Merge tag 'v3.10.94' 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 sk_type : 16;
356 kmemcheck_bitfield_end(flags);
357 int sk_wmem_queued;
358 gfp_t sk_allocation;
359 u32 sk_pacing_rate; /* bytes per second */
360 netdev_features_t sk_route_caps;
361 netdev_features_t sk_route_nocaps;
362 int sk_gso_type;
363 unsigned int sk_gso_max_size;
364 u16 sk_gso_max_segs;
365 int sk_rcvlowat;
366 unsigned long sk_lingertime;
367 struct sk_buff_head sk_error_queue;
368 struct proto *sk_prot_creator;
369 rwlock_t sk_callback_lock;
370 int sk_err,
371 sk_err_soft;
372 unsigned short sk_ack_backlog;
373 unsigned short sk_max_ack_backlog;
374 __u32 sk_priority;
375 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
376 __u32 sk_cgrp_prioidx;
377 #endif
378 struct pid *sk_peer_pid;
379 const struct cred *sk_peer_cred;
380 long sk_rcvtimeo;
381 long sk_sndtimeo;
382 void *sk_protinfo;
383 struct timer_list sk_timer;
384 ktime_t sk_stamp;
385 struct socket *sk_socket;
386 void *sk_user_data;
387 struct page_frag sk_frag;
388 struct sk_buff *sk_send_head;
389 __s32 sk_peek_off;
390 int sk_write_pending;
391 #ifdef CONFIG_SECURITY
392 void *sk_security;
393 #endif
394 __u32 sk_mark;
395 u32 sk_classid;
396 struct cg_proto *sk_cgrp;
397 void (*sk_state_change)(struct sock *sk);
398 void (*sk_data_ready)(struct sock *sk, int bytes);
399 void (*sk_write_space)(struct sock *sk);
400 void (*sk_error_report)(struct sock *sk);
401 int (*sk_backlog_rcv)(struct sock *sk,
402 struct sk_buff *skb);
403 void (*sk_destruct)(struct sock *sk);
404 };
405
406 /*
407 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
408 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
409 * on a socket means that the socket will reuse everybody else's port
410 * without looking at the other's sk_reuse value.
411 */
412
413 #define SK_NO_REUSE 0
414 #define SK_CAN_REUSE 1
415 #define SK_FORCE_REUSE 2
416
417 static inline int sk_peek_offset(struct sock *sk, int flags)
418 {
419 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
420 return sk->sk_peek_off;
421 else
422 return 0;
423 }
424
425 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
426 {
427 if (sk->sk_peek_off >= 0) {
428 if (sk->sk_peek_off >= val)
429 sk->sk_peek_off -= val;
430 else
431 sk->sk_peek_off = 0;
432 }
433 }
434
435 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
436 {
437 if (sk->sk_peek_off >= 0)
438 sk->sk_peek_off += val;
439 }
440
441 /*
442 * Hashed lists helper routines
443 */
444 static inline struct sock *sk_entry(const struct hlist_node *node)
445 {
446 return hlist_entry(node, struct sock, sk_node);
447 }
448
449 static inline struct sock *__sk_head(const struct hlist_head *head)
450 {
451 return hlist_entry(head->first, struct sock, sk_node);
452 }
453
454 static inline struct sock *sk_head(const struct hlist_head *head)
455 {
456 return hlist_empty(head) ? NULL : __sk_head(head);
457 }
458
459 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
460 {
461 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
462 }
463
464 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
465 {
466 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
467 }
468
469 static inline struct sock *sk_next(const struct sock *sk)
470 {
471 return sk->sk_node.next ?
472 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
473 }
474
475 static inline struct sock *sk_nulls_next(const struct sock *sk)
476 {
477 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
478 hlist_nulls_entry(sk->sk_nulls_node.next,
479 struct sock, sk_nulls_node) :
480 NULL;
481 }
482
483 static inline bool sk_unhashed(const struct sock *sk)
484 {
485 return hlist_unhashed(&sk->sk_node);
486 }
487
488 static inline bool sk_hashed(const struct sock *sk)
489 {
490 return !sk_unhashed(sk);
491 }
492
493 static inline void sk_node_init(struct hlist_node *node)
494 {
495 node->pprev = NULL;
496 }
497
498 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
499 {
500 node->pprev = NULL;
501 }
502
503 static inline void __sk_del_node(struct sock *sk)
504 {
505 __hlist_del(&sk->sk_node);
506 }
507
508 /* NB: equivalent to hlist_del_init_rcu */
509 static inline bool __sk_del_node_init(struct sock *sk)
510 {
511 if (sk_hashed(sk)) {
512 __sk_del_node(sk);
513 sk_node_init(&sk->sk_node);
514 return true;
515 }
516 return false;
517 }
518
519 /* Grab socket reference count. This operation is valid only
520 when sk is ALREADY grabbed f.e. it is found in hash table
521 or a list and the lookup is made under lock preventing hash table
522 modifications.
523 */
524
525 static inline void sock_hold(struct sock *sk)
526 {
527 atomic_inc(&sk->sk_refcnt);
528 }
529
530 /* Ungrab socket in the context, which assumes that socket refcnt
531 cannot hit zero, f.e. it is true in context of any socketcall.
532 */
533 static inline void __sock_put(struct sock *sk)
534 {
535 atomic_dec(&sk->sk_refcnt);
536 }
537
538 static inline bool sk_del_node_init(struct sock *sk)
539 {
540 bool rc = __sk_del_node_init(sk);
541
542 if (rc) {
543 /* paranoid for a while -acme */
544 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
545 __sock_put(sk);
546 }
547 return rc;
548 }
549 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
550
551 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
552 {
553 if (sk_hashed(sk)) {
554 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
555 return true;
556 }
557 return false;
558 }
559
560 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
561 {
562 bool rc = __sk_nulls_del_node_init_rcu(sk);
563
564 if (rc) {
565 /* paranoid for a while -acme */
566 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
567 __sock_put(sk);
568 }
569 return rc;
570 }
571
572 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
573 {
574 hlist_add_head(&sk->sk_node, list);
575 }
576
577 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
578 {
579 sock_hold(sk);
580 __sk_add_node(sk, list);
581 }
582
583 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
584 {
585 sock_hold(sk);
586 hlist_add_head_rcu(&sk->sk_node, list);
587 }
588
589 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
590 {
591 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
592 }
593
594 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
595 {
596 sock_hold(sk);
597 __sk_nulls_add_node_rcu(sk, list);
598 }
599
600 static inline void __sk_del_bind_node(struct sock *sk)
601 {
602 __hlist_del(&sk->sk_bind_node);
603 }
604
605 static inline void sk_add_bind_node(struct sock *sk,
606 struct hlist_head *list)
607 {
608 hlist_add_head(&sk->sk_bind_node, list);
609 }
610
611 #define sk_for_each(__sk, list) \
612 hlist_for_each_entry(__sk, list, sk_node)
613 #define sk_for_each_rcu(__sk, list) \
614 hlist_for_each_entry_rcu(__sk, list, sk_node)
615 #define sk_nulls_for_each(__sk, node, list) \
616 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
617 #define sk_nulls_for_each_rcu(__sk, node, list) \
618 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
619 #define sk_for_each_from(__sk) \
620 hlist_for_each_entry_from(__sk, sk_node)
621 #define sk_nulls_for_each_from(__sk, node) \
622 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
623 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
624 #define sk_for_each_safe(__sk, tmp, list) \
625 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
626 #define sk_for_each_bound(__sk, list) \
627 hlist_for_each_entry(__sk, list, sk_bind_node)
628
629 static inline struct user_namespace *sk_user_ns(struct sock *sk)
630 {
631 /* Careful only use this in a context where these parameters
632 * can not change and must all be valid, such as recvmsg from
633 * userspace.
634 */
635 return sk->sk_socket->file->f_cred->user_ns;
636 }
637
638 /* Sock flags */
639 enum sock_flags {
640 SOCK_DEAD,
641 SOCK_DONE,
642 SOCK_URGINLINE,
643 SOCK_KEEPOPEN,
644 SOCK_LINGER,
645 SOCK_DESTROY,
646 SOCK_BROADCAST,
647 SOCK_TIMESTAMP,
648 SOCK_ZAPPED,
649 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
650 SOCK_DBG, /* %SO_DEBUG setting */
651 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
652 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
653 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
654 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
655 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
656 SOCK_TIMESTAMPING_TX_HARDWARE, /* %SOF_TIMESTAMPING_TX_HARDWARE */
657 SOCK_TIMESTAMPING_TX_SOFTWARE, /* %SOF_TIMESTAMPING_TX_SOFTWARE */
658 SOCK_TIMESTAMPING_RX_HARDWARE, /* %SOF_TIMESTAMPING_RX_HARDWARE */
659 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
660 SOCK_TIMESTAMPING_SOFTWARE, /* %SOF_TIMESTAMPING_SOFTWARE */
661 SOCK_TIMESTAMPING_RAW_HARDWARE, /* %SOF_TIMESTAMPING_RAW_HARDWARE */
662 SOCK_TIMESTAMPING_SYS_HARDWARE, /* %SOF_TIMESTAMPING_SYS_HARDWARE */
663 SOCK_FASYNC, /* fasync() active */
664 SOCK_RXQ_OVFL,
665 SOCK_ZEROCOPY, /* buffers from userspace */
666 SOCK_WIFI_STATUS, /* push wifi status to userspace */
667 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
668 * Will use last 4 bytes of packet sent from
669 * user-space instead.
670 */
671 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
672 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
673 };
674
675 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
676 {
677 nsk->sk_flags = osk->sk_flags;
678 }
679
680 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
681 {
682 __set_bit(flag, &sk->sk_flags);
683 }
684
685 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
686 {
687 __clear_bit(flag, &sk->sk_flags);
688 }
689
690 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
691 {
692 return test_bit(flag, &sk->sk_flags);
693 }
694
695 #ifdef CONFIG_NET
696 extern struct static_key memalloc_socks;
697 static inline int sk_memalloc_socks(void)
698 {
699 return static_key_false(&memalloc_socks);
700 }
701 #else
702
703 static inline int sk_memalloc_socks(void)
704 {
705 return 0;
706 }
707
708 #endif
709
710 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
711 {
712 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
713 }
714
715 static inline void sk_acceptq_removed(struct sock *sk)
716 {
717 sk->sk_ack_backlog--;
718 }
719
720 static inline void sk_acceptq_added(struct sock *sk)
721 {
722 sk->sk_ack_backlog++;
723 }
724
725 static inline bool sk_acceptq_is_full(const struct sock *sk)
726 {
727 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
728 }
729
730 /*
731 * Compute minimal free write space needed to queue new packets.
732 */
733 static inline int sk_stream_min_wspace(const struct sock *sk)
734 {
735 return sk->sk_wmem_queued >> 1;
736 }
737
738 static inline int sk_stream_wspace(const struct sock *sk)
739 {
740 return sk->sk_sndbuf - sk->sk_wmem_queued;
741 }
742
743 extern void sk_stream_write_space(struct sock *sk);
744
745 static inline bool sk_stream_memory_free(const struct sock *sk)
746 {
747 return sk->sk_wmem_queued < sk->sk_sndbuf;
748 }
749
750 /* OOB backlog add */
751 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
752 {
753 /* dont let skb dst not refcounted, we are going to leave rcu lock */
754 skb_dst_force(skb);
755
756 if (!sk->sk_backlog.tail)
757 sk->sk_backlog.head = skb;
758 else
759 sk->sk_backlog.tail->next = skb;
760
761 sk->sk_backlog.tail = skb;
762 skb->next = NULL;
763 }
764
765 /*
766 * Take into account size of receive queue and backlog queue
767 * Do not take into account this skb truesize,
768 * to allow even a single big packet to come.
769 */
770 static inline bool sk_rcvqueues_full(const struct sock *sk, const struct sk_buff *skb,
771 unsigned int limit)
772 {
773 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
774
775 return qsize > limit;
776 }
777
778 /* The per-socket spinlock must be held here. */
779 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
780 unsigned int limit)
781 {
782 if (sk_rcvqueues_full(sk, skb, limit))
783 {
784 #ifdef CONFIG_MTK_NET_LOGGING
785 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);
786 #endif
787 return -ENOBUFS;
788 }
789
790 /*
791 * If the skb was allocated from pfmemalloc reserves, only
792 * allow SOCK_MEMALLOC sockets to use it as this socket is
793 * helping free memory
794 */
795 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
796 return -ENOMEM;
797
798 __sk_add_backlog(sk, skb);
799 sk->sk_backlog.len += skb->truesize;
800 return 0;
801 }
802
803 extern int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
804
805 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
806 {
807 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
808 return __sk_backlog_rcv(sk, skb);
809
810 return sk->sk_backlog_rcv(sk, skb);
811 }
812
813 static inline void sock_rps_record_flow(const struct sock *sk)
814 {
815 #ifdef CONFIG_RPS
816 struct rps_sock_flow_table *sock_flow_table;
817
818 rcu_read_lock();
819 sock_flow_table = rcu_dereference(rps_sock_flow_table);
820 rps_record_sock_flow(sock_flow_table, sk->sk_rxhash);
821 rcu_read_unlock();
822 #endif
823 }
824
825 static inline void sock_rps_reset_flow(const struct sock *sk)
826 {
827 #ifdef CONFIG_RPS
828 struct rps_sock_flow_table *sock_flow_table;
829
830 rcu_read_lock();
831 sock_flow_table = rcu_dereference(rps_sock_flow_table);
832 rps_reset_sock_flow(sock_flow_table, sk->sk_rxhash);
833 rcu_read_unlock();
834 #endif
835 }
836
837 static inline void sock_rps_save_rxhash(struct sock *sk,
838 const struct sk_buff *skb)
839 {
840 #ifdef CONFIG_RPS
841 if (unlikely(sk->sk_rxhash != skb->rxhash)) {
842 sock_rps_reset_flow(sk);
843 sk->sk_rxhash = skb->rxhash;
844 }
845 #endif
846 }
847
848 static inline void sock_rps_reset_rxhash(struct sock *sk)
849 {
850 #ifdef CONFIG_RPS
851 sock_rps_reset_flow(sk);
852 sk->sk_rxhash = 0;
853 #endif
854 }
855
856 #define sk_wait_event(__sk, __timeo, __condition) \
857 ({ int __rc; \
858 release_sock(__sk); \
859 __rc = __condition; \
860 if (!__rc) { \
861 *(__timeo) = schedule_timeout(*(__timeo)); \
862 } \
863 lock_sock(__sk); \
864 __rc = __condition; \
865 __rc; \
866 })
867
868 extern int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
869 extern int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
870 extern void sk_stream_wait_close(struct sock *sk, long timeo_p);
871 extern int sk_stream_error(struct sock *sk, int flags, int err);
872 extern void sk_stream_kill_queues(struct sock *sk);
873 extern void sk_set_memalloc(struct sock *sk);
874 extern void sk_clear_memalloc(struct sock *sk);
875
876 extern int sk_wait_data(struct sock *sk, long *timeo);
877
878 struct request_sock_ops;
879 struct timewait_sock_ops;
880 struct inet_hashinfo;
881 struct raw_hashinfo;
882 struct module;
883
884 /*
885 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
886 * un-modified. Special care is taken when initializing object to zero.
887 */
888 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
889 {
890 if (offsetof(struct sock, sk_node.next) != 0)
891 memset(sk, 0, offsetof(struct sock, sk_node.next));
892 memset(&sk->sk_node.pprev, 0,
893 size - offsetof(struct sock, sk_node.pprev));
894 }
895
896 /* Networking protocol blocks we attach to sockets.
897 * socket layer -> transport layer interface
898 * transport -> network interface is defined by struct inet_proto
899 */
900 struct proto {
901 void (*close)(struct sock *sk,
902 long timeout);
903 int (*connect)(struct sock *sk,
904 struct sockaddr *uaddr,
905 int addr_len);
906 int (*disconnect)(struct sock *sk, int flags);
907
908 struct sock * (*accept)(struct sock *sk, int flags, int *err);
909
910 int (*ioctl)(struct sock *sk, int cmd,
911 unsigned long arg);
912 int (*init)(struct sock *sk);
913 void (*destroy)(struct sock *sk);
914 void (*shutdown)(struct sock *sk, int how);
915 int (*setsockopt)(struct sock *sk, int level,
916 int optname, char __user *optval,
917 unsigned int optlen);
918 int (*getsockopt)(struct sock *sk, int level,
919 int optname, char __user *optval,
920 int __user *option);
921 #ifdef CONFIG_COMPAT
922 int (*compat_setsockopt)(struct sock *sk,
923 int level,
924 int optname, char __user *optval,
925 unsigned int optlen);
926 int (*compat_getsockopt)(struct sock *sk,
927 int level,
928 int optname, char __user *optval,
929 int __user *option);
930 int (*compat_ioctl)(struct sock *sk,
931 unsigned int cmd, unsigned long arg);
932 #endif
933 int (*sendmsg)(struct kiocb *iocb, struct sock *sk,
934 struct msghdr *msg, size_t len);
935 int (*recvmsg)(struct kiocb *iocb, struct sock *sk,
936 struct msghdr *msg,
937 size_t len, int noblock, int flags,
938 int *addr_len);
939 int (*sendpage)(struct sock *sk, struct page *page,
940 int offset, size_t size, int flags);
941 int (*bind)(struct sock *sk,
942 struct sockaddr *uaddr, int addr_len);
943
944 int (*backlog_rcv) (struct sock *sk,
945 struct sk_buff *skb);
946
947 void (*release_cb)(struct sock *sk);
948
949 /* Keeping track of sk's, looking them up, and port selection methods. */
950 void (*hash)(struct sock *sk);
951 void (*unhash)(struct sock *sk);
952 void (*rehash)(struct sock *sk);
953 int (*get_port)(struct sock *sk, unsigned short snum);
954 void (*clear_sk)(struct sock *sk, int size);
955
956 /* Keeping track of sockets in use */
957 #ifdef CONFIG_PROC_FS
958 unsigned int inuse_idx;
959 #endif
960
961 /* Memory pressure */
962 void (*enter_memory_pressure)(struct sock *sk);
963 atomic_long_t *memory_allocated; /* Current allocated memory. */
964 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
965 /*
966 * Pressure flag: try to collapse.
967 * Technical note: it is used by multiple contexts non atomically.
968 * All the __sk_mem_schedule() is of this nature: accounting
969 * is strict, actions are advisory and have some latency.
970 */
971 int *memory_pressure;
972 long *sysctl_mem;
973 int *sysctl_wmem;
974 int *sysctl_rmem;
975 int max_header;
976 bool no_autobind;
977
978 struct kmem_cache *slab;
979 unsigned int obj_size;
980 int slab_flags;
981
982 struct percpu_counter *orphan_count;
983
984 struct request_sock_ops *rsk_prot;
985 struct timewait_sock_ops *twsk_prot;
986
987 union {
988 struct inet_hashinfo *hashinfo;
989 struct udp_table *udp_table;
990 struct raw_hashinfo *raw_hash;
991 } h;
992
993 struct module *owner;
994
995 char name[32];
996
997 struct list_head node;
998 #ifdef SOCK_REFCNT_DEBUG
999 atomic_t socks;
1000 #endif
1001 #ifdef CONFIG_MEMCG_KMEM
1002 /*
1003 * cgroup specific init/deinit functions. Called once for all
1004 * protocols that implement it, from cgroups populate function.
1005 * This function has to setup any files the protocol want to
1006 * appear in the kmem cgroup filesystem.
1007 */
1008 int (*init_cgroup)(struct mem_cgroup *memcg,
1009 struct cgroup_subsys *ss);
1010 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1011 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1012 #endif
1013 };
1014
1015 /*
1016 * Bits in struct cg_proto.flags
1017 */
1018 enum cg_proto_flags {
1019 /* Currently active and new sockets should be assigned to cgroups */
1020 MEMCG_SOCK_ACTIVE,
1021 /* It was ever activated; we must disarm static keys on destruction */
1022 MEMCG_SOCK_ACTIVATED,
1023 };
1024
1025 struct cg_proto {
1026 void (*enter_memory_pressure)(struct sock *sk);
1027 struct res_counter *memory_allocated; /* Current allocated memory. */
1028 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1029 int *memory_pressure;
1030 long *sysctl_mem;
1031 unsigned long flags;
1032 /*
1033 * memcg field is used to find which memcg we belong directly
1034 * Each memcg struct can hold more than one cg_proto, so container_of
1035 * won't really cut.
1036 *
1037 * The elegant solution would be having an inverse function to
1038 * proto_cgroup in struct proto, but that means polluting the structure
1039 * for everybody, instead of just for memcg users.
1040 */
1041 struct mem_cgroup *memcg;
1042 };
1043
1044 extern int proto_register(struct proto *prot, int alloc_slab);
1045 extern void proto_unregister(struct proto *prot);
1046
1047 static inline bool memcg_proto_active(struct cg_proto *cg_proto)
1048 {
1049 return test_bit(MEMCG_SOCK_ACTIVE, &cg_proto->flags);
1050 }
1051
1052 static inline bool memcg_proto_activated(struct cg_proto *cg_proto)
1053 {
1054 return test_bit(MEMCG_SOCK_ACTIVATED, &cg_proto->flags);
1055 }
1056
1057 #ifdef SOCK_REFCNT_DEBUG
1058 static inline void sk_refcnt_debug_inc(struct sock *sk)
1059 {
1060 atomic_inc(&sk->sk_prot->socks);
1061 }
1062
1063 static inline void sk_refcnt_debug_dec(struct sock *sk)
1064 {
1065 atomic_dec(&sk->sk_prot->socks);
1066 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1067 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1068 }
1069
1070 static inline void sk_refcnt_debug_release(const struct sock *sk)
1071 {
1072 if (atomic_read(&sk->sk_refcnt) != 1)
1073 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1074 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1075 }
1076 #else /* SOCK_REFCNT_DEBUG */
1077 #define sk_refcnt_debug_inc(sk) do { } while (0)
1078 #define sk_refcnt_debug_dec(sk) do { } while (0)
1079 #define sk_refcnt_debug_release(sk) do { } while (0)
1080 #endif /* SOCK_REFCNT_DEBUG */
1081
1082 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1083 extern struct static_key memcg_socket_limit_enabled;
1084 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1085 struct cg_proto *cg_proto)
1086 {
1087 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1088 }
1089 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1090 #else
1091 #define mem_cgroup_sockets_enabled 0
1092 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1093 struct cg_proto *cg_proto)
1094 {
1095 return NULL;
1096 }
1097 #endif
1098
1099
1100 static inline bool sk_has_memory_pressure(const struct sock *sk)
1101 {
1102 return sk->sk_prot->memory_pressure != NULL;
1103 }
1104
1105 static inline bool sk_under_memory_pressure(const struct sock *sk)
1106 {
1107 if (!sk->sk_prot->memory_pressure)
1108 return false;
1109
1110 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1111 return !!*sk->sk_cgrp->memory_pressure;
1112
1113 return !!*sk->sk_prot->memory_pressure;
1114 }
1115
1116 static inline void sk_leave_memory_pressure(struct sock *sk)
1117 {
1118 int *memory_pressure = sk->sk_prot->memory_pressure;
1119
1120 if (!memory_pressure)
1121 return;
1122
1123 if (*memory_pressure)
1124 *memory_pressure = 0;
1125
1126 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1127 struct cg_proto *cg_proto = sk->sk_cgrp;
1128 struct proto *prot = sk->sk_prot;
1129
1130 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1131 if (*cg_proto->memory_pressure)
1132 *cg_proto->memory_pressure = 0;
1133 }
1134
1135 }
1136
1137 static inline void sk_enter_memory_pressure(struct sock *sk)
1138 {
1139 if (!sk->sk_prot->enter_memory_pressure)
1140 return;
1141
1142 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1143 struct cg_proto *cg_proto = sk->sk_cgrp;
1144 struct proto *prot = sk->sk_prot;
1145
1146 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1147 cg_proto->enter_memory_pressure(sk);
1148 }
1149
1150 sk->sk_prot->enter_memory_pressure(sk);
1151 }
1152
1153 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1154 {
1155 long *prot = sk->sk_prot->sysctl_mem;
1156 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1157 prot = sk->sk_cgrp->sysctl_mem;
1158 return prot[index];
1159 }
1160
1161 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1162 unsigned long amt,
1163 int *parent_status)
1164 {
1165 struct res_counter *fail;
1166 int ret;
1167
1168 ret = res_counter_charge_nofail(prot->memory_allocated,
1169 amt << PAGE_SHIFT, &fail);
1170 if (ret < 0)
1171 *parent_status = OVER_LIMIT;
1172 }
1173
1174 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1175 unsigned long amt)
1176 {
1177 res_counter_uncharge(prot->memory_allocated, amt << PAGE_SHIFT);
1178 }
1179
1180 static inline u64 memcg_memory_allocated_read(struct cg_proto *prot)
1181 {
1182 u64 ret;
1183 ret = res_counter_read_u64(prot->memory_allocated, RES_USAGE);
1184 return ret >> PAGE_SHIFT;
1185 }
1186
1187 static inline long
1188 sk_memory_allocated(const struct sock *sk)
1189 {
1190 struct proto *prot = sk->sk_prot;
1191 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1192 return memcg_memory_allocated_read(sk->sk_cgrp);
1193
1194 return atomic_long_read(prot->memory_allocated);
1195 }
1196
1197 static inline long
1198 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1199 {
1200 struct proto *prot = sk->sk_prot;
1201
1202 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1203 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1204 /* update the root cgroup regardless */
1205 atomic_long_add_return(amt, prot->memory_allocated);
1206 return memcg_memory_allocated_read(sk->sk_cgrp);
1207 }
1208
1209 return atomic_long_add_return(amt, prot->memory_allocated);
1210 }
1211
1212 static inline void
1213 sk_memory_allocated_sub(struct sock *sk, int amt)
1214 {
1215 struct proto *prot = sk->sk_prot;
1216
1217 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1218 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1219
1220 atomic_long_sub(amt, prot->memory_allocated);
1221 }
1222
1223 static inline void sk_sockets_allocated_dec(struct sock *sk)
1224 {
1225 struct proto *prot = sk->sk_prot;
1226
1227 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1228 struct cg_proto *cg_proto = sk->sk_cgrp;
1229
1230 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1231 percpu_counter_dec(cg_proto->sockets_allocated);
1232 }
1233
1234 percpu_counter_dec(prot->sockets_allocated);
1235 }
1236
1237 static inline void sk_sockets_allocated_inc(struct sock *sk)
1238 {
1239 struct proto *prot = sk->sk_prot;
1240
1241 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1242 struct cg_proto *cg_proto = sk->sk_cgrp;
1243
1244 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1245 percpu_counter_inc(cg_proto->sockets_allocated);
1246 }
1247
1248 percpu_counter_inc(prot->sockets_allocated);
1249 }
1250
1251 static inline int
1252 sk_sockets_allocated_read_positive(struct sock *sk)
1253 {
1254 struct proto *prot = sk->sk_prot;
1255
1256 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1257 return percpu_counter_read_positive(sk->sk_cgrp->sockets_allocated);
1258
1259 return percpu_counter_read_positive(prot->sockets_allocated);
1260 }
1261
1262 static inline int
1263 proto_sockets_allocated_sum_positive(struct proto *prot)
1264 {
1265 return percpu_counter_sum_positive(prot->sockets_allocated);
1266 }
1267
1268 static inline long
1269 proto_memory_allocated(struct proto *prot)
1270 {
1271 return atomic_long_read(prot->memory_allocated);
1272 }
1273
1274 static inline bool
1275 proto_memory_pressure(struct proto *prot)
1276 {
1277 if (!prot->memory_pressure)
1278 return false;
1279 return !!*prot->memory_pressure;
1280 }
1281
1282
1283 #ifdef CONFIG_PROC_FS
1284 /* Called with local bh disabled */
1285 extern void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1286 extern int sock_prot_inuse_get(struct net *net, struct proto *proto);
1287 #else
1288 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1289 int inc)
1290 {
1291 }
1292 #endif
1293
1294
1295 /* With per-bucket locks this operation is not-atomic, so that
1296 * this version is not worse.
1297 */
1298 static inline void __sk_prot_rehash(struct sock *sk)
1299 {
1300 sk->sk_prot->unhash(sk);
1301 sk->sk_prot->hash(sk);
1302 }
1303
1304 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1305
1306 /* About 10 seconds */
1307 #define SOCK_DESTROY_TIME (10*HZ)
1308
1309 /* Sockets 0-1023 can't be bound to unless you are superuser */
1310 #define PROT_SOCK 1024
1311
1312 #define SHUTDOWN_MASK 3
1313 #define RCV_SHUTDOWN 1
1314 #define SEND_SHUTDOWN 2
1315
1316 #define SOCK_SNDBUF_LOCK 1
1317 #define SOCK_RCVBUF_LOCK 2
1318 #define SOCK_BINDADDR_LOCK 4
1319 #define SOCK_BINDPORT_LOCK 8
1320
1321 /* sock_iocb: used to kick off async processing of socket ios */
1322 struct sock_iocb {
1323 struct list_head list;
1324
1325 int flags;
1326 int size;
1327 struct socket *sock;
1328 struct sock *sk;
1329 struct scm_cookie *scm;
1330 struct msghdr *msg, async_msg;
1331 struct kiocb *kiocb;
1332 };
1333
1334 static inline struct sock_iocb *kiocb_to_siocb(struct kiocb *iocb)
1335 {
1336 return (struct sock_iocb *)iocb->private;
1337 }
1338
1339 static inline struct kiocb *siocb_to_kiocb(struct sock_iocb *si)
1340 {
1341 return si->kiocb;
1342 }
1343
1344 struct socket_alloc {
1345 struct socket socket;
1346 struct inode vfs_inode;
1347 };
1348
1349 static inline struct socket *SOCKET_I(struct inode *inode)
1350 {
1351 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1352 }
1353
1354 static inline struct inode *SOCK_INODE(struct socket *socket)
1355 {
1356 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1357 }
1358
1359 /*
1360 * Functions for memory accounting
1361 */
1362 extern int __sk_mem_schedule(struct sock *sk, int size, int kind);
1363 extern void __sk_mem_reclaim(struct sock *sk);
1364
1365 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1366 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1367 #define SK_MEM_SEND 0
1368 #define SK_MEM_RECV 1
1369
1370 static inline int sk_mem_pages(int amt)
1371 {
1372 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1373 }
1374
1375 static inline bool sk_has_account(struct sock *sk)
1376 {
1377 /* return true if protocol supports memory accounting */
1378 return !!sk->sk_prot->memory_allocated;
1379 }
1380
1381 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1382 {
1383 if (!sk_has_account(sk))
1384 return true;
1385 return size <= sk->sk_forward_alloc ||
1386 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1387 }
1388
1389 static inline bool
1390 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1391 {
1392 if (!sk_has_account(sk))
1393 return true;
1394 return size<= sk->sk_forward_alloc ||
1395 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1396 skb_pfmemalloc(skb);
1397 }
1398
1399 static inline void sk_mem_reclaim(struct sock *sk)
1400 {
1401 if (!sk_has_account(sk))
1402 return;
1403 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1404 __sk_mem_reclaim(sk);
1405 }
1406
1407 static inline void sk_mem_reclaim_partial(struct sock *sk)
1408 {
1409 if (!sk_has_account(sk))
1410 return;
1411 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1412 __sk_mem_reclaim(sk);
1413 }
1414
1415 static inline void sk_mem_charge(struct sock *sk, int size)
1416 {
1417 if (!sk_has_account(sk))
1418 return;
1419 sk->sk_forward_alloc -= size;
1420 }
1421
1422 static inline void sk_mem_uncharge(struct sock *sk, int size)
1423 {
1424 if (!sk_has_account(sk))
1425 return;
1426 sk->sk_forward_alloc += size;
1427 }
1428
1429 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1430 {
1431 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1432 sk->sk_wmem_queued -= skb->truesize;
1433 sk_mem_uncharge(sk, skb->truesize);
1434 __kfree_skb(skb);
1435 }
1436
1437 /* Used by processes to "lock" a socket state, so that
1438 * interrupts and bottom half handlers won't change it
1439 * from under us. It essentially blocks any incoming
1440 * packets, so that we won't get any new data or any
1441 * packets that change the state of the socket.
1442 *
1443 * While locked, BH processing will add new packets to
1444 * the backlog queue. This queue is processed by the
1445 * owner of the socket lock right before it is released.
1446 *
1447 * Since ~2.3.5 it is also exclusive sleep lock serializing
1448 * accesses from user process context.
1449 */
1450 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1451
1452 static inline void sock_release_ownership(struct sock *sk)
1453 {
1454 sk->sk_lock.owned = 0;
1455 }
1456
1457 /*
1458 * Macro so as to not evaluate some arguments when
1459 * lockdep is not enabled.
1460 *
1461 * Mark both the sk_lock and the sk_lock.slock as a
1462 * per-address-family lock class.
1463 */
1464 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1465 do { \
1466 sk->sk_lock.owned = 0; \
1467 init_waitqueue_head(&sk->sk_lock.wq); \
1468 spin_lock_init(&(sk)->sk_lock.slock); \
1469 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1470 sizeof((sk)->sk_lock)); \
1471 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1472 (skey), (sname)); \
1473 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1474 } while (0)
1475
1476 extern void lock_sock_nested(struct sock *sk, int subclass);
1477
1478 static inline void lock_sock(struct sock *sk)
1479 {
1480 lock_sock_nested(sk, 0);
1481 }
1482
1483 extern void release_sock(struct sock *sk);
1484
1485 /* BH context may only use the following locking interface. */
1486 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1487 #define bh_lock_sock_nested(__sk) \
1488 spin_lock_nested(&((__sk)->sk_lock.slock), \
1489 SINGLE_DEPTH_NESTING)
1490 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1491
1492 extern bool lock_sock_fast(struct sock *sk);
1493 /**
1494 * unlock_sock_fast - complement of lock_sock_fast
1495 * @sk: socket
1496 * @slow: slow mode
1497 *
1498 * fast unlock socket for user context.
1499 * If slow mode is on, we call regular release_sock()
1500 */
1501 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1502 {
1503 if (slow)
1504 release_sock(sk);
1505 else
1506 spin_unlock_bh(&sk->sk_lock.slock);
1507 }
1508
1509
1510 extern struct sock *sk_alloc(struct net *net, int family,
1511 gfp_t priority,
1512 struct proto *prot);
1513 extern void sk_free(struct sock *sk);
1514 extern void sk_release_kernel(struct sock *sk);
1515 extern struct sock *sk_clone_lock(const struct sock *sk,
1516 const gfp_t priority);
1517
1518 extern struct sk_buff *sock_wmalloc(struct sock *sk,
1519 unsigned long size, int force,
1520 gfp_t priority);
1521 extern struct sk_buff *sock_rmalloc(struct sock *sk,
1522 unsigned long size, int force,
1523 gfp_t priority);
1524 extern void sock_wfree(struct sk_buff *skb);
1525 extern void sock_rfree(struct sk_buff *skb);
1526 extern void sock_edemux(struct sk_buff *skb);
1527
1528 extern int sock_setsockopt(struct socket *sock, int level,
1529 int op, char __user *optval,
1530 unsigned int optlen);
1531
1532 extern int sock_getsockopt(struct socket *sock, int level,
1533 int op, char __user *optval,
1534 int __user *optlen);
1535 extern struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1536 unsigned long size,
1537 int noblock,
1538 int *errcode);
1539 extern struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1540 unsigned long header_len,
1541 unsigned long data_len,
1542 int noblock,
1543 int *errcode);
1544 extern void *sock_kmalloc(struct sock *sk, int size,
1545 gfp_t priority);
1546 extern void sock_kfree_s(struct sock *sk, void *mem, int size);
1547 extern void sk_send_sigurg(struct sock *sk);
1548
1549 /*
1550 * Functions to fill in entries in struct proto_ops when a protocol
1551 * does not implement a particular function.
1552 */
1553 extern int sock_no_bind(struct socket *,
1554 struct sockaddr *, int);
1555 extern int sock_no_connect(struct socket *,
1556 struct sockaddr *, int, int);
1557 extern int sock_no_socketpair(struct socket *,
1558 struct socket *);
1559 extern int sock_no_accept(struct socket *,
1560 struct socket *, int);
1561 extern int sock_no_getname(struct socket *,
1562 struct sockaddr *, int *, int);
1563 extern unsigned int sock_no_poll(struct file *, struct socket *,
1564 struct poll_table_struct *);
1565 extern int sock_no_ioctl(struct socket *, unsigned int,
1566 unsigned long);
1567 extern int sock_no_listen(struct socket *, int);
1568 extern int sock_no_shutdown(struct socket *, int);
1569 extern int sock_no_getsockopt(struct socket *, int , int,
1570 char __user *, int __user *);
1571 extern int sock_no_setsockopt(struct socket *, int, int,
1572 char __user *, unsigned int);
1573 extern int sock_no_sendmsg(struct kiocb *, struct socket *,
1574 struct msghdr *, size_t);
1575 extern int sock_no_recvmsg(struct kiocb *, struct socket *,
1576 struct msghdr *, size_t, int);
1577 extern int sock_no_mmap(struct file *file,
1578 struct socket *sock,
1579 struct vm_area_struct *vma);
1580 extern ssize_t sock_no_sendpage(struct socket *sock,
1581 struct page *page,
1582 int offset, size_t size,
1583 int flags);
1584
1585 /*
1586 * Functions to fill in entries in struct proto_ops when a protocol
1587 * uses the inet style.
1588 */
1589 extern int sock_common_getsockopt(struct socket *sock, int level, int optname,
1590 char __user *optval, int __user *optlen);
1591 extern int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1592 struct msghdr *msg, size_t size, int flags);
1593 extern int sock_common_setsockopt(struct socket *sock, int level, int optname,
1594 char __user *optval, unsigned int optlen);
1595 extern int compat_sock_common_getsockopt(struct socket *sock, int level,
1596 int optname, char __user *optval, int __user *optlen);
1597 extern int compat_sock_common_setsockopt(struct socket *sock, int level,
1598 int optname, char __user *optval, unsigned int optlen);
1599
1600 extern void sk_common_release(struct sock *sk);
1601
1602 /*
1603 * Default socket callbacks and setup code
1604 */
1605
1606 /* Initialise core socket variables */
1607 extern void sock_init_data(struct socket *sock, struct sock *sk);
1608
1609 extern void sk_filter_release_rcu(struct rcu_head *rcu);
1610
1611 /**
1612 * sk_filter_release - release a socket filter
1613 * @fp: filter to remove
1614 *
1615 * Remove a filter from a socket and release its resources.
1616 */
1617
1618 static inline void sk_filter_release(struct sk_filter *fp)
1619 {
1620 if (atomic_dec_and_test(&fp->refcnt))
1621 call_rcu(&fp->rcu, sk_filter_release_rcu);
1622 }
1623
1624 static inline void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp)
1625 {
1626 unsigned int size = sk_filter_len(fp);
1627
1628 atomic_sub(size, &sk->sk_omem_alloc);
1629 sk_filter_release(fp);
1630 }
1631
1632 static inline void sk_filter_charge(struct sock *sk, struct sk_filter *fp)
1633 {
1634 atomic_inc(&fp->refcnt);
1635 atomic_add(sk_filter_len(fp), &sk->sk_omem_alloc);
1636 }
1637
1638 /*
1639 * Socket reference counting postulates.
1640 *
1641 * * Each user of socket SHOULD hold a reference count.
1642 * * Each access point to socket (an hash table bucket, reference from a list,
1643 * running timer, skb in flight MUST hold a reference count.
1644 * * When reference count hits 0, it means it will never increase back.
1645 * * When reference count hits 0, it means that no references from
1646 * outside exist to this socket and current process on current CPU
1647 * is last user and may/should destroy this socket.
1648 * * sk_free is called from any context: process, BH, IRQ. When
1649 * it is called, socket has no references from outside -> sk_free
1650 * may release descendant resources allocated by the socket, but
1651 * to the time when it is called, socket is NOT referenced by any
1652 * hash tables, lists etc.
1653 * * Packets, delivered from outside (from network or from another process)
1654 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1655 * when they sit in queue. Otherwise, packets will leak to hole, when
1656 * socket is looked up by one cpu and unhasing is made by another CPU.
1657 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1658 * (leak to backlog). Packet socket does all the processing inside
1659 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1660 * use separate SMP lock, so that they are prone too.
1661 */
1662
1663 /* Ungrab socket and destroy it, if it was the last reference. */
1664 static inline void sock_put(struct sock *sk)
1665 {
1666 if (atomic_dec_and_test(&sk->sk_refcnt))
1667 sk_free(sk);
1668 }
1669
1670 extern int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1671 const int nested);
1672
1673 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1674 {
1675 sk->sk_tx_queue_mapping = tx_queue;
1676 }
1677
1678 static inline void sk_tx_queue_clear(struct sock *sk)
1679 {
1680 sk->sk_tx_queue_mapping = -1;
1681 }
1682
1683 static inline int sk_tx_queue_get(const struct sock *sk)
1684 {
1685 return sk ? sk->sk_tx_queue_mapping : -1;
1686 }
1687
1688 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1689 {
1690 sk_tx_queue_clear(sk);
1691 sk->sk_socket = sock;
1692 }
1693
1694 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1695 {
1696 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1697 return &rcu_dereference_raw(sk->sk_wq)->wait;
1698 }
1699 /* Detach socket from process context.
1700 * Announce socket dead, detach it from wait queue and inode.
1701 * Note that parent inode held reference count on this struct sock,
1702 * we do not release it in this function, because protocol
1703 * probably wants some additional cleanups or even continuing
1704 * to work with this socket (TCP).
1705 */
1706 static inline void sock_orphan(struct sock *sk)
1707 {
1708 write_lock_bh(&sk->sk_callback_lock);
1709 sock_set_flag(sk, SOCK_DEAD);
1710 sk_set_socket(sk, NULL);
1711 sk->sk_wq = NULL;
1712 write_unlock_bh(&sk->sk_callback_lock);
1713 }
1714
1715 static inline void sock_graft(struct sock *sk, struct socket *parent)
1716 {
1717 write_lock_bh(&sk->sk_callback_lock);
1718 sk->sk_wq = parent->wq;
1719 parent->sk = sk;
1720 sk_set_socket(sk, parent);
1721 security_sock_graft(sk, parent);
1722 write_unlock_bh(&sk->sk_callback_lock);
1723 }
1724
1725 extern kuid_t sock_i_uid(struct sock *sk);
1726 extern unsigned long sock_i_ino(struct sock *sk);
1727
1728 static inline struct dst_entry *
1729 __sk_dst_get(struct sock *sk)
1730 {
1731 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1732 lockdep_is_held(&sk->sk_lock.slock));
1733 }
1734
1735 static inline struct dst_entry *
1736 sk_dst_get(struct sock *sk)
1737 {
1738 struct dst_entry *dst;
1739
1740 rcu_read_lock();
1741 dst = rcu_dereference(sk->sk_dst_cache);
1742 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1743 dst = NULL;
1744 rcu_read_unlock();
1745 return dst;
1746 }
1747
1748 extern void sk_reset_txq(struct sock *sk);
1749
1750 static inline void dst_negative_advice(struct sock *sk)
1751 {
1752 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1753
1754 if (dst && dst->ops->negative_advice) {
1755 ndst = dst->ops->negative_advice(dst);
1756
1757 if (ndst != dst) {
1758 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1759 sk_reset_txq(sk);
1760 }
1761 }
1762 }
1763
1764 static inline void
1765 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1766 {
1767 struct dst_entry *old_dst;
1768
1769 sk_tx_queue_clear(sk);
1770 /*
1771 * This can be called while sk is owned by the caller only,
1772 * with no state that can be checked in a rcu_dereference_check() cond
1773 */
1774 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1775 rcu_assign_pointer(sk->sk_dst_cache, dst);
1776 dst_release(old_dst);
1777 }
1778
1779 static inline void
1780 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1781 {
1782 struct dst_entry *old_dst;
1783
1784 sk_tx_queue_clear(sk);
1785 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1786 dst_release(old_dst);
1787 }
1788
1789 static inline void
1790 __sk_dst_reset(struct sock *sk)
1791 {
1792 __sk_dst_set(sk, NULL);
1793 }
1794
1795 static inline void
1796 sk_dst_reset(struct sock *sk)
1797 {
1798 sk_dst_set(sk, NULL);
1799 }
1800
1801 extern struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1802
1803 extern struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1804
1805 static inline bool sk_can_gso(const struct sock *sk)
1806 {
1807 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1808 }
1809
1810 extern void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1811
1812 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1813 {
1814 sk->sk_route_nocaps |= flags;
1815 sk->sk_route_caps &= ~flags;
1816 }
1817
1818 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1819 char __user *from, char *to,
1820 int copy, int offset)
1821 {
1822 if (skb->ip_summed == CHECKSUM_NONE) {
1823 int err = 0;
1824 __wsum csum = csum_and_copy_from_user(from, to, copy, 0, &err);
1825 if (err)
1826 return err;
1827 skb->csum = csum_block_add(skb->csum, csum, offset);
1828 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1829 if (!access_ok(VERIFY_READ, from, copy) ||
1830 __copy_from_user_nocache(to, from, copy))
1831 return -EFAULT;
1832 } else if (copy_from_user(to, from, copy))
1833 return -EFAULT;
1834
1835 return 0;
1836 }
1837
1838 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1839 char __user *from, int copy)
1840 {
1841 int err, offset = skb->len;
1842
1843 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1844 copy, offset);
1845 if (err)
1846 __skb_trim(skb, offset);
1847
1848 return err;
1849 }
1850
1851 static inline int skb_copy_to_page_nocache(struct sock *sk, char __user *from,
1852 struct sk_buff *skb,
1853 struct page *page,
1854 int off, int copy)
1855 {
1856 int err;
1857
1858 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1859 copy, skb->len);
1860 if (err)
1861 return err;
1862
1863 skb->len += copy;
1864 skb->data_len += copy;
1865 skb->truesize += copy;
1866 sk->sk_wmem_queued += copy;
1867 sk_mem_charge(sk, copy);
1868 return 0;
1869 }
1870
1871 static inline int skb_copy_to_page(struct sock *sk, char __user *from,
1872 struct sk_buff *skb, struct page *page,
1873 int off, int copy)
1874 {
1875 if (skb->ip_summed == CHECKSUM_NONE) {
1876 int err = 0;
1877 __wsum csum = csum_and_copy_from_user(from,
1878 page_address(page) + off,
1879 copy, 0, &err);
1880 if (err)
1881 return err;
1882 skb->csum = csum_block_add(skb->csum, csum, skb->len);
1883 } else if (copy_from_user(page_address(page) + off, from, copy))
1884 return -EFAULT;
1885
1886 skb->len += copy;
1887 skb->data_len += copy;
1888 skb->truesize += copy;
1889 sk->sk_wmem_queued += copy;
1890 sk_mem_charge(sk, copy);
1891 return 0;
1892 }
1893
1894 /**
1895 * sk_wmem_alloc_get - returns write allocations
1896 * @sk: socket
1897 *
1898 * Returns sk_wmem_alloc minus initial offset of one
1899 */
1900 static inline int sk_wmem_alloc_get(const struct sock *sk)
1901 {
1902 return atomic_read(&sk->sk_wmem_alloc) - 1;
1903 }
1904
1905 /**
1906 * sk_rmem_alloc_get - returns read allocations
1907 * @sk: socket
1908 *
1909 * Returns sk_rmem_alloc
1910 */
1911 static inline int sk_rmem_alloc_get(const struct sock *sk)
1912 {
1913 return atomic_read(&sk->sk_rmem_alloc);
1914 }
1915
1916 /**
1917 * sk_has_allocations - check if allocations are outstanding
1918 * @sk: socket
1919 *
1920 * Returns true if socket has write or read allocations
1921 */
1922 static inline bool sk_has_allocations(const struct sock *sk)
1923 {
1924 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1925 }
1926
1927 /**
1928 * wq_has_sleeper - check if there are any waiting processes
1929 * @wq: struct socket_wq
1930 *
1931 * Returns true if socket_wq has waiting processes
1932 *
1933 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1934 * barrier call. They were added due to the race found within the tcp code.
1935 *
1936 * Consider following tcp code paths:
1937 *
1938 * CPU1 CPU2
1939 *
1940 * sys_select receive packet
1941 * ... ...
1942 * __add_wait_queue update tp->rcv_nxt
1943 * ... ...
1944 * tp->rcv_nxt check sock_def_readable
1945 * ... {
1946 * schedule rcu_read_lock();
1947 * wq = rcu_dereference(sk->sk_wq);
1948 * if (wq && waitqueue_active(&wq->wait))
1949 * wake_up_interruptible(&wq->wait)
1950 * ...
1951 * }
1952 *
1953 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1954 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1955 * could then endup calling schedule and sleep forever if there are no more
1956 * data on the socket.
1957 *
1958 */
1959 static inline bool wq_has_sleeper(struct socket_wq *wq)
1960 {
1961 /* We need to be sure we are in sync with the
1962 * add_wait_queue modifications to the wait queue.
1963 *
1964 * This memory barrier is paired in the sock_poll_wait.
1965 */
1966 smp_mb();
1967 return wq && waitqueue_active(&wq->wait);
1968 }
1969
1970 /**
1971 * sock_poll_wait - place memory barrier behind the poll_wait call.
1972 * @filp: file
1973 * @wait_address: socket wait queue
1974 * @p: poll_table
1975 *
1976 * See the comments in the wq_has_sleeper function.
1977 */
1978 static inline void sock_poll_wait(struct file *filp,
1979 wait_queue_head_t *wait_address, poll_table *p)
1980 {
1981 if (!poll_does_not_wait(p) && wait_address) {
1982 poll_wait(filp, wait_address, p);
1983 /* We need to be sure we are in sync with the
1984 * socket flags modification.
1985 *
1986 * This memory barrier is paired in the wq_has_sleeper.
1987 */
1988 smp_mb();
1989 }
1990 }
1991
1992 /*
1993 * Queue a received datagram if it will fit. Stream and sequenced
1994 * protocols can't normally use this as they need to fit buffers in
1995 * and play with them.
1996 *
1997 * Inlined as it's very short and called for pretty much every
1998 * packet ever received.
1999 */
2000
2001 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
2002 {
2003 skb_orphan(skb);
2004 skb->sk = sk;
2005 skb->destructor = sock_wfree;
2006 /*
2007 * We used to take a refcount on sk, but following operation
2008 * is enough to guarantee sk_free() wont free this sock until
2009 * all in-flight packets are completed
2010 */
2011 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
2012 }
2013
2014 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2015 {
2016 skb_orphan(skb);
2017 skb->sk = sk;
2018 skb->destructor = sock_rfree;
2019 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2020 sk_mem_charge(sk, skb->truesize);
2021 }
2022
2023 extern void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2024 unsigned long expires);
2025
2026 extern void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2027
2028 extern int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2029
2030 extern int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2031
2032 /*
2033 * Recover an error report and clear atomically
2034 */
2035
2036 static inline int sock_error(struct sock *sk)
2037 {
2038 int err;
2039 if (likely(!sk->sk_err))
2040 return 0;
2041 err = xchg(&sk->sk_err, 0);
2042 return -err;
2043 }
2044
2045 static inline unsigned long sock_wspace(struct sock *sk)
2046 {
2047 int amt = 0;
2048
2049 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2050 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
2051 if (amt < 0)
2052 amt = 0;
2053 }
2054 return amt;
2055 }
2056
2057 static inline void sk_wake_async(struct sock *sk, int how, int band)
2058 {
2059 if (sock_flag(sk, SOCK_FASYNC))
2060 sock_wake_async(sk->sk_socket, how, band);
2061 }
2062
2063 #define SOCK_MIN_SNDBUF 2048
2064 /*
2065 * Since sk_rmem_alloc sums skb->truesize, even a small frame might need
2066 * sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak
2067 */
2068 #define SOCK_MIN_RCVBUF (2048 + sizeof(struct sk_buff))
2069
2070 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2071 {
2072 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2073 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2074 sk->sk_sndbuf = max(sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2075 }
2076 }
2077
2078 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp);
2079
2080 /**
2081 * sk_page_frag - return an appropriate page_frag
2082 * @sk: socket
2083 *
2084 * If socket allocation mode allows current thread to sleep, it means its
2085 * safe to use the per task page_frag instead of the per socket one.
2086 */
2087 static inline struct page_frag *sk_page_frag(struct sock *sk)
2088 {
2089 if (sk->sk_allocation & __GFP_WAIT)
2090 return &current->task_frag;
2091
2092 return &sk->sk_frag;
2093 }
2094
2095 extern bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2096
2097 /*
2098 * Default write policy as shown to user space via poll/select/SIGIO
2099 */
2100 static inline bool sock_writeable(const struct sock *sk)
2101 {
2102 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2103 }
2104
2105 static inline gfp_t gfp_any(void)
2106 {
2107 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2108 }
2109
2110 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2111 {
2112 return noblock ? 0 : sk->sk_rcvtimeo;
2113 }
2114
2115 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2116 {
2117 return noblock ? 0 : sk->sk_sndtimeo;
2118 }
2119
2120 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2121 {
2122 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2123 }
2124
2125 /* Alas, with timeout socket operations are not restartable.
2126 * Compare this to poll().
2127 */
2128 static inline int sock_intr_errno(long timeo)
2129 {
2130 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2131 }
2132
2133 extern void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2134 struct sk_buff *skb);
2135 extern void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2136 struct sk_buff *skb);
2137
2138 static inline void
2139 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2140 {
2141 ktime_t kt = skb->tstamp;
2142 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2143
2144 /*
2145 * generate control messages if
2146 * - receive time stamping in software requested (SOCK_RCVTSTAMP
2147 * or SOCK_TIMESTAMPING_RX_SOFTWARE)
2148 * - software time stamp available and wanted
2149 * (SOCK_TIMESTAMPING_SOFTWARE)
2150 * - hardware time stamps available and wanted
2151 * (SOCK_TIMESTAMPING_SYS_HARDWARE or
2152 * SOCK_TIMESTAMPING_RAW_HARDWARE)
2153 */
2154 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2155 sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE) ||
2156 (kt.tv64 && sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE)) ||
2157 (hwtstamps->hwtstamp.tv64 &&
2158 sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE)) ||
2159 (hwtstamps->syststamp.tv64 &&
2160 sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE)))
2161 __sock_recv_timestamp(msg, sk, skb);
2162 else
2163 sk->sk_stamp = kt;
2164
2165 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2166 __sock_recv_wifi_status(msg, sk, skb);
2167 }
2168
2169 extern void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2170 struct sk_buff *skb);
2171
2172 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2173 struct sk_buff *skb)
2174 {
2175 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2176 (1UL << SOCK_RCVTSTAMP) | \
2177 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
2178 (1UL << SOCK_TIMESTAMPING_SOFTWARE) | \
2179 (1UL << SOCK_TIMESTAMPING_RAW_HARDWARE) | \
2180 (1UL << SOCK_TIMESTAMPING_SYS_HARDWARE))
2181
2182 if (sk->sk_flags & FLAGS_TS_OR_DROPS)
2183 __sock_recv_ts_and_drops(msg, sk, skb);
2184 else
2185 sk->sk_stamp = skb->tstamp;
2186 }
2187
2188 /**
2189 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2190 * @sk: socket sending this packet
2191 * @tx_flags: filled with instructions for time stamping
2192 *
2193 * Currently only depends on SOCK_TIMESTAMPING* flags.
2194 */
2195 extern void sock_tx_timestamp(struct sock *sk, __u8 *tx_flags);
2196
2197 /**
2198 * sk_eat_skb - Release a skb if it is no longer needed
2199 * @sk: socket to eat this skb from
2200 * @skb: socket buffer to eat
2201 * @copied_early: flag indicating whether DMA operations copied this data early
2202 *
2203 * This routine must be called with interrupts disabled or with the socket
2204 * locked so that the sk_buff queue operation is ok.
2205 */
2206 #ifdef CONFIG_NET_DMA
2207 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2208 {
2209 __skb_unlink(skb, &sk->sk_receive_queue);
2210 if (!copied_early)
2211 __kfree_skb(skb);
2212 else
2213 __skb_queue_tail(&sk->sk_async_wait_queue, skb);
2214 }
2215 #else
2216 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb, bool copied_early)
2217 {
2218 __skb_unlink(skb, &sk->sk_receive_queue);
2219 __kfree_skb(skb);
2220 }
2221 #endif
2222
2223 static inline
2224 struct net *sock_net(const struct sock *sk)
2225 {
2226 return read_pnet(&sk->sk_net);
2227 }
2228
2229 static inline
2230 void sock_net_set(struct sock *sk, struct net *net)
2231 {
2232 write_pnet(&sk->sk_net, net);
2233 }
2234
2235 /*
2236 * Kernel sockets, f.e. rtnl or icmp_socket, are a part of a namespace.
2237 * They should not hold a reference to a namespace in order to allow
2238 * to stop it.
2239 * Sockets after sk_change_net should be released using sk_release_kernel
2240 */
2241 static inline void sk_change_net(struct sock *sk, struct net *net)
2242 {
2243 put_net(sock_net(sk));
2244 sock_net_set(sk, hold_net(net));
2245 }
2246
2247 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2248 {
2249 if (skb->sk) {
2250 struct sock *sk = skb->sk;
2251
2252 skb->destructor = NULL;
2253 skb->sk = NULL;
2254 return sk;
2255 }
2256 return NULL;
2257 }
2258
2259 extern void sock_enable_timestamp(struct sock *sk, int flag);
2260 extern int sock_get_timestamp(struct sock *, struct timeval __user *);
2261 extern int sock_get_timestampns(struct sock *, struct timespec __user *);
2262
2263 bool sk_ns_capable(const struct sock *sk,
2264 struct user_namespace *user_ns, int cap);
2265 bool sk_capable(const struct sock *sk, int cap);
2266 bool sk_net_capable(const struct sock *sk, int cap);
2267
2268 /*
2269 * Enable debug/info messages
2270 */
2271 extern int net_msg_warn;
2272 #define NETDEBUG(fmt, args...) \
2273 do { if (net_msg_warn) printk(fmt,##args); } while (0)
2274
2275 #define LIMIT_NETDEBUG(fmt, args...) \
2276 do { if (net_msg_warn && net_ratelimit()) printk(fmt,##args); } while(0)
2277
2278 extern __u32 sysctl_wmem_max;
2279 extern __u32 sysctl_rmem_max;
2280
2281 extern int sysctl_optmem_max;
2282
2283 extern __u32 sysctl_wmem_default;
2284 extern __u32 sysctl_rmem_default;
2285
2286 #endif /* _SOCK_H */