xref: /openbmc/linux/include/net/sock.h (revision 33a03aad)
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/memcontrol.h>
58 #include <linux/res_counter.h>
59 #include <linux/static_key.h>
60 #include <linux/aio.h>
61 #include <linux/sched.h>
62 
63 #include <linux/filter.h>
64 #include <linux/rculist_nulls.h>
65 #include <linux/poll.h>
66 
67 #include <linux/atomic.h>
68 #include <net/dst.h>
69 #include <net/checksum.h>
70 
71 struct cgroup;
72 struct cgroup_subsys;
73 #ifdef CONFIG_NET
74 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
75 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
76 #else
77 static inline
78 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
79 {
80 	return 0;
81 }
82 static inline
83 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
84 {
85 }
86 #endif
87 /*
88  * This structure really needs to be cleaned up.
89  * Most of it is for TCP, and not used by any of
90  * the other protocols.
91  */
92 
93 /* Define this to get the SOCK_DBG debugging facility. */
94 #define SOCK_DEBUGGING
95 #ifdef SOCK_DEBUGGING
96 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
97 					printk(KERN_DEBUG msg); } while (0)
98 #else
99 /* Validate arguments and do nothing */
100 static inline __printf(2, 3)
101 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
102 {
103 }
104 #endif
105 
106 /* This is the per-socket lock.  The spinlock provides a synchronization
107  * between user contexts and software interrupt processing, whereas the
108  * mini-semaphore synchronizes multiple users amongst themselves.
109  */
110 typedef struct {
111 	spinlock_t		slock;
112 	int			owned;
113 	wait_queue_head_t	wq;
114 	/*
115 	 * We express the mutex-alike socket_lock semantics
116 	 * to the lock validator by explicitly managing
117 	 * the slock as a lock variant (in addition to
118 	 * the slock itself):
119 	 */
120 #ifdef CONFIG_DEBUG_LOCK_ALLOC
121 	struct lockdep_map dep_map;
122 #endif
123 } socket_lock_t;
124 
125 struct sock;
126 struct proto;
127 struct net;
128 
129 /**
130  *	struct sock_common - minimal network layer representation of sockets
131  *	@skc_daddr: Foreign IPv4 addr
132  *	@skc_rcv_saddr: Bound local IPv4 addr
133  *	@skc_hash: hash value used with various protocol lookup tables
134  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
135  *	@skc_family: network address family
136  *	@skc_state: Connection state
137  *	@skc_reuse: %SO_REUSEADDR setting
138  *	@skc_bound_dev_if: bound device index if != 0
139  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
140  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
141  *	@skc_prot: protocol handlers inside a network family
142  *	@skc_net: reference to the network namespace of this socket
143  *	@skc_node: main hash linkage for various protocol lookup tables
144  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
145  *	@skc_tx_queue_mapping: tx queue number for this connection
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  */
151 struct sock_common {
152 	/* skc_daddr and skc_rcv_saddr must be grouped :
153 	 * cf INET_MATCH() and INET_TW_MATCH()
154 	 */
155 	__be32			skc_daddr;
156 	__be32			skc_rcv_saddr;
157 
158 	union  {
159 		unsigned int	skc_hash;
160 		__u16		skc_u16hashes[2];
161 	};
162 	unsigned short		skc_family;
163 	volatile unsigned char	skc_state;
164 	unsigned char		skc_reuse;
165 	int			skc_bound_dev_if;
166 	union {
167 		struct hlist_node	skc_bind_node;
168 		struct hlist_nulls_node skc_portaddr_node;
169 	};
170 	struct proto		*skc_prot;
171 #ifdef CONFIG_NET_NS
172 	struct net	 	*skc_net;
173 #endif
174 	/*
175 	 * fields between dontcopy_begin/dontcopy_end
176 	 * are not copied in sock_copy()
177 	 */
178 	/* private: */
179 	int			skc_dontcopy_begin[0];
180 	/* public: */
181 	union {
182 		struct hlist_node	skc_node;
183 		struct hlist_nulls_node skc_nulls_node;
184 	};
185 	int			skc_tx_queue_mapping;
186 	atomic_t		skc_refcnt;
187 	/* private: */
188 	int                     skc_dontcopy_end[0];
189 	/* public: */
190 };
191 
192 struct cg_proto;
193 /**
194   *	struct sock - network layer representation of sockets
195   *	@__sk_common: shared layout with inet_timewait_sock
196   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
197   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
198   *	@sk_lock:	synchronizer
199   *	@sk_rcvbuf: size of receive buffer in bytes
200   *	@sk_wq: sock wait queue and async head
201   *	@sk_dst_cache: destination cache
202   *	@sk_dst_lock: destination cache lock
203   *	@sk_policy: flow policy
204   *	@sk_receive_queue: incoming packets
205   *	@sk_wmem_alloc: transmit queue bytes committed
206   *	@sk_write_queue: Packet sending queue
207   *	@sk_async_wait_queue: DMA copied packets
208   *	@sk_omem_alloc: "o" is "option" or "other"
209   *	@sk_wmem_queued: persistent queue size
210   *	@sk_forward_alloc: space allocated forward
211   *	@sk_allocation: allocation mode
212   *	@sk_sndbuf: size of send buffer in bytes
213   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
214   *		   %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
215   *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
216   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
217   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
218   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
219   *	@sk_gso_max_size: Maximum GSO segment size to build
220   *	@sk_lingertime: %SO_LINGER l_linger setting
221   *	@sk_backlog: always used with the per-socket spinlock held
222   *	@sk_callback_lock: used with the callbacks in the end of this struct
223   *	@sk_error_queue: rarely used
224   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
225   *			  IPV6_ADDRFORM for instance)
226   *	@sk_err: last error
227   *	@sk_err_soft: errors that don't cause failure but are the cause of a
228   *		      persistent failure not just 'timed out'
229   *	@sk_drops: raw/udp drops counter
230   *	@sk_ack_backlog: current listen backlog
231   *	@sk_max_ack_backlog: listen backlog set in listen()
232   *	@sk_priority: %SO_PRIORITY setting
233   *	@sk_cgrp_prioidx: socket group's priority map index
234   *	@sk_type: socket type (%SOCK_STREAM, etc)
235   *	@sk_protocol: which protocol this socket belongs in this network family
236   *	@sk_peer_pid: &struct pid for this socket's peer
237   *	@sk_peer_cred: %SO_PEERCRED setting
238   *	@sk_rcvlowat: %SO_RCVLOWAT setting
239   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
240   *	@sk_sndtimeo: %SO_SNDTIMEO setting
241   *	@sk_rxhash: flow hash received from netif layer
242   *	@sk_filter: socket filtering instructions
243   *	@sk_protinfo: private area, net family specific, when not using slab
244   *	@sk_timer: sock cleanup timer
245   *	@sk_stamp: time stamp of last packet received
246   *	@sk_socket: Identd and reporting IO signals
247   *	@sk_user_data: RPC layer private data
248   *	@sk_sndmsg_page: cached page for sendmsg
249   *	@sk_sndmsg_off: cached offset for sendmsg
250   *	@sk_peek_off: current peek_offset value
251   *	@sk_send_head: front of stuff to transmit
252   *	@sk_security: used by security modules
253   *	@sk_mark: generic packet mark
254   *	@sk_classid: this socket's cgroup classid
255   *	@sk_cgrp: this socket's cgroup-specific proto data
256   *	@sk_write_pending: a write to stream socket waits to start
257   *	@sk_state_change: callback to indicate change in the state of the sock
258   *	@sk_data_ready: callback to indicate there is data to be processed
259   *	@sk_write_space: callback to indicate there is bf sending space available
260   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
261   *	@sk_backlog_rcv: callback to process the backlog
262   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
263  */
264 struct sock {
265 	/*
266 	 * Now struct inet_timewait_sock also uses sock_common, so please just
267 	 * don't add nothing before this first member (__sk_common) --acme
268 	 */
269 	struct sock_common	__sk_common;
270 #define sk_node			__sk_common.skc_node
271 #define sk_nulls_node		__sk_common.skc_nulls_node
272 #define sk_refcnt		__sk_common.skc_refcnt
273 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
274 
275 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
276 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
277 #define sk_hash			__sk_common.skc_hash
278 #define sk_family		__sk_common.skc_family
279 #define sk_state		__sk_common.skc_state
280 #define sk_reuse		__sk_common.skc_reuse
281 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
282 #define sk_bind_node		__sk_common.skc_bind_node
283 #define sk_prot			__sk_common.skc_prot
284 #define sk_net			__sk_common.skc_net
285 	socket_lock_t		sk_lock;
286 	struct sk_buff_head	sk_receive_queue;
287 	/*
288 	 * The backlog queue is special, it is always used with
289 	 * the per-socket spinlock held and requires low latency
290 	 * access. Therefore we special case it's implementation.
291 	 * Note : rmem_alloc is in this structure to fill a hole
292 	 * on 64bit arches, not because its logically part of
293 	 * backlog.
294 	 */
295 	struct {
296 		atomic_t	rmem_alloc;
297 		int		len;
298 		struct sk_buff	*head;
299 		struct sk_buff	*tail;
300 	} sk_backlog;
301 #define sk_rmem_alloc sk_backlog.rmem_alloc
302 	int			sk_forward_alloc;
303 #ifdef CONFIG_RPS
304 	__u32			sk_rxhash;
305 #endif
306 	atomic_t		sk_drops;
307 	int			sk_rcvbuf;
308 
309 	struct sk_filter __rcu	*sk_filter;
310 	struct socket_wq __rcu	*sk_wq;
311 
312 #ifdef CONFIG_NET_DMA
313 	struct sk_buff_head	sk_async_wait_queue;
314 #endif
315 
316 #ifdef CONFIG_XFRM
317 	struct xfrm_policy	*sk_policy[2];
318 #endif
319 	unsigned long 		sk_flags;
320 	struct dst_entry	*sk_dst_cache;
321 	spinlock_t		sk_dst_lock;
322 	atomic_t		sk_wmem_alloc;
323 	atomic_t		sk_omem_alloc;
324 	int			sk_sndbuf;
325 	struct sk_buff_head	sk_write_queue;
326 	kmemcheck_bitfield_begin(flags);
327 	unsigned int		sk_shutdown  : 2,
328 				sk_no_check  : 2,
329 				sk_userlocks : 4,
330 				sk_protocol  : 8,
331 				sk_type      : 16;
332 	kmemcheck_bitfield_end(flags);
333 	int			sk_wmem_queued;
334 	gfp_t			sk_allocation;
335 	netdev_features_t	sk_route_caps;
336 	netdev_features_t	sk_route_nocaps;
337 	int			sk_gso_type;
338 	unsigned int		sk_gso_max_size;
339 	int			sk_rcvlowat;
340 	unsigned long	        sk_lingertime;
341 	struct sk_buff_head	sk_error_queue;
342 	struct proto		*sk_prot_creator;
343 	rwlock_t		sk_callback_lock;
344 	int			sk_err,
345 				sk_err_soft;
346 	unsigned short		sk_ack_backlog;
347 	unsigned short		sk_max_ack_backlog;
348 	__u32			sk_priority;
349 #ifdef CONFIG_CGROUPS
350 	__u32			sk_cgrp_prioidx;
351 #endif
352 	struct pid		*sk_peer_pid;
353 	const struct cred	*sk_peer_cred;
354 	long			sk_rcvtimeo;
355 	long			sk_sndtimeo;
356 	void			*sk_protinfo;
357 	struct timer_list	sk_timer;
358 	ktime_t			sk_stamp;
359 	struct socket		*sk_socket;
360 	void			*sk_user_data;
361 	struct page		*sk_sndmsg_page;
362 	struct sk_buff		*sk_send_head;
363 	__u32			sk_sndmsg_off;
364 	__s32			sk_peek_off;
365 	int			sk_write_pending;
366 #ifdef CONFIG_SECURITY
367 	void			*sk_security;
368 #endif
369 	__u32			sk_mark;
370 	u32			sk_classid;
371 	struct cg_proto		*sk_cgrp;
372 	void			(*sk_state_change)(struct sock *sk);
373 	void			(*sk_data_ready)(struct sock *sk, int bytes);
374 	void			(*sk_write_space)(struct sock *sk);
375 	void			(*sk_error_report)(struct sock *sk);
376 	int			(*sk_backlog_rcv)(struct sock *sk,
377 						  struct sk_buff *skb);
378 	void                    (*sk_destruct)(struct sock *sk);
379 };
380 
381 /*
382  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
383  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
384  * on a socket means that the socket will reuse everybody else's port
385  * without looking at the other's sk_reuse value.
386  */
387 
388 #define SK_NO_REUSE	0
389 #define SK_CAN_REUSE	1
390 #define SK_FORCE_REUSE	2
391 
392 static inline int sk_peek_offset(struct sock *sk, int flags)
393 {
394 	if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
395 		return sk->sk_peek_off;
396 	else
397 		return 0;
398 }
399 
400 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
401 {
402 	if (sk->sk_peek_off >= 0) {
403 		if (sk->sk_peek_off >= val)
404 			sk->sk_peek_off -= val;
405 		else
406 			sk->sk_peek_off = 0;
407 	}
408 }
409 
410 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
411 {
412 	if (sk->sk_peek_off >= 0)
413 		sk->sk_peek_off += val;
414 }
415 
416 /*
417  * Hashed lists helper routines
418  */
419 static inline struct sock *sk_entry(const struct hlist_node *node)
420 {
421 	return hlist_entry(node, struct sock, sk_node);
422 }
423 
424 static inline struct sock *__sk_head(const struct hlist_head *head)
425 {
426 	return hlist_entry(head->first, struct sock, sk_node);
427 }
428 
429 static inline struct sock *sk_head(const struct hlist_head *head)
430 {
431 	return hlist_empty(head) ? NULL : __sk_head(head);
432 }
433 
434 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
435 {
436 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
437 }
438 
439 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
440 {
441 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
442 }
443 
444 static inline struct sock *sk_next(const struct sock *sk)
445 {
446 	return sk->sk_node.next ?
447 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
448 }
449 
450 static inline struct sock *sk_nulls_next(const struct sock *sk)
451 {
452 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
453 		hlist_nulls_entry(sk->sk_nulls_node.next,
454 				  struct sock, sk_nulls_node) :
455 		NULL;
456 }
457 
458 static inline bool sk_unhashed(const struct sock *sk)
459 {
460 	return hlist_unhashed(&sk->sk_node);
461 }
462 
463 static inline bool sk_hashed(const struct sock *sk)
464 {
465 	return !sk_unhashed(sk);
466 }
467 
468 static inline void sk_node_init(struct hlist_node *node)
469 {
470 	node->pprev = NULL;
471 }
472 
473 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
474 {
475 	node->pprev = NULL;
476 }
477 
478 static inline void __sk_del_node(struct sock *sk)
479 {
480 	__hlist_del(&sk->sk_node);
481 }
482 
483 /* NB: equivalent to hlist_del_init_rcu */
484 static inline bool __sk_del_node_init(struct sock *sk)
485 {
486 	if (sk_hashed(sk)) {
487 		__sk_del_node(sk);
488 		sk_node_init(&sk->sk_node);
489 		return true;
490 	}
491 	return false;
492 }
493 
494 /* Grab socket reference count. This operation is valid only
495    when sk is ALREADY grabbed f.e. it is found in hash table
496    or a list and the lookup is made under lock preventing hash table
497    modifications.
498  */
499 
500 static inline void sock_hold(struct sock *sk)
501 {
502 	atomic_inc(&sk->sk_refcnt);
503 }
504 
505 /* Ungrab socket in the context, which assumes that socket refcnt
506    cannot hit zero, f.e. it is true in context of any socketcall.
507  */
508 static inline void __sock_put(struct sock *sk)
509 {
510 	atomic_dec(&sk->sk_refcnt);
511 }
512 
513 static inline bool sk_del_node_init(struct sock *sk)
514 {
515 	bool rc = __sk_del_node_init(sk);
516 
517 	if (rc) {
518 		/* paranoid for a while -acme */
519 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
520 		__sock_put(sk);
521 	}
522 	return rc;
523 }
524 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
525 
526 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
527 {
528 	if (sk_hashed(sk)) {
529 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
530 		return true;
531 	}
532 	return false;
533 }
534 
535 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
536 {
537 	bool rc = __sk_nulls_del_node_init_rcu(sk);
538 
539 	if (rc) {
540 		/* paranoid for a while -acme */
541 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
542 		__sock_put(sk);
543 	}
544 	return rc;
545 }
546 
547 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
548 {
549 	hlist_add_head(&sk->sk_node, list);
550 }
551 
552 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
553 {
554 	sock_hold(sk);
555 	__sk_add_node(sk, list);
556 }
557 
558 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
559 {
560 	sock_hold(sk);
561 	hlist_add_head_rcu(&sk->sk_node, list);
562 }
563 
564 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
565 {
566 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
567 }
568 
569 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
570 {
571 	sock_hold(sk);
572 	__sk_nulls_add_node_rcu(sk, list);
573 }
574 
575 static inline void __sk_del_bind_node(struct sock *sk)
576 {
577 	__hlist_del(&sk->sk_bind_node);
578 }
579 
580 static inline void sk_add_bind_node(struct sock *sk,
581 					struct hlist_head *list)
582 {
583 	hlist_add_head(&sk->sk_bind_node, list);
584 }
585 
586 #define sk_for_each(__sk, node, list) \
587 	hlist_for_each_entry(__sk, node, list, sk_node)
588 #define sk_for_each_rcu(__sk, node, list) \
589 	hlist_for_each_entry_rcu(__sk, node, list, sk_node)
590 #define sk_nulls_for_each(__sk, node, list) \
591 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
592 #define sk_nulls_for_each_rcu(__sk, node, list) \
593 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
594 #define sk_for_each_from(__sk, node) \
595 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
596 		hlist_for_each_entry_from(__sk, node, sk_node)
597 #define sk_nulls_for_each_from(__sk, node) \
598 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
599 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
600 #define sk_for_each_safe(__sk, node, tmp, list) \
601 	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
602 #define sk_for_each_bound(__sk, node, list) \
603 	hlist_for_each_entry(__sk, node, list, sk_bind_node)
604 
605 /* Sock flags */
606 enum sock_flags {
607 	SOCK_DEAD,
608 	SOCK_DONE,
609 	SOCK_URGINLINE,
610 	SOCK_KEEPOPEN,
611 	SOCK_LINGER,
612 	SOCK_DESTROY,
613 	SOCK_BROADCAST,
614 	SOCK_TIMESTAMP,
615 	SOCK_ZAPPED,
616 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
617 	SOCK_DBG, /* %SO_DEBUG setting */
618 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
619 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
620 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
621 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
622 	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
623 	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
624 	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
625 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
626 	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
627 	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
628 	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
629 	SOCK_FASYNC, /* fasync() active */
630 	SOCK_RXQ_OVFL,
631 	SOCK_ZEROCOPY, /* buffers from userspace */
632 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
633 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
634 		     * Will use last 4 bytes of packet sent from
635 		     * user-space instead.
636 		     */
637 };
638 
639 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
640 {
641 	nsk->sk_flags = osk->sk_flags;
642 }
643 
644 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
645 {
646 	__set_bit(flag, &sk->sk_flags);
647 }
648 
649 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
650 {
651 	__clear_bit(flag, &sk->sk_flags);
652 }
653 
654 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
655 {
656 	return test_bit(flag, &sk->sk_flags);
657 }
658 
659 static inline void sk_acceptq_removed(struct sock *sk)
660 {
661 	sk->sk_ack_backlog--;
662 }
663 
664 static inline void sk_acceptq_added(struct sock *sk)
665 {
666 	sk->sk_ack_backlog++;
667 }
668 
669 static inline bool sk_acceptq_is_full(const struct sock *sk)
670 {
671 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
672 }
673 
674 /*
675  * Compute minimal free write space needed to queue new packets.
676  */
677 static inline int sk_stream_min_wspace(const struct sock *sk)
678 {
679 	return sk->sk_wmem_queued >> 1;
680 }
681 
682 static inline int sk_stream_wspace(const struct sock *sk)
683 {
684 	return sk->sk_sndbuf - sk->sk_wmem_queued;
685 }
686 
687 extern void sk_stream_write_space(struct sock *sk);
688 
689 static inline bool sk_stream_memory_free(const struct sock *sk)
690 {
691 	return sk->sk_wmem_queued < sk->sk_sndbuf;
692 }
693 
694 /* OOB backlog add */
695 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
696 {
697 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
698 	skb_dst_force(skb);
699 
700 	if (!sk->sk_backlog.tail)
701 		sk->sk_backlog.head = skb;
702 	else
703 		sk->sk_backlog.tail->next = skb;
704 
705 	sk->sk_backlog.tail = skb;
706 	skb->next = NULL;
707 }
708 
709 /*
710  * Take into account size of receive queue and backlog queue
711  * Do not take into account this skb truesize,
712  * to allow even a single big packet to come.
713  */
714 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
715 				     unsigned int limit)
716 {
717 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
718 
719 	return qsize > limit;
720 }
721 
722 /* The per-socket spinlock must be held here. */
723 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
724 					      unsigned int limit)
725 {
726 	if (sk_rcvqueues_full(sk, skb, limit))
727 		return -ENOBUFS;
728 
729 	__sk_add_backlog(sk, skb);
730 	sk->sk_backlog.len += skb->truesize;
731 	return 0;
732 }
733 
734 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
735 {
736 	return sk->sk_backlog_rcv(sk, skb);
737 }
738 
739 static inline void sock_rps_record_flow(const struct sock *sk)
740 {
741 #ifdef CONFIG_RPS
742 	struct rps_sock_flow_table *sock_flow_table;
743 
744 	rcu_read_lock();
745 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
746 	rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
747 	rcu_read_unlock();
748 #endif
749 }
750 
751 static inline void sock_rps_reset_flow(const struct sock *sk)
752 {
753 #ifdef CONFIG_RPS
754 	struct rps_sock_flow_table *sock_flow_table;
755 
756 	rcu_read_lock();
757 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
758 	rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
759 	rcu_read_unlock();
760 #endif
761 }
762 
763 static inline void sock_rps_save_rxhash(struct sock *sk,
764 					const struct sk_buff *skb)
765 {
766 #ifdef CONFIG_RPS
767 	if (unlikely(sk->sk_rxhash != skb->rxhash)) {
768 		sock_rps_reset_flow(sk);
769 		sk->sk_rxhash = skb->rxhash;
770 	}
771 #endif
772 }
773 
774 static inline void sock_rps_reset_rxhash(struct sock *sk)
775 {
776 #ifdef CONFIG_RPS
777 	sock_rps_reset_flow(sk);
778 	sk->sk_rxhash = 0;
779 #endif
780 }
781 
782 #define sk_wait_event(__sk, __timeo, __condition)			\
783 	({	int __rc;						\
784 		release_sock(__sk);					\
785 		__rc = __condition;					\
786 		if (!__rc) {						\
787 			*(__timeo) = schedule_timeout(*(__timeo));	\
788 		}							\
789 		lock_sock(__sk);					\
790 		__rc = __condition;					\
791 		__rc;							\
792 	})
793 
794 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
795 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
796 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
797 extern int sk_stream_error(struct sock *sk, int flags, int err);
798 extern void sk_stream_kill_queues(struct sock *sk);
799 
800 extern int sk_wait_data(struct sock *sk, long *timeo);
801 
802 struct request_sock_ops;
803 struct timewait_sock_ops;
804 struct inet_hashinfo;
805 struct raw_hashinfo;
806 struct module;
807 
808 /* Networking protocol blocks we attach to sockets.
809  * socket layer -> transport layer interface
810  * transport -> network interface is defined by struct inet_proto
811  */
812 struct proto {
813 	void			(*close)(struct sock *sk,
814 					long timeout);
815 	int			(*connect)(struct sock *sk,
816 					struct sockaddr *uaddr,
817 					int addr_len);
818 	int			(*disconnect)(struct sock *sk, int flags);
819 
820 	struct sock *		(*accept)(struct sock *sk, int flags, int *err);
821 
822 	int			(*ioctl)(struct sock *sk, int cmd,
823 					 unsigned long arg);
824 	int			(*init)(struct sock *sk);
825 	void			(*destroy)(struct sock *sk);
826 	void			(*shutdown)(struct sock *sk, int how);
827 	int			(*setsockopt)(struct sock *sk, int level,
828 					int optname, char __user *optval,
829 					unsigned int optlen);
830 	int			(*getsockopt)(struct sock *sk, int level,
831 					int optname, char __user *optval,
832 					int __user *option);
833 #ifdef CONFIG_COMPAT
834 	int			(*compat_setsockopt)(struct sock *sk,
835 					int level,
836 					int optname, char __user *optval,
837 					unsigned int optlen);
838 	int			(*compat_getsockopt)(struct sock *sk,
839 					int level,
840 					int optname, char __user *optval,
841 					int __user *option);
842 	int			(*compat_ioctl)(struct sock *sk,
843 					unsigned int cmd, unsigned long arg);
844 #endif
845 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
846 					   struct msghdr *msg, size_t len);
847 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
848 					   struct msghdr *msg,
849 					   size_t len, int noblock, int flags,
850 					   int *addr_len);
851 	int			(*sendpage)(struct sock *sk, struct page *page,
852 					int offset, size_t size, int flags);
853 	int			(*bind)(struct sock *sk,
854 					struct sockaddr *uaddr, int addr_len);
855 
856 	int			(*backlog_rcv) (struct sock *sk,
857 						struct sk_buff *skb);
858 
859 	/* Keeping track of sk's, looking them up, and port selection methods. */
860 	void			(*hash)(struct sock *sk);
861 	void			(*unhash)(struct sock *sk);
862 	void			(*rehash)(struct sock *sk);
863 	int			(*get_port)(struct sock *sk, unsigned short snum);
864 	void			(*clear_sk)(struct sock *sk, int size);
865 
866 	/* Keeping track of sockets in use */
867 #ifdef CONFIG_PROC_FS
868 	unsigned int		inuse_idx;
869 #endif
870 
871 	/* Memory pressure */
872 	void			(*enter_memory_pressure)(struct sock *sk);
873 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
874 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
875 	/*
876 	 * Pressure flag: try to collapse.
877 	 * Technical note: it is used by multiple contexts non atomically.
878 	 * All the __sk_mem_schedule() is of this nature: accounting
879 	 * is strict, actions are advisory and have some latency.
880 	 */
881 	int			*memory_pressure;
882 	long			*sysctl_mem;
883 	int			*sysctl_wmem;
884 	int			*sysctl_rmem;
885 	int			max_header;
886 	bool			no_autobind;
887 
888 	struct kmem_cache	*slab;
889 	unsigned int		obj_size;
890 	int			slab_flags;
891 
892 	struct percpu_counter	*orphan_count;
893 
894 	struct request_sock_ops	*rsk_prot;
895 	struct timewait_sock_ops *twsk_prot;
896 
897 	union {
898 		struct inet_hashinfo	*hashinfo;
899 		struct udp_table	*udp_table;
900 		struct raw_hashinfo	*raw_hash;
901 	} h;
902 
903 	struct module		*owner;
904 
905 	char			name[32];
906 
907 	struct list_head	node;
908 #ifdef SOCK_REFCNT_DEBUG
909 	atomic_t		socks;
910 #endif
911 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
912 	/*
913 	 * cgroup specific init/deinit functions. Called once for all
914 	 * protocols that implement it, from cgroups populate function.
915 	 * This function has to setup any files the protocol want to
916 	 * appear in the kmem cgroup filesystem.
917 	 */
918 	int			(*init_cgroup)(struct mem_cgroup *memcg,
919 					       struct cgroup_subsys *ss);
920 	void			(*destroy_cgroup)(struct mem_cgroup *memcg);
921 	struct cg_proto		*(*proto_cgroup)(struct mem_cgroup *memcg);
922 #endif
923 };
924 
925 /*
926  * Bits in struct cg_proto.flags
927  */
928 enum cg_proto_flags {
929 	/* Currently active and new sockets should be assigned to cgroups */
930 	MEMCG_SOCK_ACTIVE,
931 	/* It was ever activated; we must disarm static keys on destruction */
932 	MEMCG_SOCK_ACTIVATED,
933 };
934 
935 struct cg_proto {
936 	void			(*enter_memory_pressure)(struct sock *sk);
937 	struct res_counter	*memory_allocated;	/* Current allocated memory. */
938 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
939 	int			*memory_pressure;
940 	long			*sysctl_mem;
941 	unsigned long		flags;
942 	/*
943 	 * memcg field is used to find which memcg we belong directly
944 	 * Each memcg struct can hold more than one cg_proto, so container_of
945 	 * won't really cut.
946 	 *
947 	 * The elegant solution would be having an inverse function to
948 	 * proto_cgroup in struct proto, but that means polluting the structure
949 	 * for everybody, instead of just for memcg users.
950 	 */
951 	struct mem_cgroup	*memcg;
952 };
953 
954 extern int proto_register(struct proto *prot, int alloc_slab);
955 extern void proto_unregister(struct proto *prot);
956 
957 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
958 {
959 	return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
960 }
961 
962 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
963 {
964 	return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
965 }
966 
967 #ifdef SOCK_REFCNT_DEBUG
968 static inline void sk_refcnt_debug_inc(struct sock *sk)
969 {
970 	atomic_inc(&sk->sk_prot->socks);
971 }
972 
973 static inline void sk_refcnt_debug_dec(struct sock *sk)
974 {
975 	atomic_dec(&sk->sk_prot->socks);
976 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
977 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
978 }
979 
980 inline void sk_refcnt_debug_release(const struct sock *sk)
981 {
982 	if (atomic_read(&sk->sk_refcnt) != 1)
983 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
984 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
985 }
986 #else /* SOCK_REFCNT_DEBUG */
987 #define sk_refcnt_debug_inc(sk) do { } while (0)
988 #define sk_refcnt_debug_dec(sk) do { } while (0)
989 #define sk_refcnt_debug_release(sk) do { } while (0)
990 #endif /* SOCK_REFCNT_DEBUG */
991 
992 #if defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM) && defined(CONFIG_NET)
993 extern struct static_key memcg_socket_limit_enabled;
994 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
995 					       struct cg_proto *cg_proto)
996 {
997 	return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
998 }
999 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1000 #else
1001 #define mem_cgroup_sockets_enabled 0
1002 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1003 					       struct cg_proto *cg_proto)
1004 {
1005 	return NULL;
1006 }
1007 #endif
1008 
1009 
1010 static inline bool sk_has_memory_pressure(const struct sock *sk)
1011 {
1012 	return sk->sk_prot->memory_pressure != NULL;
1013 }
1014 
1015 static inline bool sk_under_memory_pressure(const struct sock *sk)
1016 {
1017 	if (!sk->sk_prot->memory_pressure)
1018 		return false;
1019 
1020 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1021 		return !!*sk->sk_cgrp->memory_pressure;
1022 
1023 	return !!*sk->sk_prot->memory_pressure;
1024 }
1025 
1026 static inline void sk_leave_memory_pressure(struct sock *sk)
1027 {
1028 	int *memory_pressure = sk->sk_prot->memory_pressure;
1029 
1030 	if (!memory_pressure)
1031 		return;
1032 
1033 	if (*memory_pressure)
1034 		*memory_pressure = 0;
1035 
1036 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1037 		struct cg_proto *cg_proto = sk->sk_cgrp;
1038 		struct proto *prot = sk->sk_prot;
1039 
1040 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1041 			if (*cg_proto->memory_pressure)
1042 				*cg_proto->memory_pressure = 0;
1043 	}
1044 
1045 }
1046 
1047 static inline void sk_enter_memory_pressure(struct sock *sk)
1048 {
1049 	if (!sk->sk_prot->enter_memory_pressure)
1050 		return;
1051 
1052 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1053 		struct cg_proto *cg_proto = sk->sk_cgrp;
1054 		struct proto *prot = sk->sk_prot;
1055 
1056 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1057 			cg_proto->enter_memory_pressure(sk);
1058 	}
1059 
1060 	sk->sk_prot->enter_memory_pressure(sk);
1061 }
1062 
1063 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1064 {
1065 	long *prot = sk->sk_prot->sysctl_mem;
1066 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1067 		prot = sk->sk_cgrp->sysctl_mem;
1068 	return prot[index];
1069 }
1070 
1071 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1072 					      unsigned long amt,
1073 					      int *parent_status)
1074 {
1075 	struct res_counter *fail;
1076 	int ret;
1077 
1078 	ret = res_counter_charge_nofail(prot->memory_allocated,
1079 					amt << PAGE_SHIFT, &fail);
1080 	if (ret < 0)
1081 		*parent_status = OVER_LIMIT;
1082 }
1083 
1084 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1085 					      unsigned long amt)
1086 {
1087 	res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1088 }
1089 
1090 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1091 {
1092 	u64 ret;
1093 	ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1094 	return ret >> PAGE_SHIFT;
1095 }
1096 
1097 static inline long
1098 sk_memory_allocated(const struct sock *sk)
1099 {
1100 	struct proto *prot = sk->sk_prot;
1101 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1102 		return memcg_memory_allocated_read(sk->sk_cgrp);
1103 
1104 	return atomic_long_read(prot->memory_allocated);
1105 }
1106 
1107 static inline long
1108 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1109 {
1110 	struct proto *prot = sk->sk_prot;
1111 
1112 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1113 		memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1114 		/* update the root cgroup regardless */
1115 		atomic_long_add_return(amt, prot->memory_allocated);
1116 		return memcg_memory_allocated_read(sk->sk_cgrp);
1117 	}
1118 
1119 	return atomic_long_add_return(amt, prot->memory_allocated);
1120 }
1121 
1122 static inline void
1123 sk_memory_allocated_sub(struct sock *sk, int amt)
1124 {
1125 	struct proto *prot = sk->sk_prot;
1126 
1127 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1128 		memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1129 
1130 	atomic_long_sub(amt, prot->memory_allocated);
1131 }
1132 
1133 static inline void sk_sockets_allocated_dec(struct sock *sk)
1134 {
1135 	struct proto *prot = sk->sk_prot;
1136 
1137 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1138 		struct cg_proto *cg_proto = sk->sk_cgrp;
1139 
1140 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1141 			percpu_counter_dec(cg_proto->sockets_allocated);
1142 	}
1143 
1144 	percpu_counter_dec(prot->sockets_allocated);
1145 }
1146 
1147 static inline void sk_sockets_allocated_inc(struct sock *sk)
1148 {
1149 	struct proto *prot = sk->sk_prot;
1150 
1151 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1152 		struct cg_proto *cg_proto = sk->sk_cgrp;
1153 
1154 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1155 			percpu_counter_inc(cg_proto->sockets_allocated);
1156 	}
1157 
1158 	percpu_counter_inc(prot->sockets_allocated);
1159 }
1160 
1161 static inline int
1162 sk_sockets_allocated_read_positive(struct sock *sk)
1163 {
1164 	struct proto *prot = sk->sk_prot;
1165 
1166 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1167 		return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1168 
1169 	return percpu_counter_read_positive(prot->sockets_allocated);
1170 }
1171 
1172 static inline int
1173 proto_sockets_allocated_sum_positive(struct proto *prot)
1174 {
1175 	return percpu_counter_sum_positive(prot->sockets_allocated);
1176 }
1177 
1178 static inline long
1179 proto_memory_allocated(struct proto *prot)
1180 {
1181 	return atomic_long_read(prot->memory_allocated);
1182 }
1183 
1184 static inline bool
1185 proto_memory_pressure(struct proto *prot)
1186 {
1187 	if (!prot->memory_pressure)
1188 		return false;
1189 	return !!*prot->memory_pressure;
1190 }
1191 
1192 
1193 #ifdef CONFIG_PROC_FS
1194 /* Called with local bh disabled */
1195 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1196 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1197 #else
1198 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1199 		int inc)
1200 {
1201 }
1202 #endif
1203 
1204 
1205 /* With per-bucket locks this operation is not-atomic, so that
1206  * this version is not worse.
1207  */
1208 static inline void __sk_prot_rehash(struct sock *sk)
1209 {
1210 	sk->sk_prot->unhash(sk);
1211 	sk->sk_prot->hash(sk);
1212 }
1213 
1214 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1215 
1216 /* About 10 seconds */
1217 #define SOCK_DESTROY_TIME (10*HZ)
1218 
1219 /* Sockets 0-1023 can't be bound to unless you are superuser */
1220 #define PROT_SOCK	1024
1221 
1222 #define SHUTDOWN_MASK	3
1223 #define RCV_SHUTDOWN	1
1224 #define SEND_SHUTDOWN	2
1225 
1226 #define SOCK_SNDBUF_LOCK	1
1227 #define SOCK_RCVBUF_LOCK	2
1228 #define SOCK_BINDADDR_LOCK	4
1229 #define SOCK_BINDPORT_LOCK	8
1230 
1231 /* sock_iocb: used to kick off async processing of socket ios */
1232 struct sock_iocb {
1233 	struct list_head	list;
1234 
1235 	int			flags;
1236 	int			size;
1237 	struct socket		*sock;
1238 	struct sock		*sk;
1239 	struct scm_cookie	*scm;
1240 	struct msghdr		*msg, async_msg;
1241 	struct kiocb		*kiocb;
1242 };
1243 
1244 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1245 {
1246 	return (struct sock_iocb *)iocb->private;
1247 }
1248 
1249 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1250 {
1251 	return si->kiocb;
1252 }
1253 
1254 struct socket_alloc {
1255 	struct socket socket;
1256 	struct inode vfs_inode;
1257 };
1258 
1259 static inline struct socket *SOCKET_I(struct inode *inode)
1260 {
1261 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1262 }
1263 
1264 static inline struct inode *SOCK_INODE(struct socket *socket)
1265 {
1266 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1267 }
1268 
1269 /*
1270  * Functions for memory accounting
1271  */
1272 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1273 extern void __sk_mem_reclaim(struct sock *sk);
1274 
1275 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1276 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1277 #define SK_MEM_SEND	0
1278 #define SK_MEM_RECV	1
1279 
1280 static inline int sk_mem_pages(int amt)
1281 {
1282 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1283 }
1284 
1285 static inline bool sk_has_account(struct sock *sk)
1286 {
1287 	/* return true if protocol supports memory accounting */
1288 	return !!sk->sk_prot->memory_allocated;
1289 }
1290 
1291 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1292 {
1293 	if (!sk_has_account(sk))
1294 		return true;
1295 	return size <= sk->sk_forward_alloc ||
1296 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1297 }
1298 
1299 static inline bool sk_rmem_schedule(struct sock *sk, int size)
1300 {
1301 	if (!sk_has_account(sk))
1302 		return true;
1303 	return size <= sk->sk_forward_alloc ||
1304 		__sk_mem_schedule(sk, size, SK_MEM_RECV);
1305 }
1306 
1307 static inline void sk_mem_reclaim(struct sock *sk)
1308 {
1309 	if (!sk_has_account(sk))
1310 		return;
1311 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1312 		__sk_mem_reclaim(sk);
1313 }
1314 
1315 static inline void sk_mem_reclaim_partial(struct sock *sk)
1316 {
1317 	if (!sk_has_account(sk))
1318 		return;
1319 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1320 		__sk_mem_reclaim(sk);
1321 }
1322 
1323 static inline void sk_mem_charge(struct sock *sk, int size)
1324 {
1325 	if (!sk_has_account(sk))
1326 		return;
1327 	sk->sk_forward_alloc -= size;
1328 }
1329 
1330 static inline void sk_mem_uncharge(struct sock *sk, int size)
1331 {
1332 	if (!sk_has_account(sk))
1333 		return;
1334 	sk->sk_forward_alloc += size;
1335 }
1336 
1337 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1338 {
1339 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1340 	sk->sk_wmem_queued -= skb->truesize;
1341 	sk_mem_uncharge(sk, skb->truesize);
1342 	__kfree_skb(skb);
1343 }
1344 
1345 /* Used by processes to "lock" a socket state, so that
1346  * interrupts and bottom half handlers won't change it
1347  * from under us. It essentially blocks any incoming
1348  * packets, so that we won't get any new data or any
1349  * packets that change the state of the socket.
1350  *
1351  * While locked, BH processing will add new packets to
1352  * the backlog queue.  This queue is processed by the
1353  * owner of the socket lock right before it is released.
1354  *
1355  * Since ~2.3.5 it is also exclusive sleep lock serializing
1356  * accesses from user process context.
1357  */
1358 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1359 
1360 /*
1361  * Macro so as to not evaluate some arguments when
1362  * lockdep is not enabled.
1363  *
1364  * Mark both the sk_lock and the sk_lock.slock as a
1365  * per-address-family lock class.
1366  */
1367 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1368 do {									\
1369 	sk->sk_lock.owned = 0;						\
1370 	init_waitqueue_head(&sk->sk_lock.wq);				\
1371 	spin_lock_init(&(sk)->sk_lock.slock);				\
1372 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1373 			sizeof((sk)->sk_lock));				\
1374 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1375 				(skey), (sname));				\
1376 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1377 } while (0)
1378 
1379 extern void lock_sock_nested(struct sock *sk, int subclass);
1380 
1381 static inline void lock_sock(struct sock *sk)
1382 {
1383 	lock_sock_nested(sk, 0);
1384 }
1385 
1386 extern void release_sock(struct sock *sk);
1387 
1388 /* BH context may only use the following locking interface. */
1389 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1390 #define bh_lock_sock_nested(__sk) \
1391 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1392 				SINGLE_DEPTH_NESTING)
1393 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1394 
1395 extern bool lock_sock_fast(struct sock *sk);
1396 /**
1397  * unlock_sock_fast - complement of lock_sock_fast
1398  * @sk: socket
1399  * @slow: slow mode
1400  *
1401  * fast unlock socket for user context.
1402  * If slow mode is on, we call regular release_sock()
1403  */
1404 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1405 {
1406 	if (slow)
1407 		release_sock(sk);
1408 	else
1409 		spin_unlock_bh(&sk->sk_lock.slock);
1410 }
1411 
1412 
1413 extern struct sock		*sk_alloc(struct net *net, int family,
1414 					  gfp_t priority,
1415 					  struct proto *prot);
1416 extern void			sk_free(struct sock *sk);
1417 extern void			sk_release_kernel(struct sock *sk);
1418 extern struct sock		*sk_clone_lock(const struct sock *sk,
1419 					       const gfp_t priority);
1420 
1421 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
1422 					      unsigned long size, int force,
1423 					      gfp_t priority);
1424 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
1425 					      unsigned long size, int force,
1426 					      gfp_t priority);
1427 extern void			sock_wfree(struct sk_buff *skb);
1428 extern void			sock_rfree(struct sk_buff *skb);
1429 
1430 extern int			sock_setsockopt(struct socket *sock, int level,
1431 						int op, char __user *optval,
1432 						unsigned int optlen);
1433 
1434 extern int			sock_getsockopt(struct socket *sock, int level,
1435 						int op, char __user *optval,
1436 						int __user *optlen);
1437 extern struct sk_buff		*sock_alloc_send_skb(struct sock *sk,
1438 						     unsigned long size,
1439 						     int noblock,
1440 						     int *errcode);
1441 extern struct sk_buff		*sock_alloc_send_pskb(struct sock *sk,
1442 						      unsigned long header_len,
1443 						      unsigned long data_len,
1444 						      int noblock,
1445 						      int *errcode);
1446 extern void *sock_kmalloc(struct sock *sk, int size,
1447 			  gfp_t priority);
1448 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1449 extern void sk_send_sigurg(struct sock *sk);
1450 
1451 #ifdef CONFIG_CGROUPS
1452 extern void sock_update_classid(struct sock *sk);
1453 #else
1454 static inline void sock_update_classid(struct sock *sk)
1455 {
1456 }
1457 #endif
1458 
1459 /*
1460  * Functions to fill in entries in struct proto_ops when a protocol
1461  * does not implement a particular function.
1462  */
1463 extern int                      sock_no_bind(struct socket *,
1464 					     struct sockaddr *, int);
1465 extern int                      sock_no_connect(struct socket *,
1466 						struct sockaddr *, int, int);
1467 extern int                      sock_no_socketpair(struct socket *,
1468 						   struct socket *);
1469 extern int                      sock_no_accept(struct socket *,
1470 					       struct socket *, int);
1471 extern int                      sock_no_getname(struct socket *,
1472 						struct sockaddr *, int *, int);
1473 extern unsigned int             sock_no_poll(struct file *, struct socket *,
1474 					     struct poll_table_struct *);
1475 extern int                      sock_no_ioctl(struct socket *, unsigned int,
1476 					      unsigned long);
1477 extern int			sock_no_listen(struct socket *, int);
1478 extern int                      sock_no_shutdown(struct socket *, int);
1479 extern int			sock_no_getsockopt(struct socket *, int , int,
1480 						   char __user *, int __user *);
1481 extern int			sock_no_setsockopt(struct socket *, int, int,
1482 						   char __user *, unsigned int);
1483 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1484 						struct msghdr *, size_t);
1485 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1486 						struct msghdr *, size_t, int);
1487 extern int			sock_no_mmap(struct file *file,
1488 					     struct socket *sock,
1489 					     struct vm_area_struct *vma);
1490 extern ssize_t			sock_no_sendpage(struct socket *sock,
1491 						struct page *page,
1492 						int offset, size_t size,
1493 						int flags);
1494 
1495 /*
1496  * Functions to fill in entries in struct proto_ops when a protocol
1497  * uses the inet style.
1498  */
1499 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1500 				  char __user *optval, int __user *optlen);
1501 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1502 			       struct msghdr *msg, size_t size, int flags);
1503 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1504 				  char __user *optval, unsigned int optlen);
1505 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1506 		int optname, char __user *optval, int __user *optlen);
1507 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1508 		int optname, char __user *optval, unsigned int optlen);
1509 
1510 extern void sk_common_release(struct sock *sk);
1511 
1512 /*
1513  *	Default socket callbacks and setup code
1514  */
1515 
1516 /* Initialise core socket variables */
1517 extern void sock_init_data(struct socket *sock, struct sock *sk);
1518 
1519 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1520 
1521 /**
1522  *	sk_filter_release - release a socket filter
1523  *	@fp: filter to remove
1524  *
1525  *	Remove a filter from a socket and release its resources.
1526  */
1527 
1528 static inline void sk_filter_release(struct sk_filter *fp)
1529 {
1530 	if (atomic_dec_and_test(&fp->refcnt))
1531 		call_rcu(&fp->rcu, sk_filter_release_rcu);
1532 }
1533 
1534 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1535 {
1536 	unsigned int size = sk_filter_len(fp);
1537 
1538 	atomic_sub(size, &sk->sk_omem_alloc);
1539 	sk_filter_release(fp);
1540 }
1541 
1542 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1543 {
1544 	atomic_inc(&fp->refcnt);
1545 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1546 }
1547 
1548 /*
1549  * Socket reference counting postulates.
1550  *
1551  * * Each user of socket SHOULD hold a reference count.
1552  * * Each access point to socket (an hash table bucket, reference from a list,
1553  *   running timer, skb in flight MUST hold a reference count.
1554  * * When reference count hits 0, it means it will never increase back.
1555  * * When reference count hits 0, it means that no references from
1556  *   outside exist to this socket and current process on current CPU
1557  *   is last user and may/should destroy this socket.
1558  * * sk_free is called from any context: process, BH, IRQ. When
1559  *   it is called, socket has no references from outside -> sk_free
1560  *   may release descendant resources allocated by the socket, but
1561  *   to the time when it is called, socket is NOT referenced by any
1562  *   hash tables, lists etc.
1563  * * Packets, delivered from outside (from network or from another process)
1564  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1565  *   when they sit in queue. Otherwise, packets will leak to hole, when
1566  *   socket is looked up by one cpu and unhasing is made by another CPU.
1567  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1568  *   (leak to backlog). Packet socket does all the processing inside
1569  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1570  *   use separate SMP lock, so that they are prone too.
1571  */
1572 
1573 /* Ungrab socket and destroy it, if it was the last reference. */
1574 static inline void sock_put(struct sock *sk)
1575 {
1576 	if (atomic_dec_and_test(&sk->sk_refcnt))
1577 		sk_free(sk);
1578 }
1579 
1580 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1581 			  const int nested);
1582 
1583 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1584 {
1585 	sk->sk_tx_queue_mapping = tx_queue;
1586 }
1587 
1588 static inline void sk_tx_queue_clear(struct sock *sk)
1589 {
1590 	sk->sk_tx_queue_mapping = -1;
1591 }
1592 
1593 static inline int sk_tx_queue_get(const struct sock *sk)
1594 {
1595 	return sk ? sk->sk_tx_queue_mapping : -1;
1596 }
1597 
1598 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1599 {
1600 	sk_tx_queue_clear(sk);
1601 	sk->sk_socket = sock;
1602 }
1603 
1604 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1605 {
1606 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1607 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1608 }
1609 /* Detach socket from process context.
1610  * Announce socket dead, detach it from wait queue and inode.
1611  * Note that parent inode held reference count on this struct sock,
1612  * we do not release it in this function, because protocol
1613  * probably wants some additional cleanups or even continuing
1614  * to work with this socket (TCP).
1615  */
1616 static inline void sock_orphan(struct sock *sk)
1617 {
1618 	write_lock_bh(&sk->sk_callback_lock);
1619 	sock_set_flag(sk, SOCK_DEAD);
1620 	sk_set_socket(sk, NULL);
1621 	sk->sk_wq  = NULL;
1622 	write_unlock_bh(&sk->sk_callback_lock);
1623 }
1624 
1625 static inline void sock_graft(struct sock *sk, struct socket *parent)
1626 {
1627 	write_lock_bh(&sk->sk_callback_lock);
1628 	sk->sk_wq = parent->wq;
1629 	parent->sk = sk;
1630 	sk_set_socket(sk, parent);
1631 	security_sock_graft(sk, parent);
1632 	write_unlock_bh(&sk->sk_callback_lock);
1633 }
1634 
1635 extern int sock_i_uid(struct sock *sk);
1636 extern unsigned long sock_i_ino(struct sock *sk);
1637 
1638 static inline struct dst_entry *
1639 __sk_dst_get(struct sock *sk)
1640 {
1641 	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1642 						       lockdep_is_held(&sk->sk_lock.slock));
1643 }
1644 
1645 static inline struct dst_entry *
1646 sk_dst_get(struct sock *sk)
1647 {
1648 	struct dst_entry *dst;
1649 
1650 	rcu_read_lock();
1651 	dst = rcu_dereference(sk->sk_dst_cache);
1652 	if (dst)
1653 		dst_hold(dst);
1654 	rcu_read_unlock();
1655 	return dst;
1656 }
1657 
1658 extern void sk_reset_txq(struct sock *sk);
1659 
1660 static inline void dst_negative_advice(struct sock *sk)
1661 {
1662 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1663 
1664 	if (dst && dst->ops->negative_advice) {
1665 		ndst = dst->ops->negative_advice(dst);
1666 
1667 		if (ndst != dst) {
1668 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1669 			sk_reset_txq(sk);
1670 		}
1671 	}
1672 }
1673 
1674 static inline void
1675 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1676 {
1677 	struct dst_entry *old_dst;
1678 
1679 	sk_tx_queue_clear(sk);
1680 	/*
1681 	 * This can be called while sk is owned by the caller only,
1682 	 * with no state that can be checked in a rcu_dereference_check() cond
1683 	 */
1684 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1685 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1686 	dst_release(old_dst);
1687 }
1688 
1689 static inline void
1690 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1691 {
1692 	spin_lock(&sk->sk_dst_lock);
1693 	__sk_dst_set(sk, dst);
1694 	spin_unlock(&sk->sk_dst_lock);
1695 }
1696 
1697 static inline void
1698 __sk_dst_reset(struct sock *sk)
1699 {
1700 	__sk_dst_set(sk, NULL);
1701 }
1702 
1703 static inline void
1704 sk_dst_reset(struct sock *sk)
1705 {
1706 	spin_lock(&sk->sk_dst_lock);
1707 	__sk_dst_reset(sk);
1708 	spin_unlock(&sk->sk_dst_lock);
1709 }
1710 
1711 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1712 
1713 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1714 
1715 static inline bool sk_can_gso(const struct sock *sk)
1716 {
1717 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1718 }
1719 
1720 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1721 
1722 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1723 {
1724 	sk->sk_route_nocaps |= flags;
1725 	sk->sk_route_caps &= ~flags;
1726 }
1727 
1728 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1729 					   char __user *from, char *to,
1730 					   int copy, int offset)
1731 {
1732 	if (skb->ip_summed == CHECKSUM_NONE) {
1733 		int err = 0;
1734 		__wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1735 		if (err)
1736 			return err;
1737 		skb->csum = csum_block_add(skb->csum, csum, offset);
1738 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1739 		if (!access_ok(VERIFY_READ, from, copy) ||
1740 		    __copy_from_user_nocache(to, from, copy))
1741 			return -EFAULT;
1742 	} else if (copy_from_user(to, from, copy))
1743 		return -EFAULT;
1744 
1745 	return 0;
1746 }
1747 
1748 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1749 				       char __user *from, int copy)
1750 {
1751 	int err, offset = skb->len;
1752 
1753 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1754 				       copy, offset);
1755 	if (err)
1756 		__skb_trim(skb, offset);
1757 
1758 	return err;
1759 }
1760 
1761 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1762 					   struct sk_buff *skb,
1763 					   struct page *page,
1764 					   int off, int copy)
1765 {
1766 	int err;
1767 
1768 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1769 				       copy, skb->len);
1770 	if (err)
1771 		return err;
1772 
1773 	skb->len	     += copy;
1774 	skb->data_len	     += copy;
1775 	skb->truesize	     += copy;
1776 	sk->sk_wmem_queued   += copy;
1777 	sk_mem_charge(sk, copy);
1778 	return 0;
1779 }
1780 
1781 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1782 				   struct sk_buff *skb, struct page *page,
1783 				   int off, int copy)
1784 {
1785 	if (skb->ip_summed == CHECKSUM_NONE) {
1786 		int err = 0;
1787 		__wsum csum = csum_and_copy_from_user(from,
1788 						     page_address(page) + off,
1789 							    copy, 0, &err);
1790 		if (err)
1791 			return err;
1792 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1793 	} else if (copy_from_user(page_address(page) + off, from, copy))
1794 		return -EFAULT;
1795 
1796 	skb->len	     += copy;
1797 	skb->data_len	     += copy;
1798 	skb->truesize	     += copy;
1799 	sk->sk_wmem_queued   += copy;
1800 	sk_mem_charge(sk, copy);
1801 	return 0;
1802 }
1803 
1804 /**
1805  * sk_wmem_alloc_get - returns write allocations
1806  * @sk: socket
1807  *
1808  * Returns sk_wmem_alloc minus initial offset of one
1809  */
1810 static inline int sk_wmem_alloc_get(const struct sock *sk)
1811 {
1812 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1813 }
1814 
1815 /**
1816  * sk_rmem_alloc_get - returns read allocations
1817  * @sk: socket
1818  *
1819  * Returns sk_rmem_alloc
1820  */
1821 static inline int sk_rmem_alloc_get(const struct sock *sk)
1822 {
1823 	return atomic_read(&sk->sk_rmem_alloc);
1824 }
1825 
1826 /**
1827  * sk_has_allocations - check if allocations are outstanding
1828  * @sk: socket
1829  *
1830  * Returns true if socket has write or read allocations
1831  */
1832 static inline bool sk_has_allocations(const struct sock *sk)
1833 {
1834 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1835 }
1836 
1837 /**
1838  * wq_has_sleeper - check if there are any waiting processes
1839  * @wq: struct socket_wq
1840  *
1841  * Returns true if socket_wq has waiting processes
1842  *
1843  * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1844  * barrier call. They were added due to the race found within the tcp code.
1845  *
1846  * Consider following tcp code paths:
1847  *
1848  * CPU1                  CPU2
1849  *
1850  * sys_select            receive packet
1851  *   ...                 ...
1852  *   __add_wait_queue    update tp->rcv_nxt
1853  *   ...                 ...
1854  *   tp->rcv_nxt check   sock_def_readable
1855  *   ...                 {
1856  *   schedule               rcu_read_lock();
1857  *                          wq = rcu_dereference(sk->sk_wq);
1858  *                          if (wq && waitqueue_active(&wq->wait))
1859  *                              wake_up_interruptible(&wq->wait)
1860  *                          ...
1861  *                       }
1862  *
1863  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1864  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1865  * could then endup calling schedule and sleep forever if there are no more
1866  * data on the socket.
1867  *
1868  */
1869 static inline bool wq_has_sleeper(struct socket_wq *wq)
1870 {
1871 	/* We need to be sure we are in sync with the
1872 	 * add_wait_queue modifications to the wait queue.
1873 	 *
1874 	 * This memory barrier is paired in the sock_poll_wait.
1875 	 */
1876 	smp_mb();
1877 	return wq && waitqueue_active(&wq->wait);
1878 }
1879 
1880 /**
1881  * sock_poll_wait - place memory barrier behind the poll_wait call.
1882  * @filp:           file
1883  * @wait_address:   socket wait queue
1884  * @p:              poll_table
1885  *
1886  * See the comments in the wq_has_sleeper function.
1887  */
1888 static inline void sock_poll_wait(struct file *filp,
1889 		wait_queue_head_t *wait_address, poll_table *p)
1890 {
1891 	if (!poll_does_not_wait(p) && wait_address) {
1892 		poll_wait(filp, wait_address, p);
1893 		/* We need to be sure we are in sync with the
1894 		 * socket flags modification.
1895 		 *
1896 		 * This memory barrier is paired in the wq_has_sleeper.
1897 		 */
1898 		smp_mb();
1899 	}
1900 }
1901 
1902 /*
1903  *	Queue a received datagram if it will fit. Stream and sequenced
1904  *	protocols can't normally use this as they need to fit buffers in
1905  *	and play with them.
1906  *
1907  *	Inlined as it's very short and called for pretty much every
1908  *	packet ever received.
1909  */
1910 
1911 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1912 {
1913 	skb_orphan(skb);
1914 	skb->sk = sk;
1915 	skb->destructor = sock_wfree;
1916 	/*
1917 	 * We used to take a refcount on sk, but following operation
1918 	 * is enough to guarantee sk_free() wont free this sock until
1919 	 * all in-flight packets are completed
1920 	 */
1921 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1922 }
1923 
1924 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1925 {
1926 	skb_orphan(skb);
1927 	skb->sk = sk;
1928 	skb->destructor = sock_rfree;
1929 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1930 	sk_mem_charge(sk, skb->truesize);
1931 }
1932 
1933 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1934 			   unsigned long expires);
1935 
1936 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1937 
1938 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1939 
1940 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1941 
1942 /*
1943  *	Recover an error report and clear atomically
1944  */
1945 
1946 static inline int sock_error(struct sock *sk)
1947 {
1948 	int err;
1949 	if (likely(!sk->sk_err))
1950 		return 0;
1951 	err = xchg(&sk->sk_err, 0);
1952 	return -err;
1953 }
1954 
1955 static inline unsigned long sock_wspace(struct sock *sk)
1956 {
1957 	int amt = 0;
1958 
1959 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1960 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1961 		if (amt < 0)
1962 			amt = 0;
1963 	}
1964 	return amt;
1965 }
1966 
1967 static inline void sk_wake_async(struct sock *sk, int how, int band)
1968 {
1969 	if (sock_flag(sk, SOCK_FASYNC))
1970 		sock_wake_async(sk->sk_socket, how, band);
1971 }
1972 
1973 #define SOCK_MIN_SNDBUF 2048
1974 /*
1975  * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1976  * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1977  */
1978 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1979 
1980 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1981 {
1982 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1983 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1984 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1985 	}
1986 }
1987 
1988 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1989 
1990 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1991 {
1992 	struct page *page = NULL;
1993 
1994 	page = alloc_pages(sk->sk_allocation, 0);
1995 	if (!page) {
1996 		sk_enter_memory_pressure(sk);
1997 		sk_stream_moderate_sndbuf(sk);
1998 	}
1999 	return page;
2000 }
2001 
2002 /*
2003  *	Default write policy as shown to user space via poll/select/SIGIO
2004  */
2005 static inline bool sock_writeable(const struct sock *sk)
2006 {
2007 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2008 }
2009 
2010 static inline gfp_t gfp_any(void)
2011 {
2012 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2013 }
2014 
2015 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2016 {
2017 	return noblock ? 0 : sk->sk_rcvtimeo;
2018 }
2019 
2020 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2021 {
2022 	return noblock ? 0 : sk->sk_sndtimeo;
2023 }
2024 
2025 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2026 {
2027 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2028 }
2029 
2030 /* Alas, with timeout socket operations are not restartable.
2031  * Compare this to poll().
2032  */
2033 static inline int sock_intr_errno(long timeo)
2034 {
2035 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2036 }
2037 
2038 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2039 	struct sk_buff *skb);
2040 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2041 	struct sk_buff *skb);
2042 
2043 static inline void
2044 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2045 {
2046 	ktime_t kt = skb->tstamp;
2047 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2048 
2049 	/*
2050 	 * generate control messages if
2051 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2052 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
2053 	 * - software time stamp available and wanted
2054 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
2055 	 * - hardware time stamps available and wanted
2056 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
2057 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
2058 	 */
2059 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2060 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2061 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2062 	    (hwtstamps->hwtstamp.tv64 &&
2063 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2064 	    (hwtstamps->syststamp.tv64 &&
2065 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2066 		__sock_recv_timestamp(msg, sk, skb);
2067 	else
2068 		sk->sk_stamp = kt;
2069 
2070 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2071 		__sock_recv_wifi_status(msg, sk, skb);
2072 }
2073 
2074 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2075 				     struct sk_buff *skb);
2076 
2077 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2078 					  struct sk_buff *skb)
2079 {
2080 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2081 			   (1UL << SOCK_RCVTSTAMP)			| \
2082 			   (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)	| \
2083 			   (1UL << SOCK_TIMESTAMPING_SOFTWARE)		| \
2084 			   (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE)	| \
2085 			   (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2086 
2087 	if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2088 		__sock_recv_ts_and_drops(msg, sk, skb);
2089 	else
2090 		sk->sk_stamp = skb->tstamp;
2091 }
2092 
2093 /**
2094  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2095  * @sk:		socket sending this packet
2096  * @tx_flags:	filled with instructions for time stamping
2097  *
2098  * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2099  * parameters are invalid.
2100  */
2101 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2102 
2103 /**
2104  * sk_eat_skb - Release a skb if it is no longer needed
2105  * @sk: socket to eat this skb from
2106  * @skb: socket buffer to eat
2107  * @copied_early: flag indicating whether DMA operations copied this data early
2108  *
2109  * This routine must be called with interrupts disabled or with the socket
2110  * locked so that the sk_buff queue operation is ok.
2111 */
2112 #ifdef CONFIG_NET_DMA
2113 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2114 {
2115 	__skb_unlink(skb, &sk->sk_receive_queue);
2116 	if (!copied_early)
2117 		__kfree_skb(skb);
2118 	else
2119 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
2120 }
2121 #else
2122 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2123 {
2124 	__skb_unlink(skb, &sk->sk_receive_queue);
2125 	__kfree_skb(skb);
2126 }
2127 #endif
2128 
2129 static inline
2130 struct net *sock_net(const struct sock *sk)
2131 {
2132 	return read_pnet(&sk->sk_net);
2133 }
2134 
2135 static inline
2136 void sock_net_set(struct sock *sk, struct net *net)
2137 {
2138 	write_pnet(&sk->sk_net, net);
2139 }
2140 
2141 /*
2142  * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2143  * They should not hold a reference to a namespace in order to allow
2144  * to stop it.
2145  * Sockets after sk_change_net should be released using sk_release_kernel
2146  */
2147 static inline void sk_change_net(struct sock *sk, struct net *net)
2148 {
2149 	put_net(sock_net(sk));
2150 	sock_net_set(sk, hold_net(net));
2151 }
2152 
2153 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2154 {
2155 	if (unlikely(skb->sk)) {
2156 		struct sock *sk = skb->sk;
2157 
2158 		skb->destructor = NULL;
2159 		skb->sk = NULL;
2160 		return sk;
2161 	}
2162 	return NULL;
2163 }
2164 
2165 extern void sock_enable_timestamp(struct sock *sk, int flag);
2166 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2167 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2168 
2169 /*
2170  *	Enable debug/info messages
2171  */
2172 extern int net_msg_warn;
2173 #define NETDEBUG(fmt, args...) \
2174 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
2175 
2176 #define LIMIT_NETDEBUG(fmt, args...) \
2177 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2178 
2179 extern __u32 sysctl_wmem_max;
2180 extern __u32 sysctl_rmem_max;
2181 
2182 extern void sk_init(void);
2183 
2184 extern int sysctl_optmem_max;
2185 
2186 extern __u32 sysctl_wmem_default;
2187 extern __u32 sysctl_rmem_default;
2188 
2189 #endif	/* _SOCK_H */
2190