2861 */
2862static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2863						   struct net_device *dev)
2864{
2865#ifdef CONFIG_SOCK_VALIDATE_XMIT
2866	struct sock *sk = skb->sk;
2867
2868	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2869		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2870#ifdef CONFIG_TLS_DEVICE
2871	} else if (unlikely(skb->decrypted)) {
2872		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2873		kfree_skb(skb);
2874		skb = NULL;
2875#endif
2876	}
2877#endif
2878
2879	return skb;
2880}
2881
2882/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2883 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2884 */
2885static inline bool sk_listener(const struct sock *sk)
2886{
2887	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2888}
2889
 
 
 
 
 
 
 
 
 
 
2890void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2891int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2892		       int type);
2893
2894bool sk_ns_capable(const struct sock *sk,
2895		   struct user_namespace *user_ns, int cap);
2896bool sk_capable(const struct sock *sk, int cap);
2897bool sk_net_capable(const struct sock *sk, int cap);
2898
2899void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2900
2901/* Take into consideration the size of the struct sk_buff overhead in the
2902 * determination of these values, since that is non-constant across
2903 * platforms.  This makes socket queueing behavior and performance
2904 * not depend upon such differences.
2905 */
2906#define _SK_MEM_PACKETS		256
2907#define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2908#define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2909#define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2910
2911extern __u32 sysctl_wmem_max;
2912extern __u32 sysctl_rmem_max;
2913
2914extern int sysctl_tstamp_allow_data;
2915
2916extern __u32 sysctl_wmem_default;
2917extern __u32 sysctl_rmem_default;
2918
2919#define SKB_FRAG_PAGE_ORDER	get_order(32768)
2920DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2921
2922static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2923{
2924	/* Does this proto have per netns sysctl_wmem ? */
2925	if (proto->sysctl_wmem_offset)
2926		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2927
2928	return READ_ONCE(*proto->sysctl_wmem);
2929}
2930
2931static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2932{
2933	/* Does this proto have per netns sysctl_rmem ? */
2934	if (proto->sysctl_rmem_offset)
2935		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2936
2937	return READ_ONCE(*proto->sysctl_rmem);
2938}
2939
2940/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2941 * Some wifi drivers need to tweak it to get more chunks.
2942 * They can use this helper from their ndo_start_xmit()
2943 */
2944static inline void sk_pacing_shift_update(struct sock *sk, int val)
2945{
2946	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2947		return;
2948	WRITE_ONCE(sk->sk_pacing_shift, val);
2949}
2950
2951/* if a socket is bound to a device, check that the given device
2952 * index is either the same or that the socket is bound to an L3
2953 * master device and the given device index is also enslaved to
2954 * that L3 master
2955 */
2956static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2957{
2958	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2959	int mdif;
2960
2961	if (!bound_dev_if || bound_dev_if == dif)
2962		return true;
2963
2964	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2965	if (mdif && mdif == bound_dev_if)
2966		return true;
2967
2968	return false;
2969}
2970
2971void sock_def_readable(struct sock *sk);
2972
2973int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2974void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2975int sock_set_timestamping(struct sock *sk, int optname,
2976			  struct so_timestamping timestamping);
2977
2978void sock_enable_timestamps(struct sock *sk);
2979void sock_no_linger(struct sock *sk);
2980void sock_set_keepalive(struct sock *sk);
2981void sock_set_priority(struct sock *sk, u32 priority);
2982void sock_set_rcvbuf(struct sock *sk, int val);
2983void sock_set_mark(struct sock *sk, u32 val);
2984void sock_set_reuseaddr(struct sock *sk);
2985void sock_set_reuseport(struct sock *sk);
2986void sock_set_sndtimeo(struct sock *sk, s64 secs);
2987
2988int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2989
2990int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2991int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2992			   sockptr_t optval, int optlen, bool old_timeval);
2993
2994int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
2995		     void __user *arg, void *karg, size_t size);
2996int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
2997static inline bool sk_is_readable(struct sock *sk)
2998{
2999	if (sk->sk_prot->sock_is_readable)
3000		return sk->sk_prot->sock_is_readable(sk);
3001	return false;
3002}
3003#endif	/* _SOCK_H */