xref: /openbmc/linux/include/net/sock.h (revision 9a69abf8)
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 typedef __u32 __bitwise __portpair;
130 typedef __u64 __bitwise __addrpair;
131 
132 /**
133  *	struct sock_common - minimal network layer representation of sockets
134  *	@skc_daddr: Foreign IPv4 addr
135  *	@skc_rcv_saddr: Bound local IPv4 addr
136  *	@skc_hash: hash value used with various protocol lookup tables
137  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
138  *	@skc_dport: placeholder for inet_dport/tw_dport
139  *	@skc_num: placeholder for inet_num/tw_num
140  *	@skc_family: network address family
141  *	@skc_state: Connection state
142  *	@skc_reuse: %SO_REUSEADDR setting
143  *	@skc_reuseport: %SO_REUSEPORT setting
144  *	@skc_bound_dev_if: bound device index if != 0
145  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
146  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
147  *	@skc_prot: protocol handlers inside a network family
148  *	@skc_net: reference to the network namespace of this socket
149  *	@skc_node: main hash linkage for various protocol lookup tables
150  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
151  *	@skc_tx_queue_mapping: tx queue number for this connection
152  *	@skc_refcnt: reference count
153  *
154  *	This is the minimal network layer representation of sockets, the header
155  *	for struct sock and struct inet_timewait_sock.
156  */
157 struct sock_common {
158 	/* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
159 	 * address on 64bit arches : cf INET_MATCH() and INET_TW_MATCH()
160 	 */
161 	union {
162 		__addrpair	skc_addrpair;
163 		struct {
164 			__be32	skc_daddr;
165 			__be32	skc_rcv_saddr;
166 		};
167 	};
168 	union  {
169 		unsigned int	skc_hash;
170 		__u16		skc_u16hashes[2];
171 	};
172 	/* skc_dport && skc_num must be grouped as well */
173 	union {
174 		__portpair	skc_portpair;
175 		struct {
176 			__be16	skc_dport;
177 			__u16	skc_num;
178 		};
179 	};
180 
181 	unsigned short		skc_family;
182 	volatile unsigned char	skc_state;
183 	unsigned char		skc_reuse:4;
184 	unsigned char		skc_reuseport:4;
185 	int			skc_bound_dev_if;
186 	union {
187 		struct hlist_node	skc_bind_node;
188 		struct hlist_nulls_node skc_portaddr_node;
189 	};
190 	struct proto		*skc_prot;
191 #ifdef CONFIG_NET_NS
192 	struct net	 	*skc_net;
193 #endif
194 	/*
195 	 * fields between dontcopy_begin/dontcopy_end
196 	 * are not copied in sock_copy()
197 	 */
198 	/* private: */
199 	int			skc_dontcopy_begin[0];
200 	/* public: */
201 	union {
202 		struct hlist_node	skc_node;
203 		struct hlist_nulls_node skc_nulls_node;
204 	};
205 	int			skc_tx_queue_mapping;
206 	atomic_t		skc_refcnt;
207 	/* private: */
208 	int                     skc_dontcopy_end[0];
209 	/* public: */
210 };
211 
212 struct cg_proto;
213 /**
214   *	struct sock - network layer representation of sockets
215   *	@__sk_common: shared layout with inet_timewait_sock
216   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
217   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
218   *	@sk_lock:	synchronizer
219   *	@sk_rcvbuf: size of receive buffer in bytes
220   *	@sk_wq: sock wait queue and async head
221   *	@sk_rx_dst: receive input route used by early tcp demux
222   *	@sk_dst_cache: destination cache
223   *	@sk_dst_lock: destination cache lock
224   *	@sk_policy: flow policy
225   *	@sk_receive_queue: incoming packets
226   *	@sk_wmem_alloc: transmit queue bytes committed
227   *	@sk_write_queue: Packet sending queue
228   *	@sk_async_wait_queue: DMA copied packets
229   *	@sk_omem_alloc: "o" is "option" or "other"
230   *	@sk_wmem_queued: persistent queue size
231   *	@sk_forward_alloc: space allocated forward
232   *	@sk_allocation: allocation mode
233   *	@sk_sndbuf: size of send buffer in bytes
234   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
235   *		   %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
236   *	@sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets
237   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
238   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
239   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
240   *	@sk_gso_max_size: Maximum GSO segment size to build
241   *	@sk_gso_max_segs: Maximum number of GSO segments
242   *	@sk_lingertime: %SO_LINGER l_linger setting
243   *	@sk_backlog: always used with the per-socket spinlock held
244   *	@sk_callback_lock: used with the callbacks in the end of this struct
245   *	@sk_error_queue: rarely used
246   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
247   *			  IPV6_ADDRFORM for instance)
248   *	@sk_err: last error
249   *	@sk_err_soft: errors that don't cause failure but are the cause of a
250   *		      persistent failure not just 'timed out'
251   *	@sk_drops: raw/udp drops counter
252   *	@sk_ack_backlog: current listen backlog
253   *	@sk_max_ack_backlog: listen backlog set in listen()
254   *	@sk_priority: %SO_PRIORITY setting
255   *	@sk_cgrp_prioidx: socket group's priority map index
256   *	@sk_type: socket type (%SOCK_STREAM, etc)
257   *	@sk_protocol: which protocol this socket belongs in this network family
258   *	@sk_peer_pid: &struct pid for this socket's peer
259   *	@sk_peer_cred: %SO_PEERCRED setting
260   *	@sk_rcvlowat: %SO_RCVLOWAT setting
261   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
262   *	@sk_sndtimeo: %SO_SNDTIMEO setting
263   *	@sk_rxhash: flow hash received from netif layer
264   *	@sk_filter: socket filtering instructions
265   *	@sk_protinfo: private area, net family specific, when not using slab
266   *	@sk_timer: sock cleanup timer
267   *	@sk_stamp: time stamp of last packet received
268   *	@sk_socket: Identd and reporting IO signals
269   *	@sk_user_data: RPC layer private data
270   *	@sk_frag: cached page frag
271   *	@sk_peek_off: current peek_offset value
272   *	@sk_send_head: front of stuff to transmit
273   *	@sk_security: used by security modules
274   *	@sk_mark: generic packet mark
275   *	@sk_classid: this socket's cgroup classid
276   *	@sk_cgrp: this socket's cgroup-specific proto data
277   *	@sk_write_pending: a write to stream socket waits to start
278   *	@sk_state_change: callback to indicate change in the state of the sock
279   *	@sk_data_ready: callback to indicate there is data to be processed
280   *	@sk_write_space: callback to indicate there is bf sending space available
281   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
282   *	@sk_backlog_rcv: callback to process the backlog
283   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
284  */
285 struct sock {
286 	/*
287 	 * Now struct inet_timewait_sock also uses sock_common, so please just
288 	 * don't add nothing before this first member (__sk_common) --acme
289 	 */
290 	struct sock_common	__sk_common;
291 #define sk_node			__sk_common.skc_node
292 #define sk_nulls_node		__sk_common.skc_nulls_node
293 #define sk_refcnt		__sk_common.skc_refcnt
294 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
295 
296 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
297 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
298 #define sk_hash			__sk_common.skc_hash
299 #define sk_family		__sk_common.skc_family
300 #define sk_state		__sk_common.skc_state
301 #define sk_reuse		__sk_common.skc_reuse
302 #define sk_reuseport		__sk_common.skc_reuseport
303 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
304 #define sk_bind_node		__sk_common.skc_bind_node
305 #define sk_prot			__sk_common.skc_prot
306 #define sk_net			__sk_common.skc_net
307 	socket_lock_t		sk_lock;
308 	struct sk_buff_head	sk_receive_queue;
309 	/*
310 	 * The backlog queue is special, it is always used with
311 	 * the per-socket spinlock held and requires low latency
312 	 * access. Therefore we special case it's implementation.
313 	 * Note : rmem_alloc is in this structure to fill a hole
314 	 * on 64bit arches, not because its logically part of
315 	 * backlog.
316 	 */
317 	struct {
318 		atomic_t	rmem_alloc;
319 		int		len;
320 		struct sk_buff	*head;
321 		struct sk_buff	*tail;
322 	} sk_backlog;
323 #define sk_rmem_alloc sk_backlog.rmem_alloc
324 	int			sk_forward_alloc;
325 #ifdef CONFIG_RPS
326 	__u32			sk_rxhash;
327 #endif
328 	atomic_t		sk_drops;
329 	int			sk_rcvbuf;
330 
331 	struct sk_filter __rcu	*sk_filter;
332 	struct socket_wq __rcu	*sk_wq;
333 
334 #ifdef CONFIG_NET_DMA
335 	struct sk_buff_head	sk_async_wait_queue;
336 #endif
337 
338 #ifdef CONFIG_XFRM
339 	struct xfrm_policy	*sk_policy[2];
340 #endif
341 	unsigned long 		sk_flags;
342 	struct dst_entry	*sk_rx_dst;
343 	struct dst_entry __rcu	*sk_dst_cache;
344 	spinlock_t		sk_dst_lock;
345 	atomic_t		sk_wmem_alloc;
346 	atomic_t		sk_omem_alloc;
347 	int			sk_sndbuf;
348 	struct sk_buff_head	sk_write_queue;
349 	kmemcheck_bitfield_begin(flags);
350 	unsigned int		sk_shutdown  : 2,
351 				sk_no_check  : 2,
352 				sk_userlocks : 4,
353 				sk_protocol  : 8,
354 				sk_type      : 16;
355 	kmemcheck_bitfield_end(flags);
356 	int			sk_wmem_queued;
357 	gfp_t			sk_allocation;
358 	netdev_features_t	sk_route_caps;
359 	netdev_features_t	sk_route_nocaps;
360 	int			sk_gso_type;
361 	unsigned int		sk_gso_max_size;
362 	u16			sk_gso_max_segs;
363 	int			sk_rcvlowat;
364 	unsigned long	        sk_lingertime;
365 	struct sk_buff_head	sk_error_queue;
366 	struct proto		*sk_prot_creator;
367 	rwlock_t		sk_callback_lock;
368 	int			sk_err,
369 				sk_err_soft;
370 	unsigned short		sk_ack_backlog;
371 	unsigned short		sk_max_ack_backlog;
372 	__u32			sk_priority;
373 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
374 	__u32			sk_cgrp_prioidx;
375 #endif
376 	struct pid		*sk_peer_pid;
377 	const struct cred	*sk_peer_cred;
378 	long			sk_rcvtimeo;
379 	long			sk_sndtimeo;
380 	void			*sk_protinfo;
381 	struct timer_list	sk_timer;
382 	ktime_t			sk_stamp;
383 	struct socket		*sk_socket;
384 	void			*sk_user_data;
385 	struct page_frag	sk_frag;
386 	struct sk_buff		*sk_send_head;
387 	__s32			sk_peek_off;
388 	int			sk_write_pending;
389 #ifdef CONFIG_SECURITY
390 	void			*sk_security;
391 #endif
392 	__u32			sk_mark;
393 	u32			sk_classid;
394 	struct cg_proto		*sk_cgrp;
395 	void			(*sk_state_change)(struct sock *sk);
396 	void			(*sk_data_ready)(struct sock *sk, int bytes);
397 	void			(*sk_write_space)(struct sock *sk);
398 	void			(*sk_error_report)(struct sock *sk);
399 	int			(*sk_backlog_rcv)(struct sock *sk,
400 						  struct sk_buff *skb);
401 	void                    (*sk_destruct)(struct sock *sk);
402 };
403 
404 /*
405  * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
406  * or not whether his port will be reused by someone else. SK_FORCE_REUSE
407  * on a socket means that the socket will reuse everybody else's port
408  * without looking at the other's sk_reuse value.
409  */
410 
411 #define SK_NO_REUSE	0
412 #define SK_CAN_REUSE	1
413 #define SK_FORCE_REUSE	2
414 
415 static inline int sk_peek_offset(struct sock *sk, int flags)
416 {
417 	if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
418 		return sk->sk_peek_off;
419 	else
420 		return 0;
421 }
422 
423 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
424 {
425 	if (sk->sk_peek_off >= 0) {
426 		if (sk->sk_peek_off >= val)
427 			sk->sk_peek_off -= val;
428 		else
429 			sk->sk_peek_off = 0;
430 	}
431 }
432 
433 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
434 {
435 	if (sk->sk_peek_off >= 0)
436 		sk->sk_peek_off += val;
437 }
438 
439 /*
440  * Hashed lists helper routines
441  */
442 static inline struct sock *sk_entry(const struct hlist_node *node)
443 {
444 	return hlist_entry(node, struct sock, sk_node);
445 }
446 
447 static inline struct sock *__sk_head(const struct hlist_head *head)
448 {
449 	return hlist_entry(head->first, struct sock, sk_node);
450 }
451 
452 static inline struct sock *sk_head(const struct hlist_head *head)
453 {
454 	return hlist_empty(head) ? NULL : __sk_head(head);
455 }
456 
457 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
458 {
459 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
460 }
461 
462 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
463 {
464 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
465 }
466 
467 static inline struct sock *sk_next(const struct sock *sk)
468 {
469 	return sk->sk_node.next ?
470 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
471 }
472 
473 static inline struct sock *sk_nulls_next(const struct sock *sk)
474 {
475 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
476 		hlist_nulls_entry(sk->sk_nulls_node.next,
477 				  struct sock, sk_nulls_node) :
478 		NULL;
479 }
480 
481 static inline bool sk_unhashed(const struct sock *sk)
482 {
483 	return hlist_unhashed(&sk->sk_node);
484 }
485 
486 static inline bool sk_hashed(const struct sock *sk)
487 {
488 	return !sk_unhashed(sk);
489 }
490 
491 static inline void sk_node_init(struct hlist_node *node)
492 {
493 	node->pprev = NULL;
494 }
495 
496 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
497 {
498 	node->pprev = NULL;
499 }
500 
501 static inline void __sk_del_node(struct sock *sk)
502 {
503 	__hlist_del(&sk->sk_node);
504 }
505 
506 /* NB: equivalent to hlist_del_init_rcu */
507 static inline bool __sk_del_node_init(struct sock *sk)
508 {
509 	if (sk_hashed(sk)) {
510 		__sk_del_node(sk);
511 		sk_node_init(&sk->sk_node);
512 		return true;
513 	}
514 	return false;
515 }
516 
517 /* Grab socket reference count. This operation is valid only
518    when sk is ALREADY grabbed f.e. it is found in hash table
519    or a list and the lookup is made under lock preventing hash table
520    modifications.
521  */
522 
523 static inline void sock_hold(struct sock *sk)
524 {
525 	atomic_inc(&sk->sk_refcnt);
526 }
527 
528 /* Ungrab socket in the context, which assumes that socket refcnt
529    cannot hit zero, f.e. it is true in context of any socketcall.
530  */
531 static inline void __sock_put(struct sock *sk)
532 {
533 	atomic_dec(&sk->sk_refcnt);
534 }
535 
536 static inline bool sk_del_node_init(struct sock *sk)
537 {
538 	bool rc = __sk_del_node_init(sk);
539 
540 	if (rc) {
541 		/* paranoid for a while -acme */
542 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
543 		__sock_put(sk);
544 	}
545 	return rc;
546 }
547 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
548 
549 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
550 {
551 	if (sk_hashed(sk)) {
552 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
553 		return true;
554 	}
555 	return false;
556 }
557 
558 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
559 {
560 	bool rc = __sk_nulls_del_node_init_rcu(sk);
561 
562 	if (rc) {
563 		/* paranoid for a while -acme */
564 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
565 		__sock_put(sk);
566 	}
567 	return rc;
568 }
569 
570 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
571 {
572 	hlist_add_head(&sk->sk_node, list);
573 }
574 
575 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
576 {
577 	sock_hold(sk);
578 	__sk_add_node(sk, list);
579 }
580 
581 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
582 {
583 	sock_hold(sk);
584 	hlist_add_head_rcu(&sk->sk_node, list);
585 }
586 
587 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
588 {
589 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
590 }
591 
592 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
593 {
594 	sock_hold(sk);
595 	__sk_nulls_add_node_rcu(sk, list);
596 }
597 
598 static inline void __sk_del_bind_node(struct sock *sk)
599 {
600 	__hlist_del(&sk->sk_bind_node);
601 }
602 
603 static inline void sk_add_bind_node(struct sock *sk,
604 					struct hlist_head *list)
605 {
606 	hlist_add_head(&sk->sk_bind_node, list);
607 }
608 
609 #define sk_for_each(__sk, list) \
610 	hlist_for_each_entry(__sk, list, sk_node)
611 #define sk_for_each_rcu(__sk, list) \
612 	hlist_for_each_entry_rcu(__sk, list, sk_node)
613 #define sk_nulls_for_each(__sk, node, list) \
614 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
615 #define sk_nulls_for_each_rcu(__sk, node, list) \
616 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
617 #define sk_for_each_from(__sk) \
618 	hlist_for_each_entry_from(__sk, sk_node)
619 #define sk_nulls_for_each_from(__sk, node) \
620 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
621 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
622 #define sk_for_each_safe(__sk, tmp, list) \
623 	hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
624 #define sk_for_each_bound(__sk, list) \
625 	hlist_for_each_entry(__sk, list, sk_bind_node)
626 
627 static inline struct user_namespace *sk_user_ns(struct sock *sk)
628 {
629 	/* Careful only use this in a context where these parameters
630 	 * can not change and must all be valid, such as recvmsg from
631 	 * userspace.
632 	 */
633 	return sk->sk_socket->file->f_cred->user_ns;
634 }
635 
636 /* Sock flags */
637 enum sock_flags {
638 	SOCK_DEAD,
639 	SOCK_DONE,
640 	SOCK_URGINLINE,
641 	SOCK_KEEPOPEN,
642 	SOCK_LINGER,
643 	SOCK_DESTROY,
644 	SOCK_BROADCAST,
645 	SOCK_TIMESTAMP,
646 	SOCK_ZAPPED,
647 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
648 	SOCK_DBG, /* %SO_DEBUG setting */
649 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
650 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
651 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
652 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
653 	SOCK_MEMALLOC, /* VM depends on this socket for swapping */
654 	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
655 	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
656 	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
657 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
658 	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
659 	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
660 	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
661 	SOCK_FASYNC, /* fasync() active */
662 	SOCK_RXQ_OVFL,
663 	SOCK_ZEROCOPY, /* buffers from userspace */
664 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
665 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
666 		     * Will use last 4 bytes of packet sent from
667 		     * user-space instead.
668 		     */
669 	SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
670 };
671 
672 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
673 {
674 	nsk->sk_flags = osk->sk_flags;
675 }
676 
677 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
678 {
679 	__set_bit(flag, &sk->sk_flags);
680 }
681 
682 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
683 {
684 	__clear_bit(flag, &sk->sk_flags);
685 }
686 
687 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
688 {
689 	return test_bit(flag, &sk->sk_flags);
690 }
691 
692 #ifdef CONFIG_NET
693 extern struct static_key memalloc_socks;
694 static inline int sk_memalloc_socks(void)
695 {
696 	return static_key_false(&memalloc_socks);
697 }
698 #else
699 
700 static inline int sk_memalloc_socks(void)
701 {
702 	return 0;
703 }
704 
705 #endif
706 
707 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
708 {
709 	return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
710 }
711 
712 static inline void sk_acceptq_removed(struct sock *sk)
713 {
714 	sk->sk_ack_backlog--;
715 }
716 
717 static inline void sk_acceptq_added(struct sock *sk)
718 {
719 	sk->sk_ack_backlog++;
720 }
721 
722 static inline bool sk_acceptq_is_full(const struct sock *sk)
723 {
724 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
725 }
726 
727 /*
728  * Compute minimal free write space needed to queue new packets.
729  */
730 static inline int sk_stream_min_wspace(const struct sock *sk)
731 {
732 	return sk->sk_wmem_queued >> 1;
733 }
734 
735 static inline int sk_stream_wspace(const struct sock *sk)
736 {
737 	return sk->sk_sndbuf - sk->sk_wmem_queued;
738 }
739 
740 extern void sk_stream_write_space(struct sock *sk);
741 
742 static inline bool sk_stream_memory_free(const struct sock *sk)
743 {
744 	return sk->sk_wmem_queued < sk->sk_sndbuf;
745 }
746 
747 /* OOB backlog add */
748 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
749 {
750 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
751 	skb_dst_force(skb);
752 
753 	if (!sk->sk_backlog.tail)
754 		sk->sk_backlog.head = skb;
755 	else
756 		sk->sk_backlog.tail->next = skb;
757 
758 	sk->sk_backlog.tail = skb;
759 	skb->next = NULL;
760 }
761 
762 /*
763  * Take into account size of receive queue and backlog queue
764  * Do not take into account this skb truesize,
765  * to allow even a single big packet to come.
766  */
767 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
768 				     unsigned int limit)
769 {
770 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
771 
772 	return qsize > limit;
773 }
774 
775 /* The per-socket spinlock must be held here. */
776 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
777 					      unsigned int limit)
778 {
779 	if (sk_rcvqueues_full(sk, skb, limit))
780 		return -ENOBUFS;
781 
782 	__sk_add_backlog(sk, skb);
783 	sk->sk_backlog.len += skb->truesize;
784 	return 0;
785 }
786 
787 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
788 
789 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
790 {
791 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
792 		return __sk_backlog_rcv(sk, skb);
793 
794 	return sk->sk_backlog_rcv(sk, skb);
795 }
796 
797 static inline void sock_rps_record_flow(const struct sock *sk)
798 {
799 #ifdef CONFIG_RPS
800 	struct rps_sock_flow_table *sock_flow_table;
801 
802 	rcu_read_lock();
803 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
804 	rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
805 	rcu_read_unlock();
806 #endif
807 }
808 
809 static inline void sock_rps_reset_flow(const struct sock *sk)
810 {
811 #ifdef CONFIG_RPS
812 	struct rps_sock_flow_table *sock_flow_table;
813 
814 	rcu_read_lock();
815 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
816 	rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
817 	rcu_read_unlock();
818 #endif
819 }
820 
821 static inline void sock_rps_save_rxhash(struct sock *sk,
822 					const struct sk_buff *skb)
823 {
824 #ifdef CONFIG_RPS
825 	if (unlikely(sk->sk_rxhash != skb->rxhash)) {
826 		sock_rps_reset_flow(sk);
827 		sk->sk_rxhash = skb->rxhash;
828 	}
829 #endif
830 }
831 
832 static inline void sock_rps_reset_rxhash(struct sock *sk)
833 {
834 #ifdef CONFIG_RPS
835 	sock_rps_reset_flow(sk);
836 	sk->sk_rxhash = 0;
837 #endif
838 }
839 
840 #define sk_wait_event(__sk, __timeo, __condition)			\
841 	({	int __rc;						\
842 		release_sock(__sk);					\
843 		__rc = __condition;					\
844 		if (!__rc) {						\
845 			*(__timeo) = schedule_timeout(*(__timeo));	\
846 		}							\
847 		lock_sock(__sk);					\
848 		__rc = __condition;					\
849 		__rc;							\
850 	})
851 
852 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
853 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
854 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
855 extern int sk_stream_error(struct sock *sk, int flags, int err);
856 extern void sk_stream_kill_queues(struct sock *sk);
857 extern void sk_set_memalloc(struct sock *sk);
858 extern void sk_clear_memalloc(struct sock *sk);
859 
860 extern int sk_wait_data(struct sock *sk, long *timeo);
861 
862 struct request_sock_ops;
863 struct timewait_sock_ops;
864 struct inet_hashinfo;
865 struct raw_hashinfo;
866 struct module;
867 
868 /* Networking protocol blocks we attach to sockets.
869  * socket layer -> transport layer interface
870  * transport -> network interface is defined by struct inet_proto
871  */
872 struct proto {
873 	void			(*close)(struct sock *sk,
874 					long timeout);
875 	int			(*connect)(struct sock *sk,
876 					struct sockaddr *uaddr,
877 					int addr_len);
878 	int			(*disconnect)(struct sock *sk, int flags);
879 
880 	struct sock *		(*accept)(struct sock *sk, int flags, int *err);
881 
882 	int			(*ioctl)(struct sock *sk, int cmd,
883 					 unsigned long arg);
884 	int			(*init)(struct sock *sk);
885 	void			(*destroy)(struct sock *sk);
886 	void			(*shutdown)(struct sock *sk, int how);
887 	int			(*setsockopt)(struct sock *sk, int level,
888 					int optname, char __user *optval,
889 					unsigned int optlen);
890 	int			(*getsockopt)(struct sock *sk, int level,
891 					int optname, char __user *optval,
892 					int __user *option);
893 #ifdef CONFIG_COMPAT
894 	int			(*compat_setsockopt)(struct sock *sk,
895 					int level,
896 					int optname, char __user *optval,
897 					unsigned int optlen);
898 	int			(*compat_getsockopt)(struct sock *sk,
899 					int level,
900 					int optname, char __user *optval,
901 					int __user *option);
902 	int			(*compat_ioctl)(struct sock *sk,
903 					unsigned int cmd, unsigned long arg);
904 #endif
905 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
906 					   struct msghdr *msg, size_t len);
907 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
908 					   struct msghdr *msg,
909 					   size_t len, int noblock, int flags,
910 					   int *addr_len);
911 	int			(*sendpage)(struct sock *sk, struct page *page,
912 					int offset, size_t size, int flags);
913 	int			(*bind)(struct sock *sk,
914 					struct sockaddr *uaddr, int addr_len);
915 
916 	int			(*backlog_rcv) (struct sock *sk,
917 						struct sk_buff *skb);
918 
919 	void		(*release_cb)(struct sock *sk);
920 	void		(*mtu_reduced)(struct sock *sk);
921 
922 	/* Keeping track of sk's, looking them up, and port selection methods. */
923 	void			(*hash)(struct sock *sk);
924 	void			(*unhash)(struct sock *sk);
925 	void			(*rehash)(struct sock *sk);
926 	int			(*get_port)(struct sock *sk, unsigned short snum);
927 	void			(*clear_sk)(struct sock *sk, int size);
928 
929 	/* Keeping track of sockets in use */
930 #ifdef CONFIG_PROC_FS
931 	unsigned int		inuse_idx;
932 #endif
933 
934 	/* Memory pressure */
935 	void			(*enter_memory_pressure)(struct sock *sk);
936 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
937 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
938 	/*
939 	 * Pressure flag: try to collapse.
940 	 * Technical note: it is used by multiple contexts non atomically.
941 	 * All the __sk_mem_schedule() is of this nature: accounting
942 	 * is strict, actions are advisory and have some latency.
943 	 */
944 	int			*memory_pressure;
945 	long			*sysctl_mem;
946 	int			*sysctl_wmem;
947 	int			*sysctl_rmem;
948 	int			max_header;
949 	bool			no_autobind;
950 
951 	struct kmem_cache	*slab;
952 	unsigned int		obj_size;
953 	int			slab_flags;
954 
955 	struct percpu_counter	*orphan_count;
956 
957 	struct request_sock_ops	*rsk_prot;
958 	struct timewait_sock_ops *twsk_prot;
959 
960 	union {
961 		struct inet_hashinfo	*hashinfo;
962 		struct udp_table	*udp_table;
963 		struct raw_hashinfo	*raw_hash;
964 	} h;
965 
966 	struct module		*owner;
967 
968 	char			name[32];
969 
970 	struct list_head	node;
971 #ifdef SOCK_REFCNT_DEBUG
972 	atomic_t		socks;
973 #endif
974 #ifdef CONFIG_MEMCG_KMEM
975 	/*
976 	 * cgroup specific init/deinit functions. Called once for all
977 	 * protocols that implement it, from cgroups populate function.
978 	 * This function has to setup any files the protocol want to
979 	 * appear in the kmem cgroup filesystem.
980 	 */
981 	int			(*init_cgroup)(struct mem_cgroup *memcg,
982 					       struct cgroup_subsys *ss);
983 	void			(*destroy_cgroup)(struct mem_cgroup *memcg);
984 	struct cg_proto		*(*proto_cgroup)(struct mem_cgroup *memcg);
985 #endif
986 };
987 
988 /*
989  * Bits in struct cg_proto.flags
990  */
991 enum cg_proto_flags {
992 	/* Currently active and new sockets should be assigned to cgroups */
993 	MEMCG_SOCK_ACTIVE,
994 	/* It was ever activated; we must disarm static keys on destruction */
995 	MEMCG_SOCK_ACTIVATED,
996 };
997 
998 struct cg_proto {
999 	void			(*enter_memory_pressure)(struct sock *sk);
1000 	struct res_counter	*memory_allocated;	/* Current allocated memory. */
1001 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1002 	int			*memory_pressure;
1003 	long			*sysctl_mem;
1004 	unsigned long		flags;
1005 	/*
1006 	 * memcg field is used to find which memcg we belong directly
1007 	 * Each memcg struct can hold more than one cg_proto, so container_of
1008 	 * won't really cut.
1009 	 *
1010 	 * The elegant solution would be having an inverse function to
1011 	 * proto_cgroup in struct proto, but that means polluting the structure
1012 	 * for everybody, instead of just for memcg users.
1013 	 */
1014 	struct mem_cgroup	*memcg;
1015 };
1016 
1017 extern int proto_register(struct proto *prot, int alloc_slab);
1018 extern void proto_unregister(struct proto *prot);
1019 
1020 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1021 {
1022 	return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1023 }
1024 
1025 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1026 {
1027 	return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1028 }
1029 
1030 #ifdef SOCK_REFCNT_DEBUG
1031 static inline void sk_refcnt_debug_inc(struct sock *sk)
1032 {
1033 	atomic_inc(&sk->sk_prot->socks);
1034 }
1035 
1036 static inline void sk_refcnt_debug_dec(struct sock *sk)
1037 {
1038 	atomic_dec(&sk->sk_prot->socks);
1039 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1040 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1041 }
1042 
1043 static inline void sk_refcnt_debug_release(const struct sock *sk)
1044 {
1045 	if (atomic_read(&sk->sk_refcnt) != 1)
1046 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1047 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1048 }
1049 #else /* SOCK_REFCNT_DEBUG */
1050 #define sk_refcnt_debug_inc(sk) do { } while (0)
1051 #define sk_refcnt_debug_dec(sk) do { } while (0)
1052 #define sk_refcnt_debug_release(sk) do { } while (0)
1053 #endif /* SOCK_REFCNT_DEBUG */
1054 
1055 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1056 extern struct static_key memcg_socket_limit_enabled;
1057 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1058 					       struct cg_proto *cg_proto)
1059 {
1060 	return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1061 }
1062 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1063 #else
1064 #define mem_cgroup_sockets_enabled 0
1065 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1066 					       struct cg_proto *cg_proto)
1067 {
1068 	return NULL;
1069 }
1070 #endif
1071 
1072 
1073 static inline bool sk_has_memory_pressure(const struct sock *sk)
1074 {
1075 	return sk->sk_prot->memory_pressure != NULL;
1076 }
1077 
1078 static inline bool sk_under_memory_pressure(const struct sock *sk)
1079 {
1080 	if (!sk->sk_prot->memory_pressure)
1081 		return false;
1082 
1083 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1084 		return !!*sk->sk_cgrp->memory_pressure;
1085 
1086 	return !!*sk->sk_prot->memory_pressure;
1087 }
1088 
1089 static inline void sk_leave_memory_pressure(struct sock *sk)
1090 {
1091 	int *memory_pressure = sk->sk_prot->memory_pressure;
1092 
1093 	if (!memory_pressure)
1094 		return;
1095 
1096 	if (*memory_pressure)
1097 		*memory_pressure = 0;
1098 
1099 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1100 		struct cg_proto *cg_proto = sk->sk_cgrp;
1101 		struct proto *prot = sk->sk_prot;
1102 
1103 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1104 			if (*cg_proto->memory_pressure)
1105 				*cg_proto->memory_pressure = 0;
1106 	}
1107 
1108 }
1109 
1110 static inline void sk_enter_memory_pressure(struct sock *sk)
1111 {
1112 	if (!sk->sk_prot->enter_memory_pressure)
1113 		return;
1114 
1115 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1116 		struct cg_proto *cg_proto = sk->sk_cgrp;
1117 		struct proto *prot = sk->sk_prot;
1118 
1119 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1120 			cg_proto->enter_memory_pressure(sk);
1121 	}
1122 
1123 	sk->sk_prot->enter_memory_pressure(sk);
1124 }
1125 
1126 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1127 {
1128 	long *prot = sk->sk_prot->sysctl_mem;
1129 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1130 		prot = sk->sk_cgrp->sysctl_mem;
1131 	return prot[index];
1132 }
1133 
1134 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1135 					      unsigned long amt,
1136 					      int *parent_status)
1137 {
1138 	struct res_counter *fail;
1139 	int ret;
1140 
1141 	ret = res_counter_charge_nofail(prot->memory_allocated,
1142 					amt << PAGE_SHIFT, &fail);
1143 	if (ret < 0)
1144 		*parent_status = OVER_LIMIT;
1145 }
1146 
1147 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1148 					      unsigned long amt)
1149 {
1150 	res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1151 }
1152 
1153 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1154 {
1155 	u64 ret;
1156 	ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1157 	return ret >> PAGE_SHIFT;
1158 }
1159 
1160 static inline long
1161 sk_memory_allocated(const struct sock *sk)
1162 {
1163 	struct proto *prot = sk->sk_prot;
1164 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1165 		return memcg_memory_allocated_read(sk->sk_cgrp);
1166 
1167 	return atomic_long_read(prot->memory_allocated);
1168 }
1169 
1170 static inline long
1171 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1172 {
1173 	struct proto *prot = sk->sk_prot;
1174 
1175 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1176 		memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1177 		/* update the root cgroup regardless */
1178 		atomic_long_add_return(amt, prot->memory_allocated);
1179 		return memcg_memory_allocated_read(sk->sk_cgrp);
1180 	}
1181 
1182 	return atomic_long_add_return(amt, prot->memory_allocated);
1183 }
1184 
1185 static inline void
1186 sk_memory_allocated_sub(struct sock *sk, int amt)
1187 {
1188 	struct proto *prot = sk->sk_prot;
1189 
1190 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1191 		memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1192 
1193 	atomic_long_sub(amt, prot->memory_allocated);
1194 }
1195 
1196 static inline void sk_sockets_allocated_dec(struct sock *sk)
1197 {
1198 	struct proto *prot = sk->sk_prot;
1199 
1200 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1201 		struct cg_proto *cg_proto = sk->sk_cgrp;
1202 
1203 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1204 			percpu_counter_dec(cg_proto->sockets_allocated);
1205 	}
1206 
1207 	percpu_counter_dec(prot->sockets_allocated);
1208 }
1209 
1210 static inline void sk_sockets_allocated_inc(struct sock *sk)
1211 {
1212 	struct proto *prot = sk->sk_prot;
1213 
1214 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1215 		struct cg_proto *cg_proto = sk->sk_cgrp;
1216 
1217 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1218 			percpu_counter_inc(cg_proto->sockets_allocated);
1219 	}
1220 
1221 	percpu_counter_inc(prot->sockets_allocated);
1222 }
1223 
1224 static inline int
1225 sk_sockets_allocated_read_positive(struct sock *sk)
1226 {
1227 	struct proto *prot = sk->sk_prot;
1228 
1229 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1230 		return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1231 
1232 	return percpu_counter_read_positive(prot->sockets_allocated);
1233 }
1234 
1235 static inline int
1236 proto_sockets_allocated_sum_positive(struct proto *prot)
1237 {
1238 	return percpu_counter_sum_positive(prot->sockets_allocated);
1239 }
1240 
1241 static inline long
1242 proto_memory_allocated(struct proto *prot)
1243 {
1244 	return atomic_long_read(prot->memory_allocated);
1245 }
1246 
1247 static inline bool
1248 proto_memory_pressure(struct proto *prot)
1249 {
1250 	if (!prot->memory_pressure)
1251 		return false;
1252 	return !!*prot->memory_pressure;
1253 }
1254 
1255 
1256 #ifdef CONFIG_PROC_FS
1257 /* Called with local bh disabled */
1258 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1259 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1260 #else
1261 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1262 		int inc)
1263 {
1264 }
1265 #endif
1266 
1267 
1268 /* With per-bucket locks this operation is not-atomic, so that
1269  * this version is not worse.
1270  */
1271 static inline void __sk_prot_rehash(struct sock *sk)
1272 {
1273 	sk->sk_prot->unhash(sk);
1274 	sk->sk_prot->hash(sk);
1275 }
1276 
1277 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1278 
1279 /* About 10 seconds */
1280 #define SOCK_DESTROY_TIME (10*HZ)
1281 
1282 /* Sockets 0-1023 can't be bound to unless you are superuser */
1283 #define PROT_SOCK	1024
1284 
1285 #define SHUTDOWN_MASK	3
1286 #define RCV_SHUTDOWN	1
1287 #define SEND_SHUTDOWN	2
1288 
1289 #define SOCK_SNDBUF_LOCK	1
1290 #define SOCK_RCVBUF_LOCK	2
1291 #define SOCK_BINDADDR_LOCK	4
1292 #define SOCK_BINDPORT_LOCK	8
1293 
1294 /* sock_iocb: used to kick off async processing of socket ios */
1295 struct sock_iocb {
1296 	struct list_head	list;
1297 
1298 	int			flags;
1299 	int			size;
1300 	struct socket		*sock;
1301 	struct sock		*sk;
1302 	struct scm_cookie	*scm;
1303 	struct msghdr		*msg, async_msg;
1304 	struct kiocb		*kiocb;
1305 };
1306 
1307 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1308 {
1309 	return (struct sock_iocb *)iocb->private;
1310 }
1311 
1312 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1313 {
1314 	return si->kiocb;
1315 }
1316 
1317 struct socket_alloc {
1318 	struct socket socket;
1319 	struct inode vfs_inode;
1320 };
1321 
1322 static inline struct socket *SOCKET_I(struct inode *inode)
1323 {
1324 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1325 }
1326 
1327 static inline struct inode *SOCK_INODE(struct socket *socket)
1328 {
1329 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1330 }
1331 
1332 /*
1333  * Functions for memory accounting
1334  */
1335 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1336 extern void __sk_mem_reclaim(struct sock *sk);
1337 
1338 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1339 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1340 #define SK_MEM_SEND	0
1341 #define SK_MEM_RECV	1
1342 
1343 static inline int sk_mem_pages(int amt)
1344 {
1345 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1346 }
1347 
1348 static inline bool sk_has_account(struct sock *sk)
1349 {
1350 	/* return true if protocol supports memory accounting */
1351 	return !!sk->sk_prot->memory_allocated;
1352 }
1353 
1354 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1355 {
1356 	if (!sk_has_account(sk))
1357 		return true;
1358 	return size <= sk->sk_forward_alloc ||
1359 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1360 }
1361 
1362 static inline bool
1363 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1364 {
1365 	if (!sk_has_account(sk))
1366 		return true;
1367 	return size<= sk->sk_forward_alloc ||
1368 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1369 		skb_pfmemalloc(skb);
1370 }
1371 
1372 static inline void sk_mem_reclaim(struct sock *sk)
1373 {
1374 	if (!sk_has_account(sk))
1375 		return;
1376 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1377 		__sk_mem_reclaim(sk);
1378 }
1379 
1380 static inline void sk_mem_reclaim_partial(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);
1386 }
1387 
1388 static inline void sk_mem_charge(struct sock *sk, int size)
1389 {
1390 	if (!sk_has_account(sk))
1391 		return;
1392 	sk->sk_forward_alloc -= size;
1393 }
1394 
1395 static inline void sk_mem_uncharge(struct sock *sk, int size)
1396 {
1397 	if (!sk_has_account(sk))
1398 		return;
1399 	sk->sk_forward_alloc += size;
1400 }
1401 
1402 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1403 {
1404 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1405 	sk->sk_wmem_queued -= skb->truesize;
1406 	sk_mem_uncharge(sk, skb->truesize);
1407 	__kfree_skb(skb);
1408 }
1409 
1410 /* Used by processes to "lock" a socket state, so that
1411  * interrupts and bottom half handlers won't change it
1412  * from under us. It essentially blocks any incoming
1413  * packets, so that we won't get any new data or any
1414  * packets that change the state of the socket.
1415  *
1416  * While locked, BH processing will add new packets to
1417  * the backlog queue.  This queue is processed by the
1418  * owner of the socket lock right before it is released.
1419  *
1420  * Since ~2.3.5 it is also exclusive sleep lock serializing
1421  * accesses from user process context.
1422  */
1423 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1424 
1425 /*
1426  * Macro so as to not evaluate some arguments when
1427  * lockdep is not enabled.
1428  *
1429  * Mark both the sk_lock and the sk_lock.slock as a
1430  * per-address-family lock class.
1431  */
1432 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1433 do {									\
1434 	sk->sk_lock.owned = 0;						\
1435 	init_waitqueue_head(&sk->sk_lock.wq);				\
1436 	spin_lock_init(&(sk)->sk_lock.slock);				\
1437 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1438 			sizeof((sk)->sk_lock));				\
1439 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1440 				(skey), (sname));				\
1441 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1442 } while (0)
1443 
1444 extern void lock_sock_nested(struct sock *sk, int subclass);
1445 
1446 static inline void lock_sock(struct sock *sk)
1447 {
1448 	lock_sock_nested(sk, 0);
1449 }
1450 
1451 extern void release_sock(struct sock *sk);
1452 
1453 /* BH context may only use the following locking interface. */
1454 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1455 #define bh_lock_sock_nested(__sk) \
1456 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1457 				SINGLE_DEPTH_NESTING)
1458 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1459 
1460 extern bool lock_sock_fast(struct sock *sk);
1461 /**
1462  * unlock_sock_fast - complement of lock_sock_fast
1463  * @sk: socket
1464  * @slow: slow mode
1465  *
1466  * fast unlock socket for user context.
1467  * If slow mode is on, we call regular release_sock()
1468  */
1469 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1470 {
1471 	if (slow)
1472 		release_sock(sk);
1473 	else
1474 		spin_unlock_bh(&sk->sk_lock.slock);
1475 }
1476 
1477 
1478 extern struct sock		*sk_alloc(struct net *net, int family,
1479 					  gfp_t priority,
1480 					  struct proto *prot);
1481 extern void			sk_free(struct sock *sk);
1482 extern void			sk_release_kernel(struct sock *sk);
1483 extern struct sock		*sk_clone_lock(const struct sock *sk,
1484 					       const gfp_t priority);
1485 
1486 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
1487 					      unsigned long size, int force,
1488 					      gfp_t priority);
1489 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
1490 					      unsigned long size, int force,
1491 					      gfp_t priority);
1492 extern void			sock_wfree(struct sk_buff *skb);
1493 extern void			sock_rfree(struct sk_buff *skb);
1494 extern void			sock_edemux(struct sk_buff *skb);
1495 
1496 extern int			sock_setsockopt(struct socket *sock, int level,
1497 						int op, char __user *optval,
1498 						unsigned int optlen);
1499 
1500 extern int			sock_getsockopt(struct socket *sock, int level,
1501 						int op, char __user *optval,
1502 						int __user *optlen);
1503 extern struct sk_buff		*sock_alloc_send_skb(struct sock *sk,
1504 						     unsigned long size,
1505 						     int noblock,
1506 						     int *errcode);
1507 extern struct sk_buff		*sock_alloc_send_pskb(struct sock *sk,
1508 						      unsigned long header_len,
1509 						      unsigned long data_len,
1510 						      int noblock,
1511 						      int *errcode);
1512 extern void *sock_kmalloc(struct sock *sk, int size,
1513 			  gfp_t priority);
1514 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1515 extern void sk_send_sigurg(struct sock *sk);
1516 
1517 /*
1518  * Functions to fill in entries in struct proto_ops when a protocol
1519  * does not implement a particular function.
1520  */
1521 extern int                      sock_no_bind(struct socket *,
1522 					     struct sockaddr *, int);
1523 extern int                      sock_no_connect(struct socket *,
1524 						struct sockaddr *, int, int);
1525 extern int                      sock_no_socketpair(struct socket *,
1526 						   struct socket *);
1527 extern int                      sock_no_accept(struct socket *,
1528 					       struct socket *, int);
1529 extern int                      sock_no_getname(struct socket *,
1530 						struct sockaddr *, int *, int);
1531 extern unsigned int             sock_no_poll(struct file *, struct socket *,
1532 					     struct poll_table_struct *);
1533 extern int                      sock_no_ioctl(struct socket *, unsigned int,
1534 					      unsigned long);
1535 extern int			sock_no_listen(struct socket *, int);
1536 extern int                      sock_no_shutdown(struct socket *, int);
1537 extern int			sock_no_getsockopt(struct socket *, int , int,
1538 						   char __user *, int __user *);
1539 extern int			sock_no_setsockopt(struct socket *, int, int,
1540 						   char __user *, unsigned int);
1541 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1542 						struct msghdr *, size_t);
1543 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1544 						struct msghdr *, size_t, int);
1545 extern int			sock_no_mmap(struct file *file,
1546 					     struct socket *sock,
1547 					     struct vm_area_struct *vma);
1548 extern ssize_t			sock_no_sendpage(struct socket *sock,
1549 						struct page *page,
1550 						int offset, size_t size,
1551 						int flags);
1552 
1553 /*
1554  * Functions to fill in entries in struct proto_ops when a protocol
1555  * uses the inet style.
1556  */
1557 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1558 				  char __user *optval, int __user *optlen);
1559 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1560 			       struct msghdr *msg, size_t size, int flags);
1561 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1562 				  char __user *optval, unsigned int optlen);
1563 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1564 		int optname, char __user *optval, int __user *optlen);
1565 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1566 		int optname, char __user *optval, unsigned int optlen);
1567 
1568 extern void sk_common_release(struct sock *sk);
1569 
1570 /*
1571  *	Default socket callbacks and setup code
1572  */
1573 
1574 /* Initialise core socket variables */
1575 extern void sock_init_data(struct socket *sock, struct sock *sk);
1576 
1577 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1578 
1579 /**
1580  *	sk_filter_release - release a socket filter
1581  *	@fp: filter to remove
1582  *
1583  *	Remove a filter from a socket and release its resources.
1584  */
1585 
1586 static inline void sk_filter_release(struct sk_filter *fp)
1587 {
1588 	if (atomic_dec_and_test(&fp->refcnt))
1589 		call_rcu(&fp->rcu, sk_filter_release_rcu);
1590 }
1591 
1592 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1593 {
1594 	unsigned int size = sk_filter_len(fp);
1595 
1596 	atomic_sub(size, &sk->sk_omem_alloc);
1597 	sk_filter_release(fp);
1598 }
1599 
1600 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1601 {
1602 	atomic_inc(&fp->refcnt);
1603 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1604 }
1605 
1606 /*
1607  * Socket reference counting postulates.
1608  *
1609  * * Each user of socket SHOULD hold a reference count.
1610  * * Each access point to socket (an hash table bucket, reference from a list,
1611  *   running timer, skb in flight MUST hold a reference count.
1612  * * When reference count hits 0, it means it will never increase back.
1613  * * When reference count hits 0, it means that no references from
1614  *   outside exist to this socket and current process on current CPU
1615  *   is last user and may/should destroy this socket.
1616  * * sk_free is called from any context: process, BH, IRQ. When
1617  *   it is called, socket has no references from outside -> sk_free
1618  *   may release descendant resources allocated by the socket, but
1619  *   to the time when it is called, socket is NOT referenced by any
1620  *   hash tables, lists etc.
1621  * * Packets, delivered from outside (from network or from another process)
1622  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1623  *   when they sit in queue. Otherwise, packets will leak to hole, when
1624  *   socket is looked up by one cpu and unhasing is made by another CPU.
1625  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1626  *   (leak to backlog). Packet socket does all the processing inside
1627  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1628  *   use separate SMP lock, so that they are prone too.
1629  */
1630 
1631 /* Ungrab socket and destroy it, if it was the last reference. */
1632 static inline void sock_put(struct sock *sk)
1633 {
1634 	if (atomic_dec_and_test(&sk->sk_refcnt))
1635 		sk_free(sk);
1636 }
1637 
1638 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1639 			  const int nested);
1640 
1641 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1642 {
1643 	sk->sk_tx_queue_mapping = tx_queue;
1644 }
1645 
1646 static inline void sk_tx_queue_clear(struct sock *sk)
1647 {
1648 	sk->sk_tx_queue_mapping = -1;
1649 }
1650 
1651 static inline int sk_tx_queue_get(const struct sock *sk)
1652 {
1653 	return sk ? sk->sk_tx_queue_mapping : -1;
1654 }
1655 
1656 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1657 {
1658 	sk_tx_queue_clear(sk);
1659 	sk->sk_socket = sock;
1660 }
1661 
1662 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1663 {
1664 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1665 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1666 }
1667 /* Detach socket from process context.
1668  * Announce socket dead, detach it from wait queue and inode.
1669  * Note that parent inode held reference count on this struct sock,
1670  * we do not release it in this function, because protocol
1671  * probably wants some additional cleanups or even continuing
1672  * to work with this socket (TCP).
1673  */
1674 static inline void sock_orphan(struct sock *sk)
1675 {
1676 	write_lock_bh(&sk->sk_callback_lock);
1677 	sock_set_flag(sk, SOCK_DEAD);
1678 	sk_set_socket(sk, NULL);
1679 	sk->sk_wq  = NULL;
1680 	write_unlock_bh(&sk->sk_callback_lock);
1681 }
1682 
1683 static inline void sock_graft(struct sock *sk, struct socket *parent)
1684 {
1685 	write_lock_bh(&sk->sk_callback_lock);
1686 	sk->sk_wq = parent->wq;
1687 	parent->sk = sk;
1688 	sk_set_socket(sk, parent);
1689 	security_sock_graft(sk, parent);
1690 	write_unlock_bh(&sk->sk_callback_lock);
1691 }
1692 
1693 extern kuid_t sock_i_uid(struct sock *sk);
1694 extern unsigned long sock_i_ino(struct sock *sk);
1695 
1696 static inline struct dst_entry *
1697 __sk_dst_get(struct sock *sk)
1698 {
1699 	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1700 						       lockdep_is_held(&sk->sk_lock.slock));
1701 }
1702 
1703 static inline struct dst_entry *
1704 sk_dst_get(struct sock *sk)
1705 {
1706 	struct dst_entry *dst;
1707 
1708 	rcu_read_lock();
1709 	dst = rcu_dereference(sk->sk_dst_cache);
1710 	if (dst)
1711 		dst_hold(dst);
1712 	rcu_read_unlock();
1713 	return dst;
1714 }
1715 
1716 extern void sk_reset_txq(struct sock *sk);
1717 
1718 static inline void dst_negative_advice(struct sock *sk)
1719 {
1720 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1721 
1722 	if (dst && dst->ops->negative_advice) {
1723 		ndst = dst->ops->negative_advice(dst);
1724 
1725 		if (ndst != dst) {
1726 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1727 			sk_reset_txq(sk);
1728 		}
1729 	}
1730 }
1731 
1732 static inline void
1733 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1734 {
1735 	struct dst_entry *old_dst;
1736 
1737 	sk_tx_queue_clear(sk);
1738 	/*
1739 	 * This can be called while sk is owned by the caller only,
1740 	 * with no state that can be checked in a rcu_dereference_check() cond
1741 	 */
1742 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1743 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1744 	dst_release(old_dst);
1745 }
1746 
1747 static inline void
1748 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1749 {
1750 	spin_lock(&sk->sk_dst_lock);
1751 	__sk_dst_set(sk, dst);
1752 	spin_unlock(&sk->sk_dst_lock);
1753 }
1754 
1755 static inline void
1756 __sk_dst_reset(struct sock *sk)
1757 {
1758 	__sk_dst_set(sk, NULL);
1759 }
1760 
1761 static inline void
1762 sk_dst_reset(struct sock *sk)
1763 {
1764 	spin_lock(&sk->sk_dst_lock);
1765 	__sk_dst_reset(sk);
1766 	spin_unlock(&sk->sk_dst_lock);
1767 }
1768 
1769 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1770 
1771 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1772 
1773 static inline bool sk_can_gso(const struct sock *sk)
1774 {
1775 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1776 }
1777 
1778 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1779 
1780 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1781 {
1782 	sk->sk_route_nocaps |= flags;
1783 	sk->sk_route_caps &= ~flags;
1784 }
1785 
1786 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1787 					   char __user *from, char *to,
1788 					   int copy, int offset)
1789 {
1790 	if (skb->ip_summed == CHECKSUM_NONE) {
1791 		int err = 0;
1792 		__wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1793 		if (err)
1794 			return err;
1795 		skb->csum = csum_block_add(skb->csum, csum, offset);
1796 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1797 		if (!access_ok(VERIFY_READ, from, copy) ||
1798 		    __copy_from_user_nocache(to, from, copy))
1799 			return -EFAULT;
1800 	} else if (copy_from_user(to, from, copy))
1801 		return -EFAULT;
1802 
1803 	return 0;
1804 }
1805 
1806 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1807 				       char __user *from, int copy)
1808 {
1809 	int err, offset = skb->len;
1810 
1811 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1812 				       copy, offset);
1813 	if (err)
1814 		__skb_trim(skb, offset);
1815 
1816 	return err;
1817 }
1818 
1819 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1820 					   struct sk_buff *skb,
1821 					   struct page *page,
1822 					   int off, int copy)
1823 {
1824 	int err;
1825 
1826 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1827 				       copy, skb->len);
1828 	if (err)
1829 		return err;
1830 
1831 	skb->len	     += copy;
1832 	skb->data_len	     += copy;
1833 	skb->truesize	     += copy;
1834 	sk->sk_wmem_queued   += copy;
1835 	sk_mem_charge(sk, copy);
1836 	return 0;
1837 }
1838 
1839 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1840 				   struct sk_buff *skb, struct page *page,
1841 				   int off, int copy)
1842 {
1843 	if (skb->ip_summed == CHECKSUM_NONE) {
1844 		int err = 0;
1845 		__wsum csum = csum_and_copy_from_user(from,
1846 						     page_address(page) + off,
1847 							    copy, 0, &err);
1848 		if (err)
1849 			return err;
1850 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1851 	} else if (copy_from_user(page_address(page) + off, from, copy))
1852 		return -EFAULT;
1853 
1854 	skb->len	     += copy;
1855 	skb->data_len	     += copy;
1856 	skb->truesize	     += copy;
1857 	sk->sk_wmem_queued   += copy;
1858 	sk_mem_charge(sk, copy);
1859 	return 0;
1860 }
1861 
1862 /**
1863  * sk_wmem_alloc_get - returns write allocations
1864  * @sk: socket
1865  *
1866  * Returns sk_wmem_alloc minus initial offset of one
1867  */
1868 static inline int sk_wmem_alloc_get(const struct sock *sk)
1869 {
1870 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1871 }
1872 
1873 /**
1874  * sk_rmem_alloc_get - returns read allocations
1875  * @sk: socket
1876  *
1877  * Returns sk_rmem_alloc
1878  */
1879 static inline int sk_rmem_alloc_get(const struct sock *sk)
1880 {
1881 	return atomic_read(&sk->sk_rmem_alloc);
1882 }
1883 
1884 /**
1885  * sk_has_allocations - check if allocations are outstanding
1886  * @sk: socket
1887  *
1888  * Returns true if socket has write or read allocations
1889  */
1890 static inline bool sk_has_allocations(const struct sock *sk)
1891 {
1892 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1893 }
1894 
1895 /**
1896  * wq_has_sleeper - check if there are any waiting processes
1897  * @wq: struct socket_wq
1898  *
1899  * Returns true if socket_wq has waiting processes
1900  *
1901  * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1902  * barrier call. They were added due to the race found within the tcp code.
1903  *
1904  * Consider following tcp code paths:
1905  *
1906  * CPU1                  CPU2
1907  *
1908  * sys_select            receive packet
1909  *   ...                 ...
1910  *   __add_wait_queue    update tp->rcv_nxt
1911  *   ...                 ...
1912  *   tp->rcv_nxt check   sock_def_readable
1913  *   ...                 {
1914  *   schedule               rcu_read_lock();
1915  *                          wq = rcu_dereference(sk->sk_wq);
1916  *                          if (wq && waitqueue_active(&wq->wait))
1917  *                              wake_up_interruptible(&wq->wait)
1918  *                          ...
1919  *                       }
1920  *
1921  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1922  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1923  * could then endup calling schedule and sleep forever if there are no more
1924  * data on the socket.
1925  *
1926  */
1927 static inline bool wq_has_sleeper(struct socket_wq *wq)
1928 {
1929 	/* We need to be sure we are in sync with the
1930 	 * add_wait_queue modifications to the wait queue.
1931 	 *
1932 	 * This memory barrier is paired in the sock_poll_wait.
1933 	 */
1934 	smp_mb();
1935 	return wq && waitqueue_active(&wq->wait);
1936 }
1937 
1938 /**
1939  * sock_poll_wait - place memory barrier behind the poll_wait call.
1940  * @filp:           file
1941  * @wait_address:   socket wait queue
1942  * @p:              poll_table
1943  *
1944  * See the comments in the wq_has_sleeper function.
1945  */
1946 static inline void sock_poll_wait(struct file *filp,
1947 		wait_queue_head_t *wait_address, poll_table *p)
1948 {
1949 	if (!poll_does_not_wait(p) && wait_address) {
1950 		poll_wait(filp, wait_address, p);
1951 		/* We need to be sure we are in sync with the
1952 		 * socket flags modification.
1953 		 *
1954 		 * This memory barrier is paired in the wq_has_sleeper.
1955 		 */
1956 		smp_mb();
1957 	}
1958 }
1959 
1960 /*
1961  *	Queue a received datagram if it will fit. Stream and sequenced
1962  *	protocols can't normally use this as they need to fit buffers in
1963  *	and play with them.
1964  *
1965  *	Inlined as it's very short and called for pretty much every
1966  *	packet ever received.
1967  */
1968 
1969 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1970 {
1971 	skb_orphan(skb);
1972 	skb->sk = sk;
1973 	skb->destructor = sock_wfree;
1974 	/*
1975 	 * We used to take a refcount on sk, but following operation
1976 	 * is enough to guarantee sk_free() wont free this sock until
1977 	 * all in-flight packets are completed
1978 	 */
1979 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1980 }
1981 
1982 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1983 {
1984 	skb_orphan(skb);
1985 	skb->sk = sk;
1986 	skb->destructor = sock_rfree;
1987 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1988 	sk_mem_charge(sk, skb->truesize);
1989 }
1990 
1991 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1992 			   unsigned long expires);
1993 
1994 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1995 
1996 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1997 
1998 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1999 
2000 /*
2001  *	Recover an error report and clear atomically
2002  */
2003 
2004 static inline int sock_error(struct sock *sk)
2005 {
2006 	int err;
2007 	if (likely(!sk->sk_err))
2008 		return 0;
2009 	err = xchg(&sk->sk_err, 0);
2010 	return -err;
2011 }
2012 
2013 static inline unsigned long sock_wspace(struct sock *sk)
2014 {
2015 	int amt = 0;
2016 
2017 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2018 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2019 		if (amt < 0)
2020 			amt = 0;
2021 	}
2022 	return amt;
2023 }
2024 
2025 static inline void sk_wake_async(struct sock *sk, int how, int band)
2026 {
2027 	if (sock_flag(sk, SOCK_FASYNC))
2028 		sock_wake_async(sk->sk_socket, how, band);
2029 }
2030 
2031 #define SOCK_MIN_SNDBUF 2048
2032 /*
2033  * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
2034  * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
2035  */
2036 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
2037 
2038 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2039 {
2040 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2041 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2042 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2043 	}
2044 }
2045 
2046 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2047 
2048 /**
2049  * sk_page_frag - return an appropriate page_frag
2050  * @sk: socket
2051  *
2052  * If socket allocation mode allows current thread to sleep, it means its
2053  * safe to use the per task page_frag instead of the per socket one.
2054  */
2055 static inline struct page_frag *sk_page_frag(struct sock *sk)
2056 {
2057 	if (sk->sk_allocation & __GFP_WAIT)
2058 		return &current->task_frag;
2059 
2060 	return &sk->sk_frag;
2061 }
2062 
2063 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2064 
2065 /*
2066  *	Default write policy as shown to user space via poll/select/SIGIO
2067  */
2068 static inline bool sock_writeable(const struct sock *sk)
2069 {
2070 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2071 }
2072 
2073 static inline gfp_t gfp_any(void)
2074 {
2075 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2076 }
2077 
2078 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2079 {
2080 	return noblock ? 0 : sk->sk_rcvtimeo;
2081 }
2082 
2083 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2084 {
2085 	return noblock ? 0 : sk->sk_sndtimeo;
2086 }
2087 
2088 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2089 {
2090 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2091 }
2092 
2093 /* Alas, with timeout socket operations are not restartable.
2094  * Compare this to poll().
2095  */
2096 static inline int sock_intr_errno(long timeo)
2097 {
2098 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2099 }
2100 
2101 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2102 	struct sk_buff *skb);
2103 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2104 	struct sk_buff *skb);
2105 
2106 static inline void
2107 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2108 {
2109 	ktime_t kt = skb->tstamp;
2110 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2111 
2112 	/*
2113 	 * generate control messages if
2114 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2115 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
2116 	 * - software time stamp available and wanted
2117 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
2118 	 * - hardware time stamps available and wanted
2119 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
2120 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
2121 	 */
2122 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2123 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2124 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2125 	    (hwtstamps->hwtstamp.tv64 &&
2126 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2127 	    (hwtstamps->syststamp.tv64 &&
2128 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2129 		__sock_recv_timestamp(msg, sk, skb);
2130 	else
2131 		sk->sk_stamp = kt;
2132 
2133 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2134 		__sock_recv_wifi_status(msg, sk, skb);
2135 }
2136 
2137 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2138 				     struct sk_buff *skb);
2139 
2140 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2141 					  struct sk_buff *skb)
2142 {
2143 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2144 			   (1UL << SOCK_RCVTSTAMP)			| \
2145 			   (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)	| \
2146 			   (1UL << SOCK_TIMESTAMPING_SOFTWARE)		| \
2147 			   (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE)	| \
2148 			   (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2149 
2150 	if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2151 		__sock_recv_ts_and_drops(msg, sk, skb);
2152 	else
2153 		sk->sk_stamp = skb->tstamp;
2154 }
2155 
2156 /**
2157  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2158  * @sk:		socket sending this packet
2159  * @tx_flags:	filled with instructions for time stamping
2160  *
2161  * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2162  * parameters are invalid.
2163  */
2164 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2165 
2166 /**
2167  * sk_eat_skb - Release a skb if it is no longer needed
2168  * @sk: socket to eat this skb from
2169  * @skb: socket buffer to eat
2170  * @copied_early: flag indicating whether DMA operations copied this data early
2171  *
2172  * This routine must be called with interrupts disabled or with the socket
2173  * locked so that the sk_buff queue operation is ok.
2174 */
2175 #ifdef CONFIG_NET_DMA
2176 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2177 {
2178 	__skb_unlink(skb, &sk->sk_receive_queue);
2179 	if (!copied_early)
2180 		__kfree_skb(skb);
2181 	else
2182 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
2183 }
2184 #else
2185 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2186 {
2187 	__skb_unlink(skb, &sk->sk_receive_queue);
2188 	__kfree_skb(skb);
2189 }
2190 #endif
2191 
2192 static inline
2193 struct net *sock_net(const struct sock *sk)
2194 {
2195 	return read_pnet(&sk->sk_net);
2196 }
2197 
2198 static inline
2199 void sock_net_set(struct sock *sk, struct net *net)
2200 {
2201 	write_pnet(&sk->sk_net, net);
2202 }
2203 
2204 /*
2205  * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2206  * They should not hold a reference to a namespace in order to allow
2207  * to stop it.
2208  * Sockets after sk_change_net should be released using sk_release_kernel
2209  */
2210 static inline void sk_change_net(struct sock *sk, struct net *net)
2211 {
2212 	put_net(sock_net(sk));
2213 	sock_net_set(sk, hold_net(net));
2214 }
2215 
2216 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2217 {
2218 	if (skb->sk) {
2219 		struct sock *sk = skb->sk;
2220 
2221 		skb->destructor = NULL;
2222 		skb->sk = NULL;
2223 		return sk;
2224 	}
2225 	return NULL;
2226 }
2227 
2228 extern void sock_enable_timestamp(struct sock *sk, int flag);
2229 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2230 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2231 
2232 /*
2233  *	Enable debug/info messages
2234  */
2235 extern int net_msg_warn;
2236 #define NETDEBUG(fmt, args...) \
2237 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
2238 
2239 #define LIMIT_NETDEBUG(fmt, args...) \
2240 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2241 
2242 extern __u32 sysctl_wmem_max;
2243 extern __u32 sysctl_rmem_max;
2244 
2245 extern int sysctl_optmem_max;
2246 
2247 extern __u32 sysctl_wmem_default;
2248 extern __u32 sysctl_rmem_default;
2249 
2250 #endif	/* _SOCK_H */
2251