Merge tag 'v3.10.95' into update
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / include / net / sock.h
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_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
234 * @sk_sndbuf: size of send buffer in bytes
235 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
236 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
237 * @sk_no_check: %SO_NO_CHECK setting, whether or not checkup packets
238 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
239 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
240 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
241 * @sk_gso_max_size: Maximum GSO segment size to build
242 * @sk_gso_max_segs: Maximum number of GSO segments
243 * @sk_lingertime: %SO_LINGER l_linger setting
244 * @sk_backlog: always used with the per-socket spinlock held
245 * @sk_callback_lock: used with the callbacks in the end of this struct
246 * @sk_error_queue: rarely used
247 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
248 * IPV6_ADDRFORM for instance)
249 * @sk_err: last error
250 * @sk_err_soft: errors that don't cause failure but are the cause of a
251 * persistent failure not just 'timed out'
252 * @sk_drops: raw/udp drops counter
253 * @sk_ack_backlog: current listen backlog
254 * @sk_max_ack_backlog: listen backlog set in listen()
255 * @sk_priority: %SO_PRIORITY setting
256 * @sk_cgrp_prioidx: socket group's priority map index
257 * @sk_type: socket type (%SOCK_STREAM, etc)
258 * @sk_protocol: which protocol this socket belongs in this network family
259 * @sk_peer_pid: &struct pid for this socket's peer
260 * @sk_peer_cred: %SO_PEERCRED setting
261 * @sk_rcvlowat: %SO_RCVLOWAT setting
262 * @sk_rcvtimeo: %SO_RCVTIMEO setting
263 * @sk_sndtimeo: %SO_SNDTIMEO setting
264 * @sk_rxhash: flow hash received from netif layer
265 * @sk_filter: socket filtering instructions
266 * @sk_protinfo: private area, net family specific, when not using slab
267 * @sk_timer: sock cleanup timer
268 * @sk_stamp: time stamp of last packet received
269 * @sk_socket: Identd and reporting IO signals
270 * @sk_user_data: RPC layer private data
271 * @sk_frag: cached page frag
272 * @sk_peek_off: current peek_offset value
273 * @sk_send_head: front of stuff to transmit
274 * @sk_security: used by security modules
275 * @sk_mark: generic packet mark
276 * @sk_classid: this socket's cgroup classid
277 * @sk_cgrp: this socket's cgroup-specific proto data
278 * @sk_write_pending: a write to stream socket waits to start
279 * @sk_state_change: callback to indicate change in the state of the sock
280 * @sk_data_ready: callback to indicate there is data to be processed
281 * @sk_write_space: callback to indicate there is bf sending space available
282 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
283 * @sk_backlog_rcv: callback to process the backlog
284 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
285 */
286 struct sock {
287 /*
288 * Now struct inet_timewait_sock also uses sock_common, so please just
289 * don't add nothing before this first member (__sk_common) --acme
290 */
291 struct sock_common __sk_common;
292 #define sk_node __sk_common.skc_node
293 #define sk_nulls_node __sk_common.skc_nulls_node
294 #define sk_refcnt __sk_common.skc_refcnt
295 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
296
297 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
298 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
299 #define sk_hash __sk_common.skc_hash
300 #define sk_family __sk_common.skc_family
301 #define sk_state __sk_common.skc_state
302 #define sk_reuse __sk_common.skc_reuse
303 #define sk_reuseport __sk_common.skc_reuseport
304 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
305 #define sk_bind_node __sk_common.skc_bind_node
306 #define sk_prot __sk_common.skc_prot
307 #define sk_net __sk_common.skc_net
308 socket_lock_t sk_lock;
309 struct sk_buff_head sk_receive_queue;
310 /*
311 * The backlog queue is special, it is always used with
312 * the per-socket spinlock held and requires low latency
313 * access. Therefore we special case it's implementation.
314 * Note : rmem_alloc is in this structure to fill a hole
315 * on 64bit arches, not because its logically part of
316 * backlog.
317 */
318 struct {
319 atomic_t rmem_alloc;
320 int len;
321 struct sk_buff *head;
322 struct sk_buff *tail;
323 } sk_backlog;
324 #define sk_rmem_alloc sk_backlog.rmem_alloc
325 int sk_forward_alloc;
326 #ifdef CONFIG_RPS
327 __u32 sk_rxhash;
328 #endif
329 atomic_t sk_drops;
330 int sk_rcvbuf;
331
332 struct sk_filter __rcu *sk_filter;
333 struct socket_wq __rcu *sk_wq;
334
335 #ifdef CONFIG_NET_DMA
336 struct sk_buff_head sk_async_wait_queue;
337 #endif
338
339 #ifdef CONFIG_XFRM
340 struct xfrm_policy *sk_policy[2];
341 #endif
342 unsigned long sk_flags;
343 struct dst_entry *sk_rx_dst;
344 struct dst_entry __rcu *sk_dst_cache;
345 spinlock_t sk_dst_lock;
346 atomic_t sk_wmem_alloc;
347 atomic_t sk_omem_alloc;
348 int sk_sndbuf;
349 struct sk_buff_head sk_write_queue;
350 kmemcheck_bitfield_begin(flags);
351 unsigned int sk_shutdown : 2,
352 sk_no_check : 2,
353 sk_userlocks : 4,
354 sk_protocol : 8,
355 #define SK_PROTOCOL_MAX U8_MAX
356 sk_type : 16;
357 kmemcheck_bitfield_end(flags);
358 int sk_wmem_queued;
359 gfp_t sk_allocation;
360 u32 sk_pacing_rate; /* bytes per second */
361 netdev_features_t sk_route_caps;
362 netdev_features_t sk_route_nocaps;
363 int sk_gso_type;
364 unsigned int sk_gso_max_size;
365 u16 sk_gso_max_segs;
366 int sk_rcvlowat;
367 unsigned long sk_lingertime;
368 struct sk_buff_head sk_error_queue;
369 struct proto *sk_prot_creator;
370 rwlock_t sk_callback_lock;
371 int sk_err,
372 sk_err_soft;
373 unsigned short sk_ack_backlog;
374 unsigned short sk_max_ack_backlog;
375 __u32 sk_priority;
376 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
377 __u32 sk_cgrp_prioidx;
378 #endif
379 struct pid *sk_peer_pid;
380 const struct cred *sk_peer_cred;
381 long sk_rcvtimeo;
382 long sk_sndtimeo;
383 void *sk_protinfo;
384 struct timer_list sk_timer;
385 ktime_t sk_stamp;
386 struct socket *sk_socket;
387 void *sk_user_data;
388 struct page_frag sk_frag;
389 struct sk_buff *sk_send_head;
390 __s32 sk_peek_off;
391 int sk_write_pending;
392 #ifdef CONFIG_SECURITY
393 void *sk_security;
394 #endif
395 __u32 sk_mark;
396 u32 sk_classid;
397 struct cg_proto *sk_cgrp;
398 void (*sk_state_change)(struct sock *sk);
399 void (*sk_data_ready)(struct sock *sk, int bytes);
400 void (*sk_write_space)(struct sock *sk);
401 void (*sk_error_report)(struct sock *sk);
402 int (*sk_backlog_rcv)(struct sock *sk,
403 struct sk_buff *skb);
404 void (*sk_destruct)(struct sock *sk);
405 };
406
407 /*
408 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
409 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
410 * on a socket means that the socket will reuse everybody else's port
411 * without looking at the other's sk_reuse value.
412 */
413
414 #define SK_NO_REUSE 0
415 #define SK_CAN_REUSE 1
416 #define SK_FORCE_REUSE 2
417
418 static inline int sk_peek_offset(struct sock *sk, int flags)
419 {
420 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
421 return sk->sk_peek_off;
422 else
423 return 0;
424 }
425
426 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
427 {
428 if (sk->sk_peek_off >= 0) {
429 if (sk->sk_peek_off >= val)
430 sk->sk_peek_off -= val;
431 else
432 sk->sk_peek_off = 0;
433 }
434 }
435
436 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
437 {
438 if (sk->sk_peek_off >= 0)
439 sk->sk_peek_off += val;
440 }
441
442 /*
443 * Hashed lists helper routines
444 */
445 static inline struct sock *sk_entry(const struct hlist_node *node)
446 {
447 return hlist_entry(node, struct sock, sk_node);
448 }
449
450 static inline struct sock *__sk_head(const struct hlist_head *head)
451 {
452 return hlist_entry(head->first, struct sock, sk_node);
453 }
454
455 static inline struct sock *sk_head(const struct hlist_head *head)
456 {
457 return hlist_empty(head) ? NULL : __sk_head(head);
458 }
459
460 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
461 {
462 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
463 }
464
465 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
466 {
467 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
468 }
469
470 static inline struct sock *sk_next(const struct sock *sk)
471 {
472 return sk->sk_node.next ?
473 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
474 }
475
476 static inline struct sock *sk_nulls_next(const struct sock *sk)
477 {
478 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
479 hlist_nulls_entry(sk->sk_nulls_node.next,
480 struct sock, sk_nulls_node) :
481 NULL;
482 }
483
484 static inline bool sk_unhashed(const struct sock *sk)
485 {
486 return hlist_unhashed(&sk->sk_node);
487 }
488
489 static inline bool sk_hashed(const struct sock *sk)
490 {
491 return !sk_unhashed(sk);
492 }
493
494 static inline void sk_node_init(struct hlist_node *node)
495 {
496 node->pprev = NULL;
497 }
498
499 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
500 {
501 node->pprev = NULL;
502 }
503
504 static inline void __sk_del_node(struct sock *sk)
505 {
506 __hlist_del(&sk->sk_node);
507 }
508
509 /* NB: equivalent to hlist_del_init_rcu */
510 static inline bool __sk_del_node_init(struct sock *sk)
511 {
512 if (sk_hashed(sk)) {
513 __sk_del_node(sk);
514 sk_node_init(&sk->sk_node);
515 return true;
516 }
517 return false;
518 }
519
520 /* Grab socket reference count. This operation is valid only
521 when sk is ALREADY grabbed f.e. it is found in hash table
522 or a list and the lookup is made under lock preventing hash table
523 modifications.
524 */
525
526 static inline void sock_hold(struct sock *sk)
527 {
528 atomic_inc(&sk->sk_refcnt);
529 }
530
531 /* Ungrab socket in the context, which assumes that socket refcnt
532 cannot hit zero, f.e. it is true in context of any socketcall.
533 */
534 static inline void __sock_put(struct sock *sk)
535 {
536 atomic_dec(&sk->sk_refcnt);
537 }
538
539 static inline bool sk_del_node_init(struct sock *sk)
540 {
541 bool rc = __sk_del_node_init(sk);
542
543 if (rc) {
544 /* paranoid for a while -acme */
545 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
546 __sock_put(sk);
547 }
548 return rc;
549 }
550 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
551
552 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
553 {
554 if (sk_hashed(sk)) {
555 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
556 return true;
557 }
558 return false;
559 }
560
561 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
562 {
563 bool rc = __sk_nulls_del_node_init_rcu(sk);
564
565 if (rc) {
566 /* paranoid for a while -acme */
567 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
568 __sock_put(sk);
569 }
570 return rc;
571 }
572
573 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
574 {
575 hlist_add_head(&sk->sk_node, list);
576 }
577
578 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
579 {
580 sock_hold(sk);
581 __sk_add_node(sk, list);
582 }
583
584 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
585 {
586 sock_hold(sk);
587 hlist_add_head_rcu(&sk->sk_node, list);
588 }
589
590 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
591 {
592 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
593 }
594
595 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
596 {
597 sock_hold(sk);
598 __sk_nulls_add_node_rcu(sk, list);
599 }
600
601 static inline void __sk_del_bind_node(struct sock *sk)
602 {
603 __hlist_del(&sk->sk_bind_node);
604 }
605
606 static inline void sk_add_bind_node(struct sock *sk,
607 struct hlist_head *list)
608 {
609 hlist_add_head(&sk->sk_bind_node, list);
610 }
611
612 #define sk_for_each(__sk, list) \
613 hlist_for_each_entry(__sk, list, sk_node)
614 #define sk_for_each_rcu(__sk, list) \
615 hlist_for_each_entry_rcu(__sk, list, sk_node)
616 #define sk_nulls_for_each(__sk, node, list) \
617 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
618 #define sk_nulls_for_each_rcu(__sk, node, list) \
619 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
620 #define sk_for_each_from(__sk) \
621 hlist_for_each_entry_from(__sk, sk_node)
622 #define sk_nulls_for_each_from(__sk, node) \
623 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
624 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
625 #define sk_for_each_safe(__sk, tmp, list) \
626 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
627 #define sk_for_each_bound(__sk, list) \
628 hlist_for_each_entry(__sk, list, sk_bind_node)
629
630 static inline struct user_namespace *sk_user_ns(struct sock *sk)
631 {
632 /* Careful only use this in a context where these parameters
633 * can not change and must all be valid, such as recvmsg from
634 * userspace.
635 */
636 return sk->sk_socket->file->f_cred->user_ns;
637 }
638
639 /* Sock flags */
640 enum sock_flags {
641 SOCK_DEAD,
642 SOCK_DONE,
643 SOCK_URGINLINE,
644 SOCK_KEEPOPEN,
645 SOCK_LINGER,
646 SOCK_DESTROY,
647 SOCK_BROADCAST,
648 SOCK_TIMESTAMP,
649 SOCK_ZAPPED,
650 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
651 SOCK_DBG, /* %SO_DEBUG setting */
652 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
653 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
654 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
655 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
656 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
657 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
658 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
659 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
660 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
661 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
662 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
663 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
664 SOCK_FASYNC, /* fasync() active */
665 SOCK_RXQ_OVFL,
666 SOCK_ZEROCOPY, /* buffers from userspace */
667 SOCK_WIFI_STATUS, /* push wifi status to userspace */
668 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
669 * Will use last 4 bytes of packet sent from
670 * user-space instead.
671 */
672 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
673 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
674 };
675
676 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
677
678 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
679 {
680 nsk->sk_flags = osk->sk_flags;
681 }
682
683 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
684 {
685 __set_bit(flag, &sk->sk_flags);
686 }
687
688 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
689 {
690 __clear_bit(flag, &sk->sk_flags);
691 }
692
693 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
694 {
695 return test_bit(flag, &sk->sk_flags);
696 }
697
698 #ifdef CONFIG_NET
699 extern struct static_key memalloc_socks;
700 static inline int sk_memalloc_socks(void)
701 {
702 return static_key_false(&memalloc_socks);
703 }
704 #else
705
706 static inline int sk_memalloc_socks(void)
707 {
708 return 0;
709 }
710
711 #endif
712
713 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
714 {
715 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
716 }
717
718 static inline void sk_acceptq_removed(struct sock *sk)
719 {
720 sk->sk_ack_backlog--;
721 }
722
723 static inline void sk_acceptq_added(struct sock *sk)
724 {
725 sk->sk_ack_backlog++;
726 }
727
728 static inline bool sk_acceptq_is_full(const struct sock *sk)
729 {
730 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
731 }
732
733 /*
734 * Compute minimal free write space needed to queue new packets.
735 */
736 static inline int sk_stream_min_wspace(const struct sock *sk)
737 {
738 return sk->sk_wmem_queued >> 1;
739 }
740
741 static inline int sk_stream_wspace(const struct sock *sk)
742 {
743 return sk->sk_sndbuf - sk->sk_wmem_queued;
744 }
745
746 extern void sk_stream_write_space(struct sock *sk);
747
748 static inline bool sk_stream_memory_free(const struct sock *sk)
749 {
750 return sk->sk_wmem_queued < sk->sk_sndbuf;
751 }
752
753 /* OOB backlog add */
754 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
755 {
756 /* dont let skb dst not refcounted, we are going to leave rcu lock */
757 skb_dst_force(skb);
758
759 if (!sk->sk_backlog.tail)
760 sk->sk_backlog.head = skb;
761 else
762 sk->sk_backlog.tail->next = skb;
763
764 sk->sk_backlog.tail = skb;
765 skb->next = NULL;
766 }
767
768 /*
769 * Take into account size of receive queue and backlog queue
770 * Do not take into account this skb truesize,
771 * to allow even a single big packet to come.
772 */
773 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
774 unsigned int limit)
775 {
776 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
777
778 return qsize > limit;
779 }
780
781 /* The per-socket spinlock must be held here. */
782 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
783 unsigned int limit)
784 {
785 if (sk_rcvqueues_full(sk, skb, limit))
786 {
787 #ifdef CONFIG_MTK_NET_LOGGING
788 printk(KERN_ERR "[mtk_net][sock]sk_add_backlog->sk_rcvqueues_full sk->sk_rcvbuf:%d,sk->sk_sndbuf:%d ",sk->sk_rcvbuf,sk->sk_sndbuf);
789 #endif
790 return -ENOBUFS;
791 }
792
793 /*
794 * If the skb was allocated from pfmemalloc reserves, only
795 * allow SOCK_MEMALLOC sockets to use it as this socket is
796 * helping free memory
797 */
798 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
799 return -ENOMEM;
800
801 __sk_add_backlog(sk, skb);
802 sk->sk_backlog.len += skb->truesize;
803 return 0;
804 }
805
806 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
807
808 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
809 {
810 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
811 return __sk_backlog_rcv(sk, skb);
812
813 return sk->sk_backlog_rcv(sk, skb);
814 }
815
816 static inline void sock_rps_record_flow(const struct sock *sk)
817 {
818 #ifdef CONFIG_RPS
819 struct rps_sock_flow_table *sock_flow_table;
820
821 rcu_read_lock();
822 sock_flow_table = rcu_dereference(rps_sock_flow_table);
823 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
824 rcu_read_unlock();
825 #endif
826 }
827
828 static inline void sock_rps_reset_flow(const struct sock *sk)
829 {
830 #ifdef CONFIG_RPS
831 struct rps_sock_flow_table *sock_flow_table;
832
833 rcu_read_lock();
834 sock_flow_table = rcu_dereference(rps_sock_flow_table);
835 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
836 rcu_read_unlock();
837 #endif
838 }
839
840 static inline void sock_rps_save_rxhash(struct sock *sk,
841 const struct sk_buff *skb)
842 {
843 #ifdef CONFIG_RPS
844 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
845 sock_rps_reset_flow(sk);
846 sk->sk_rxhash = skb->rxhash;
847 }
848 #endif
849 }
850
851 static inline void sock_rps_reset_rxhash(struct sock *sk)
852 {
853 #ifdef CONFIG_RPS
854 sock_rps_reset_flow(sk);
855 sk->sk_rxhash = 0;
856 #endif
857 }
858
859 #define sk_wait_event(__sk, __timeo, __condition) \
860 ({ int __rc; \
861 release_sock(__sk); \
862 __rc = __condition; \
863 if (!__rc) { \
864 *(__timeo) = schedule_timeout(*(__timeo)); \
865 } \
866 lock_sock(__sk); \
867 __rc = __condition; \
868 __rc; \
869 })
870
871 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
872 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
873 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
874 extern int sk_stream_error(struct sock *sk, int flags, int err);
875 extern void sk_stream_kill_queues(struct sock *sk);
876 extern void sk_set_memalloc(struct sock *sk);
877 extern void sk_clear_memalloc(struct sock *sk);
878
879 extern int sk_wait_data(struct sock *sk, long *timeo);
880
881 struct request_sock_ops;
882 struct timewait_sock_ops;
883 struct inet_hashinfo;
884 struct raw_hashinfo;
885 struct module;
886
887 /*
888 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
889 * un-modified. Special care is taken when initializing object to zero.
890 */
891 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
892 {
893 if (offsetof(struct sock, sk_node.next) != 0)
894 memset(sk, 0, offsetof(struct sock, sk_node.next));
895 memset(&sk->sk_node.pprev, 0,
896 size - offsetof(struct sock, sk_node.pprev));
897 }
898
899 /* Networking protocol blocks we attach to sockets.
900 * socket layer -> transport layer interface
901 * transport -> network interface is defined by struct inet_proto
902 */
903 struct proto {
904 void (*close)(struct sock *sk,
905 long timeout);
906 int (*connect)(struct sock *sk,
907 struct sockaddr *uaddr,
908 int addr_len);
909 int (*disconnect)(struct sock *sk, int flags);
910
911 struct sock * (*accept)(struct sock *sk, int flags, int *err);
912
913 int (*ioctl)(struct sock *sk, int cmd,
914 unsigned long arg);
915 int (*init)(struct sock *sk);
916 void (*destroy)(struct sock *sk);
917 void (*shutdown)(struct sock *sk, int how);
918 int (*setsockopt)(struct sock *sk, int level,
919 int optname, char __user *optval,
920 unsigned int optlen);
921 int (*getsockopt)(struct sock *sk, int level,
922 int optname, char __user *optval,
923 int __user *option);
924 #ifdef CONFIG_COMPAT
925 int (*compat_setsockopt)(struct sock *sk,
926 int level,
927 int optname, char __user *optval,
928 unsigned int optlen);
929 int (*compat_getsockopt)(struct sock *sk,
930 int level,
931 int optname, char __user *optval,
932 int __user *option);
933 int (*compat_ioctl)(struct sock *sk,
934 unsigned int cmd, unsigned long arg);
935 #endif
936 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
937 struct msghdr *msg, size_t len);
938 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
939 struct msghdr *msg,
940 size_t len, int noblock, int flags,
941 int *addr_len);
942 int (*sendpage)(struct sock *sk, struct page *page,
943 int offset, size_t size, int flags);
944 int (*bind)(struct sock *sk,
945 struct sockaddr *uaddr, int addr_len);
946
947 int (*backlog_rcv) (struct sock *sk,
948 struct sk_buff *skb);
949
950 void (*release_cb)(struct sock *sk);
951
952 /* Keeping track of sk's, looking them up, and port selection methods. */
953 void (*hash)(struct sock *sk);
954 void (*unhash)(struct sock *sk);
955 void (*rehash)(struct sock *sk);
956 int (*get_port)(struct sock *sk, unsigned short snum);
957 void (*clear_sk)(struct sock *sk, int size);
958
959 /* Keeping track of sockets in use */
960 #ifdef CONFIG_PROC_FS
961 unsigned int inuse_idx;
962 #endif
963
964 /* Memory pressure */
965 void (*enter_memory_pressure)(struct sock *sk);
966 atomic_long_t *memory_allocated; /* Current allocated memory. */
967 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
968 /*
969 * Pressure flag: try to collapse.
970 * Technical note: it is used by multiple contexts non atomically.
971 * All the __sk_mem_schedule() is of this nature: accounting
972 * is strict, actions are advisory and have some latency.
973 */
974 int *memory_pressure;
975 long *sysctl_mem;
976 int *sysctl_wmem;
977 int *sysctl_rmem;
978 int max_header;
979 bool no_autobind;
980
981 struct kmem_cache *slab;
982 unsigned int obj_size;
983 int slab_flags;
984
985 struct percpu_counter *orphan_count;
986
987 struct request_sock_ops *rsk_prot;
988 struct timewait_sock_ops *twsk_prot;
989
990 union {
991 struct inet_hashinfo *hashinfo;
992 struct udp_table *udp_table;
993 struct raw_hashinfo *raw_hash;
994 } h;
995
996 struct module *owner;
997
998 char name[32];
999
1000 struct list_head node;
1001 #ifdef SOCK_REFCNT_DEBUG
1002 atomic_t socks;
1003 #endif
1004 #ifdef CONFIG_MEMCG_KMEM
1005 /*
1006 * cgroup specific init/deinit functions. Called once for all
1007 * protocols that implement it, from cgroups populate function.
1008 * This function has to setup any files the protocol want to
1009 * appear in the kmem cgroup filesystem.
1010 */
1011 int (*init_cgroup)(struct mem_cgroup *memcg,
1012 struct cgroup_subsys *ss);
1013 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1014 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1015 #endif
1016 };
1017
1018 /*
1019 * Bits in struct cg_proto.flags
1020 */
1021 enum cg_proto_flags {
1022 /* Currently active and new sockets should be assigned to cgroups */
1023 MEMCG_SOCK_ACTIVE,
1024 /* It was ever activated; we must disarm static keys on destruction */
1025 MEMCG_SOCK_ACTIVATED,
1026 };
1027
1028 struct cg_proto {
1029 void (*enter_memory_pressure)(struct sock *sk);
1030 struct res_counter *memory_allocated; /* Current allocated memory. */
1031 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1032 int *memory_pressure;
1033 long *sysctl_mem;
1034 unsigned long flags;
1035 /*
1036 * memcg field is used to find which memcg we belong directly
1037 * Each memcg struct can hold more than one cg_proto, so container_of
1038 * won't really cut.
1039 *
1040 * The elegant solution would be having an inverse function to
1041 * proto_cgroup in struct proto, but that means polluting the structure
1042 * for everybody, instead of just for memcg users.
1043 */
1044 struct mem_cgroup *memcg;
1045 };
1046
1047 extern int proto_register(struct proto *prot, int alloc_slab);
1048 extern void proto_unregister(struct proto *prot);
1049
1050 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1051 {
1052 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1053 }
1054
1055 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1056 {
1057 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1058 }
1059
1060 #ifdef SOCK_REFCNT_DEBUG
1061 static inline void sk_refcnt_debug_inc(struct sock *sk)
1062 {
1063 atomic_inc(&sk->sk_prot->socks);
1064 }
1065
1066 static inline void sk_refcnt_debug_dec(struct sock *sk)
1067 {
1068 atomic_dec(&sk->sk_prot->socks);
1069 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1070 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1071 }
1072
1073 static inline void sk_refcnt_debug_release(const struct sock *sk)
1074 {
1075 if (atomic_read(&sk->sk_refcnt) != 1)
1076 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1077 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1078 }
1079 #else /* SOCK_REFCNT_DEBUG */
1080 #define sk_refcnt_debug_inc(sk) do { } while (0)
1081 #define sk_refcnt_debug_dec(sk) do { } while (0)
1082 #define sk_refcnt_debug_release(sk) do { } while (0)
1083 #endif /* SOCK_REFCNT_DEBUG */
1084
1085 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1086 extern struct static_key memcg_socket_limit_enabled;
1087 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1088 struct cg_proto *cg_proto)
1089 {
1090 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1091 }
1092 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1093 #else
1094 #define mem_cgroup_sockets_enabled 0
1095 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1096 struct cg_proto *cg_proto)
1097 {
1098 return NULL;
1099 }
1100 #endif
1101
1102
1103 static inline bool sk_has_memory_pressure(const struct sock *sk)
1104 {
1105 return sk->sk_prot->memory_pressure != NULL;
1106 }
1107
1108 static inline bool sk_under_memory_pressure(const struct sock *sk)
1109 {
1110 if (!sk->sk_prot->memory_pressure)
1111 return false;
1112
1113 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1114 return !!*sk->sk_cgrp->memory_pressure;
1115
1116 return !!*sk->sk_prot->memory_pressure;
1117 }
1118
1119 static inline void sk_leave_memory_pressure(struct sock *sk)
1120 {
1121 int *memory_pressure = sk->sk_prot->memory_pressure;
1122
1123 if (!memory_pressure)
1124 return;
1125
1126 if (*memory_pressure)
1127 *memory_pressure = 0;
1128
1129 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1130 struct cg_proto *cg_proto = sk->sk_cgrp;
1131 struct proto *prot = sk->sk_prot;
1132
1133 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1134 if (*cg_proto->memory_pressure)
1135 *cg_proto->memory_pressure = 0;
1136 }
1137
1138 }
1139
1140 static inline void sk_enter_memory_pressure(struct sock *sk)
1141 {
1142 if (!sk->sk_prot->enter_memory_pressure)
1143 return;
1144
1145 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1146 struct cg_proto *cg_proto = sk->sk_cgrp;
1147 struct proto *prot = sk->sk_prot;
1148
1149 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1150 cg_proto->enter_memory_pressure(sk);
1151 }
1152
1153 sk->sk_prot->enter_memory_pressure(sk);
1154 }
1155
1156 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1157 {
1158 long *prot = sk->sk_prot->sysctl_mem;
1159 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1160 prot = sk->sk_cgrp->sysctl_mem;
1161 return prot[index];
1162 }
1163
1164 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1165 unsigned long amt,
1166 int *parent_status)
1167 {
1168 struct res_counter *fail;
1169 int ret;
1170
1171 ret = res_counter_charge_nofail(prot->memory_allocated,
1172 amt << PAGE_SHIFT, &fail);
1173 if (ret < 0)
1174 *parent_status = OVER_LIMIT;
1175 }
1176
1177 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1178 unsigned long amt)
1179 {
1180 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1181 }
1182
1183 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1184 {
1185 u64 ret;
1186 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1187 return ret >> PAGE_SHIFT;
1188 }
1189
1190 static inline long
1191 sk_memory_allocated(const struct sock *sk)
1192 {
1193 struct proto *prot = sk->sk_prot;
1194 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1195 return memcg_memory_allocated_read(sk->sk_cgrp);
1196
1197 return atomic_long_read(prot->memory_allocated);
1198 }
1199
1200 static inline long
1201 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1202 {
1203 struct proto *prot = sk->sk_prot;
1204
1205 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1206 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1207 /* update the root cgroup regardless */
1208 atomic_long_add_return(amt, prot->memory_allocated);
1209 return memcg_memory_allocated_read(sk->sk_cgrp);
1210 }
1211
1212 return atomic_long_add_return(amt, prot->memory_allocated);
1213 }
1214
1215 static inline void
1216 sk_memory_allocated_sub(struct sock *sk, int amt)
1217 {
1218 struct proto *prot = sk->sk_prot;
1219
1220 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1221 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1222
1223 atomic_long_sub(amt, prot->memory_allocated);
1224 }
1225
1226 static inline void sk_sockets_allocated_dec(struct sock *sk)
1227 {
1228 struct proto *prot = sk->sk_prot;
1229
1230 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1231 struct cg_proto *cg_proto = sk->sk_cgrp;
1232
1233 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1234 percpu_counter_dec(cg_proto->sockets_allocated);
1235 }
1236
1237 percpu_counter_dec(prot->sockets_allocated);
1238 }
1239
1240 static inline void sk_sockets_allocated_inc(struct sock *sk)
1241 {
1242 struct proto *prot = sk->sk_prot;
1243
1244 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1245 struct cg_proto *cg_proto = sk->sk_cgrp;
1246
1247 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1248 percpu_counter_inc(cg_proto->sockets_allocated);
1249 }
1250
1251 percpu_counter_inc(prot->sockets_allocated);
1252 }
1253
1254 static inline int
1255 sk_sockets_allocated_read_positive(struct sock *sk)
1256 {
1257 struct proto *prot = sk->sk_prot;
1258
1259 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1260 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1261
1262 return percpu_counter_read_positive(prot->sockets_allocated);
1263 }
1264
1265 static inline int
1266 proto_sockets_allocated_sum_positive(struct proto *prot)
1267 {
1268 return percpu_counter_sum_positive(prot->sockets_allocated);
1269 }
1270
1271 static inline long
1272 proto_memory_allocated(struct proto *prot)
1273 {
1274 return atomic_long_read(prot->memory_allocated);
1275 }
1276
1277 static inline bool
1278 proto_memory_pressure(struct proto *prot)
1279 {
1280 if (!prot->memory_pressure)
1281 return false;
1282 return !!*prot->memory_pressure;
1283 }
1284
1285
1286 #ifdef CONFIG_PROC_FS
1287 /* Called with local bh disabled */
1288 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1289 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1290 #else
1291 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1292 int inc)
1293 {
1294 }
1295 #endif
1296
1297
1298 /* With per-bucket locks this operation is not-atomic, so that
1299 * this version is not worse.
1300 */
1301 static inline void __sk_prot_rehash(struct sock *sk)
1302 {
1303 sk->sk_prot->unhash(sk);
1304 sk->sk_prot->hash(sk);
1305 }
1306
1307 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1308
1309 /* About 10 seconds */
1310 #define SOCK_DESTROY_TIME (10*HZ)
1311
1312 /* Sockets 0-1023 can't be bound to unless you are superuser */
1313 #define PROT_SOCK 1024
1314
1315 #define SHUTDOWN_MASK 3
1316 #define RCV_SHUTDOWN 1
1317 #define SEND_SHUTDOWN 2
1318
1319 #define SOCK_SNDBUF_LOCK 1
1320 #define SOCK_RCVBUF_LOCK 2
1321 #define SOCK_BINDADDR_LOCK 4
1322 #define SOCK_BINDPORT_LOCK 8
1323
1324 /* sock_iocb: used to kick off async processing of socket ios */
1325 struct sock_iocb {
1326 struct list_head list;
1327
1328 int flags;
1329 int size;
1330 struct socket *sock;
1331 struct sock *sk;
1332 struct scm_cookie *scm;
1333 struct msghdr *msg, async_msg;
1334 struct kiocb *kiocb;
1335 };
1336
1337 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1338 {
1339 return (struct sock_iocb *)iocb->private;
1340 }
1341
1342 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1343 {
1344 return si->kiocb;
1345 }
1346
1347 struct socket_alloc {
1348 struct socket socket;
1349 struct inode vfs_inode;
1350 };
1351
1352 static inline struct socket *SOCKET_I(struct inode *inode)
1353 {
1354 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1355 }
1356
1357 static inline struct inode *SOCK_INODE(struct socket *socket)
1358 {
1359 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1360 }
1361
1362 /*
1363 * Functions for memory accounting
1364 */
1365 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1366 extern void __sk_mem_reclaim(struct sock *sk);
1367
1368 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1369 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1370 #define SK_MEM_SEND 0
1371 #define SK_MEM_RECV 1
1372
1373 static inline int sk_mem_pages(int amt)
1374 {
1375 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1376 }
1377
1378 static inline bool sk_has_account(struct sock *sk)
1379 {
1380 /* return true if protocol supports memory accounting */
1381 return !!sk->sk_prot->memory_allocated;
1382 }
1383
1384 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1385 {
1386 if (!sk_has_account(sk))
1387 return true;
1388 return size <= sk->sk_forward_alloc ||
1389 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1390 }
1391
1392 static inline bool
1393 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1394 {
1395 if (!sk_has_account(sk))
1396 return true;
1397 return size<= sk->sk_forward_alloc ||
1398 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1399 skb_pfmemalloc(skb);
1400 }
1401
1402 static inline void sk_mem_reclaim(struct sock *sk)
1403 {
1404 if (!sk_has_account(sk))
1405 return;
1406 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1407 __sk_mem_reclaim(sk);
1408 }
1409
1410 static inline void sk_mem_reclaim_partial(struct sock *sk)
1411 {
1412 if (!sk_has_account(sk))
1413 return;
1414 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1415 __sk_mem_reclaim(sk);
1416 }
1417
1418 static inline void sk_mem_charge(struct sock *sk, int size)
1419 {
1420 if (!sk_has_account(sk))
1421 return;
1422 sk->sk_forward_alloc -= size;
1423 }
1424
1425 static inline void sk_mem_uncharge(struct sock *sk, int size)
1426 {
1427 if (!sk_has_account(sk))
1428 return;
1429 sk->sk_forward_alloc += size;
1430 }
1431
1432 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1433 {
1434 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1435 sk->sk_wmem_queued -= skb->truesize;
1436 sk_mem_uncharge(sk, skb->truesize);
1437 __kfree_skb(skb);
1438 }
1439
1440 /* Used by processes to "lock" a socket state, so that
1441 * interrupts and bottom half handlers won't change it
1442 * from under us. It essentially blocks any incoming
1443 * packets, so that we won't get any new data or any
1444 * packets that change the state of the socket.
1445 *
1446 * While locked, BH processing will add new packets to
1447 * the backlog queue. This queue is processed by the
1448 * owner of the socket lock right before it is released.
1449 *
1450 * Since ~2.3.5 it is also exclusive sleep lock serializing
1451 * accesses from user process context.
1452 */
1453 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1454
1455 static inline void sock_release_ownership(struct sock *sk)
1456 {
1457 sk->sk_lock.owned = 0;
1458 }
1459
1460 /*
1461 * Macro so as to not evaluate some arguments when
1462 * lockdep is not enabled.
1463 *
1464 * Mark both the sk_lock and the sk_lock.slock as a
1465 * per-address-family lock class.
1466 */
1467 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1468 do { \
1469 sk->sk_lock.owned = 0; \
1470 init_waitqueue_head(&sk->sk_lock.wq); \
1471 spin_lock_init(&(sk)->sk_lock.slock); \
1472 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1473 sizeof((sk)->sk_lock)); \
1474 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1475 (skey), (sname)); \
1476 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1477 } while (0)
1478
1479 extern void lock_sock_nested(struct sock *sk, int subclass);
1480
1481 static inline void lock_sock(struct sock *sk)
1482 {
1483 lock_sock_nested(sk, 0);
1484 }
1485
1486 extern void release_sock(struct sock *sk);
1487
1488 /* BH context may only use the following locking interface. */
1489 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1490 #define bh_lock_sock_nested(__sk) \
1491 spin_lock_nested(&((__sk)->sk_lock.slock), \
1492 SINGLE_DEPTH_NESTING)
1493 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1494
1495 extern bool lock_sock_fast(struct sock *sk);
1496 /**
1497 * unlock_sock_fast - complement of lock_sock_fast
1498 * @sk: socket
1499 * @slow: slow mode
1500 *
1501 * fast unlock socket for user context.
1502 * If slow mode is on, we call regular release_sock()
1503 */
1504 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1505 {
1506 if (slow)
1507 release_sock(sk);
1508 else
1509 spin_unlock_bh(&sk->sk_lock.slock);
1510 }
1511
1512
1513 extern struct sock *sk_alloc(struct net *net, int family,
1514 gfp_t priority,
1515 struct proto *prot);
1516 extern void sk_free(struct sock *sk);
1517 extern void sk_release_kernel(struct sock *sk);
1518 extern struct sock *sk_clone_lock(const struct sock *sk,
1519 const gfp_t priority);
1520
1521 extern struct sk_buff *sock_wmalloc(struct sock *sk,
1522 unsigned long size, int force,
1523 gfp_t priority);
1524 extern struct sk_buff *sock_rmalloc(struct sock *sk,
1525 unsigned long size, int force,
1526 gfp_t priority);
1527 extern void sock_wfree(struct sk_buff *skb);
1528 extern void sock_rfree(struct sk_buff *skb);
1529 extern void sock_edemux(struct sk_buff *skb);
1530
1531 extern int sock_setsockopt(struct socket *sock, int level,
1532 int op, char __user *optval,
1533 unsigned int optlen);
1534
1535 extern int sock_getsockopt(struct socket *sock, int level,
1536 int op, char __user *optval,
1537 int __user *optlen);
1538 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1539 unsigned long size,
1540 int noblock,
1541 int *errcode);
1542 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1543 unsigned long header_len,
1544 unsigned long data_len,
1545 int noblock,
1546 int *errcode);
1547 extern void *sock_kmalloc(struct sock *sk, int size,
1548 gfp_t priority);
1549 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1550 extern void sk_send_sigurg(struct sock *sk);
1551
1552 /*
1553 * Functions to fill in entries in struct proto_ops when a protocol
1554 * does not implement a particular function.
1555 */
1556 extern int sock_no_bind(struct socket *,
1557 struct sockaddr *, int);
1558 extern int sock_no_connect(struct socket *,
1559 struct sockaddr *, int, int);
1560 extern int sock_no_socketpair(struct socket *,
1561 struct socket *);
1562 extern int sock_no_accept(struct socket *,
1563 struct socket *, int);
1564 extern int sock_no_getname(struct socket *,
1565 struct sockaddr *, int *, int);
1566 extern unsigned int sock_no_poll(struct file *, struct socket *,
1567 struct poll_table_struct *);
1568 extern int sock_no_ioctl(struct socket *, unsigned int,
1569 unsigned long);
1570 extern int sock_no_listen(struct socket *, int);
1571 extern int sock_no_shutdown(struct socket *, int);
1572 extern int sock_no_getsockopt(struct socket *, int , int,
1573 char __user *, int __user *);
1574 extern int sock_no_setsockopt(struct socket *, int, int,
1575 char __user *, unsigned int);
1576 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1577 struct msghdr *, size_t);
1578 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1579 struct msghdr *, size_t, int);
1580 extern int sock_no_mmap(struct file *file,
1581 struct socket *sock,
1582 struct vm_area_struct *vma);
1583 extern ssize_t sock_no_sendpage(struct socket *sock,
1584 struct page *page,
1585 int offset, size_t size,
1586 int flags);
1587
1588 /*
1589 * Functions to fill in entries in struct proto_ops when a protocol
1590 * uses the inet style.
1591 */
1592 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1593 char __user *optval, int __user *optlen);
1594 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1595 struct msghdr *msg, size_t size, int flags);
1596 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1597 char __user *optval, unsigned int optlen);
1598 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1599 int optname, char __user *optval, int __user *optlen);
1600 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1601 int optname, char __user *optval, unsigned int optlen);
1602
1603 extern void sk_common_release(struct sock *sk);
1604
1605 /*
1606 * Default socket callbacks and setup code
1607 */
1608
1609 /* Initialise core socket variables */
1610 extern void sock_init_data(struct socket *sock, struct sock *sk);
1611
1612 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1613
1614 /**
1615 * sk_filter_release - release a socket filter
1616 * @fp: filter to remove
1617 *
1618 * Remove a filter from a socket and release its resources.
1619 */
1620
1621 static inline void sk_filter_release(struct sk_filter *fp)
1622 {
1623 if (atomic_dec_and_test(&fp->refcnt))
1624 call_rcu(&fp->rcu, sk_filter_release_rcu);
1625 }
1626
1627 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1628 {
1629 unsigned int size = sk_filter_len(fp);
1630
1631 atomic_sub(size, &sk->sk_omem_alloc);
1632 sk_filter_release(fp);
1633 }
1634
1635 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1636 {
1637 atomic_inc(&fp->refcnt);
1638 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1639 }
1640
1641 /*
1642 * Socket reference counting postulates.
1643 *
1644 * * Each user of socket SHOULD hold a reference count.
1645 * * Each access point to socket (an hash table bucket, reference from a list,
1646 * running timer, skb in flight MUST hold a reference count.
1647 * * When reference count hits 0, it means it will never increase back.
1648 * * When reference count hits 0, it means that no references from
1649 * outside exist to this socket and current process on current CPU
1650 * is last user and may/should destroy this socket.
1651 * * sk_free is called from any context: process, BH, IRQ. When
1652 * it is called, socket has no references from outside -> sk_free
1653 * may release descendant resources allocated by the socket, but
1654 * to the time when it is called, socket is NOT referenced by any
1655 * hash tables, lists etc.
1656 * * Packets, delivered from outside (from network or from another process)
1657 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1658 * when they sit in queue. Otherwise, packets will leak to hole, when
1659 * socket is looked up by one cpu and unhasing is made by another CPU.
1660 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1661 * (leak to backlog). Packet socket does all the processing inside
1662 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1663 * use separate SMP lock, so that they are prone too.
1664 */
1665
1666 /* Ungrab socket and destroy it, if it was the last reference. */
1667 static inline void sock_put(struct sock *sk)
1668 {
1669 if (atomic_dec_and_test(&sk->sk_refcnt))
1670 sk_free(sk);
1671 }
1672
1673 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1674 const int nested);
1675
1676 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1677 {
1678 sk->sk_tx_queue_mapping = tx_queue;
1679 }
1680
1681 static inline void sk_tx_queue_clear(struct sock *sk)
1682 {
1683 sk->sk_tx_queue_mapping = -1;
1684 }
1685
1686 static inline int sk_tx_queue_get(const struct sock *sk)
1687 {
1688 return sk ? sk->sk_tx_queue_mapping : -1;
1689 }
1690
1691 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1692 {
1693 sk_tx_queue_clear(sk);
1694 sk->sk_socket = sock;
1695 }
1696
1697 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1698 {
1699 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1700 return &rcu_dereference_raw(sk->sk_wq)->wait;
1701 }
1702 /* Detach socket from process context.
1703 * Announce socket dead, detach it from wait queue and inode.
1704 * Note that parent inode held reference count on this struct sock,
1705 * we do not release it in this function, because protocol
1706 * probably wants some additional cleanups or even continuing
1707 * to work with this socket (TCP).
1708 */
1709 static inline void sock_orphan(struct sock *sk)
1710 {
1711 write_lock_bh(&sk->sk_callback_lock);
1712 sock_set_flag(sk, SOCK_DEAD);
1713 sk_set_socket(sk, NULL);
1714 sk->sk_wq = NULL;
1715 write_unlock_bh(&sk->sk_callback_lock);
1716 }
1717
1718 static inline void sock_graft(struct sock *sk, struct socket *parent)
1719 {
1720 write_lock_bh(&sk->sk_callback_lock);
1721 sk->sk_wq = parent->wq;
1722 parent->sk = sk;
1723 sk_set_socket(sk, parent);
1724 security_sock_graft(sk, parent);
1725 write_unlock_bh(&sk->sk_callback_lock);
1726 }
1727
1728 extern kuid_t sock_i_uid(struct sock *sk);
1729 extern unsigned long sock_i_ino(struct sock *sk);
1730
1731 static inline struct dst_entry *
1732 __sk_dst_get(struct sock *sk)
1733 {
1734 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1735 lockdep_is_held(&sk->sk_lock.slock));
1736 }
1737
1738 static inline struct dst_entry *
1739 sk_dst_get(struct sock *sk)
1740 {
1741 struct dst_entry *dst;
1742
1743 rcu_read_lock();
1744 dst = rcu_dereference(sk->sk_dst_cache);
1745 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1746 dst = NULL;
1747 rcu_read_unlock();
1748 return dst;
1749 }
1750
1751 extern void sk_reset_txq(struct sock *sk);
1752
1753 static inline void dst_negative_advice(struct sock *sk)
1754 {
1755 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1756
1757 if (dst && dst->ops->negative_advice) {
1758 ndst = dst->ops->negative_advice(dst);
1759
1760 if (ndst != dst) {
1761 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1762 sk_reset_txq(sk);
1763 }
1764 }
1765 }
1766
1767 static inline void
1768 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1769 {
1770 struct dst_entry *old_dst;
1771
1772 sk_tx_queue_clear(sk);
1773 /*
1774 * This can be called while sk is owned by the caller only,
1775 * with no state that can be checked in a rcu_dereference_check() cond
1776 */
1777 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1778 rcu_assign_pointer(sk->sk_dst_cache, dst);
1779 dst_release(old_dst);
1780 }
1781
1782 static inline void
1783 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1784 {
1785 struct dst_entry *old_dst;
1786
1787 sk_tx_queue_clear(sk);
1788 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1789 dst_release(old_dst);
1790 }
1791
1792 static inline void
1793 __sk_dst_reset(struct sock *sk)
1794 {
1795 __sk_dst_set(sk, NULL);
1796 }
1797
1798 static inline void
1799 sk_dst_reset(struct sock *sk)
1800 {
1801 sk_dst_set(sk, NULL);
1802 }
1803
1804 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1805
1806 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1807
1808 static inline bool sk_can_gso(const struct sock *sk)
1809 {
1810 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1811 }
1812
1813 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1814
1815 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1816 {
1817 sk->sk_route_nocaps |= flags;
1818 sk->sk_route_caps &= ~flags;
1819 }
1820
1821 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1822 char __user *from, char *to,
1823 int copy, int offset)
1824 {
1825 if (skb->ip_summed == CHECKSUM_NONE) {
1826 int err = 0;
1827 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1828 if (err)
1829 return err;
1830 skb->csum = csum_block_add(skb->csum, csum, offset);
1831 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1832 if (!access_ok(VERIFY_READ, from, copy) ||
1833 __copy_from_user_nocache(to, from, copy))
1834 return -EFAULT;
1835 } else if (copy_from_user(to, from, copy))
1836 return -EFAULT;
1837
1838 return 0;
1839 }
1840
1841 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1842 char __user *from, int copy)
1843 {
1844 int err, offset = skb->len;
1845
1846 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1847 copy, offset);
1848 if (err)
1849 __skb_trim(skb, offset);
1850
1851 return err;
1852 }
1853
1854 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1855 struct sk_buff *skb,
1856 struct page *page,
1857 int off, int copy)
1858 {
1859 int err;
1860
1861 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1862 copy, skb->len);
1863 if (err)
1864 return err;
1865
1866 skb->len += copy;
1867 skb->data_len += copy;
1868 skb->truesize += copy;
1869 sk->sk_wmem_queued += copy;
1870 sk_mem_charge(sk, copy);
1871 return 0;
1872 }
1873
1874 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1875 struct sk_buff *skb, struct page *page,
1876 int off, int copy)
1877 {
1878 if (skb->ip_summed == CHECKSUM_NONE) {
1879 int err = 0;
1880 __wsum csum = csum_and_copy_from_user(from,
1881 page_address(page) + off,
1882 copy, 0, &err);
1883 if (err)
1884 return err;
1885 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1886 } else if (copy_from_user(page_address(page) + off, from, copy))
1887 return -EFAULT;
1888
1889 skb->len += copy;
1890 skb->data_len += copy;
1891 skb->truesize += copy;
1892 sk->sk_wmem_queued += copy;
1893 sk_mem_charge(sk, copy);
1894 return 0;
1895 }
1896
1897 /**
1898 * sk_wmem_alloc_get - returns write allocations
1899 * @sk: socket
1900 *
1901 * Returns sk_wmem_alloc minus initial offset of one
1902 */
1903 static inline int sk_wmem_alloc_get(const struct sock *sk)
1904 {
1905 return atomic_read(&sk->sk_wmem_alloc) - 1;
1906 }
1907
1908 /**
1909 * sk_rmem_alloc_get - returns read allocations
1910 * @sk: socket
1911 *
1912 * Returns sk_rmem_alloc
1913 */
1914 static inline int sk_rmem_alloc_get(const struct sock *sk)
1915 {
1916 return atomic_read(&sk->sk_rmem_alloc);
1917 }
1918
1919 /**
1920 * sk_has_allocations - check if allocations are outstanding
1921 * @sk: socket
1922 *
1923 * Returns true if socket has write or read allocations
1924 */
1925 static inline bool sk_has_allocations(const struct sock *sk)
1926 {
1927 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1928 }
1929
1930 /**
1931 * wq_has_sleeper - check if there are any waiting processes
1932 * @wq: struct socket_wq
1933 *
1934 * Returns true if socket_wq has waiting processes
1935 *
1936 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1937 * barrier call. They were added due to the race found within the tcp code.
1938 *
1939 * Consider following tcp code paths:
1940 *
1941 * CPU1 CPU2
1942 *
1943 * sys_select receive packet
1944 * ... ...
1945 * __add_wait_queue update tp->rcv_nxt
1946 * ... ...
1947 * tp->rcv_nxt check sock_def_readable
1948 * ... {
1949 * schedule rcu_read_lock();
1950 * wq = rcu_dereference(sk->sk_wq);
1951 * if (wq && waitqueue_active(&wq->wait))
1952 * wake_up_interruptible(&wq->wait)
1953 * ...
1954 * }
1955 *
1956 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1957 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1958 * could then endup calling schedule and sleep forever if there are no more
1959 * data on the socket.
1960 *
1961 */
1962 static inline bool wq_has_sleeper(struct socket_wq *wq)
1963 {
1964 /* We need to be sure we are in sync with the
1965 * add_wait_queue modifications to the wait queue.
1966 *
1967 * This memory barrier is paired in the sock_poll_wait.
1968 */
1969 smp_mb();
1970 return wq && waitqueue_active(&wq->wait);
1971 }
1972
1973 /**
1974 * sock_poll_wait - place memory barrier behind the poll_wait call.
1975 * @filp: file
1976 * @wait_address: socket wait queue
1977 * @p: poll_table
1978 *
1979 * See the comments in the wq_has_sleeper function.
1980 */
1981 static inline void sock_poll_wait(struct file *filp,
1982 wait_queue_head_t *wait_address, poll_table *p)
1983 {
1984 if (!poll_does_not_wait(p) && wait_address) {
1985 poll_wait(filp, wait_address, p);
1986 /* We need to be sure we are in sync with the
1987 * socket flags modification.
1988 *
1989 * This memory barrier is paired in the wq_has_sleeper.
1990 */
1991 smp_mb();
1992 }
1993 }
1994
1995 /*
1996 * Queue a received datagram if it will fit. Stream and sequenced
1997 * protocols can't normally use this as they need to fit buffers in
1998 * and play with them.
1999 *
2000 * Inlined as it's very short and called for pretty much every
2001 * packet ever received.
2002 */
2003
2004 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2005 {
2006 skb_orphan(skb);
2007 skb->sk = sk;
2008 skb->destructor = sock_wfree;
2009 /*
2010 * We used to take a refcount on sk, but following operation
2011 * is enough to guarantee sk_free() wont free this sock until
2012 * all in-flight packets are completed
2013 */
2014 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
2015 }
2016
2017 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2018 {
2019 skb_orphan(skb);
2020 skb->sk = sk;
2021 skb->destructor = sock_rfree;
2022 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2023 sk_mem_charge(sk, skb->truesize);
2024 }
2025
2026 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2027 unsigned long expires);
2028
2029 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2030
2031 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2032
2033 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2034
2035 /*
2036 * Recover an error report and clear atomically
2037 */
2038
2039 static inline int sock_error(struct sock *sk)
2040 {
2041 int err;
2042 if (likely(!sk->sk_err))
2043 return 0;
2044 err = xchg(&sk->sk_err, 0);
2045 return -err;
2046 }
2047
2048 static inline unsigned long sock_wspace(struct sock *sk)
2049 {
2050 int amt = 0;
2051
2052 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2053 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2054 if (amt < 0)
2055 amt = 0;
2056 }
2057 return amt;
2058 }
2059
2060 static inline void sk_wake_async(struct sock *sk, int how, int band)
2061 {
2062 if (sock_flag(sk, SOCK_FASYNC))
2063 sock_wake_async(sk->sk_socket, how, band);
2064 }
2065
2066 #define SOCK_MIN_SNDBUF 2048
2067 /*
2068 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
2069 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
2070 */
2071 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
2072
2073 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2074 {
2075 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2076 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2077 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2078 }
2079 }
2080
2081 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2082
2083 /**
2084 * sk_page_frag - return an appropriate page_frag
2085 * @sk: socket
2086 *
2087 * If socket allocation mode allows current thread to sleep, it means its
2088 * safe to use the per task page_frag instead of the per socket one.
2089 */
2090 static inline struct page_frag *sk_page_frag(struct sock *sk)
2091 {
2092 if (sk->sk_allocation & __GFP_WAIT)
2093 return &current->task_frag;
2094
2095 return &sk->sk_frag;
2096 }
2097
2098 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2099
2100 /*
2101 * Default write policy as shown to user space via poll/select/SIGIO
2102 */
2103 static inline bool sock_writeable(const struct sock *sk)
2104 {
2105 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2106 }
2107
2108 static inline gfp_t gfp_any(void)
2109 {
2110 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2111 }
2112
2113 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2114 {
2115 return noblock ? 0 : sk->sk_rcvtimeo;
2116 }
2117
2118 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2119 {
2120 return noblock ? 0 : sk->sk_sndtimeo;
2121 }
2122
2123 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2124 {
2125 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2126 }
2127
2128 /* Alas, with timeout socket operations are not restartable.
2129 * Compare this to poll().
2130 */
2131 static inline int sock_intr_errno(long timeo)
2132 {
2133 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2134 }
2135
2136 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2137 struct sk_buff *skb);
2138 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2139 struct sk_buff *skb);
2140
2141 static inline void
2142 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2143 {
2144 ktime_t kt = skb->tstamp;
2145 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2146
2147 /*
2148 * generate control messages if
2149 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2150 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2151 * - software time stamp available and wanted
2152 * (SOCK_TIMESTAMPING_SOFTWARE)
2153 * - hardware time stamps available and wanted
2154 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2155 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2156 */
2157 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2158 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2159 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2160 (hwtstamps->hwtstamp.tv64 &&
2161 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2162 (hwtstamps->syststamp.tv64 &&
2163 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2164 __sock_recv_timestamp(msg, sk, skb);
2165 else
2166 sk->sk_stamp = kt;
2167
2168 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2169 __sock_recv_wifi_status(msg, sk, skb);
2170 }
2171
2172 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2173 struct sk_buff *skb);
2174
2175 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2176 struct sk_buff *skb)
2177 {
2178 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2179 (1UL << SOCK_RCVTSTAMP) | \
2180 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
2181 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2182 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2183 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2184
2185 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2186 __sock_recv_ts_and_drops(msg, sk, skb);
2187 else
2188 sk->sk_stamp = skb->tstamp;
2189 }
2190
2191 /**
2192 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2193 * @sk: socket sending this packet
2194 * @tx_flags: filled with instructions for time stamping
2195 *
2196 * Currently only depends on SOCK_TIMESTAMPING* flags.
2197 */
2198 extern void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2199
2200 /**
2201 * sk_eat_skb - Release a skb if it is no longer needed
2202 * @sk: socket to eat this skb from
2203 * @skb: socket buffer to eat
2204 * @copied_early: flag indicating whether DMA operations copied this data early
2205 *
2206 * This routine must be called with interrupts disabled or with the socket
2207 * locked so that the sk_buff queue operation is ok.
2208 */
2209 #ifdef CONFIG_NET_DMA
2210 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2211 {
2212 __skb_unlink(skb, &sk->sk_receive_queue);
2213 if (!copied_early)
2214 __kfree_skb(skb);
2215 else
2216 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2217 }
2218 #else
2219 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2220 {
2221 __skb_unlink(skb, &sk->sk_receive_queue);
2222 __kfree_skb(skb);
2223 }
2224 #endif
2225
2226 static inline
2227 struct net *sock_net(const struct sock *sk)
2228 {
2229 return read_pnet(&sk->sk_net);
2230 }
2231
2232 static inline
2233 void sock_net_set(struct sock *sk, struct net *net)
2234 {
2235 write_pnet(&sk->sk_net, net);
2236 }
2237
2238 /*
2239 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2240 * They should not hold a reference to a namespace in order to allow
2241 * to stop it.
2242 * Sockets after sk_change_net should be released using sk_release_kernel
2243 */
2244 static inline void sk_change_net(struct sock *sk, struct net *net)
2245 {
2246 put_net(sock_net(sk));
2247 sock_net_set(sk, hold_net(net));
2248 }
2249
2250 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2251 {
2252 if (skb->sk) {
2253 struct sock *sk = skb->sk;
2254
2255 skb->destructor = NULL;
2256 skb->sk = NULL;
2257 return sk;
2258 }
2259 return NULL;
2260 }
2261
2262 extern void sock_enable_timestamp(struct sock *sk, int flag);
2263 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2264 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2265
2266 bool sk_ns_capable(const struct sock *sk,
2267 struct user_namespace *user_ns, int cap);
2268 bool sk_capable(const struct sock *sk, int cap);
2269 bool sk_net_capable(const struct sock *sk, int cap);
2270
2271 /*
2272 * Enable debug/info messages
2273 */
2274 extern int net_msg_warn;
2275 #define NETDEBUG(fmt, args...) \
2276 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2277
2278 #define LIMIT_NETDEBUG(fmt, args...) \
2279 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2280
2281 extern __u32 sysctl_wmem_max;
2282 extern __u32 sysctl_rmem_max;
2283
2284 extern int sysctl_optmem_max;
2285
2286 extern __u32 sysctl_wmem_default;
2287 extern __u32 sysctl_rmem_default;
2288
2289 #endif /* _SOCK_H */