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/lockdep.h>
50 #include <linux/netdevice.h>
51 #include <linux/skbuff.h>	/* struct sk_buff */
52 #include <linux/mm.h>
53 #include <linux/security.h>
54 #include <linux/slab.h>
55 #include <linux/uaccess.h>
56 #include <linux/memcontrol.h>
57 #include <linux/res_counter.h>
58 #include <linux/static_key.h>
59 #include <linux/aio.h>
60 #include <linux/sched.h>
61 
62 #include <linux/filter.h>
63 #include <linux/rculist_nulls.h>
64 #include <linux/poll.h>
65 
66 #include <linux/atomic.h>
67 #include <net/dst.h>
68 #include <net/checksum.h>
69 
70 struct cgroup;
71 struct cgroup_subsys;
72 #ifdef CONFIG_NET
73 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss);
74 void mem_cgroup_sockets_destroy(struct cgroup *cgrp);
75 #else
76 static inline
mem_cgroup_sockets_init(struct cgroup * cgrp,struct cgroup_subsys * ss)77 int mem_cgroup_sockets_init(struct cgroup *cgrp, struct cgroup_subsys *ss)
78 {
79 	return 0;
80 }
81 static inline
mem_cgroup_sockets_destroy(struct cgroup * cgrp)82 void mem_cgroup_sockets_destroy(struct cgroup *cgrp)
83 {
84 }
85 #endif
86 /*
87  * This structure really needs to be cleaned up.
88  * Most of it is for TCP, and not used by any of
89  * the other protocols.
90  */
91 
92 /* Define this to get the SOCK_DBG debugging facility. */
93 #define SOCK_DEBUGGING
94 #ifdef SOCK_DEBUGGING
95 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
96 					printk(KERN_DEBUG msg); } while (0)
97 #else
98 /* Validate arguments and do nothing */
99 static inline __printf(2, 3)
SOCK_DEBUG(struct sock * sk,const char * msg,...)100 void SOCK_DEBUG(struct sock *sk, const char *msg, ...)
101 {
102 }
103 #endif
104 
105 /* This is the per-socket lock.  The spinlock provides a synchronization
106  * between user contexts and software interrupt processing, whereas the
107  * mini-semaphore synchronizes multiple users amongst themselves.
108  */
109 typedef struct {
110 	spinlock_t		slock;
111 	int			owned;
112 	wait_queue_head_t	wq;
113 	/*
114 	 * We express the mutex-alike socket_lock semantics
115 	 * to the lock validator by explicitly managing
116 	 * the slock as a lock variant (in addition to
117 	 * the slock itself):
118 	 */
119 #ifdef CONFIG_DEBUG_LOCK_ALLOC
120 	struct lockdep_map dep_map;
121 #endif
122 } socket_lock_t;
123 
124 struct sock;
125 struct proto;
126 struct net;
127 
128 /**
129  *	struct sock_common - minimal network layer representation of sockets
130  *	@skc_daddr: Foreign IPv4 addr
131  *	@skc_rcv_saddr: Bound local IPv4 addr
132  *	@skc_hash: hash value used with various protocol lookup tables
133  *	@skc_u16hashes: two u16 hash values used by UDP lookup tables
134  *	@skc_family: network address family
135  *	@skc_state: Connection state
136  *	@skc_reuse: %SO_REUSEADDR setting
137  *	@skc_bound_dev_if: bound device index if != 0
138  *	@skc_bind_node: bind hash linkage for various protocol lookup tables
139  *	@skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
140  *	@skc_prot: protocol handlers inside a network family
141  *	@skc_net: reference to the network namespace of this socket
142  *	@skc_node: main hash linkage for various protocol lookup tables
143  *	@skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
144  *	@skc_tx_queue_mapping: tx queue number for this connection
145  *	@skc_refcnt: reference count
146  *
147  *	This is the minimal network layer representation of sockets, the header
148  *	for struct sock and struct inet_timewait_sock.
149  */
150 struct sock_common {
151 	/* skc_daddr and skc_rcv_saddr must be grouped :
152 	 * cf INET_MATCH() and INET_TW_MATCH()
153 	 */
154 	__be32			skc_daddr;
155 	__be32			skc_rcv_saddr;
156 
157 	union  {
158 		unsigned int	skc_hash;
159 		__u16		skc_u16hashes[2];
160 	};
161 	unsigned short		skc_family;
162 	volatile unsigned char	skc_state;
163 	unsigned char		skc_reuse;
164 	int			skc_bound_dev_if;
165 	union {
166 		struct hlist_node	skc_bind_node;
167 		struct hlist_nulls_node skc_portaddr_node;
168 	};
169 	struct proto		*skc_prot;
170 #ifdef CONFIG_NET_NS
171 	struct net	 	*skc_net;
172 #endif
173 	/*
174 	 * fields between dontcopy_begin/dontcopy_end
175 	 * are not copied in sock_copy()
176 	 */
177 	/* private: */
178 	int			skc_dontcopy_begin[0];
179 	/* public: */
180 	union {
181 		struct hlist_node	skc_node;
182 		struct hlist_nulls_node skc_nulls_node;
183 	};
184 	int			skc_tx_queue_mapping;
185 	atomic_t		skc_refcnt;
186 	/* private: */
187 	int                     skc_dontcopy_end[0];
188 	/* public: */
189 };
190 
191 struct cg_proto;
192 /**
193   *	struct sock - network layer representation of sockets
194   *	@__sk_common: shared layout with inet_timewait_sock
195   *	@sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
196   *	@sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
197   *	@sk_lock:	synchronizer
198   *	@sk_rcvbuf: size of receive buffer in bytes
199   *	@sk_wq: sock wait queue and async head
200   *	@sk_dst_cache: destination cache
201   *	@sk_dst_lock: destination cache lock
202   *	@sk_policy: flow policy
203   *	@sk_receive_queue: incoming packets
204   *	@sk_wmem_alloc: transmit queue bytes committed
205   *	@sk_write_queue: Packet sending queue
206   *	@sk_async_wait_queue: DMA copied packets
207   *	@sk_omem_alloc: "o" is "option" or "other"
208   *	@sk_wmem_queued: persistent queue size
209   *	@sk_forward_alloc: space allocated forward
210   *	@sk_allocation: allocation mode
211   *	@sk_sndbuf: size of send buffer in bytes
212   *	@sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
213   *		   %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
214   *	@sk_no_check: %SO_NO_CHECK setting, wether or not checkup packets
215   *	@sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
216   *	@sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
217   *	@sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
218   *	@sk_gso_max_size: Maximum GSO segment size to build
219   *	@sk_gso_max_segs: Maximum number of GSO segments
220   *	@sk_lingertime: %SO_LINGER l_linger setting
221   *	@sk_backlog: always used with the per-socket spinlock held
222   *	@sk_callback_lock: used with the callbacks in the end of this struct
223   *	@sk_error_queue: rarely used
224   *	@sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
225   *			  IPV6_ADDRFORM for instance)
226   *	@sk_err: last error
227   *	@sk_err_soft: errors that don't cause failure but are the cause of a
228   *		      persistent failure not just 'timed out'
229   *	@sk_drops: raw/udp drops counter
230   *	@sk_ack_backlog: current listen backlog
231   *	@sk_max_ack_backlog: listen backlog set in listen()
232   *	@sk_priority: %SO_PRIORITY setting
233   *	@sk_cgrp_prioidx: socket group's priority map index
234   *	@sk_type: socket type (%SOCK_STREAM, etc)
235   *	@sk_protocol: which protocol this socket belongs in this network family
236   *	@sk_peer_pid: &struct pid for this socket's peer
237   *	@sk_peer_cred: %SO_PEERCRED setting
238   *	@sk_rcvlowat: %SO_RCVLOWAT setting
239   *	@sk_rcvtimeo: %SO_RCVTIMEO setting
240   *	@sk_sndtimeo: %SO_SNDTIMEO setting
241   *	@sk_rxhash: flow hash received from netif layer
242   *	@sk_filter: socket filtering instructions
243   *	@sk_protinfo: private area, net family specific, when not using slab
244   *	@sk_timer: sock cleanup timer
245   *	@sk_stamp: time stamp of last packet received
246   *	@sk_socket: Identd and reporting IO signals
247   *	@sk_user_data: RPC layer private data
248   *	@sk_sndmsg_page: cached page for sendmsg
249   *	@sk_sndmsg_off: cached offset for sendmsg
250   *	@sk_peek_off: current peek_offset value
251   *	@sk_send_head: front of stuff to transmit
252   *	@sk_security: used by security modules
253   *	@sk_mark: generic packet mark
254   *	@sk_classid: this socket's cgroup classid
255   *	@sk_cgrp: this socket's cgroup-specific proto data
256   *	@sk_write_pending: a write to stream socket waits to start
257   *	@sk_state_change: callback to indicate change in the state of the sock
258   *	@sk_data_ready: callback to indicate there is data to be processed
259   *	@sk_write_space: callback to indicate there is bf sending space available
260   *	@sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
261   *	@sk_backlog_rcv: callback to process the backlog
262   *	@sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
263  */
264 struct sock {
265 	/*
266 	 * Now struct inet_timewait_sock also uses sock_common, so please just
267 	 * don't add nothing before this first member (__sk_common) --acme
268 	 */
269 	struct sock_common	__sk_common;
270 #define sk_node			__sk_common.skc_node
271 #define sk_nulls_node		__sk_common.skc_nulls_node
272 #define sk_refcnt		__sk_common.skc_refcnt
273 #define sk_tx_queue_mapping	__sk_common.skc_tx_queue_mapping
274 
275 #define sk_dontcopy_begin	__sk_common.skc_dontcopy_begin
276 #define sk_dontcopy_end		__sk_common.skc_dontcopy_end
277 #define sk_hash			__sk_common.skc_hash
278 #define sk_family		__sk_common.skc_family
279 #define sk_state		__sk_common.skc_state
280 #define sk_reuse		__sk_common.skc_reuse
281 #define sk_bound_dev_if		__sk_common.skc_bound_dev_if
282 #define sk_bind_node		__sk_common.skc_bind_node
283 #define sk_prot			__sk_common.skc_prot
284 #define sk_net			__sk_common.skc_net
285 	socket_lock_t		sk_lock;
286 	struct sk_buff_head	sk_receive_queue;
287 	/*
288 	 * The backlog queue is special, it is always used with
289 	 * the per-socket spinlock held and requires low latency
290 	 * access. Therefore we special case it's implementation.
291 	 * Note : rmem_alloc is in this structure to fill a hole
292 	 * on 64bit arches, not because its logically part of
293 	 * backlog.
294 	 */
295 	struct {
296 		atomic_t	rmem_alloc;
297 		int		len;
298 		struct sk_buff	*head;
299 		struct sk_buff	*tail;
300 	} sk_backlog;
301 #define sk_rmem_alloc sk_backlog.rmem_alloc
302 	int			sk_forward_alloc;
303 #ifdef CONFIG_RPS
304 	__u32			sk_rxhash;
305 #endif
306 	atomic_t		sk_drops;
307 	int			sk_rcvbuf;
308 
309 	struct sk_filter __rcu	*sk_filter;
310 	struct socket_wq __rcu	*sk_wq;
311 
312 #ifdef CONFIG_NET_DMA
313 	struct sk_buff_head	sk_async_wait_queue;
314 #endif
315 
316 #ifdef CONFIG_XFRM
317 	struct xfrm_policy	*sk_policy[2];
318 #endif
319 	unsigned long 		sk_flags;
320 	struct dst_entry	*sk_dst_cache;
321 	spinlock_t		sk_dst_lock;
322 	atomic_t		sk_wmem_alloc;
323 	atomic_t		sk_omem_alloc;
324 	int			sk_sndbuf;
325 	struct sk_buff_head	sk_write_queue;
326 	kmemcheck_bitfield_begin(flags);
327 	unsigned int		sk_shutdown  : 2,
328 				sk_no_check  : 2,
329 				sk_userlocks : 4,
330 				sk_protocol  : 8,
331 				sk_type      : 16;
332 	kmemcheck_bitfield_end(flags);
333 	int			sk_wmem_queued;
334 	gfp_t			sk_allocation;
335 	netdev_features_t	sk_route_caps;
336 	netdev_features_t	sk_route_nocaps;
337 	int			sk_gso_type;
338 	unsigned int		sk_gso_max_size;
339 	u16			sk_gso_max_segs;
340 	int			sk_rcvlowat;
341 	unsigned long	        sk_lingertime;
342 	struct sk_buff_head	sk_error_queue;
343 	struct proto		*sk_prot_creator;
344 	rwlock_t		sk_callback_lock;
345 	int			sk_err,
346 				sk_err_soft;
347 	unsigned short		sk_ack_backlog;
348 	unsigned short		sk_max_ack_backlog;
349 	__u32			sk_priority;
350 #ifdef CONFIG_CGROUPS
351 	__u32			sk_cgrp_prioidx;
352 #endif
353 	struct pid		*sk_peer_pid;
354 	const struct cred	*sk_peer_cred;
355 	long			sk_rcvtimeo;
356 	long			sk_sndtimeo;
357 	void			*sk_protinfo;
358 	struct timer_list	sk_timer;
359 	ktime_t			sk_stamp;
360 	struct socket		*sk_socket;
361 	void			*sk_user_data;
362 	struct page		*sk_sndmsg_page;
363 	struct sk_buff		*sk_send_head;
364 	__u32			sk_sndmsg_off;
365 	__s32			sk_peek_off;
366 	int			sk_write_pending;
367 #ifdef CONFIG_SECURITY
368 	void			*sk_security;
369 #endif
370 	__u32			sk_mark;
371 	u32			sk_classid;
372 	struct cg_proto		*sk_cgrp;
373 	void			(*sk_state_change)(struct sock *sk);
374 	void			(*sk_data_ready)(struct sock *sk, int bytes);
375 	void			(*sk_write_space)(struct sock *sk);
376 	void			(*sk_error_report)(struct sock *sk);
377   	int			(*sk_backlog_rcv)(struct sock *sk,
378 						  struct sk_buff *skb);
379 	void                    (*sk_destruct)(struct sock *sk);
380 };
381 
sk_peek_offset(struct sock * sk,int flags)382 static inline int sk_peek_offset(struct sock *sk, int flags)
383 {
384 	if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
385 		return sk->sk_peek_off;
386 	else
387 		return 0;
388 }
389 
sk_peek_offset_bwd(struct sock * sk,int val)390 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
391 {
392 	if (sk->sk_peek_off >= 0) {
393 		if (sk->sk_peek_off >= val)
394 			sk->sk_peek_off -= val;
395 		else
396 			sk->sk_peek_off = 0;
397 	}
398 }
399 
sk_peek_offset_fwd(struct sock * sk,int val)400 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
401 {
402 	if (sk->sk_peek_off >= 0)
403 		sk->sk_peek_off += val;
404 }
405 
406 /*
407  * Hashed lists helper routines
408  */
sk_entry(const struct hlist_node * node)409 static inline struct sock *sk_entry(const struct hlist_node *node)
410 {
411 	return hlist_entry(node, struct sock, sk_node);
412 }
413 
__sk_head(const struct hlist_head * head)414 static inline struct sock *__sk_head(const struct hlist_head *head)
415 {
416 	return hlist_entry(head->first, struct sock, sk_node);
417 }
418 
sk_head(const struct hlist_head * head)419 static inline struct sock *sk_head(const struct hlist_head *head)
420 {
421 	return hlist_empty(head) ? NULL : __sk_head(head);
422 }
423 
__sk_nulls_head(const struct hlist_nulls_head * head)424 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
425 {
426 	return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
427 }
428 
sk_nulls_head(const struct hlist_nulls_head * head)429 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
430 {
431 	return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
432 }
433 
sk_next(const struct sock * sk)434 static inline struct sock *sk_next(const struct sock *sk)
435 {
436 	return sk->sk_node.next ?
437 		hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
438 }
439 
sk_nulls_next(const struct sock * sk)440 static inline struct sock *sk_nulls_next(const struct sock *sk)
441 {
442 	return (!is_a_nulls(sk->sk_nulls_node.next)) ?
443 		hlist_nulls_entry(sk->sk_nulls_node.next,
444 				  struct sock, sk_nulls_node) :
445 		NULL;
446 }
447 
sk_unhashed(const struct sock * sk)448 static inline int sk_unhashed(const struct sock *sk)
449 {
450 	return hlist_unhashed(&sk->sk_node);
451 }
452 
sk_hashed(const struct sock * sk)453 static inline int sk_hashed(const struct sock *sk)
454 {
455 	return !sk_unhashed(sk);
456 }
457 
sk_node_init(struct hlist_node * node)458 static __inline__ void sk_node_init(struct hlist_node *node)
459 {
460 	node->pprev = NULL;
461 }
462 
sk_nulls_node_init(struct hlist_nulls_node * node)463 static __inline__ void sk_nulls_node_init(struct hlist_nulls_node *node)
464 {
465 	node->pprev = NULL;
466 }
467 
__sk_del_node(struct sock * sk)468 static __inline__ void __sk_del_node(struct sock *sk)
469 {
470 	__hlist_del(&sk->sk_node);
471 }
472 
473 /* NB: equivalent to hlist_del_init_rcu */
__sk_del_node_init(struct sock * sk)474 static __inline__ int __sk_del_node_init(struct sock *sk)
475 {
476 	if (sk_hashed(sk)) {
477 		__sk_del_node(sk);
478 		sk_node_init(&sk->sk_node);
479 		return 1;
480 	}
481 	return 0;
482 }
483 
484 /* Grab socket reference count. This operation is valid only
485    when sk is ALREADY grabbed f.e. it is found in hash table
486    or a list and the lookup is made under lock preventing hash table
487    modifications.
488  */
489 
sock_hold(struct sock * sk)490 static inline void sock_hold(struct sock *sk)
491 {
492 	atomic_inc(&sk->sk_refcnt);
493 }
494 
495 /* Ungrab socket in the context, which assumes that socket refcnt
496    cannot hit zero, f.e. it is true in context of any socketcall.
497  */
__sock_put(struct sock * sk)498 static inline void __sock_put(struct sock *sk)
499 {
500 	atomic_dec(&sk->sk_refcnt);
501 }
502 
sk_del_node_init(struct sock * sk)503 static __inline__ int sk_del_node_init(struct sock *sk)
504 {
505 	int rc = __sk_del_node_init(sk);
506 
507 	if (rc) {
508 		/* paranoid for a while -acme */
509 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
510 		__sock_put(sk);
511 	}
512 	return rc;
513 }
514 #define sk_del_node_init_rcu(sk)	sk_del_node_init(sk)
515 
__sk_nulls_del_node_init_rcu(struct sock * sk)516 static __inline__ int __sk_nulls_del_node_init_rcu(struct sock *sk)
517 {
518 	if (sk_hashed(sk)) {
519 		hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
520 		return 1;
521 	}
522 	return 0;
523 }
524 
sk_nulls_del_node_init_rcu(struct sock * sk)525 static __inline__ int sk_nulls_del_node_init_rcu(struct sock *sk)
526 {
527 	int rc = __sk_nulls_del_node_init_rcu(sk);
528 
529 	if (rc) {
530 		/* paranoid for a while -acme */
531 		WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
532 		__sock_put(sk);
533 	}
534 	return rc;
535 }
536 
__sk_add_node(struct sock * sk,struct hlist_head * list)537 static __inline__ void __sk_add_node(struct sock *sk, struct hlist_head *list)
538 {
539 	hlist_add_head(&sk->sk_node, list);
540 }
541 
sk_add_node(struct sock * sk,struct hlist_head * list)542 static __inline__ void sk_add_node(struct sock *sk, struct hlist_head *list)
543 {
544 	sock_hold(sk);
545 	__sk_add_node(sk, list);
546 }
547 
sk_add_node_rcu(struct sock * sk,struct hlist_head * list)548 static __inline__ void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
549 {
550 	sock_hold(sk);
551 	hlist_add_head_rcu(&sk->sk_node, list);
552 }
553 
__sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)554 static __inline__ void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
555 {
556 	hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
557 }
558 
sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)559 static __inline__ void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
560 {
561 	sock_hold(sk);
562 	__sk_nulls_add_node_rcu(sk, list);
563 }
564 
__sk_del_bind_node(struct sock * sk)565 static __inline__ void __sk_del_bind_node(struct sock *sk)
566 {
567 	__hlist_del(&sk->sk_bind_node);
568 }
569 
sk_add_bind_node(struct sock * sk,struct hlist_head * list)570 static __inline__ void sk_add_bind_node(struct sock *sk,
571 					struct hlist_head *list)
572 {
573 	hlist_add_head(&sk->sk_bind_node, list);
574 }
575 
576 #define sk_for_each(__sk, node, list) \
577 	hlist_for_each_entry(__sk, node, list, sk_node)
578 #define sk_for_each_rcu(__sk, node, list) \
579 	hlist_for_each_entry_rcu(__sk, node, list, sk_node)
580 #define sk_nulls_for_each(__sk, node, list) \
581 	hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
582 #define sk_nulls_for_each_rcu(__sk, node, list) \
583 	hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
584 #define sk_for_each_from(__sk, node) \
585 	if (__sk && ({ node = &(__sk)->sk_node; 1; })) \
586 		hlist_for_each_entry_from(__sk, node, sk_node)
587 #define sk_nulls_for_each_from(__sk, node) \
588 	if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
589 		hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
590 #define sk_for_each_safe(__sk, node, tmp, list) \
591 	hlist_for_each_entry_safe(__sk, node, tmp, list, sk_node)
592 #define sk_for_each_bound(__sk, node, list) \
593 	hlist_for_each_entry(__sk, node, list, sk_bind_node)
594 
595 /* Sock flags */
596 enum sock_flags {
597 	SOCK_DEAD,
598 	SOCK_DONE,
599 	SOCK_URGINLINE,
600 	SOCK_KEEPOPEN,
601 	SOCK_LINGER,
602 	SOCK_DESTROY,
603 	SOCK_BROADCAST,
604 	SOCK_TIMESTAMP,
605 	SOCK_ZAPPED,
606 	SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
607 	SOCK_DBG, /* %SO_DEBUG setting */
608 	SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
609 	SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
610 	SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
611 	SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
612 	SOCK_TIMESTAMPING_TX_HARDWARE,  /* %SOF_TIMESTAMPING_TX_HARDWARE */
613 	SOCK_TIMESTAMPING_TX_SOFTWARE,  /* %SOF_TIMESTAMPING_TX_SOFTWARE */
614 	SOCK_TIMESTAMPING_RX_HARDWARE,  /* %SOF_TIMESTAMPING_RX_HARDWARE */
615 	SOCK_TIMESTAMPING_RX_SOFTWARE,  /* %SOF_TIMESTAMPING_RX_SOFTWARE */
616 	SOCK_TIMESTAMPING_SOFTWARE,     /* %SOF_TIMESTAMPING_SOFTWARE */
617 	SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
618 	SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
619 	SOCK_FASYNC, /* fasync() active */
620 	SOCK_RXQ_OVFL,
621 	SOCK_ZEROCOPY, /* buffers from userspace */
622 	SOCK_WIFI_STATUS, /* push wifi status to userspace */
623 	SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
624 		     * Will use last 4 bytes of packet sent from
625 		     * user-space instead.
626 		     */
627 };
628 
sock_copy_flags(struct sock * nsk,struct sock * osk)629 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
630 {
631 	nsk->sk_flags = osk->sk_flags;
632 }
633 
sock_set_flag(struct sock * sk,enum sock_flags flag)634 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
635 {
636 	__set_bit(flag, &sk->sk_flags);
637 }
638 
sock_reset_flag(struct sock * sk,enum sock_flags flag)639 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
640 {
641 	__clear_bit(flag, &sk->sk_flags);
642 }
643 
sock_flag(struct sock * sk,enum sock_flags flag)644 static inline int sock_flag(struct sock *sk, enum sock_flags flag)
645 {
646 	return test_bit(flag, &sk->sk_flags);
647 }
648 
sk_acceptq_removed(struct sock * sk)649 static inline void sk_acceptq_removed(struct sock *sk)
650 {
651 	sk->sk_ack_backlog--;
652 }
653 
sk_acceptq_added(struct sock * sk)654 static inline void sk_acceptq_added(struct sock *sk)
655 {
656 	sk->sk_ack_backlog++;
657 }
658 
sk_acceptq_is_full(struct sock * sk)659 static inline int sk_acceptq_is_full(struct sock *sk)
660 {
661 	return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
662 }
663 
664 /*
665  * Compute minimal free write space needed to queue new packets.
666  */
sk_stream_min_wspace(struct sock * sk)667 static inline int sk_stream_min_wspace(struct sock *sk)
668 {
669 	return sk->sk_wmem_queued >> 1;
670 }
671 
sk_stream_wspace(struct sock * sk)672 static inline int sk_stream_wspace(struct sock *sk)
673 {
674 	return sk->sk_sndbuf - sk->sk_wmem_queued;
675 }
676 
677 extern void sk_stream_write_space(struct sock *sk);
678 
sk_stream_memory_free(struct sock * sk)679 static inline int sk_stream_memory_free(struct sock *sk)
680 {
681 	return sk->sk_wmem_queued < sk->sk_sndbuf;
682 }
683 
684 /* OOB backlog add */
__sk_add_backlog(struct sock * sk,struct sk_buff * skb)685 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
686 {
687 	/* dont let skb dst not refcounted, we are going to leave rcu lock */
688 	skb_dst_force(skb);
689 
690 	if (!sk->sk_backlog.tail)
691 		sk->sk_backlog.head = skb;
692 	else
693 		sk->sk_backlog.tail->next = skb;
694 
695 	sk->sk_backlog.tail = skb;
696 	skb->next = NULL;
697 }
698 
699 /*
700  * Take into account size of receive queue and backlog queue
701  * Do not take into account this skb truesize,
702  * to allow even a single big packet to come.
703  */
sk_rcvqueues_full(const struct sock * sk,const struct sk_buff * skb)704 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb)
705 {
706 	unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
707 
708 	return qsize > sk->sk_rcvbuf;
709 }
710 
711 /* The per-socket spinlock must be held here. */
sk_add_backlog(struct sock * sk,struct sk_buff * skb)712 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb)
713 {
714 	if (sk_rcvqueues_full(sk, skb))
715 		return -ENOBUFS;
716 
717 	__sk_add_backlog(sk, skb);
718 	sk->sk_backlog.len += skb->truesize;
719 	return 0;
720 }
721 
sk_backlog_rcv(struct sock * sk,struct sk_buff * skb)722 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
723 {
724 	return sk->sk_backlog_rcv(sk, skb);
725 }
726 
sock_rps_record_flow(const struct sock * sk)727 static inline void sock_rps_record_flow(const struct sock *sk)
728 {
729 #ifdef CONFIG_RPS
730 	struct rps_sock_flow_table *sock_flow_table;
731 
732 	rcu_read_lock();
733 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
734 	rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
735 	rcu_read_unlock();
736 #endif
737 }
738 
sock_rps_reset_flow(const struct sock * sk)739 static inline void sock_rps_reset_flow(const struct sock *sk)
740 {
741 #ifdef CONFIG_RPS
742 	struct rps_sock_flow_table *sock_flow_table;
743 
744 	rcu_read_lock();
745 	sock_flow_table = rcu_dereference(rps_sock_flow_table);
746 	rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
747 	rcu_read_unlock();
748 #endif
749 }
750 
sock_rps_save_rxhash(struct sock * sk,const struct sk_buff * skb)751 static inline void sock_rps_save_rxhash(struct sock *sk,
752 					const struct sk_buff *skb)
753 {
754 #ifdef CONFIG_RPS
755 	if (unlikely(sk->sk_rxhash != skb->rxhash)) {
756 		sock_rps_reset_flow(sk);
757 		sk->sk_rxhash = skb->rxhash;
758 	}
759 #endif
760 }
761 
sock_rps_reset_rxhash(struct sock * sk)762 static inline void sock_rps_reset_rxhash(struct sock *sk)
763 {
764 #ifdef CONFIG_RPS
765 	sock_rps_reset_flow(sk);
766 	sk->sk_rxhash = 0;
767 #endif
768 }
769 
770 #define sk_wait_event(__sk, __timeo, __condition)			\
771 	({	int __rc;						\
772 		release_sock(__sk);					\
773 		__rc = __condition;					\
774 		if (!__rc) {						\
775 			*(__timeo) = schedule_timeout(*(__timeo));	\
776 		}							\
777 		lock_sock(__sk);					\
778 		__rc = __condition;					\
779 		__rc;							\
780 	})
781 
782 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
783 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
784 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
785 extern int sk_stream_error(struct sock *sk, int flags, int err);
786 extern void sk_stream_kill_queues(struct sock *sk);
787 
788 extern int sk_wait_data(struct sock *sk, long *timeo);
789 
790 struct request_sock_ops;
791 struct timewait_sock_ops;
792 struct inet_hashinfo;
793 struct raw_hashinfo;
794 struct module;
795 
796 /*
797  * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
798  * un-modified. Special care is taken when initializing object to zero.
799  */
sk_prot_clear_nulls(struct sock * sk,int size)800 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
801 {
802 	if (offsetof(struct sock, sk_node.next) != 0)
803 		memset(sk, 0, offsetof(struct sock, sk_node.next));
804 	memset(&sk->sk_node.pprev, 0,
805 	       size - offsetof(struct sock, sk_node.pprev));
806 }
807 
808 /* Networking protocol blocks we attach to sockets.
809  * socket layer -> transport layer interface
810  * transport -> network interface is defined by struct inet_proto
811  */
812 struct proto {
813 	void			(*close)(struct sock *sk,
814 					long timeout);
815 	int			(*connect)(struct sock *sk,
816 				        struct sockaddr *uaddr,
817 					int addr_len);
818 	int			(*disconnect)(struct sock *sk, int flags);
819 
820 	struct sock *		(*accept) (struct sock *sk, int flags, int *err);
821 
822 	int			(*ioctl)(struct sock *sk, int cmd,
823 					 unsigned long arg);
824 	int			(*init)(struct sock *sk);
825 	void			(*destroy)(struct sock *sk);
826 	void			(*shutdown)(struct sock *sk, int how);
827 	int			(*setsockopt)(struct sock *sk, int level,
828 					int optname, char __user *optval,
829 					unsigned int optlen);
830 	int			(*getsockopt)(struct sock *sk, int level,
831 					int optname, char __user *optval,
832 					int __user *option);
833 #ifdef CONFIG_COMPAT
834 	int			(*compat_setsockopt)(struct sock *sk,
835 					int level,
836 					int optname, char __user *optval,
837 					unsigned int optlen);
838 	int			(*compat_getsockopt)(struct sock *sk,
839 					int level,
840 					int optname, char __user *optval,
841 					int __user *option);
842 	int			(*compat_ioctl)(struct sock *sk,
843 					unsigned int cmd, unsigned long arg);
844 #endif
845 	int			(*sendmsg)(struct kiocb *iocb, struct sock *sk,
846 					   struct msghdr *msg, size_t len);
847 	int			(*recvmsg)(struct kiocb *iocb, struct sock *sk,
848 					   struct msghdr *msg,
849 					size_t len, int noblock, int flags,
850 					int *addr_len);
851 	int			(*sendpage)(struct sock *sk, struct page *page,
852 					int offset, size_t size, int flags);
853 	int			(*bind)(struct sock *sk,
854 					struct sockaddr *uaddr, int addr_len);
855 
856 	int			(*backlog_rcv) (struct sock *sk,
857 						struct sk_buff *skb);
858 
859 	/* Keeping track of sk's, looking them up, and port selection methods. */
860 	void			(*hash)(struct sock *sk);
861 	void			(*unhash)(struct sock *sk);
862 	void			(*rehash)(struct sock *sk);
863 	int			(*get_port)(struct sock *sk, unsigned short snum);
864 	void			(*clear_sk)(struct sock *sk, int size);
865 
866 	/* Keeping track of sockets in use */
867 #ifdef CONFIG_PROC_FS
868 	unsigned int		inuse_idx;
869 #endif
870 
871 	/* Memory pressure */
872 	void			(*enter_memory_pressure)(struct sock *sk);
873 	atomic_long_t		*memory_allocated;	/* Current allocated memory. */
874 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
875 	/*
876 	 * Pressure flag: try to collapse.
877 	 * Technical note: it is used by multiple contexts non atomically.
878 	 * All the __sk_mem_schedule() is of this nature: accounting
879 	 * is strict, actions are advisory and have some latency.
880 	 */
881 	int			*memory_pressure;
882 	long			*sysctl_mem;
883 	int			*sysctl_wmem;
884 	int			*sysctl_rmem;
885 	int			max_header;
886 	bool			no_autobind;
887 
888 	struct kmem_cache	*slab;
889 	unsigned int		obj_size;
890 	int			slab_flags;
891 
892 	struct percpu_counter	*orphan_count;
893 
894 	struct request_sock_ops	*rsk_prot;
895 	struct timewait_sock_ops *twsk_prot;
896 
897 	union {
898 		struct inet_hashinfo	*hashinfo;
899 		struct udp_table	*udp_table;
900 		struct raw_hashinfo	*raw_hash;
901 	} h;
902 
903 	struct module		*owner;
904 
905 	char			name[32];
906 
907 	struct list_head	node;
908 #ifdef SOCK_REFCNT_DEBUG
909 	atomic_t		socks;
910 #endif
911 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
912 	/*
913 	 * cgroup specific init/deinit functions. Called once for all
914 	 * protocols that implement it, from cgroups populate function.
915 	 * This function has to setup any files the protocol want to
916 	 * appear in the kmem cgroup filesystem.
917 	 */
918 	int			(*init_cgroup)(struct cgroup *cgrp,
919 					       struct cgroup_subsys *ss);
920 	void			(*destroy_cgroup)(struct cgroup *cgrp);
921 	struct cg_proto		*(*proto_cgroup)(struct mem_cgroup *memcg);
922 #endif
923 };
924 
925 struct cg_proto {
926 	void			(*enter_memory_pressure)(struct sock *sk);
927 	struct res_counter	*memory_allocated;	/* Current allocated memory. */
928 	struct percpu_counter	*sockets_allocated;	/* Current number of sockets. */
929 	int			*memory_pressure;
930 	long			*sysctl_mem;
931 	/*
932 	 * memcg field is used to find which memcg we belong directly
933 	 * Each memcg struct can hold more than one cg_proto, so container_of
934 	 * won't really cut.
935 	 *
936 	 * The elegant solution would be having an inverse function to
937 	 * proto_cgroup in struct proto, but that means polluting the structure
938 	 * for everybody, instead of just for memcg users.
939 	 */
940 	struct mem_cgroup	*memcg;
941 };
942 
943 extern int proto_register(struct proto *prot, int alloc_slab);
944 extern void proto_unregister(struct proto *prot);
945 
946 #ifdef SOCK_REFCNT_DEBUG
sk_refcnt_debug_inc(struct sock * sk)947 static inline void sk_refcnt_debug_inc(struct sock *sk)
948 {
949 	atomic_inc(&sk->sk_prot->socks);
950 }
951 
sk_refcnt_debug_dec(struct sock * sk)952 static inline void sk_refcnt_debug_dec(struct sock *sk)
953 {
954 	atomic_dec(&sk->sk_prot->socks);
955 	printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
956 	       sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
957 }
958 
sk_refcnt_debug_release(const struct sock * sk)959 static inline void sk_refcnt_debug_release(const struct sock *sk)
960 {
961 	if (atomic_read(&sk->sk_refcnt) != 1)
962 		printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
963 		       sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
964 }
965 #else /* SOCK_REFCNT_DEBUG */
966 #define sk_refcnt_debug_inc(sk) do { } while (0)
967 #define sk_refcnt_debug_dec(sk) do { } while (0)
968 #define sk_refcnt_debug_release(sk) do { } while (0)
969 #endif /* SOCK_REFCNT_DEBUG */
970 
971 #if defined(CONFIG_CGROUP_MEM_RES_CTLR_KMEM) && defined(CONFIG_NET)
972 extern struct static_key memcg_socket_limit_enabled;
parent_cg_proto(struct proto * proto,struct cg_proto * cg_proto)973 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
974 					       struct cg_proto *cg_proto)
975 {
976 	return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
977 }
978 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
979 #else
980 #define mem_cgroup_sockets_enabled 0
parent_cg_proto(struct proto * proto,struct cg_proto * cg_proto)981 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
982 					       struct cg_proto *cg_proto)
983 {
984 	return NULL;
985 }
986 #endif
987 
988 
sk_has_memory_pressure(const struct sock * sk)989 static inline bool sk_has_memory_pressure(const struct sock *sk)
990 {
991 	return sk->sk_prot->memory_pressure != NULL;
992 }
993 
sk_under_memory_pressure(const struct sock * sk)994 static inline bool sk_under_memory_pressure(const struct sock *sk)
995 {
996 	if (!sk->sk_prot->memory_pressure)
997 		return false;
998 
999 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1000 		return !!*sk->sk_cgrp->memory_pressure;
1001 
1002 	return !!*sk->sk_prot->memory_pressure;
1003 }
1004 
sk_leave_memory_pressure(struct sock * sk)1005 static inline void sk_leave_memory_pressure(struct sock *sk)
1006 {
1007 	int *memory_pressure = sk->sk_prot->memory_pressure;
1008 
1009 	if (!memory_pressure)
1010 		return;
1011 
1012 	if (*memory_pressure)
1013 		*memory_pressure = 0;
1014 
1015 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1016 		struct cg_proto *cg_proto = sk->sk_cgrp;
1017 		struct proto *prot = sk->sk_prot;
1018 
1019 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1020 			if (*cg_proto->memory_pressure)
1021 				*cg_proto->memory_pressure = 0;
1022 	}
1023 
1024 }
1025 
sk_enter_memory_pressure(struct sock * sk)1026 static inline void sk_enter_memory_pressure(struct sock *sk)
1027 {
1028 	if (!sk->sk_prot->enter_memory_pressure)
1029 		return;
1030 
1031 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1032 		struct cg_proto *cg_proto = sk->sk_cgrp;
1033 		struct proto *prot = sk->sk_prot;
1034 
1035 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1036 			cg_proto->enter_memory_pressure(sk);
1037 	}
1038 
1039 	sk->sk_prot->enter_memory_pressure(sk);
1040 }
1041 
sk_prot_mem_limits(const struct sock * sk,int index)1042 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1043 {
1044 	long *prot = sk->sk_prot->sysctl_mem;
1045 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1046 		prot = sk->sk_cgrp->sysctl_mem;
1047 	return prot[index];
1048 }
1049 
memcg_memory_allocated_add(struct cg_proto * prot,unsigned long amt,int * parent_status)1050 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1051 					      unsigned long amt,
1052 					      int *parent_status)
1053 {
1054 	struct res_counter *fail;
1055 	int ret;
1056 
1057 	ret = res_counter_charge_nofail(prot->memory_allocated,
1058 					amt << PAGE_SHIFT, &fail);
1059 	if (ret < 0)
1060 		*parent_status = OVER_LIMIT;
1061 }
1062 
memcg_memory_allocated_sub(struct cg_proto * prot,unsigned long amt)1063 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1064 					      unsigned long amt)
1065 {
1066 	res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1067 }
1068 
memcg_memory_allocated_read(struct cg_proto * prot)1069 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1070 {
1071 	u64 ret;
1072 	ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1073 	return ret >> PAGE_SHIFT;
1074 }
1075 
1076 static inline long
sk_memory_allocated(const struct sock * sk)1077 sk_memory_allocated(const struct sock *sk)
1078 {
1079 	struct proto *prot = sk->sk_prot;
1080 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1081 		return memcg_memory_allocated_read(sk->sk_cgrp);
1082 
1083 	return atomic_long_read(prot->memory_allocated);
1084 }
1085 
1086 static inline long
sk_memory_allocated_add(struct sock * sk,int amt,int * parent_status)1087 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1088 {
1089 	struct proto *prot = sk->sk_prot;
1090 
1091 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1092 		memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1093 		/* update the root cgroup regardless */
1094 		atomic_long_add_return(amt, prot->memory_allocated);
1095 		return memcg_memory_allocated_read(sk->sk_cgrp);
1096 	}
1097 
1098 	return atomic_long_add_return(amt, prot->memory_allocated);
1099 }
1100 
1101 static inline void
sk_memory_allocated_sub(struct sock * sk,int amt)1102 sk_memory_allocated_sub(struct sock *sk, int amt)
1103 {
1104 	struct proto *prot = sk->sk_prot;
1105 
1106 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1107 		memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1108 
1109 	atomic_long_sub(amt, prot->memory_allocated);
1110 }
1111 
sk_sockets_allocated_dec(struct sock * sk)1112 static inline void sk_sockets_allocated_dec(struct sock *sk)
1113 {
1114 	struct proto *prot = sk->sk_prot;
1115 
1116 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1117 		struct cg_proto *cg_proto = sk->sk_cgrp;
1118 
1119 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1120 			percpu_counter_dec(cg_proto->sockets_allocated);
1121 	}
1122 
1123 	percpu_counter_dec(prot->sockets_allocated);
1124 }
1125 
sk_sockets_allocated_inc(struct sock * sk)1126 static inline void sk_sockets_allocated_inc(struct sock *sk)
1127 {
1128 	struct proto *prot = sk->sk_prot;
1129 
1130 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1131 		struct cg_proto *cg_proto = sk->sk_cgrp;
1132 
1133 		for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1134 			percpu_counter_inc(cg_proto->sockets_allocated);
1135 	}
1136 
1137 	percpu_counter_inc(prot->sockets_allocated);
1138 }
1139 
1140 static inline int
sk_sockets_allocated_read_positive(struct sock * sk)1141 sk_sockets_allocated_read_positive(struct sock *sk)
1142 {
1143 	struct proto *prot = sk->sk_prot;
1144 
1145 	if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1146 		return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1147 
1148 	return percpu_counter_read_positive(prot->sockets_allocated);
1149 }
1150 
1151 static inline int
proto_sockets_allocated_sum_positive(struct proto * prot)1152 proto_sockets_allocated_sum_positive(struct proto *prot)
1153 {
1154 	return percpu_counter_sum_positive(prot->sockets_allocated);
1155 }
1156 
1157 static inline long
proto_memory_allocated(struct proto * prot)1158 proto_memory_allocated(struct proto *prot)
1159 {
1160 	return atomic_long_read(prot->memory_allocated);
1161 }
1162 
1163 static inline bool
proto_memory_pressure(struct proto * prot)1164 proto_memory_pressure(struct proto *prot)
1165 {
1166 	if (!prot->memory_pressure)
1167 		return false;
1168 	return !!*prot->memory_pressure;
1169 }
1170 
1171 
1172 #ifdef CONFIG_PROC_FS
1173 /* Called with local bh disabled */
1174 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1175 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1176 #else
sock_prot_inuse_add(struct net * net,struct proto * prot,int inc)1177 static void inline sock_prot_inuse_add(struct net *net, struct proto *prot,
1178 		int inc)
1179 {
1180 }
1181 #endif
1182 
1183 
1184 /* With per-bucket locks this operation is not-atomic, so that
1185  * this version is not worse.
1186  */
__sk_prot_rehash(struct sock * sk)1187 static inline void __sk_prot_rehash(struct sock *sk)
1188 {
1189 	sk->sk_prot->unhash(sk);
1190 	sk->sk_prot->hash(sk);
1191 }
1192 
1193 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1194 
1195 /* About 10 seconds */
1196 #define SOCK_DESTROY_TIME (10*HZ)
1197 
1198 /* Sockets 0-1023 can't be bound to unless you are superuser */
1199 #define PROT_SOCK	1024
1200 
1201 #define SHUTDOWN_MASK	3
1202 #define RCV_SHUTDOWN	1
1203 #define SEND_SHUTDOWN	2
1204 
1205 #define SOCK_SNDBUF_LOCK	1
1206 #define SOCK_RCVBUF_LOCK	2
1207 #define SOCK_BINDADDR_LOCK	4
1208 #define SOCK_BINDPORT_LOCK	8
1209 
1210 /* sock_iocb: used to kick off async processing of socket ios */
1211 struct sock_iocb {
1212 	struct list_head	list;
1213 
1214 	int			flags;
1215 	int			size;
1216 	struct socket		*sock;
1217 	struct sock		*sk;
1218 	struct scm_cookie	*scm;
1219 	struct msghdr		*msg, async_msg;
1220 	struct kiocb		*kiocb;
1221 };
1222 
kiocb_to_siocb(struct kiocb * iocb)1223 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1224 {
1225 	return (struct sock_iocb *)iocb->private;
1226 }
1227 
siocb_to_kiocb(struct sock_iocb * si)1228 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1229 {
1230 	return si->kiocb;
1231 }
1232 
1233 struct socket_alloc {
1234 	struct socket socket;
1235 	struct inode vfs_inode;
1236 };
1237 
SOCKET_I(struct inode * inode)1238 static inline struct socket *SOCKET_I(struct inode *inode)
1239 {
1240 	return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1241 }
1242 
SOCK_INODE(struct socket * socket)1243 static inline struct inode *SOCK_INODE(struct socket *socket)
1244 {
1245 	return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1246 }
1247 
1248 /*
1249  * Functions for memory accounting
1250  */
1251 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1252 extern void __sk_mem_reclaim(struct sock *sk);
1253 
1254 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1255 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1256 #define SK_MEM_SEND	0
1257 #define SK_MEM_RECV	1
1258 
sk_mem_pages(int amt)1259 static inline int sk_mem_pages(int amt)
1260 {
1261 	return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1262 }
1263 
sk_has_account(struct sock * sk)1264 static inline int sk_has_account(struct sock *sk)
1265 {
1266 	/* return true if protocol supports memory accounting */
1267 	return !!sk->sk_prot->memory_allocated;
1268 }
1269 
sk_wmem_schedule(struct sock * sk,int size)1270 static inline int sk_wmem_schedule(struct sock *sk, int size)
1271 {
1272 	if (!sk_has_account(sk))
1273 		return 1;
1274 	return size <= sk->sk_forward_alloc ||
1275 		__sk_mem_schedule(sk, size, SK_MEM_SEND);
1276 }
1277 
sk_rmem_schedule(struct sock * sk,int size)1278 static inline int sk_rmem_schedule(struct sock *sk, int size)
1279 {
1280 	if (!sk_has_account(sk))
1281 		return 1;
1282 	return size <= sk->sk_forward_alloc ||
1283 		__sk_mem_schedule(sk, size, SK_MEM_RECV);
1284 }
1285 
sk_mem_reclaim(struct sock * sk)1286 static inline void sk_mem_reclaim(struct sock *sk)
1287 {
1288 	if (!sk_has_account(sk))
1289 		return;
1290 	if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1291 		__sk_mem_reclaim(sk);
1292 }
1293 
sk_mem_reclaim_partial(struct sock * sk)1294 static inline void sk_mem_reclaim_partial(struct sock *sk)
1295 {
1296 	if (!sk_has_account(sk))
1297 		return;
1298 	if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1299 		__sk_mem_reclaim(sk);
1300 }
1301 
sk_mem_charge(struct sock * sk,int size)1302 static inline void sk_mem_charge(struct sock *sk, int size)
1303 {
1304 	if (!sk_has_account(sk))
1305 		return;
1306 	sk->sk_forward_alloc -= size;
1307 }
1308 
sk_mem_uncharge(struct sock * sk,int size)1309 static inline void sk_mem_uncharge(struct sock *sk, int size)
1310 {
1311 	if (!sk_has_account(sk))
1312 		return;
1313 	sk->sk_forward_alloc += size;
1314 }
1315 
sk_wmem_free_skb(struct sock * sk,struct sk_buff * skb)1316 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1317 {
1318 	sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1319 	sk->sk_wmem_queued -= skb->truesize;
1320 	sk_mem_uncharge(sk, skb->truesize);
1321 	__kfree_skb(skb);
1322 }
1323 
1324 /* Used by processes to "lock" a socket state, so that
1325  * interrupts and bottom half handlers won't change it
1326  * from under us. It essentially blocks any incoming
1327  * packets, so that we won't get any new data or any
1328  * packets that change the state of the socket.
1329  *
1330  * While locked, BH processing will add new packets to
1331  * the backlog queue.  This queue is processed by the
1332  * owner of the socket lock right before it is released.
1333  *
1334  * Since ~2.3.5 it is also exclusive sleep lock serializing
1335  * accesses from user process context.
1336  */
1337 #define sock_owned_by_user(sk)	((sk)->sk_lock.owned)
1338 
1339 /*
1340  * Macro so as to not evaluate some arguments when
1341  * lockdep is not enabled.
1342  *
1343  * Mark both the sk_lock and the sk_lock.slock as a
1344  * per-address-family lock class.
1345  */
1346 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) 	\
1347 do {									\
1348 	sk->sk_lock.owned = 0;						\
1349 	init_waitqueue_head(&sk->sk_lock.wq);				\
1350 	spin_lock_init(&(sk)->sk_lock.slock);				\
1351 	debug_check_no_locks_freed((void *)&(sk)->sk_lock,		\
1352 			sizeof((sk)->sk_lock));				\
1353 	lockdep_set_class_and_name(&(sk)->sk_lock.slock,		\
1354 		       	(skey), (sname));				\
1355 	lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0);	\
1356 } while (0)
1357 
1358 extern void lock_sock_nested(struct sock *sk, int subclass);
1359 
lock_sock(struct sock * sk)1360 static inline void lock_sock(struct sock *sk)
1361 {
1362 	lock_sock_nested(sk, 0);
1363 }
1364 
1365 extern void release_sock(struct sock *sk);
1366 
1367 /* BH context may only use the following locking interface. */
1368 #define bh_lock_sock(__sk)	spin_lock(&((__sk)->sk_lock.slock))
1369 #define bh_lock_sock_nested(__sk) \
1370 				spin_lock_nested(&((__sk)->sk_lock.slock), \
1371 				SINGLE_DEPTH_NESTING)
1372 #define bh_unlock_sock(__sk)	spin_unlock(&((__sk)->sk_lock.slock))
1373 
1374 extern bool lock_sock_fast(struct sock *sk);
1375 /**
1376  * unlock_sock_fast - complement of lock_sock_fast
1377  * @sk: socket
1378  * @slow: slow mode
1379  *
1380  * fast unlock socket for user context.
1381  * If slow mode is on, we call regular release_sock()
1382  */
unlock_sock_fast(struct sock * sk,bool slow)1383 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1384 {
1385 	if (slow)
1386 		release_sock(sk);
1387 	else
1388 		spin_unlock_bh(&sk->sk_lock.slock);
1389 }
1390 
1391 
1392 extern struct sock		*sk_alloc(struct net *net, int family,
1393 					  gfp_t priority,
1394 					  struct proto *prot);
1395 extern void			sk_free(struct sock *sk);
1396 extern void			sk_release_kernel(struct sock *sk);
1397 extern struct sock		*sk_clone_lock(const struct sock *sk,
1398 					       const gfp_t priority);
1399 
1400 extern struct sk_buff		*sock_wmalloc(struct sock *sk,
1401 					      unsigned long size, int force,
1402 					      gfp_t priority);
1403 extern struct sk_buff		*sock_rmalloc(struct sock *sk,
1404 					      unsigned long size, int force,
1405 					      gfp_t priority);
1406 extern void			sock_wfree(struct sk_buff *skb);
1407 extern void			sock_rfree(struct sk_buff *skb);
1408 
1409 extern int			sock_setsockopt(struct socket *sock, int level,
1410 						int op, char __user *optval,
1411 						unsigned int optlen);
1412 
1413 extern int			sock_getsockopt(struct socket *sock, int level,
1414 						int op, char __user *optval,
1415 						int __user *optlen);
1416 extern struct sk_buff 		*sock_alloc_send_skb(struct sock *sk,
1417 						     unsigned long size,
1418 						     int noblock,
1419 						     int *errcode);
1420 extern struct sk_buff 		*sock_alloc_send_pskb(struct sock *sk,
1421 						      unsigned long header_len,
1422 						      unsigned long data_len,
1423 						      int noblock,
1424 						      int *errcode);
1425 extern void *sock_kmalloc(struct sock *sk, int size,
1426 			  gfp_t priority);
1427 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1428 extern void sk_send_sigurg(struct sock *sk);
1429 
1430 #ifdef CONFIG_CGROUPS
1431 extern void sock_update_classid(struct sock *sk);
1432 #else
sock_update_classid(struct sock * sk)1433 static inline void sock_update_classid(struct sock *sk)
1434 {
1435 }
1436 #endif
1437 
1438 /*
1439  * Functions to fill in entries in struct proto_ops when a protocol
1440  * does not implement a particular function.
1441  */
1442 extern int                      sock_no_bind(struct socket *,
1443 					     struct sockaddr *, int);
1444 extern int                      sock_no_connect(struct socket *,
1445 						struct sockaddr *, int, int);
1446 extern int                      sock_no_socketpair(struct socket *,
1447 						   struct socket *);
1448 extern int                      sock_no_accept(struct socket *,
1449 					       struct socket *, int);
1450 extern int                      sock_no_getname(struct socket *,
1451 						struct sockaddr *, int *, int);
1452 extern unsigned int             sock_no_poll(struct file *, struct socket *,
1453 					     struct poll_table_struct *);
1454 extern int                      sock_no_ioctl(struct socket *, unsigned int,
1455 					      unsigned long);
1456 extern int			sock_no_listen(struct socket *, int);
1457 extern int                      sock_no_shutdown(struct socket *, int);
1458 extern int			sock_no_getsockopt(struct socket *, int , int,
1459 						   char __user *, int __user *);
1460 extern int			sock_no_setsockopt(struct socket *, int, int,
1461 						   char __user *, unsigned int);
1462 extern int                      sock_no_sendmsg(struct kiocb *, struct socket *,
1463 						struct msghdr *, size_t);
1464 extern int                      sock_no_recvmsg(struct kiocb *, struct socket *,
1465 						struct msghdr *, size_t, int);
1466 extern int			sock_no_mmap(struct file *file,
1467 					     struct socket *sock,
1468 					     struct vm_area_struct *vma);
1469 extern ssize_t			sock_no_sendpage(struct socket *sock,
1470 						struct page *page,
1471 						int offset, size_t size,
1472 						int flags);
1473 
1474 /*
1475  * Functions to fill in entries in struct proto_ops when a protocol
1476  * uses the inet style.
1477  */
1478 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1479 				  char __user *optval, int __user *optlen);
1480 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1481 			       struct msghdr *msg, size_t size, int flags);
1482 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1483 				  char __user *optval, unsigned int optlen);
1484 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1485 		int optname, char __user *optval, int __user *optlen);
1486 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1487 		int optname, char __user *optval, unsigned int optlen);
1488 
1489 extern void sk_common_release(struct sock *sk);
1490 
1491 /*
1492  *	Default socket callbacks and setup code
1493  */
1494 
1495 /* Initialise core socket variables */
1496 extern void sock_init_data(struct socket *sock, struct sock *sk);
1497 
1498 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1499 
1500 /**
1501  *	sk_filter_release - release a socket filter
1502  *	@fp: filter to remove
1503  *
1504  *	Remove a filter from a socket and release its resources.
1505  */
1506 
sk_filter_release(struct sk_filter * fp)1507 static inline void sk_filter_release(struct sk_filter *fp)
1508 {
1509 	if (atomic_dec_and_test(&fp->refcnt))
1510 		call_rcu(&fp->rcu, sk_filter_release_rcu);
1511 }
1512 
sk_filter_uncharge(struct sock * sk,struct sk_filter * fp)1513 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1514 {
1515 	unsigned int size = sk_filter_len(fp);
1516 
1517 	atomic_sub(size, &sk->sk_omem_alloc);
1518 	sk_filter_release(fp);
1519 }
1520 
sk_filter_charge(struct sock * sk,struct sk_filter * fp)1521 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1522 {
1523 	atomic_inc(&fp->refcnt);
1524 	atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1525 }
1526 
1527 /*
1528  * Socket reference counting postulates.
1529  *
1530  * * Each user of socket SHOULD hold a reference count.
1531  * * Each access point to socket (an hash table bucket, reference from a list,
1532  *   running timer, skb in flight MUST hold a reference count.
1533  * * When reference count hits 0, it means it will never increase back.
1534  * * When reference count hits 0, it means that no references from
1535  *   outside exist to this socket and current process on current CPU
1536  *   is last user and may/should destroy this socket.
1537  * * sk_free is called from any context: process, BH, IRQ. When
1538  *   it is called, socket has no references from outside -> sk_free
1539  *   may release descendant resources allocated by the socket, but
1540  *   to the time when it is called, socket is NOT referenced by any
1541  *   hash tables, lists etc.
1542  * * Packets, delivered from outside (from network or from another process)
1543  *   and enqueued on receive/error queues SHOULD NOT grab reference count,
1544  *   when they sit in queue. Otherwise, packets will leak to hole, when
1545  *   socket is looked up by one cpu and unhasing is made by another CPU.
1546  *   It is true for udp/raw, netlink (leak to receive and error queues), tcp
1547  *   (leak to backlog). Packet socket does all the processing inside
1548  *   BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1549  *   use separate SMP lock, so that they are prone too.
1550  */
1551 
1552 /* Ungrab socket and destroy it, if it was the last reference. */
sock_put(struct sock * sk)1553 static inline void sock_put(struct sock *sk)
1554 {
1555 	if (atomic_dec_and_test(&sk->sk_refcnt))
1556 		sk_free(sk);
1557 }
1558 
1559 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1560 			  const int nested);
1561 
sk_tx_queue_set(struct sock * sk,int tx_queue)1562 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1563 {
1564 	sk->sk_tx_queue_mapping = tx_queue;
1565 }
1566 
sk_tx_queue_clear(struct sock * sk)1567 static inline void sk_tx_queue_clear(struct sock *sk)
1568 {
1569 	sk->sk_tx_queue_mapping = -1;
1570 }
1571 
sk_tx_queue_get(const struct sock * sk)1572 static inline int sk_tx_queue_get(const struct sock *sk)
1573 {
1574 	return sk ? sk->sk_tx_queue_mapping : -1;
1575 }
1576 
sk_set_socket(struct sock * sk,struct socket * sock)1577 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1578 {
1579 	sk_tx_queue_clear(sk);
1580 	sk->sk_socket = sock;
1581 }
1582 
sk_sleep(struct sock * sk)1583 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1584 {
1585 	BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1586 	return &rcu_dereference_raw(sk->sk_wq)->wait;
1587 }
1588 /* Detach socket from process context.
1589  * Announce socket dead, detach it from wait queue and inode.
1590  * Note that parent inode held reference count on this struct sock,
1591  * we do not release it in this function, because protocol
1592  * probably wants some additional cleanups or even continuing
1593  * to work with this socket (TCP).
1594  */
sock_orphan(struct sock * sk)1595 static inline void sock_orphan(struct sock *sk)
1596 {
1597 	write_lock_bh(&sk->sk_callback_lock);
1598 	sock_set_flag(sk, SOCK_DEAD);
1599 	sk_set_socket(sk, NULL);
1600 	sk->sk_wq  = NULL;
1601 	write_unlock_bh(&sk->sk_callback_lock);
1602 }
1603 
sock_graft(struct sock * sk,struct socket * parent)1604 static inline void sock_graft(struct sock *sk, struct socket *parent)
1605 {
1606 	write_lock_bh(&sk->sk_callback_lock);
1607 	sk->sk_wq = parent->wq;
1608 	parent->sk = sk;
1609 	sk_set_socket(sk, parent);
1610 	security_sock_graft(sk, parent);
1611 	write_unlock_bh(&sk->sk_callback_lock);
1612 }
1613 
1614 extern int sock_i_uid(struct sock *sk);
1615 extern unsigned long sock_i_ino(struct sock *sk);
1616 
1617 static inline struct dst_entry *
__sk_dst_get(struct sock * sk)1618 __sk_dst_get(struct sock *sk)
1619 {
1620 	return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1621 						       lockdep_is_held(&sk->sk_lock.slock));
1622 }
1623 
1624 static inline struct dst_entry *
sk_dst_get(struct sock * sk)1625 sk_dst_get(struct sock *sk)
1626 {
1627 	struct dst_entry *dst;
1628 
1629 	rcu_read_lock();
1630 	dst = rcu_dereference(sk->sk_dst_cache);
1631 	if (dst)
1632 		dst_hold(dst);
1633 	rcu_read_unlock();
1634 	return dst;
1635 }
1636 
1637 extern void sk_reset_txq(struct sock *sk);
1638 
dst_negative_advice(struct sock * sk)1639 static inline void dst_negative_advice(struct sock *sk)
1640 {
1641 	struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1642 
1643 	if (dst && dst->ops->negative_advice) {
1644 		ndst = dst->ops->negative_advice(dst);
1645 
1646 		if (ndst != dst) {
1647 			rcu_assign_pointer(sk->sk_dst_cache, ndst);
1648 			sk_reset_txq(sk);
1649 		}
1650 	}
1651 }
1652 
1653 static inline void
__sk_dst_set(struct sock * sk,struct dst_entry * dst)1654 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1655 {
1656 	struct dst_entry *old_dst;
1657 
1658 	sk_tx_queue_clear(sk);
1659 	/*
1660 	 * This can be called while sk is owned by the caller only,
1661 	 * with no state that can be checked in a rcu_dereference_check() cond
1662 	 */
1663 	old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1664 	rcu_assign_pointer(sk->sk_dst_cache, dst);
1665 	dst_release(old_dst);
1666 }
1667 
1668 static inline void
sk_dst_set(struct sock * sk,struct dst_entry * dst)1669 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1670 {
1671 	spin_lock(&sk->sk_dst_lock);
1672 	__sk_dst_set(sk, dst);
1673 	spin_unlock(&sk->sk_dst_lock);
1674 }
1675 
1676 static inline void
__sk_dst_reset(struct sock * sk)1677 __sk_dst_reset(struct sock *sk)
1678 {
1679 	__sk_dst_set(sk, NULL);
1680 }
1681 
1682 static inline void
sk_dst_reset(struct sock * sk)1683 sk_dst_reset(struct sock *sk)
1684 {
1685 	spin_lock(&sk->sk_dst_lock);
1686 	__sk_dst_reset(sk);
1687 	spin_unlock(&sk->sk_dst_lock);
1688 }
1689 
1690 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1691 
1692 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1693 
sk_can_gso(const struct sock * sk)1694 static inline int sk_can_gso(const struct sock *sk)
1695 {
1696 	return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1697 }
1698 
1699 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1700 
sk_nocaps_add(struct sock * sk,netdev_features_t flags)1701 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1702 {
1703 	sk->sk_route_nocaps |= flags;
1704 	sk->sk_route_caps &= ~flags;
1705 }
1706 
skb_do_copy_data_nocache(struct sock * sk,struct sk_buff * skb,char __user * from,char * to,int copy,int offset)1707 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1708 					   char __user *from, char *to,
1709 					   int copy, int offset)
1710 {
1711 	if (skb->ip_summed == CHECKSUM_NONE) {
1712 		int err = 0;
1713 		__wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1714 		if (err)
1715 			return err;
1716 		skb->csum = csum_block_add(skb->csum, csum, offset);
1717 	} else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1718 		if (!access_ok(VERIFY_READ, from, copy) ||
1719 		    __copy_from_user_nocache(to, from, copy))
1720 			return -EFAULT;
1721 	} else if (copy_from_user(to, from, copy))
1722 		return -EFAULT;
1723 
1724 	return 0;
1725 }
1726 
skb_add_data_nocache(struct sock * sk,struct sk_buff * skb,char __user * from,int copy)1727 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1728 				       char __user *from, int copy)
1729 {
1730 	int err, offset = skb->len;
1731 
1732 	err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1733 				       copy, offset);
1734 	if (err)
1735 		__skb_trim(skb, offset);
1736 
1737 	return err;
1738 }
1739 
skb_copy_to_page_nocache(struct sock * sk,char __user * from,struct sk_buff * skb,struct page * page,int off,int copy)1740 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1741 					   struct sk_buff *skb,
1742 					   struct page *page,
1743 					   int off, int copy)
1744 {
1745 	int err;
1746 
1747 	err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1748 				       copy, skb->len);
1749 	if (err)
1750 		return err;
1751 
1752 	skb->len	     += copy;
1753 	skb->data_len	     += copy;
1754 	skb->truesize	     += copy;
1755 	sk->sk_wmem_queued   += copy;
1756 	sk_mem_charge(sk, copy);
1757 	return 0;
1758 }
1759 
skb_copy_to_page(struct sock * sk,char __user * from,struct sk_buff * skb,struct page * page,int off,int copy)1760 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1761 				   struct sk_buff *skb, struct page *page,
1762 				   int off, int copy)
1763 {
1764 	if (skb->ip_summed == CHECKSUM_NONE) {
1765 		int err = 0;
1766 		__wsum csum = csum_and_copy_from_user(from,
1767 						     page_address(page) + off,
1768 							    copy, 0, &err);
1769 		if (err)
1770 			return err;
1771 		skb->csum = csum_block_add(skb->csum, csum, skb->len);
1772 	} else if (copy_from_user(page_address(page) + off, from, copy))
1773 		return -EFAULT;
1774 
1775 	skb->len	     += copy;
1776 	skb->data_len	     += copy;
1777 	skb->truesize	     += copy;
1778 	sk->sk_wmem_queued   += copy;
1779 	sk_mem_charge(sk, copy);
1780 	return 0;
1781 }
1782 
1783 /**
1784  * sk_wmem_alloc_get - returns write allocations
1785  * @sk: socket
1786  *
1787  * Returns sk_wmem_alloc minus initial offset of one
1788  */
sk_wmem_alloc_get(const struct sock * sk)1789 static inline int sk_wmem_alloc_get(const struct sock *sk)
1790 {
1791 	return atomic_read(&sk->sk_wmem_alloc) - 1;
1792 }
1793 
1794 /**
1795  * sk_rmem_alloc_get - returns read allocations
1796  * @sk: socket
1797  *
1798  * Returns sk_rmem_alloc
1799  */
sk_rmem_alloc_get(const struct sock * sk)1800 static inline int sk_rmem_alloc_get(const struct sock *sk)
1801 {
1802 	return atomic_read(&sk->sk_rmem_alloc);
1803 }
1804 
1805 /**
1806  * sk_has_allocations - check if allocations are outstanding
1807  * @sk: socket
1808  *
1809  * Returns true if socket has write or read allocations
1810  */
sk_has_allocations(const struct sock * sk)1811 static inline int sk_has_allocations(const struct sock *sk)
1812 {
1813 	return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1814 }
1815 
1816 /**
1817  * wq_has_sleeper - check if there are any waiting processes
1818  * @wq: struct socket_wq
1819  *
1820  * Returns true if socket_wq has waiting processes
1821  *
1822  * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1823  * barrier call. They were added due to the race found within the tcp code.
1824  *
1825  * Consider following tcp code paths:
1826  *
1827  * CPU1                  CPU2
1828  *
1829  * sys_select            receive packet
1830  *   ...                 ...
1831  *   __add_wait_queue    update tp->rcv_nxt
1832  *   ...                 ...
1833  *   tp->rcv_nxt check   sock_def_readable
1834  *   ...                 {
1835  *   schedule               rcu_read_lock();
1836  *                          wq = rcu_dereference(sk->sk_wq);
1837  *                          if (wq && waitqueue_active(&wq->wait))
1838  *                              wake_up_interruptible(&wq->wait)
1839  *                          ...
1840  *                       }
1841  *
1842  * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1843  * in its cache, and so does the tp->rcv_nxt update on CPU2 side.  The CPU1
1844  * could then endup calling schedule and sleep forever if there are no more
1845  * data on the socket.
1846  *
1847  */
wq_has_sleeper(struct socket_wq * wq)1848 static inline bool wq_has_sleeper(struct socket_wq *wq)
1849 {
1850 
1851 	/*
1852 	 * We need to be sure we are in sync with the
1853 	 * add_wait_queue modifications to the wait queue.
1854 	 *
1855 	 * This memory barrier is paired in the sock_poll_wait.
1856 	 */
1857 	smp_mb();
1858 	return wq && waitqueue_active(&wq->wait);
1859 }
1860 
1861 /**
1862  * sock_poll_wait - place memory barrier behind the poll_wait call.
1863  * @filp:           file
1864  * @wait_address:   socket wait queue
1865  * @p:              poll_table
1866  *
1867  * See the comments in the wq_has_sleeper function.
1868  */
sock_poll_wait(struct file * filp,wait_queue_head_t * wait_address,poll_table * p)1869 static inline void sock_poll_wait(struct file *filp,
1870 		wait_queue_head_t *wait_address, poll_table *p)
1871 {
1872 	if (!poll_does_not_wait(p) && wait_address) {
1873 		poll_wait(filp, wait_address, p);
1874 		/*
1875 		 * We need to be sure we are in sync with the
1876 		 * socket flags modification.
1877 		 *
1878 		 * This memory barrier is paired in the wq_has_sleeper.
1879 		*/
1880 		smp_mb();
1881 	}
1882 }
1883 
1884 /*
1885  * 	Queue a received datagram if it will fit. Stream and sequenced
1886  *	protocols can't normally use this as they need to fit buffers in
1887  *	and play with them.
1888  *
1889  * 	Inlined as it's very short and called for pretty much every
1890  *	packet ever received.
1891  */
1892 
skb_set_owner_w(struct sk_buff * skb,struct sock * sk)1893 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1894 {
1895 	skb_orphan(skb);
1896 	skb->sk = sk;
1897 	skb->destructor = sock_wfree;
1898 	/*
1899 	 * We used to take a refcount on sk, but following operation
1900 	 * is enough to guarantee sk_free() wont free this sock until
1901 	 * all in-flight packets are completed
1902 	 */
1903 	atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1904 }
1905 
skb_set_owner_r(struct sk_buff * skb,struct sock * sk)1906 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1907 {
1908 	skb_orphan(skb);
1909 	skb->sk = sk;
1910 	skb->destructor = sock_rfree;
1911 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1912 	sk_mem_charge(sk, skb->truesize);
1913 }
1914 
1915 extern void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1916 			   unsigned long expires);
1917 
1918 extern void sk_stop_timer(struct sock *sk, struct timer_list* timer);
1919 
1920 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1921 
1922 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1923 
1924 /*
1925  *	Recover an error report and clear atomically
1926  */
1927 
sock_error(struct sock * sk)1928 static inline int sock_error(struct sock *sk)
1929 {
1930 	int err;
1931 	if (likely(!sk->sk_err))
1932 		return 0;
1933 	err = xchg(&sk->sk_err, 0);
1934 	return -err;
1935 }
1936 
sock_wspace(struct sock * sk)1937 static inline unsigned long sock_wspace(struct sock *sk)
1938 {
1939 	int amt = 0;
1940 
1941 	if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1942 		amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1943 		if (amt < 0)
1944 			amt = 0;
1945 	}
1946 	return amt;
1947 }
1948 
sk_wake_async(struct sock * sk,int how,int band)1949 static inline void sk_wake_async(struct sock *sk, int how, int band)
1950 {
1951 	if (sock_flag(sk, SOCK_FASYNC))
1952 		sock_wake_async(sk->sk_socket, how, band);
1953 }
1954 
1955 #define SOCK_MIN_SNDBUF 2048
1956 /*
1957  * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
1958  * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
1959  */
1960 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
1961 
sk_stream_moderate_sndbuf(struct sock * sk)1962 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
1963 {
1964 	if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
1965 		sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
1966 		sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
1967 	}
1968 }
1969 
1970 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
1971 
sk_stream_alloc_page(struct sock * sk)1972 static inline struct page *sk_stream_alloc_page(struct sock *sk)
1973 {
1974 	struct page *page = NULL;
1975 
1976 	page = alloc_pages(sk->sk_allocation, 0);
1977 	if (!page) {
1978 		sk_enter_memory_pressure(sk);
1979 		sk_stream_moderate_sndbuf(sk);
1980 	}
1981 	return page;
1982 }
1983 
1984 /*
1985  *	Default write policy as shown to user space via poll/select/SIGIO
1986  */
sock_writeable(const struct sock * sk)1987 static inline int sock_writeable(const struct sock *sk)
1988 {
1989 	return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
1990 }
1991 
gfp_any(void)1992 static inline gfp_t gfp_any(void)
1993 {
1994 	return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
1995 }
1996 
sock_rcvtimeo(const struct sock * sk,int noblock)1997 static inline long sock_rcvtimeo(const struct sock *sk, int noblock)
1998 {
1999 	return noblock ? 0 : sk->sk_rcvtimeo;
2000 }
2001 
sock_sndtimeo(const struct sock * sk,int noblock)2002 static inline long sock_sndtimeo(const struct sock *sk, int noblock)
2003 {
2004 	return noblock ? 0 : sk->sk_sndtimeo;
2005 }
2006 
sock_rcvlowat(const struct sock * sk,int waitall,int len)2007 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2008 {
2009 	return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2010 }
2011 
2012 /* Alas, with timeout socket operations are not restartable.
2013  * Compare this to poll().
2014  */
sock_intr_errno(long timeo)2015 static inline int sock_intr_errno(long timeo)
2016 {
2017 	return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2018 }
2019 
2020 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2021 	struct sk_buff *skb);
2022 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2023 	struct sk_buff *skb);
2024 
2025 static __inline__ void
sock_recv_timestamp(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2026 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2027 {
2028 	ktime_t kt = skb->tstamp;
2029 	struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2030 
2031 	/*
2032 	 * generate control messages if
2033 	 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2034 	 *   or SOCK_TIMESTAMPING_RX_SOFTWARE)
2035 	 * - software time stamp available and wanted
2036 	 *   (SOCK_TIMESTAMPING_SOFTWARE)
2037 	 * - hardware time stamps available and wanted
2038 	 *   (SOCK_TIMESTAMPING_SYS_HARDWARE or
2039 	 *   SOCK_TIMESTAMPING_RAW_HARDWARE)
2040 	 */
2041 	if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2042 	    sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2043 	    (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2044 	    (hwtstamps->hwtstamp.tv64 &&
2045 	     sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2046 	    (hwtstamps->syststamp.tv64 &&
2047 	     sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2048 		__sock_recv_timestamp(msg, sk, skb);
2049 	else
2050 		sk->sk_stamp = kt;
2051 
2052 	if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2053 		__sock_recv_wifi_status(msg, sk, skb);
2054 }
2055 
2056 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2057 				     struct sk_buff *skb);
2058 
sock_recv_ts_and_drops(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2059 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2060 					  struct sk_buff *skb)
2061 {
2062 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL)			| \
2063 			   (1UL << SOCK_RCVTSTAMP)			| \
2064 			   (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE)	| \
2065 			   (1UL << SOCK_TIMESTAMPING_SOFTWARE)		| \
2066 			   (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) 	| \
2067 			   (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2068 
2069 	if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2070 		__sock_recv_ts_and_drops(msg, sk, skb);
2071 	else
2072 		sk->sk_stamp = skb->tstamp;
2073 }
2074 
2075 /**
2076  * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2077  * @sk:		socket sending this packet
2078  * @tx_flags:	filled with instructions for time stamping
2079  *
2080  * Currently only depends on SOCK_TIMESTAMPING* flags. Returns error code if
2081  * parameters are invalid.
2082  */
2083 extern int sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2084 
2085 /**
2086  * sk_eat_skb - Release a skb if it is no longer needed
2087  * @sk: socket to eat this skb from
2088  * @skb: socket buffer to eat
2089  * @copied_early: flag indicating whether DMA operations copied this data early
2090  *
2091  * This routine must be called with interrupts disabled or with the socket
2092  * locked so that the sk_buff queue operation is ok.
2093 */
2094 #ifdef CONFIG_NET_DMA
sk_eat_skb(struct sock * sk,struct sk_buff * skb,int copied_early)2095 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
2096 {
2097 	__skb_unlink(skb, &sk->sk_receive_queue);
2098 	if (!copied_early)
2099 		__kfree_skb(skb);
2100 	else
2101 		__skb_queue_tail(&sk->sk_async_wait_queue, skb);
2102 }
2103 #else
sk_eat_skb(struct sock * sk,struct sk_buff * skb,int copied_early)2104 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, int copied_early)
2105 {
2106 	__skb_unlink(skb, &sk->sk_receive_queue);
2107 	__kfree_skb(skb);
2108 }
2109 #endif
2110 
2111 static inline
sock_net(const struct sock * sk)2112 struct net *sock_net(const struct sock *sk)
2113 {
2114 	return read_pnet(&sk->sk_net);
2115 }
2116 
2117 static inline
sock_net_set(struct sock * sk,struct net * net)2118 void sock_net_set(struct sock *sk, struct net *net)
2119 {
2120 	write_pnet(&sk->sk_net, net);
2121 }
2122 
2123 /*
2124  * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2125  * They should not hold a reference to a namespace in order to allow
2126  * to stop it.
2127  * Sockets after sk_change_net should be released using sk_release_kernel
2128  */
sk_change_net(struct sock * sk,struct net * net)2129 static inline void sk_change_net(struct sock *sk, struct net *net)
2130 {
2131 	put_net(sock_net(sk));
2132 	sock_net_set(sk, hold_net(net));
2133 }
2134 
skb_steal_sock(struct sk_buff * skb)2135 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2136 {
2137 	if (unlikely(skb->sk)) {
2138 		struct sock *sk = skb->sk;
2139 
2140 		skb->destructor = NULL;
2141 		skb->sk = NULL;
2142 		return sk;
2143 	}
2144 	return NULL;
2145 }
2146 
2147 extern void sock_enable_timestamp(struct sock *sk, int flag);
2148 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2149 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2150 
2151 /*
2152  *	Enable debug/info messages
2153  */
2154 extern int net_msg_warn;
2155 #define NETDEBUG(fmt, args...) \
2156 	do { if (net_msg_warn) printk(fmt,##args); } while (0)
2157 
2158 #define LIMIT_NETDEBUG(fmt, args...) \
2159 	do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2160 
2161 extern __u32 sysctl_wmem_max;
2162 extern __u32 sysctl_rmem_max;
2163 
2164 extern void sk_init(void);
2165 
2166 extern int sysctl_optmem_max;
2167 
2168 extern __u32 sysctl_wmem_default;
2169 extern __u32 sysctl_rmem_default;
2170 
2171 #endif	/* _SOCK_H */
2172