Linux Audio

Check our new training course

Loading...
v3.1
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		Definitions for the AF_INET socket handler.
   7 *
   8 * Version:	@(#)sock.h	1.0.4	05/13/93
   9 *
  10 * Authors:	Ross Biro
  11 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
  13 *		Florian La Roche <flla@stud.uni-sb.de>
  14 *
  15 * Fixes:
  16 *		Alan Cox	:	Volatiles in skbuff pointers. See
  17 *					skbuff comments. May be overdone,
  18 *					better to prove they can be removed
  19 *					than the reverse.
  20 *		Alan Cox	:	Added a zapped field for tcp to note
  21 *					a socket is reset and must stay shut up
  22 *		Alan Cox	:	New fields for options
  23 *	Pauline Middelink	:	identd support
  24 *		Alan Cox	:	Eliminate low level recv/recvfrom
  25 *		David S. Miller	:	New socket lookup architecture.
  26 *              Steve Whitehouse:       Default routines for sock_ops
  27 *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
  28 *              			protinfo be just a void pointer, as the
  29 *              			protocol specific parts were moved to
  30 *              			respective headers and ipv4/v6, etc now
  31 *              			use private slabcaches for its socks
  32 *              Pedro Hortas	:	New flags field for socket options
  33 *
  34 *
  35 *		This program is free software; you can redistribute it and/or
  36 *		modify it under the terms of the GNU General Public License
  37 *		as published by the Free Software Foundation; either version
  38 *		2 of the License, or (at your option) any later version.
  39 */
  40#ifndef _SOCK_H
  41#define _SOCK_H
  42
  43#include <linux/hardirq.h>
  44#include <linux/kernel.h>
  45#include <linux/list.h>
  46#include <linux/list_nulls.h>
  47#include <linux/timer.h>
  48#include <linux/cache.h>
  49#include <linux/module.h>
  50#include <linux/lockdep.h>
  51#include <linux/netdevice.h>
  52#include <linux/skbuff.h>	/* struct sk_buff */
  53#include <linux/mm.h>
  54#include <linux/security.h>
  55#include <linux/slab.h>
  56#include <linux/uaccess.h>
  57
 
 
 
 
 
 
  58#include <linux/filter.h>
  59#include <linux/rculist_nulls.h>
  60#include <linux/poll.h>
  61
  62#include <linux/atomic.h>
 
  63#include <net/dst.h>
  64#include <net/checksum.h>
 
 
 
 
  65
  66/*
  67 * This structure really needs to be cleaned up.
  68 * Most of it is for TCP, and not used by any of
  69 * the other protocols.
  70 */
  71
  72/* Define this to get the SOCK_DBG debugging facility. */
  73#define SOCK_DEBUGGING
  74#ifdef SOCK_DEBUGGING
  75#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
  76					printk(KERN_DEBUG msg); } while (0)
  77#else
  78/* Validate arguments and do nothing */
  79static inline void __attribute__ ((format (printf, 2, 3)))
  80SOCK_DEBUG(struct sock *sk, const char *msg, ...)
  81{
  82}
  83#endif
  84
  85/* This is the per-socket lock.  The spinlock provides a synchronization
  86 * between user contexts and software interrupt processing, whereas the
  87 * mini-semaphore synchronizes multiple users amongst themselves.
  88 */
  89typedef struct {
  90	spinlock_t		slock;
  91	int			owned;
  92	wait_queue_head_t	wq;
  93	/*
  94	 * We express the mutex-alike socket_lock semantics
  95	 * to the lock validator by explicitly managing
  96	 * the slock as a lock variant (in addition to
  97	 * the slock itself):
  98	 */
  99#ifdef CONFIG_DEBUG_LOCK_ALLOC
 100	struct lockdep_map dep_map;
 101#endif
 102} socket_lock_t;
 103
 104struct sock;
 105struct proto;
 106struct net;
 107
 
 
 
 108/**
 109 *	struct sock_common - minimal network layer representation of sockets
 110 *	@skc_daddr: Foreign IPv4 addr
 111 *	@skc_rcv_saddr: Bound local IPv4 addr
 112 *	@skc_hash: hash value used with various protocol lookup tables
 113 *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
 
 
 114 *	@skc_family: network address family
 115 *	@skc_state: Connection state
 116 *	@skc_reuse: %SO_REUSEADDR setting
 
 117 *	@skc_bound_dev_if: bound device index if != 0
 118 *	@skc_bind_node: bind hash linkage for various protocol lookup tables
 119 *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
 120 *	@skc_prot: protocol handlers inside a network family
 121 *	@skc_net: reference to the network namespace of this socket
 122 *	@skc_node: main hash linkage for various protocol lookup tables
 123 *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
 124 *	@skc_tx_queue_mapping: tx queue number for this connection
 
 
 
 
 125 *	@skc_refcnt: reference count
 126 *
 127 *	This is the minimal network layer representation of sockets, the header
 128 *	for struct sock and struct inet_timewait_sock.
 129 */
 130struct sock_common {
 131	/* skc_daddr and skc_rcv_saddr must be grouped :
 132	 * cf INET_MATCH() and INET_TW_MATCH()
 133	 */
 134	__be32			skc_daddr;
 135	__be32			skc_rcv_saddr;
 136
 
 
 
 
 137	union  {
 138		unsigned int	skc_hash;
 139		__u16		skc_u16hashes[2];
 140	};
 
 
 
 
 
 
 
 
 
 141	unsigned short		skc_family;
 142	volatile unsigned char	skc_state;
 143	unsigned char		skc_reuse;
 
 
 
 144	int			skc_bound_dev_if;
 145	union {
 146		struct hlist_node	skc_bind_node;
 147		struct hlist_nulls_node skc_portaddr_node;
 148	};
 149	struct proto		*skc_prot;
 150#ifdef CONFIG_NET_NS
 151	struct net	 	*skc_net;
 
 
 
 152#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 153	/*
 154	 * fields between dontcopy_begin/dontcopy_end
 155	 * are not copied in sock_copy()
 156	 */
 157	/* private: */
 158	int			skc_dontcopy_begin[0];
 159	/* public: */
 160	union {
 161		struct hlist_node	skc_node;
 162		struct hlist_nulls_node skc_nulls_node;
 163	};
 164	int			skc_tx_queue_mapping;
 165	atomic_t		skc_refcnt;
 
 
 
 
 
 
 166	/* private: */
 167	int                     skc_dontcopy_end[0];
 
 
 
 
 
 168	/* public: */
 169};
 170
 171/**
 172  *	struct sock - network layer representation of sockets
 173  *	@__sk_common: shared layout with inet_timewait_sock
 174  *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
 175  *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
 176  *	@sk_lock:	synchronizer
 
 177  *	@sk_rcvbuf: size of receive buffer in bytes
 178  *	@sk_wq: sock wait queue and async head
 
 179  *	@sk_dst_cache: destination cache
 180  *	@sk_dst_lock: destination cache lock
 181  *	@sk_policy: flow policy
 182  *	@sk_receive_queue: incoming packets
 183  *	@sk_wmem_alloc: transmit queue bytes committed
 
 184  *	@sk_write_queue: Packet sending queue
 185  *	@sk_async_wait_queue: DMA copied packets
 186  *	@sk_omem_alloc: "o" is "option" or "other"
 187  *	@sk_wmem_queued: persistent queue size
 188  *	@sk_forward_alloc: space allocated forward
 
 
 189  *	@sk_allocation: allocation mode
 
 
 
 190  *	@sk_sndbuf: size of send buffer in bytes
 191  *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
 192  *		   %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
 193  *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
 
 194  *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
 195  *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
 196  *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
 197  *	@sk_gso_max_size: Maximum GSO segment size to build
 
 
 198  *	@sk_lingertime: %SO_LINGER l_linger setting
 199  *	@sk_backlog: always used with the per-socket spinlock held
 200  *	@sk_callback_lock: used with the callbacks in the end of this struct
 201  *	@sk_error_queue: rarely used
 202  *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
 203  *			  IPV6_ADDRFORM for instance)
 204  *	@sk_err: last error
 205  *	@sk_err_soft: errors that don't cause failure but are the cause of a
 206  *		      persistent failure not just 'timed out'
 207  *	@sk_drops: raw/udp drops counter
 208  *	@sk_ack_backlog: current listen backlog
 209  *	@sk_max_ack_backlog: listen backlog set in listen()
 
 210  *	@sk_priority: %SO_PRIORITY setting
 211  *	@sk_type: socket type (%SOCK_STREAM, etc)
 212  *	@sk_protocol: which protocol this socket belongs in this network family
 213  *	@sk_peer_pid: &struct pid for this socket's peer
 214  *	@sk_peer_cred: %SO_PEERCRED setting
 215  *	@sk_rcvlowat: %SO_RCVLOWAT setting
 216  *	@sk_rcvtimeo: %SO_RCVTIMEO setting
 217  *	@sk_sndtimeo: %SO_SNDTIMEO setting
 218  *	@sk_rxhash: flow hash received from netif layer
 219  *	@sk_filter: socket filtering instructions
 220  *	@sk_protinfo: private area, net family specific, when not using slab
 221  *	@sk_timer: sock cleanup timer
 222  *	@sk_stamp: time stamp of last packet received
 
 
 
 223  *	@sk_socket: Identd and reporting IO signals
 224  *	@sk_user_data: RPC layer private data
 225  *	@sk_sndmsg_page: cached page for sendmsg
 226  *	@sk_sndmsg_off: cached offset for sendmsg
 227  *	@sk_send_head: front of stuff to transmit
 228  *	@sk_security: used by security modules
 229  *	@sk_mark: generic packet mark
 230  *	@sk_classid: this socket's cgroup classid
 
 231  *	@sk_write_pending: a write to stream socket waits to start
 232  *	@sk_state_change: callback to indicate change in the state of the sock
 233  *	@sk_data_ready: callback to indicate there is data to be processed
 234  *	@sk_write_space: callback to indicate there is bf sending space available
 235  *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
 236  *	@sk_backlog_rcv: callback to process the backlog
 237  *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
 238 */
 
 
 239struct sock {
 240	/*
 241	 * Now struct inet_timewait_sock also uses sock_common, so please just
 242	 * don't add nothing before this first member (__sk_common) --acme
 243	 */
 244	struct sock_common	__sk_common;
 245#define sk_node			__sk_common.skc_node
 246#define sk_nulls_node		__sk_common.skc_nulls_node
 247#define sk_refcnt		__sk_common.skc_refcnt
 248#define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
 249
 250#define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
 251#define sk_dontcopy_end		__sk_common.skc_dontcopy_end
 252#define sk_hash			__sk_common.skc_hash
 
 
 
 
 
 
 253#define sk_family		__sk_common.skc_family
 254#define sk_state		__sk_common.skc_state
 255#define sk_reuse		__sk_common.skc_reuse
 
 
 
 256#define sk_bound_dev_if		__sk_common.skc_bound_dev_if
 257#define sk_bind_node		__sk_common.skc_bind_node
 258#define sk_prot			__sk_common.skc_prot
 259#define sk_net			__sk_common.skc_net
 
 
 
 
 
 
 
 260	socket_lock_t		sk_lock;
 
 
 
 261	struct sk_buff_head	sk_receive_queue;
 262	/*
 263	 * The backlog queue is special, it is always used with
 264	 * the per-socket spinlock held and requires low latency
 265	 * access. Therefore we special case it's implementation.
 266	 * Note : rmem_alloc is in this structure to fill a hole
 267	 * on 64bit arches, not because its logically part of
 268	 * backlog.
 269	 */
 270	struct {
 271		atomic_t	rmem_alloc;
 272		int		len;
 273		struct sk_buff	*head;
 274		struct sk_buff	*tail;
 275	} sk_backlog;
 276#define sk_rmem_alloc sk_backlog.rmem_alloc
 
 277	int			sk_forward_alloc;
 278#ifdef CONFIG_RPS
 279	__u32			sk_rxhash;
 
 
 280#endif
 281	atomic_t		sk_drops;
 282	int			sk_rcvbuf;
 283
 284	struct sk_filter __rcu	*sk_filter;
 285	struct socket_wq __rcu	*sk_wq;
 286
 287#ifdef CONFIG_NET_DMA
 288	struct sk_buff_head	sk_async_wait_queue;
 289#endif
 290
 291#ifdef CONFIG_XFRM
 292	struct xfrm_policy	*sk_policy[2];
 293#endif
 294	unsigned long 		sk_flags;
 295	struct dst_entry	*sk_dst_cache;
 296	spinlock_t		sk_dst_lock;
 297	atomic_t		sk_wmem_alloc;
 298	atomic_t		sk_omem_alloc;
 299	int			sk_sndbuf;
 
 
 
 
 
 
 
 
 
 300	struct sk_buff_head	sk_write_queue;
 301	kmemcheck_bitfield_begin(flags);
 302	unsigned int		sk_shutdown  : 2,
 303				sk_no_check  : 2,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 304				sk_userlocks : 4,
 305				sk_protocol  : 8,
 306				sk_type      : 16;
 307	kmemcheck_bitfield_end(flags);
 308	int			sk_wmem_queued;
 309	gfp_t			sk_allocation;
 310	int			sk_route_caps;
 311	int			sk_route_nocaps;
 312	int			sk_gso_type;
 313	unsigned int		sk_gso_max_size;
 314	int			sk_rcvlowat;
 315	unsigned long	        sk_lingertime;
 316	struct sk_buff_head	sk_error_queue;
 317	struct proto		*sk_prot_creator;
 318	rwlock_t		sk_callback_lock;
 319	int			sk_err,
 320				sk_err_soft;
 321	unsigned short		sk_ack_backlog;
 322	unsigned short		sk_max_ack_backlog;
 323	__u32			sk_priority;
 324	struct pid		*sk_peer_pid;
 325	const struct cred	*sk_peer_cred;
 326	long			sk_rcvtimeo;
 327	long			sk_sndtimeo;
 328	void			*sk_protinfo;
 329	struct timer_list	sk_timer;
 330	ktime_t			sk_stamp;
 
 
 
 
 331	struct socket		*sk_socket;
 332	void			*sk_user_data;
 333	struct page		*sk_sndmsg_page;
 334	struct sk_buff		*sk_send_head;
 335	__u32			sk_sndmsg_off;
 336	int			sk_write_pending;
 337#ifdef CONFIG_SECURITY
 338	void			*sk_security;
 339#endif
 340	__u32			sk_mark;
 341	u32			sk_classid;
 342	void			(*sk_state_change)(struct sock *sk);
 343	void			(*sk_data_ready)(struct sock *sk, int bytes);
 344	void			(*sk_write_space)(struct sock *sk);
 345	void			(*sk_error_report)(struct sock *sk);
 346  	int			(*sk_backlog_rcv)(struct sock *sk,
 347						  struct sk_buff *skb);  
 348	void                    (*sk_destruct)(struct sock *sk);
 
 
 
 
 
 
 
 
 349};
 350
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 351/*
 352 * Hashed lists helper routines
 353 */
 354static inline struct sock *sk_entry(const struct hlist_node *node)
 355{
 356	return hlist_entry(node, struct sock, sk_node);
 357}
 358
 359static inline struct sock *__sk_head(const struct hlist_head *head)
 360{
 361	return hlist_entry(head->first, struct sock, sk_node);
 362}
 363
 364static inline struct sock *sk_head(const struct hlist_head *head)
 365{
 366	return hlist_empty(head) ? NULL : __sk_head(head);
 367}
 368
 369static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
 370{
 371	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
 372}
 373
 374static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
 375{
 376	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
 377}
 378
 379static inline struct sock *sk_next(const struct sock *sk)
 380{
 381	return sk->sk_node.next ?
 382		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
 383}
 384
 385static inline struct sock *sk_nulls_next(const struct sock *sk)
 386{
 387	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
 388		hlist_nulls_entry(sk->sk_nulls_node.next,
 389				  struct sock, sk_nulls_node) :
 390		NULL;
 391}
 392
 393static inline int sk_unhashed(const struct sock *sk)
 394{
 395	return hlist_unhashed(&sk->sk_node);
 396}
 397
 398static inline int sk_hashed(const struct sock *sk)
 399{
 400	return !sk_unhashed(sk);
 401}
 402
 403static __inline__ void sk_node_init(struct hlist_node *node)
 404{
 405	node->pprev = NULL;
 406}
 407
 408static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
 409{
 410	node->pprev = NULL;
 411}
 412
 413static __inline__ void __sk_del_node(struct sock *sk)
 414{
 415	__hlist_del(&sk->sk_node);
 416}
 417
 418/* NB: equivalent to hlist_del_init_rcu */
 419static __inline__ int __sk_del_node_init(struct sock *sk)
 420{
 421	if (sk_hashed(sk)) {
 422		__sk_del_node(sk);
 423		sk_node_init(&sk->sk_node);
 424		return 1;
 425	}
 426	return 0;
 427}
 428
 429/* Grab socket reference count. This operation is valid only
 430   when sk is ALREADY grabbed f.e. it is found in hash table
 431   or a list and the lookup is made under lock preventing hash table
 432   modifications.
 433 */
 434
 435static inline void sock_hold(struct sock *sk)
 436{
 437	atomic_inc(&sk->sk_refcnt);
 438}
 439
 440/* Ungrab socket in the context, which assumes that socket refcnt
 441   cannot hit zero, f.e. it is true in context of any socketcall.
 442 */
 443static inline void __sock_put(struct sock *sk)
 444{
 445	atomic_dec(&sk->sk_refcnt);
 446}
 447
 448static __inline__ int sk_del_node_init(struct sock *sk)
 449{
 450	int rc = __sk_del_node_init(sk);
 451
 452	if (rc) {
 453		/* paranoid for a while -acme */
 454		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
 455		__sock_put(sk);
 456	}
 457	return rc;
 458}
 459#define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
 460
 461static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
 462{
 463	if (sk_hashed(sk)) {
 464		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
 465		return 1;
 466	}
 467	return 0;
 468}
 469
 470static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
 471{
 472	int rc = __sk_nulls_del_node_init_rcu(sk);
 473
 474	if (rc) {
 475		/* paranoid for a while -acme */
 476		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
 477		__sock_put(sk);
 478	}
 479	return rc;
 480}
 481
 482static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
 483{
 484	hlist_add_head(&sk->sk_node, list);
 485}
 486
 487static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
 488{
 489	sock_hold(sk);
 490	__sk_add_node(sk, list);
 491}
 492
 493static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
 494{
 495	sock_hold(sk);
 496	hlist_add_head_rcu(&sk->sk_node, list);
 
 
 
 
 497}
 498
 499static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 500{
 501	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
 502}
 503
 504static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 505{
 506	sock_hold(sk);
 507	__sk_nulls_add_node_rcu(sk, list);
 508}
 509
 510static __inline__ void __sk_del_bind_node(struct sock *sk)
 511{
 512	__hlist_del(&sk->sk_bind_node);
 513}
 514
 515static __inline__ void sk_add_bind_node(struct sock *sk,
 516					struct hlist_head *list)
 517{
 518	hlist_add_head(&sk->sk_bind_node, list);
 519}
 520
 521#define sk_for_each(__sk, node, list) \
 522	hlist_for_each_entry(__sk, node, list, sk_node)
 523#define sk_for_each_rcu(__sk, node, list) \
 524	hlist_for_each_entry_rcu(__sk, node, list, sk_node)
 525#define sk_nulls_for_each(__sk, node, list) \
 526	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
 527#define sk_nulls_for_each_rcu(__sk, node, list) \
 528	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
 529#define sk_for_each_from(__sk, node) \
 530	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
 531		hlist_for_each_entry_from(__sk, node, sk_node)
 532#define sk_nulls_for_each_from(__sk, node) \
 533	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
 534		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
 535#define sk_for_each_safe(__sk, node, tmp, list) \
 536	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
 537#define sk_for_each_bound(__sk, node, list) \
 538	hlist_for_each_entry(__sk, node, list, sk_bind_node)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 539
 540/* Sock flags */
 541enum sock_flags {
 542	SOCK_DEAD,
 543	SOCK_DONE,
 544	SOCK_URGINLINE,
 545	SOCK_KEEPOPEN,
 546	SOCK_LINGER,
 547	SOCK_DESTROY,
 548	SOCK_BROADCAST,
 549	SOCK_TIMESTAMP,
 550	SOCK_ZAPPED,
 551	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
 552	SOCK_DBG, /* %SO_DEBUG setting */
 553	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
 554	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
 555	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
 556	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
 557	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
 558	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
 559	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
 560	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
 561	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
 562	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
 563	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
 564	SOCK_FASYNC, /* fasync() active */
 565	SOCK_RXQ_OVFL,
 566	SOCK_ZEROCOPY, /* buffers from userspace */
 
 
 
 
 
 
 
 
 567};
 568
 
 
 569static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
 570{
 571	nsk->sk_flags = osk->sk_flags;
 572}
 573
 574static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
 575{
 576	__set_bit(flag, &sk->sk_flags);
 577}
 578
 579static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
 580{
 581	__clear_bit(flag, &sk->sk_flags);
 582}
 583
 584static inline int sock_flag(struct sock *sk, enum sock_flags flag)
 585{
 586	return test_bit(flag, &sk->sk_flags);
 587}
 588
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 589static inline void sk_acceptq_removed(struct sock *sk)
 590{
 591	sk->sk_ack_backlog--;
 592}
 593
 594static inline void sk_acceptq_added(struct sock *sk)
 595{
 596	sk->sk_ack_backlog++;
 597}
 598
 599static inline int sk_acceptq_is_full(struct sock *sk)
 600{
 601	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
 602}
 603
 604/*
 605 * Compute minimal free write space needed to queue new packets.
 606 */
 607static inline int sk_stream_min_wspace(struct sock *sk)
 608{
 609	return sk->sk_wmem_queued >> 1;
 610}
 611
 612static inline int sk_stream_wspace(struct sock *sk)
 613{
 614	return sk->sk_sndbuf - sk->sk_wmem_queued;
 615}
 616
 617extern void sk_stream_write_space(struct sock *sk);
 618
 619static inline int sk_stream_memory_free(struct sock *sk)
 620{
 621	return sk->sk_wmem_queued < sk->sk_sndbuf;
 622}
 623
 624/* OOB backlog add */
 625static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
 626{
 627	/* dont let skb dst not refcounted, we are going to leave rcu lock */
 628	skb_dst_force(skb);
 629
 630	if (!sk->sk_backlog.tail)
 631		sk->sk_backlog.head = skb;
 632	else
 633		sk->sk_backlog.tail->next = skb;
 634
 635	sk->sk_backlog.tail = skb;
 636	skb->next = NULL;
 637}
 638
 639/*
 640 * Take into account size of receive queue and backlog queue
 
 
 641 */
 642static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb)
 643{
 644	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
 645
 646	return qsize + skb->truesize > sk->sk_rcvbuf;
 647}
 648
 649/* The per-socket spinlock must be held here. */
 650static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
 
 651{
 652	if (sk_rcvqueues_full(sk, skb))
 653		return -ENOBUFS;
 654
 
 
 
 
 
 
 
 
 655	__sk_add_backlog(sk, skb);
 656	sk->sk_backlog.len += skb->truesize;
 657	return 0;
 658}
 659
 
 
 660static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 661{
 
 
 
 662	return sk->sk_backlog_rcv(sk, skb);
 663}
 664
 665static inline void sock_rps_record_flow(const struct sock *sk)
 
 
 
 
 
 
 
 
 666{
 667#ifdef CONFIG_RPS
 668	struct rps_sock_flow_table *sock_flow_table;
 669
 670	rcu_read_lock();
 671	sock_flow_table = rcu_dereference(rps_sock_flow_table);
 672	rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
 673	rcu_read_unlock();
 674#endif
 675}
 676
 677static inline void sock_rps_reset_flow(const struct sock *sk)
 678{
 679#ifdef CONFIG_RPS
 680	struct rps_sock_flow_table *sock_flow_table;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 681
 682	rcu_read_lock();
 683	sock_flow_table = rcu_dereference(rps_sock_flow_table);
 684	rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
 685	rcu_read_unlock();
 
 
 686#endif
 687}
 688
 689static inline void sock_rps_save_rxhash(struct sock *sk, u32 rxhash)
 690{
 691#ifdef CONFIG_RPS
 692	if (unlikely(sk->sk_rxhash != rxhash)) {
 693		sock_rps_reset_flow(sk);
 694		sk->sk_rxhash = rxhash;
 695	}
 696#endif
 697}
 698
 699#define sk_wait_event(__sk, __timeo, __condition)			\
 700	({	int __rc;						\
 701		release_sock(__sk);					\
 702		__rc = __condition;					\
 703		if (!__rc) {						\
 704			*(__timeo) = schedule_timeout(*(__timeo));	\
 
 
 705		}							\
 
 706		lock_sock(__sk);					\
 707		__rc = __condition;					\
 708		__rc;							\
 709	})
 710
 711extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
 712extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
 713extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
 714extern int sk_stream_error(struct sock *sk, int flags, int err);
 715extern void sk_stream_kill_queues(struct sock *sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 716
 717extern int sk_wait_data(struct sock *sk, long *timeo);
 718
 719struct request_sock_ops;
 720struct timewait_sock_ops;
 721struct inet_hashinfo;
 722struct raw_hashinfo;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 723
 724/* Networking protocol blocks we attach to sockets.
 725 * socket layer -> transport layer interface
 726 * transport -> network interface is defined by struct inet_proto
 727 */
 728struct proto {
 729	void			(*close)(struct sock *sk, 
 730					long timeout);
 
 
 
 731	int			(*connect)(struct sock *sk,
 732				        struct sockaddr *uaddr, 
 733					int addr_len);
 734	int			(*disconnect)(struct sock *sk, int flags);
 735
 736	struct sock *		(*accept) (struct sock *sk, int flags, int *err);
 
 737
 738	int			(*ioctl)(struct sock *sk, int cmd,
 739					 unsigned long arg);
 740	int			(*init)(struct sock *sk);
 741	void			(*destroy)(struct sock *sk);
 742	void			(*shutdown)(struct sock *sk, int how);
 743	int			(*setsockopt)(struct sock *sk, int level, 
 744					int optname, char __user *optval,
 745					unsigned int optlen);
 746	int			(*getsockopt)(struct sock *sk, int level, 
 747					int optname, char __user *optval, 
 748					int __user *option);  	 
 
 749#ifdef CONFIG_COMPAT
 750	int			(*compat_setsockopt)(struct sock *sk,
 751					int level,
 752					int optname, char __user *optval,
 753					unsigned int optlen);
 754	int			(*compat_getsockopt)(struct sock *sk,
 755					int level,
 756					int optname, char __user *optval,
 757					int __user *option);
 758	int			(*compat_ioctl)(struct sock *sk,
 759					unsigned int cmd, unsigned long arg);
 760#endif
 761	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
 762					   struct msghdr *msg, size_t len);
 763	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
 764					   struct msghdr *msg,
 765					size_t len, int noblock, int flags, 
 766					int *addr_len);
 767	int			(*sendpage)(struct sock *sk, struct page *page,
 768					int offset, size_t size, int flags);
 769	int			(*bind)(struct sock *sk, 
 770					struct sockaddr *uaddr, int addr_len);
 771
 772	int			(*backlog_rcv) (struct sock *sk, 
 773						struct sk_buff *skb);
 774
 
 
 775	/* Keeping track of sk's, looking them up, and port selection methods. */
 776	void			(*hash)(struct sock *sk);
 777	void			(*unhash)(struct sock *sk);
 778	void			(*rehash)(struct sock *sk);
 779	int			(*get_port)(struct sock *sk, unsigned short snum);
 780	void			(*clear_sk)(struct sock *sk, int size);
 781
 782	/* Keeping track of sockets in use */
 783#ifdef CONFIG_PROC_FS
 784	unsigned int		inuse_idx;
 785#endif
 786
 
 
 787	/* Memory pressure */
 788	void			(*enter_memory_pressure)(struct sock *sk);
 
 789	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
 790	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
 791	/*
 792	 * Pressure flag: try to collapse.
 793	 * Technical note: it is used by multiple contexts non atomically.
 794	 * All the __sk_mem_schedule() is of this nature: accounting
 795	 * is strict, actions are advisory and have some latency.
 796	 */
 797	int			*memory_pressure;
 798	long			*sysctl_mem;
 
 799	int			*sysctl_wmem;
 800	int			*sysctl_rmem;
 
 
 
 801	int			max_header;
 802	bool			no_autobind;
 803
 804	struct kmem_cache	*slab;
 805	unsigned int		obj_size;
 806	int			slab_flags;
 
 
 807
 808	struct percpu_counter	*orphan_count;
 809
 810	struct request_sock_ops	*rsk_prot;
 811	struct timewait_sock_ops *twsk_prot;
 812
 813	union {
 814		struct inet_hashinfo	*hashinfo;
 815		struct udp_table	*udp_table;
 816		struct raw_hashinfo	*raw_hash;
 
 817	} h;
 818
 819	struct module		*owner;
 820
 821	char			name[32];
 822
 823	struct list_head	node;
 824#ifdef SOCK_REFCNT_DEBUG
 825	atomic_t		socks;
 826#endif
 827};
 
 828
 829extern int proto_register(struct proto *prot, int alloc_slab);
 830extern void proto_unregister(struct proto *prot);
 
 831
 832#ifdef SOCK_REFCNT_DEBUG
 833static inline void sk_refcnt_debug_inc(struct sock *sk)
 834{
 835	atomic_inc(&sk->sk_prot->socks);
 836}
 837
 838static inline void sk_refcnt_debug_dec(struct sock *sk)
 839{
 840	atomic_dec(&sk->sk_prot->socks);
 841	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
 842	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
 843}
 844
 845static inline void sk_refcnt_debug_release(const struct sock *sk)
 846{
 847	if (atomic_read(&sk->sk_refcnt) != 1)
 848		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
 849		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
 850}
 851#else /* SOCK_REFCNT_DEBUG */
 852#define sk_refcnt_debug_inc(sk) do { } while (0)
 853#define sk_refcnt_debug_dec(sk) do { } while (0)
 854#define sk_refcnt_debug_release(sk) do { } while (0)
 855#endif /* SOCK_REFCNT_DEBUG */
 856
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 857
 858#ifdef CONFIG_PROC_FS
 859/* Called with local bh disabled */
 860extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
 861extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
 
 862#else
 863static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
 864		int inc)
 865{
 866}
 867#endif
 868
 869
 870/* With per-bucket locks this operation is not-atomic, so that
 871 * this version is not worse.
 872 */
 873static inline void __sk_prot_rehash(struct sock *sk)
 874{
 875	sk->sk_prot->unhash(sk);
 876	sk->sk_prot->hash(sk);
 877}
 878
 879void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
 880
 881/* About 10 seconds */
 882#define SOCK_DESTROY_TIME (10*HZ)
 883
 884/* Sockets 0-1023 can't be bound to unless you are superuser */
 885#define PROT_SOCK	1024
 886
 887#define SHUTDOWN_MASK	3
 888#define RCV_SHUTDOWN	1
 889#define SEND_SHUTDOWN	2
 890
 891#define SOCK_SNDBUF_LOCK	1
 892#define SOCK_RCVBUF_LOCK	2
 893#define SOCK_BINDADDR_LOCK	4
 894#define SOCK_BINDPORT_LOCK	8
 895
 896/* sock_iocb: used to kick off async processing of socket ios */
 897struct sock_iocb {
 898	struct list_head	list;
 899
 900	int			flags;
 901	int			size;
 902	struct socket		*sock;
 903	struct sock		*sk;
 904	struct scm_cookie	*scm;
 905	struct msghdr		*msg, async_msg;
 906	struct kiocb		*kiocb;
 907};
 908
 909static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
 910{
 911	return (struct sock_iocb *)iocb->private;
 912}
 913
 914static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
 915{
 916	return si->kiocb;
 917}
 918
 919struct socket_alloc {
 920	struct socket socket;
 921	struct inode vfs_inode;
 922};
 923
 924static inline struct socket *SOCKET_I(struct inode *inode)
 925{
 926	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
 927}
 928
 929static inline struct inode *SOCK_INODE(struct socket *socket)
 930{
 931	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
 932}
 933
 934/*
 935 * Functions for memory accounting
 936 */
 937extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
 938extern void __sk_mem_reclaim(struct sock *sk);
 
 
 939
 940#define SK_MEM_QUANTUM ((int)PAGE_SIZE)
 
 
 
 941#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
 942#define SK_MEM_SEND	0
 943#define SK_MEM_RECV	1
 944
 
 
 
 
 
 
 
 
 
 
 
 
 
 945static inline int sk_mem_pages(int amt)
 946{
 947	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
 948}
 949
 950static inline int sk_has_account(struct sock *sk)
 951{
 952	/* return true if protocol supports memory accounting */
 953	return !!sk->sk_prot->memory_allocated;
 954}
 955
 956static inline int sk_wmem_schedule(struct sock *sk, int size)
 957{
 958	if (!sk_has_account(sk))
 959		return 1;
 960	return size <= sk->sk_forward_alloc ||
 961		__sk_mem_schedule(sk, size, SK_MEM_SEND);
 962}
 963
 964static inline int sk_rmem_schedule(struct sock *sk, int size)
 
 965{
 966	if (!sk_has_account(sk))
 967		return 1;
 968	return size <= sk->sk_forward_alloc ||
 969		__sk_mem_schedule(sk, size, SK_MEM_RECV);
 
 970}
 971
 972static inline void sk_mem_reclaim(struct sock *sk)
 973{
 974	if (!sk_has_account(sk))
 975		return;
 976	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
 977		__sk_mem_reclaim(sk);
 978}
 979
 980static inline void sk_mem_reclaim_partial(struct sock *sk)
 981{
 982	if (!sk_has_account(sk))
 983		return;
 984	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
 985		__sk_mem_reclaim(sk);
 986}
 987
 988static inline void sk_mem_charge(struct sock *sk, int size)
 989{
 990	if (!sk_has_account(sk))
 991		return;
 992	sk->sk_forward_alloc -= size;
 993}
 994
 995static inline void sk_mem_uncharge(struct sock *sk, int size)
 996{
 997	if (!sk_has_account(sk))
 998		return;
 999	sk->sk_forward_alloc += size;
 
 
 
 
 
 
 
 
 
 
1000}
1001
1002static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1003{
1004	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1005	sk->sk_wmem_queued -= skb->truesize;
1006	sk_mem_uncharge(sk, skb->truesize);
1007	__kfree_skb(skb);
1008}
1009
1010/* Used by processes to "lock" a socket state, so that
1011 * interrupts and bottom half handlers won't change it
1012 * from under us. It essentially blocks any incoming
1013 * packets, so that we won't get any new data or any
1014 * packets that change the state of the socket.
1015 *
1016 * While locked, BH processing will add new packets to
1017 * the backlog queue.  This queue is processed by the
1018 * owner of the socket lock right before it is released.
1019 *
1020 * Since ~2.3.5 it is also exclusive sleep lock serializing
1021 * accesses from user process context.
1022 */
1023#define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1024
1025/*
1026 * Macro so as to not evaluate some arguments when
1027 * lockdep is not enabled.
1028 *
1029 * Mark both the sk_lock and the sk_lock.slock as a
1030 * per-address-family lock class.
1031 */
1032#define sock_lock_init_class_and_name(sk, sname, skey, name, key) 	\
1033do {									\
1034	sk->sk_lock.owned = 0;						\
1035	init_waitqueue_head(&sk->sk_lock.wq);				\
1036	spin_lock_init(&(sk)->sk_lock.slock);				\
1037	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1038			sizeof((sk)->sk_lock));				\
1039	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1040		       	(skey), (sname));				\
1041	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1042} while (0)
1043
1044extern void lock_sock_nested(struct sock *sk, int subclass);
 
 
 
 
 
 
 
 
1045
1046static inline void lock_sock(struct sock *sk)
1047{
1048	lock_sock_nested(sk, 0);
1049}
1050
1051extern void release_sock(struct sock *sk);
1052
1053/* BH context may only use the following locking interface. */
1054#define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1055#define bh_lock_sock_nested(__sk) \
1056				spin_lock_nested(&((__sk)->sk_lock.slock), \
1057				SINGLE_DEPTH_NESTING)
1058#define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1059
1060extern bool lock_sock_fast(struct sock *sk);
1061/**
1062 * unlock_sock_fast - complement of lock_sock_fast
1063 * @sk: socket
1064 * @slow: slow mode
1065 *
1066 * fast unlock socket for user context.
1067 * If slow mode is on, we call regular release_sock()
1068 */
1069static inline void unlock_sock_fast(struct sock *sk, bool slow)
1070{
1071	if (slow)
1072		release_sock(sk);
1073	else
1074		spin_unlock_bh(&sk->sk_lock.slock);
1075}
1076
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1077
1078extern struct sock		*sk_alloc(struct net *net, int family,
1079					  gfp_t priority,
1080					  struct proto *prot);
1081extern void			sk_free(struct sock *sk);
1082extern void			sk_release_kernel(struct sock *sk);
1083extern struct sock		*sk_clone(const struct sock *sk,
1084					  const gfp_t priority);
1085
1086extern struct sk_buff		*sock_wmalloc(struct sock *sk,
1087					      unsigned long size, int force,
1088					      gfp_t priority);
1089extern struct sk_buff		*sock_rmalloc(struct sock *sk,
1090					      unsigned long size, int force,
1091					      gfp_t priority);
1092extern void			sock_wfree(struct sk_buff *skb);
1093extern void			sock_rfree(struct sk_buff *skb);
1094
1095extern int			sock_setsockopt(struct socket *sock, int level,
1096						int op, char __user *optval,
1097						unsigned int optlen);
1098
1099extern int			sock_getsockopt(struct socket *sock, int level,
1100						int op, char __user *optval, 
1101						int __user *optlen);
1102extern struct sk_buff 		*sock_alloc_send_skb(struct sock *sk,
1103						     unsigned long size,
1104						     int noblock,
1105						     int *errcode);
1106extern struct sk_buff 		*sock_alloc_send_pskb(struct sock *sk,
1107						      unsigned long header_len,
1108						      unsigned long data_len,
1109						      int noblock,
1110						      int *errcode);
1111extern void *sock_kmalloc(struct sock *sk, int size,
1112			  gfp_t priority);
1113extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1114extern void sk_send_sigurg(struct sock *sk);
1115
1116#ifdef CONFIG_CGROUPS
1117extern void sock_update_classid(struct sock *sk);
1118#else
1119static inline void sock_update_classid(struct sock *sk)
1120{
 
 
 
1121}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1122#endif
1123
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1124/*
1125 * Functions to fill in entries in struct proto_ops when a protocol
1126 * does not implement a particular function.
1127 */
1128extern int                      sock_no_bind(struct socket *, 
1129					     struct sockaddr *, int);
1130extern int                      sock_no_connect(struct socket *,
1131						struct sockaddr *, int, int);
1132extern int                      sock_no_socketpair(struct socket *,
1133						   struct socket *);
1134extern int                      sock_no_accept(struct socket *,
1135					       struct socket *, int);
1136extern int                      sock_no_getname(struct socket *,
1137						struct sockaddr *, int *, int);
1138extern unsigned int             sock_no_poll(struct file *, struct socket *,
1139					     struct poll_table_struct *);
1140extern int                      sock_no_ioctl(struct socket *, unsigned int,
1141					      unsigned long);
1142extern int			sock_no_listen(struct socket *, int);
1143extern int                      sock_no_shutdown(struct socket *, int);
1144extern int			sock_no_getsockopt(struct socket *, int , int,
1145						   char __user *, int __user *);
1146extern int			sock_no_setsockopt(struct socket *, int, int,
1147						   char __user *, unsigned int);
1148extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1149						struct msghdr *, size_t);
1150extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1151						struct msghdr *, size_t, int);
1152extern int			sock_no_mmap(struct file *file,
1153					     struct socket *sock,
1154					     struct vm_area_struct *vma);
1155extern ssize_t			sock_no_sendpage(struct socket *sock,
1156						struct page *page,
1157						int offset, size_t size, 
1158						int flags);
1159
1160/*
1161 * Functions to fill in entries in struct proto_ops when a protocol
1162 * uses the inet style.
1163 */
1164extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1165				  char __user *optval, int __user *optlen);
1166extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1167			       struct msghdr *msg, size_t size, int flags);
1168extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1169				  char __user *optval, unsigned int optlen);
1170extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1171		int optname, char __user *optval, int __user *optlen);
1172extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1173		int optname, char __user *optval, unsigned int optlen);
1174
1175extern void sk_common_release(struct sock *sk);
1176
1177/*
1178 *	Default socket callbacks and setup code
1179 */
1180 
1181/* Initialise core socket variables */
1182extern void sock_init_data(struct socket *sock, struct sock *sk);
1183
1184extern void sk_filter_release_rcu(struct rcu_head *rcu);
1185
1186/**
1187 *	sk_filter_release - release a socket filter
1188 *	@fp: filter to remove
1189 *
1190 *	Remove a filter from a socket and release its resources.
1191 */
1192
1193static inline void sk_filter_release(struct sk_filter *fp)
1194{
1195	if (atomic_dec_and_test(&fp->refcnt))
1196		call_rcu(&fp->rcu, sk_filter_release_rcu);
1197}
1198
1199static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1200{
1201	unsigned int size = sk_filter_len(fp);
1202
1203	atomic_sub(size, &sk->sk_omem_alloc);
1204	sk_filter_release(fp);
1205}
1206
1207static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1208{
1209	atomic_inc(&fp->refcnt);
1210	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1211}
1212
1213/*
1214 * Socket reference counting postulates.
1215 *
1216 * * Each user of socket SHOULD hold a reference count.
1217 * * Each access point to socket (an hash table bucket, reference from a list,
1218 *   running timer, skb in flight MUST hold a reference count.
1219 * * When reference count hits 0, it means it will never increase back.
1220 * * When reference count hits 0, it means that no references from
1221 *   outside exist to this socket and current process on current CPU
1222 *   is last user and may/should destroy this socket.
1223 * * sk_free is called from any context: process, BH, IRQ. When
1224 *   it is called, socket has no references from outside -> sk_free
1225 *   may release descendant resources allocated by the socket, but
1226 *   to the time when it is called, socket is NOT referenced by any
1227 *   hash tables, lists etc.
1228 * * Packets, delivered from outside (from network or from another process)
1229 *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1230 *   when they sit in queue. Otherwise, packets will leak to hole, when
1231 *   socket is looked up by one cpu and unhasing is made by another CPU.
1232 *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1233 *   (leak to backlog). Packet socket does all the processing inside
1234 *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1235 *   use separate SMP lock, so that they are prone too.
1236 */
1237
1238/* Ungrab socket and destroy it, if it was the last reference. */
1239static inline void sock_put(struct sock *sk)
1240{
1241	if (atomic_dec_and_test(&sk->sk_refcnt))
1242		sk_free(sk);
1243}
 
 
 
 
1244
1245extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1246			  const int nested);
 
 
 
 
 
1247
1248static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1249{
1250	sk->sk_tx_queue_mapping = tx_queue;
1251}
1252
1253static inline void sk_tx_queue_clear(struct sock *sk)
1254{
1255	sk->sk_tx_queue_mapping = -1;
1256}
1257
1258static inline int sk_tx_queue_get(const struct sock *sk)
1259{
1260	return sk ? sk->sk_tx_queue_mapping : -1;
1261}
1262
1263static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1264{
1265	sk_tx_queue_clear(sk);
1266	sk->sk_socket = sock;
1267}
1268
1269static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1270{
1271	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1272	return &rcu_dereference_raw(sk->sk_wq)->wait;
1273}
1274/* Detach socket from process context.
1275 * Announce socket dead, detach it from wait queue and inode.
1276 * Note that parent inode held reference count on this struct sock,
1277 * we do not release it in this function, because protocol
1278 * probably wants some additional cleanups or even continuing
1279 * to work with this socket (TCP).
1280 */
1281static inline void sock_orphan(struct sock *sk)
1282{
1283	write_lock_bh(&sk->sk_callback_lock);
1284	sock_set_flag(sk, SOCK_DEAD);
1285	sk_set_socket(sk, NULL);
1286	sk->sk_wq  = NULL;
1287	write_unlock_bh(&sk->sk_callback_lock);
1288}
1289
1290static inline void sock_graft(struct sock *sk, struct socket *parent)
1291{
 
1292	write_lock_bh(&sk->sk_callback_lock);
1293	sk->sk_wq = parent->wq;
1294	parent->sk = sk;
1295	sk_set_socket(sk, parent);
 
1296	security_sock_graft(sk, parent);
1297	write_unlock_bh(&sk->sk_callback_lock);
1298}
1299
1300extern int sock_i_uid(struct sock *sk);
1301extern unsigned long sock_i_ino(struct sock *sk);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1302
1303static inline struct dst_entry *
1304__sk_dst_get(struct sock *sk)
1305{
1306	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1307						       lockdep_is_held(&sk->sk_lock.slock));
1308}
1309
1310static inline struct dst_entry *
1311sk_dst_get(struct sock *sk)
1312{
1313	struct dst_entry *dst;
1314
1315	rcu_read_lock();
1316	dst = rcu_dereference(sk->sk_dst_cache);
1317	if (dst)
1318		dst_hold(dst);
1319	rcu_read_unlock();
1320	return dst;
1321}
1322
1323extern void sk_reset_txq(struct sock *sk);
1324
1325static inline void dst_negative_advice(struct sock *sk)
1326{
1327	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1328
 
 
1329	if (dst && dst->ops->negative_advice) {
1330		ndst = dst->ops->negative_advice(dst);
1331
1332		if (ndst != dst) {
1333			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1334			sk_reset_txq(sk);
 
1335		}
1336	}
1337}
1338
1339static inline void
1340__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1341{
1342	struct dst_entry *old_dst;
1343
1344	sk_tx_queue_clear(sk);
1345	/*
1346	 * This can be called while sk is owned by the caller only,
1347	 * with no state that can be checked in a rcu_dereference_check() cond
1348	 */
1349	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1350	rcu_assign_pointer(sk->sk_dst_cache, dst);
1351	dst_release(old_dst);
1352}
1353
1354static inline void
1355sk_dst_set(struct sock *sk, struct dst_entry *dst)
1356{
1357	spin_lock(&sk->sk_dst_lock);
1358	__sk_dst_set(sk, dst);
1359	spin_unlock(&sk->sk_dst_lock);
 
 
 
1360}
1361
1362static inline void
1363__sk_dst_reset(struct sock *sk)
1364{
1365	__sk_dst_set(sk, NULL);
1366}
1367
1368static inline void
1369sk_dst_reset(struct sock *sk)
1370{
1371	spin_lock(&sk->sk_dst_lock);
1372	__sk_dst_reset(sk);
1373	spin_unlock(&sk->sk_dst_lock);
1374}
1375
1376extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
 
 
1377
1378extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
 
 
 
 
1379
1380static inline int sk_can_gso(const struct sock *sk)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1381{
1382	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1383}
1384
1385extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1386
1387static inline void sk_nocaps_add(struct sock *sk, int flags)
1388{
1389	sk->sk_route_nocaps |= flags;
1390	sk->sk_route_caps &= ~flags;
1391}
1392
1393static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1394					   char __user *from, char *to,
1395					   int copy, int offset)
1396{
1397	if (skb->ip_summed == CHECKSUM_NONE) {
1398		int err = 0;
1399		__wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1400		if (err)
1401			return err;
1402		skb->csum = csum_block_add(skb->csum, csum, offset);
1403	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1404		if (!access_ok(VERIFY_READ, from, copy) ||
1405		    __copy_from_user_nocache(to, from, copy))
1406			return -EFAULT;
1407	} else if (copy_from_user(to, from, copy))
1408		return -EFAULT;
1409
1410	return 0;
1411}
1412
1413static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1414				       char __user *from, int copy)
1415{
1416	int err, offset = skb->len;
1417
1418	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1419				       copy, offset);
1420	if (err)
1421		__skb_trim(skb, offset);
1422
1423	return err;
1424}
1425
1426static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1427					   struct sk_buff *skb,
1428					   struct page *page,
1429					   int off, int copy)
1430{
1431	int err;
1432
1433	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1434				       copy, skb->len);
1435	if (err)
1436		return err;
1437
1438	skb->len	     += copy;
1439	skb->data_len	     += copy;
1440	skb->truesize	     += copy;
1441	sk->sk_wmem_queued   += copy;
1442	sk_mem_charge(sk, copy);
1443	return 0;
1444}
1445
1446static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1447				   struct sk_buff *skb, struct page *page,
1448				   int off, int copy)
1449{
1450	if (skb->ip_summed == CHECKSUM_NONE) {
1451		int err = 0;
1452		__wsum csum = csum_and_copy_from_user(from,
1453						     page_address(page) + off,
1454							    copy, 0, &err);
1455		if (err)
1456			return err;
1457		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1458	} else if (copy_from_user(page_address(page) + off, from, copy))
1459		return -EFAULT;
1460
1461	skb->len	     += copy;
1462	skb->data_len	     += copy;
1463	skb->truesize	     += copy;
1464	sk->sk_wmem_queued   += copy;
1465	sk_mem_charge(sk, copy);
1466	return 0;
1467}
1468
1469/**
1470 * sk_wmem_alloc_get - returns write allocations
1471 * @sk: socket
1472 *
1473 * Returns sk_wmem_alloc minus initial offset of one
1474 */
1475static inline int sk_wmem_alloc_get(const struct sock *sk)
1476{
1477	return atomic_read(&sk->sk_wmem_alloc) - 1;
1478}
1479
1480/**
1481 * sk_rmem_alloc_get - returns read allocations
1482 * @sk: socket
1483 *
1484 * Returns sk_rmem_alloc
1485 */
1486static inline int sk_rmem_alloc_get(const struct sock *sk)
1487{
1488	return atomic_read(&sk->sk_rmem_alloc);
1489}
1490
1491/**
1492 * sk_has_allocations - check if allocations are outstanding
1493 * @sk: socket
1494 *
1495 * Returns true if socket has write or read allocations
1496 */
1497static inline int sk_has_allocations(const struct sock *sk)
1498{
1499	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1500}
1501
1502/**
1503 * wq_has_sleeper - check if there are any waiting processes
1504 * @wq: struct socket_wq
1505 *
1506 * Returns true if socket_wq has waiting processes
1507 *
1508 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1509 * barrier call. They were added due to the race found within the tcp code.
1510 *
1511 * Consider following tcp code paths:
1512 *
1513 * CPU1                  CPU2
1514 *
1515 * sys_select            receive packet
1516 *   ...                 ...
1517 *   __add_wait_queue    update tp->rcv_nxt
1518 *   ...                 ...
1519 *   tp->rcv_nxt check   sock_def_readable
1520 *   ...                 {
1521 *   schedule               rcu_read_lock();
1522 *                          wq = rcu_dereference(sk->sk_wq);
1523 *                          if (wq && waitqueue_active(&wq->wait))
1524 *                              wake_up_interruptible(&wq->wait)
1525 *                          ...
1526 *                       }
1527 *
1528 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1529 * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1530 * could then endup calling schedule and sleep forever if there are no more
1531 * data on the socket.
1532 *
1533 */
1534static inline bool wq_has_sleeper(struct socket_wq *wq)
1535{
1536
1537	/*
1538	 * We need to be sure we are in sync with the
1539	 * add_wait_queue modifications to the wait queue.
1540	 *
1541	 * This memory barrier is paired in the sock_poll_wait.
1542	 */
1543	smp_mb();
1544	return wq && waitqueue_active(&wq->wait);
1545}
1546
1547/**
1548 * sock_poll_wait - place memory barrier behind the poll_wait call.
1549 * @filp:           file
1550 * @wait_address:   socket wait queue
1551 * @p:              poll_table
1552 *
1553 * See the comments in the wq_has_sleeper function.
1554 */
1555static inline void sock_poll_wait(struct file *filp,
1556		wait_queue_head_t *wait_address, poll_table *p)
1557{
1558	if (p && wait_address) {
1559		poll_wait(filp, wait_address, p);
1560		/*
1561		 * We need to be sure we are in sync with the
1562		 * socket flags modification.
1563		 *
1564		 * This memory barrier is paired in the wq_has_sleeper.
1565		*/
1566		smp_mb();
1567	}
1568}
1569
 
 
 
 
 
 
 
 
 
 
1570/*
1571 * 	Queue a received datagram if it will fit. Stream and sequenced
1572 *	protocols can't normally use this as they need to fit buffers in
1573 *	and play with them.
1574 *
1575 * 	Inlined as it's very short and called for pretty much every
1576 *	packet ever received.
1577 */
1578
1579static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1580{
1581	skb_orphan(skb);
1582	skb->sk = sk;
1583	skb->destructor = sock_wfree;
1584	/*
1585	 * We used to take a refcount on sk, but following operation
1586	 * is enough to guarantee sk_free() wont free this sock until
1587	 * all in-flight packets are completed
1588	 */
1589	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1590}
1591
1592static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1593{
1594	skb_orphan(skb);
1595	skb->sk = sk;
1596	skb->destructor = sock_rfree;
1597	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1598	sk_mem_charge(sk, skb->truesize);
1599}
1600
1601extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1602			   unsigned long expires);
1603
1604extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1605
1606extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
 
 
 
 
 
1607
1608extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
 
1609
1610/*
1611 *	Recover an error report and clear atomically
1612 */
1613 
1614static inline int sock_error(struct sock *sk)
1615{
1616	int err;
1617	if (likely(!sk->sk_err))
1618		return 0;
1619	err = xchg(&sk->sk_err, 0);
1620	return -err;
1621}
1622
1623static inline unsigned long sock_wspace(struct sock *sk)
1624{
1625	int amt = 0;
1626
1627	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1628		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1629		if (amt < 0) 
1630			amt = 0;
1631	}
1632	return amt;
1633}
1634
1635static inline void sk_wake_async(struct sock *sk, int how, int band)
 
 
 
 
1636{
1637	if (sock_flag(sk, SOCK_FASYNC))
1638		sock_wake_async(sk->sk_socket, how, band);
 
 
 
1639}
1640
1641#define SOCK_MIN_SNDBUF 2048
1642/*
1643 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1644 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1645 */
1646#define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
 
 
 
1647
1648static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1649{
1650	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1651		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1652		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1653	}
1654}
1655
1656struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
 
1657
1658static inline struct page *sk_stream_alloc_page(struct sock *sk)
 
 
 
 
 
 
 
1659{
1660	struct page *page = NULL;
 
1661
1662	page = alloc_pages(sk->sk_allocation, 0);
1663	if (!page) {
1664		sk->sk_prot->enter_memory_pressure(sk);
1665		sk_stream_moderate_sndbuf(sk);
1666	}
1667	return page;
1668}
1669
 
 
 
 
 
 
1670/*
1671 *	Default write policy as shown to user space via poll/select/SIGIO
1672 */
1673static inline int sock_writeable(const struct sock *sk) 
1674{
1675	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1676}
1677
1678static inline gfp_t gfp_any(void)
1679{
1680	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1681}
1682
1683static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1684{
1685	return noblock ? 0 : sk->sk_rcvtimeo;
1686}
1687
1688static inline long sock_sndtimeo(const struct sock *sk, int noblock)
1689{
1690	return noblock ? 0 : sk->sk_sndtimeo;
1691}
1692
1693static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
1694{
1695	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
1696}
1697
1698/* Alas, with timeout socket operations are not restartable.
1699 * Compare this to poll().
1700 */
1701static inline int sock_intr_errno(long timeo)
1702{
1703	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
1704}
1705
1706extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
1707	struct sk_buff *skb);
 
 
 
 
 
 
 
 
 
 
 
1708
1709static __inline__ void
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1710sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
1711{
1712	ktime_t kt = skb->tstamp;
1713	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
1714
1715	/*
1716	 * generate control messages if
1717	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
1718	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
1719	 * - software time stamp available and wanted
1720	 *   (SOCK_TIMESTAMPING_SOFTWARE)
1721	 * - hardware time stamps available and wanted
1722	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
1723	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
1724	 */
1725	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
1726	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
1727	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
1728	    (hwtstamps->hwtstamp.tv64 &&
1729	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
1730	    (hwtstamps->syststamp.tv64 &&
1731	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
1732		__sock_recv_timestamp(msg, sk, skb);
1733	else
1734		sk->sk_stamp = kt;
 
 
 
1735}
1736
1737extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
1738				     struct sk_buff *skb);
1739
 
1740static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
1741					  struct sk_buff *skb)
1742{
1743#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
1744			   (1UL << SOCK_RCVTSTAMP)			| \
1745			   (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)	| \
1746			   (1UL << SOCK_TIMESTAMPING_SOFTWARE)		| \
1747			   (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) 	| \
1748			   (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
1749
1750	if (sk->sk_flags & FLAGS_TS_OR_DROPS)
1751		__sock_recv_ts_and_drops(msg, sk, skb);
1752	else
1753		sk->sk_stamp = skb->tstamp;
 
 
1754}
1755
 
 
1756/**
1757 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
1758 * @sk:		socket sending this packet
1759 * @tx_flags:	filled with instructions for time stamping
 
1760 *
1761 * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
1762 * parameters are invalid.
1763 */
1764extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
 
 
 
 
 
 
 
1765
1766/**
1767 * sk_eat_skb - Release a skb if it is no longer needed
1768 * @sk: socket to eat this skb from
1769 * @skb: socket buffer to eat
1770 * @copied_early: flag indicating whether DMA operations copied this data early
1771 *
1772 * This routine must be called with interrupts disabled or with the socket
1773 * locked so that the sk_buff queue operation is ok.
1774*/
1775#ifdef CONFIG_NET_DMA
1776static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1777{
1778	__skb_unlink(skb, &sk->sk_receive_queue);
1779	if (!copied_early)
1780		__kfree_skb(skb);
1781	else
1782		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
1783}
1784#else
1785static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
1786{
1787	__skb_unlink(skb, &sk->sk_receive_queue);
1788	__kfree_skb(skb);
1789}
1790#endif
1791
1792static inline
1793struct net *sock_net(const struct sock *sk)
1794{
1795	return read_pnet(&sk->sk_net);
1796}
1797
1798static inline
1799void sock_net_set(struct sock *sk, struct net *net)
1800{
1801	write_pnet(&sk->sk_net, net);
1802}
1803
1804/*
1805 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
1806 * They should not hold a reference to a namespace in order to allow
1807 * to stop it.
1808 * Sockets after sk_change_net should be released using sk_release_kernel
1809 */
1810static inline void sk_change_net(struct sock *sk, struct net *net)
1811{
1812	put_net(sock_net(sk));
1813	sock_net_set(sk, hold_net(net));
1814}
1815
1816static inline struct sock *skb_steal_sock(struct sk_buff *skb)
1817{
1818	if (unlikely(skb->sk)) {
1819		struct sock *sk = skb->sk;
1820
1821		skb->destructor = NULL;
1822		skb->sk = NULL;
1823		return sk;
1824	}
1825	return NULL;
1826}
1827
1828extern void sock_enable_timestamp(struct sock *sk, int flag);
1829extern int sock_get_timestamp(struct sock *, struct timeval __user *);
1830extern int sock_get_timestampns(struct sock *, struct timespec __user *);
1831
1832/* 
1833 *	Enable debug/info messages 
1834 */
1835extern int net_msg_warn;
1836#define NETDEBUG(fmt, args...) \
1837	do { if (net_msg_warn) printk(fmt,##args); } while (0)
1838
1839#define LIMIT_NETDEBUG(fmt, args...) \
1840	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1841
1842extern __u32 sysctl_wmem_max;
1843extern __u32 sysctl_rmem_max;
1844
1845extern void sk_init(void);
1846
1847extern int sysctl_optmem_max;
1848
1849extern __u32 sysctl_wmem_default;
1850extern __u32 sysctl_rmem_default;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1851
1852#endif	/* _SOCK_H */
v4.17
   1/*
   2 * INET		An implementation of the TCP/IP protocol suite for the LINUX
   3 *		operating system.  INET is implemented using the  BSD Socket
   4 *		interface as the means of communication with the user level.
   5 *
   6 *		Definitions for the AF_INET socket handler.
   7 *
   8 * Version:	@(#)sock.h	1.0.4	05/13/93
   9 *
  10 * Authors:	Ross Biro
  11 *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12 *		Corey Minyard <wf-rch!minyard@relay.EU.net>
  13 *		Florian La Roche <flla@stud.uni-sb.de>
  14 *
  15 * Fixes:
  16 *		Alan Cox	:	Volatiles in skbuff pointers. See
  17 *					skbuff comments. May be overdone,
  18 *					better to prove they can be removed
  19 *					than the reverse.
  20 *		Alan Cox	:	Added a zapped field for tcp to note
  21 *					a socket is reset and must stay shut up
  22 *		Alan Cox	:	New fields for options
  23 *	Pauline Middelink	:	identd support
  24 *		Alan Cox	:	Eliminate low level recv/recvfrom
  25 *		David S. Miller	:	New socket lookup architecture.
  26 *              Steve Whitehouse:       Default routines for sock_ops
  27 *              Arnaldo C. Melo :	removed net_pinfo, tp_pinfo and made
  28 *              			protinfo be just a void pointer, as the
  29 *              			protocol specific parts were moved to
  30 *              			respective headers and ipv4/v6, etc now
  31 *              			use private slabcaches for its socks
  32 *              Pedro Hortas	:	New flags field for socket options
  33 *
  34 *
  35 *		This program is free software; you can redistribute it and/or
  36 *		modify it under the terms of the GNU General Public License
  37 *		as published by the Free Software Foundation; either version
  38 *		2 of the License, or (at your option) any later version.
  39 */
  40#ifndef _SOCK_H
  41#define _SOCK_H
  42
  43#include <linux/hardirq.h>
  44#include <linux/kernel.h>
  45#include <linux/list.h>
  46#include <linux/list_nulls.h>
  47#include <linux/timer.h>
  48#include <linux/cache.h>
  49#include <linux/bitops.h>
  50#include <linux/lockdep.h>
  51#include <linux/netdevice.h>
  52#include <linux/skbuff.h>	/* struct sk_buff */
  53#include <linux/mm.h>
  54#include <linux/security.h>
  55#include <linux/slab.h>
  56#include <linux/uaccess.h>
  57#include <linux/page_counter.h>
  58#include <linux/memcontrol.h>
  59#include <linux/static_key.h>
  60#include <linux/sched.h>
  61#include <linux/wait.h>
  62#include <linux/cgroup-defs.h>
  63#include <linux/rbtree.h>
  64#include <linux/filter.h>
  65#include <linux/rculist_nulls.h>
  66#include <linux/poll.h>
  67
  68#include <linux/atomic.h>
  69#include <linux/refcount.h>
  70#include <net/dst.h>
  71#include <net/checksum.h>
  72#include <net/tcp_states.h>
  73#include <linux/net_tstamp.h>
  74#include <net/smc.h>
  75#include <net/l3mdev.h>
  76
  77/*
  78 * This structure really needs to be cleaned up.
  79 * Most of it is for TCP, and not used by any of
  80 * the other protocols.
  81 */
  82
  83/* Define this to get the SOCK_DBG debugging facility. */
  84#define SOCK_DEBUGGING
  85#ifdef SOCK_DEBUGGING
  86#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
  87					printk(KERN_DEBUG msg); } while (0)
  88#else
  89/* Validate arguments and do nothing */
  90static inline __printf(2, 3)
  91void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
  92{
  93}
  94#endif
  95
  96/* This is the per-socket lock.  The spinlock provides a synchronization
  97 * between user contexts and software interrupt processing, whereas the
  98 * mini-semaphore synchronizes multiple users amongst themselves.
  99 */
 100typedef struct {
 101	spinlock_t		slock;
 102	int			owned;
 103	wait_queue_head_t	wq;
 104	/*
 105	 * We express the mutex-alike socket_lock semantics
 106	 * to the lock validator by explicitly managing
 107	 * the slock as a lock variant (in addition to
 108	 * the slock itself):
 109	 */
 110#ifdef CONFIG_DEBUG_LOCK_ALLOC
 111	struct lockdep_map dep_map;
 112#endif
 113} socket_lock_t;
 114
 115struct sock;
 116struct proto;
 117struct net;
 118
 119typedef __u32 __bitwise __portpair;
 120typedef __u64 __bitwise __addrpair;
 121
 122/**
 123 *	struct sock_common - minimal network layer representation of sockets
 124 *	@skc_daddr: Foreign IPv4 addr
 125 *	@skc_rcv_saddr: Bound local IPv4 addr
 126 *	@skc_hash: hash value used with various protocol lookup tables
 127 *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
 128 *	@skc_dport: placeholder for inet_dport/tw_dport
 129 *	@skc_num: placeholder for inet_num/tw_num
 130 *	@skc_family: network address family
 131 *	@skc_state: Connection state
 132 *	@skc_reuse: %SO_REUSEADDR setting
 133 *	@skc_reuseport: %SO_REUSEPORT setting
 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_node: main hash linkage for various protocol lookup tables
 140 *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
 141 *	@skc_tx_queue_mapping: tx queue number for this connection
 142 *	@skc_flags: place holder for sk_flags
 143 *		%SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
 144 *		%SO_OOBINLINE settings, %SO_TIMESTAMPING settings
 145 *	@skc_incoming_cpu: record/match cpu processing incoming packets
 146 *	@skc_refcnt: reference count
 147 *
 148 *	This is the minimal network layer representation of sockets, the header
 149 *	for struct sock and struct inet_timewait_sock.
 150 */
 151struct sock_common {
 152	/* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
 153	 * address on 64bit arches : cf INET_MATCH()
 154	 */
 155	union {
 156		__addrpair	skc_addrpair;
 157		struct {
 158			__be32	skc_daddr;
 159			__be32	skc_rcv_saddr;
 160		};
 161	};
 162	union  {
 163		unsigned int	skc_hash;
 164		__u16		skc_u16hashes[2];
 165	};
 166	/* skc_dport && skc_num must be grouped as well */
 167	union {
 168		__portpair	skc_portpair;
 169		struct {
 170			__be16	skc_dport;
 171			__u16	skc_num;
 172		};
 173	};
 174
 175	unsigned short		skc_family;
 176	volatile unsigned char	skc_state;
 177	unsigned char		skc_reuse:4;
 178	unsigned char		skc_reuseport:1;
 179	unsigned char		skc_ipv6only:1;
 180	unsigned char		skc_net_refcnt:1;
 181	int			skc_bound_dev_if;
 182	union {
 183		struct hlist_node	skc_bind_node;
 184		struct hlist_node	skc_portaddr_node;
 185	};
 186	struct proto		*skc_prot;
 187	possible_net_t		skc_net;
 188
 189#if IS_ENABLED(CONFIG_IPV6)
 190	struct in6_addr		skc_v6_daddr;
 191	struct in6_addr		skc_v6_rcv_saddr;
 192#endif
 193
 194	atomic64_t		skc_cookie;
 195
 196	/* following fields are padding to force
 197	 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
 198	 * assuming IPV6 is enabled. We use this padding differently
 199	 * for different kind of 'sockets'
 200	 */
 201	union {
 202		unsigned long	skc_flags;
 203		struct sock	*skc_listener; /* request_sock */
 204		struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
 205	};
 206	/*
 207	 * fields between dontcopy_begin/dontcopy_end
 208	 * are not copied in sock_copy()
 209	 */
 210	/* private: */
 211	int			skc_dontcopy_begin[0];
 212	/* public: */
 213	union {
 214		struct hlist_node	skc_node;
 215		struct hlist_nulls_node skc_nulls_node;
 216	};
 217	int			skc_tx_queue_mapping;
 218	union {
 219		int		skc_incoming_cpu;
 220		u32		skc_rcv_wnd;
 221		u32		skc_tw_rcv_nxt; /* struct tcp_timewait_sock  */
 222	};
 223
 224	refcount_t		skc_refcnt;
 225	/* private: */
 226	int                     skc_dontcopy_end[0];
 227	union {
 228		u32		skc_rxhash;
 229		u32		skc_window_clamp;
 230		u32		skc_tw_snd_nxt; /* struct tcp_timewait_sock */
 231	};
 232	/* public: */
 233};
 234
 235/**
 236  *	struct sock - network layer representation of sockets
 237  *	@__sk_common: shared layout with inet_timewait_sock
 238  *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
 239  *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
 240  *	@sk_lock:	synchronizer
 241  *	@sk_kern_sock: True if sock is using kernel lock classes
 242  *	@sk_rcvbuf: size of receive buffer in bytes
 243  *	@sk_wq: sock wait queue and async head
 244  *	@sk_rx_dst: receive input route used by early demux
 245  *	@sk_dst_cache: destination cache
 246  *	@sk_dst_pending_confirm: need to confirm neighbour
 247  *	@sk_policy: flow policy
 248  *	@sk_receive_queue: incoming packets
 249  *	@sk_wmem_alloc: transmit queue bytes committed
 250  *	@sk_tsq_flags: TCP Small Queues flags
 251  *	@sk_write_queue: Packet sending queue
 
 252  *	@sk_omem_alloc: "o" is "option" or "other"
 253  *	@sk_wmem_queued: persistent queue size
 254  *	@sk_forward_alloc: space allocated forward
 255  *	@sk_napi_id: id of the last napi context to receive data for sk
 256  *	@sk_ll_usec: usecs to busypoll when there is no data
 257  *	@sk_allocation: allocation mode
 258  *	@sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
 259  *	@sk_pacing_status: Pacing status (requested, handled by sch_fq)
 260  *	@sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
 261  *	@sk_sndbuf: size of send buffer in bytes
 262  *	@__sk_flags_offset: empty field used to determine location of bitfield
 263  *	@sk_padding: unused element for alignment
 264  *	@sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
 265  *	@sk_no_check_rx: allow zero checksum in RX packets
 266  *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
 267  *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
 268  *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
 269  *	@sk_gso_max_size: Maximum GSO segment size to build
 270  *	@sk_gso_max_segs: Maximum number of GSO segments
 271  *	@sk_pacing_shift: scaling factor for TCP Small Queues
 272  *	@sk_lingertime: %SO_LINGER l_linger setting
 273  *	@sk_backlog: always used with the per-socket spinlock held
 274  *	@sk_callback_lock: used with the callbacks in the end of this struct
 275  *	@sk_error_queue: rarely used
 276  *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
 277  *			  IPV6_ADDRFORM for instance)
 278  *	@sk_err: last error
 279  *	@sk_err_soft: errors that don't cause failure but are the cause of a
 280  *		      persistent failure not just 'timed out'
 281  *	@sk_drops: raw/udp drops counter
 282  *	@sk_ack_backlog: current listen backlog
 283  *	@sk_max_ack_backlog: listen backlog set in listen()
 284  *	@sk_uid: user id of owner
 285  *	@sk_priority: %SO_PRIORITY setting
 286  *	@sk_type: socket type (%SOCK_STREAM, etc)
 287  *	@sk_protocol: which protocol this socket belongs in this network family
 288  *	@sk_peer_pid: &struct pid for this socket's peer
 289  *	@sk_peer_cred: %SO_PEERCRED setting
 290  *	@sk_rcvlowat: %SO_RCVLOWAT setting
 291  *	@sk_rcvtimeo: %SO_RCVTIMEO setting
 292  *	@sk_sndtimeo: %SO_SNDTIMEO setting
 293  *	@sk_txhash: computed flow hash for use on transmit
 294  *	@sk_filter: socket filtering instructions
 
 295  *	@sk_timer: sock cleanup timer
 296  *	@sk_stamp: time stamp of last packet received
 297  *	@sk_tsflags: SO_TIMESTAMPING socket options
 298  *	@sk_tskey: counter to disambiguate concurrent tstamp requests
 299  *	@sk_zckey: counter to order MSG_ZEROCOPY notifications
 300  *	@sk_socket: Identd and reporting IO signals
 301  *	@sk_user_data: RPC layer private data
 302  *	@sk_frag: cached page frag
 303  *	@sk_peek_off: current peek_offset value
 304  *	@sk_send_head: front of stuff to transmit
 305  *	@sk_security: used by security modules
 306  *	@sk_mark: generic packet mark
 307  *	@sk_cgrp_data: cgroup data for this cgroup
 308  *	@sk_memcg: this socket's memory cgroup association
 309  *	@sk_write_pending: a write to stream socket waits to start
 310  *	@sk_state_change: callback to indicate change in the state of the sock
 311  *	@sk_data_ready: callback to indicate there is data to be processed
 312  *	@sk_write_space: callback to indicate there is bf sending space available
 313  *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
 314  *	@sk_backlog_rcv: callback to process the backlog
 315  *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
 316  *	@sk_reuseport_cb: reuseport group container
 317  *	@sk_rcu: used during RCU grace period
 318  */
 319struct sock {
 320	/*
 321	 * Now struct inet_timewait_sock also uses sock_common, so please just
 322	 * don't add nothing before this first member (__sk_common) --acme
 323	 */
 324	struct sock_common	__sk_common;
 325#define sk_node			__sk_common.skc_node
 326#define sk_nulls_node		__sk_common.skc_nulls_node
 327#define sk_refcnt		__sk_common.skc_refcnt
 328#define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
 329
 330#define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
 331#define sk_dontcopy_end		__sk_common.skc_dontcopy_end
 332#define sk_hash			__sk_common.skc_hash
 333#define sk_portpair		__sk_common.skc_portpair
 334#define sk_num			__sk_common.skc_num
 335#define sk_dport		__sk_common.skc_dport
 336#define sk_addrpair		__sk_common.skc_addrpair
 337#define sk_daddr		__sk_common.skc_daddr
 338#define sk_rcv_saddr		__sk_common.skc_rcv_saddr
 339#define sk_family		__sk_common.skc_family
 340#define sk_state		__sk_common.skc_state
 341#define sk_reuse		__sk_common.skc_reuse
 342#define sk_reuseport		__sk_common.skc_reuseport
 343#define sk_ipv6only		__sk_common.skc_ipv6only
 344#define sk_net_refcnt		__sk_common.skc_net_refcnt
 345#define sk_bound_dev_if		__sk_common.skc_bound_dev_if
 346#define sk_bind_node		__sk_common.skc_bind_node
 347#define sk_prot			__sk_common.skc_prot
 348#define sk_net			__sk_common.skc_net
 349#define sk_v6_daddr		__sk_common.skc_v6_daddr
 350#define sk_v6_rcv_saddr	__sk_common.skc_v6_rcv_saddr
 351#define sk_cookie		__sk_common.skc_cookie
 352#define sk_incoming_cpu		__sk_common.skc_incoming_cpu
 353#define sk_flags		__sk_common.skc_flags
 354#define sk_rxhash		__sk_common.skc_rxhash
 355
 356	socket_lock_t		sk_lock;
 357	atomic_t		sk_drops;
 358	int			sk_rcvlowat;
 359	struct sk_buff_head	sk_error_queue;
 360	struct sk_buff_head	sk_receive_queue;
 361	/*
 362	 * The backlog queue is special, it is always used with
 363	 * the per-socket spinlock held and requires low latency
 364	 * access. Therefore we special case it's implementation.
 365	 * Note : rmem_alloc is in this structure to fill a hole
 366	 * on 64bit arches, not because its logically part of
 367	 * backlog.
 368	 */
 369	struct {
 370		atomic_t	rmem_alloc;
 371		int		len;
 372		struct sk_buff	*head;
 373		struct sk_buff	*tail;
 374	} sk_backlog;
 375#define sk_rmem_alloc sk_backlog.rmem_alloc
 376
 377	int			sk_forward_alloc;
 378#ifdef CONFIG_NET_RX_BUSY_POLL
 379	unsigned int		sk_ll_usec;
 380	/* ===== mostly read cache line ===== */
 381	unsigned int		sk_napi_id;
 382#endif
 
 383	int			sk_rcvbuf;
 384
 385	struct sk_filter __rcu	*sk_filter;
 386	union {
 387		struct socket_wq __rcu	*sk_wq;
 388		struct socket_wq	*sk_wq_raw;
 389	};
 
 
 390#ifdef CONFIG_XFRM
 391	struct xfrm_policy __rcu *sk_policy[2];
 392#endif
 393	struct dst_entry	*sk_rx_dst;
 394	struct dst_entry __rcu	*sk_dst_cache;
 
 
 395	atomic_t		sk_omem_alloc;
 396	int			sk_sndbuf;
 397
 398	/* ===== cache line for TX ===== */
 399	int			sk_wmem_queued;
 400	refcount_t		sk_wmem_alloc;
 401	unsigned long		sk_tsq_flags;
 402	union {
 403		struct sk_buff	*sk_send_head;
 404		struct rb_root	tcp_rtx_queue;
 405	};
 406	struct sk_buff_head	sk_write_queue;
 407	__s32			sk_peek_off;
 408	int			sk_write_pending;
 409	__u32			sk_dst_pending_confirm;
 410	u32			sk_pacing_status; /* see enum sk_pacing */
 411	long			sk_sndtimeo;
 412	struct timer_list	sk_timer;
 413	__u32			sk_priority;
 414	__u32			sk_mark;
 415	u32			sk_pacing_rate; /* bytes per second */
 416	u32			sk_max_pacing_rate;
 417	struct page_frag	sk_frag;
 418	netdev_features_t	sk_route_caps;
 419	netdev_features_t	sk_route_nocaps;
 420	netdev_features_t	sk_route_forced_caps;
 421	int			sk_gso_type;
 422	unsigned int		sk_gso_max_size;
 423	gfp_t			sk_allocation;
 424	__u32			sk_txhash;
 425
 426	/*
 427	 * Because of non atomicity rules, all
 428	 * changes are protected by socket lock.
 429	 */
 430	unsigned int		__sk_flags_offset[0];
 431#ifdef __BIG_ENDIAN_BITFIELD
 432#define SK_FL_PROTO_SHIFT  16
 433#define SK_FL_PROTO_MASK   0x00ff0000
 434
 435#define SK_FL_TYPE_SHIFT   0
 436#define SK_FL_TYPE_MASK    0x0000ffff
 437#else
 438#define SK_FL_PROTO_SHIFT  8
 439#define SK_FL_PROTO_MASK   0x0000ff00
 440
 441#define SK_FL_TYPE_SHIFT   16
 442#define SK_FL_TYPE_MASK    0xffff0000
 443#endif
 444
 445	unsigned int		sk_padding : 1,
 446				sk_kern_sock : 1,
 447				sk_no_check_tx : 1,
 448				sk_no_check_rx : 1,
 449				sk_userlocks : 4,
 450				sk_protocol  : 8,
 451				sk_type      : 16;
 452#define SK_PROTOCOL_MAX U8_MAX
 453	u16			sk_gso_max_segs;
 454	u8			sk_pacing_shift;
 
 
 
 
 
 455	unsigned long	        sk_lingertime;
 
 456	struct proto		*sk_prot_creator;
 457	rwlock_t		sk_callback_lock;
 458	int			sk_err,
 459				sk_err_soft;
 460	u32			sk_ack_backlog;
 461	u32			sk_max_ack_backlog;
 462	kuid_t			sk_uid;
 463	struct pid		*sk_peer_pid;
 464	const struct cred	*sk_peer_cred;
 465	long			sk_rcvtimeo;
 
 
 
 466	ktime_t			sk_stamp;
 467	u16			sk_tsflags;
 468	u8			sk_shutdown;
 469	u32			sk_tskey;
 470	atomic_t		sk_zckey;
 471	struct socket		*sk_socket;
 472	void			*sk_user_data;
 
 
 
 
 473#ifdef CONFIG_SECURITY
 474	void			*sk_security;
 475#endif
 476	struct sock_cgroup_data	sk_cgrp_data;
 477	struct mem_cgroup	*sk_memcg;
 478	void			(*sk_state_change)(struct sock *sk);
 479	void			(*sk_data_ready)(struct sock *sk);
 480	void			(*sk_write_space)(struct sock *sk);
 481	void			(*sk_error_report)(struct sock *sk);
 482	int			(*sk_backlog_rcv)(struct sock *sk,
 483						  struct sk_buff *skb);
 484	void                    (*sk_destruct)(struct sock *sk);
 485	struct sock_reuseport __rcu	*sk_reuseport_cb;
 486	struct rcu_head		sk_rcu;
 487};
 488
 489enum sk_pacing {
 490	SK_PACING_NONE		= 0,
 491	SK_PACING_NEEDED	= 1,
 492	SK_PACING_FQ		= 2,
 493};
 494
 495#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
 496
 497#define rcu_dereference_sk_user_data(sk)	rcu_dereference(__sk_user_data((sk)))
 498#define rcu_assign_sk_user_data(sk, ptr)	rcu_assign_pointer(__sk_user_data((sk)), ptr)
 499
 500/*
 501 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
 502 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
 503 * on a socket means that the socket will reuse everybody else's port
 504 * without looking at the other's sk_reuse value.
 505 */
 506
 507#define SK_NO_REUSE	0
 508#define SK_CAN_REUSE	1
 509#define SK_FORCE_REUSE	2
 510
 511int sk_set_peek_off(struct sock *sk, int val);
 512
 513static inline int sk_peek_offset(struct sock *sk, int flags)
 514{
 515	if (unlikely(flags & MSG_PEEK)) {
 516		return READ_ONCE(sk->sk_peek_off);
 517	}
 518
 519	return 0;
 520}
 521
 522static inline void sk_peek_offset_bwd(struct sock *sk, int val)
 523{
 524	s32 off = READ_ONCE(sk->sk_peek_off);
 525
 526	if (unlikely(off >= 0)) {
 527		off = max_t(s32, off - val, 0);
 528		WRITE_ONCE(sk->sk_peek_off, off);
 529	}
 530}
 531
 532static inline void sk_peek_offset_fwd(struct sock *sk, int val)
 533{
 534	sk_peek_offset_bwd(sk, -val);
 535}
 536
 537/*
 538 * Hashed lists helper routines
 539 */
 540static inline struct sock *sk_entry(const struct hlist_node *node)
 541{
 542	return hlist_entry(node, struct sock, sk_node);
 543}
 544
 545static inline struct sock *__sk_head(const struct hlist_head *head)
 546{
 547	return hlist_entry(head->first, struct sock, sk_node);
 548}
 549
 550static inline struct sock *sk_head(const struct hlist_head *head)
 551{
 552	return hlist_empty(head) ? NULL : __sk_head(head);
 553}
 554
 555static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
 556{
 557	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
 558}
 559
 560static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
 561{
 562	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
 563}
 564
 565static inline struct sock *sk_next(const struct sock *sk)
 566{
 567	return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
 
 568}
 569
 570static inline struct sock *sk_nulls_next(const struct sock *sk)
 571{
 572	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
 573		hlist_nulls_entry(sk->sk_nulls_node.next,
 574				  struct sock, sk_nulls_node) :
 575		NULL;
 576}
 577
 578static inline bool sk_unhashed(const struct sock *sk)
 579{
 580	return hlist_unhashed(&sk->sk_node);
 581}
 582
 583static inline bool sk_hashed(const struct sock *sk)
 584{
 585	return !sk_unhashed(sk);
 586}
 587
 588static inline void sk_node_init(struct hlist_node *node)
 589{
 590	node->pprev = NULL;
 591}
 592
 593static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
 594{
 595	node->pprev = NULL;
 596}
 597
 598static inline void __sk_del_node(struct sock *sk)
 599{
 600	__hlist_del(&sk->sk_node);
 601}
 602
 603/* NB: equivalent to hlist_del_init_rcu */
 604static inline bool __sk_del_node_init(struct sock *sk)
 605{
 606	if (sk_hashed(sk)) {
 607		__sk_del_node(sk);
 608		sk_node_init(&sk->sk_node);
 609		return true;
 610	}
 611	return false;
 612}
 613
 614/* Grab socket reference count. This operation is valid only
 615   when sk is ALREADY grabbed f.e. it is found in hash table
 616   or a list and the lookup is made under lock preventing hash table
 617   modifications.
 618 */
 619
 620static __always_inline void sock_hold(struct sock *sk)
 621{
 622	refcount_inc(&sk->sk_refcnt);
 623}
 624
 625/* Ungrab socket in the context, which assumes that socket refcnt
 626   cannot hit zero, f.e. it is true in context of any socketcall.
 627 */
 628static __always_inline void __sock_put(struct sock *sk)
 629{
 630	refcount_dec(&sk->sk_refcnt);
 631}
 632
 633static inline bool sk_del_node_init(struct sock *sk)
 634{
 635	bool rc = __sk_del_node_init(sk);
 636
 637	if (rc) {
 638		/* paranoid for a while -acme */
 639		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 640		__sock_put(sk);
 641	}
 642	return rc;
 643}
 644#define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
 645
 646static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
 647{
 648	if (sk_hashed(sk)) {
 649		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
 650		return true;
 651	}
 652	return false;
 653}
 654
 655static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
 656{
 657	bool rc = __sk_nulls_del_node_init_rcu(sk);
 658
 659	if (rc) {
 660		/* paranoid for a while -acme */
 661		WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
 662		__sock_put(sk);
 663	}
 664	return rc;
 665}
 666
 667static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
 668{
 669	hlist_add_head(&sk->sk_node, list);
 670}
 671
 672static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
 673{
 674	sock_hold(sk);
 675	__sk_add_node(sk, list);
 676}
 677
 678static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
 679{
 680	sock_hold(sk);
 681	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
 682	    sk->sk_family == AF_INET6)
 683		hlist_add_tail_rcu(&sk->sk_node, list);
 684	else
 685		hlist_add_head_rcu(&sk->sk_node, list);
 686}
 687
 688static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 689{
 690	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
 691}
 692
 693static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
 694{
 695	sock_hold(sk);
 696	__sk_nulls_add_node_rcu(sk, list);
 697}
 698
 699static inline void __sk_del_bind_node(struct sock *sk)
 700{
 701	__hlist_del(&sk->sk_bind_node);
 702}
 703
 704static inline void sk_add_bind_node(struct sock *sk,
 705					struct hlist_head *list)
 706{
 707	hlist_add_head(&sk->sk_bind_node, list);
 708}
 709
 710#define sk_for_each(__sk, list) \
 711	hlist_for_each_entry(__sk, list, sk_node)
 712#define sk_for_each_rcu(__sk, list) \
 713	hlist_for_each_entry_rcu(__sk, list, sk_node)
 714#define sk_nulls_for_each(__sk, node, list) \
 715	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
 716#define sk_nulls_for_each_rcu(__sk, node, list) \
 717	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
 718#define sk_for_each_from(__sk) \
 719	hlist_for_each_entry_from(__sk, sk_node)
 
 720#define sk_nulls_for_each_from(__sk, node) \
 721	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
 722		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
 723#define sk_for_each_safe(__sk, tmp, list) \
 724	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
 725#define sk_for_each_bound(__sk, list) \
 726	hlist_for_each_entry(__sk, list, sk_bind_node)
 727
 728/**
 729 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
 730 * @tpos:	the type * to use as a loop cursor.
 731 * @pos:	the &struct hlist_node to use as a loop cursor.
 732 * @head:	the head for your list.
 733 * @offset:	offset of hlist_node within the struct.
 734 *
 735 */
 736#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset)		       \
 737	for (pos = rcu_dereference(hlist_first_rcu(head));		       \
 738	     pos != NULL &&						       \
 739		({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;});       \
 740	     pos = rcu_dereference(hlist_next_rcu(pos)))
 741
 742static inline struct user_namespace *sk_user_ns(struct sock *sk)
 743{
 744	/* Careful only use this in a context where these parameters
 745	 * can not change and must all be valid, such as recvmsg from
 746	 * userspace.
 747	 */
 748	return sk->sk_socket->file->f_cred->user_ns;
 749}
 750
 751/* Sock flags */
 752enum sock_flags {
 753	SOCK_DEAD,
 754	SOCK_DONE,
 755	SOCK_URGINLINE,
 756	SOCK_KEEPOPEN,
 757	SOCK_LINGER,
 758	SOCK_DESTROY,
 759	SOCK_BROADCAST,
 760	SOCK_TIMESTAMP,
 761	SOCK_ZAPPED,
 762	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
 763	SOCK_DBG, /* %SO_DEBUG setting */
 764	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
 765	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
 766	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
 767	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
 768	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
 
 
 769	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
 
 
 
 770	SOCK_FASYNC, /* fasync() active */
 771	SOCK_RXQ_OVFL,
 772	SOCK_ZEROCOPY, /* buffers from userspace */
 773	SOCK_WIFI_STATUS, /* push wifi status to userspace */
 774	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
 775		     * Will use last 4 bytes of packet sent from
 776		     * user-space instead.
 777		     */
 778	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
 779	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
 780	SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
 781};
 782
 783#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
 784
 785static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
 786{
 787	nsk->sk_flags = osk->sk_flags;
 788}
 789
 790static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
 791{
 792	__set_bit(flag, &sk->sk_flags);
 793}
 794
 795static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
 796{
 797	__clear_bit(flag, &sk->sk_flags);
 798}
 799
 800static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
 801{
 802	return test_bit(flag, &sk->sk_flags);
 803}
 804
 805#ifdef CONFIG_NET
 806extern struct static_key memalloc_socks;
 807static inline int sk_memalloc_socks(void)
 808{
 809	return static_key_false(&memalloc_socks);
 810}
 811#else
 812
 813static inline int sk_memalloc_socks(void)
 814{
 815	return 0;
 816}
 817
 818#endif
 819
 820static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
 821{
 822	return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
 823}
 824
 825static inline void sk_acceptq_removed(struct sock *sk)
 826{
 827	sk->sk_ack_backlog--;
 828}
 829
 830static inline void sk_acceptq_added(struct sock *sk)
 831{
 832	sk->sk_ack_backlog++;
 833}
 834
 835static inline bool sk_acceptq_is_full(const struct sock *sk)
 836{
 837	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
 838}
 839
 840/*
 841 * Compute minimal free write space needed to queue new packets.
 842 */
 843static inline int sk_stream_min_wspace(const struct sock *sk)
 844{
 845	return sk->sk_wmem_queued >> 1;
 846}
 847
 848static inline int sk_stream_wspace(const struct sock *sk)
 849{
 850	return sk->sk_sndbuf - sk->sk_wmem_queued;
 851}
 852
 853void sk_stream_write_space(struct sock *sk);
 
 
 
 
 
 854
 855/* OOB backlog add */
 856static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
 857{
 858	/* dont let skb dst not refcounted, we are going to leave rcu lock */
 859	skb_dst_force(skb);
 860
 861	if (!sk->sk_backlog.tail)
 862		sk->sk_backlog.head = skb;
 863	else
 864		sk->sk_backlog.tail->next = skb;
 865
 866	sk->sk_backlog.tail = skb;
 867	skb->next = NULL;
 868}
 869
 870/*
 871 * Take into account size of receive queue and backlog queue
 872 * Do not take into account this skb truesize,
 873 * to allow even a single big packet to come.
 874 */
 875static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
 876{
 877	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
 878
 879	return qsize > limit;
 880}
 881
 882/* The per-socket spinlock must be held here. */
 883static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
 884					      unsigned int limit)
 885{
 886	if (sk_rcvqueues_full(sk, limit))
 887		return -ENOBUFS;
 888
 889	/*
 890	 * If the skb was allocated from pfmemalloc reserves, only
 891	 * allow SOCK_MEMALLOC sockets to use it as this socket is
 892	 * helping free memory
 893	 */
 894	if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
 895		return -ENOMEM;
 896
 897	__sk_add_backlog(sk, skb);
 898	sk->sk_backlog.len += skb->truesize;
 899	return 0;
 900}
 901
 902int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
 903
 904static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 905{
 906	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
 907		return __sk_backlog_rcv(sk, skb);
 908
 909	return sk->sk_backlog_rcv(sk, skb);
 910}
 911
 912static inline void sk_incoming_cpu_update(struct sock *sk)
 913{
 914	int cpu = raw_smp_processor_id();
 915
 916	if (unlikely(sk->sk_incoming_cpu != cpu))
 917		sk->sk_incoming_cpu = cpu;
 918}
 919
 920static inline void sock_rps_record_flow_hash(__u32 hash)
 921{
 922#ifdef CONFIG_RPS
 923	struct rps_sock_flow_table *sock_flow_table;
 924
 925	rcu_read_lock();
 926	sock_flow_table = rcu_dereference(rps_sock_flow_table);
 927	rps_record_sock_flow(sock_flow_table, hash);
 928	rcu_read_unlock();
 929#endif
 930}
 931
 932static inline void sock_rps_record_flow(const struct sock *sk)
 933{
 934#ifdef CONFIG_RPS
 935	if (static_key_false(&rfs_needed)) {
 936		/* Reading sk->sk_rxhash might incur an expensive cache line
 937		 * miss.
 938		 *
 939		 * TCP_ESTABLISHED does cover almost all states where RFS
 940		 * might be useful, and is cheaper [1] than testing :
 941		 *	IPv4: inet_sk(sk)->inet_daddr
 942		 * 	IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
 943		 * OR	an additional socket flag
 944		 * [1] : sk_state and sk_prot are in the same cache line.
 945		 */
 946		if (sk->sk_state == TCP_ESTABLISHED)
 947			sock_rps_record_flow_hash(sk->sk_rxhash);
 948	}
 949#endif
 950}
 951
 952static inline void sock_rps_save_rxhash(struct sock *sk,
 953					const struct sk_buff *skb)
 954{
 955#ifdef CONFIG_RPS
 956	if (unlikely(sk->sk_rxhash != skb->hash))
 957		sk->sk_rxhash = skb->hash;
 958#endif
 959}
 960
 961static inline void sock_rps_reset_rxhash(struct sock *sk)
 962{
 963#ifdef CONFIG_RPS
 964	sk->sk_rxhash = 0;
 
 
 
 965#endif
 966}
 967
 968#define sk_wait_event(__sk, __timeo, __condition, __wait)		\
 969	({	int __rc;						\
 970		release_sock(__sk);					\
 971		__rc = __condition;					\
 972		if (!__rc) {						\
 973			*(__timeo) = wait_woken(__wait,			\
 974						TASK_INTERRUPTIBLE,	\
 975						*(__timeo));		\
 976		}							\
 977		sched_annotate_sleep();					\
 978		lock_sock(__sk);					\
 979		__rc = __condition;					\
 980		__rc;							\
 981	})
 982
 983int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
 984int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
 985void sk_stream_wait_close(struct sock *sk, long timeo_p);
 986int sk_stream_error(struct sock *sk, int flags, int err);
 987void sk_stream_kill_queues(struct sock *sk);
 988void sk_set_memalloc(struct sock *sk);
 989void sk_clear_memalloc(struct sock *sk);
 990
 991void __sk_flush_backlog(struct sock *sk);
 992
 993static inline bool sk_flush_backlog(struct sock *sk)
 994{
 995	if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
 996		__sk_flush_backlog(sk);
 997		return true;
 998	}
 999	return false;
1000}
1001
1002int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1003
1004struct request_sock_ops;
1005struct timewait_sock_ops;
1006struct inet_hashinfo;
1007struct raw_hashinfo;
1008struct smc_hashinfo;
1009struct module;
1010
1011/*
1012 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1013 * un-modified. Special care is taken when initializing object to zero.
1014 */
1015static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1016{
1017	if (offsetof(struct sock, sk_node.next) != 0)
1018		memset(sk, 0, offsetof(struct sock, sk_node.next));
1019	memset(&sk->sk_node.pprev, 0,
1020	       size - offsetof(struct sock, sk_node.pprev));
1021}
1022
1023/* Networking protocol blocks we attach to sockets.
1024 * socket layer -> transport layer interface
 
1025 */
1026struct proto {
1027	void			(*close)(struct sock *sk,
1028					long timeout);
1029	int			(*pre_connect)(struct sock *sk,
1030					struct sockaddr *uaddr,
1031					int addr_len);
1032	int			(*connect)(struct sock *sk,
1033					struct sockaddr *uaddr,
1034					int addr_len);
1035	int			(*disconnect)(struct sock *sk, int flags);
1036
1037	struct sock *		(*accept)(struct sock *sk, int flags, int *err,
1038					  bool kern);
1039
1040	int			(*ioctl)(struct sock *sk, int cmd,
1041					 unsigned long arg);
1042	int			(*init)(struct sock *sk);
1043	void			(*destroy)(struct sock *sk);
1044	void			(*shutdown)(struct sock *sk, int how);
1045	int			(*setsockopt)(struct sock *sk, int level,
1046					int optname, char __user *optval,
1047					unsigned int optlen);
1048	int			(*getsockopt)(struct sock *sk, int level,
1049					int optname, char __user *optval,
1050					int __user *option);
1051	void			(*keepalive)(struct sock *sk, int valbool);
1052#ifdef CONFIG_COMPAT
1053	int			(*compat_setsockopt)(struct sock *sk,
1054					int level,
1055					int optname, char __user *optval,
1056					unsigned int optlen);
1057	int			(*compat_getsockopt)(struct sock *sk,
1058					int level,
1059					int optname, char __user *optval,
1060					int __user *option);
1061	int			(*compat_ioctl)(struct sock *sk,
1062					unsigned int cmd, unsigned long arg);
1063#endif
1064	int			(*sendmsg)(struct sock *sk, struct msghdr *msg,
1065					   size_t len);
1066	int			(*recvmsg)(struct sock *sk, struct msghdr *msg,
1067					   size_t len, int noblock, int flags,
1068					   int *addr_len);
 
1069	int			(*sendpage)(struct sock *sk, struct page *page,
1070					int offset, size_t size, int flags);
1071	int			(*bind)(struct sock *sk,
1072					struct sockaddr *uaddr, int addr_len);
1073
1074	int			(*backlog_rcv) (struct sock *sk,
1075						struct sk_buff *skb);
1076
1077	void		(*release_cb)(struct sock *sk);
1078
1079	/* Keeping track of sk's, looking them up, and port selection methods. */
1080	int			(*hash)(struct sock *sk);
1081	void			(*unhash)(struct sock *sk);
1082	void			(*rehash)(struct sock *sk);
1083	int			(*get_port)(struct sock *sk, unsigned short snum);
 
1084
1085	/* Keeping track of sockets in use */
1086#ifdef CONFIG_PROC_FS
1087	unsigned int		inuse_idx;
1088#endif
1089
1090	bool			(*stream_memory_free)(const struct sock *sk);
1091	bool			(*stream_memory_read)(const struct sock *sk);
1092	/* Memory pressure */
1093	void			(*enter_memory_pressure)(struct sock *sk);
1094	void			(*leave_memory_pressure)(struct sock *sk);
1095	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
1096	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1097	/*
1098	 * Pressure flag: try to collapse.
1099	 * Technical note: it is used by multiple contexts non atomically.
1100	 * All the __sk_mem_schedule() is of this nature: accounting
1101	 * is strict, actions are advisory and have some latency.
1102	 */
1103	unsigned long		*memory_pressure;
1104	long			*sysctl_mem;
1105
1106	int			*sysctl_wmem;
1107	int			*sysctl_rmem;
1108	u32			sysctl_wmem_offset;
1109	u32			sysctl_rmem_offset;
1110
1111	int			max_header;
1112	bool			no_autobind;
1113
1114	struct kmem_cache	*slab;
1115	unsigned int		obj_size;
1116	slab_flags_t		slab_flags;
1117	unsigned int		useroffset;	/* Usercopy region offset */
1118	unsigned int		usersize;	/* Usercopy region size */
1119
1120	struct percpu_counter	*orphan_count;
1121
1122	struct request_sock_ops	*rsk_prot;
1123	struct timewait_sock_ops *twsk_prot;
1124
1125	union {
1126		struct inet_hashinfo	*hashinfo;
1127		struct udp_table	*udp_table;
1128		struct raw_hashinfo	*raw_hash;
1129		struct smc_hashinfo	*smc_hash;
1130	} h;
1131
1132	struct module		*owner;
1133
1134	char			name[32];
1135
1136	struct list_head	node;
1137#ifdef SOCK_REFCNT_DEBUG
1138	atomic_t		socks;
1139#endif
1140	int			(*diag_destroy)(struct sock *sk, int err);
1141} __randomize_layout;
1142
1143int proto_register(struct proto *prot, int alloc_slab);
1144void proto_unregister(struct proto *prot);
1145int sock_load_diag_module(int family, int protocol);
1146
1147#ifdef SOCK_REFCNT_DEBUG
1148static inline void sk_refcnt_debug_inc(struct sock *sk)
1149{
1150	atomic_inc(&sk->sk_prot->socks);
1151}
1152
1153static inline void sk_refcnt_debug_dec(struct sock *sk)
1154{
1155	atomic_dec(&sk->sk_prot->socks);
1156	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1157	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1158}
1159
1160static inline void sk_refcnt_debug_release(const struct sock *sk)
1161{
1162	if (refcount_read(&sk->sk_refcnt) != 1)
1163		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1164		       sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1165}
1166#else /* SOCK_REFCNT_DEBUG */
1167#define sk_refcnt_debug_inc(sk) do { } while (0)
1168#define sk_refcnt_debug_dec(sk) do { } while (0)
1169#define sk_refcnt_debug_release(sk) do { } while (0)
1170#endif /* SOCK_REFCNT_DEBUG */
1171
1172static inline bool sk_stream_memory_free(const struct sock *sk)
1173{
1174	if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1175		return false;
1176
1177	return sk->sk_prot->stream_memory_free ?
1178		sk->sk_prot->stream_memory_free(sk) : true;
1179}
1180
1181static inline bool sk_stream_is_writeable(const struct sock *sk)
1182{
1183	return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1184	       sk_stream_memory_free(sk);
1185}
1186
1187static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1188					    struct cgroup *ancestor)
1189{
1190#ifdef CONFIG_SOCK_CGROUP_DATA
1191	return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1192				    ancestor);
1193#else
1194	return -ENOTSUPP;
1195#endif
1196}
1197
1198static inline bool sk_has_memory_pressure(const struct sock *sk)
1199{
1200	return sk->sk_prot->memory_pressure != NULL;
1201}
1202
1203static inline bool sk_under_memory_pressure(const struct sock *sk)
1204{
1205	if (!sk->sk_prot->memory_pressure)
1206		return false;
1207
1208	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1209	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
1210		return true;
1211
1212	return !!*sk->sk_prot->memory_pressure;
1213}
1214
1215static inline long
1216sk_memory_allocated(const struct sock *sk)
1217{
1218	return atomic_long_read(sk->sk_prot->memory_allocated);
1219}
1220
1221static inline long
1222sk_memory_allocated_add(struct sock *sk, int amt)
1223{
1224	return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1225}
1226
1227static inline void
1228sk_memory_allocated_sub(struct sock *sk, int amt)
1229{
1230	atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1231}
1232
1233static inline void sk_sockets_allocated_dec(struct sock *sk)
1234{
1235	percpu_counter_dec(sk->sk_prot->sockets_allocated);
1236}
1237
1238static inline void sk_sockets_allocated_inc(struct sock *sk)
1239{
1240	percpu_counter_inc(sk->sk_prot->sockets_allocated);
1241}
1242
1243static inline int
1244sk_sockets_allocated_read_positive(struct sock *sk)
1245{
1246	return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1247}
1248
1249static inline int
1250proto_sockets_allocated_sum_positive(struct proto *prot)
1251{
1252	return percpu_counter_sum_positive(prot->sockets_allocated);
1253}
1254
1255static inline long
1256proto_memory_allocated(struct proto *prot)
1257{
1258	return atomic_long_read(prot->memory_allocated);
1259}
1260
1261static inline bool
1262proto_memory_pressure(struct proto *prot)
1263{
1264	if (!prot->memory_pressure)
1265		return false;
1266	return !!*prot->memory_pressure;
1267}
1268
1269
1270#ifdef CONFIG_PROC_FS
1271/* Called with local bh disabled */
1272void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1273int sock_prot_inuse_get(struct net *net, struct proto *proto);
1274int sock_inuse_get(struct net *net);
1275#else
1276static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1277		int inc)
1278{
1279}
1280#endif
1281
1282
1283/* With per-bucket locks this operation is not-atomic, so that
1284 * this version is not worse.
1285 */
1286static inline int __sk_prot_rehash(struct sock *sk)
1287{
1288	sk->sk_prot->unhash(sk);
1289	return sk->sk_prot->hash(sk);
1290}
1291
 
 
1292/* About 10 seconds */
1293#define SOCK_DESTROY_TIME (10*HZ)
1294
1295/* Sockets 0-1023 can't be bound to unless you are superuser */
1296#define PROT_SOCK	1024
1297
1298#define SHUTDOWN_MASK	3
1299#define RCV_SHUTDOWN	1
1300#define SEND_SHUTDOWN	2
1301
1302#define SOCK_SNDBUF_LOCK	1
1303#define SOCK_RCVBUF_LOCK	2
1304#define SOCK_BINDADDR_LOCK	4
1305#define SOCK_BINDPORT_LOCK	8
1306
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1307struct socket_alloc {
1308	struct socket socket;
1309	struct inode vfs_inode;
1310};
1311
1312static inline struct socket *SOCKET_I(struct inode *inode)
1313{
1314	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1315}
1316
1317static inline struct inode *SOCK_INODE(struct socket *socket)
1318{
1319	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1320}
1321
1322/*
1323 * Functions for memory accounting
1324 */
1325int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1326int __sk_mem_schedule(struct sock *sk, int size, int kind);
1327void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1328void __sk_mem_reclaim(struct sock *sk, int amount);
1329
1330/* We used to have PAGE_SIZE here, but systems with 64KB pages
1331 * do not necessarily have 16x time more memory than 4KB ones.
1332 */
1333#define SK_MEM_QUANTUM 4096
1334#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1335#define SK_MEM_SEND	0
1336#define SK_MEM_RECV	1
1337
1338/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1339static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1340{
1341	long val = sk->sk_prot->sysctl_mem[index];
1342
1343#if PAGE_SIZE > SK_MEM_QUANTUM
1344	val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1345#elif PAGE_SIZE < SK_MEM_QUANTUM
1346	val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1347#endif
1348	return val;
1349}
1350
1351static inline int sk_mem_pages(int amt)
1352{
1353	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1354}
1355
1356static inline bool sk_has_account(struct sock *sk)
1357{
1358	/* return true if protocol supports memory accounting */
1359	return !!sk->sk_prot->memory_allocated;
1360}
1361
1362static inline bool sk_wmem_schedule(struct sock *sk, int size)
1363{
1364	if (!sk_has_account(sk))
1365		return true;
1366	return size <= sk->sk_forward_alloc ||
1367		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1368}
1369
1370static inline bool
1371sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1372{
1373	if (!sk_has_account(sk))
1374		return true;
1375	return size<= sk->sk_forward_alloc ||
1376		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1377		skb_pfmemalloc(skb);
1378}
1379
1380static inline void sk_mem_reclaim(struct sock *sk)
1381{
1382	if (!sk_has_account(sk))
1383		return;
1384	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1385		__sk_mem_reclaim(sk, sk->sk_forward_alloc);
1386}
1387
1388static inline void sk_mem_reclaim_partial(struct sock *sk)
1389{
1390	if (!sk_has_account(sk))
1391		return;
1392	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1393		__sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1394}
1395
1396static inline void sk_mem_charge(struct sock *sk, int size)
1397{
1398	if (!sk_has_account(sk))
1399		return;
1400	sk->sk_forward_alloc -= size;
1401}
1402
1403static inline void sk_mem_uncharge(struct sock *sk, int size)
1404{
1405	if (!sk_has_account(sk))
1406		return;
1407	sk->sk_forward_alloc += size;
1408
1409	/* Avoid a possible overflow.
1410	 * TCP send queues can make this happen, if sk_mem_reclaim()
1411	 * is not called and more than 2 GBytes are released at once.
1412	 *
1413	 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1414	 * no need to hold that much forward allocation anyway.
1415	 */
1416	if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1417		__sk_mem_reclaim(sk, 1 << 20);
1418}
1419
1420static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1421{
1422	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1423	sk->sk_wmem_queued -= skb->truesize;
1424	sk_mem_uncharge(sk, skb->truesize);
1425	__kfree_skb(skb);
1426}
1427
1428static inline void sock_release_ownership(struct sock *sk)
1429{
1430	if (sk->sk_lock.owned) {
1431		sk->sk_lock.owned = 0;
1432
1433		/* The sk_lock has mutex_unlock() semantics: */
1434		mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1435	}
1436}
 
 
 
 
 
1437
1438/*
1439 * Macro so as to not evaluate some arguments when
1440 * lockdep is not enabled.
1441 *
1442 * Mark both the sk_lock and the sk_lock.slock as a
1443 * per-address-family lock class.
1444 */
1445#define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1446do {									\
1447	sk->sk_lock.owned = 0;						\
1448	init_waitqueue_head(&sk->sk_lock.wq);				\
1449	spin_lock_init(&(sk)->sk_lock.slock);				\
1450	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1451			sizeof((sk)->sk_lock));				\
1452	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1453				(skey), (sname));				\
1454	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1455} while (0)
1456
1457#ifdef CONFIG_LOCKDEP
1458static inline bool lockdep_sock_is_held(const struct sock *sk)
1459{
1460	return lockdep_is_held(&sk->sk_lock) ||
1461	       lockdep_is_held(&sk->sk_lock.slock);
1462}
1463#endif
1464
1465void lock_sock_nested(struct sock *sk, int subclass);
1466
1467static inline void lock_sock(struct sock *sk)
1468{
1469	lock_sock_nested(sk, 0);
1470}
1471
1472void release_sock(struct sock *sk);
1473
1474/* BH context may only use the following locking interface. */
1475#define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1476#define bh_lock_sock_nested(__sk) \
1477				spin_lock_nested(&((__sk)->sk_lock.slock), \
1478				SINGLE_DEPTH_NESTING)
1479#define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1480
1481bool lock_sock_fast(struct sock *sk);
1482/**
1483 * unlock_sock_fast - complement of lock_sock_fast
1484 * @sk: socket
1485 * @slow: slow mode
1486 *
1487 * fast unlock socket for user context.
1488 * If slow mode is on, we call regular release_sock()
1489 */
1490static inline void unlock_sock_fast(struct sock *sk, bool slow)
1491{
1492	if (slow)
1493		release_sock(sk);
1494	else
1495		spin_unlock_bh(&sk->sk_lock.slock);
1496}
1497
1498/* Used by processes to "lock" a socket state, so that
1499 * interrupts and bottom half handlers won't change it
1500 * from under us. It essentially blocks any incoming
1501 * packets, so that we won't get any new data or any
1502 * packets that change the state of the socket.
1503 *
1504 * While locked, BH processing will add new packets to
1505 * the backlog queue.  This queue is processed by the
1506 * owner of the socket lock right before it is released.
1507 *
1508 * Since ~2.3.5 it is also exclusive sleep lock serializing
1509 * accesses from user process context.
1510 */
1511
1512static inline void sock_owned_by_me(const struct sock *sk)
1513{
1514#ifdef CONFIG_LOCKDEP
1515	WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1516#endif
1517}
1518
1519static inline bool sock_owned_by_user(const struct sock *sk)
1520{
1521	sock_owned_by_me(sk);
1522	return sk->sk_lock.owned;
1523}
1524
1525static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1526{
1527	return sk->sk_lock.owned;
1528}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1529
1530/* no reclassification while locks are held */
1531static inline bool sock_allow_reclassification(const struct sock *csk)
 
 
1532{
1533	struct sock *sk = (struct sock *)csk;
1534
1535	return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1536}
1537
1538struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1539		      struct proto *prot, int kern);
1540void sk_free(struct sock *sk);
1541void sk_destruct(struct sock *sk);
1542struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1543void sk_free_unlock_clone(struct sock *sk);
1544
1545struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1546			     gfp_t priority);
1547void __sock_wfree(struct sk_buff *skb);
1548void sock_wfree(struct sk_buff *skb);
1549struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1550			     gfp_t priority);
1551void skb_orphan_partial(struct sk_buff *skb);
1552void sock_rfree(struct sk_buff *skb);
1553void sock_efree(struct sk_buff *skb);
1554#ifdef CONFIG_INET
1555void sock_edemux(struct sk_buff *skb);
1556#else
1557#define sock_edemux sock_efree
1558#endif
1559
1560int sock_setsockopt(struct socket *sock, int level, int op,
1561		    char __user *optval, unsigned int optlen);
1562
1563int sock_getsockopt(struct socket *sock, int level, int op,
1564		    char __user *optval, int __user *optlen);
1565struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1566				    int noblock, int *errcode);
1567struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1568				     unsigned long data_len, int noblock,
1569				     int *errcode, int max_page_order);
1570void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1571void sock_kfree_s(struct sock *sk, void *mem, int size);
1572void sock_kzfree_s(struct sock *sk, void *mem, int size);
1573void sk_send_sigurg(struct sock *sk);
1574
1575struct sockcm_cookie {
1576	u32 mark;
1577	u16 tsflags;
1578};
1579
1580int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1581		     struct sockcm_cookie *sockc);
1582int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1583		   struct sockcm_cookie *sockc);
1584
1585/*
1586 * Functions to fill in entries in struct proto_ops when a protocol
1587 * does not implement a particular function.
1588 */
1589int sock_no_bind(struct socket *, struct sockaddr *, int);
1590int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1591int sock_no_socketpair(struct socket *, struct socket *);
1592int sock_no_accept(struct socket *, struct socket *, int, bool);
1593int sock_no_getname(struct socket *, struct sockaddr *, int);
1594__poll_t sock_no_poll(struct file *, struct socket *,
1595			  struct poll_table_struct *);
1596int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1597int sock_no_listen(struct socket *, int);
1598int sock_no_shutdown(struct socket *, int);
1599int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1600int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1601int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1602int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1603int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1604int sock_no_mmap(struct file *file, struct socket *sock,
1605		 struct vm_area_struct *vma);
1606ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1607			 size_t size, int flags);
1608ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1609				int offset, size_t size, int flags);
 
 
 
 
 
 
 
 
 
 
1610
1611/*
1612 * Functions to fill in entries in struct proto_ops when a protocol
1613 * uses the inet style.
1614 */
1615int sock_common_getsockopt(struct socket *sock, int level, int optname,
1616				  char __user *optval, int __user *optlen);
1617int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1618			int flags);
1619int sock_common_setsockopt(struct socket *sock, int level, int optname,
1620				  char __user *optval, unsigned int optlen);
1621int compat_sock_common_getsockopt(struct socket *sock, int level,
1622		int optname, char __user *optval, int __user *optlen);
1623int compat_sock_common_setsockopt(struct socket *sock, int level,
1624		int optname, char __user *optval, unsigned int optlen);
1625
1626void sk_common_release(struct sock *sk);
1627
1628/*
1629 *	Default socket callbacks and setup code
1630 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1631
1632/* Initialise core socket variables */
1633void sock_init_data(struct socket *sock, struct sock *sk);
 
 
 
1634
1635/*
1636 * Socket reference counting postulates.
1637 *
1638 * * Each user of socket SHOULD hold a reference count.
1639 * * Each access point to socket (an hash table bucket, reference from a list,
1640 *   running timer, skb in flight MUST hold a reference count.
1641 * * When reference count hits 0, it means it will never increase back.
1642 * * When reference count hits 0, it means that no references from
1643 *   outside exist to this socket and current process on current CPU
1644 *   is last user and may/should destroy this socket.
1645 * * sk_free is called from any context: process, BH, IRQ. When
1646 *   it is called, socket has no references from outside -> sk_free
1647 *   may release descendant resources allocated by the socket, but
1648 *   to the time when it is called, socket is NOT referenced by any
1649 *   hash tables, lists etc.
1650 * * Packets, delivered from outside (from network or from another process)
1651 *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1652 *   when they sit in queue. Otherwise, packets will leak to hole, when
1653 *   socket is looked up by one cpu and unhasing is made by another CPU.
1654 *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1655 *   (leak to backlog). Packet socket does all the processing inside
1656 *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1657 *   use separate SMP lock, so that they are prone too.
1658 */
1659
1660/* Ungrab socket and destroy it, if it was the last reference. */
1661static inline void sock_put(struct sock *sk)
1662{
1663	if (refcount_dec_and_test(&sk->sk_refcnt))
1664		sk_free(sk);
1665}
1666/* Generic version of sock_put(), dealing with all sockets
1667 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1668 */
1669void sock_gen_put(struct sock *sk);
1670
1671int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1672		     unsigned int trim_cap, bool refcounted);
1673static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1674				 const int nested)
1675{
1676	return __sk_receive_skb(sk, skb, nested, 1, true);
1677}
1678
1679static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1680{
1681	sk->sk_tx_queue_mapping = tx_queue;
1682}
1683
1684static inline void sk_tx_queue_clear(struct sock *sk)
1685{
1686	sk->sk_tx_queue_mapping = -1;
1687}
1688
1689static inline int sk_tx_queue_get(const struct sock *sk)
1690{
1691	return sk ? sk->sk_tx_queue_mapping : -1;
1692}
1693
1694static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1695{
1696	sk_tx_queue_clear(sk);
1697	sk->sk_socket = sock;
1698}
1699
1700static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1701{
1702	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1703	return &rcu_dereference_raw(sk->sk_wq)->wait;
1704}
1705/* Detach socket from process context.
1706 * Announce socket dead, detach it from wait queue and inode.
1707 * Note that parent inode held reference count on this struct sock,
1708 * we do not release it in this function, because protocol
1709 * probably wants some additional cleanups or even continuing
1710 * to work with this socket (TCP).
1711 */
1712static inline void sock_orphan(struct sock *sk)
1713{
1714	write_lock_bh(&sk->sk_callback_lock);
1715	sock_set_flag(sk, SOCK_DEAD);
1716	sk_set_socket(sk, NULL);
1717	sk->sk_wq  = NULL;
1718	write_unlock_bh(&sk->sk_callback_lock);
1719}
1720
1721static inline void sock_graft(struct sock *sk, struct socket *parent)
1722{
1723	WARN_ON(parent->sk);
1724	write_lock_bh(&sk->sk_callback_lock);
1725	sk->sk_wq = parent->wq;
1726	parent->sk = sk;
1727	sk_set_socket(sk, parent);
1728	sk->sk_uid = SOCK_INODE(parent)->i_uid;
1729	security_sock_graft(sk, parent);
1730	write_unlock_bh(&sk->sk_callback_lock);
1731}
1732
1733kuid_t sock_i_uid(struct sock *sk);
1734unsigned long sock_i_ino(struct sock *sk);
1735
1736static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1737{
1738	return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1739}
1740
1741static inline u32 net_tx_rndhash(void)
1742{
1743	u32 v = prandom_u32();
1744
1745	return v ?: 1;
1746}
1747
1748static inline void sk_set_txhash(struct sock *sk)
1749{
1750	sk->sk_txhash = net_tx_rndhash();
1751}
1752
1753static inline void sk_rethink_txhash(struct sock *sk)
1754{
1755	if (sk->sk_txhash)
1756		sk_set_txhash(sk);
1757}
1758
1759static inline struct dst_entry *
1760__sk_dst_get(struct sock *sk)
1761{
1762	return rcu_dereference_check(sk->sk_dst_cache,
1763				     lockdep_sock_is_held(sk));
1764}
1765
1766static inline struct dst_entry *
1767sk_dst_get(struct sock *sk)
1768{
1769	struct dst_entry *dst;
1770
1771	rcu_read_lock();
1772	dst = rcu_dereference(sk->sk_dst_cache);
1773	if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1774		dst = NULL;
1775	rcu_read_unlock();
1776	return dst;
1777}
1778
 
 
1779static inline void dst_negative_advice(struct sock *sk)
1780{
1781	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1782
1783	sk_rethink_txhash(sk);
1784
1785	if (dst && dst->ops->negative_advice) {
1786		ndst = dst->ops->negative_advice(dst);
1787
1788		if (ndst != dst) {
1789			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1790			sk_tx_queue_clear(sk);
1791			sk->sk_dst_pending_confirm = 0;
1792		}
1793	}
1794}
1795
1796static inline void
1797__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1798{
1799	struct dst_entry *old_dst;
1800
1801	sk_tx_queue_clear(sk);
1802	sk->sk_dst_pending_confirm = 0;
1803	old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1804					    lockdep_sock_is_held(sk));
 
 
1805	rcu_assign_pointer(sk->sk_dst_cache, dst);
1806	dst_release(old_dst);
1807}
1808
1809static inline void
1810sk_dst_set(struct sock *sk, struct dst_entry *dst)
1811{
1812	struct dst_entry *old_dst;
1813
1814	sk_tx_queue_clear(sk);
1815	sk->sk_dst_pending_confirm = 0;
1816	old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1817	dst_release(old_dst);
1818}
1819
1820static inline void
1821__sk_dst_reset(struct sock *sk)
1822{
1823	__sk_dst_set(sk, NULL);
1824}
1825
1826static inline void
1827sk_dst_reset(struct sock *sk)
1828{
1829	sk_dst_set(sk, NULL);
 
 
1830}
1831
1832struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1833
1834struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1835
1836static inline void sk_dst_confirm(struct sock *sk)
1837{
1838	if (!sk->sk_dst_pending_confirm)
1839		sk->sk_dst_pending_confirm = 1;
1840}
1841
1842static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1843{
1844	if (skb_get_dst_pending_confirm(skb)) {
1845		struct sock *sk = skb->sk;
1846		unsigned long now = jiffies;
1847
1848		/* avoid dirtying neighbour */
1849		if (n->confirmed != now)
1850			n->confirmed = now;
1851		if (sk && sk->sk_dst_pending_confirm)
1852			sk->sk_dst_pending_confirm = 0;
1853	}
1854}
1855
1856bool sk_mc_loop(struct sock *sk);
1857
1858static inline bool sk_can_gso(const struct sock *sk)
1859{
1860	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1861}
1862
1863void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1864
1865static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1866{
1867	sk->sk_route_nocaps |= flags;
1868	sk->sk_route_caps &= ~flags;
1869}
1870
1871static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1872					   struct iov_iter *from, char *to,
1873					   int copy, int offset)
1874{
1875	if (skb->ip_summed == CHECKSUM_NONE) {
1876		__wsum csum = 0;
1877		if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1878			return -EFAULT;
 
1879		skb->csum = csum_block_add(skb->csum, csum, offset);
1880	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1881		if (!copy_from_iter_full_nocache(to, copy, from))
 
1882			return -EFAULT;
1883	} else if (!copy_from_iter_full(to, copy, from))
1884		return -EFAULT;
1885
1886	return 0;
1887}
1888
1889static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1890				       struct iov_iter *from, int copy)
1891{
1892	int err, offset = skb->len;
1893
1894	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1895				       copy, offset);
1896	if (err)
1897		__skb_trim(skb, offset);
1898
1899	return err;
1900}
1901
1902static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1903					   struct sk_buff *skb,
1904					   struct page *page,
1905					   int off, int copy)
1906{
1907	int err;
1908
1909	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1910				       copy, skb->len);
1911	if (err)
1912		return err;
1913
1914	skb->len	     += copy;
1915	skb->data_len	     += copy;
1916	skb->truesize	     += copy;
1917	sk->sk_wmem_queued   += copy;
1918	sk_mem_charge(sk, copy);
1919	return 0;
1920}
1921
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1922/**
1923 * sk_wmem_alloc_get - returns write allocations
1924 * @sk: socket
1925 *
1926 * Returns sk_wmem_alloc minus initial offset of one
1927 */
1928static inline int sk_wmem_alloc_get(const struct sock *sk)
1929{
1930	return refcount_read(&sk->sk_wmem_alloc) - 1;
1931}
1932
1933/**
1934 * sk_rmem_alloc_get - returns read allocations
1935 * @sk: socket
1936 *
1937 * Returns sk_rmem_alloc
1938 */
1939static inline int sk_rmem_alloc_get(const struct sock *sk)
1940{
1941	return atomic_read(&sk->sk_rmem_alloc);
1942}
1943
1944/**
1945 * sk_has_allocations - check if allocations are outstanding
1946 * @sk: socket
1947 *
1948 * Returns true if socket has write or read allocations
1949 */
1950static inline bool sk_has_allocations(const struct sock *sk)
1951{
1952	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1953}
1954
1955/**
1956 * skwq_has_sleeper - check if there are any waiting processes
1957 * @wq: struct socket_wq
1958 *
1959 * Returns true if socket_wq has waiting processes
1960 *
1961 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1962 * barrier call. They were added due to the race found within the tcp code.
1963 *
1964 * Consider following tcp code paths::
1965 *
1966 *   CPU1                CPU2
1967 *   sys_select          receive packet
 
1968 *   ...                 ...
1969 *   __add_wait_queue    update tp->rcv_nxt
1970 *   ...                 ...
1971 *   tp->rcv_nxt check   sock_def_readable
1972 *   ...                 {
1973 *   schedule               rcu_read_lock();
1974 *                          wq = rcu_dereference(sk->sk_wq);
1975 *                          if (wq && waitqueue_active(&wq->wait))
1976 *                              wake_up_interruptible(&wq->wait)
1977 *                          ...
1978 *                       }
1979 *
1980 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1981 * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1982 * could then endup calling schedule and sleep forever if there are no more
1983 * data on the socket.
1984 *
1985 */
1986static inline bool skwq_has_sleeper(struct socket_wq *wq)
1987{
1988	return wq && wq_has_sleeper(&wq->wait);
 
 
 
 
 
 
 
 
1989}
1990
1991/**
1992 * sock_poll_wait - place memory barrier behind the poll_wait call.
1993 * @filp:           file
1994 * @wait_address:   socket wait queue
1995 * @p:              poll_table
1996 *
1997 * See the comments in the wq_has_sleeper function.
1998 */
1999static inline void sock_poll_wait(struct file *filp,
2000		wait_queue_head_t *wait_address, poll_table *p)
2001{
2002	if (!poll_does_not_wait(p) && wait_address) {
2003		poll_wait(filp, wait_address, p);
2004		/* We need to be sure we are in sync with the
 
2005		 * socket flags modification.
2006		 *
2007		 * This memory barrier is paired in the wq_has_sleeper.
2008		 */
2009		smp_mb();
2010	}
2011}
2012
2013static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2014{
2015	if (sk->sk_txhash) {
2016		skb->l4_hash = 1;
2017		skb->hash = sk->sk_txhash;
2018	}
2019}
2020
2021void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2022
2023/*
2024 *	Queue a received datagram if it will fit. Stream and sequenced
2025 *	protocols can't normally use this as they need to fit buffers in
2026 *	and play with them.
2027 *
2028 *	Inlined as it's very short and called for pretty much every
2029 *	packet ever received.
2030 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2031static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2032{
2033	skb_orphan(skb);
2034	skb->sk = sk;
2035	skb->destructor = sock_rfree;
2036	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2037	sk_mem_charge(sk, skb->truesize);
2038}
2039
2040void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2041		    unsigned long expires);
2042
2043void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2044
2045int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2046			struct sk_buff *skb, unsigned int flags,
2047			void (*destructor)(struct sock *sk,
2048					   struct sk_buff *skb));
2049int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2050int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2051
2052int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2053struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2054
2055/*
2056 *	Recover an error report and clear atomically
2057 */
2058
2059static inline int sock_error(struct sock *sk)
2060{
2061	int err;
2062	if (likely(!sk->sk_err))
2063		return 0;
2064	err = xchg(&sk->sk_err, 0);
2065	return -err;
2066}
2067
2068static inline unsigned long sock_wspace(struct sock *sk)
2069{
2070	int amt = 0;
2071
2072	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2073		amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2074		if (amt < 0)
2075			amt = 0;
2076	}
2077	return amt;
2078}
2079
2080/* Note:
2081 *  We use sk->sk_wq_raw, from contexts knowing this
2082 *  pointer is not NULL and cannot disappear/change.
2083 */
2084static inline void sk_set_bit(int nr, struct sock *sk)
2085{
2086	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2087	    !sock_flag(sk, SOCK_FASYNC))
2088		return;
2089
2090	set_bit(nr, &sk->sk_wq_raw->flags);
2091}
2092
2093static inline void sk_clear_bit(int nr, struct sock *sk)
2094{
2095	if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2096	    !sock_flag(sk, SOCK_FASYNC))
2097		return;
2098
2099	clear_bit(nr, &sk->sk_wq_raw->flags);
2100}
2101
2102static inline void sk_wake_async(const struct sock *sk, int how, int band)
2103{
2104	if (sock_flag(sk, SOCK_FASYNC)) {
2105		rcu_read_lock();
2106		sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2107		rcu_read_unlock();
2108	}
2109}
2110
2111/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2112 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2113 * Note: for send buffers, TCP works better if we can build two skbs at
2114 * minimum.
2115 */
2116#define TCP_SKB_MIN_TRUESIZE	(2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2117
2118#define SOCK_MIN_SNDBUF		(TCP_SKB_MIN_TRUESIZE * 2)
2119#define SOCK_MIN_RCVBUF		 TCP_SKB_MIN_TRUESIZE
2120
2121static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2122{
2123	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2124		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2125		sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2126	}
2127}
2128
2129struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2130				    bool force_schedule);
2131
2132/**
2133 * sk_page_frag - return an appropriate page_frag
2134 * @sk: socket
2135 *
2136 * If socket allocation mode allows current thread to sleep, it means its
2137 * safe to use the per task page_frag instead of the per socket one.
2138 */
2139static inline struct page_frag *sk_page_frag(struct sock *sk)
2140{
2141	if (gfpflags_allow_blocking(sk->sk_allocation))
2142		return &current->task_frag;
2143
2144	return &sk->sk_frag;
 
 
 
 
 
2145}
2146
2147bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2148
2149int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2150		int sg_start, int *sg_curr, unsigned int *sg_size,
2151		int first_coalesce);
2152
2153/*
2154 *	Default write policy as shown to user space via poll/select/SIGIO
2155 */
2156static inline bool sock_writeable(const struct sock *sk)
2157{
2158	return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2159}
2160
2161static inline gfp_t gfp_any(void)
2162{
2163	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2164}
2165
2166static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2167{
2168	return noblock ? 0 : sk->sk_rcvtimeo;
2169}
2170
2171static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2172{
2173	return noblock ? 0 : sk->sk_sndtimeo;
2174}
2175
2176static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2177{
2178	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2179}
2180
2181/* Alas, with timeout socket operations are not restartable.
2182 * Compare this to poll().
2183 */
2184static inline int sock_intr_errno(long timeo)
2185{
2186	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2187}
2188
2189struct sock_skb_cb {
2190	u32 dropcount;
2191};
2192
2193/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2194 * using skb->cb[] would keep using it directly and utilize its
2195 * alignement guarantee.
2196 */
2197#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2198			    sizeof(struct sock_skb_cb)))
2199
2200#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2201			    SOCK_SKB_CB_OFFSET))
2202
2203#define sock_skb_cb_check_size(size) \
2204	BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2205
2206static inline void
2207sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2208{
2209	SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2210						atomic_read(&sk->sk_drops) : 0;
2211}
2212
2213static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2214{
2215	int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2216
2217	atomic_add(segs, &sk->sk_drops);
2218}
2219
2220void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2221			   struct sk_buff *skb);
2222void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2223			     struct sk_buff *skb);
2224
2225static inline void
2226sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2227{
2228	ktime_t kt = skb->tstamp;
2229	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2230
2231	/*
2232	 * generate control messages if
2233	 * - receive time stamping in software requested
 
2234	 * - software time stamp available and wanted
 
2235	 * - hardware time stamps available and wanted
 
 
2236	 */
2237	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2238	    (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2239	    (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2240	    (hwtstamps->hwtstamp &&
2241	     (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
 
 
2242		__sock_recv_timestamp(msg, sk, skb);
2243	else
2244		sk->sk_stamp = kt;
2245
2246	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2247		__sock_recv_wifi_status(msg, sk, skb);
2248}
2249
2250void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2251			      struct sk_buff *skb);
2252
2253#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2254static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2255					  struct sk_buff *skb)
2256{
2257#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2258			   (1UL << SOCK_RCVTSTAMP))
2259#define TSFLAGS_ANY	  (SOF_TIMESTAMPING_SOFTWARE			| \
2260			   SOF_TIMESTAMPING_RAW_HARDWARE)
 
 
2261
2262	if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2263		__sock_recv_ts_and_drops(msg, sk, skb);
2264	else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2265		sk->sk_stamp = skb->tstamp;
2266	else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2267		sk->sk_stamp = 0;
2268}
2269
2270void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2271
2272/**
2273 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2274 * @sk:		socket sending this packet
2275 * @tsflags:	timestamping flags to use
2276 * @tx_flags:	completed with instructions for time stamping
2277 *
2278 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
 
2279 */
2280static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2281				     __u8 *tx_flags)
2282{
2283	if (unlikely(tsflags))
2284		__sock_tx_timestamp(tsflags, tx_flags);
2285	if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2286		*tx_flags |= SKBTX_WIFI_STATUS;
2287}
2288
2289/**
2290 * sk_eat_skb - Release a skb if it is no longer needed
2291 * @sk: socket to eat this skb from
2292 * @skb: socket buffer to eat
 
2293 *
2294 * This routine must be called with interrupts disabled or with the socket
2295 * locked so that the sk_buff queue operation is ok.
2296*/
2297static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
 
 
 
 
 
 
 
 
 
 
2298{
2299	__skb_unlink(skb, &sk->sk_receive_queue);
2300	__kfree_skb(skb);
2301}
 
2302
2303static inline
2304struct net *sock_net(const struct sock *sk)
2305{
2306	return read_pnet(&sk->sk_net);
2307}
2308
2309static inline
2310void sock_net_set(struct sock *sk, struct net *net)
2311{
2312	write_pnet(&sk->sk_net, net);
2313}
2314
 
 
 
 
 
 
 
 
 
 
 
 
2315static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2316{
2317	if (skb->sk) {
2318		struct sock *sk = skb->sk;
2319
2320		skb->destructor = NULL;
2321		skb->sk = NULL;
2322		return sk;
2323	}
2324	return NULL;
2325}
2326
2327/* This helper checks if a socket is a full socket,
2328 * ie _not_ a timewait or request socket.
2329 */
2330static inline bool sk_fullsock(const struct sock *sk)
2331{
2332	return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2333}
 
 
 
2334
2335/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2336 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2337 */
2338static inline bool sk_listener(const struct sock *sk)
2339{
2340	return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2341}
2342
2343void sock_enable_timestamp(struct sock *sk, int flag);
2344int sock_get_timestamp(struct sock *, struct timeval __user *);
2345int sock_get_timestampns(struct sock *, struct timespec __user *);
2346int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2347		       int type);
2348
2349bool sk_ns_capable(const struct sock *sk,
2350		   struct user_namespace *user_ns, int cap);
2351bool sk_capable(const struct sock *sk, int cap);
2352bool sk_net_capable(const struct sock *sk, int cap);
2353
2354void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2355
2356/* Take into consideration the size of the struct sk_buff overhead in the
2357 * determination of these values, since that is non-constant across
2358 * platforms.  This makes socket queueing behavior and performance
2359 * not depend upon such differences.
2360 */
2361#define _SK_MEM_PACKETS		256
2362#define _SK_MEM_OVERHEAD	SKB_TRUESIZE(256)
2363#define SK_WMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2364#define SK_RMEM_MAX		(_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2365
2366extern __u32 sysctl_wmem_max;
2367extern __u32 sysctl_rmem_max;
2368
2369extern int sysctl_tstamp_allow_data;
 
2370extern int sysctl_optmem_max;
2371
2372extern __u32 sysctl_wmem_default;
2373extern __u32 sysctl_rmem_default;
2374
2375static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2376{
2377	/* Does this proto have per netns sysctl_wmem ? */
2378	if (proto->sysctl_wmem_offset)
2379		return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2380
2381	return *proto->sysctl_wmem;
2382}
2383
2384static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2385{
2386	/* Does this proto have per netns sysctl_rmem ? */
2387	if (proto->sysctl_rmem_offset)
2388		return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2389
2390	return *proto->sysctl_rmem;
2391}
2392
2393/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2394 * Some wifi drivers need to tweak it to get more chunks.
2395 * They can use this helper from their ndo_start_xmit()
2396 */
2397static inline void sk_pacing_shift_update(struct sock *sk, int val)
2398{
2399	if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2400		return;
2401	sk->sk_pacing_shift = val;
2402}
2403
2404/* if a socket is bound to a device, check that the given device
2405 * index is either the same or that the socket is bound to an L3
2406 * master device and the given device index is also enslaved to
2407 * that L3 master
2408 */
2409static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2410{
2411	int mdif;
2412
2413	if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2414		return true;
2415
2416	mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2417	if (mdif && mdif == sk->sk_bound_dev_if)
2418		return true;
2419
2420	return false;
2421}
2422
2423#endif	/* _SOCK_H */