xref: /openbmc/linux/include/net/sock.h (revision f35e839a)
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 	SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
671 };
672 
673 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
674 {
675 	nsk->sk_flags = osk->sk_flags;
676 }
677 
678 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
679 {
680 	__set_bit(flag, &sk->sk_flags);
681 }
682 
683 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
684 {
685 	__clear_bit(flag, &sk->sk_flags);
686 }
687 
688 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
689 {
690 	return test_bit(flag, &sk->sk_flags);
691 }
692 
693 #ifdef CONFIG_NET
694 extern struct static_key memalloc_socks;
695 static inline int sk_memalloc_socks(void)
696 {
697 	return static_key_false(&memalloc_socks);
698 }
699 #else
700 
701 static inline int sk_memalloc_socks(void)
702 {
703 	return 0;
704 }
705 
706 #endif
707 
708 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
709 {
710 	return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
711 }
712 
713 static inline void sk_acceptq_removed(struct sock *sk)
714 {
715 	sk->sk_ack_backlog--;
716 }
717 
718 static inline void sk_acceptq_added(struct sock *sk)
719 {
720 	sk->sk_ack_backlog++;
721 }
722 
723 static inline bool sk_acceptq_is_full(const struct sock *sk)
724 {
725 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
726 }
727 
728 /*
729  * Compute minimal free write space needed to queue new packets.
730  */
731 static inline int sk_stream_min_wspace(const struct sock *sk)
732 {
733 	return sk->sk_wmem_queued >> 1;
734 }
735 
736 static inline int sk_stream_wspace(const struct sock *sk)
737 {
738 	return sk->sk_sndbuf - sk->sk_wmem_queued;
739 }
740 
741 extern void sk_stream_write_space(struct sock *sk);
742 
743 static inline bool sk_stream_memory_free(const struct sock *sk)
744 {
745 	return sk->sk_wmem_queued < sk->sk_sndbuf;
746 }
747 
748 /* OOB backlog add */
749 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
750 {
751 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
752 	skb_dst_force(skb);
753 
754 	if (!sk->sk_backlog.tail)
755 		sk->sk_backlog.head = skb;
756 	else
757 		sk->sk_backlog.tail->next = skb;
758 
759 	sk->sk_backlog.tail = skb;
760 	skb->next = NULL;
761 }
762 
763 /*
764  * Take into account size of receive queue and backlog queue
765  * Do not take into account this skb truesize,
766  * to allow even a single big packet to come.
767  */
768 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
769 				     unsigned int limit)
770 {
771 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
772 
773 	return qsize > limit;
774 }
775 
776 /* The per-socket spinlock must be held here. */
777 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
778 					      unsigned int limit)
779 {
780 	if (sk_rcvqueues_full(sk, skb, limit))
781 		return -ENOBUFS;
782 
783 	__sk_add_backlog(sk, skb);
784 	sk->sk_backlog.len += skb->truesize;
785 	return 0;
786 }
787 
788 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
789 
790 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
791 {
792 	if (sk_memalloc_socks() && skb_pfmemalloc(skb))
793 		return __sk_backlog_rcv(sk, skb);
794 
795 	return sk->sk_backlog_rcv(sk, skb);
796 }
797 
798 static inline void sock_rps_record_flow(const struct sock *sk)
799 {
800 #ifdef CONFIG_RPS
801 	struct rps_sock_flow_table *sock_flow_table;
802 
803 	rcu_read_lock();
804 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
805 	rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
806 	rcu_read_unlock();
807 #endif
808 }
809 
810 static inline void sock_rps_reset_flow(const struct sock *sk)
811 {
812 #ifdef CONFIG_RPS
813 	struct rps_sock_flow_table *sock_flow_table;
814 
815 	rcu_read_lock();
816 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
817 	rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
818 	rcu_read_unlock();
819 #endif
820 }
821 
822 static inline void sock_rps_save_rxhash(struct sock *sk,
823 					const struct sk_buff *skb)
824 {
825 #ifdef CONFIG_RPS
826 	if (unlikely(sk->sk_rxhash != skb->rxhash)) {
827 		sock_rps_reset_flow(sk);
828 		sk->sk_rxhash = skb->rxhash;
829 	}
830 #endif
831 }
832 
833 static inline void sock_rps_reset_rxhash(struct sock *sk)
834 {
835 #ifdef CONFIG_RPS
836 	sock_rps_reset_flow(sk);
837 	sk->sk_rxhash = 0;
838 #endif
839 }
840 
841 #define sk_wait_event(__sk, __timeo, __condition)			\
842 	({	int __rc;						\
843 		release_sock(__sk);					\
844 		__rc = __condition;					\
845 		if (!__rc) {						\
846 			*(__timeo) = schedule_timeout(*(__timeo));	\
847 		}							\
848 		lock_sock(__sk);					\
849 		__rc = __condition;					\
850 		__rc;							\
851 	})
852 
853 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
854 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
855 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
856 extern int sk_stream_error(struct sock *sk, int flags, int err);
857 extern void sk_stream_kill_queues(struct sock *sk);
858 extern void sk_set_memalloc(struct sock *sk);
859 extern void sk_clear_memalloc(struct sock *sk);
860 
861 extern int sk_wait_data(struct sock *sk, long *timeo);
862 
863 struct request_sock_ops;
864 struct timewait_sock_ops;
865 struct inet_hashinfo;
866 struct raw_hashinfo;
867 struct module;
868 
869 /* Networking protocol blocks we attach to sockets.
870  * socket layer -> transport layer interface
871  * transport -> network interface is defined by struct inet_proto
872  */
873 struct proto {
874 	void			(*close)(struct sock *sk,
875 					long timeout);
876 	int			(*connect)(struct sock *sk,
877 					struct sockaddr *uaddr,
878 					int addr_len);
879 	int			(*disconnect)(struct sock *sk, int flags);
880 
881 	struct sock *		(*accept)(struct sock *sk, int flags, int *err);
882 
883 	int			(*ioctl)(struct sock *sk, int cmd,
884 					 unsigned long arg);
885 	int			(*init)(struct sock *sk);
886 	void			(*destroy)(struct sock *sk);
887 	void			(*shutdown)(struct sock *sk, int how);
888 	int			(*setsockopt)(struct sock *sk, int level,
889 					int optname, char __user *optval,
890 					unsigned int optlen);
891 	int			(*getsockopt)(struct sock *sk, int level,
892 					int optname, char __user *optval,
893 					int __user *option);
894 #ifdef CONFIG_COMPAT
895 	int			(*compat_setsockopt)(struct sock *sk,
896 					int level,
897 					int optname, char __user *optval,
898 					unsigned int optlen);
899 	int			(*compat_getsockopt)(struct sock *sk,
900 					int level,
901 					int optname, char __user *optval,
902 					int __user *option);
903 	int			(*compat_ioctl)(struct sock *sk,
904 					unsigned int cmd, unsigned long arg);
905 #endif
906 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
907 					   struct msghdr *msg, size_t len);
908 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
909 					   struct msghdr *msg,
910 					   size_t len, int noblock, int flags,
911 					   int *addr_len);
912 	int			(*sendpage)(struct sock *sk, struct page *page,
913 					int offset, size_t size, int flags);
914 	int			(*bind)(struct sock *sk,
915 					struct sockaddr *uaddr, int addr_len);
916 
917 	int			(*backlog_rcv) (struct sock *sk,
918 						struct sk_buff *skb);
919 
920 	void		(*release_cb)(struct sock *sk);
921 	void		(*mtu_reduced)(struct sock *sk);
922 
923 	/* Keeping track of sk's, looking them up, and port selection methods. */
924 	void			(*hash)(struct sock *sk);
925 	void			(*unhash)(struct sock *sk);
926 	void			(*rehash)(struct sock *sk);
927 	int			(*get_port)(struct sock *sk, unsigned short snum);
928 	void			(*clear_sk)(struct sock *sk, int size);
929 
930 	/* Keeping track of sockets in use */
931 #ifdef CONFIG_PROC_FS
932 	unsigned int		inuse_idx;
933 #endif
934 
935 	/* Memory pressure */
936 	void			(*enter_memory_pressure)(struct sock *sk);
937 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
938 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
939 	/*
940 	 * Pressure flag: try to collapse.
941 	 * Technical note: it is used by multiple contexts non atomically.
942 	 * All the __sk_mem_schedule() is of this nature: accounting
943 	 * is strict, actions are advisory and have some latency.
944 	 */
945 	int			*memory_pressure;
946 	long			*sysctl_mem;
947 	int			*sysctl_wmem;
948 	int			*sysctl_rmem;
949 	int			max_header;
950 	bool			no_autobind;
951 
952 	struct kmem_cache	*slab;
953 	unsigned int		obj_size;
954 	int			slab_flags;
955 
956 	struct percpu_counter	*orphan_count;
957 
958 	struct request_sock_ops	*rsk_prot;
959 	struct timewait_sock_ops *twsk_prot;
960 
961 	union {
962 		struct inet_hashinfo	*hashinfo;
963 		struct udp_table	*udp_table;
964 		struct raw_hashinfo	*raw_hash;
965 	} h;
966 
967 	struct module		*owner;
968 
969 	char			name[32];
970 
971 	struct list_head	node;
972 #ifdef SOCK_REFCNT_DEBUG
973 	atomic_t		socks;
974 #endif
975 #ifdef CONFIG_MEMCG_KMEM
976 	/*
977 	 * cgroup specific init/deinit functions. Called once for all
978 	 * protocols that implement it, from cgroups populate function.
979 	 * This function has to setup any files the protocol want to
980 	 * appear in the kmem cgroup filesystem.
981 	 */
982 	int			(*init_cgroup)(struct mem_cgroup *memcg,
983 					       struct cgroup_subsys *ss);
984 	void			(*destroy_cgroup)(struct mem_cgroup *memcg);
985 	struct cg_proto		*(*proto_cgroup)(struct mem_cgroup *memcg);
986 #endif
987 };
988 
989 /*
990  * Bits in struct cg_proto.flags
991  */
992 enum cg_proto_flags {
993 	/* Currently active and new sockets should be assigned to cgroups */
994 	MEMCG_SOCK_ACTIVE,
995 	/* It was ever activated; we must disarm static keys on destruction */
996 	MEMCG_SOCK_ACTIVATED,
997 };
998 
999 struct cg_proto {
1000 	void			(*enter_memory_pressure)(struct sock *sk);
1001 	struct res_counter	*memory_allocated;	/* Current allocated memory. */
1002 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
1003 	int			*memory_pressure;
1004 	long			*sysctl_mem;
1005 	unsigned long		flags;
1006 	/*
1007 	 * memcg field is used to find which memcg we belong directly
1008 	 * Each memcg struct can hold more than one cg_proto, so container_of
1009 	 * won't really cut.
1010 	 *
1011 	 * The elegant solution would be having an inverse function to
1012 	 * proto_cgroup in struct proto, but that means polluting the structure
1013 	 * for everybody, instead of just for memcg users.
1014 	 */
1015 	struct mem_cgroup	*memcg;
1016 };
1017 
1018 extern int proto_register(struct proto *prot, int alloc_slab);
1019 extern void proto_unregister(struct proto *prot);
1020 
1021 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1022 {
1023 	return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1024 }
1025 
1026 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1027 {
1028 	return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1029 }
1030 
1031 #ifdef SOCK_REFCNT_DEBUG
1032 static inline void sk_refcnt_debug_inc(struct sock *sk)
1033 {
1034 	atomic_inc(&sk->sk_prot->socks);
1035 }
1036 
1037 static inline void sk_refcnt_debug_dec(struct sock *sk)
1038 {
1039 	atomic_dec(&sk->sk_prot->socks);
1040 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1041 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1042 }
1043 
1044 static inline void sk_refcnt_debug_release(const struct sock *sk)
1045 {
1046 	if (atomic_read(&sk->sk_refcnt) != 1)
1047 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1048 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1049 }
1050 #else /* SOCK_REFCNT_DEBUG */
1051 #define sk_refcnt_debug_inc(sk) do { } while (0)
1052 #define sk_refcnt_debug_dec(sk) do { } while (0)
1053 #define sk_refcnt_debug_release(sk) do { } while (0)
1054 #endif /* SOCK_REFCNT_DEBUG */
1055 
1056 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1057 extern struct static_key memcg_socket_limit_enabled;
1058 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1059 					       struct cg_proto *cg_proto)
1060 {
1061 	return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1062 }
1063 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1064 #else
1065 #define mem_cgroup_sockets_enabled 0
1066 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1067 					       struct cg_proto *cg_proto)
1068 {
1069 	return NULL;
1070 }
1071 #endif
1072 
1073 
1074 static inline bool sk_has_memory_pressure(const struct sock *sk)
1075 {
1076 	return sk->sk_prot->memory_pressure != NULL;
1077 }
1078 
1079 static inline bool sk_under_memory_pressure(const struct sock *sk)
1080 {
1081 	if (!sk->sk_prot->memory_pressure)
1082 		return false;
1083 
1084 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1085 		return !!*sk->sk_cgrp->memory_pressure;
1086 
1087 	return !!*sk->sk_prot->memory_pressure;
1088 }
1089 
1090 static inline void sk_leave_memory_pressure(struct sock *sk)
1091 {
1092 	int *memory_pressure = sk->sk_prot->memory_pressure;
1093 
1094 	if (!memory_pressure)
1095 		return;
1096 
1097 	if (*memory_pressure)
1098 		*memory_pressure = 0;
1099 
1100 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1101 		struct cg_proto *cg_proto = sk->sk_cgrp;
1102 		struct proto *prot = sk->sk_prot;
1103 
1104 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1105 			if (*cg_proto->memory_pressure)
1106 				*cg_proto->memory_pressure = 0;
1107 	}
1108 
1109 }
1110 
1111 static inline void sk_enter_memory_pressure(struct sock *sk)
1112 {
1113 	if (!sk->sk_prot->enter_memory_pressure)
1114 		return;
1115 
1116 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1117 		struct cg_proto *cg_proto = sk->sk_cgrp;
1118 		struct proto *prot = sk->sk_prot;
1119 
1120 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1121 			cg_proto->enter_memory_pressure(sk);
1122 	}
1123 
1124 	sk->sk_prot->enter_memory_pressure(sk);
1125 }
1126 
1127 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1128 {
1129 	long *prot = sk->sk_prot->sysctl_mem;
1130 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1131 		prot = sk->sk_cgrp->sysctl_mem;
1132 	return prot[index];
1133 }
1134 
1135 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1136 					      unsigned long amt,
1137 					      int *parent_status)
1138 {
1139 	struct res_counter *fail;
1140 	int ret;
1141 
1142 	ret = res_counter_charge_nofail(prot->memory_allocated,
1143 					amt << PAGE_SHIFT, &fail);
1144 	if (ret < 0)
1145 		*parent_status = OVER_LIMIT;
1146 }
1147 
1148 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1149 					      unsigned long amt)
1150 {
1151 	res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1152 }
1153 
1154 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1155 {
1156 	u64 ret;
1157 	ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1158 	return ret >> PAGE_SHIFT;
1159 }
1160 
1161 static inline long
1162 sk_memory_allocated(const struct sock *sk)
1163 {
1164 	struct proto *prot = sk->sk_prot;
1165 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1166 		return memcg_memory_allocated_read(sk->sk_cgrp);
1167 
1168 	return atomic_long_read(prot->memory_allocated);
1169 }
1170 
1171 static inline long
1172 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1173 {
1174 	struct proto *prot = sk->sk_prot;
1175 
1176 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1177 		memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1178 		/* update the root cgroup regardless */
1179 		atomic_long_add_return(amt, prot->memory_allocated);
1180 		return memcg_memory_allocated_read(sk->sk_cgrp);
1181 	}
1182 
1183 	return atomic_long_add_return(amt, prot->memory_allocated);
1184 }
1185 
1186 static inline void
1187 sk_memory_allocated_sub(struct sock *sk, int amt)
1188 {
1189 	struct proto *prot = sk->sk_prot;
1190 
1191 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1192 		memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1193 
1194 	atomic_long_sub(amt, prot->memory_allocated);
1195 }
1196 
1197 static inline void sk_sockets_allocated_dec(struct sock *sk)
1198 {
1199 	struct proto *prot = sk->sk_prot;
1200 
1201 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1202 		struct cg_proto *cg_proto = sk->sk_cgrp;
1203 
1204 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1205 			percpu_counter_dec(cg_proto->sockets_allocated);
1206 	}
1207 
1208 	percpu_counter_dec(prot->sockets_allocated);
1209 }
1210 
1211 static inline void sk_sockets_allocated_inc(struct sock *sk)
1212 {
1213 	struct proto *prot = sk->sk_prot;
1214 
1215 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1216 		struct cg_proto *cg_proto = sk->sk_cgrp;
1217 
1218 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1219 			percpu_counter_inc(cg_proto->sockets_allocated);
1220 	}
1221 
1222 	percpu_counter_inc(prot->sockets_allocated);
1223 }
1224 
1225 static inline int
1226 sk_sockets_allocated_read_positive(struct sock *sk)
1227 {
1228 	struct proto *prot = sk->sk_prot;
1229 
1230 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1231 		return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1232 
1233 	return percpu_counter_read_positive(prot->sockets_allocated);
1234 }
1235 
1236 static inline int
1237 proto_sockets_allocated_sum_positive(struct proto *prot)
1238 {
1239 	return percpu_counter_sum_positive(prot->sockets_allocated);
1240 }
1241 
1242 static inline long
1243 proto_memory_allocated(struct proto *prot)
1244 {
1245 	return atomic_long_read(prot->memory_allocated);
1246 }
1247 
1248 static inline bool
1249 proto_memory_pressure(struct proto *prot)
1250 {
1251 	if (!prot->memory_pressure)
1252 		return false;
1253 	return !!*prot->memory_pressure;
1254 }
1255 
1256 
1257 #ifdef CONFIG_PROC_FS
1258 /* Called with local bh disabled */
1259 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1260 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1261 #else
1262 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1263 		int inc)
1264 {
1265 }
1266 #endif
1267 
1268 
1269 /* With per-bucket locks this operation is not-atomic, so that
1270  * this version is not worse.
1271  */
1272 static inline void __sk_prot_rehash(struct sock *sk)
1273 {
1274 	sk->sk_prot->unhash(sk);
1275 	sk->sk_prot->hash(sk);
1276 }
1277 
1278 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1279 
1280 /* About 10 seconds */
1281 #define SOCK_DESTROY_TIME (10*HZ)
1282 
1283 /* Sockets 0-1023 can't be bound to unless you are superuser */
1284 #define PROT_SOCK	1024
1285 
1286 #define SHUTDOWN_MASK	3
1287 #define RCV_SHUTDOWN	1
1288 #define SEND_SHUTDOWN	2
1289 
1290 #define SOCK_SNDBUF_LOCK	1
1291 #define SOCK_RCVBUF_LOCK	2
1292 #define SOCK_BINDADDR_LOCK	4
1293 #define SOCK_BINDPORT_LOCK	8
1294 
1295 /* sock_iocb: used to kick off async processing of socket ios */
1296 struct sock_iocb {
1297 	struct list_head	list;
1298 
1299 	int			flags;
1300 	int			size;
1301 	struct socket		*sock;
1302 	struct sock		*sk;
1303 	struct scm_cookie	*scm;
1304 	struct msghdr		*msg, async_msg;
1305 	struct kiocb		*kiocb;
1306 };
1307 
1308 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1309 {
1310 	return (struct sock_iocb *)iocb->private;
1311 }
1312 
1313 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1314 {
1315 	return si->kiocb;
1316 }
1317 
1318 struct socket_alloc {
1319 	struct socket socket;
1320 	struct inode vfs_inode;
1321 };
1322 
1323 static inline struct socket *SOCKET_I(struct inode *inode)
1324 {
1325 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1326 }
1327 
1328 static inline struct inode *SOCK_INODE(struct socket *socket)
1329 {
1330 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1331 }
1332 
1333 /*
1334  * Functions for memory accounting
1335  */
1336 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1337 extern void __sk_mem_reclaim(struct sock *sk);
1338 
1339 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1340 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1341 #define SK_MEM_SEND	0
1342 #define SK_MEM_RECV	1
1343 
1344 static inline int sk_mem_pages(int amt)
1345 {
1346 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1347 }
1348 
1349 static inline bool sk_has_account(struct sock *sk)
1350 {
1351 	/* return true if protocol supports memory accounting */
1352 	return !!sk->sk_prot->memory_allocated;
1353 }
1354 
1355 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1356 {
1357 	if (!sk_has_account(sk))
1358 		return true;
1359 	return size <= sk->sk_forward_alloc ||
1360 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1361 }
1362 
1363 static inline bool
1364 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1365 {
1366 	if (!sk_has_account(sk))
1367 		return true;
1368 	return size<= sk->sk_forward_alloc ||
1369 		__sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1370 		skb_pfmemalloc(skb);
1371 }
1372 
1373 static inline void sk_mem_reclaim(struct sock *sk)
1374 {
1375 	if (!sk_has_account(sk))
1376 		return;
1377 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1378 		__sk_mem_reclaim(sk);
1379 }
1380 
1381 static inline void sk_mem_reclaim_partial(struct sock *sk)
1382 {
1383 	if (!sk_has_account(sk))
1384 		return;
1385 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1386 		__sk_mem_reclaim(sk);
1387 }
1388 
1389 static inline void sk_mem_charge(struct sock *sk, int size)
1390 {
1391 	if (!sk_has_account(sk))
1392 		return;
1393 	sk->sk_forward_alloc -= size;
1394 }
1395 
1396 static inline void sk_mem_uncharge(struct sock *sk, int size)
1397 {
1398 	if (!sk_has_account(sk))
1399 		return;
1400 	sk->sk_forward_alloc += size;
1401 }
1402 
1403 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1404 {
1405 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1406 	sk->sk_wmem_queued -= skb->truesize;
1407 	sk_mem_uncharge(sk, skb->truesize);
1408 	__kfree_skb(skb);
1409 }
1410 
1411 /* Used by processes to "lock" a socket state, so that
1412  * interrupts and bottom half handlers won't change it
1413  * from under us. It essentially blocks any incoming
1414  * packets, so that we won't get any new data or any
1415  * packets that change the state of the socket.
1416  *
1417  * While locked, BH processing will add new packets to
1418  * the backlog queue.  This queue is processed by the
1419  * owner of the socket lock right before it is released.
1420  *
1421  * Since ~2.3.5 it is also exclusive sleep lock serializing
1422  * accesses from user process context.
1423  */
1424 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1425 
1426 /*
1427  * Macro so as to not evaluate some arguments when
1428  * lockdep is not enabled.
1429  *
1430  * Mark both the sk_lock and the sk_lock.slock as a
1431  * per-address-family lock class.
1432  */
1433 #define sock_lock_init_class_and_name(sk, sname, skey, name, key)	\
1434 do {									\
1435 	sk->sk_lock.owned = 0;						\
1436 	init_waitqueue_head(&sk->sk_lock.wq);				\
1437 	spin_lock_init(&(sk)->sk_lock.slock);				\
1438 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1439 			sizeof((sk)->sk_lock));				\
1440 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1441 				(skey), (sname));				\
1442 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1443 } while (0)
1444 
1445 extern void lock_sock_nested(struct sock *sk, int subclass);
1446 
1447 static inline void lock_sock(struct sock *sk)
1448 {
1449 	lock_sock_nested(sk, 0);
1450 }
1451 
1452 extern void release_sock(struct sock *sk);
1453 
1454 /* BH context may only use the following locking interface. */
1455 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1456 #define bh_lock_sock_nested(__sk) \
1457 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1458 				SINGLE_DEPTH_NESTING)
1459 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1460 
1461 extern bool lock_sock_fast(struct sock *sk);
1462 /**
1463  * unlock_sock_fast - complement of lock_sock_fast
1464  * @sk: socket
1465  * @slow: slow mode
1466  *
1467  * fast unlock socket for user context.
1468  * If slow mode is on, we call regular release_sock()
1469  */
1470 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1471 {
1472 	if (slow)
1473 		release_sock(sk);
1474 	else
1475 		spin_unlock_bh(&sk->sk_lock.slock);
1476 }
1477 
1478 
1479 extern struct sock		*sk_alloc(struct net *net, int family,
1480 					  gfp_t priority,
1481 					  struct proto *prot);
1482 extern void			sk_free(struct sock *sk);
1483 extern void			sk_release_kernel(struct sock *sk);
1484 extern struct sock		*sk_clone_lock(const struct sock *sk,
1485 					       const gfp_t priority);
1486 
1487 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
1488 					      unsigned long size, int force,
1489 					      gfp_t priority);
1490 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
1491 					      unsigned long size, int force,
1492 					      gfp_t priority);
1493 extern void			sock_wfree(struct sk_buff *skb);
1494 extern void			sock_rfree(struct sk_buff *skb);
1495 extern void			sock_edemux(struct sk_buff *skb);
1496 
1497 extern int			sock_setsockopt(struct socket *sock, int level,
1498 						int op, char __user *optval,
1499 						unsigned int optlen);
1500 
1501 extern int			sock_getsockopt(struct socket *sock, int level,
1502 						int op, char __user *optval,
1503 						int __user *optlen);
1504 extern struct sk_buff		*sock_alloc_send_skb(struct sock *sk,
1505 						     unsigned long size,
1506 						     int noblock,
1507 						     int *errcode);
1508 extern struct sk_buff		*sock_alloc_send_pskb(struct sock *sk,
1509 						      unsigned long header_len,
1510 						      unsigned long data_len,
1511 						      int noblock,
1512 						      int *errcode);
1513 extern void *sock_kmalloc(struct sock *sk, int size,
1514 			  gfp_t priority);
1515 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1516 extern void sk_send_sigurg(struct sock *sk);
1517 
1518 /*
1519  * Functions to fill in entries in struct proto_ops when a protocol
1520  * does not implement a particular function.
1521  */
1522 extern int                      sock_no_bind(struct socket *,
1523 					     struct sockaddr *, int);
1524 extern int                      sock_no_connect(struct socket *,
1525 						struct sockaddr *, int, int);
1526 extern int                      sock_no_socketpair(struct socket *,
1527 						   struct socket *);
1528 extern int                      sock_no_accept(struct socket *,
1529 					       struct socket *, int);
1530 extern int                      sock_no_getname(struct socket *,
1531 						struct sockaddr *, int *, int);
1532 extern unsigned int             sock_no_poll(struct file *, struct socket *,
1533 					     struct poll_table_struct *);
1534 extern int                      sock_no_ioctl(struct socket *, unsigned int,
1535 					      unsigned long);
1536 extern int			sock_no_listen(struct socket *, int);
1537 extern int                      sock_no_shutdown(struct socket *, int);
1538 extern int			sock_no_getsockopt(struct socket *, int , int,
1539 						   char __user *, int __user *);
1540 extern int			sock_no_setsockopt(struct socket *, int, int,
1541 						   char __user *, unsigned int);
1542 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1543 						struct msghdr *, size_t);
1544 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1545 						struct msghdr *, size_t, int);
1546 extern int			sock_no_mmap(struct file *file,
1547 					     struct socket *sock,
1548 					     struct vm_area_struct *vma);
1549 extern ssize_t			sock_no_sendpage(struct socket *sock,
1550 						struct page *page,
1551 						int offset, size_t size,
1552 						int flags);
1553 
1554 /*
1555  * Functions to fill in entries in struct proto_ops when a protocol
1556  * uses the inet style.
1557  */
1558 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1559 				  char __user *optval, int __user *optlen);
1560 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1561 			       struct msghdr *msg, size_t size, int flags);
1562 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1563 				  char __user *optval, unsigned int optlen);
1564 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1565 		int optname, char __user *optval, int __user *optlen);
1566 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1567 		int optname, char __user *optval, unsigned int optlen);
1568 
1569 extern void sk_common_release(struct sock *sk);
1570 
1571 /*
1572  *	Default socket callbacks and setup code
1573  */
1574 
1575 /* Initialise core socket variables */
1576 extern void sock_init_data(struct socket *sock, struct sock *sk);
1577 
1578 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1579 
1580 /**
1581  *	sk_filter_release - release a socket filter
1582  *	@fp: filter to remove
1583  *
1584  *	Remove a filter from a socket and release its resources.
1585  */
1586 
1587 static inline void sk_filter_release(struct sk_filter *fp)
1588 {
1589 	if (atomic_dec_and_test(&fp->refcnt))
1590 		call_rcu(&fp->rcu, sk_filter_release_rcu);
1591 }
1592 
1593 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1594 {
1595 	unsigned int size = sk_filter_len(fp);
1596 
1597 	atomic_sub(size, &sk->sk_omem_alloc);
1598 	sk_filter_release(fp);
1599 }
1600 
1601 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1602 {
1603 	atomic_inc(&fp->refcnt);
1604 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1605 }
1606 
1607 /*
1608  * Socket reference counting postulates.
1609  *
1610  * * Each user of socket SHOULD hold a reference count.
1611  * * Each access point to socket (an hash table bucket, reference from a list,
1612  *   running timer, skb in flight MUST hold a reference count.
1613  * * When reference count hits 0, it means it will never increase back.
1614  * * When reference count hits 0, it means that no references from
1615  *   outside exist to this socket and current process on current CPU
1616  *   is last user and may/should destroy this socket.
1617  * * sk_free is called from any context: process, BH, IRQ. When
1618  *   it is called, socket has no references from outside -> sk_free
1619  *   may release descendant resources allocated by the socket, but
1620  *   to the time when it is called, socket is NOT referenced by any
1621  *   hash tables, lists etc.
1622  * * Packets, delivered from outside (from network or from another process)
1623  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1624  *   when they sit in queue. Otherwise, packets will leak to hole, when
1625  *   socket is looked up by one cpu and unhasing is made by another CPU.
1626  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1627  *   (leak to backlog). Packet socket does all the processing inside
1628  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1629  *   use separate SMP lock, so that they are prone too.
1630  */
1631 
1632 /* Ungrab socket and destroy it, if it was the last reference. */
1633 static inline void sock_put(struct sock *sk)
1634 {
1635 	if (atomic_dec_and_test(&sk->sk_refcnt))
1636 		sk_free(sk);
1637 }
1638 
1639 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1640 			  const int nested);
1641 
1642 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1643 {
1644 	sk->sk_tx_queue_mapping = tx_queue;
1645 }
1646 
1647 static inline void sk_tx_queue_clear(struct sock *sk)
1648 {
1649 	sk->sk_tx_queue_mapping = -1;
1650 }
1651 
1652 static inline int sk_tx_queue_get(const struct sock *sk)
1653 {
1654 	return sk ? sk->sk_tx_queue_mapping : -1;
1655 }
1656 
1657 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1658 {
1659 	sk_tx_queue_clear(sk);
1660 	sk->sk_socket = sock;
1661 }
1662 
1663 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1664 {
1665 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1666 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1667 }
1668 /* Detach socket from process context.
1669  * Announce socket dead, detach it from wait queue and inode.
1670  * Note that parent inode held reference count on this struct sock,
1671  * we do not release it in this function, because protocol
1672  * probably wants some additional cleanups or even continuing
1673  * to work with this socket (TCP).
1674  */
1675 static inline void sock_orphan(struct sock *sk)
1676 {
1677 	write_lock_bh(&sk->sk_callback_lock);
1678 	sock_set_flag(sk, SOCK_DEAD);
1679 	sk_set_socket(sk, NULL);
1680 	sk->sk_wq  = NULL;
1681 	write_unlock_bh(&sk->sk_callback_lock);
1682 }
1683 
1684 static inline void sock_graft(struct sock *sk, struct socket *parent)
1685 {
1686 	write_lock_bh(&sk->sk_callback_lock);
1687 	sk->sk_wq = parent->wq;
1688 	parent->sk = sk;
1689 	sk_set_socket(sk, parent);
1690 	security_sock_graft(sk, parent);
1691 	write_unlock_bh(&sk->sk_callback_lock);
1692 }
1693 
1694 extern kuid_t sock_i_uid(struct sock *sk);
1695 extern unsigned long sock_i_ino(struct sock *sk);
1696 
1697 static inline struct dst_entry *
1698 __sk_dst_get(struct sock *sk)
1699 {
1700 	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1701 						       lockdep_is_held(&sk->sk_lock.slock));
1702 }
1703 
1704 static inline struct dst_entry *
1705 sk_dst_get(struct sock *sk)
1706 {
1707 	struct dst_entry *dst;
1708 
1709 	rcu_read_lock();
1710 	dst = rcu_dereference(sk->sk_dst_cache);
1711 	if (dst)
1712 		dst_hold(dst);
1713 	rcu_read_unlock();
1714 	return dst;
1715 }
1716 
1717 extern void sk_reset_txq(struct sock *sk);
1718 
1719 static inline void dst_negative_advice(struct sock *sk)
1720 {
1721 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1722 
1723 	if (dst && dst->ops->negative_advice) {
1724 		ndst = dst->ops->negative_advice(dst);
1725 
1726 		if (ndst != dst) {
1727 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1728 			sk_reset_txq(sk);
1729 		}
1730 	}
1731 }
1732 
1733 static inline void
1734 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1735 {
1736 	struct dst_entry *old_dst;
1737 
1738 	sk_tx_queue_clear(sk);
1739 	/*
1740 	 * This can be called while sk is owned by the caller only,
1741 	 * with no state that can be checked in a rcu_dereference_check() cond
1742 	 */
1743 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1744 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1745 	dst_release(old_dst);
1746 }
1747 
1748 static inline void
1749 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1750 {
1751 	spin_lock(&sk->sk_dst_lock);
1752 	__sk_dst_set(sk, dst);
1753 	spin_unlock(&sk->sk_dst_lock);
1754 }
1755 
1756 static inline void
1757 __sk_dst_reset(struct sock *sk)
1758 {
1759 	__sk_dst_set(sk, NULL);
1760 }
1761 
1762 static inline void
1763 sk_dst_reset(struct sock *sk)
1764 {
1765 	spin_lock(&sk->sk_dst_lock);
1766 	__sk_dst_reset(sk);
1767 	spin_unlock(&sk->sk_dst_lock);
1768 }
1769 
1770 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1771 
1772 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1773 
1774 static inline bool sk_can_gso(const struct sock *sk)
1775 {
1776 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1777 }
1778 
1779 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1780 
1781 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1782 {
1783 	sk->sk_route_nocaps |= flags;
1784 	sk->sk_route_caps &= ~flags;
1785 }
1786 
1787 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1788 					   char __user *from, char *to,
1789 					   int copy, int offset)
1790 {
1791 	if (skb->ip_summed == CHECKSUM_NONE) {
1792 		int err = 0;
1793 		__wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1794 		if (err)
1795 			return err;
1796 		skb->csum = csum_block_add(skb->csum, csum, offset);
1797 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1798 		if (!access_ok(VERIFY_READ, from, copy) ||
1799 		    __copy_from_user_nocache(to, from, copy))
1800 			return -EFAULT;
1801 	} else if (copy_from_user(to, from, copy))
1802 		return -EFAULT;
1803 
1804 	return 0;
1805 }
1806 
1807 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1808 				       char __user *from, int copy)
1809 {
1810 	int err, offset = skb->len;
1811 
1812 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1813 				       copy, offset);
1814 	if (err)
1815 		__skb_trim(skb, offset);
1816 
1817 	return err;
1818 }
1819 
1820 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1821 					   struct sk_buff *skb,
1822 					   struct page *page,
1823 					   int off, int copy)
1824 {
1825 	int err;
1826 
1827 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1828 				       copy, skb->len);
1829 	if (err)
1830 		return err;
1831 
1832 	skb->len	     += copy;
1833 	skb->data_len	     += copy;
1834 	skb->truesize	     += copy;
1835 	sk->sk_wmem_queued   += copy;
1836 	sk_mem_charge(sk, copy);
1837 	return 0;
1838 }
1839 
1840 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1841 				   struct sk_buff *skb, struct page *page,
1842 				   int off, int copy)
1843 {
1844 	if (skb->ip_summed == CHECKSUM_NONE) {
1845 		int err = 0;
1846 		__wsum csum = csum_and_copy_from_user(from,
1847 						     page_address(page) + off,
1848 							    copy, 0, &err);
1849 		if (err)
1850 			return err;
1851 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1852 	} else if (copy_from_user(page_address(page) + off, from, copy))
1853 		return -EFAULT;
1854 
1855 	skb->len	     += copy;
1856 	skb->data_len	     += copy;
1857 	skb->truesize	     += copy;
1858 	sk->sk_wmem_queued   += copy;
1859 	sk_mem_charge(sk, copy);
1860 	return 0;
1861 }
1862 
1863 /**
1864  * sk_wmem_alloc_get - returns write allocations
1865  * @sk: socket
1866  *
1867  * Returns sk_wmem_alloc minus initial offset of one
1868  */
1869 static inline int sk_wmem_alloc_get(const struct sock *sk)
1870 {
1871 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1872 }
1873 
1874 /**
1875  * sk_rmem_alloc_get - returns read allocations
1876  * @sk: socket
1877  *
1878  * Returns sk_rmem_alloc
1879  */
1880 static inline int sk_rmem_alloc_get(const struct sock *sk)
1881 {
1882 	return atomic_read(&sk->sk_rmem_alloc);
1883 }
1884 
1885 /**
1886  * sk_has_allocations - check if allocations are outstanding
1887  * @sk: socket
1888  *
1889  * Returns true if socket has write or read allocations
1890  */
1891 static inline bool sk_has_allocations(const struct sock *sk)
1892 {
1893 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1894 }
1895 
1896 /**
1897  * wq_has_sleeper - check if there are any waiting processes
1898  * @wq: struct socket_wq
1899  *
1900  * Returns true if socket_wq has waiting processes
1901  *
1902  * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1903  * barrier call. They were added due to the race found within the tcp code.
1904  *
1905  * Consider following tcp code paths:
1906  *
1907  * CPU1                  CPU2
1908  *
1909  * sys_select            receive packet
1910  *   ...                 ...
1911  *   __add_wait_queue    update tp->rcv_nxt
1912  *   ...                 ...
1913  *   tp->rcv_nxt check   sock_def_readable
1914  *   ...                 {
1915  *   schedule               rcu_read_lock();
1916  *                          wq = rcu_dereference(sk->sk_wq);
1917  *                          if (wq && waitqueue_active(&wq->wait))
1918  *                              wake_up_interruptible(&wq->wait)
1919  *                          ...
1920  *                       }
1921  *
1922  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1923  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1924  * could then endup calling schedule and sleep forever if there are no more
1925  * data on the socket.
1926  *
1927  */
1928 static inline bool wq_has_sleeper(struct socket_wq *wq)
1929 {
1930 	/* We need to be sure we are in sync with the
1931 	 * add_wait_queue modifications to the wait queue.
1932 	 *
1933 	 * This memory barrier is paired in the sock_poll_wait.
1934 	 */
1935 	smp_mb();
1936 	return wq && waitqueue_active(&wq->wait);
1937 }
1938 
1939 /**
1940  * sock_poll_wait - place memory barrier behind the poll_wait call.
1941  * @filp:           file
1942  * @wait_address:   socket wait queue
1943  * @p:              poll_table
1944  *
1945  * See the comments in the wq_has_sleeper function.
1946  */
1947 static inline void sock_poll_wait(struct file *filp,
1948 		wait_queue_head_t *wait_address, poll_table *p)
1949 {
1950 	if (!poll_does_not_wait(p) && wait_address) {
1951 		poll_wait(filp, wait_address, p);
1952 		/* We need to be sure we are in sync with the
1953 		 * socket flags modification.
1954 		 *
1955 		 * This memory barrier is paired in the wq_has_sleeper.
1956 		 */
1957 		smp_mb();
1958 	}
1959 }
1960 
1961 /*
1962  *	Queue a received datagram if it will fit. Stream and sequenced
1963  *	protocols can't normally use this as they need to fit buffers in
1964  *	and play with them.
1965  *
1966  *	Inlined as it's very short and called for pretty much every
1967  *	packet ever received.
1968  */
1969 
1970 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1971 {
1972 	skb_orphan(skb);
1973 	skb->sk = sk;
1974 	skb->destructor = sock_wfree;
1975 	/*
1976 	 * We used to take a refcount on sk, but following operation
1977 	 * is enough to guarantee sk_free() wont free this sock until
1978 	 * all in-flight packets are completed
1979 	 */
1980 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1981 }
1982 
1983 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1984 {
1985 	skb_orphan(skb);
1986 	skb->sk = sk;
1987 	skb->destructor = sock_rfree;
1988 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1989 	sk_mem_charge(sk, skb->truesize);
1990 }
1991 
1992 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1993 			   unsigned long expires);
1994 
1995 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1996 
1997 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1998 
1999 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2000 
2001 /*
2002  *	Recover an error report and clear atomically
2003  */
2004 
2005 static inline int sock_error(struct sock *sk)
2006 {
2007 	int err;
2008 	if (likely(!sk->sk_err))
2009 		return 0;
2010 	err = xchg(&sk->sk_err, 0);
2011 	return -err;
2012 }
2013 
2014 static inline unsigned long sock_wspace(struct sock *sk)
2015 {
2016 	int amt = 0;
2017 
2018 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2019 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2020 		if (amt < 0)
2021 			amt = 0;
2022 	}
2023 	return amt;
2024 }
2025 
2026 static inline void sk_wake_async(struct sock *sk, int how, int band)
2027 {
2028 	if (sock_flag(sk, SOCK_FASYNC))
2029 		sock_wake_async(sk->sk_socket, how, band);
2030 }
2031 
2032 #define SOCK_MIN_SNDBUF 2048
2033 /*
2034  * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
2035  * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
2036  */
2037 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
2038 
2039 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2040 {
2041 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2042 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2043 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2044 	}
2045 }
2046 
2047 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2048 
2049 /**
2050  * sk_page_frag - return an appropriate page_frag
2051  * @sk: socket
2052  *
2053  * If socket allocation mode allows current thread to sleep, it means its
2054  * safe to use the per task page_frag instead of the per socket one.
2055  */
2056 static inline struct page_frag *sk_page_frag(struct sock *sk)
2057 {
2058 	if (sk->sk_allocation & __GFP_WAIT)
2059 		return &current->task_frag;
2060 
2061 	return &sk->sk_frag;
2062 }
2063 
2064 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2065 
2066 /*
2067  *	Default write policy as shown to user space via poll/select/SIGIO
2068  */
2069 static inline bool sock_writeable(const struct sock *sk)
2070 {
2071 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2072 }
2073 
2074 static inline gfp_t gfp_any(void)
2075 {
2076 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2077 }
2078 
2079 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2080 {
2081 	return noblock ? 0 : sk->sk_rcvtimeo;
2082 }
2083 
2084 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2085 {
2086 	return noblock ? 0 : sk->sk_sndtimeo;
2087 }
2088 
2089 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2090 {
2091 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2092 }
2093 
2094 /* Alas, with timeout socket operations are not restartable.
2095  * Compare this to poll().
2096  */
2097 static inline int sock_intr_errno(long timeo)
2098 {
2099 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2100 }
2101 
2102 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2103 	struct sk_buff *skb);
2104 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2105 	struct sk_buff *skb);
2106 
2107 static inline void
2108 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2109 {
2110 	ktime_t kt = skb->tstamp;
2111 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2112 
2113 	/*
2114 	 * generate control messages if
2115 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2116 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
2117 	 * - software time stamp available and wanted
2118 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
2119 	 * - hardware time stamps available and wanted
2120 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
2121 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
2122 	 */
2123 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2124 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2125 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2126 	    (hwtstamps->hwtstamp.tv64 &&
2127 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2128 	    (hwtstamps->syststamp.tv64 &&
2129 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2130 		__sock_recv_timestamp(msg, sk, skb);
2131 	else
2132 		sk->sk_stamp = kt;
2133 
2134 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2135 		__sock_recv_wifi_status(msg, sk, skb);
2136 }
2137 
2138 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2139 				     struct sk_buff *skb);
2140 
2141 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2142 					  struct sk_buff *skb)
2143 {
2144 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2145 			   (1UL << SOCK_RCVTSTAMP)			| \
2146 			   (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)	| \
2147 			   (1UL << SOCK_TIMESTAMPING_SOFTWARE)		| \
2148 			   (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE)	| \
2149 			   (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2150 
2151 	if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2152 		__sock_recv_ts_and_drops(msg, sk, skb);
2153 	else
2154 		sk->sk_stamp = skb->tstamp;
2155 }
2156 
2157 /**
2158  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2159  * @sk:		socket sending this packet
2160  * @tx_flags:	filled with instructions for time stamping
2161  *
2162  * Currently only depends on SOCK_TIMESTAMPING* flags.
2163  */
2164 extern void 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