Linux Audio

Check our new training course

Loading...
v5.4
   1/* SPDX-License-Identifier: GPL-2.0-or-later */
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Definitions for the AF_INET socket handler.
   8 *
   9 * Version:	@(#)sock.h	1.0.4	05/13/93
  10 *
  11 * Authors:	Ross Biro
  12 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  13 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
  14 *		Florian La Roche <flla@stud.uni-sb.de>
  15 *
  16 * Fixes:
  17 *		Alan Cox	:	Volatiles in skbuff pointers. See
  18 *					skbuff comments. May be overdone,
  19 *					better to prove they can be removed
  20 *					than the reverse.
  21 *		Alan Cox	:	Added a zapped field for tcp to note
  22 *					a socket is reset and must stay shut up
  23 *		Alan Cox	:	New fields for options
  24 *	Pauline Middelink	:	identd support
  25 *		Alan Cox	:	Eliminate low level recv/recvfrom
  26 *		David S. Miller	:	New socket lookup architecture.
  27 *              Steve Whitehouse:       Default routines for sock_ops
  28 *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
  29 *              			protinfo be just a void pointer, as the
  30 *              			protocol specific parts were moved to
  31 *              			respective headers and ipv4/v6, etc now
  32 *              			use private slabcaches for its socks
  33 *              Pedro Hortas	:	New flags field for socket options
  34 */
  35#ifndef _SOCK_H
  36#define _SOCK_H
  37
  38#include <linux/hardirq.h>
  39#include <linux/kernel.h>
  40#include <linux/list.h>
  41#include <linux/list_nulls.h>
  42#include <linux/timer.h>
  43#include <linux/cache.h>
  44#include <linux/bitops.h>
  45#include <linux/lockdep.h>
  46#include <linux/netdevice.h>
  47#include <linux/skbuff.h>	/* struct sk_buff */
  48#include <linux/mm.h>
  49#include <linux/security.h>
  50#include <linux/slab.h>
  51#include <linux/uaccess.h>
  52#include <linux/page_counter.h>
  53#include <linux/memcontrol.h>
  54#include <linux/static_key.h>
  55#include <linux/sched.h>
  56#include <linux/wait.h>
  57#include <linux/cgroup-defs.h>
  58#include <linux/rbtree.h>
  59#include <linux/filter.h>
  60#include <linux/rculist_nulls.h>
  61#include <linux/poll.h>
  62
 
  63#include <linux/atomic.h>
  64#include <linux/refcount.h>
 
  65#include <net/dst.h>
  66#include <net/checksum.h>
  67#include <net/tcp_states.h>
  68#include <linux/net_tstamp.h>
  69#include <net/smc.h>
  70#include <net/l3mdev.h>
 
  71
  72/*
  73 * This structure really needs to be cleaned up.
  74 * Most of it is for TCP, and not used by any of
  75 * the other protocols.
  76 */
  77
  78/* Define this to get the SOCK_DBG debugging facility. */
  79#define SOCK_DEBUGGING
  80#ifdef SOCK_DEBUGGING
  81#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
  82					printk(KERN_DEBUG msg); } while (0)
  83#else
  84/* Validate arguments and do nothing */
  85static inline __printf(2, 3)
  86void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
  87{
  88}
  89#endif
  90
  91/* This is the per-socket lock.  The spinlock provides a synchronization
  92 * between user contexts and software interrupt processing, whereas the
  93 * mini-semaphore synchronizes multiple users amongst themselves.
  94 */
  95typedef struct {
  96	spinlock_t		slock;
  97	int			owned;
  98	wait_queue_head_t	wq;
  99	/*
 100	 * We express the mutex-alike socket_lock semantics
 101	 * to the lock validator by explicitly managing
 102	 * the slock as a lock variant (in addition to
 103	 * the slock itself):
 104	 */
 105#ifdef CONFIG_DEBUG_LOCK_ALLOC
 106	struct lockdep_map dep_map;
 107#endif
 108} socket_lock_t;
 109
 110struct sock;
 111struct proto;
 112struct net;
 113
 114typedef __u32 __bitwise __portpair;
 115typedef __u64 __bitwise __addrpair;
 116
 117/**
 118 *	struct sock_common - minimal network layer representation of sockets
 119 *	@skc_daddr: Foreign IPv4 addr
 120 *	@skc_rcv_saddr: Bound local IPv4 addr
 
 121 *	@skc_hash: hash value used with various protocol lookup tables
 122 *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
 123 *	@skc_dport: placeholder for inet_dport/tw_dport
 124 *	@skc_num: placeholder for inet_num/tw_num
 
 125 *	@skc_family: network address family
 126 *	@skc_state: Connection state
 127 *	@skc_reuse: %SO_REUSEADDR setting
 128 *	@skc_reuseport: %SO_REUSEPORT setting
 
 
 129 *	@skc_bound_dev_if: bound device index if != 0
 130 *	@skc_bind_node: bind hash linkage for various protocol lookup tables
 131 *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
 132 *	@skc_prot: protocol handlers inside a network family
 133 *	@skc_net: reference to the network namespace of this socket
 
 
 
 134 *	@skc_node: main hash linkage for various protocol lookup tables
 135 *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
 136 *	@skc_tx_queue_mapping: tx queue number for this connection
 137 *	@skc_rx_queue_mapping: rx queue number for this connection
 138 *	@skc_flags: place holder for sk_flags
 139 *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
 140 *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
 
 
 
 
 141 *	@skc_incoming_cpu: record/match cpu processing incoming packets
 
 
 
 
 142 *	@skc_refcnt: reference count
 143 *
 144 *	This is the minimal network layer representation of sockets, the header
 145 *	for struct sock and struct inet_timewait_sock.
 146 */
 147struct sock_common {
 148	/* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
 149	 * address on 64bit arches : cf INET_MATCH()
 150	 */
 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_XPS
 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_sk_storage;
 
 235
 236/**
 237  *	struct sock - network layer representation of sockets
 238  *	@__sk_common: shared layout with inet_timewait_sock
 239  *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
 240  *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
 241  *	@sk_lock:	synchronizer
 242  *	@sk_kern_sock: True if sock is using kernel lock classes
 243  *	@sk_rcvbuf: size of receive buffer in bytes
 244  *	@sk_wq: sock wait queue and async head
 245  *	@sk_rx_dst: receive input route used by early demux
 
 
 246  *	@sk_dst_cache: destination cache
 247  *	@sk_dst_pending_confirm: need to confirm neighbour
 248  *	@sk_policy: flow policy
 249  *	@sk_receive_queue: incoming packets
 250  *	@sk_wmem_alloc: transmit queue bytes committed
 251  *	@sk_tsq_flags: TCP Small Queues flags
 252  *	@sk_write_queue: Packet sending queue
 253  *	@sk_omem_alloc: "o" is "option" or "other"
 254  *	@sk_wmem_queued: persistent queue size
 255  *	@sk_forward_alloc: space allocated forward
 
 256  *	@sk_napi_id: id of the last napi context to receive data for sk
 257  *	@sk_ll_usec: usecs to busypoll when there is no data
 258  *	@sk_allocation: allocation mode
 259  *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
 260  *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
 261  *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
 262  *	@sk_sndbuf: size of send buffer in bytes
 263  *	@__sk_flags_offset: empty field used to determine location of bitfield
 264  *	@sk_padding: unused element for alignment
 265  *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
 266  *	@sk_no_check_rx: allow zero checksum in RX packets
 267  *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
 268  *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
 269  *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
 270  *	@sk_gso_max_size: Maximum GSO segment size to build
 271  *	@sk_gso_max_segs: Maximum number of GSO segments
 272  *	@sk_pacing_shift: scaling factor for TCP Small Queues
 273  *	@sk_lingertime: %SO_LINGER l_linger setting
 274  *	@sk_backlog: always used with the per-socket spinlock held
 275  *	@sk_callback_lock: used with the callbacks in the end of this struct
 276  *	@sk_error_queue: rarely used
 277  *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
 278  *			  IPV6_ADDRFORM for instance)
 279  *	@sk_err: last error
 280  *	@sk_err_soft: errors that don't cause failure but are the cause of a
 281  *		      persistent failure not just 'timed out'
 282  *	@sk_drops: raw/udp drops counter
 283  *	@sk_ack_backlog: current listen backlog
 284  *	@sk_max_ack_backlog: listen backlog set in listen()
 285  *	@sk_uid: user id of owner
 
 
 286  *	@sk_priority: %SO_PRIORITY setting
 287  *	@sk_type: socket type (%SOCK_STREAM, etc)
 288  *	@sk_protocol: which protocol this socket belongs in this network family
 
 289  *	@sk_peer_pid: &struct pid for this socket's peer
 290  *	@sk_peer_cred: %SO_PEERCRED setting
 291  *	@sk_rcvlowat: %SO_RCVLOWAT setting
 292  *	@sk_rcvtimeo: %SO_RCVTIMEO setting
 293  *	@sk_sndtimeo: %SO_SNDTIMEO setting
 294  *	@sk_txhash: computed flow hash for use on transmit
 
 295  *	@sk_filter: socket filtering instructions
 296  *	@sk_timer: sock cleanup timer
 297  *	@sk_stamp: time stamp of last packet received
 298  *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
 299  *	@sk_tsflags: SO_TIMESTAMPING socket options
 
 
 
 
 
 300  *	@sk_tskey: counter to disambiguate concurrent tstamp requests
 301  *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
 302  *	@sk_socket: Identd and reporting IO signals
 303  *	@sk_user_data: RPC layer private data
 304  *	@sk_frag: cached page frag
 305  *	@sk_peek_off: current peek_offset value
 306  *	@sk_send_head: front of stuff to transmit
 
 307  *	@sk_security: used by security modules
 308  *	@sk_mark: generic packet mark
 309  *	@sk_cgrp_data: cgroup data for this cgroup
 310  *	@sk_memcg: this socket's memory cgroup association
 311  *	@sk_write_pending: a write to stream socket waits to start
 312  *	@sk_state_change: callback to indicate change in the state of the sock
 313  *	@sk_data_ready: callback to indicate there is data to be processed
 314  *	@sk_write_space: callback to indicate there is bf sending space available
 315  *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
 316  *	@sk_backlog_rcv: callback to process the backlog
 
 317  *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
 318  *	@sk_reuseport_cb: reuseport group container
 
 319  *	@sk_rcu: used during RCU grace period
 320  *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
 321  *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
 
 322  *	@sk_txtime_unused: unused txtime flags
 
 
 323  */
 324struct sock {
 325	/*
 326	 * Now struct inet_timewait_sock also uses sock_common, so please just
 327	 * don't add nothing before this first member (__sk_common) --acme
 328	 */
 329	struct sock_common	__sk_common;
 330#define sk_node			__sk_common.skc_node
 331#define sk_nulls_node		__sk_common.skc_nulls_node
 332#define sk_refcnt		__sk_common.skc_refcnt
 333#define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
 334#ifdef CONFIG_XPS
 335#define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
 336#endif
 337
 338#define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
 339#define sk_dontcopy_end		__sk_common.skc_dontcopy_end
 340#define sk_hash			__sk_common.skc_hash
 341#define sk_portpair		__sk_common.skc_portpair
 342#define sk_num			__sk_common.skc_num
 343#define sk_dport		__sk_common.skc_dport
 344#define sk_addrpair		__sk_common.skc_addrpair
 345#define sk_daddr		__sk_common.skc_daddr
 346#define sk_rcv_saddr		__sk_common.skc_rcv_saddr
 347#define sk_family		__sk_common.skc_family
 348#define sk_state		__sk_common.skc_state
 349#define sk_reuse		__sk_common.skc_reuse
 350#define sk_reuseport		__sk_common.skc_reuseport
 351#define sk_ipv6only		__sk_common.skc_ipv6only
 352#define sk_net_refcnt		__sk_common.skc_net_refcnt
 353#define sk_bound_dev_if		__sk_common.skc_bound_dev_if
 354#define sk_bind_node		__sk_common.skc_bind_node
 355#define sk_prot			__sk_common.skc_prot
 356#define sk_net			__sk_common.skc_net
 357#define sk_v6_daddr		__sk_common.skc_v6_daddr
 358#define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
 359#define sk_cookie		__sk_common.skc_cookie
 360#define sk_incoming_cpu		__sk_common.skc_incoming_cpu
 361#define sk_flags		__sk_common.skc_flags
 362#define sk_rxhash		__sk_common.skc_rxhash
 363
 
 
 
 
 
 364	socket_lock_t		sk_lock;
 365	atomic_t		sk_drops;
 366	int			sk_rcvlowat;
 367	struct sk_buff_head	sk_error_queue;
 368	struct sk_buff		*sk_rx_skb_cache;
 369	struct sk_buff_head	sk_receive_queue;
 370	/*
 371	 * The backlog queue is special, it is always used with
 372	 * the per-socket spinlock held and requires low latency
 373	 * access. Therefore we special case it's implementation.
 374	 * Note : rmem_alloc is in this structure to fill a hole
 375	 * on 64bit arches, not because its logically part of
 376	 * backlog.
 377	 */
 378	struct {
 379		atomic_t	rmem_alloc;
 380		int		len;
 381		struct sk_buff	*head;
 382		struct sk_buff	*tail;
 383	} sk_backlog;
 
 384#define sk_rmem_alloc sk_backlog.rmem_alloc
 385
 386	int			sk_forward_alloc;
 
 387#ifdef CONFIG_NET_RX_BUSY_POLL
 388	unsigned int		sk_ll_usec;
 389	/* ===== mostly read cache line ===== */
 390	unsigned int		sk_napi_id;
 391#endif
 392	int			sk_rcvbuf;
 393
 394	struct sk_filter __rcu	*sk_filter;
 395	union {
 396		struct socket_wq __rcu	*sk_wq;
 
 397		struct socket_wq	*sk_wq_raw;
 
 398	};
 399#ifdef CONFIG_XFRM
 400	struct xfrm_policy __rcu *sk_policy[2];
 401#endif
 402	struct dst_entry	*sk_rx_dst;
 403	struct dst_entry __rcu	*sk_dst_cache;
 404	atomic_t		sk_omem_alloc;
 405	int			sk_sndbuf;
 406
 407	/* ===== cache line for TX ===== */
 408	int			sk_wmem_queued;
 409	refcount_t		sk_wmem_alloc;
 410	unsigned long		sk_tsq_flags;
 411	union {
 412		struct sk_buff	*sk_send_head;
 413		struct rb_root	tcp_rtx_queue;
 414	};
 415	struct sk_buff		*sk_tx_skb_cache;
 416	struct sk_buff_head	sk_write_queue;
 417	__s32			sk_peek_off;
 418	int			sk_write_pending;
 419	__u32			sk_dst_pending_confirm;
 420	u32			sk_pacing_status; /* see enum sk_pacing */
 421	long			sk_sndtimeo;
 422	struct timer_list	sk_timer;
 423	__u32			sk_priority;
 424	__u32			sk_mark;
 425	unsigned long		sk_pacing_rate; /* bytes per second */
 426	unsigned long		sk_max_pacing_rate;
 427	struct page_frag	sk_frag;
 428	netdev_features_t	sk_route_caps;
 429	netdev_features_t	sk_route_nocaps;
 430	netdev_features_t	sk_route_forced_caps;
 431	int			sk_gso_type;
 432	unsigned int		sk_gso_max_size;
 433	gfp_t			sk_allocation;
 434	__u32			sk_txhash;
 435
 436	/*
 437	 * Because of non atomicity rules, all
 438	 * changes are protected by socket lock.
 439	 */
 440	unsigned int		__sk_flags_offset[0];
 441#ifdef __BIG_ENDIAN_BITFIELD
 442#define SK_FL_PROTO_SHIFT  16
 443#define SK_FL_PROTO_MASK   0x00ff0000
 444
 445#define SK_FL_TYPE_SHIFT   0
 446#define SK_FL_TYPE_MASK    0x0000ffff
 447#else
 448#define SK_FL_PROTO_SHIFT  8
 449#define SK_FL_PROTO_MASK   0x0000ff00
 450
 451#define SK_FL_TYPE_SHIFT   16
 452#define SK_FL_TYPE_MASK    0xffff0000
 453#endif
 454
 455	unsigned int		sk_padding : 1,
 456				sk_kern_sock : 1,
 457				sk_no_check_tx : 1,
 458				sk_no_check_rx : 1,
 459				sk_userlocks : 4,
 460				sk_protocol  : 8,
 461				sk_type      : 16;
 462#define SK_PROTOCOL_MAX U8_MAX
 463	u16			sk_gso_max_segs;
 464	u8			sk_pacing_shift;
 
 
 
 465	unsigned long	        sk_lingertime;
 466	struct proto		*sk_prot_creator;
 467	rwlock_t		sk_callback_lock;
 468	int			sk_err,
 469				sk_err_soft;
 470	u32			sk_ack_backlog;
 471	u32			sk_max_ack_backlog;
 472	kuid_t			sk_uid;
 
 
 
 
 
 
 
 473	struct pid		*sk_peer_pid;
 474	const struct cred	*sk_peer_cred;
 
 475	long			sk_rcvtimeo;
 476	ktime_t			sk_stamp;
 477#if BITS_PER_LONG==32
 478	seqlock_t		sk_stamp_seq;
 479#endif
 480	u16			sk_tsflags;
 481	u8			sk_shutdown;
 482	u32			sk_tskey;
 483	atomic_t		sk_zckey;
 
 
 484
 485	u8			sk_clockid;
 486	u8			sk_txtime_deadline_mode : 1,
 487				sk_txtime_report_errors : 1,
 488				sk_txtime_unused : 6;
 
 489
 490	struct socket		*sk_socket;
 491	void			*sk_user_data;
 492#ifdef CONFIG_SECURITY
 493	void			*sk_security;
 494#endif
 495	struct sock_cgroup_data	sk_cgrp_data;
 496	struct mem_cgroup	*sk_memcg;
 497	void			(*sk_state_change)(struct sock *sk);
 498	void			(*sk_data_ready)(struct sock *sk);
 499	void			(*sk_write_space)(struct sock *sk);
 500	void			(*sk_error_report)(struct sock *sk);
 501	int			(*sk_backlog_rcv)(struct sock *sk,
 502						  struct sk_buff *skb);
 503#ifdef CONFIG_SOCK_VALIDATE_XMIT
 504	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
 505							struct net_device *dev,
 506							struct sk_buff *skb);
 507#endif
 508	void                    (*sk_destruct)(struct sock *sk);
 509	struct sock_reuseport __rcu	*sk_reuseport_cb;
 510#ifdef CONFIG_BPF_SYSCALL
 511	struct bpf_sk_storage __rcu	*sk_bpf_storage;
 512#endif
 513	struct rcu_head		sk_rcu;
 
 
 514};
 515
 516enum sk_pacing {
 517	SK_PACING_NONE		= 0,
 518	SK_PACING_NEEDED	= 1,
 519	SK_PACING_FQ		= 2,
 520};
 521
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 522#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
 523
 524#define rcu_dereference_sk_user_data(sk)	rcu_dereference(__sk_user_data((sk)))
 525#define rcu_assign_sk_user_data(sk, ptr)	rcu_assign_pointer(__sk_user_data((sk)), ptr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 526
 527/*
 528 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
 529 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
 530 * on a socket means that the socket will reuse everybody else's port
 531 * without looking at the other's sk_reuse value.
 532 */
 533
 534#define SK_NO_REUSE	0
 535#define SK_CAN_REUSE	1
 536#define SK_FORCE_REUSE	2
 537
 538int sk_set_peek_off(struct sock *sk, int val);
 539
 540static inline int sk_peek_offset(struct sock *sk, int flags)
 541{
 542	if (unlikely(flags & MSG_PEEK)) {
 543		return READ_ONCE(sk->sk_peek_off);
 544	}
 545
 546	return 0;
 547}
 548
 549static inline void sk_peek_offset_bwd(struct sock *sk, int val)
 550{
 551	s32 off = READ_ONCE(sk->sk_peek_off);
 552
 553	if (unlikely(off >= 0)) {
 554		off = max_t(s32, off - val, 0);
 555		WRITE_ONCE(sk->sk_peek_off, off);
 556	}
 557}
 558
 559static inline void sk_peek_offset_fwd(struct sock *sk, int val)
 560{
 561	sk_peek_offset_bwd(sk, -val);
 562}
 563
 564/*
 565 * Hashed lists helper routines
 566 */
 567static inline struct sock *sk_entry(const struct hlist_node *node)
 568{
 569	return hlist_entry(node, struct sock, sk_node);
 570}
 571
 572static inline struct sock *__sk_head(const struct hlist_head *head)
 573{
 574	return hlist_entry(head->first, struct sock, sk_node);
 575}
 576
 577static inline struct sock *sk_head(const struct hlist_head *head)
 578{
 579	return hlist_empty(head) ? NULL : __sk_head(head);
 580}
 581
 582static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
 583{
 584	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
 585}
 586
 587static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
 588{
 589	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
 590}
 591
 592static inline struct sock *sk_next(const struct sock *sk)
 593{
 594	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
 595}
 596
 597static inline struct sock *sk_nulls_next(const struct sock *sk)
 598{
 599	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
 600		hlist_nulls_entry(sk->sk_nulls_node.next,
 601				  struct sock, sk_nulls_node) :
 602		NULL;
 603}
 604
 605static inline bool sk_unhashed(const struct sock *sk)
 606{
 607	return hlist_unhashed(&sk->sk_node);
 608}
 609
 610static inline bool sk_hashed(const struct sock *sk)
 611{
 612	return !sk_unhashed(sk);
 613}
 614
 615static inline void sk_node_init(struct hlist_node *node)
 616{
 617	node->pprev = NULL;
 618}
 619
 620static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
 621{
 622	node->pprev = NULL;
 623}
 624
 625static inline void __sk_del_node(struct sock *sk)
 626{
 627	__hlist_del(&sk->sk_node);
 628}
 629
 630/* NB: equivalent to hlist_del_init_rcu */
 631static inline bool __sk_del_node_init(struct sock *sk)
 632{
 633	if (sk_hashed(sk)) {
 634		__sk_del_node(sk);
 635		sk_node_init(&sk->sk_node);
 636		return true;
 637	}
 638	return false;
 639}
 640
 641/* Grab socket reference count. This operation is valid only
 642   when sk is ALREADY grabbed f.e. it is found in hash table
 643   or a list and the lookup is made under lock preventing hash table
 644   modifications.
 645 */
 646
 647static __always_inline void sock_hold(struct sock *sk)
 648{
 649	refcount_inc(&sk->sk_refcnt);
 650}
 651
 652/* Ungrab socket in the context, which assumes that socket refcnt
 653   cannot hit zero, f.e. it is true in context of any socketcall.
 654 */
 655static __always_inline void __sock_put(struct sock *sk)
 656{
 657	refcount_dec(&sk->sk_refcnt);
 658}
 659
 660static inline bool sk_del_node_init(struct sock *sk)
 661{
 662	bool rc = __sk_del_node_init(sk);
 663
 664	if (rc) {
 665		/* paranoid for a while -acme */
 666		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 667		__sock_put(sk);
 668	}
 669	return rc;
 670}
 671#define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
 672
 673static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
 674{
 675	if (sk_hashed(sk)) {
 676		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
 677		return true;
 678	}
 679	return false;
 680}
 681
 682static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
 683{
 684	bool rc = __sk_nulls_del_node_init_rcu(sk);
 685
 686	if (rc) {
 687		/* paranoid for a while -acme */
 688		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 689		__sock_put(sk);
 690	}
 691	return rc;
 692}
 693
 694static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
 695{
 696	hlist_add_head(&sk->sk_node, list);
 697}
 698
 699static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
 700{
 701	sock_hold(sk);
 702	__sk_add_node(sk, list);
 703}
 704
 705static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
 706{
 707	sock_hold(sk);
 708	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 709	    sk->sk_family == AF_INET6)
 710		hlist_add_tail_rcu(&sk->sk_node, list);
 711	else
 712		hlist_add_head_rcu(&sk->sk_node, list);
 713}
 714
 715static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
 716{
 717	sock_hold(sk);
 718	hlist_add_tail_rcu(&sk->sk_node, list);
 719}
 720
 721static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 722{
 723	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
 724}
 725
 
 
 
 
 
 726static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 727{
 728	sock_hold(sk);
 729	__sk_nulls_add_node_rcu(sk, list);
 730}
 731
 732static inline void __sk_del_bind_node(struct sock *sk)
 733{
 734	__hlist_del(&sk->sk_bind_node);
 735}
 736
 737static inline void sk_add_bind_node(struct sock *sk,
 738					struct hlist_head *list)
 739{
 740	hlist_add_head(&sk->sk_bind_node, list);
 741}
 742
 
 
 
 
 
 
 
 
 
 
 743#define sk_for_each(__sk, list) \
 744	hlist_for_each_entry(__sk, list, sk_node)
 745#define sk_for_each_rcu(__sk, list) \
 746	hlist_for_each_entry_rcu(__sk, list, sk_node)
 747#define sk_nulls_for_each(__sk, node, list) \
 748	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
 749#define sk_nulls_for_each_rcu(__sk, node, list) \
 750	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
 751#define sk_for_each_from(__sk) \
 752	hlist_for_each_entry_from(__sk, sk_node)
 753#define sk_nulls_for_each_from(__sk, node) \
 754	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
 755		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
 756#define sk_for_each_safe(__sk, tmp, list) \
 757	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
 758#define sk_for_each_bound(__sk, list) \
 759	hlist_for_each_entry(__sk, list, sk_bind_node)
 
 
 760
 761/**
 762 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
 763 * @tpos:	the type * to use as a loop cursor.
 764 * @pos:	the &struct hlist_node to use as a loop cursor.
 765 * @head:	the head for your list.
 766 * @offset:	offset of hlist_node within the struct.
 767 *
 768 */
 769#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
 770	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
 771	     pos != NULL &&						       \
 772		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
 773	     pos = rcu_dereference(hlist_next_rcu(pos)))
 774
 775static inline struct user_namespace *sk_user_ns(struct sock *sk)
 776{
 777	/* Careful only use this in a context where these parameters
 778	 * can not change and must all be valid, such as recvmsg from
 779	 * userspace.
 780	 */
 781	return sk->sk_socket->file->f_cred->user_ns;
 782}
 783
 784/* Sock flags */
 785enum sock_flags {
 786	SOCK_DEAD,
 787	SOCK_DONE,
 788	SOCK_URGINLINE,
 789	SOCK_KEEPOPEN,
 790	SOCK_LINGER,
 791	SOCK_DESTROY,
 792	SOCK_BROADCAST,
 793	SOCK_TIMESTAMP,
 794	SOCK_ZAPPED,
 795	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
 796	SOCK_DBG, /* %SO_DEBUG setting */
 797	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
 798	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
 799	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
 800	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
 801	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
 802	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
 803	SOCK_FASYNC, /* fasync() active */
 804	SOCK_RXQ_OVFL,
 805	SOCK_ZEROCOPY, /* buffers from userspace */
 806	SOCK_WIFI_STATUS, /* push wifi status to userspace */
 807	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
 808		     * Will use last 4 bytes of packet sent from
 809		     * user-space instead.
 810		     */
 811	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
 812	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
 813	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
 814	SOCK_TXTIME,
 815	SOCK_XDP, /* XDP is attached */
 816	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
 
 817};
 818
 819#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
 820
 821static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
 822{
 823	nsk->sk_flags = osk->sk_flags;
 824}
 825
 826static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
 827{
 828	__set_bit(flag, &sk->sk_flags);
 829}
 830
 831static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
 832{
 833	__clear_bit(flag, &sk->sk_flags);
 834}
 835
 
 
 
 
 
 
 
 
 
 836static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
 837{
 838	return test_bit(flag, &sk->sk_flags);
 839}
 840
 841#ifdef CONFIG_NET
 842DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
 843static inline int sk_memalloc_socks(void)
 844{
 845	return static_branch_unlikely(&memalloc_socks_key);
 846}
 
 
 847#else
 848
 849static inline int sk_memalloc_socks(void)
 850{
 851	return 0;
 852}
 853
 
 
 854#endif
 855
 856static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
 857{
 858	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
 859}
 860
 861static inline void sk_acceptq_removed(struct sock *sk)
 862{
 863	sk->sk_ack_backlog--;
 864}
 865
 866static inline void sk_acceptq_added(struct sock *sk)
 867{
 868	sk->sk_ack_backlog++;
 869}
 870
 
 
 
 
 871static inline bool sk_acceptq_is_full(const struct sock *sk)
 872{
 873	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
 874}
 875
 876/*
 877 * Compute minimal free write space needed to queue new packets.
 878 */
 879static inline int sk_stream_min_wspace(const struct sock *sk)
 880{
 881	return READ_ONCE(sk->sk_wmem_queued) >> 1;
 882}
 883
 884static inline int sk_stream_wspace(const struct sock *sk)
 885{
 886	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
 887}
 888
 889static inline void sk_wmem_queued_add(struct sock *sk, int val)
 890{
 891	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
 892}
 893
 894void sk_stream_write_space(struct sock *sk);
 895
 896/* OOB backlog add */
 897static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
 898{
 899	/* dont let skb dst not refcounted, we are going to leave rcu lock */
 900	skb_dst_force(skb);
 901
 902	if (!sk->sk_backlog.tail)
 903		sk->sk_backlog.head = skb;
 904	else
 905		sk->sk_backlog.tail->next = skb;
 906
 907	sk->sk_backlog.tail = skb;
 908	skb->next = NULL;
 909}
 910
 911/*
 912 * Take into account size of receive queue and backlog queue
 913 * Do not take into account this skb truesize,
 914 * to allow even a single big packet to come.
 915 */
 916static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
 917{
 918	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
 919
 920	return qsize > limit;
 921}
 922
 923/* The per-socket spinlock must be held here. */
 924static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
 925					      unsigned int limit)
 926{
 927	if (sk_rcvqueues_full(sk, limit))
 928		return -ENOBUFS;
 929
 930	/*
 931	 * If the skb was allocated from pfmemalloc reserves, only
 932	 * allow SOCK_MEMALLOC sockets to use it as this socket is
 933	 * helping free memory
 934	 */
 935	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
 936		return -ENOMEM;
 937
 938	__sk_add_backlog(sk, skb);
 939	sk->sk_backlog.len += skb->truesize;
 940	return 0;
 941}
 942
 943int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
 944
 
 
 
 945static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 946{
 947	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
 948		return __sk_backlog_rcv(sk, skb);
 949
 950	return sk->sk_backlog_rcv(sk, skb);
 
 
 
 951}
 952
 953static inline void sk_incoming_cpu_update(struct sock *sk)
 954{
 955	int cpu = raw_smp_processor_id();
 956
 957	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
 958		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
 959}
 960
 961static inline void sock_rps_record_flow_hash(__u32 hash)
 962{
 963#ifdef CONFIG_RPS
 964	struct rps_sock_flow_table *sock_flow_table;
 965
 966	rcu_read_lock();
 967	sock_flow_table = rcu_dereference(rps_sock_flow_table);
 968	rps_record_sock_flow(sock_flow_table, hash);
 969	rcu_read_unlock();
 970#endif
 971}
 972
 973static inline void sock_rps_record_flow(const struct sock *sk)
 974{
 975#ifdef CONFIG_RPS
 976	if (static_branch_unlikely(&rfs_needed)) {
 977		/* Reading sk->sk_rxhash might incur an expensive cache line
 978		 * miss.
 979		 *
 980		 * TCP_ESTABLISHED does cover almost all states where RFS
 981		 * might be useful, and is cheaper [1] than testing :
 982		 *	IPv4: inet_sk(sk)->inet_daddr
 983		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
 984		 * OR	an additional socket flag
 985		 * [1] : sk_state and sk_prot are in the same cache line.
 986		 */
 987		if (sk->sk_state == TCP_ESTABLISHED)
 988			sock_rps_record_flow_hash(sk->sk_rxhash);
 989	}
 990#endif
 991}
 992
 993static inline void sock_rps_save_rxhash(struct sock *sk,
 994					const struct sk_buff *skb)
 995{
 996#ifdef CONFIG_RPS
 997	if (unlikely(sk->sk_rxhash != skb->hash))
 998		sk->sk_rxhash = skb->hash;
 999#endif
1000}
1001
1002static inline void sock_rps_reset_rxhash(struct sock *sk)
1003{
1004#ifdef CONFIG_RPS
1005	sk->sk_rxhash = 0;
1006#endif
1007}
1008
1009#define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1010	({	int __rc;						\
1011		release_sock(__sk);					\
1012		__rc = __condition;					\
1013		if (!__rc) {						\
1014			*(__timeo) = wait_woken(__wait,			\
1015						TASK_INTERRUPTIBLE,	\
1016						*(__timeo));		\
1017		}							\
1018		sched_annotate_sleep();					\
1019		lock_sock(__sk);					\
1020		__rc = __condition;					\
1021		__rc;							\
1022	})
1023
1024int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1025int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1026void sk_stream_wait_close(struct sock *sk, long timeo_p);
1027int sk_stream_error(struct sock *sk, int flags, int err);
1028void sk_stream_kill_queues(struct sock *sk);
1029void sk_set_memalloc(struct sock *sk);
1030void sk_clear_memalloc(struct sock *sk);
1031
1032void __sk_flush_backlog(struct sock *sk);
1033
1034static inline bool sk_flush_backlog(struct sock *sk)
1035{
1036	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1037		__sk_flush_backlog(sk);
1038		return true;
1039	}
1040	return false;
1041}
1042
1043int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1044
1045struct request_sock_ops;
1046struct timewait_sock_ops;
1047struct inet_hashinfo;
1048struct raw_hashinfo;
1049struct smc_hashinfo;
1050struct module;
 
1051
1052/*
1053 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1054 * un-modified. Special care is taken when initializing object to zero.
1055 */
1056static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1057{
1058	if (offsetof(struct sock, sk_node.next) != 0)
1059		memset(sk, 0, offsetof(struct sock, sk_node.next));
1060	memset(&sk->sk_node.pprev, 0,
1061	       size - offsetof(struct sock, sk_node.pprev));
1062}
1063
1064/* Networking protocol blocks we attach to sockets.
1065 * socket layer -> transport layer interface
1066 */
1067struct proto {
1068	void			(*close)(struct sock *sk,
1069					long timeout);
1070	int			(*pre_connect)(struct sock *sk,
1071					struct sockaddr *uaddr,
1072					int addr_len);
1073	int			(*connect)(struct sock *sk,
1074					struct sockaddr *uaddr,
1075					int addr_len);
1076	int			(*disconnect)(struct sock *sk, int flags);
1077
1078	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1079					  bool kern);
1080
1081	int			(*ioctl)(struct sock *sk, int cmd,
1082					 unsigned long arg);
1083	int			(*init)(struct sock *sk);
1084	void			(*destroy)(struct sock *sk);
1085	void			(*shutdown)(struct sock *sk, int how);
1086	int			(*setsockopt)(struct sock *sk, int level,
1087					int optname, char __user *optval,
1088					unsigned int optlen);
1089	int			(*getsockopt)(struct sock *sk, int level,
1090					int optname, char __user *optval,
1091					int __user *option);
1092	void			(*keepalive)(struct sock *sk, int valbool);
1093#ifdef CONFIG_COMPAT
1094	int			(*compat_setsockopt)(struct sock *sk,
1095					int level,
1096					int optname, char __user *optval,
1097					unsigned int optlen);
1098	int			(*compat_getsockopt)(struct sock *sk,
1099					int level,
1100					int optname, char __user *optval,
1101					int __user *option);
1102	int			(*compat_ioctl)(struct sock *sk,
1103					unsigned int cmd, unsigned long arg);
1104#endif
1105	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1106					   size_t len);
1107	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1108					   size_t len, int noblock, int flags,
1109					   int *addr_len);
1110	int			(*sendpage)(struct sock *sk, struct page *page,
1111					int offset, size_t size, int flags);
1112	int			(*bind)(struct sock *sk,
1113					struct sockaddr *uaddr, int addr_len);
 
 
1114
1115	int			(*backlog_rcv) (struct sock *sk,
1116						struct sk_buff *skb);
 
 
1117
1118	void		(*release_cb)(struct sock *sk);
1119
1120	/* Keeping track of sk's, looking them up, and port selection methods. */
1121	int			(*hash)(struct sock *sk);
1122	void			(*unhash)(struct sock *sk);
1123	void			(*rehash)(struct sock *sk);
1124	int			(*get_port)(struct sock *sk, unsigned short snum);
 
 
 
 
 
 
1125
1126	/* Keeping track of sockets in use */
1127#ifdef CONFIG_PROC_FS
1128	unsigned int		inuse_idx;
1129#endif
1130
 
 
 
 
1131	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1132	bool			(*stream_memory_read)(const struct sock *sk);
1133	/* Memory pressure */
1134	void			(*enter_memory_pressure)(struct sock *sk);
1135	void			(*leave_memory_pressure)(struct sock *sk);
1136	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
 
1137	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
 
1138	/*
1139	 * Pressure flag: try to collapse.
1140	 * Technical note: it is used by multiple contexts non atomically.
1141	 * All the __sk_mem_schedule() is of this nature: accounting
1142	 * is strict, actions are advisory and have some latency.
1143	 */
1144	unsigned long		*memory_pressure;
1145	long			*sysctl_mem;
1146
1147	int			*sysctl_wmem;
1148	int			*sysctl_rmem;
1149	u32			sysctl_wmem_offset;
1150	u32			sysctl_rmem_offset;
1151
1152	int			max_header;
1153	bool			no_autobind;
1154
1155	struct kmem_cache	*slab;
1156	unsigned int		obj_size;
1157	slab_flags_t		slab_flags;
1158	unsigned int		useroffset;	/* Usercopy region offset */
1159	unsigned int		usersize;	/* Usercopy region size */
1160
1161	struct percpu_counter	*orphan_count;
1162
1163	struct request_sock_ops	*rsk_prot;
1164	struct timewait_sock_ops *twsk_prot;
1165
1166	union {
1167		struct inet_hashinfo	*hashinfo;
1168		struct udp_table	*udp_table;
1169		struct raw_hashinfo	*raw_hash;
1170		struct smc_hashinfo	*smc_hash;
1171	} h;
1172
1173	struct module		*owner;
1174
1175	char			name[32];
1176
1177	struct list_head	node;
1178#ifdef SOCK_REFCNT_DEBUG
1179	atomic_t		socks;
1180#endif
1181	int			(*diag_destroy)(struct sock *sk, int err);
1182} __randomize_layout;
1183
1184int proto_register(struct proto *prot, int alloc_slab);
1185void proto_unregister(struct proto *prot);
1186int sock_load_diag_module(int family, int protocol);
1187
1188#ifdef SOCK_REFCNT_DEBUG
1189static inline void sk_refcnt_debug_inc(struct sock *sk)
1190{
1191	atomic_inc(&sk->sk_prot->socks);
1192}
1193
1194static inline void sk_refcnt_debug_dec(struct sock *sk)
1195{
1196	atomic_dec(&sk->sk_prot->socks);
1197	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1198	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1199}
1200
1201static inline void sk_refcnt_debug_release(const struct sock *sk)
1202{
1203	if (refcount_read(&sk->sk_refcnt) != 1)
1204		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1205		       sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1206}
1207#else /* SOCK_REFCNT_DEBUG */
1208#define sk_refcnt_debug_inc(sk) do { } while (0)
1209#define sk_refcnt_debug_dec(sk) do { } while (0)
1210#define sk_refcnt_debug_release(sk) do { } while (0)
1211#endif /* SOCK_REFCNT_DEBUG */
1212
 
 
 
 
 
 
 
 
 
 
 
1213static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1214{
1215	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1216		return false;
1217
1218	return sk->sk_prot->stream_memory_free ?
1219		sk->sk_prot->stream_memory_free(sk, wake) : true;
 
1220}
1221
1222static inline bool sk_stream_memory_free(const struct sock *sk)
1223{
1224	return __sk_stream_memory_free(sk, 0);
1225}
1226
1227static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1228{
1229	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1230	       __sk_stream_memory_free(sk, wake);
1231}
1232
1233static inline bool sk_stream_is_writeable(const struct sock *sk)
1234{
1235	return __sk_stream_is_writeable(sk, 0);
1236}
1237
1238static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1239					    struct cgroup *ancestor)
1240{
1241#ifdef CONFIG_SOCK_CGROUP_DATA
1242	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1243				    ancestor);
1244#else
1245	return -ENOTSUPP;
1246#endif
1247}
1248
1249static inline bool sk_has_memory_pressure(const struct sock *sk)
1250{
1251	return sk->sk_prot->memory_pressure != NULL;
1252}
1253
1254static inline bool sk_under_memory_pressure(const struct sock *sk)
1255{
1256	if (!sk->sk_prot->memory_pressure)
1257		return false;
1258
1259	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1260	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1261		return true;
1262
1263	return !!*sk->sk_prot->memory_pressure;
1264}
1265
1266static inline long
1267sk_memory_allocated(const struct sock *sk)
1268{
1269	return atomic_long_read(sk->sk_prot->memory_allocated);
1270}
1271
1272static inline long
 
 
 
 
 
 
 
 
 
1273sk_memory_allocated_add(struct sock *sk, int amt)
1274{
1275	return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
 
 
 
 
 
 
 
 
1276}
1277
1278static inline void
1279sk_memory_allocated_sub(struct sock *sk, int amt)
1280{
1281	atomic_long_sub(amt, sk->sk_prot->memory_allocated);
 
 
 
 
 
 
 
 
1282}
1283
 
 
1284static inline void sk_sockets_allocated_dec(struct sock *sk)
1285{
1286	percpu_counter_dec(sk->sk_prot->sockets_allocated);
 
1287}
1288
1289static inline void sk_sockets_allocated_inc(struct sock *sk)
1290{
1291	percpu_counter_inc(sk->sk_prot->sockets_allocated);
 
1292}
1293
1294static inline u64
1295sk_sockets_allocated_read_positive(struct sock *sk)
1296{
1297	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1298}
1299
1300static inline int
1301proto_sockets_allocated_sum_positive(struct proto *prot)
1302{
1303	return percpu_counter_sum_positive(prot->sockets_allocated);
1304}
1305
1306static inline long
1307proto_memory_allocated(struct proto *prot)
1308{
1309	return atomic_long_read(prot->memory_allocated);
1310}
1311
1312static inline bool
1313proto_memory_pressure(struct proto *prot)
1314{
1315	if (!prot->memory_pressure)
1316		return false;
1317	return !!*prot->memory_pressure;
1318}
1319
1320
1321#ifdef CONFIG_PROC_FS
1322/* Called with local bh disabled */
1323void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1324int sock_prot_inuse_get(struct net *net, struct proto *proto);
1325int sock_inuse_get(struct net *net);
1326#else
1327static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1328		int inc)
 
 
 
 
1329{
1330}
1331#endif
1332
1333
1334/* With per-bucket locks this operation is not-atomic, so that
1335 * this version is not worse.
1336 */
1337static inline int __sk_prot_rehash(struct sock *sk)
1338{
1339	sk->sk_prot->unhash(sk);
1340	return sk->sk_prot->hash(sk);
1341}
1342
1343/* About 10 seconds */
1344#define SOCK_DESTROY_TIME (10*HZ)
1345
1346/* Sockets 0-1023 can't be bound to unless you are superuser */
1347#define PROT_SOCK	1024
1348
1349#define SHUTDOWN_MASK	3
1350#define RCV_SHUTDOWN	1
1351#define SEND_SHUTDOWN	2
1352
1353#define SOCK_SNDBUF_LOCK	1
1354#define SOCK_RCVBUF_LOCK	2
1355#define SOCK_BINDADDR_LOCK	4
1356#define SOCK_BINDPORT_LOCK	8
1357
1358struct socket_alloc {
1359	struct socket socket;
1360	struct inode vfs_inode;
1361};
1362
1363static inline struct socket *SOCKET_I(struct inode *inode)
1364{
1365	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1366}
1367
1368static inline struct inode *SOCK_INODE(struct socket *socket)
1369{
1370	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1371}
1372
1373/*
1374 * Functions for memory accounting
1375 */
1376int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1377int __sk_mem_schedule(struct sock *sk, int size, int kind);
1378void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1379void __sk_mem_reclaim(struct sock *sk, int amount);
1380
1381/* We used to have PAGE_SIZE here, but systems with 64KB pages
1382 * do not necessarily have 16x time more memory than 4KB ones.
1383 */
1384#define SK_MEM_QUANTUM 4096
1385#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1386#define SK_MEM_SEND	0
1387#define SK_MEM_RECV	1
1388
1389/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1390static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1391{
1392	long val = sk->sk_prot->sysctl_mem[index];
1393
1394#if PAGE_SIZE > SK_MEM_QUANTUM
1395	val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1396#elif PAGE_SIZE < SK_MEM_QUANTUM
1397	val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1398#endif
1399	return val;
1400}
1401
1402static inline int sk_mem_pages(int amt)
1403{
1404	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1405}
1406
1407static inline bool sk_has_account(struct sock *sk)
1408{
1409	/* return true if protocol supports memory accounting */
1410	return !!sk->sk_prot->memory_allocated;
1411}
1412
1413static inline bool sk_wmem_schedule(struct sock *sk, int size)
1414{
 
 
1415	if (!sk_has_account(sk))
1416		return true;
1417	return size <= sk->sk_forward_alloc ||
1418		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1419}
1420
1421static inline bool
1422sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1423{
 
 
1424	if (!sk_has_account(sk))
1425		return true;
1426	return size<= sk->sk_forward_alloc ||
1427		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1428		skb_pfmemalloc(skb);
1429}
1430
 
 
 
 
 
 
 
 
 
 
 
 
 
1431static inline void sk_mem_reclaim(struct sock *sk)
1432{
 
 
1433	if (!sk_has_account(sk))
1434		return;
1435	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1436		__sk_mem_reclaim(sk, sk->sk_forward_alloc);
 
 
 
1437}
1438
1439static inline void sk_mem_reclaim_partial(struct sock *sk)
1440{
1441	if (!sk_has_account(sk))
1442		return;
1443	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1444		__sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1445}
1446
1447static inline void sk_mem_charge(struct sock *sk, int size)
1448{
1449	if (!sk_has_account(sk))
1450		return;
1451	sk->sk_forward_alloc -= size;
1452}
1453
1454static inline void sk_mem_uncharge(struct sock *sk, int size)
1455{
1456	if (!sk_has_account(sk))
1457		return;
1458	sk->sk_forward_alloc += size;
1459
1460	/* Avoid a possible overflow.
1461	 * TCP send queues can make this happen, if sk_mem_reclaim()
1462	 * is not called and more than 2 GBytes are released at once.
1463	 *
1464	 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1465	 * no need to hold that much forward allocation anyway.
1466	 */
1467	if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1468		__sk_mem_reclaim(sk, 1 << 20);
1469}
1470
1471DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
1472static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1473{
1474	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1475	sk_wmem_queued_add(sk, -skb->truesize);
1476	sk_mem_uncharge(sk, skb->truesize);
1477	if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1478	    !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1479		skb_zcopy_clear(skb, true);
1480		sk->sk_tx_skb_cache = skb;
1481		return;
1482	}
1483	__kfree_skb(skb);
1484}
1485
1486static inline void sock_release_ownership(struct sock *sk)
1487{
1488	if (sk->sk_lock.owned) {
1489		sk->sk_lock.owned = 0;
1490
1491		/* The sk_lock has mutex_unlock() semantics: */
1492		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1493	}
1494}
1495
1496/*
1497 * Macro so as to not evaluate some arguments when
1498 * lockdep is not enabled.
1499 *
1500 * Mark both the sk_lock and the sk_lock.slock as a
1501 * per-address-family lock class.
1502 */
1503#define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1504do {									\
1505	sk->sk_lock.owned = 0;						\
1506	init_waitqueue_head(&sk->sk_lock.wq);				\
1507	spin_lock_init(&(sk)->sk_lock.slock);				\
1508	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1509			sizeof((sk)->sk_lock));				\
1510	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1511				(skey), (sname));				\
1512	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1513} while (0)
1514
1515#ifdef CONFIG_LOCKDEP
1516static inline bool lockdep_sock_is_held(const struct sock *sk)
1517{
1518	return lockdep_is_held(&sk->sk_lock) ||
1519	       lockdep_is_held(&sk->sk_lock.slock);
1520}
1521#endif
1522
1523void lock_sock_nested(struct sock *sk, int subclass);
1524
1525static inline void lock_sock(struct sock *sk)
1526{
1527	lock_sock_nested(sk, 0);
1528}
1529
 
1530void __release_sock(struct sock *sk);
1531void release_sock(struct sock *sk);
1532
1533/* BH context may only use the following locking interface. */
1534#define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1535#define bh_lock_sock_nested(__sk) \
1536				spin_lock_nested(&((__sk)->sk_lock.slock), \
1537				SINGLE_DEPTH_NESTING)
1538#define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1539
1540bool lock_sock_fast(struct sock *sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1541/**
1542 * unlock_sock_fast - complement of lock_sock_fast
1543 * @sk: socket
1544 * @slow: slow mode
1545 *
1546 * fast unlock socket for user context.
1547 * If slow mode is on, we call regular release_sock()
1548 */
1549static inline void unlock_sock_fast(struct sock *sk, bool slow)
 
1550{
1551	if (slow)
1552		release_sock(sk);
1553	else
 
 
1554		spin_unlock_bh(&sk->sk_lock.slock);
 
1555}
1556
 
 
 
 
 
1557/* Used by processes to "lock" a socket state, so that
1558 * interrupts and bottom half handlers won't change it
1559 * from under us. It essentially blocks any incoming
1560 * packets, so that we won't get any new data or any
1561 * packets that change the state of the socket.
1562 *
1563 * While locked, BH processing will add new packets to
1564 * the backlog queue.  This queue is processed by the
1565 * owner of the socket lock right before it is released.
1566 *
1567 * Since ~2.3.5 it is also exclusive sleep lock serializing
1568 * accesses from user process context.
1569 */
1570
1571static inline void sock_owned_by_me(const struct sock *sk)
1572{
1573#ifdef CONFIG_LOCKDEP
1574	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1575#endif
1576}
1577
1578static inline bool sock_owned_by_user(const struct sock *sk)
1579{
1580	sock_owned_by_me(sk);
1581	return sk->sk_lock.owned;
1582}
1583
1584static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1585{
1586	return sk->sk_lock.owned;
1587}
1588
 
 
 
 
 
 
 
 
 
 
1589/* no reclassification while locks are held */
1590static inline bool sock_allow_reclassification(const struct sock *csk)
1591{
1592	struct sock *sk = (struct sock *)csk;
1593
1594	return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
 
1595}
1596
1597struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1598		      struct proto *prot, int kern);
1599void sk_free(struct sock *sk);
1600void sk_destruct(struct sock *sk);
1601struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1602void sk_free_unlock_clone(struct sock *sk);
1603
1604struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1605			     gfp_t priority);
1606void __sock_wfree(struct sk_buff *skb);
1607void sock_wfree(struct sk_buff *skb);
1608struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1609			     gfp_t priority);
1610void skb_orphan_partial(struct sk_buff *skb);
1611void sock_rfree(struct sk_buff *skb);
1612void sock_efree(struct sk_buff *skb);
1613#ifdef CONFIG_INET
1614void sock_edemux(struct sk_buff *skb);
 
1615#else
1616#define sock_edemux sock_efree
1617#endif
1618
 
 
1619int sock_setsockopt(struct socket *sock, int level, int op,
1620		    char __user *optval, unsigned int optlen);
1621
 
 
1622int sock_getsockopt(struct socket *sock, int level, int op,
1623		    char __user *optval, int __user *optlen);
1624int sock_gettstamp(struct socket *sock, void __user *userstamp,
1625		   bool timeval, bool time32);
1626struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1627				    int noblock, int *errcode);
1628struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1629				     unsigned long data_len, int noblock,
1630				     int *errcode, int max_page_order);
 
 
 
 
 
 
 
 
1631void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1632void sock_kfree_s(struct sock *sk, void *mem, int size);
1633void sock_kzfree_s(struct sock *sk, void *mem, int size);
1634void sk_send_sigurg(struct sock *sk);
1635
 
 
 
 
 
 
 
1636struct sockcm_cookie {
1637	u64 transmit_time;
1638	u32 mark;
1639	u16 tsflags;
1640};
1641
1642static inline void sockcm_init(struct sockcm_cookie *sockc,
1643			       const struct sock *sk)
1644{
1645	*sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1646}
1647
1648int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1649		     struct sockcm_cookie *sockc);
1650int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1651		   struct sockcm_cookie *sockc);
1652
1653/*
1654 * Functions to fill in entries in struct proto_ops when a protocol
1655 * does not implement a particular function.
1656 */
1657int sock_no_bind(struct socket *, struct sockaddr *, int);
1658int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1659int sock_no_socketpair(struct socket *, struct socket *);
1660int sock_no_accept(struct socket *, struct socket *, int, bool);
1661int sock_no_getname(struct socket *, struct sockaddr *, int);
1662int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1663int sock_no_listen(struct socket *, int);
1664int sock_no_shutdown(struct socket *, int);
1665int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1666int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1667int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1668int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1669int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1670int sock_no_mmap(struct file *file, struct socket *sock,
1671		 struct vm_area_struct *vma);
1672ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1673			 size_t size, int flags);
1674ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1675				int offset, size_t size, int flags);
1676
1677/*
1678 * Functions to fill in entries in struct proto_ops when a protocol
1679 * uses the inet style.
1680 */
1681int sock_common_getsockopt(struct socket *sock, int level, int optname,
1682				  char __user *optval, int __user *optlen);
1683int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1684			int flags);
1685int sock_common_setsockopt(struct socket *sock, int level, int optname,
1686				  char __user *optval, unsigned int optlen);
1687int compat_sock_common_getsockopt(struct socket *sock, int level,
1688		int optname, char __user *optval, int __user *optlen);
1689int compat_sock_common_setsockopt(struct socket *sock, int level,
1690		int optname, char __user *optval, unsigned int optlen);
1691
1692void sk_common_release(struct sock *sk);
1693
1694/*
1695 *	Default socket callbacks and setup code
1696 */
1697
1698/* Initialise core socket variables */
1699void sock_init_data(struct socket *sock, struct sock *sk);
1700
1701/*
1702 * Socket reference counting postulates.
1703 *
1704 * * Each user of socket SHOULD hold a reference count.
1705 * * Each access point to socket (an hash table bucket, reference from a list,
1706 *   running timer, skb in flight MUST hold a reference count.
1707 * * When reference count hits 0, it means it will never increase back.
1708 * * When reference count hits 0, it means that no references from
1709 *   outside exist to this socket and current process on current CPU
1710 *   is last user and may/should destroy this socket.
1711 * * sk_free is called from any context: process, BH, IRQ. When
1712 *   it is called, socket has no references from outside -> sk_free
1713 *   may release descendant resources allocated by the socket, but
1714 *   to the time when it is called, socket is NOT referenced by any
1715 *   hash tables, lists etc.
1716 * * Packets, delivered from outside (from network or from another process)
1717 *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1718 *   when they sit in queue. Otherwise, packets will leak to hole, when
1719 *   socket is looked up by one cpu and unhasing is made by another CPU.
1720 *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1721 *   (leak to backlog). Packet socket does all the processing inside
1722 *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1723 *   use separate SMP lock, so that they are prone too.
1724 */
1725
1726/* Ungrab socket and destroy it, if it was the last reference. */
1727static inline void sock_put(struct sock *sk)
1728{
1729	if (refcount_dec_and_test(&sk->sk_refcnt))
1730		sk_free(sk);
1731}
1732/* Generic version of sock_put(), dealing with all sockets
1733 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1734 */
1735void sock_gen_put(struct sock *sk);
1736
1737int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1738		     unsigned int trim_cap, bool refcounted);
1739static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1740				 const int nested)
1741{
1742	return __sk_receive_skb(sk, skb, nested, 1, true);
1743}
1744
1745static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1746{
1747	/* sk_tx_queue_mapping accept only upto a 16-bit value */
1748	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1749		return;
1750	sk->sk_tx_queue_mapping = tx_queue;
1751}
1752
1753#define NO_QUEUE_MAPPING	USHRT_MAX
1754
1755static inline void sk_tx_queue_clear(struct sock *sk)
1756{
1757	sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1758}
1759
1760static inline int sk_tx_queue_get(const struct sock *sk)
1761{
1762	if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1763		return sk->sk_tx_queue_mapping;
1764
1765	return -1;
1766}
1767
1768static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
 
 
1769{
1770#ifdef CONFIG_XPS
1771	if (skb_rx_queue_recorded(skb)) {
1772		u16 rx_queue = skb_get_rx_queue(skb);
1773
1774		if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1775			return;
1776
1777		sk->sk_rx_queue_mapping = rx_queue;
1778	}
1779#endif
1780}
1781
 
 
 
 
 
 
 
 
 
 
1782static inline void sk_rx_queue_clear(struct sock *sk)
1783{
1784#ifdef CONFIG_XPS
1785	sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1786#endif
1787}
1788
1789#ifdef CONFIG_XPS
1790static inline int sk_rx_queue_get(const struct sock *sk)
1791{
1792	if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1793		return sk->sk_rx_queue_mapping;
 
 
 
 
 
 
1794
1795	return -1;
1796}
1797#endif
1798
1799static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1800{
1801	sk_tx_queue_clear(sk);
1802	sk->sk_socket = sock;
1803}
1804
1805static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1806{
1807	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1808	return &rcu_dereference_raw(sk->sk_wq)->wait;
1809}
1810/* Detach socket from process context.
1811 * Announce socket dead, detach it from wait queue and inode.
1812 * Note that parent inode held reference count on this struct sock,
1813 * we do not release it in this function, because protocol
1814 * probably wants some additional cleanups or even continuing
1815 * to work with this socket (TCP).
1816 */
1817static inline void sock_orphan(struct sock *sk)
1818{
1819	write_lock_bh(&sk->sk_callback_lock);
1820	sock_set_flag(sk, SOCK_DEAD);
1821	sk_set_socket(sk, NULL);
1822	sk->sk_wq  = NULL;
1823	write_unlock_bh(&sk->sk_callback_lock);
1824}
1825
1826static inline void sock_graft(struct sock *sk, struct socket *parent)
1827{
1828	WARN_ON(parent->sk);
1829	write_lock_bh(&sk->sk_callback_lock);
1830	rcu_assign_pointer(sk->sk_wq, &parent->wq);
1831	parent->sk = sk;
1832	sk_set_socket(sk, parent);
1833	sk->sk_uid = SOCK_INODE(parent)->i_uid;
1834	security_sock_graft(sk, parent);
1835	write_unlock_bh(&sk->sk_callback_lock);
1836}
1837
1838kuid_t sock_i_uid(struct sock *sk);
1839unsigned long sock_i_ino(struct sock *sk);
1840
1841static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1842{
1843	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1844}
1845
1846static inline u32 net_tx_rndhash(void)
1847{
1848	u32 v = prandom_u32();
1849
1850	return v ?: 1;
1851}
1852
1853static inline void sk_set_txhash(struct sock *sk)
1854{
1855	sk->sk_txhash = net_tx_rndhash();
 
1856}
1857
1858static inline void sk_rethink_txhash(struct sock *sk)
1859{
1860	if (sk->sk_txhash)
1861		sk_set_txhash(sk);
 
 
 
1862}
1863
1864static inline struct dst_entry *
1865__sk_dst_get(struct sock *sk)
1866{
1867	return rcu_dereference_check(sk->sk_dst_cache,
1868				     lockdep_sock_is_held(sk));
1869}
1870
1871static inline struct dst_entry *
1872sk_dst_get(struct sock *sk)
1873{
1874	struct dst_entry *dst;
1875
1876	rcu_read_lock();
1877	dst = rcu_dereference(sk->sk_dst_cache);
1878	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1879		dst = NULL;
1880	rcu_read_unlock();
1881	return dst;
1882}
1883
1884static inline void dst_negative_advice(struct sock *sk)
1885{
1886	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1887
1888	sk_rethink_txhash(sk);
1889
1890	if (dst && dst->ops->negative_advice) {
1891		ndst = dst->ops->negative_advice(dst);
1892
1893		if (ndst != dst) {
1894			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1895			sk_tx_queue_clear(sk);
1896			sk->sk_dst_pending_confirm = 0;
1897		}
1898	}
1899}
1900
 
 
 
 
 
 
1901static inline void
1902__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1903{
1904	struct dst_entry *old_dst;
1905
1906	sk_tx_queue_clear(sk);
1907	sk->sk_dst_pending_confirm = 0;
1908	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1909					    lockdep_sock_is_held(sk));
1910	rcu_assign_pointer(sk->sk_dst_cache, dst);
1911	dst_release(old_dst);
1912}
1913
1914static inline void
1915sk_dst_set(struct sock *sk, struct dst_entry *dst)
1916{
1917	struct dst_entry *old_dst;
1918
1919	sk_tx_queue_clear(sk);
1920	sk->sk_dst_pending_confirm = 0;
1921	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1922	dst_release(old_dst);
1923}
1924
1925static inline void
1926__sk_dst_reset(struct sock *sk)
1927{
1928	__sk_dst_set(sk, NULL);
1929}
1930
1931static inline void
1932sk_dst_reset(struct sock *sk)
1933{
1934	sk_dst_set(sk, NULL);
1935}
1936
1937struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1938
1939struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1940
1941static inline void sk_dst_confirm(struct sock *sk)
1942{
1943	if (!sk->sk_dst_pending_confirm)
1944		sk->sk_dst_pending_confirm = 1;
1945}
1946
1947static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1948{
1949	if (skb_get_dst_pending_confirm(skb)) {
1950		struct sock *sk = skb->sk;
1951		unsigned long now = jiffies;
1952
1953		/* avoid dirtying neighbour */
1954		if (n->confirmed != now)
1955			n->confirmed = now;
1956		if (sk && sk->sk_dst_pending_confirm)
1957			sk->sk_dst_pending_confirm = 0;
1958	}
1959}
1960
1961bool sk_mc_loop(struct sock *sk);
1962
1963static inline bool sk_can_gso(const struct sock *sk)
1964{
1965	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1966}
1967
1968void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1969
1970static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1971{
1972	sk->sk_route_nocaps |= flags;
1973	sk->sk_route_caps &= ~flags;
1974}
1975
1976static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1977					   struct iov_iter *from, char *to,
1978					   int copy, int offset)
1979{
1980	if (skb->ip_summed == CHECKSUM_NONE) {
1981		__wsum csum = 0;
1982		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1983			return -EFAULT;
1984		skb->csum = csum_block_add(skb->csum, csum, offset);
1985	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1986		if (!copy_from_iter_full_nocache(to, copy, from))
1987			return -EFAULT;
1988	} else if (!copy_from_iter_full(to, copy, from))
1989		return -EFAULT;
1990
1991	return 0;
1992}
1993
1994static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1995				       struct iov_iter *from, int copy)
1996{
1997	int err, offset = skb->len;
1998
1999	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2000				       copy, offset);
2001	if (err)
2002		__skb_trim(skb, offset);
2003
2004	return err;
2005}
2006
2007static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2008					   struct sk_buff *skb,
2009					   struct page *page,
2010					   int off, int copy)
2011{
2012	int err;
2013
2014	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2015				       copy, skb->len);
2016	if (err)
2017		return err;
2018
2019	skb->len	     += copy;
2020	skb->data_len	     += copy;
2021	skb->truesize	     += copy;
2022	sk_wmem_queued_add(sk, copy);
2023	sk_mem_charge(sk, copy);
2024	return 0;
2025}
2026
2027/**
2028 * sk_wmem_alloc_get - returns write allocations
2029 * @sk: socket
2030 *
2031 * Returns sk_wmem_alloc minus initial offset of one
2032 */
2033static inline int sk_wmem_alloc_get(const struct sock *sk)
2034{
2035	return refcount_read(&sk->sk_wmem_alloc) - 1;
2036}
2037
2038/**
2039 * sk_rmem_alloc_get - returns read allocations
2040 * @sk: socket
2041 *
2042 * Returns sk_rmem_alloc
2043 */
2044static inline int sk_rmem_alloc_get(const struct sock *sk)
2045{
2046	return atomic_read(&sk->sk_rmem_alloc);
2047}
2048
2049/**
2050 * sk_has_allocations - check if allocations are outstanding
2051 * @sk: socket
2052 *
2053 * Returns true if socket has write or read allocations
2054 */
2055static inline bool sk_has_allocations(const struct sock *sk)
2056{
2057	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2058}
2059
2060/**
2061 * skwq_has_sleeper - check if there are any waiting processes
2062 * @wq: struct socket_wq
2063 *
2064 * Returns true if socket_wq has waiting processes
2065 *
2066 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2067 * barrier call. They were added due to the race found within the tcp code.
2068 *
2069 * Consider following tcp code paths::
2070 *
2071 *   CPU1                CPU2
2072 *   sys_select          receive packet
2073 *   ...                 ...
2074 *   __add_wait_queue    update tp->rcv_nxt
2075 *   ...                 ...
2076 *   tp->rcv_nxt check   sock_def_readable
2077 *   ...                 {
2078 *   schedule               rcu_read_lock();
2079 *                          wq = rcu_dereference(sk->sk_wq);
2080 *                          if (wq && waitqueue_active(&wq->wait))
2081 *                              wake_up_interruptible(&wq->wait)
2082 *                          ...
2083 *                       }
2084 *
2085 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2086 * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2087 * could then endup calling schedule and sleep forever if there are no more
2088 * data on the socket.
2089 *
2090 */
2091static inline bool skwq_has_sleeper(struct socket_wq *wq)
2092{
2093	return wq && wq_has_sleeper(&wq->wait);
2094}
2095
2096/**
2097 * sock_poll_wait - place memory barrier behind the poll_wait call.
2098 * @filp:           file
2099 * @sock:           socket to wait on
2100 * @p:              poll_table
2101 *
2102 * See the comments in the wq_has_sleeper function.
2103 */
2104static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2105				  poll_table *p)
2106{
2107	if (!poll_does_not_wait(p)) {
2108		poll_wait(filp, &sock->wq.wait, p);
2109		/* We need to be sure we are in sync with the
2110		 * socket flags modification.
2111		 *
2112		 * This memory barrier is paired in the wq_has_sleeper.
2113		 */
2114		smp_mb();
2115	}
2116}
2117
2118static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2119{
2120	if (sk->sk_txhash) {
 
 
 
2121		skb->l4_hash = 1;
2122		skb->hash = sk->sk_txhash;
2123	}
2124}
2125
2126void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2127
2128/*
2129 *	Queue a received datagram if it will fit. Stream and sequenced
2130 *	protocols can't normally use this as they need to fit buffers in
2131 *	and play with them.
2132 *
2133 *	Inlined as it's very short and called for pretty much every
2134 *	packet ever received.
2135 */
2136static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2137{
2138	skb_orphan(skb);
2139	skb->sk = sk;
2140	skb->destructor = sock_rfree;
2141	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2142	sk_mem_charge(sk, skb->truesize);
2143}
2144
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2145void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2146		    unsigned long expires);
2147
2148void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2149
 
 
2150int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2151			struct sk_buff *skb, unsigned int flags,
2152			void (*destructor)(struct sock *sk,
2153					   struct sk_buff *skb));
2154int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2155int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 
 
 
 
 
 
 
2156
2157int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2158struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2159
2160/*
2161 *	Recover an error report and clear atomically
2162 */
2163
2164static inline int sock_error(struct sock *sk)
2165{
2166	int err;
2167	if (likely(!sk->sk_err))
 
 
 
 
2168		return 0;
 
2169	err = xchg(&sk->sk_err, 0);
2170	return -err;
2171}
2172
 
 
2173static inline unsigned long sock_wspace(struct sock *sk)
2174{
2175	int amt = 0;
2176
2177	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2178		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2179		if (amt < 0)
2180			amt = 0;
2181	}
2182	return amt;
2183}
2184
2185/* Note:
2186 *  We use sk->sk_wq_raw, from contexts knowing this
2187 *  pointer is not NULL and cannot disappear/change.
2188 */
2189static inline void sk_set_bit(int nr, struct sock *sk)
2190{
2191	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2192	    !sock_flag(sk, SOCK_FASYNC))
2193		return;
2194
2195	set_bit(nr, &sk->sk_wq_raw->flags);
2196}
2197
2198static inline void sk_clear_bit(int nr, struct sock *sk)
2199{
2200	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2201	    !sock_flag(sk, SOCK_FASYNC))
2202		return;
2203
2204	clear_bit(nr, &sk->sk_wq_raw->flags);
2205}
2206
2207static inline void sk_wake_async(const struct sock *sk, int how, int band)
2208{
2209	if (sock_flag(sk, SOCK_FASYNC)) {
2210		rcu_read_lock();
2211		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2212		rcu_read_unlock();
2213	}
2214}
2215
2216/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2217 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2218 * Note: for send buffers, TCP works better if we can build two skbs at
2219 * minimum.
2220 */
2221#define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2222
2223#define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2224#define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2225
2226static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2227{
2228	u32 val;
2229
2230	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2231		return;
2232
2233	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
 
2234
2235	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2236}
2237
2238struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2239				    bool force_schedule);
2240
2241/**
2242 * sk_page_frag - return an appropriate page_frag
2243 * @sk: socket
2244 *
2245 * Use the per task page_frag instead of the per socket one for
2246 * optimization when we know that we're in the normal context and owns
2247 * everything that's associated with %current.
2248 *
2249 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2250 * inside other socket operations and end up recursing into sk_page_frag()
2251 * while it's already in use.
 
 
 
 
2252 */
2253static inline struct page_frag *sk_page_frag(struct sock *sk)
2254{
2255	if (gfpflags_normal_context(sk->sk_allocation))
2256		return &current->task_frag;
2257
2258	return &sk->sk_frag;
2259}
2260
2261bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2262
2263/*
2264 *	Default write policy as shown to user space via poll/select/SIGIO
2265 */
2266static inline bool sock_writeable(const struct sock *sk)
2267{
2268	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2269}
2270
2271static inline gfp_t gfp_any(void)
2272{
2273	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2274}
2275
 
 
 
 
 
2276static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2277{
2278	return noblock ? 0 : sk->sk_rcvtimeo;
2279}
2280
2281static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2282{
2283	return noblock ? 0 : sk->sk_sndtimeo;
2284}
2285
2286static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2287{
2288	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2289
2290	return v ?: 1;
2291}
2292
2293/* Alas, with timeout socket operations are not restartable.
2294 * Compare this to poll().
2295 */
2296static inline int sock_intr_errno(long timeo)
2297{
2298	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2299}
2300
2301struct sock_skb_cb {
2302	u32 dropcount;
2303};
2304
2305/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2306 * using skb->cb[] would keep using it directly and utilize its
2307 * alignement guarantee.
2308 */
2309#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2310			    sizeof(struct sock_skb_cb)))
2311
2312#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2313			    SOCK_SKB_CB_OFFSET))
2314
2315#define sock_skb_cb_check_size(size) \
2316	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2317
2318static inline void
2319sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2320{
2321	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2322						atomic_read(&sk->sk_drops) : 0;
2323}
2324
2325static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2326{
2327	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2328
2329	atomic_add(segs, &sk->sk_drops);
2330}
2331
2332static inline ktime_t sock_read_timestamp(struct sock *sk)
2333{
2334#if BITS_PER_LONG==32
2335	unsigned int seq;
2336	ktime_t kt;
2337
2338	do {
2339		seq = read_seqbegin(&sk->sk_stamp_seq);
2340		kt = sk->sk_stamp;
2341	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2342
2343	return kt;
2344#else
2345	return READ_ONCE(sk->sk_stamp);
2346#endif
2347}
2348
2349static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2350{
2351#if BITS_PER_LONG==32
2352	write_seqlock(&sk->sk_stamp_seq);
2353	sk->sk_stamp = kt;
2354	write_sequnlock(&sk->sk_stamp_seq);
2355#else
2356	WRITE_ONCE(sk->sk_stamp, kt);
2357#endif
2358}
2359
2360void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2361			   struct sk_buff *skb);
2362void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2363			     struct sk_buff *skb);
2364
2365static inline void
2366sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2367{
2368	ktime_t kt = skb->tstamp;
2369	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2370
2371	/*
2372	 * generate control messages if
2373	 * - receive time stamping in software requested
2374	 * - software time stamp available and wanted
2375	 * - hardware time stamps available and wanted
2376	 */
2377	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2378	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2379	    (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2380	    (hwtstamps->hwtstamp &&
2381	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2382		__sock_recv_timestamp(msg, sk, skb);
2383	else
2384		sock_write_timestamp(sk, kt);
2385
2386	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2387		__sock_recv_wifi_status(msg, sk, skb);
2388}
2389
2390void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2391			      struct sk_buff *skb);
2392
2393#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2394static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2395					  struct sk_buff *skb)
2396{
2397#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2398			   (1UL << SOCK_RCVTSTAMP))
 
2399#define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2400			   SOF_TIMESTAMPING_RAW_HARDWARE)
2401
2402	if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2403		__sock_recv_ts_and_drops(msg, sk, skb);
2404	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2405		sock_write_timestamp(sk, skb->tstamp);
2406	else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2407		sock_write_timestamp(sk, 0);
2408}
2409
2410void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2411
2412/**
2413 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2414 * @sk:		socket sending this packet
2415 * @tsflags:	timestamping flags to use
2416 * @tx_flags:	completed with instructions for time stamping
2417 * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2418 *
2419 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2420 */
2421static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2422				      __u8 *tx_flags, __u32 *tskey)
2423{
2424	if (unlikely(tsflags)) {
2425		__sock_tx_timestamp(tsflags, tx_flags);
2426		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2427		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2428			*tskey = sk->sk_tskey++;
2429	}
2430	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2431		*tx_flags |= SKBTX_WIFI_STATUS;
2432}
2433
2434static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2435				     __u8 *tx_flags)
2436{
2437	_sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2438}
2439
2440static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2441{
2442	_sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2443			   &skb_shinfo(skb)->tskey);
2444}
2445
 
 
 
 
 
2446/**
2447 * sk_eat_skb - Release a skb if it is no longer needed
2448 * @sk: socket to eat this skb from
2449 * @skb: socket buffer to eat
2450 *
2451 * This routine must be called with interrupts disabled or with the socket
2452 * locked so that the sk_buff queue operation is ok.
2453*/
2454DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2455static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2456{
2457	__skb_unlink(skb, &sk->sk_receive_queue);
2458	if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2459	    !sk->sk_rx_skb_cache) {
2460		sk->sk_rx_skb_cache = skb;
2461		skb_orphan(skb);
2462		return;
2463	}
2464	__kfree_skb(skb);
2465}
2466
2467static inline
2468struct net *sock_net(const struct sock *sk)
2469{
2470	return read_pnet(&sk->sk_net);
 
 
 
 
2471}
2472
2473static inline
2474void sock_net_set(struct sock *sk, struct net *net)
 
 
2475{
2476	write_pnet(&sk->sk_net, net);
 
 
 
 
 
 
 
2477}
2478
2479static inline struct sock *skb_steal_sock(struct sk_buff *skb)
 
 
 
 
 
 
2480{
2481	if (skb->sk) {
2482		struct sock *sk = skb->sk;
2483
 
 
 
2484		skb->destructor = NULL;
2485		skb->sk = NULL;
2486		return sk;
2487	}
 
2488	return NULL;
2489}
2490
2491/* This helper checks if a socket is a full socket,
2492 * ie _not_ a timewait or request socket.
2493 */
2494static inline bool sk_fullsock(const struct sock *sk)
2495{
2496	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2497}
2498
2499/* Checks if this SKB belongs to an HW offloaded socket
2500 * and whether any SW fallbacks are required based on dev.
2501 * Check decrypted mark in case skb_orphan() cleared socket.
2502 */
2503static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2504						   struct net_device *dev)
2505{
2506#ifdef CONFIG_SOCK_VALIDATE_XMIT
2507	struct sock *sk = skb->sk;
2508
2509	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2510		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2511#ifdef CONFIG_TLS_DEVICE
2512	} else if (unlikely(skb->decrypted)) {
2513		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2514		kfree_skb(skb);
2515		skb = NULL;
2516#endif
2517	}
2518#endif
2519
2520	return skb;
2521}
2522
2523/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2524 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2525 */
2526static inline bool sk_listener(const struct sock *sk)
2527{
2528	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2529}
2530
2531void sock_enable_timestamp(struct sock *sk, int flag);
2532int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2533		       int type);
2534
2535bool sk_ns_capable(const struct sock *sk,
2536		   struct user_namespace *user_ns, int cap);
2537bool sk_capable(const struct sock *sk, int cap);
2538bool sk_net_capable(const struct sock *sk, int cap);
2539
2540void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2541
2542/* Take into consideration the size of the struct sk_buff overhead in the
2543 * determination of these values, since that is non-constant across
2544 * platforms.  This makes socket queueing behavior and performance
2545 * not depend upon such differences.
2546 */
2547#define _SK_MEM_PACKETS		256
2548#define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2549#define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2550#define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2551
2552extern __u32 sysctl_wmem_max;
2553extern __u32 sysctl_rmem_max;
2554
2555extern int sysctl_tstamp_allow_data;
2556extern int sysctl_optmem_max;
2557
2558extern __u32 sysctl_wmem_default;
2559extern __u32 sysctl_rmem_default;
2560
 
2561DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2562
2563static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2564{
2565	/* Does this proto have per netns sysctl_wmem ? */
2566	if (proto->sysctl_wmem_offset)
2567		return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2568
2569	return *proto->sysctl_wmem;
2570}
2571
2572static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2573{
2574	/* Does this proto have per netns sysctl_rmem ? */
2575	if (proto->sysctl_rmem_offset)
2576		return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2577
2578	return *proto->sysctl_rmem;
2579}
2580
2581/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2582 * Some wifi drivers need to tweak it to get more chunks.
2583 * They can use this helper from their ndo_start_xmit()
2584 */
2585static inline void sk_pacing_shift_update(struct sock *sk, int val)
2586{
2587	if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2588		return;
2589	sk->sk_pacing_shift = val;
2590}
2591
2592/* if a socket is bound to a device, check that the given device
2593 * index is either the same or that the socket is bound to an L3
2594 * master device and the given device index is also enslaved to
2595 * that L3 master
2596 */
2597static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2598{
 
2599	int mdif;
2600
2601	if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2602		return true;
2603
2604	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2605	if (mdif && mdif == sk->sk_bound_dev_if)
2606		return true;
2607
2608	return false;
2609}
2610
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2611#endif	/* _SOCK_H */
v6.2
   1/* SPDX-License-Identifier: GPL-2.0-or-later */
   2/*
   3 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   4 *		operating system.  INET is implemented using the  BSD Socket
   5 *		interface as the means of communication with the user level.
   6 *
   7 *		Definitions for the AF_INET socket handler.
   8 *
   9 * Version:	@(#)sock.h	1.0.4	05/13/93
  10 *
  11 * Authors:	Ross Biro
  12 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  13 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
  14 *		Florian La Roche <flla@stud.uni-sb.de>
  15 *
  16 * Fixes:
  17 *		Alan Cox	:	Volatiles in skbuff pointers. See
  18 *					skbuff comments. May be overdone,
  19 *					better to prove they can be removed
  20 *					than the reverse.
  21 *		Alan Cox	:	Added a zapped field for tcp to note
  22 *					a socket is reset and must stay shut up
  23 *		Alan Cox	:	New fields for options
  24 *	Pauline Middelink	:	identd support
  25 *		Alan Cox	:	Eliminate low level recv/recvfrom
  26 *		David S. Miller	:	New socket lookup architecture.
  27 *              Steve Whitehouse:       Default routines for sock_ops
  28 *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
  29 *              			protinfo be just a void pointer, as the
  30 *              			protocol specific parts were moved to
  31 *              			respective headers and ipv4/v6, etc now
  32 *              			use private slabcaches for its socks
  33 *              Pedro Hortas	:	New flags field for socket options
  34 */
  35#ifndef _SOCK_H
  36#define _SOCK_H
  37
  38#include <linux/hardirq.h>
  39#include <linux/kernel.h>
  40#include <linux/list.h>
  41#include <linux/list_nulls.h>
  42#include <linux/timer.h>
  43#include <linux/cache.h>
  44#include <linux/bitops.h>
  45#include <linux/lockdep.h>
  46#include <linux/netdevice.h>
  47#include <linux/skbuff.h>	/* struct sk_buff */
  48#include <linux/mm.h>
  49#include <linux/security.h>
  50#include <linux/slab.h>
  51#include <linux/uaccess.h>
  52#include <linux/page_counter.h>
  53#include <linux/memcontrol.h>
  54#include <linux/static_key.h>
  55#include <linux/sched.h>
  56#include <linux/wait.h>
  57#include <linux/cgroup-defs.h>
  58#include <linux/rbtree.h>
 
  59#include <linux/rculist_nulls.h>
  60#include <linux/poll.h>
  61#include <linux/sockptr.h>
  62#include <linux/indirect_call_wrapper.h>
  63#include <linux/atomic.h>
  64#include <linux/refcount.h>
  65#include <linux/llist.h>
  66#include <net/dst.h>
  67#include <net/checksum.h>
  68#include <net/tcp_states.h>
  69#include <linux/net_tstamp.h>
 
  70#include <net/l3mdev.h>
  71#include <uapi/linux/socket.h>
  72
  73/*
  74 * This structure really needs to be cleaned up.
  75 * Most of it is for TCP, and not used by any of
  76 * the other protocols.
  77 */
  78
  79/* Define this to get the SOCK_DBG debugging facility. */
  80#define SOCK_DEBUGGING
  81#ifdef SOCK_DEBUGGING
  82#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
  83					printk(KERN_DEBUG msg); } while (0)
  84#else
  85/* Validate arguments and do nothing */
  86static inline __printf(2, 3)
  87void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
  88{
  89}
  90#endif
  91
  92/* This is the per-socket lock.  The spinlock provides a synchronization
  93 * between user contexts and software interrupt processing, whereas the
  94 * mini-semaphore synchronizes multiple users amongst themselves.
  95 */
  96typedef struct {
  97	spinlock_t		slock;
  98	int			owned;
  99	wait_queue_head_t	wq;
 100	/*
 101	 * We express the mutex-alike socket_lock semantics
 102	 * to the lock validator by explicitly managing
 103	 * the slock as a lock variant (in addition to
 104	 * the slock itself):
 105	 */
 106#ifdef CONFIG_DEBUG_LOCK_ALLOC
 107	struct lockdep_map dep_map;
 108#endif
 109} socket_lock_t;
 110
 111struct sock;
 112struct proto;
 113struct net;
 114
 115typedef __u32 __bitwise __portpair;
 116typedef __u64 __bitwise __addrpair;
 117
 118/**
 119 *	struct sock_common - minimal network layer representation of sockets
 120 *	@skc_daddr: Foreign IPv4 addr
 121 *	@skc_rcv_saddr: Bound local IPv4 addr
 122 *	@skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
 123 *	@skc_hash: hash value used with various protocol lookup tables
 124 *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
 125 *	@skc_dport: placeholder for inet_dport/tw_dport
 126 *	@skc_num: placeholder for inet_num/tw_num
 127 *	@skc_portpair: __u32 union of @skc_dport & @skc_num
 128 *	@skc_family: network address family
 129 *	@skc_state: Connection state
 130 *	@skc_reuse: %SO_REUSEADDR setting
 131 *	@skc_reuseport: %SO_REUSEPORT setting
 132 *	@skc_ipv6only: socket is IPV6 only
 133 *	@skc_net_refcnt: socket is using net ref counting
 134 *	@skc_bound_dev_if: bound device index if != 0
 135 *	@skc_bind_node: bind hash linkage for various protocol lookup tables
 136 *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
 137 *	@skc_prot: protocol handlers inside a network family
 138 *	@skc_net: reference to the network namespace of this socket
 139 *	@skc_v6_daddr: IPV6 destination address
 140 *	@skc_v6_rcv_saddr: IPV6 source address
 141 *	@skc_cookie: socket's cookie value
 142 *	@skc_node: main hash linkage for various protocol lookup tables
 143 *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
 144 *	@skc_tx_queue_mapping: tx queue number for this connection
 145 *	@skc_rx_queue_mapping: rx queue number for this connection
 146 *	@skc_flags: place holder for sk_flags
 147 *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
 148 *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
 149 *	@skc_listener: connection request listener socket (aka rsk_listener)
 150 *		[union with @skc_flags]
 151 *	@skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
 152 *		[union with @skc_flags]
 153 *	@skc_incoming_cpu: record/match cpu processing incoming packets
 154 *	@skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
 155 *		[union with @skc_incoming_cpu]
 156 *	@skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
 157 *		[union with @skc_incoming_cpu]
 158 *	@skc_refcnt: reference count
 159 *
 160 *	This is the minimal network layer representation of sockets, the header
 161 *	for struct sock and struct inet_timewait_sock.
 162 */
 163struct sock_common {
 
 
 
 164	union {
 165		__addrpair	skc_addrpair;
 166		struct {
 167			__be32	skc_daddr;
 168			__be32	skc_rcv_saddr;
 169		};
 170	};
 171	union  {
 172		unsigned int	skc_hash;
 173		__u16		skc_u16hashes[2];
 174	};
 175	/* skc_dport && skc_num must be grouped as well */
 176	union {
 177		__portpair	skc_portpair;
 178		struct {
 179			__be16	skc_dport;
 180			__u16	skc_num;
 181		};
 182	};
 183
 184	unsigned short		skc_family;
 185	volatile unsigned char	skc_state;
 186	unsigned char		skc_reuse:4;
 187	unsigned char		skc_reuseport:1;
 188	unsigned char		skc_ipv6only:1;
 189	unsigned char		skc_net_refcnt:1;
 190	int			skc_bound_dev_if;
 191	union {
 192		struct hlist_node	skc_bind_node;
 193		struct hlist_node	skc_portaddr_node;
 194	};
 195	struct proto		*skc_prot;
 196	possible_net_t		skc_net;
 197
 198#if IS_ENABLED(CONFIG_IPV6)
 199	struct in6_addr		skc_v6_daddr;
 200	struct in6_addr		skc_v6_rcv_saddr;
 201#endif
 202
 203	atomic64_t		skc_cookie;
 204
 205	/* following fields are padding to force
 206	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
 207	 * assuming IPV6 is enabled. We use this padding differently
 208	 * for different kind of 'sockets'
 209	 */
 210	union {
 211		unsigned long	skc_flags;
 212		struct sock	*skc_listener; /* request_sock */
 213		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
 214	};
 215	/*
 216	 * fields between dontcopy_begin/dontcopy_end
 217	 * are not copied in sock_copy()
 218	 */
 219	/* private: */
 220	int			skc_dontcopy_begin[0];
 221	/* public: */
 222	union {
 223		struct hlist_node	skc_node;
 224		struct hlist_nulls_node skc_nulls_node;
 225	};
 226	unsigned short		skc_tx_queue_mapping;
 227#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
 228	unsigned short		skc_rx_queue_mapping;
 229#endif
 230	union {
 231		int		skc_incoming_cpu;
 232		u32		skc_rcv_wnd;
 233		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
 234	};
 235
 236	refcount_t		skc_refcnt;
 237	/* private: */
 238	int                     skc_dontcopy_end[0];
 239	union {
 240		u32		skc_rxhash;
 241		u32		skc_window_clamp;
 242		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
 243	};
 244	/* public: */
 245};
 246
 247struct bpf_local_storage;
 248struct sk_filter;
 249
 250/**
 251  *	struct sock - network layer representation of sockets
 252  *	@__sk_common: shared layout with inet_timewait_sock
 253  *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
 254  *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
 255  *	@sk_lock:	synchronizer
 256  *	@sk_kern_sock: True if sock is using kernel lock classes
 257  *	@sk_rcvbuf: size of receive buffer in bytes
 258  *	@sk_wq: sock wait queue and async head
 259  *	@sk_rx_dst: receive input route used by early demux
 260  *	@sk_rx_dst_ifindex: ifindex for @sk_rx_dst
 261  *	@sk_rx_dst_cookie: cookie for @sk_rx_dst
 262  *	@sk_dst_cache: destination cache
 263  *	@sk_dst_pending_confirm: need to confirm neighbour
 264  *	@sk_policy: flow policy
 265  *	@sk_receive_queue: incoming packets
 266  *	@sk_wmem_alloc: transmit queue bytes committed
 267  *	@sk_tsq_flags: TCP Small Queues flags
 268  *	@sk_write_queue: Packet sending queue
 269  *	@sk_omem_alloc: "o" is "option" or "other"
 270  *	@sk_wmem_queued: persistent queue size
 271  *	@sk_forward_alloc: space allocated forward
 272  *	@sk_reserved_mem: space reserved and non-reclaimable for the socket
 273  *	@sk_napi_id: id of the last napi context to receive data for sk
 274  *	@sk_ll_usec: usecs to busypoll when there is no data
 275  *	@sk_allocation: allocation mode
 276  *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
 277  *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
 278  *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
 279  *	@sk_sndbuf: size of send buffer in bytes
 280  *	@__sk_flags_offset: empty field used to determine location of bitfield
 281  *	@sk_padding: unused element for alignment
 282  *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
 283  *	@sk_no_check_rx: allow zero checksum in RX packets
 284  *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
 285  *	@sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
 286  *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
 287  *	@sk_gso_max_size: Maximum GSO segment size to build
 288  *	@sk_gso_max_segs: Maximum number of GSO segments
 289  *	@sk_pacing_shift: scaling factor for TCP Small Queues
 290  *	@sk_lingertime: %SO_LINGER l_linger setting
 291  *	@sk_backlog: always used with the per-socket spinlock held
 292  *	@sk_callback_lock: used with the callbacks in the end of this struct
 293  *	@sk_error_queue: rarely used
 294  *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
 295  *			  IPV6_ADDRFORM for instance)
 296  *	@sk_err: last error
 297  *	@sk_err_soft: errors that don't cause failure but are the cause of a
 298  *		      persistent failure not just 'timed out'
 299  *	@sk_drops: raw/udp drops counter
 300  *	@sk_ack_backlog: current listen backlog
 301  *	@sk_max_ack_backlog: listen backlog set in listen()
 302  *	@sk_uid: user id of owner
 303  *	@sk_prefer_busy_poll: prefer busypolling over softirq processing
 304  *	@sk_busy_poll_budget: napi processing budget when busypolling
 305  *	@sk_priority: %SO_PRIORITY setting
 306  *	@sk_type: socket type (%SOCK_STREAM, etc)
 307  *	@sk_protocol: which protocol this socket belongs in this network family
 308  *	@sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
 309  *	@sk_peer_pid: &struct pid for this socket's peer
 310  *	@sk_peer_cred: %SO_PEERCRED setting
 311  *	@sk_rcvlowat: %SO_RCVLOWAT setting
 312  *	@sk_rcvtimeo: %SO_RCVTIMEO setting
 313  *	@sk_sndtimeo: %SO_SNDTIMEO setting
 314  *	@sk_txhash: computed flow hash for use on transmit
 315  *	@sk_txrehash: enable TX hash rethink
 316  *	@sk_filter: socket filtering instructions
 317  *	@sk_timer: sock cleanup timer
 318  *	@sk_stamp: time stamp of last packet received
 319  *	@sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
 320  *	@sk_tsflags: SO_TIMESTAMPING flags
 321  *	@sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
 322  *			   Sockets that can be used under memory reclaim should
 323  *			   set this to false.
 324  *	@sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
 325  *	              for timestamping
 326  *	@sk_tskey: counter to disambiguate concurrent tstamp requests
 327  *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
 328  *	@sk_socket: Identd and reporting IO signals
 329  *	@sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
 330  *	@sk_frag: cached page frag
 331  *	@sk_peek_off: current peek_offset value
 332  *	@sk_send_head: front of stuff to transmit
 333  *	@tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
 334  *	@sk_security: used by security modules
 335  *	@sk_mark: generic packet mark
 336  *	@sk_cgrp_data: cgroup data for this cgroup
 337  *	@sk_memcg: this socket's memory cgroup association
 338  *	@sk_write_pending: a write to stream socket waits to start
 339  *	@sk_state_change: callback to indicate change in the state of the sock
 340  *	@sk_data_ready: callback to indicate there is data to be processed
 341  *	@sk_write_space: callback to indicate there is bf sending space available
 342  *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
 343  *	@sk_backlog_rcv: callback to process the backlog
 344  *	@sk_validate_xmit_skb: ptr to an optional validate function
 345  *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
 346  *	@sk_reuseport_cb: reuseport group container
 347  *	@sk_bpf_storage: ptr to cache and control for bpf_sk_storage
 348  *	@sk_rcu: used during RCU grace period
 349  *	@sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
 350  *	@sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
 351  *	@sk_txtime_report_errors: set report errors mode for SO_TXTIME
 352  *	@sk_txtime_unused: unused txtime flags
 353  *	@ns_tracker: tracker for netns reference
 354  *	@sk_bind2_node: bind node in the bhash2 table
 355  */
 356struct sock {
 357	/*
 358	 * Now struct inet_timewait_sock also uses sock_common, so please just
 359	 * don't add nothing before this first member (__sk_common) --acme
 360	 */
 361	struct sock_common	__sk_common;
 362#define sk_node			__sk_common.skc_node
 363#define sk_nulls_node		__sk_common.skc_nulls_node
 364#define sk_refcnt		__sk_common.skc_refcnt
 365#define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
 366#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
 367#define sk_rx_queue_mapping	__sk_common.skc_rx_queue_mapping
 368#endif
 369
 370#define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
 371#define sk_dontcopy_end		__sk_common.skc_dontcopy_end
 372#define sk_hash			__sk_common.skc_hash
 373#define sk_portpair		__sk_common.skc_portpair
 374#define sk_num			__sk_common.skc_num
 375#define sk_dport		__sk_common.skc_dport
 376#define sk_addrpair		__sk_common.skc_addrpair
 377#define sk_daddr		__sk_common.skc_daddr
 378#define sk_rcv_saddr		__sk_common.skc_rcv_saddr
 379#define sk_family		__sk_common.skc_family
 380#define sk_state		__sk_common.skc_state
 381#define sk_reuse		__sk_common.skc_reuse
 382#define sk_reuseport		__sk_common.skc_reuseport
 383#define sk_ipv6only		__sk_common.skc_ipv6only
 384#define sk_net_refcnt		__sk_common.skc_net_refcnt
 385#define sk_bound_dev_if		__sk_common.skc_bound_dev_if
 386#define sk_bind_node		__sk_common.skc_bind_node
 387#define sk_prot			__sk_common.skc_prot
 388#define sk_net			__sk_common.skc_net
 389#define sk_v6_daddr		__sk_common.skc_v6_daddr
 390#define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
 391#define sk_cookie		__sk_common.skc_cookie
 392#define sk_incoming_cpu		__sk_common.skc_incoming_cpu
 393#define sk_flags		__sk_common.skc_flags
 394#define sk_rxhash		__sk_common.skc_rxhash
 395
 396	/* early demux fields */
 397	struct dst_entry __rcu	*sk_rx_dst;
 398	int			sk_rx_dst_ifindex;
 399	u32			sk_rx_dst_cookie;
 400
 401	socket_lock_t		sk_lock;
 402	atomic_t		sk_drops;
 403	int			sk_rcvlowat;
 404	struct sk_buff_head	sk_error_queue;
 
 405	struct sk_buff_head	sk_receive_queue;
 406	/*
 407	 * The backlog queue is special, it is always used with
 408	 * the per-socket spinlock held and requires low latency
 409	 * access. Therefore we special case it's implementation.
 410	 * Note : rmem_alloc is in this structure to fill a hole
 411	 * on 64bit arches, not because its logically part of
 412	 * backlog.
 413	 */
 414	struct {
 415		atomic_t	rmem_alloc;
 416		int		len;
 417		struct sk_buff	*head;
 418		struct sk_buff	*tail;
 419	} sk_backlog;
 420
 421#define sk_rmem_alloc sk_backlog.rmem_alloc
 422
 423	int			sk_forward_alloc;
 424	u32			sk_reserved_mem;
 425#ifdef CONFIG_NET_RX_BUSY_POLL
 426	unsigned int		sk_ll_usec;
 427	/* ===== mostly read cache line ===== */
 428	unsigned int		sk_napi_id;
 429#endif
 430	int			sk_rcvbuf;
 431
 432	struct sk_filter __rcu	*sk_filter;
 433	union {
 434		struct socket_wq __rcu	*sk_wq;
 435		/* private: */
 436		struct socket_wq	*sk_wq_raw;
 437		/* public: */
 438	};
 439#ifdef CONFIG_XFRM
 440	struct xfrm_policy __rcu *sk_policy[2];
 441#endif
 442
 443	struct dst_entry __rcu	*sk_dst_cache;
 444	atomic_t		sk_omem_alloc;
 445	int			sk_sndbuf;
 446
 447	/* ===== cache line for TX ===== */
 448	int			sk_wmem_queued;
 449	refcount_t		sk_wmem_alloc;
 450	unsigned long		sk_tsq_flags;
 451	union {
 452		struct sk_buff	*sk_send_head;
 453		struct rb_root	tcp_rtx_queue;
 454	};
 
 455	struct sk_buff_head	sk_write_queue;
 456	__s32			sk_peek_off;
 457	int			sk_write_pending;
 458	__u32			sk_dst_pending_confirm;
 459	u32			sk_pacing_status; /* see enum sk_pacing */
 460	long			sk_sndtimeo;
 461	struct timer_list	sk_timer;
 462	__u32			sk_priority;
 463	__u32			sk_mark;
 464	unsigned long		sk_pacing_rate; /* bytes per second */
 465	unsigned long		sk_max_pacing_rate;
 466	struct page_frag	sk_frag;
 467	netdev_features_t	sk_route_caps;
 
 
 468	int			sk_gso_type;
 469	unsigned int		sk_gso_max_size;
 470	gfp_t			sk_allocation;
 471	__u32			sk_txhash;
 472
 473	/*
 474	 * Because of non atomicity rules, all
 475	 * changes are protected by socket lock.
 476	 */
 477	u8			sk_gso_disabled : 1,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 478				sk_kern_sock : 1,
 479				sk_no_check_tx : 1,
 480				sk_no_check_rx : 1,
 481				sk_userlocks : 4;
 
 
 
 
 482	u8			sk_pacing_shift;
 483	u16			sk_type;
 484	u16			sk_protocol;
 485	u16			sk_gso_max_segs;
 486	unsigned long	        sk_lingertime;
 487	struct proto		*sk_prot_creator;
 488	rwlock_t		sk_callback_lock;
 489	int			sk_err,
 490				sk_err_soft;
 491	u32			sk_ack_backlog;
 492	u32			sk_max_ack_backlog;
 493	kuid_t			sk_uid;
 494	u8			sk_txrehash;
 495#ifdef CONFIG_NET_RX_BUSY_POLL
 496	u8			sk_prefer_busy_poll;
 497	u16			sk_busy_poll_budget;
 498#endif
 499	spinlock_t		sk_peer_lock;
 500	int			sk_bind_phc;
 501	struct pid		*sk_peer_pid;
 502	const struct cred	*sk_peer_cred;
 503
 504	long			sk_rcvtimeo;
 505	ktime_t			sk_stamp;
 506#if BITS_PER_LONG==32
 507	seqlock_t		sk_stamp_seq;
 508#endif
 509	atomic_t		sk_tskey;
 
 
 510	atomic_t		sk_zckey;
 511	u32			sk_tsflags;
 512	u8			sk_shutdown;
 513
 514	u8			sk_clockid;
 515	u8			sk_txtime_deadline_mode : 1,
 516				sk_txtime_report_errors : 1,
 517				sk_txtime_unused : 6;
 518	bool			sk_use_task_frag;
 519
 520	struct socket		*sk_socket;
 521	void			*sk_user_data;
 522#ifdef CONFIG_SECURITY
 523	void			*sk_security;
 524#endif
 525	struct sock_cgroup_data	sk_cgrp_data;
 526	struct mem_cgroup	*sk_memcg;
 527	void			(*sk_state_change)(struct sock *sk);
 528	void			(*sk_data_ready)(struct sock *sk);
 529	void			(*sk_write_space)(struct sock *sk);
 530	void			(*sk_error_report)(struct sock *sk);
 531	int			(*sk_backlog_rcv)(struct sock *sk,
 532						  struct sk_buff *skb);
 533#ifdef CONFIG_SOCK_VALIDATE_XMIT
 534	struct sk_buff*		(*sk_validate_xmit_skb)(struct sock *sk,
 535							struct net_device *dev,
 536							struct sk_buff *skb);
 537#endif
 538	void                    (*sk_destruct)(struct sock *sk);
 539	struct sock_reuseport __rcu	*sk_reuseport_cb;
 540#ifdef CONFIG_BPF_SYSCALL
 541	struct bpf_local_storage __rcu	*sk_bpf_storage;
 542#endif
 543	struct rcu_head		sk_rcu;
 544	netns_tracker		ns_tracker;
 545	struct hlist_node	sk_bind2_node;
 546};
 547
 548enum sk_pacing {
 549	SK_PACING_NONE		= 0,
 550	SK_PACING_NEEDED	= 1,
 551	SK_PACING_FQ		= 2,
 552};
 553
 554/* flag bits in sk_user_data
 555 *
 556 * - SK_USER_DATA_NOCOPY:      Pointer stored in sk_user_data might
 557 *   not be suitable for copying when cloning the socket. For instance,
 558 *   it can point to a reference counted object. sk_user_data bottom
 559 *   bit is set if pointer must not be copied.
 560 *
 561 * - SK_USER_DATA_BPF:         Mark whether sk_user_data field is
 562 *   managed/owned by a BPF reuseport array. This bit should be set
 563 *   when sk_user_data's sk is added to the bpf's reuseport_array.
 564 *
 565 * - SK_USER_DATA_PSOCK:       Mark whether pointer stored in
 566 *   sk_user_data points to psock type. This bit should be set
 567 *   when sk_user_data is assigned to a psock object.
 568 */
 569#define SK_USER_DATA_NOCOPY	1UL
 570#define SK_USER_DATA_BPF	2UL
 571#define SK_USER_DATA_PSOCK	4UL
 572#define SK_USER_DATA_PTRMASK	~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
 573				  SK_USER_DATA_PSOCK)
 574
 575/**
 576 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
 577 * @sk: socket
 578 */
 579static inline bool sk_user_data_is_nocopy(const struct sock *sk)
 580{
 581	return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
 582}
 583
 584#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
 585
 586/**
 587 * __locked_read_sk_user_data_with_flags - return the pointer
 588 * only if argument flags all has been set in sk_user_data. Otherwise
 589 * return NULL
 590 *
 591 * @sk: socket
 592 * @flags: flag bits
 593 *
 594 * The caller must be holding sk->sk_callback_lock.
 595 */
 596static inline void *
 597__locked_read_sk_user_data_with_flags(const struct sock *sk,
 598				      uintptr_t flags)
 599{
 600	uintptr_t sk_user_data =
 601		(uintptr_t)rcu_dereference_check(__sk_user_data(sk),
 602						 lockdep_is_held(&sk->sk_callback_lock));
 603
 604	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
 605
 606	if ((sk_user_data & flags) == flags)
 607		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
 608	return NULL;
 609}
 610
 611/**
 612 * __rcu_dereference_sk_user_data_with_flags - return the pointer
 613 * only if argument flags all has been set in sk_user_data. Otherwise
 614 * return NULL
 615 *
 616 * @sk: socket
 617 * @flags: flag bits
 618 */
 619static inline void *
 620__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
 621					  uintptr_t flags)
 622{
 623	uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
 624
 625	WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
 626
 627	if ((sk_user_data & flags) == flags)
 628		return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
 629	return NULL;
 630}
 631
 632#define rcu_dereference_sk_user_data(sk)				\
 633	__rcu_dereference_sk_user_data_with_flags(sk, 0)
 634#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags)		\
 635({									\
 636	uintptr_t __tmp1 = (uintptr_t)(ptr),				\
 637		  __tmp2 = (uintptr_t)(flags);				\
 638	WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK);			\
 639	WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK);			\
 640	rcu_assign_pointer(__sk_user_data((sk)),			\
 641			   __tmp1 | __tmp2);				\
 642})
 643#define rcu_assign_sk_user_data(sk, ptr)				\
 644	__rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
 645
 646static inline
 647struct net *sock_net(const struct sock *sk)
 648{
 649	return read_pnet(&sk->sk_net);
 650}
 651
 652static inline
 653void sock_net_set(struct sock *sk, struct net *net)
 654{
 655	write_pnet(&sk->sk_net, net);
 656}
 657
 658/*
 659 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
 660 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
 661 * on a socket means that the socket will reuse everybody else's port
 662 * without looking at the other's sk_reuse value.
 663 */
 664
 665#define SK_NO_REUSE	0
 666#define SK_CAN_REUSE	1
 667#define SK_FORCE_REUSE	2
 668
 669int sk_set_peek_off(struct sock *sk, int val);
 670
 671static inline int sk_peek_offset(const struct sock *sk, int flags)
 672{
 673	if (unlikely(flags & MSG_PEEK)) {
 674		return READ_ONCE(sk->sk_peek_off);
 675	}
 676
 677	return 0;
 678}
 679
 680static inline void sk_peek_offset_bwd(struct sock *sk, int val)
 681{
 682	s32 off = READ_ONCE(sk->sk_peek_off);
 683
 684	if (unlikely(off >= 0)) {
 685		off = max_t(s32, off - val, 0);
 686		WRITE_ONCE(sk->sk_peek_off, off);
 687	}
 688}
 689
 690static inline void sk_peek_offset_fwd(struct sock *sk, int val)
 691{
 692	sk_peek_offset_bwd(sk, -val);
 693}
 694
 695/*
 696 * Hashed lists helper routines
 697 */
 698static inline struct sock *sk_entry(const struct hlist_node *node)
 699{
 700	return hlist_entry(node, struct sock, sk_node);
 701}
 702
 703static inline struct sock *__sk_head(const struct hlist_head *head)
 704{
 705	return hlist_entry(head->first, struct sock, sk_node);
 706}
 707
 708static inline struct sock *sk_head(const struct hlist_head *head)
 709{
 710	return hlist_empty(head) ? NULL : __sk_head(head);
 711}
 712
 713static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
 714{
 715	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
 716}
 717
 718static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
 719{
 720	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
 721}
 722
 723static inline struct sock *sk_next(const struct sock *sk)
 724{
 725	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
 726}
 727
 728static inline struct sock *sk_nulls_next(const struct sock *sk)
 729{
 730	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
 731		hlist_nulls_entry(sk->sk_nulls_node.next,
 732				  struct sock, sk_nulls_node) :
 733		NULL;
 734}
 735
 736static inline bool sk_unhashed(const struct sock *sk)
 737{
 738	return hlist_unhashed(&sk->sk_node);
 739}
 740
 741static inline bool sk_hashed(const struct sock *sk)
 742{
 743	return !sk_unhashed(sk);
 744}
 745
 746static inline void sk_node_init(struct hlist_node *node)
 747{
 748	node->pprev = NULL;
 749}
 750
 
 
 
 
 
 751static inline void __sk_del_node(struct sock *sk)
 752{
 753	__hlist_del(&sk->sk_node);
 754}
 755
 756/* NB: equivalent to hlist_del_init_rcu */
 757static inline bool __sk_del_node_init(struct sock *sk)
 758{
 759	if (sk_hashed(sk)) {
 760		__sk_del_node(sk);
 761		sk_node_init(&sk->sk_node);
 762		return true;
 763	}
 764	return false;
 765}
 766
 767/* Grab socket reference count. This operation is valid only
 768   when sk is ALREADY grabbed f.e. it is found in hash table
 769   or a list and the lookup is made under lock preventing hash table
 770   modifications.
 771 */
 772
 773static __always_inline void sock_hold(struct sock *sk)
 774{
 775	refcount_inc(&sk->sk_refcnt);
 776}
 777
 778/* Ungrab socket in the context, which assumes that socket refcnt
 779   cannot hit zero, f.e. it is true in context of any socketcall.
 780 */
 781static __always_inline void __sock_put(struct sock *sk)
 782{
 783	refcount_dec(&sk->sk_refcnt);
 784}
 785
 786static inline bool sk_del_node_init(struct sock *sk)
 787{
 788	bool rc = __sk_del_node_init(sk);
 789
 790	if (rc) {
 791		/* paranoid for a while -acme */
 792		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 793		__sock_put(sk);
 794	}
 795	return rc;
 796}
 797#define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
 798
 799static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
 800{
 801	if (sk_hashed(sk)) {
 802		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
 803		return true;
 804	}
 805	return false;
 806}
 807
 808static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
 809{
 810	bool rc = __sk_nulls_del_node_init_rcu(sk);
 811
 812	if (rc) {
 813		/* paranoid for a while -acme */
 814		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 815		__sock_put(sk);
 816	}
 817	return rc;
 818}
 819
 820static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
 821{
 822	hlist_add_head(&sk->sk_node, list);
 823}
 824
 825static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
 826{
 827	sock_hold(sk);
 828	__sk_add_node(sk, list);
 829}
 830
 831static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
 832{
 833	sock_hold(sk);
 834	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 835	    sk->sk_family == AF_INET6)
 836		hlist_add_tail_rcu(&sk->sk_node, list);
 837	else
 838		hlist_add_head_rcu(&sk->sk_node, list);
 839}
 840
 841static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
 842{
 843	sock_hold(sk);
 844	hlist_add_tail_rcu(&sk->sk_node, list);
 845}
 846
 847static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 848{
 849	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
 850}
 851
 852static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
 853{
 854	hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
 855}
 856
 857static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 858{
 859	sock_hold(sk);
 860	__sk_nulls_add_node_rcu(sk, list);
 861}
 862
 863static inline void __sk_del_bind_node(struct sock *sk)
 864{
 865	__hlist_del(&sk->sk_bind_node);
 866}
 867
 868static inline void sk_add_bind_node(struct sock *sk,
 869					struct hlist_head *list)
 870{
 871	hlist_add_head(&sk->sk_bind_node, list);
 872}
 873
 874static inline void __sk_del_bind2_node(struct sock *sk)
 875{
 876	__hlist_del(&sk->sk_bind2_node);
 877}
 878
 879static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list)
 880{
 881	hlist_add_head(&sk->sk_bind2_node, list);
 882}
 883
 884#define sk_for_each(__sk, list) \
 885	hlist_for_each_entry(__sk, list, sk_node)
 886#define sk_for_each_rcu(__sk, list) \
 887	hlist_for_each_entry_rcu(__sk, list, sk_node)
 888#define sk_nulls_for_each(__sk, node, list) \
 889	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
 890#define sk_nulls_for_each_rcu(__sk, node, list) \
 891	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
 892#define sk_for_each_from(__sk) \
 893	hlist_for_each_entry_from(__sk, sk_node)
 894#define sk_nulls_for_each_from(__sk, node) \
 895	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
 896		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
 897#define sk_for_each_safe(__sk, tmp, list) \
 898	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
 899#define sk_for_each_bound(__sk, list) \
 900	hlist_for_each_entry(__sk, list, sk_bind_node)
 901#define sk_for_each_bound_bhash2(__sk, list) \
 902	hlist_for_each_entry(__sk, list, sk_bind2_node)
 903
 904/**
 905 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
 906 * @tpos:	the type * to use as a loop cursor.
 907 * @pos:	the &struct hlist_node to use as a loop cursor.
 908 * @head:	the head for your list.
 909 * @offset:	offset of hlist_node within the struct.
 910 *
 911 */
 912#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
 913	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
 914	     pos != NULL &&						       \
 915		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
 916	     pos = rcu_dereference(hlist_next_rcu(pos)))
 917
 918static inline struct user_namespace *sk_user_ns(const struct sock *sk)
 919{
 920	/* Careful only use this in a context where these parameters
 921	 * can not change and must all be valid, such as recvmsg from
 922	 * userspace.
 923	 */
 924	return sk->sk_socket->file->f_cred->user_ns;
 925}
 926
 927/* Sock flags */
 928enum sock_flags {
 929	SOCK_DEAD,
 930	SOCK_DONE,
 931	SOCK_URGINLINE,
 932	SOCK_KEEPOPEN,
 933	SOCK_LINGER,
 934	SOCK_DESTROY,
 935	SOCK_BROADCAST,
 936	SOCK_TIMESTAMP,
 937	SOCK_ZAPPED,
 938	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
 939	SOCK_DBG, /* %SO_DEBUG setting */
 940	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
 941	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
 942	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
 
 943	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
 944	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
 945	SOCK_FASYNC, /* fasync() active */
 946	SOCK_RXQ_OVFL,
 947	SOCK_ZEROCOPY, /* buffers from userspace */
 948	SOCK_WIFI_STATUS, /* push wifi status to userspace */
 949	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
 950		     * Will use last 4 bytes of packet sent from
 951		     * user-space instead.
 952		     */
 953	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
 954	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
 955	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
 956	SOCK_TXTIME,
 957	SOCK_XDP, /* XDP is attached */
 958	SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
 959	SOCK_RCVMARK, /* Receive SO_MARK  ancillary data with packet */
 960};
 961
 962#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
 963
 964static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
 965{
 966	nsk->sk_flags = osk->sk_flags;
 967}
 968
 969static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
 970{
 971	__set_bit(flag, &sk->sk_flags);
 972}
 973
 974static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
 975{
 976	__clear_bit(flag, &sk->sk_flags);
 977}
 978
 979static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
 980				     int valbool)
 981{
 982	if (valbool)
 983		sock_set_flag(sk, bit);
 984	else
 985		sock_reset_flag(sk, bit);
 986}
 987
 988static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
 989{
 990	return test_bit(flag, &sk->sk_flags);
 991}
 992
 993#ifdef CONFIG_NET
 994DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
 995static inline int sk_memalloc_socks(void)
 996{
 997	return static_branch_unlikely(&memalloc_socks_key);
 998}
 999
1000void __receive_sock(struct file *file);
1001#else
1002
1003static inline int sk_memalloc_socks(void)
1004{
1005	return 0;
1006}
1007
1008static inline void __receive_sock(struct file *file)
1009{ }
1010#endif
1011
1012static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1013{
1014	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1015}
1016
1017static inline void sk_acceptq_removed(struct sock *sk)
1018{
1019	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1020}
1021
1022static inline void sk_acceptq_added(struct sock *sk)
1023{
1024	WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1025}
1026
1027/* Note: If you think the test should be:
1028 *	return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1029 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1030 */
1031static inline bool sk_acceptq_is_full(const struct sock *sk)
1032{
1033	return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1034}
1035
1036/*
1037 * Compute minimal free write space needed to queue new packets.
1038 */
1039static inline int sk_stream_min_wspace(const struct sock *sk)
1040{
1041	return READ_ONCE(sk->sk_wmem_queued) >> 1;
1042}
1043
1044static inline int sk_stream_wspace(const struct sock *sk)
1045{
1046	return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1047}
1048
1049static inline void sk_wmem_queued_add(struct sock *sk, int val)
1050{
1051	WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1052}
1053
1054void sk_stream_write_space(struct sock *sk);
1055
1056/* OOB backlog add */
1057static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1058{
1059	/* dont let skb dst not refcounted, we are going to leave rcu lock */
1060	skb_dst_force(skb);
1061
1062	if (!sk->sk_backlog.tail)
1063		WRITE_ONCE(sk->sk_backlog.head, skb);
1064	else
1065		sk->sk_backlog.tail->next = skb;
1066
1067	WRITE_ONCE(sk->sk_backlog.tail, skb);
1068	skb->next = NULL;
1069}
1070
1071/*
1072 * Take into account size of receive queue and backlog queue
1073 * Do not take into account this skb truesize,
1074 * to allow even a single big packet to come.
1075 */
1076static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1077{
1078	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1079
1080	return qsize > limit;
1081}
1082
1083/* The per-socket spinlock must be held here. */
1084static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1085					      unsigned int limit)
1086{
1087	if (sk_rcvqueues_full(sk, limit))
1088		return -ENOBUFS;
1089
1090	/*
1091	 * If the skb was allocated from pfmemalloc reserves, only
1092	 * allow SOCK_MEMALLOC sockets to use it as this socket is
1093	 * helping free memory
1094	 */
1095	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1096		return -ENOMEM;
1097
1098	__sk_add_backlog(sk, skb);
1099	sk->sk_backlog.len += skb->truesize;
1100	return 0;
1101}
1102
1103int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1104
1105INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1106INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1107
1108static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1109{
1110	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1111		return __sk_backlog_rcv(sk, skb);
1112
1113	return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1114				  tcp_v6_do_rcv,
1115				  tcp_v4_do_rcv,
1116				  sk, skb);
1117}
1118
1119static inline void sk_incoming_cpu_update(struct sock *sk)
1120{
1121	int cpu = raw_smp_processor_id();
1122
1123	if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1124		WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1125}
1126
1127static inline void sock_rps_record_flow_hash(__u32 hash)
1128{
1129#ifdef CONFIG_RPS
1130	struct rps_sock_flow_table *sock_flow_table;
1131
1132	rcu_read_lock();
1133	sock_flow_table = rcu_dereference(rps_sock_flow_table);
1134	rps_record_sock_flow(sock_flow_table, hash);
1135	rcu_read_unlock();
1136#endif
1137}
1138
1139static inline void sock_rps_record_flow(const struct sock *sk)
1140{
1141#ifdef CONFIG_RPS
1142	if (static_branch_unlikely(&rfs_needed)) {
1143		/* Reading sk->sk_rxhash might incur an expensive cache line
1144		 * miss.
1145		 *
1146		 * TCP_ESTABLISHED does cover almost all states where RFS
1147		 * might be useful, and is cheaper [1] than testing :
1148		 *	IPv4: inet_sk(sk)->inet_daddr
1149		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1150		 * OR	an additional socket flag
1151		 * [1] : sk_state and sk_prot are in the same cache line.
1152		 */
1153		if (sk->sk_state == TCP_ESTABLISHED)
1154			sock_rps_record_flow_hash(sk->sk_rxhash);
1155	}
1156#endif
1157}
1158
1159static inline void sock_rps_save_rxhash(struct sock *sk,
1160					const struct sk_buff *skb)
1161{
1162#ifdef CONFIG_RPS
1163	if (unlikely(sk->sk_rxhash != skb->hash))
1164		sk->sk_rxhash = skb->hash;
1165#endif
1166}
1167
1168static inline void sock_rps_reset_rxhash(struct sock *sk)
1169{
1170#ifdef CONFIG_RPS
1171	sk->sk_rxhash = 0;
1172#endif
1173}
1174
1175#define sk_wait_event(__sk, __timeo, __condition, __wait)		\
1176	({	int __rc;						\
1177		release_sock(__sk);					\
1178		__rc = __condition;					\
1179		if (!__rc) {						\
1180			*(__timeo) = wait_woken(__wait,			\
1181						TASK_INTERRUPTIBLE,	\
1182						*(__timeo));		\
1183		}							\
1184		sched_annotate_sleep();					\
1185		lock_sock(__sk);					\
1186		__rc = __condition;					\
1187		__rc;							\
1188	})
1189
1190int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1191int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1192void sk_stream_wait_close(struct sock *sk, long timeo_p);
1193int sk_stream_error(struct sock *sk, int flags, int err);
1194void sk_stream_kill_queues(struct sock *sk);
1195void sk_set_memalloc(struct sock *sk);
1196void sk_clear_memalloc(struct sock *sk);
1197
1198void __sk_flush_backlog(struct sock *sk);
1199
1200static inline bool sk_flush_backlog(struct sock *sk)
1201{
1202	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1203		__sk_flush_backlog(sk);
1204		return true;
1205	}
1206	return false;
1207}
1208
1209int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1210
1211struct request_sock_ops;
1212struct timewait_sock_ops;
1213struct inet_hashinfo;
1214struct raw_hashinfo;
1215struct smc_hashinfo;
1216struct module;
1217struct sk_psock;
1218
1219/*
1220 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1221 * un-modified. Special care is taken when initializing object to zero.
1222 */
1223static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1224{
1225	if (offsetof(struct sock, sk_node.next) != 0)
1226		memset(sk, 0, offsetof(struct sock, sk_node.next));
1227	memset(&sk->sk_node.pprev, 0,
1228	       size - offsetof(struct sock, sk_node.pprev));
1229}
1230
1231/* Networking protocol blocks we attach to sockets.
1232 * socket layer -> transport layer interface
1233 */
1234struct proto {
1235	void			(*close)(struct sock *sk,
1236					long timeout);
1237	int			(*pre_connect)(struct sock *sk,
1238					struct sockaddr *uaddr,
1239					int addr_len);
1240	int			(*connect)(struct sock *sk,
1241					struct sockaddr *uaddr,
1242					int addr_len);
1243	int			(*disconnect)(struct sock *sk, int flags);
1244
1245	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1246					  bool kern);
1247
1248	int			(*ioctl)(struct sock *sk, int cmd,
1249					 unsigned long arg);
1250	int			(*init)(struct sock *sk);
1251	void			(*destroy)(struct sock *sk);
1252	void			(*shutdown)(struct sock *sk, int how);
1253	int			(*setsockopt)(struct sock *sk, int level,
1254					int optname, sockptr_t optval,
1255					unsigned int optlen);
1256	int			(*getsockopt)(struct sock *sk, int level,
1257					int optname, char __user *optval,
1258					int __user *option);
1259	void			(*keepalive)(struct sock *sk, int valbool);
1260#ifdef CONFIG_COMPAT
 
 
 
 
 
 
 
 
1261	int			(*compat_ioctl)(struct sock *sk,
1262					unsigned int cmd, unsigned long arg);
1263#endif
1264	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1265					   size_t len);
1266	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1267					   size_t len, int flags, int *addr_len);
 
1268	int			(*sendpage)(struct sock *sk, struct page *page,
1269					int offset, size_t size, int flags);
1270	int			(*bind)(struct sock *sk,
1271					struct sockaddr *addr, int addr_len);
1272	int			(*bind_add)(struct sock *sk,
1273					struct sockaddr *addr, int addr_len);
1274
1275	int			(*backlog_rcv) (struct sock *sk,
1276						struct sk_buff *skb);
1277	bool			(*bpf_bypass_getsockopt)(int level,
1278							 int optname);
1279
1280	void		(*release_cb)(struct sock *sk);
1281
1282	/* Keeping track of sk's, looking them up, and port selection methods. */
1283	int			(*hash)(struct sock *sk);
1284	void			(*unhash)(struct sock *sk);
1285	void			(*rehash)(struct sock *sk);
1286	int			(*get_port)(struct sock *sk, unsigned short snum);
1287	void			(*put_port)(struct sock *sk);
1288#ifdef CONFIG_BPF_SYSCALL
1289	int			(*psock_update_sk_prot)(struct sock *sk,
1290							struct sk_psock *psock,
1291							bool restore);
1292#endif
1293
1294	/* Keeping track of sockets in use */
1295#ifdef CONFIG_PROC_FS
1296	unsigned int		inuse_idx;
1297#endif
1298
1299#if IS_ENABLED(CONFIG_MPTCP)
1300	int			(*forward_alloc_get)(const struct sock *sk);
1301#endif
1302
1303	bool			(*stream_memory_free)(const struct sock *sk, int wake);
1304	bool			(*sock_is_readable)(struct sock *sk);
1305	/* Memory pressure */
1306	void			(*enter_memory_pressure)(struct sock *sk);
1307	void			(*leave_memory_pressure)(struct sock *sk);
1308	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1309	int  __percpu		*per_cpu_fw_alloc;
1310	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1311
1312	/*
1313	 * Pressure flag: try to collapse.
1314	 * Technical note: it is used by multiple contexts non atomically.
1315	 * All the __sk_mem_schedule() is of this nature: accounting
1316	 * is strict, actions are advisory and have some latency.
1317	 */
1318	unsigned long		*memory_pressure;
1319	long			*sysctl_mem;
1320
1321	int			*sysctl_wmem;
1322	int			*sysctl_rmem;
1323	u32			sysctl_wmem_offset;
1324	u32			sysctl_rmem_offset;
1325
1326	int			max_header;
1327	bool			no_autobind;
1328
1329	struct kmem_cache	*slab;
1330	unsigned int		obj_size;
1331	slab_flags_t		slab_flags;
1332	unsigned int		useroffset;	/* Usercopy region offset */
1333	unsigned int		usersize;	/* Usercopy region size */
1334
1335	unsigned int __percpu	*orphan_count;
1336
1337	struct request_sock_ops	*rsk_prot;
1338	struct timewait_sock_ops *twsk_prot;
1339
1340	union {
1341		struct inet_hashinfo	*hashinfo;
1342		struct udp_table	*udp_table;
1343		struct raw_hashinfo	*raw_hash;
1344		struct smc_hashinfo	*smc_hash;
1345	} h;
1346
1347	struct module		*owner;
1348
1349	char			name[32];
1350
1351	struct list_head	node;
1352#ifdef SOCK_REFCNT_DEBUG
1353	atomic_t		socks;
1354#endif
1355	int			(*diag_destroy)(struct sock *sk, int err);
1356} __randomize_layout;
1357
1358int proto_register(struct proto *prot, int alloc_slab);
1359void proto_unregister(struct proto *prot);
1360int sock_load_diag_module(int family, int protocol);
1361
1362#ifdef SOCK_REFCNT_DEBUG
1363static inline void sk_refcnt_debug_inc(struct sock *sk)
1364{
1365	atomic_inc(&sk->sk_prot->socks);
1366}
1367
1368static inline void sk_refcnt_debug_dec(struct sock *sk)
1369{
1370	atomic_dec(&sk->sk_prot->socks);
1371	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1372	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1373}
1374
1375static inline void sk_refcnt_debug_release(const struct sock *sk)
1376{
1377	if (refcount_read(&sk->sk_refcnt) != 1)
1378		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1379		       sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1380}
1381#else /* SOCK_REFCNT_DEBUG */
1382#define sk_refcnt_debug_inc(sk) do { } while (0)
1383#define sk_refcnt_debug_dec(sk) do { } while (0)
1384#define sk_refcnt_debug_release(sk) do { } while (0)
1385#endif /* SOCK_REFCNT_DEBUG */
1386
1387INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1388
1389static inline int sk_forward_alloc_get(const struct sock *sk)
1390{
1391#if IS_ENABLED(CONFIG_MPTCP)
1392	if (sk->sk_prot->forward_alloc_get)
1393		return sk->sk_prot->forward_alloc_get(sk);
1394#endif
1395	return sk->sk_forward_alloc;
1396}
1397
1398static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1399{
1400	if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1401		return false;
1402
1403	return sk->sk_prot->stream_memory_free ?
1404		INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1405				     tcp_stream_memory_free, sk, wake) : true;
1406}
1407
1408static inline bool sk_stream_memory_free(const struct sock *sk)
1409{
1410	return __sk_stream_memory_free(sk, 0);
1411}
1412
1413static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1414{
1415	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1416	       __sk_stream_memory_free(sk, wake);
1417}
1418
1419static inline bool sk_stream_is_writeable(const struct sock *sk)
1420{
1421	return __sk_stream_is_writeable(sk, 0);
1422}
1423
1424static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1425					    struct cgroup *ancestor)
1426{
1427#ifdef CONFIG_SOCK_CGROUP_DATA
1428	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1429				    ancestor);
1430#else
1431	return -ENOTSUPP;
1432#endif
1433}
1434
1435static inline bool sk_has_memory_pressure(const struct sock *sk)
1436{
1437	return sk->sk_prot->memory_pressure != NULL;
1438}
1439
1440static inline bool sk_under_memory_pressure(const struct sock *sk)
1441{
1442	if (!sk->sk_prot->memory_pressure)
1443		return false;
1444
1445	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1446	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1447		return true;
1448
1449	return !!*sk->sk_prot->memory_pressure;
1450}
1451
1452static inline long
1453proto_memory_allocated(const struct proto *prot)
1454{
1455	return max(0L, atomic_long_read(prot->memory_allocated));
1456}
1457
1458static inline long
1459sk_memory_allocated(const struct sock *sk)
1460{
1461	return proto_memory_allocated(sk->sk_prot);
1462}
1463
1464/* 1 MB per cpu, in page units */
1465#define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT))
1466
1467static inline void
1468sk_memory_allocated_add(struct sock *sk, int amt)
1469{
1470	int local_reserve;
1471
1472	preempt_disable();
1473	local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1474	if (local_reserve >= SK_MEMORY_PCPU_RESERVE) {
1475		__this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1476		atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1477	}
1478	preempt_enable();
1479}
1480
1481static inline void
1482sk_memory_allocated_sub(struct sock *sk, int amt)
1483{
1484	int local_reserve;
1485
1486	preempt_disable();
1487	local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1488	if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) {
1489		__this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1490		atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1491	}
1492	preempt_enable();
1493}
1494
1495#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1496
1497static inline void sk_sockets_allocated_dec(struct sock *sk)
1498{
1499	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1500				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1501}
1502
1503static inline void sk_sockets_allocated_inc(struct sock *sk)
1504{
1505	percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1506				 SK_ALLOC_PERCPU_COUNTER_BATCH);
1507}
1508
1509static inline u64
1510sk_sockets_allocated_read_positive(struct sock *sk)
1511{
1512	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1513}
1514
1515static inline int
1516proto_sockets_allocated_sum_positive(struct proto *prot)
1517{
1518	return percpu_counter_sum_positive(prot->sockets_allocated);
1519}
1520
 
 
 
 
 
 
1521static inline bool
1522proto_memory_pressure(struct proto *prot)
1523{
1524	if (!prot->memory_pressure)
1525		return false;
1526	return !!*prot->memory_pressure;
1527}
1528
1529
1530#ifdef CONFIG_PROC_FS
1531#define PROTO_INUSE_NR	64	/* should be enough for the first time */
1532struct prot_inuse {
1533	int all;
1534	int val[PROTO_INUSE_NR];
1535};
1536
1537static inline void sock_prot_inuse_add(const struct net *net,
1538				       const struct proto *prot, int val)
1539{
1540	this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1541}
1542
1543static inline void sock_inuse_add(const struct net *net, int val)
1544{
1545	this_cpu_add(net->core.prot_inuse->all, val);
1546}
1547
1548int sock_prot_inuse_get(struct net *net, struct proto *proto);
1549int sock_inuse_get(struct net *net);
1550#else
1551static inline void sock_prot_inuse_add(const struct net *net,
1552				       const struct proto *prot, int val)
1553{
1554}
1555
1556static inline void sock_inuse_add(const struct net *net, int val)
1557{
1558}
1559#endif
1560
1561
1562/* With per-bucket locks this operation is not-atomic, so that
1563 * this version is not worse.
1564 */
1565static inline int __sk_prot_rehash(struct sock *sk)
1566{
1567	sk->sk_prot->unhash(sk);
1568	return sk->sk_prot->hash(sk);
1569}
1570
1571/* About 10 seconds */
1572#define SOCK_DESTROY_TIME (10*HZ)
1573
1574/* Sockets 0-1023 can't be bound to unless you are superuser */
1575#define PROT_SOCK	1024
1576
1577#define SHUTDOWN_MASK	3
1578#define RCV_SHUTDOWN	1
1579#define SEND_SHUTDOWN	2
1580
 
 
1581#define SOCK_BINDADDR_LOCK	4
1582#define SOCK_BINDPORT_LOCK	8
1583
1584struct socket_alloc {
1585	struct socket socket;
1586	struct inode vfs_inode;
1587};
1588
1589static inline struct socket *SOCKET_I(struct inode *inode)
1590{
1591	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1592}
1593
1594static inline struct inode *SOCK_INODE(struct socket *socket)
1595{
1596	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1597}
1598
1599/*
1600 * Functions for memory accounting
1601 */
1602int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1603int __sk_mem_schedule(struct sock *sk, int size, int kind);
1604void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1605void __sk_mem_reclaim(struct sock *sk, int amount);
1606
 
 
 
 
 
1607#define SK_MEM_SEND	0
1608#define SK_MEM_RECV	1
1609
1610/* sysctl_mem values are in pages */
1611static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1612{
1613	return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
 
 
 
 
 
 
 
1614}
1615
1616static inline int sk_mem_pages(int amt)
1617{
1618	return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1619}
1620
1621static inline bool sk_has_account(struct sock *sk)
1622{
1623	/* return true if protocol supports memory accounting */
1624	return !!sk->sk_prot->memory_allocated;
1625}
1626
1627static inline bool sk_wmem_schedule(struct sock *sk, int size)
1628{
1629	int delta;
1630
1631	if (!sk_has_account(sk))
1632		return true;
1633	delta = size - sk->sk_forward_alloc;
1634	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1635}
1636
1637static inline bool
1638sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1639{
1640	int delta;
1641
1642	if (!sk_has_account(sk))
1643		return true;
1644	delta = size - sk->sk_forward_alloc;
1645	return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1646		skb_pfmemalloc(skb);
1647}
1648
1649static inline int sk_unused_reserved_mem(const struct sock *sk)
1650{
1651	int unused_mem;
1652
1653	if (likely(!sk->sk_reserved_mem))
1654		return 0;
1655
1656	unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1657			atomic_read(&sk->sk_rmem_alloc);
1658
1659	return unused_mem > 0 ? unused_mem : 0;
1660}
1661
1662static inline void sk_mem_reclaim(struct sock *sk)
1663{
1664	int reclaimable;
1665
1666	if (!sk_has_account(sk))
1667		return;
1668
1669	reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1670
1671	if (reclaimable >= (int)PAGE_SIZE)
1672		__sk_mem_reclaim(sk, reclaimable);
1673}
1674
1675static inline void sk_mem_reclaim_final(struct sock *sk)
1676{
1677	sk->sk_reserved_mem = 0;
1678	sk_mem_reclaim(sk);
 
 
1679}
1680
1681static inline void sk_mem_charge(struct sock *sk, int size)
1682{
1683	if (!sk_has_account(sk))
1684		return;
1685	sk->sk_forward_alloc -= size;
1686}
1687
1688static inline void sk_mem_uncharge(struct sock *sk, int size)
1689{
1690	if (!sk_has_account(sk))
1691		return;
1692	sk->sk_forward_alloc += size;
1693	sk_mem_reclaim(sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1694}
1695
1696/*
1697 * Macro so as to not evaluate some arguments when
1698 * lockdep is not enabled.
1699 *
1700 * Mark both the sk_lock and the sk_lock.slock as a
1701 * per-address-family lock class.
1702 */
1703#define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1704do {									\
1705	sk->sk_lock.owned = 0;						\
1706	init_waitqueue_head(&sk->sk_lock.wq);				\
1707	spin_lock_init(&(sk)->sk_lock.slock);				\
1708	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1709			sizeof((sk)->sk_lock));				\
1710	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1711				(skey), (sname));				\
1712	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1713} while (0)
1714
 
1715static inline bool lockdep_sock_is_held(const struct sock *sk)
1716{
1717	return lockdep_is_held(&sk->sk_lock) ||
1718	       lockdep_is_held(&sk->sk_lock.slock);
1719}
 
1720
1721void lock_sock_nested(struct sock *sk, int subclass);
1722
1723static inline void lock_sock(struct sock *sk)
1724{
1725	lock_sock_nested(sk, 0);
1726}
1727
1728void __lock_sock(struct sock *sk);
1729void __release_sock(struct sock *sk);
1730void release_sock(struct sock *sk);
1731
1732/* BH context may only use the following locking interface. */
1733#define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1734#define bh_lock_sock_nested(__sk) \
1735				spin_lock_nested(&((__sk)->sk_lock.slock), \
1736				SINGLE_DEPTH_NESTING)
1737#define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1738
1739bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1740
1741/**
1742 * lock_sock_fast - fast version of lock_sock
1743 * @sk: socket
1744 *
1745 * This version should be used for very small section, where process wont block
1746 * return false if fast path is taken:
1747 *
1748 *   sk_lock.slock locked, owned = 0, BH disabled
1749 *
1750 * return true if slow path is taken:
1751 *
1752 *   sk_lock.slock unlocked, owned = 1, BH enabled
1753 */
1754static inline bool lock_sock_fast(struct sock *sk)
1755{
1756	/* The sk_lock has mutex_lock() semantics here. */
1757	mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1758
1759	return __lock_sock_fast(sk);
1760}
1761
1762/* fast socket lock variant for caller already holding a [different] socket lock */
1763static inline bool lock_sock_fast_nested(struct sock *sk)
1764{
1765	mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1766
1767	return __lock_sock_fast(sk);
1768}
1769
1770/**
1771 * unlock_sock_fast - complement of lock_sock_fast
1772 * @sk: socket
1773 * @slow: slow mode
1774 *
1775 * fast unlock socket for user context.
1776 * If slow mode is on, we call regular release_sock()
1777 */
1778static inline void unlock_sock_fast(struct sock *sk, bool slow)
1779	__releases(&sk->sk_lock.slock)
1780{
1781	if (slow) {
1782		release_sock(sk);
1783		__release(&sk->sk_lock.slock);
1784	} else {
1785		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1786		spin_unlock_bh(&sk->sk_lock.slock);
1787	}
1788}
1789
1790void sockopt_lock_sock(struct sock *sk);
1791void sockopt_release_sock(struct sock *sk);
1792bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1793bool sockopt_capable(int cap);
1794
1795/* Used by processes to "lock" a socket state, so that
1796 * interrupts and bottom half handlers won't change it
1797 * from under us. It essentially blocks any incoming
1798 * packets, so that we won't get any new data or any
1799 * packets that change the state of the socket.
1800 *
1801 * While locked, BH processing will add new packets to
1802 * the backlog queue.  This queue is processed by the
1803 * owner of the socket lock right before it is released.
1804 *
1805 * Since ~2.3.5 it is also exclusive sleep lock serializing
1806 * accesses from user process context.
1807 */
1808
1809static inline void sock_owned_by_me(const struct sock *sk)
1810{
1811#ifdef CONFIG_LOCKDEP
1812	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1813#endif
1814}
1815
1816static inline bool sock_owned_by_user(const struct sock *sk)
1817{
1818	sock_owned_by_me(sk);
1819	return sk->sk_lock.owned;
1820}
1821
1822static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1823{
1824	return sk->sk_lock.owned;
1825}
1826
1827static inline void sock_release_ownership(struct sock *sk)
1828{
1829	if (sock_owned_by_user_nocheck(sk)) {
1830		sk->sk_lock.owned = 0;
1831
1832		/* The sk_lock has mutex_unlock() semantics: */
1833		mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1834	}
1835}
1836
1837/* no reclassification while locks are held */
1838static inline bool sock_allow_reclassification(const struct sock *csk)
1839{
1840	struct sock *sk = (struct sock *)csk;
1841
1842	return !sock_owned_by_user_nocheck(sk) &&
1843		!spin_is_locked(&sk->sk_lock.slock);
1844}
1845
1846struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1847		      struct proto *prot, int kern);
1848void sk_free(struct sock *sk);
1849void sk_destruct(struct sock *sk);
1850struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1851void sk_free_unlock_clone(struct sock *sk);
1852
1853struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1854			     gfp_t priority);
1855void __sock_wfree(struct sk_buff *skb);
1856void sock_wfree(struct sk_buff *skb);
1857struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1858			     gfp_t priority);
1859void skb_orphan_partial(struct sk_buff *skb);
1860void sock_rfree(struct sk_buff *skb);
1861void sock_efree(struct sk_buff *skb);
1862#ifdef CONFIG_INET
1863void sock_edemux(struct sk_buff *skb);
1864void sock_pfree(struct sk_buff *skb);
1865#else
1866#define sock_edemux sock_efree
1867#endif
1868
1869int sk_setsockopt(struct sock *sk, int level, int optname,
1870		  sockptr_t optval, unsigned int optlen);
1871int sock_setsockopt(struct socket *sock, int level, int op,
1872		    sockptr_t optval, unsigned int optlen);
1873
1874int sk_getsockopt(struct sock *sk, int level, int optname,
1875		  sockptr_t optval, sockptr_t optlen);
1876int sock_getsockopt(struct socket *sock, int level, int op,
1877		    char __user *optval, int __user *optlen);
1878int sock_gettstamp(struct socket *sock, void __user *userstamp,
1879		   bool timeval, bool time32);
 
 
1880struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1881				     unsigned long data_len, int noblock,
1882				     int *errcode, int max_page_order);
1883
1884static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1885						  unsigned long size,
1886						  int noblock, int *errcode)
1887{
1888	return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1889}
1890
1891void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1892void sock_kfree_s(struct sock *sk, void *mem, int size);
1893void sock_kzfree_s(struct sock *sk, void *mem, int size);
1894void sk_send_sigurg(struct sock *sk);
1895
1896static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1897{
1898	if (sk->sk_socket)
1899		clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1900	WRITE_ONCE(sk->sk_prot, proto);
1901}
1902
1903struct sockcm_cookie {
1904	u64 transmit_time;
1905	u32 mark;
1906	u32 tsflags;
1907};
1908
1909static inline void sockcm_init(struct sockcm_cookie *sockc,
1910			       const struct sock *sk)
1911{
1912	*sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1913}
1914
1915int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1916		     struct sockcm_cookie *sockc);
1917int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1918		   struct sockcm_cookie *sockc);
1919
1920/*
1921 * Functions to fill in entries in struct proto_ops when a protocol
1922 * does not implement a particular function.
1923 */
1924int sock_no_bind(struct socket *, struct sockaddr *, int);
1925int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1926int sock_no_socketpair(struct socket *, struct socket *);
1927int sock_no_accept(struct socket *, struct socket *, int, bool);
1928int sock_no_getname(struct socket *, struct sockaddr *, int);
1929int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1930int sock_no_listen(struct socket *, int);
1931int sock_no_shutdown(struct socket *, int);
 
 
1932int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1933int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1934int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1935int sock_no_mmap(struct file *file, struct socket *sock,
1936		 struct vm_area_struct *vma);
1937ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1938			 size_t size, int flags);
1939ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1940				int offset, size_t size, int flags);
1941
1942/*
1943 * Functions to fill in entries in struct proto_ops when a protocol
1944 * uses the inet style.
1945 */
1946int sock_common_getsockopt(struct socket *sock, int level, int optname,
1947				  char __user *optval, int __user *optlen);
1948int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1949			int flags);
1950int sock_common_setsockopt(struct socket *sock, int level, int optname,
1951			   sockptr_t optval, unsigned int optlen);
 
 
 
 
1952
1953void sk_common_release(struct sock *sk);
1954
1955/*
1956 *	Default socket callbacks and setup code
1957 */
1958
1959/* Initialise core socket variables */
1960void sock_init_data(struct socket *sock, struct sock *sk);
1961
1962/*
1963 * Socket reference counting postulates.
1964 *
1965 * * Each user of socket SHOULD hold a reference count.
1966 * * Each access point to socket (an hash table bucket, reference from a list,
1967 *   running timer, skb in flight MUST hold a reference count.
1968 * * When reference count hits 0, it means it will never increase back.
1969 * * When reference count hits 0, it means that no references from
1970 *   outside exist to this socket and current process on current CPU
1971 *   is last user and may/should destroy this socket.
1972 * * sk_free is called from any context: process, BH, IRQ. When
1973 *   it is called, socket has no references from outside -> sk_free
1974 *   may release descendant resources allocated by the socket, but
1975 *   to the time when it is called, socket is NOT referenced by any
1976 *   hash tables, lists etc.
1977 * * Packets, delivered from outside (from network or from another process)
1978 *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1979 *   when they sit in queue. Otherwise, packets will leak to hole, when
1980 *   socket is looked up by one cpu and unhasing is made by another CPU.
1981 *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1982 *   (leak to backlog). Packet socket does all the processing inside
1983 *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1984 *   use separate SMP lock, so that they are prone too.
1985 */
1986
1987/* Ungrab socket and destroy it, if it was the last reference. */
1988static inline void sock_put(struct sock *sk)
1989{
1990	if (refcount_dec_and_test(&sk->sk_refcnt))
1991		sk_free(sk);
1992}
1993/* Generic version of sock_put(), dealing with all sockets
1994 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1995 */
1996void sock_gen_put(struct sock *sk);
1997
1998int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1999		     unsigned int trim_cap, bool refcounted);
2000static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
2001				 const int nested)
2002{
2003	return __sk_receive_skb(sk, skb, nested, 1, true);
2004}
2005
2006static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
2007{
2008	/* sk_tx_queue_mapping accept only upto a 16-bit value */
2009	if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
2010		return;
2011	sk->sk_tx_queue_mapping = tx_queue;
2012}
2013
2014#define NO_QUEUE_MAPPING	USHRT_MAX
2015
2016static inline void sk_tx_queue_clear(struct sock *sk)
2017{
2018	sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
2019}
2020
2021static inline int sk_tx_queue_get(const struct sock *sk)
2022{
2023	if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
2024		return sk->sk_tx_queue_mapping;
2025
2026	return -1;
2027}
2028
2029static inline void __sk_rx_queue_set(struct sock *sk,
2030				     const struct sk_buff *skb,
2031				     bool force_set)
2032{
2033#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2034	if (skb_rx_queue_recorded(skb)) {
2035		u16 rx_queue = skb_get_rx_queue(skb);
2036
2037		if (force_set ||
2038		    unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2039			WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
 
2040	}
2041#endif
2042}
2043
2044static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2045{
2046	__sk_rx_queue_set(sk, skb, true);
2047}
2048
2049static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2050{
2051	__sk_rx_queue_set(sk, skb, false);
2052}
2053
2054static inline void sk_rx_queue_clear(struct sock *sk)
2055{
2056#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2057	WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2058#endif
2059}
2060
 
2061static inline int sk_rx_queue_get(const struct sock *sk)
2062{
2063#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2064	if (sk) {
2065		int res = READ_ONCE(sk->sk_rx_queue_mapping);
2066
2067		if (res != NO_QUEUE_MAPPING)
2068			return res;
2069	}
2070#endif
2071
2072	return -1;
2073}
 
2074
2075static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2076{
 
2077	sk->sk_socket = sock;
2078}
2079
2080static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2081{
2082	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2083	return &rcu_dereference_raw(sk->sk_wq)->wait;
2084}
2085/* Detach socket from process context.
2086 * Announce socket dead, detach it from wait queue and inode.
2087 * Note that parent inode held reference count on this struct sock,
2088 * we do not release it in this function, because protocol
2089 * probably wants some additional cleanups or even continuing
2090 * to work with this socket (TCP).
2091 */
2092static inline void sock_orphan(struct sock *sk)
2093{
2094	write_lock_bh(&sk->sk_callback_lock);
2095	sock_set_flag(sk, SOCK_DEAD);
2096	sk_set_socket(sk, NULL);
2097	sk->sk_wq  = NULL;
2098	write_unlock_bh(&sk->sk_callback_lock);
2099}
2100
2101static inline void sock_graft(struct sock *sk, struct socket *parent)
2102{
2103	WARN_ON(parent->sk);
2104	write_lock_bh(&sk->sk_callback_lock);
2105	rcu_assign_pointer(sk->sk_wq, &parent->wq);
2106	parent->sk = sk;
2107	sk_set_socket(sk, parent);
2108	sk->sk_uid = SOCK_INODE(parent)->i_uid;
2109	security_sock_graft(sk, parent);
2110	write_unlock_bh(&sk->sk_callback_lock);
2111}
2112
2113kuid_t sock_i_uid(struct sock *sk);
2114unsigned long sock_i_ino(struct sock *sk);
2115
2116static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2117{
2118	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2119}
2120
2121static inline u32 net_tx_rndhash(void)
2122{
2123	u32 v = get_random_u32();
2124
2125	return v ?: 1;
2126}
2127
2128static inline void sk_set_txhash(struct sock *sk)
2129{
2130	/* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2131	WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2132}
2133
2134static inline bool sk_rethink_txhash(struct sock *sk)
2135{
2136	if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2137		sk_set_txhash(sk);
2138		return true;
2139	}
2140	return false;
2141}
2142
2143static inline struct dst_entry *
2144__sk_dst_get(struct sock *sk)
2145{
2146	return rcu_dereference_check(sk->sk_dst_cache,
2147				     lockdep_sock_is_held(sk));
2148}
2149
2150static inline struct dst_entry *
2151sk_dst_get(struct sock *sk)
2152{
2153	struct dst_entry *dst;
2154
2155	rcu_read_lock();
2156	dst = rcu_dereference(sk->sk_dst_cache);
2157	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
2158		dst = NULL;
2159	rcu_read_unlock();
2160	return dst;
2161}
2162
2163static inline void __dst_negative_advice(struct sock *sk)
2164{
2165	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2166
 
 
2167	if (dst && dst->ops->negative_advice) {
2168		ndst = dst->ops->negative_advice(dst);
2169
2170		if (ndst != dst) {
2171			rcu_assign_pointer(sk->sk_dst_cache, ndst);
2172			sk_tx_queue_clear(sk);
2173			sk->sk_dst_pending_confirm = 0;
2174		}
2175	}
2176}
2177
2178static inline void dst_negative_advice(struct sock *sk)
2179{
2180	sk_rethink_txhash(sk);
2181	__dst_negative_advice(sk);
2182}
2183
2184static inline void
2185__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2186{
2187	struct dst_entry *old_dst;
2188
2189	sk_tx_queue_clear(sk);
2190	sk->sk_dst_pending_confirm = 0;
2191	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2192					    lockdep_sock_is_held(sk));
2193	rcu_assign_pointer(sk->sk_dst_cache, dst);
2194	dst_release(old_dst);
2195}
2196
2197static inline void
2198sk_dst_set(struct sock *sk, struct dst_entry *dst)
2199{
2200	struct dst_entry *old_dst;
2201
2202	sk_tx_queue_clear(sk);
2203	sk->sk_dst_pending_confirm = 0;
2204	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2205	dst_release(old_dst);
2206}
2207
2208static inline void
2209__sk_dst_reset(struct sock *sk)
2210{
2211	__sk_dst_set(sk, NULL);
2212}
2213
2214static inline void
2215sk_dst_reset(struct sock *sk)
2216{
2217	sk_dst_set(sk, NULL);
2218}
2219
2220struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2221
2222struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2223
2224static inline void sk_dst_confirm(struct sock *sk)
2225{
2226	if (!READ_ONCE(sk->sk_dst_pending_confirm))
2227		WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2228}
2229
2230static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2231{
2232	if (skb_get_dst_pending_confirm(skb)) {
2233		struct sock *sk = skb->sk;
 
2234
2235		if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2236			WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2237		neigh_confirm(n);
 
 
2238	}
2239}
2240
2241bool sk_mc_loop(struct sock *sk);
2242
2243static inline bool sk_can_gso(const struct sock *sk)
2244{
2245	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2246}
2247
2248void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2249
2250static inline void sk_gso_disable(struct sock *sk)
2251{
2252	sk->sk_gso_disabled = 1;
2253	sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2254}
2255
2256static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2257					   struct iov_iter *from, char *to,
2258					   int copy, int offset)
2259{
2260	if (skb->ip_summed == CHECKSUM_NONE) {
2261		__wsum csum = 0;
2262		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2263			return -EFAULT;
2264		skb->csum = csum_block_add(skb->csum, csum, offset);
2265	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2266		if (!copy_from_iter_full_nocache(to, copy, from))
2267			return -EFAULT;
2268	} else if (!copy_from_iter_full(to, copy, from))
2269		return -EFAULT;
2270
2271	return 0;
2272}
2273
2274static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2275				       struct iov_iter *from, int copy)
2276{
2277	int err, offset = skb->len;
2278
2279	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2280				       copy, offset);
2281	if (err)
2282		__skb_trim(skb, offset);
2283
2284	return err;
2285}
2286
2287static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2288					   struct sk_buff *skb,
2289					   struct page *page,
2290					   int off, int copy)
2291{
2292	int err;
2293
2294	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2295				       copy, skb->len);
2296	if (err)
2297		return err;
2298
2299	skb_len_add(skb, copy);
 
 
2300	sk_wmem_queued_add(sk, copy);
2301	sk_mem_charge(sk, copy);
2302	return 0;
2303}
2304
2305/**
2306 * sk_wmem_alloc_get - returns write allocations
2307 * @sk: socket
2308 *
2309 * Return: sk_wmem_alloc minus initial offset of one
2310 */
2311static inline int sk_wmem_alloc_get(const struct sock *sk)
2312{
2313	return refcount_read(&sk->sk_wmem_alloc) - 1;
2314}
2315
2316/**
2317 * sk_rmem_alloc_get - returns read allocations
2318 * @sk: socket
2319 *
2320 * Return: sk_rmem_alloc
2321 */
2322static inline int sk_rmem_alloc_get(const struct sock *sk)
2323{
2324	return atomic_read(&sk->sk_rmem_alloc);
2325}
2326
2327/**
2328 * sk_has_allocations - check if allocations are outstanding
2329 * @sk: socket
2330 *
2331 * Return: true if socket has write or read allocations
2332 */
2333static inline bool sk_has_allocations(const struct sock *sk)
2334{
2335	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2336}
2337
2338/**
2339 * skwq_has_sleeper - check if there are any waiting processes
2340 * @wq: struct socket_wq
2341 *
2342 * Return: true if socket_wq has waiting processes
2343 *
2344 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2345 * barrier call. They were added due to the race found within the tcp code.
2346 *
2347 * Consider following tcp code paths::
2348 *
2349 *   CPU1                CPU2
2350 *   sys_select          receive packet
2351 *   ...                 ...
2352 *   __add_wait_queue    update tp->rcv_nxt
2353 *   ...                 ...
2354 *   tp->rcv_nxt check   sock_def_readable
2355 *   ...                 {
2356 *   schedule               rcu_read_lock();
2357 *                          wq = rcu_dereference(sk->sk_wq);
2358 *                          if (wq && waitqueue_active(&wq->wait))
2359 *                              wake_up_interruptible(&wq->wait)
2360 *                          ...
2361 *                       }
2362 *
2363 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2364 * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
2365 * could then endup calling schedule and sleep forever if there are no more
2366 * data on the socket.
2367 *
2368 */
2369static inline bool skwq_has_sleeper(struct socket_wq *wq)
2370{
2371	return wq && wq_has_sleeper(&wq->wait);
2372}
2373
2374/**
2375 * sock_poll_wait - place memory barrier behind the poll_wait call.
2376 * @filp:           file
2377 * @sock:           socket to wait on
2378 * @p:              poll_table
2379 *
2380 * See the comments in the wq_has_sleeper function.
2381 */
2382static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2383				  poll_table *p)
2384{
2385	if (!poll_does_not_wait(p)) {
2386		poll_wait(filp, &sock->wq.wait, p);
2387		/* We need to be sure we are in sync with the
2388		 * socket flags modification.
2389		 *
2390		 * This memory barrier is paired in the wq_has_sleeper.
2391		 */
2392		smp_mb();
2393	}
2394}
2395
2396static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2397{
2398	/* This pairs with WRITE_ONCE() in sk_set_txhash() */
2399	u32 txhash = READ_ONCE(sk->sk_txhash);
2400
2401	if (txhash) {
2402		skb->l4_hash = 1;
2403		skb->hash = txhash;
2404	}
2405}
2406
2407void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2408
2409/*
2410 *	Queue a received datagram if it will fit. Stream and sequenced
2411 *	protocols can't normally use this as they need to fit buffers in
2412 *	and play with them.
2413 *
2414 *	Inlined as it's very short and called for pretty much every
2415 *	packet ever received.
2416 */
2417static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2418{
2419	skb_orphan(skb);
2420	skb->sk = sk;
2421	skb->destructor = sock_rfree;
2422	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2423	sk_mem_charge(sk, skb->truesize);
2424}
2425
2426static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2427{
2428	if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2429		skb_orphan(skb);
2430		skb->destructor = sock_efree;
2431		skb->sk = sk;
2432		return true;
2433	}
2434	return false;
2435}
2436
2437static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2438{
2439	skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2440	if (skb) {
2441		if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2442			skb_set_owner_r(skb, sk);
2443			return skb;
2444		}
2445		__kfree_skb(skb);
2446	}
2447	return NULL;
2448}
2449
2450static inline void skb_prepare_for_gro(struct sk_buff *skb)
2451{
2452	if (skb->destructor != sock_wfree) {
2453		skb_orphan(skb);
2454		return;
2455	}
2456	skb->slow_gro = 1;
2457}
2458
2459void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2460		    unsigned long expires);
2461
2462void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2463
2464void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2465
2466int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2467			struct sk_buff *skb, unsigned int flags,
2468			void (*destructor)(struct sock *sk,
2469					   struct sk_buff *skb));
2470int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2471
2472int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2473			      enum skb_drop_reason *reason);
2474
2475static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2476{
2477	return sock_queue_rcv_skb_reason(sk, skb, NULL);
2478}
2479
2480int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2481struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2482
2483/*
2484 *	Recover an error report and clear atomically
2485 */
2486
2487static inline int sock_error(struct sock *sk)
2488{
2489	int err;
2490
2491	/* Avoid an atomic operation for the common case.
2492	 * This is racy since another cpu/thread can change sk_err under us.
2493	 */
2494	if (likely(data_race(!sk->sk_err)))
2495		return 0;
2496
2497	err = xchg(&sk->sk_err, 0);
2498	return -err;
2499}
2500
2501void sk_error_report(struct sock *sk);
2502
2503static inline unsigned long sock_wspace(struct sock *sk)
2504{
2505	int amt = 0;
2506
2507	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2508		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2509		if (amt < 0)
2510			amt = 0;
2511	}
2512	return amt;
2513}
2514
2515/* Note:
2516 *  We use sk->sk_wq_raw, from contexts knowing this
2517 *  pointer is not NULL and cannot disappear/change.
2518 */
2519static inline void sk_set_bit(int nr, struct sock *sk)
2520{
2521	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2522	    !sock_flag(sk, SOCK_FASYNC))
2523		return;
2524
2525	set_bit(nr, &sk->sk_wq_raw->flags);
2526}
2527
2528static inline void sk_clear_bit(int nr, struct sock *sk)
2529{
2530	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2531	    !sock_flag(sk, SOCK_FASYNC))
2532		return;
2533
2534	clear_bit(nr, &sk->sk_wq_raw->flags);
2535}
2536
2537static inline void sk_wake_async(const struct sock *sk, int how, int band)
2538{
2539	if (sock_flag(sk, SOCK_FASYNC)) {
2540		rcu_read_lock();
2541		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2542		rcu_read_unlock();
2543	}
2544}
2545
2546/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2547 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2548 * Note: for send buffers, TCP works better if we can build two skbs at
2549 * minimum.
2550 */
2551#define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2552
2553#define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2554#define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2555
2556static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2557{
2558	u32 val;
2559
2560	if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2561		return;
2562
2563	val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2564	val = max_t(u32, val, sk_unused_reserved_mem(sk));
2565
2566	WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2567}
2568
 
 
 
2569/**
2570 * sk_page_frag - return an appropriate page_frag
2571 * @sk: socket
2572 *
2573 * Use the per task page_frag instead of the per socket one for
2574 * optimization when we know that we're in process context and own
2575 * everything that's associated with %current.
2576 *
2577 * Both direct reclaim and page faults can nest inside other
2578 * socket operations and end up recursing into sk_page_frag()
2579 * while it's already in use: explicitly avoid task page_frag
2580 * when users disable sk_use_task_frag.
2581 *
2582 * Return: a per task page_frag if context allows that,
2583 * otherwise a per socket one.
2584 */
2585static inline struct page_frag *sk_page_frag(struct sock *sk)
2586{
2587	if (sk->sk_use_task_frag)
2588		return &current->task_frag;
2589
2590	return &sk->sk_frag;
2591}
2592
2593bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2594
2595/*
2596 *	Default write policy as shown to user space via poll/select/SIGIO
2597 */
2598static inline bool sock_writeable(const struct sock *sk)
2599{
2600	return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2601}
2602
2603static inline gfp_t gfp_any(void)
2604{
2605	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2606}
2607
2608static inline gfp_t gfp_memcg_charge(void)
2609{
2610	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2611}
2612
2613static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2614{
2615	return noblock ? 0 : sk->sk_rcvtimeo;
2616}
2617
2618static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2619{
2620	return noblock ? 0 : sk->sk_sndtimeo;
2621}
2622
2623static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2624{
2625	int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2626
2627	return v ?: 1;
2628}
2629
2630/* Alas, with timeout socket operations are not restartable.
2631 * Compare this to poll().
2632 */
2633static inline int sock_intr_errno(long timeo)
2634{
2635	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2636}
2637
2638struct sock_skb_cb {
2639	u32 dropcount;
2640};
2641
2642/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2643 * using skb->cb[] would keep using it directly and utilize its
2644 * alignement guarantee.
2645 */
2646#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2647			    sizeof(struct sock_skb_cb)))
2648
2649#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2650			    SOCK_SKB_CB_OFFSET))
2651
2652#define sock_skb_cb_check_size(size) \
2653	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2654
2655static inline void
2656sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2657{
2658	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2659						atomic_read(&sk->sk_drops) : 0;
2660}
2661
2662static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2663{
2664	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2665
2666	atomic_add(segs, &sk->sk_drops);
2667}
2668
2669static inline ktime_t sock_read_timestamp(struct sock *sk)
2670{
2671#if BITS_PER_LONG==32
2672	unsigned int seq;
2673	ktime_t kt;
2674
2675	do {
2676		seq = read_seqbegin(&sk->sk_stamp_seq);
2677		kt = sk->sk_stamp;
2678	} while (read_seqretry(&sk->sk_stamp_seq, seq));
2679
2680	return kt;
2681#else
2682	return READ_ONCE(sk->sk_stamp);
2683#endif
2684}
2685
2686static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2687{
2688#if BITS_PER_LONG==32
2689	write_seqlock(&sk->sk_stamp_seq);
2690	sk->sk_stamp = kt;
2691	write_sequnlock(&sk->sk_stamp_seq);
2692#else
2693	WRITE_ONCE(sk->sk_stamp, kt);
2694#endif
2695}
2696
2697void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2698			   struct sk_buff *skb);
2699void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2700			     struct sk_buff *skb);
2701
2702static inline void
2703sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2704{
2705	ktime_t kt = skb->tstamp;
2706	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2707
2708	/*
2709	 * generate control messages if
2710	 * - receive time stamping in software requested
2711	 * - software time stamp available and wanted
2712	 * - hardware time stamps available and wanted
2713	 */
2714	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2715	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2716	    (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2717	    (hwtstamps->hwtstamp &&
2718	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2719		__sock_recv_timestamp(msg, sk, skb);
2720	else
2721		sock_write_timestamp(sk, kt);
2722
2723	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2724		__sock_recv_wifi_status(msg, sk, skb);
2725}
2726
2727void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2728		       struct sk_buff *skb);
2729
2730#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2731static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2732				   struct sk_buff *skb)
2733{
2734#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL)			| \
2735			   (1UL << SOCK_RCVTSTAMP)			| \
2736			   (1UL << SOCK_RCVMARK))
2737#define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2738			   SOF_TIMESTAMPING_RAW_HARDWARE)
2739
2740	if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY)
2741		__sock_recv_cmsgs(msg, sk, skb);
2742	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2743		sock_write_timestamp(sk, skb->tstamp);
2744	else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2745		sock_write_timestamp(sk, 0);
2746}
2747
2748void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2749
2750/**
2751 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2752 * @sk:		socket sending this packet
2753 * @tsflags:	timestamping flags to use
2754 * @tx_flags:	completed with instructions for time stamping
2755 * @tskey:      filled in with next sk_tskey (not for TCP, which uses seqno)
2756 *
2757 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2758 */
2759static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2760				      __u8 *tx_flags, __u32 *tskey)
2761{
2762	if (unlikely(tsflags)) {
2763		__sock_tx_timestamp(tsflags, tx_flags);
2764		if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2765		    tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2766			*tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2767	}
2768	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2769		*tx_flags |= SKBTX_WIFI_STATUS;
2770}
2771
2772static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2773				     __u8 *tx_flags)
2774{
2775	_sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2776}
2777
2778static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2779{
2780	_sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2781			   &skb_shinfo(skb)->tskey);
2782}
2783
2784static inline bool sk_is_tcp(const struct sock *sk)
2785{
2786	return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP;
2787}
2788
2789/**
2790 * sk_eat_skb - Release a skb if it is no longer needed
2791 * @sk: socket to eat this skb from
2792 * @skb: socket buffer to eat
2793 *
2794 * This routine must be called with interrupts disabled or with the socket
2795 * locked so that the sk_buff queue operation is ok.
2796*/
 
2797static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2798{
2799	__skb_unlink(skb, &sk->sk_receive_queue);
 
 
 
 
 
 
2800	__kfree_skb(skb);
2801}
2802
2803static inline bool
2804skb_sk_is_prefetched(struct sk_buff *skb)
2805{
2806#ifdef CONFIG_INET
2807	return skb->destructor == sock_pfree;
2808#else
2809	return false;
2810#endif /* CONFIG_INET */
2811}
2812
2813/* This helper checks if a socket is a full socket,
2814 * ie _not_ a timewait or request socket.
2815 */
2816static inline bool sk_fullsock(const struct sock *sk)
2817{
2818	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2819}
2820
2821static inline bool
2822sk_is_refcounted(struct sock *sk)
2823{
2824	/* Only full sockets have sk->sk_flags. */
2825	return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2826}
2827
2828/**
2829 * skb_steal_sock - steal a socket from an sk_buff
2830 * @skb: sk_buff to steal the socket from
2831 * @refcounted: is set to true if the socket is reference-counted
2832 */
2833static inline struct sock *
2834skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2835{
2836	if (skb->sk) {
2837		struct sock *sk = skb->sk;
2838
2839		*refcounted = true;
2840		if (skb_sk_is_prefetched(skb))
2841			*refcounted = sk_is_refcounted(sk);
2842		skb->destructor = NULL;
2843		skb->sk = NULL;
2844		return sk;
2845	}
2846	*refcounted = false;
2847	return NULL;
2848}
2849
 
 
 
 
 
 
 
 
2850/* Checks if this SKB belongs to an HW offloaded socket
2851 * and whether any SW fallbacks are required based on dev.
2852 * Check decrypted mark in case skb_orphan() cleared socket.
2853 */
2854static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2855						   struct net_device *dev)
2856{
2857#ifdef CONFIG_SOCK_VALIDATE_XMIT
2858	struct sock *sk = skb->sk;
2859
2860	if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2861		skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2862#ifdef CONFIG_TLS_DEVICE
2863	} else if (unlikely(skb->decrypted)) {
2864		pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2865		kfree_skb(skb);
2866		skb = NULL;
2867#endif
2868	}
2869#endif
2870
2871	return skb;
2872}
2873
2874/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2875 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2876 */
2877static inline bool sk_listener(const struct sock *sk)
2878{
2879	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2880}
2881
2882void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2883int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2884		       int type);
2885
2886bool sk_ns_capable(const struct sock *sk,
2887		   struct user_namespace *user_ns, int cap);
2888bool sk_capable(const struct sock *sk, int cap);
2889bool sk_net_capable(const struct sock *sk, int cap);
2890
2891void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2892
2893/* Take into consideration the size of the struct sk_buff overhead in the
2894 * determination of these values, since that is non-constant across
2895 * platforms.  This makes socket queueing behavior and performance
2896 * not depend upon such differences.
2897 */
2898#define _SK_MEM_PACKETS		256
2899#define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2900#define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2901#define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2902
2903extern __u32 sysctl_wmem_max;
2904extern __u32 sysctl_rmem_max;
2905
2906extern int sysctl_tstamp_allow_data;
2907extern int sysctl_optmem_max;
2908
2909extern __u32 sysctl_wmem_default;
2910extern __u32 sysctl_rmem_default;
2911
2912#define SKB_FRAG_PAGE_ORDER	get_order(32768)
2913DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2914
2915static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2916{
2917	/* Does this proto have per netns sysctl_wmem ? */
2918	if (proto->sysctl_wmem_offset)
2919		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2920
2921	return READ_ONCE(*proto->sysctl_wmem);
2922}
2923
2924static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2925{
2926	/* Does this proto have per netns sysctl_rmem ? */
2927	if (proto->sysctl_rmem_offset)
2928		return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2929
2930	return READ_ONCE(*proto->sysctl_rmem);
2931}
2932
2933/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2934 * Some wifi drivers need to tweak it to get more chunks.
2935 * They can use this helper from their ndo_start_xmit()
2936 */
2937static inline void sk_pacing_shift_update(struct sock *sk, int val)
2938{
2939	if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2940		return;
2941	WRITE_ONCE(sk->sk_pacing_shift, val);
2942}
2943
2944/* if a socket is bound to a device, check that the given device
2945 * index is either the same or that the socket is bound to an L3
2946 * master device and the given device index is also enslaved to
2947 * that L3 master
2948 */
2949static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2950{
2951	int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2952	int mdif;
2953
2954	if (!bound_dev_if || bound_dev_if == dif)
2955		return true;
2956
2957	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2958	if (mdif && mdif == bound_dev_if)
2959		return true;
2960
2961	return false;
2962}
2963
2964void sock_def_readable(struct sock *sk);
2965
2966int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2967void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2968int sock_set_timestamping(struct sock *sk, int optname,
2969			  struct so_timestamping timestamping);
2970
2971void sock_enable_timestamps(struct sock *sk);
2972void sock_no_linger(struct sock *sk);
2973void sock_set_keepalive(struct sock *sk);
2974void sock_set_priority(struct sock *sk, u32 priority);
2975void sock_set_rcvbuf(struct sock *sk, int val);
2976void sock_set_mark(struct sock *sk, u32 val);
2977void sock_set_reuseaddr(struct sock *sk);
2978void sock_set_reuseport(struct sock *sk);
2979void sock_set_sndtimeo(struct sock *sk, s64 secs);
2980
2981int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2982
2983int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2984int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2985			   sockptr_t optval, int optlen, bool old_timeval);
2986
2987static inline bool sk_is_readable(struct sock *sk)
2988{
2989	if (sk->sk_prot->sock_is_readable)
2990		return sk->sk_prot->sock_is_readable(sk);
2991	return false;
2992}
2993#endif	/* _SOCK_H */