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/page_counter.h>
58 #include <linux/memcontrol.h>
59 #include <linux/static_key.h>
60 #include <linux/sched.h>
61
62 #include <linux/filter.h>
63 #include <linux/rculist_nulls.h>
64 #include <linux/poll.h>
65
66 #include <linux/atomic.h>
67 #include <net/dst.h>
68 #include <net/checksum.h>
69 #include <net/tcp_states.h>
70 #include <linux/net_tstamp.h>
71
72 struct cgroup;
73 struct cgroup_subsys;
74 #ifdef CONFIG_NET
75 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss);
76 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg);
77 #else
78 static inline
79 int mem_cgroup_sockets_init(struct mem_cgroup *memcg, struct cgroup_subsys *ss)
80 {
81 return 0;
82 }
83 static inline
84 void mem_cgroup_sockets_destroy(struct mem_cgroup *memcg)
85 {
86 }
87 #endif
88 /*
89 * This structure really needs to be cleaned up.
90 * Most of it is for TCP, and not used by any of
91 * the other protocols.
92 */
93
94 /* Define this to get the SOCK_DBG debugging facility. */
95 #define SOCK_DEBUGGING
96 #ifdef SOCK_DEBUGGING
97 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
98 printk(KERN_DEBUG msg); } while (0)
99 #else
100 /* Validate arguments and do nothing */
101 static inline __printf(2, 3)
102 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
103 {
104 }
105 #endif
106
107 /* This is the per-socket lock. The spinlock provides a synchronization
108 * between user contexts and software interrupt processing, whereas the
109 * mini-semaphore synchronizes multiple users amongst themselves.
110 */
111 typedef struct {
112 spinlock_t slock;
113 int owned;
114 wait_queue_head_t wq;
115 /*
116 * We express the mutex-alike socket_lock semantics
117 * to the lock validator by explicitly managing
118 * the slock as a lock variant (in addition to
119 * the slock itself):
120 */
121 #ifdef CONFIG_DEBUG_LOCK_ALLOC
122 struct lockdep_map dep_map;
123 #endif
124 } socket_lock_t;
125
126 struct sock;
127 struct proto;
128 struct net;
129
130 typedef __u32 __bitwise __portpair;
131 typedef __u64 __bitwise __addrpair;
132
133 /**
134 * struct sock_common - minimal network layer representation of sockets
135 * @skc_daddr: Foreign IPv4 addr
136 * @skc_rcv_saddr: Bound local IPv4 addr
137 * @skc_hash: hash value used with various protocol lookup tables
138 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
139 * @skc_dport: placeholder for inet_dport/tw_dport
140 * @skc_num: placeholder for inet_num/tw_num
141 * @skc_family: network address family
142 * @skc_state: Connection state
143 * @skc_reuse: %SO_REUSEADDR setting
144 * @skc_reuseport: %SO_REUSEPORT setting
145 * @skc_bound_dev_if: bound device index if != 0
146 * @skc_bind_node: bind hash linkage for various protocol lookup tables
147 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
148 * @skc_prot: protocol handlers inside a network family
149 * @skc_net: reference to the network namespace of this socket
150 * @skc_node: main hash linkage for various protocol lookup tables
151 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
152 * @skc_tx_queue_mapping: tx queue number for this connection
153 * @skc_refcnt: reference count
154 *
155 * This is the minimal network layer representation of sockets, the header
156 * for struct sock and struct inet_timewait_sock.
157 */
158 struct sock_common {
159 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
160 * address on 64bit arches : cf INET_MATCH()
161 */
162 union {
163 __addrpair skc_addrpair;
164 struct {
165 __be32 skc_daddr;
166 __be32 skc_rcv_saddr;
167 };
168 };
169 union {
170 unsigned int skc_hash;
171 __u16 skc_u16hashes[2];
172 };
173 /* skc_dport && skc_num must be grouped as well */
174 union {
175 __portpair skc_portpair;
176 struct {
177 __be16 skc_dport;
178 __u16 skc_num;
179 };
180 };
181
182 unsigned short skc_family;
183 volatile unsigned char skc_state;
184 unsigned char skc_reuse:4;
185 unsigned char skc_reuseport:1;
186 unsigned char skc_ipv6only:1;
187 unsigned char skc_net_refcnt:1;
188 int skc_bound_dev_if;
189 union {
190 struct hlist_node skc_bind_node;
191 struct hlist_nulls_node skc_portaddr_node;
192 };
193 struct proto *skc_prot;
194 possible_net_t skc_net;
195
196 #if IS_ENABLED(CONFIG_IPV6)
197 struct in6_addr skc_v6_daddr;
198 struct in6_addr skc_v6_rcv_saddr;
199 #endif
200
201 atomic64_t skc_cookie;
202
203 /*
204 * fields between dontcopy_begin/dontcopy_end
205 * are not copied in sock_copy()
206 */
207 /* private: */
208 int skc_dontcopy_begin[0];
209 /* public: */
210 union {
211 struct hlist_node skc_node;
212 struct hlist_nulls_node skc_nulls_node;
213 };
214 int skc_tx_queue_mapping;
215 atomic_t skc_refcnt;
216 /* private: */
217 int skc_dontcopy_end[0];
218 /* public: */
219 };
220
221 struct cg_proto;
222 /**
223 * struct sock - network layer representation of sockets
224 * @__sk_common: shared layout with inet_timewait_sock
225 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
226 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
227 * @sk_lock: synchronizer
228 * @sk_rcvbuf: size of receive buffer in bytes
229 * @sk_wq: sock wait queue and async head
230 * @sk_rx_dst: receive input route used by early demux
231 * @sk_dst_cache: destination cache
232 * @sk_dst_lock: destination cache lock
233 * @sk_policy: flow policy
234 * @sk_receive_queue: incoming packets
235 * @sk_wmem_alloc: transmit queue bytes committed
236 * @sk_write_queue: Packet sending queue
237 * @sk_omem_alloc: "o" is "option" or "other"
238 * @sk_wmem_queued: persistent queue size
239 * @sk_forward_alloc: space allocated forward
240 * @sk_napi_id: id of the last napi context to receive data for sk
241 * @sk_ll_usec: usecs to busypoll when there is no data
242 * @sk_allocation: allocation mode
243 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
244 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
245 * @sk_sndbuf: size of send buffer in bytes
246 * @sk_flags: %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
247 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
248 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
249 * @sk_no_check_rx: allow zero checksum in RX packets
250 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
251 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
252 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
253 * @sk_gso_max_size: Maximum GSO segment size to build
254 * @sk_gso_max_segs: Maximum number of GSO segments
255 * @sk_lingertime: %SO_LINGER l_linger setting
256 * @sk_backlog: always used with the per-socket spinlock held
257 * @sk_callback_lock: used with the callbacks in the end of this struct
258 * @sk_error_queue: rarely used
259 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
260 * IPV6_ADDRFORM for instance)
261 * @sk_err: last error
262 * @sk_err_soft: errors that don't cause failure but are the cause of a
263 * persistent failure not just 'timed out'
264 * @sk_drops: raw/udp drops counter
265 * @sk_ack_backlog: current listen backlog
266 * @sk_max_ack_backlog: listen backlog set in listen()
267 * @sk_priority: %SO_PRIORITY setting
268 * @sk_cgrp_prioidx: socket group's priority map index
269 * @sk_type: socket type (%SOCK_STREAM, etc)
270 * @sk_protocol: which protocol this socket belongs in this network family
271 * @sk_peer_pid: &struct pid for this socket's peer
272 * @sk_peer_cred: %SO_PEERCRED setting
273 * @sk_rcvlowat: %SO_RCVLOWAT setting
274 * @sk_rcvtimeo: %SO_RCVTIMEO setting
275 * @sk_sndtimeo: %SO_SNDTIMEO setting
276 * @sk_rxhash: flow hash received from netif layer
277 * @sk_incoming_cpu: record cpu processing incoming packets
278 * @sk_txhash: computed flow hash for use on transmit
279 * @sk_filter: socket filtering instructions
280 * @sk_timer: sock cleanup timer
281 * @sk_stamp: time stamp of last packet received
282 * @sk_tsflags: SO_TIMESTAMPING socket options
283 * @sk_tskey: counter to disambiguate concurrent tstamp requests
284 * @sk_socket: Identd and reporting IO signals
285 * @sk_user_data: RPC layer private data
286 * @sk_frag: cached page frag
287 * @sk_peek_off: current peek_offset value
288 * @sk_send_head: front of stuff to transmit
289 * @sk_security: used by security modules
290 * @sk_mark: generic packet mark
291 * @sk_classid: this socket's cgroup classid
292 * @sk_cgrp: this socket's cgroup-specific proto data
293 * @sk_write_pending: a write to stream socket waits to start
294 * @sk_state_change: callback to indicate change in the state of the sock
295 * @sk_data_ready: callback to indicate there is data to be processed
296 * @sk_write_space: callback to indicate there is bf sending space available
297 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
298 * @sk_backlog_rcv: callback to process the backlog
299 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
300 */
301 struct sock {
302 /*
303 * Now struct inet_timewait_sock also uses sock_common, so please just
304 * don't add nothing before this first member (__sk_common) --acme
305 */
306 struct sock_common __sk_common;
307 #define sk_node __sk_common.skc_node
308 #define sk_nulls_node __sk_common.skc_nulls_node
309 #define sk_refcnt __sk_common.skc_refcnt
310 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
311
312 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
313 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
314 #define sk_hash __sk_common.skc_hash
315 #define sk_portpair __sk_common.skc_portpair
316 #define sk_num __sk_common.skc_num
317 #define sk_dport __sk_common.skc_dport
318 #define sk_addrpair __sk_common.skc_addrpair
319 #define sk_daddr __sk_common.skc_daddr
320 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
321 #define sk_family __sk_common.skc_family
322 #define sk_state __sk_common.skc_state
323 #define sk_reuse __sk_common.skc_reuse
324 #define sk_reuseport __sk_common.skc_reuseport
325 #define sk_ipv6only __sk_common.skc_ipv6only
326 #define sk_net_refcnt __sk_common.skc_net_refcnt
327 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
328 #define sk_bind_node __sk_common.skc_bind_node
329 #define sk_prot __sk_common.skc_prot
330 #define sk_net __sk_common.skc_net
331 #define sk_v6_daddr __sk_common.skc_v6_daddr
332 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
333 #define sk_cookie __sk_common.skc_cookie
334
335 socket_lock_t sk_lock;
336 struct sk_buff_head sk_receive_queue;
337 /*
338 * The backlog queue is special, it is always used with
339 * the per-socket spinlock held and requires low latency
340 * access. Therefore we special case it's implementation.
341 * Note : rmem_alloc is in this structure to fill a hole
342 * on 64bit arches, not because its logically part of
343 * backlog.
344 */
345 struct {
346 atomic_t rmem_alloc;
347 int len;
348 struct sk_buff *head;
349 struct sk_buff *tail;
350 } sk_backlog;
351 #define sk_rmem_alloc sk_backlog.rmem_alloc
352 int sk_forward_alloc;
353 #ifdef CONFIG_RPS
354 __u32 sk_rxhash;
355 #endif
356 u16 sk_incoming_cpu;
357 /* 16bit hole
358 * Warned : sk_incoming_cpu can be set from softirq,
359 * Do not use this hole without fully understanding possible issues.
360 */
361
362 __u32 sk_txhash;
363 #ifdef CONFIG_NET_RX_BUSY_POLL
364 unsigned int sk_napi_id;
365 unsigned int sk_ll_usec;
366 #endif
367 atomic_t sk_drops;
368 int sk_rcvbuf;
369
370 struct sk_filter __rcu *sk_filter;
371 struct socket_wq __rcu *sk_wq;
372
373 #ifdef CONFIG_XFRM
374 struct xfrm_policy *sk_policy[2];
375 #endif
376 unsigned long sk_flags;
377 struct dst_entry *sk_rx_dst;
378 struct dst_entry __rcu *sk_dst_cache;
379 spinlock_t sk_dst_lock;
380 atomic_t sk_wmem_alloc;
381 atomic_t sk_omem_alloc;
382 int sk_sndbuf;
383 struct sk_buff_head sk_write_queue;
384 kmemcheck_bitfield_begin(flags);
385 unsigned int sk_shutdown : 2,
386 sk_no_check_tx : 1,
387 sk_no_check_rx : 1,
388 sk_userlocks : 4,
389 sk_protocol : 8,
390 #define SK_PROTOCOL_MAX U8_MAX
391 sk_type : 16;
392 kmemcheck_bitfield_end(flags);
393 int sk_wmem_queued;
394 gfp_t sk_allocation;
395 u32 sk_pacing_rate; /* bytes per second */
396 u32 sk_max_pacing_rate;
397 netdev_features_t sk_route_caps;
398 netdev_features_t sk_route_nocaps;
399 int sk_gso_type;
400 unsigned int sk_gso_max_size;
401 u16 sk_gso_max_segs;
402 int sk_rcvlowat;
403 unsigned long sk_lingertime;
404 struct sk_buff_head sk_error_queue;
405 struct proto *sk_prot_creator;
406 rwlock_t sk_callback_lock;
407 int sk_err,
408 sk_err_soft;
409 u32 sk_ack_backlog;
410 u32 sk_max_ack_backlog;
411 __u32 sk_priority;
412 #if IS_ENABLED(CONFIG_CGROUP_NET_PRIO)
413 __u32 sk_cgrp_prioidx;
414 #endif
415 struct pid *sk_peer_pid;
416 const struct cred *sk_peer_cred;
417 long sk_rcvtimeo;
418 long sk_sndtimeo;
419 struct timer_list sk_timer;
420 ktime_t sk_stamp;
421 u16 sk_tsflags;
422 u32 sk_tskey;
423 struct socket *sk_socket;
424 void *sk_user_data;
425 struct page_frag sk_frag;
426 struct sk_buff *sk_send_head;
427 __s32 sk_peek_off;
428 int sk_write_pending;
429 #ifdef CONFIG_SECURITY
430 void *sk_security;
431 #endif
432 __u32 sk_mark;
433 #ifdef CONFIG_CGROUP_NET_CLASSID
434 u32 sk_classid;
435 #endif
436 struct cg_proto *sk_cgrp;
437 void (*sk_state_change)(struct sock *sk);
438 void (*sk_data_ready)(struct sock *sk);
439 void (*sk_write_space)(struct sock *sk);
440 void (*sk_error_report)(struct sock *sk);
441 int (*sk_backlog_rcv)(struct sock *sk,
442 struct sk_buff *skb);
443 void (*sk_destruct)(struct sock *sk);
444 };
445
446 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
447
448 #define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
449 #define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
450
451 /*
452 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
453 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
454 * on a socket means that the socket will reuse everybody else's port
455 * without looking at the other's sk_reuse value.
456 */
457
458 #define SK_NO_REUSE 0
459 #define SK_CAN_REUSE 1
460 #define SK_FORCE_REUSE 2
461
462 static inline int sk_peek_offset(struct sock *sk, int flags)
463 {
464 if ((flags & MSG_PEEK) && (sk->sk_peek_off >= 0))
465 return sk->sk_peek_off;
466 else
467 return 0;
468 }
469
470 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
471 {
472 if (sk->sk_peek_off >= 0) {
473 if (sk->sk_peek_off >= val)
474 sk->sk_peek_off -= val;
475 else
476 sk->sk_peek_off = 0;
477 }
478 }
479
480 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
481 {
482 if (sk->sk_peek_off >= 0)
483 sk->sk_peek_off += val;
484 }
485
486 /*
487 * Hashed lists helper routines
488 */
489 static inline struct sock *sk_entry(const struct hlist_node *node)
490 {
491 return hlist_entry(node, struct sock, sk_node);
492 }
493
494 static inline struct sock *__sk_head(const struct hlist_head *head)
495 {
496 return hlist_entry(head->first, struct sock, sk_node);
497 }
498
499 static inline struct sock *sk_head(const struct hlist_head *head)
500 {
501 return hlist_empty(head) ? NULL : __sk_head(head);
502 }
503
504 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
505 {
506 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
507 }
508
509 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
510 {
511 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
512 }
513
514 static inline struct sock *sk_next(const struct sock *sk)
515 {
516 return sk->sk_node.next ?
517 hlist_entry(sk->sk_node.next, struct sock, sk_node) : NULL;
518 }
519
520 static inline struct sock *sk_nulls_next(const struct sock *sk)
521 {
522 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
523 hlist_nulls_entry(sk->sk_nulls_node.next,
524 struct sock, sk_nulls_node) :
525 NULL;
526 }
527
528 static inline bool sk_unhashed(const struct sock *sk)
529 {
530 return hlist_unhashed(&sk->sk_node);
531 }
532
533 static inline bool sk_hashed(const struct sock *sk)
534 {
535 return !sk_unhashed(sk);
536 }
537
538 static inline void sk_node_init(struct hlist_node *node)
539 {
540 node->pprev = NULL;
541 }
542
543 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
544 {
545 node->pprev = NULL;
546 }
547
548 static inline void __sk_del_node(struct sock *sk)
549 {
550 __hlist_del(&sk->sk_node);
551 }
552
553 /* NB: equivalent to hlist_del_init_rcu */
554 static inline bool __sk_del_node_init(struct sock *sk)
555 {
556 if (sk_hashed(sk)) {
557 __sk_del_node(sk);
558 sk_node_init(&sk->sk_node);
559 return true;
560 }
561 return false;
562 }
563
564 /* Grab socket reference count. This operation is valid only
565 when sk is ALREADY grabbed f.e. it is found in hash table
566 or a list and the lookup is made under lock preventing hash table
567 modifications.
568 */
569
570 static inline void sock_hold(struct sock *sk)
571 {
572 atomic_inc(&sk->sk_refcnt);
573 }
574
575 /* Ungrab socket in the context, which assumes that socket refcnt
576 cannot hit zero, f.e. it is true in context of any socketcall.
577 */
578 static inline void __sock_put(struct sock *sk)
579 {
580 atomic_dec(&sk->sk_refcnt);
581 }
582
583 static inline bool sk_del_node_init(struct sock *sk)
584 {
585 bool rc = __sk_del_node_init(sk);
586
587 if (rc) {
588 /* paranoid for a while -acme */
589 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
590 __sock_put(sk);
591 }
592 return rc;
593 }
594 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
595
596 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
597 {
598 if (sk_hashed(sk)) {
599 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
600 return true;
601 }
602 return false;
603 }
604
605 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
606 {
607 bool rc = __sk_nulls_del_node_init_rcu(sk);
608
609 if (rc) {
610 /* paranoid for a while -acme */
611 WARN_ON(atomic_read(&sk->sk_refcnt) == 1);
612 __sock_put(sk);
613 }
614 return rc;
615 }
616
617 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
618 {
619 hlist_add_head(&sk->sk_node, list);
620 }
621
622 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
623 {
624 sock_hold(sk);
625 __sk_add_node(sk, list);
626 }
627
628 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
629 {
630 sock_hold(sk);
631 hlist_add_head_rcu(&sk->sk_node, list);
632 }
633
634 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
635 {
636 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
637 }
638
639 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
640 {
641 sock_hold(sk);
642 __sk_nulls_add_node_rcu(sk, list);
643 }
644
645 static inline void __sk_del_bind_node(struct sock *sk)
646 {
647 __hlist_del(&sk->sk_bind_node);
648 }
649
650 static inline void sk_add_bind_node(struct sock *sk,
651 struct hlist_head *list)
652 {
653 hlist_add_head(&sk->sk_bind_node, list);
654 }
655
656 #define sk_for_each(__sk, list) \
657 hlist_for_each_entry(__sk, list, sk_node)
658 #define sk_for_each_rcu(__sk, list) \
659 hlist_for_each_entry_rcu(__sk, list, sk_node)
660 #define sk_nulls_for_each(__sk, node, list) \
661 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
662 #define sk_nulls_for_each_rcu(__sk, node, list) \
663 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
664 #define sk_for_each_from(__sk) \
665 hlist_for_each_entry_from(__sk, sk_node)
666 #define sk_nulls_for_each_from(__sk, node) \
667 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
668 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
669 #define sk_for_each_safe(__sk, tmp, list) \
670 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
671 #define sk_for_each_bound(__sk, list) \
672 hlist_for_each_entry(__sk, list, sk_bind_node)
673
674 /**
675 * sk_nulls_for_each_entry_offset - iterate over a list at a given struct offset
676 * @tpos: the type * to use as a loop cursor.
677 * @pos: the &struct hlist_node to use as a loop cursor.
678 * @head: the head for your list.
679 * @offset: offset of hlist_node within the struct.
680 *
681 */
682 #define sk_nulls_for_each_entry_offset(tpos, pos, head, offset) \
683 for (pos = (head)->first; \
684 (!is_a_nulls(pos)) && \
685 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
686 pos = pos->next)
687
688 static inline struct user_namespace *sk_user_ns(struct sock *sk)
689 {
690 /* Careful only use this in a context where these parameters
691 * can not change and must all be valid, such as recvmsg from
692 * userspace.
693 */
694 return sk->sk_socket->file->f_cred->user_ns;
695 }
696
697 /* Sock flags */
698 enum sock_flags {
699 SOCK_DEAD,
700 SOCK_DONE,
701 SOCK_URGINLINE,
702 SOCK_KEEPOPEN,
703 SOCK_LINGER,
704 SOCK_DESTROY,
705 SOCK_BROADCAST,
706 SOCK_TIMESTAMP,
707 SOCK_ZAPPED,
708 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
709 SOCK_DBG, /* %SO_DEBUG setting */
710 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
711 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
712 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
713 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
714 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
715 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
716 SOCK_FASYNC, /* fasync() active */
717 SOCK_RXQ_OVFL,
718 SOCK_ZEROCOPY, /* buffers from userspace */
719 SOCK_WIFI_STATUS, /* push wifi status to userspace */
720 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
721 * Will use last 4 bytes of packet sent from
722 * user-space instead.
723 */
724 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
725 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
726 };
727
728 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
729
730 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
731 {
732 nsk->sk_flags = osk->sk_flags;
733 }
734
735 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
736 {
737 __set_bit(flag, &sk->sk_flags);
738 }
739
740 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
741 {
742 __clear_bit(flag, &sk->sk_flags);
743 }
744
745 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
746 {
747 return test_bit(flag, &sk->sk_flags);
748 }
749
750 #ifdef CONFIG_NET
751 extern struct static_key memalloc_socks;
752 static inline int sk_memalloc_socks(void)
753 {
754 return static_key_false(&memalloc_socks);
755 }
756 #else
757
758 static inline int sk_memalloc_socks(void)
759 {
760 return 0;
761 }
762
763 #endif
764
765 static inline gfp_t sk_gfp_atomic(struct sock *sk, gfp_t gfp_mask)
766 {
767 return GFP_ATOMIC | (sk->sk_allocation & __GFP_MEMALLOC);
768 }
769
770 static inline void sk_acceptq_removed(struct sock *sk)
771 {
772 sk->sk_ack_backlog--;
773 }
774
775 static inline void sk_acceptq_added(struct sock *sk)
776 {
777 sk->sk_ack_backlog++;
778 }
779
780 static inline bool sk_acceptq_is_full(const struct sock *sk)
781 {
782 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
783 }
784
785 /*
786 * Compute minimal free write space needed to queue new packets.
787 */
788 static inline int sk_stream_min_wspace(const struct sock *sk)
789 {
790 return sk->sk_wmem_queued >> 1;
791 }
792
793 static inline int sk_stream_wspace(const struct sock *sk)
794 {
795 return sk->sk_sndbuf - sk->sk_wmem_queued;
796 }
797
798 void sk_stream_write_space(struct sock *sk);
799
800 /* OOB backlog add */
801 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
802 {
803 /* dont let skb dst not refcounted, we are going to leave rcu lock */
804 skb_dst_force_safe(skb);
805
806 if (!sk->sk_backlog.tail)
807 sk->sk_backlog.head = skb;
808 else
809 sk->sk_backlog.tail->next = skb;
810
811 sk->sk_backlog.tail = skb;
812 skb->next = NULL;
813 }
814
815 /*
816 * Take into account size of receive queue and backlog queue
817 * Do not take into account this skb truesize,
818 * to allow even a single big packet to come.
819 */
820 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
821 {
822 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
823
824 return qsize > limit;
825 }
826
827 /* The per-socket spinlock must be held here. */
828 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
829 unsigned int limit)
830 {
831 if (sk_rcvqueues_full(sk, limit))
832 return -ENOBUFS;
833
834 /*
835 * If the skb was allocated from pfmemalloc reserves, only
836 * allow SOCK_MEMALLOC sockets to use it as this socket is
837 * helping free memory
838 */
839 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
840 return -ENOMEM;
841
842 __sk_add_backlog(sk, skb);
843 sk->sk_backlog.len += skb->truesize;
844 return 0;
845 }
846
847 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
848
849 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
850 {
851 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
852 return __sk_backlog_rcv(sk, skb);
853
854 return sk->sk_backlog_rcv(sk, skb);
855 }
856
857 static inline void sk_incoming_cpu_update(struct sock *sk)
858 {
859 sk->sk_incoming_cpu = raw_smp_processor_id();
860 }
861
862 static inline void sock_rps_record_flow_hash(__u32 hash)
863 {
864 #ifdef CONFIG_RPS
865 struct rps_sock_flow_table *sock_flow_table;
866
867 rcu_read_lock();
868 sock_flow_table = rcu_dereference(rps_sock_flow_table);
869 rps_record_sock_flow(sock_flow_table, hash);
870 rcu_read_unlock();
871 #endif
872 }
873
874 static inline void sock_rps_record_flow(const struct sock *sk)
875 {
876 #ifdef CONFIG_RPS
877 sock_rps_record_flow_hash(sk->sk_rxhash);
878 #endif
879 }
880
881 static inline void sock_rps_save_rxhash(struct sock *sk,
882 const struct sk_buff *skb)
883 {
884 #ifdef CONFIG_RPS
885 if (unlikely(sk->sk_rxhash != skb->hash))
886 sk->sk_rxhash = skb->hash;
887 #endif
888 }
889
890 static inline void sock_rps_reset_rxhash(struct sock *sk)
891 {
892 #ifdef CONFIG_RPS
893 sk->sk_rxhash = 0;
894 #endif
895 }
896
897 #define sk_wait_event(__sk, __timeo, __condition) \
898 ({ int __rc; \
899 release_sock(__sk); \
900 __rc = __condition; \
901 if (!__rc) { \
902 *(__timeo) = schedule_timeout(*(__timeo)); \
903 } \
904 sched_annotate_sleep(); \
905 lock_sock(__sk); \
906 __rc = __condition; \
907 __rc; \
908 })
909
910 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
911 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
912 void sk_stream_wait_close(struct sock *sk, long timeo_p);
913 int sk_stream_error(struct sock *sk, int flags, int err);
914 void sk_stream_kill_queues(struct sock *sk);
915 void sk_set_memalloc(struct sock *sk);
916 void sk_clear_memalloc(struct sock *sk);
917
918 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
919
920 struct request_sock_ops;
921 struct timewait_sock_ops;
922 struct inet_hashinfo;
923 struct raw_hashinfo;
924 struct module;
925
926 /*
927 * caches using SLAB_DESTROY_BY_RCU should let .next pointer from nulls nodes
928 * un-modified. Special care is taken when initializing object to zero.
929 */
930 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
931 {
932 if (offsetof(struct sock, sk_node.next) != 0)
933 memset(sk, 0, offsetof(struct sock, sk_node.next));
934 memset(&sk->sk_node.pprev, 0,
935 size - offsetof(struct sock, sk_node.pprev));
936 }
937
938 /* Networking protocol blocks we attach to sockets.
939 * socket layer -> transport layer interface
940 */
941 struct proto {
942 void (*close)(struct sock *sk,
943 long timeout);
944 int (*connect)(struct sock *sk,
945 struct sockaddr *uaddr,
946 int addr_len);
947 int (*disconnect)(struct sock *sk, int flags);
948
949 struct sock * (*accept)(struct sock *sk, int flags, int *err);
950
951 int (*ioctl)(struct sock *sk, int cmd,
952 unsigned long arg);
953 int (*init)(struct sock *sk);
954 void (*destroy)(struct sock *sk);
955 void (*shutdown)(struct sock *sk, int how);
956 int (*setsockopt)(struct sock *sk, int level,
957 int optname, char __user *optval,
958 unsigned int optlen);
959 int (*getsockopt)(struct sock *sk, int level,
960 int optname, char __user *optval,
961 int __user *option);
962 #ifdef CONFIG_COMPAT
963 int (*compat_setsockopt)(struct sock *sk,
964 int level,
965 int optname, char __user *optval,
966 unsigned int optlen);
967 int (*compat_getsockopt)(struct sock *sk,
968 int level,
969 int optname, char __user *optval,
970 int __user *option);
971 int (*compat_ioctl)(struct sock *sk,
972 unsigned int cmd, unsigned long arg);
973 #endif
974 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
975 size_t len);
976 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
977 size_t len, int noblock, int flags,
978 int *addr_len);
979 int (*sendpage)(struct sock *sk, struct page *page,
980 int offset, size_t size, int flags);
981 int (*bind)(struct sock *sk,
982 struct sockaddr *uaddr, int addr_len);
983
984 int (*backlog_rcv) (struct sock *sk,
985 struct sk_buff *skb);
986
987 void (*release_cb)(struct sock *sk);
988
989 /* Keeping track of sk's, looking them up, and port selection methods. */
990 void (*hash)(struct sock *sk);
991 void (*unhash)(struct sock *sk);
992 void (*rehash)(struct sock *sk);
993 int (*get_port)(struct sock *sk, unsigned short snum);
994 void (*clear_sk)(struct sock *sk, int size);
995
996 /* Keeping track of sockets in use */
997 #ifdef CONFIG_PROC_FS
998 unsigned int inuse_idx;
999 #endif
1000
1001 bool (*stream_memory_free)(const struct sock *sk);
1002 /* Memory pressure */
1003 void (*enter_memory_pressure)(struct sock *sk);
1004 atomic_long_t *memory_allocated; /* Current allocated memory. */
1005 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1006 /*
1007 * Pressure flag: try to collapse.
1008 * Technical note: it is used by multiple contexts non atomically.
1009 * All the __sk_mem_schedule() is of this nature: accounting
1010 * is strict, actions are advisory and have some latency.
1011 */
1012 int *memory_pressure;
1013 long *sysctl_mem;
1014 int *sysctl_wmem;
1015 int *sysctl_rmem;
1016 int max_header;
1017 bool no_autobind;
1018
1019 struct kmem_cache *slab;
1020 unsigned int obj_size;
1021 int slab_flags;
1022
1023 struct percpu_counter *orphan_count;
1024
1025 struct request_sock_ops *rsk_prot;
1026 struct timewait_sock_ops *twsk_prot;
1027
1028 union {
1029 struct inet_hashinfo *hashinfo;
1030 struct udp_table *udp_table;
1031 struct raw_hashinfo *raw_hash;
1032 } h;
1033
1034 struct module *owner;
1035
1036 char name[32];
1037
1038 struct list_head node;
1039 #ifdef SOCK_REFCNT_DEBUG
1040 atomic_t socks;
1041 #endif
1042 #ifdef CONFIG_MEMCG_KMEM
1043 /*
1044 * cgroup specific init/deinit functions. Called once for all
1045 * protocols that implement it, from cgroups populate function.
1046 * This function has to setup any files the protocol want to
1047 * appear in the kmem cgroup filesystem.
1048 */
1049 int (*init_cgroup)(struct mem_cgroup *memcg,
1050 struct cgroup_subsys *ss);
1051 void (*destroy_cgroup)(struct mem_cgroup *memcg);
1052 struct cg_proto *(*proto_cgroup)(struct mem_cgroup *memcg);
1053 #endif
1054 };
1055
1056 int proto_register(struct proto *prot, int alloc_slab);
1057 void proto_unregister(struct proto *prot);
1058
1059 #ifdef SOCK_REFCNT_DEBUG
1060 static inline void sk_refcnt_debug_inc(struct sock *sk)
1061 {
1062 atomic_inc(&sk->sk_prot->socks);
1063 }
1064
1065 static inline void sk_refcnt_debug_dec(struct sock *sk)
1066 {
1067 atomic_dec(&sk->sk_prot->socks);
1068 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1069 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1070 }
1071
1072 static inline void sk_refcnt_debug_release(const struct sock *sk)
1073 {
1074 if (atomic_read(&sk->sk_refcnt) != 1)
1075 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1076 sk->sk_prot->name, sk, atomic_read(&sk->sk_refcnt));
1077 }
1078 #else /* SOCK_REFCNT_DEBUG */
1079 #define sk_refcnt_debug_inc(sk) do { } while (0)
1080 #define sk_refcnt_debug_dec(sk) do { } while (0)
1081 #define sk_refcnt_debug_release(sk) do { } while (0)
1082 #endif /* SOCK_REFCNT_DEBUG */
1083
1084 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_NET)
1085 extern struct static_key memcg_socket_limit_enabled;
1086 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1087 struct cg_proto *cg_proto)
1088 {
1089 return proto->proto_cgroup(parent_mem_cgroup(cg_proto->memcg));
1090 }
1091 #define mem_cgroup_sockets_enabled static_key_false(&memcg_socket_limit_enabled)
1092 #else
1093 #define mem_cgroup_sockets_enabled 0
1094 static inline struct cg_proto *parent_cg_proto(struct proto *proto,
1095 struct cg_proto *cg_proto)
1096 {
1097 return NULL;
1098 }
1099 #endif
1100
1101 static inline bool sk_stream_memory_free(const struct sock *sk)
1102 {
1103 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1104 return false;
1105
1106 return sk->sk_prot->stream_memory_free ?
1107 sk->sk_prot->stream_memory_free(sk) : true;
1108 }
1109
1110 static inline bool sk_stream_is_writeable(const struct sock *sk)
1111 {
1112 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1113 sk_stream_memory_free(sk);
1114 }
1115
1116
1117 static inline bool sk_has_memory_pressure(const struct sock *sk)
1118 {
1119 return sk->sk_prot->memory_pressure != NULL;
1120 }
1121
1122 static inline bool sk_under_memory_pressure(const struct sock *sk)
1123 {
1124 if (!sk->sk_prot->memory_pressure)
1125 return false;
1126
1127 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1128 return !!sk->sk_cgrp->memory_pressure;
1129
1130 return !!*sk->sk_prot->memory_pressure;
1131 }
1132
1133 static inline void sk_leave_memory_pressure(struct sock *sk)
1134 {
1135 int *memory_pressure = sk->sk_prot->memory_pressure;
1136
1137 if (!memory_pressure)
1138 return;
1139
1140 if (*memory_pressure)
1141 *memory_pressure = 0;
1142
1143 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1144 struct cg_proto *cg_proto = sk->sk_cgrp;
1145 struct proto *prot = sk->sk_prot;
1146
1147 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1148 cg_proto->memory_pressure = 0;
1149 }
1150
1151 }
1152
1153 static inline void sk_enter_memory_pressure(struct sock *sk)
1154 {
1155 if (!sk->sk_prot->enter_memory_pressure)
1156 return;
1157
1158 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1159 struct cg_proto *cg_proto = sk->sk_cgrp;
1160 struct proto *prot = sk->sk_prot;
1161
1162 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1163 cg_proto->memory_pressure = 1;
1164 }
1165
1166 sk->sk_prot->enter_memory_pressure(sk);
1167 }
1168
1169 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1170 {
1171 long *prot = sk->sk_prot->sysctl_mem;
1172 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1173 prot = sk->sk_cgrp->sysctl_mem;
1174 return prot[index];
1175 }
1176
1177 static inline void memcg_memory_allocated_add(struct cg_proto *prot,
1178 unsigned long amt,
1179 int *parent_status)
1180 {
1181 page_counter_charge(&prot->memory_allocated, amt);
1182
1183 if (page_counter_read(&prot->memory_allocated) >
1184 prot->memory_allocated.limit)
1185 *parent_status = OVER_LIMIT;
1186 }
1187
1188 static inline void memcg_memory_allocated_sub(struct cg_proto *prot,
1189 unsigned long amt)
1190 {
1191 page_counter_uncharge(&prot->memory_allocated, amt);
1192 }
1193
1194 static inline long
1195 sk_memory_allocated(const struct sock *sk)
1196 {
1197 struct proto *prot = sk->sk_prot;
1198
1199 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1200 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1201
1202 return atomic_long_read(prot->memory_allocated);
1203 }
1204
1205 static inline long
1206 sk_memory_allocated_add(struct sock *sk, int amt, int *parent_status)
1207 {
1208 struct proto *prot = sk->sk_prot;
1209
1210 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1211 memcg_memory_allocated_add(sk->sk_cgrp, amt, parent_status);
1212 /* update the root cgroup regardless */
1213 atomic_long_add_return(amt, prot->memory_allocated);
1214 return page_counter_read(&sk->sk_cgrp->memory_allocated);
1215 }
1216
1217 return atomic_long_add_return(amt, prot->memory_allocated);
1218 }
1219
1220 static inline void
1221 sk_memory_allocated_sub(struct sock *sk, int amt)
1222 {
1223 struct proto *prot = sk->sk_prot;
1224
1225 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1226 memcg_memory_allocated_sub(sk->sk_cgrp, amt);
1227
1228 atomic_long_sub(amt, prot->memory_allocated);
1229 }
1230
1231 static inline void sk_sockets_allocated_dec(struct sock *sk)
1232 {
1233 struct proto *prot = sk->sk_prot;
1234
1235 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1236 struct cg_proto *cg_proto = sk->sk_cgrp;
1237
1238 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1239 percpu_counter_dec(&cg_proto->sockets_allocated);
1240 }
1241
1242 percpu_counter_dec(prot->sockets_allocated);
1243 }
1244
1245 static inline void sk_sockets_allocated_inc(struct sock *sk)
1246 {
1247 struct proto *prot = sk->sk_prot;
1248
1249 if (mem_cgroup_sockets_enabled && sk->sk_cgrp) {
1250 struct cg_proto *cg_proto = sk->sk_cgrp;
1251
1252 for (; cg_proto; cg_proto = parent_cg_proto(prot, cg_proto))
1253 percpu_counter_inc(&cg_proto->sockets_allocated);
1254 }
1255
1256 percpu_counter_inc(prot->sockets_allocated);
1257 }
1258
1259 static inline int
1260 sk_sockets_allocated_read_positive(struct sock *sk)
1261 {
1262 struct proto *prot = sk->sk_prot;
1263
1264 if (mem_cgroup_sockets_enabled && sk->sk_cgrp)
1265 return percpu_counter_read_positive(&sk->sk_cgrp->sockets_allocated);
1266
1267 return percpu_counter_read_positive(prot->sockets_allocated);
1268 }
1269
1270 static inline int
1271 proto_sockets_allocated_sum_positive(struct proto *prot)
1272 {
1273 return percpu_counter_sum_positive(prot->sockets_allocated);
1274 }
1275
1276 static inline long
1277 proto_memory_allocated(struct proto *prot)
1278 {
1279 return atomic_long_read(prot->memory_allocated);
1280 }
1281
1282 static inline bool
1283 proto_memory_pressure(struct proto *prot)
1284 {
1285 if (!prot->memory_pressure)
1286 return false;
1287 return !!*prot->memory_pressure;
1288 }
1289
1290
1291 #ifdef CONFIG_PROC_FS
1292 /* Called with local bh disabled */
1293 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1294 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1295 #else
1296 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1297 int inc)
1298 {
1299 }
1300 #endif
1301
1302
1303 /* With per-bucket locks this operation is not-atomic, so that
1304 * this version is not worse.
1305 */
1306 static inline void __sk_prot_rehash(struct sock *sk)
1307 {
1308 sk->sk_prot->unhash(sk);
1309 sk->sk_prot->hash(sk);
1310 }
1311
1312 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size);
1313
1314 /* About 10 seconds */
1315 #define SOCK_DESTROY_TIME (10*HZ)
1316
1317 /* Sockets 0-1023 can't be bound to unless you are superuser */
1318 #define PROT_SOCK 1024
1319
1320 #define SHUTDOWN_MASK 3
1321 #define RCV_SHUTDOWN 1
1322 #define SEND_SHUTDOWN 2
1323
1324 #define SOCK_SNDBUF_LOCK 1
1325 #define SOCK_RCVBUF_LOCK 2
1326 #define SOCK_BINDADDR_LOCK 4
1327 #define SOCK_BINDPORT_LOCK 8
1328
1329 struct socket_alloc {
1330 struct socket socket;
1331 struct inode vfs_inode;
1332 };
1333
1334 static inline struct socket *SOCKET_I(struct inode *inode)
1335 {
1336 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1337 }
1338
1339 static inline struct inode *SOCK_INODE(struct socket *socket)
1340 {
1341 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1342 }
1343
1344 /*
1345 * Functions for memory accounting
1346 */
1347 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1348 void __sk_mem_reclaim(struct sock *sk, int amount);
1349
1350 #define SK_MEM_QUANTUM ((int)PAGE_SIZE)
1351 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1352 #define SK_MEM_SEND 0
1353 #define SK_MEM_RECV 1
1354
1355 static inline int sk_mem_pages(int amt)
1356 {
1357 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1358 }
1359
1360 static inline bool sk_has_account(struct sock *sk)
1361 {
1362 /* return true if protocol supports memory accounting */
1363 return !!sk->sk_prot->memory_allocated;
1364 }
1365
1366 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1367 {
1368 if (!sk_has_account(sk))
1369 return true;
1370 return size <= sk->sk_forward_alloc ||
1371 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1372 }
1373
1374 static inline bool
1375 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1376 {
1377 if (!sk_has_account(sk))
1378 return true;
1379 return size<= sk->sk_forward_alloc ||
1380 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1381 skb_pfmemalloc(skb);
1382 }
1383
1384 static inline void sk_mem_reclaim(struct sock *sk)
1385 {
1386 if (!sk_has_account(sk))
1387 return;
1388 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1389 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1390 }
1391
1392 static inline void sk_mem_reclaim_partial(struct sock *sk)
1393 {
1394 if (!sk_has_account(sk))
1395 return;
1396 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1397 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1398 }
1399
1400 static inline void sk_mem_charge(struct sock *sk, int size)
1401 {
1402 if (!sk_has_account(sk))
1403 return;
1404 sk->sk_forward_alloc -= size;
1405 }
1406
1407 static inline void sk_mem_uncharge(struct sock *sk, int size)
1408 {
1409 if (!sk_has_account(sk))
1410 return;
1411 sk->sk_forward_alloc += size;
1412 }
1413
1414 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1415 {
1416 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1417 sk->sk_wmem_queued -= skb->truesize;
1418 sk_mem_uncharge(sk, skb->truesize);
1419 __kfree_skb(skb);
1420 }
1421
1422 /* Used by processes to "lock" a socket state, so that
1423 * interrupts and bottom half handlers won't change it
1424 * from under us. It essentially blocks any incoming
1425 * packets, so that we won't get any new data or any
1426 * packets that change the state of the socket.
1427 *
1428 * While locked, BH processing will add new packets to
1429 * the backlog queue. This queue is processed by the
1430 * owner of the socket lock right before it is released.
1431 *
1432 * Since ~2.3.5 it is also exclusive sleep lock serializing
1433 * accesses from user process context.
1434 */
1435 #define sock_owned_by_user(sk) ((sk)->sk_lock.owned)
1436
1437 static inline void sock_release_ownership(struct sock *sk)
1438 {
1439 sk->sk_lock.owned = 0;
1440 }
1441
1442 /*
1443 * Macro so as to not evaluate some arguments when
1444 * lockdep is not enabled.
1445 *
1446 * Mark both the sk_lock and the sk_lock.slock as a
1447 * per-address-family lock class.
1448 */
1449 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1450 do { \
1451 sk->sk_lock.owned = 0; \
1452 init_waitqueue_head(&sk->sk_lock.wq); \
1453 spin_lock_init(&(sk)->sk_lock.slock); \
1454 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1455 sizeof((sk)->sk_lock)); \
1456 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1457 (skey), (sname)); \
1458 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1459 } while (0)
1460
1461 void lock_sock_nested(struct sock *sk, int subclass);
1462
1463 static inline void lock_sock(struct sock *sk)
1464 {
1465 lock_sock_nested(sk, 0);
1466 }
1467
1468 void release_sock(struct sock *sk);
1469
1470 /* BH context may only use the following locking interface. */
1471 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1472 #define bh_lock_sock_nested(__sk) \
1473 spin_lock_nested(&((__sk)->sk_lock.slock), \
1474 SINGLE_DEPTH_NESTING)
1475 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1476
1477 bool lock_sock_fast(struct sock *sk);
1478 /**
1479 * unlock_sock_fast - complement of lock_sock_fast
1480 * @sk: socket
1481 * @slow: slow mode
1482 *
1483 * fast unlock socket for user context.
1484 * If slow mode is on, we call regular release_sock()
1485 */
1486 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1487 {
1488 if (slow)
1489 release_sock(sk);
1490 else
1491 spin_unlock_bh(&sk->sk_lock.slock);
1492 }
1493
1494
1495 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1496 struct proto *prot, int kern);
1497 void sk_free(struct sock *sk);
1498 void sk_destruct(struct sock *sk);
1499 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1500
1501 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1502 gfp_t priority);
1503 void sock_wfree(struct sk_buff *skb);
1504 void skb_orphan_partial(struct sk_buff *skb);
1505 void sock_rfree(struct sk_buff *skb);
1506 void sock_efree(struct sk_buff *skb);
1507 #ifdef CONFIG_INET
1508 void sock_edemux(struct sk_buff *skb);
1509 #else
1510 #define sock_edemux(skb) sock_efree(skb)
1511 #endif
1512
1513 int sock_setsockopt(struct socket *sock, int level, int op,
1514 char __user *optval, unsigned int optlen);
1515
1516 int sock_getsockopt(struct socket *sock, int level, int op,
1517 char __user *optval, int __user *optlen);
1518 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1519 int noblock, int *errcode);
1520 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1521 unsigned long data_len, int noblock,
1522 int *errcode, int max_page_order);
1523 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1524 void sock_kfree_s(struct sock *sk, void *mem, int size);
1525 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1526 void sk_send_sigurg(struct sock *sk);
1527
1528 /*
1529 * Functions to fill in entries in struct proto_ops when a protocol
1530 * does not implement a particular function.
1531 */
1532 int sock_no_bind(struct socket *, struct sockaddr *, int);
1533 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1534 int sock_no_socketpair(struct socket *, struct socket *);
1535 int sock_no_accept(struct socket *, struct socket *, int);
1536 int sock_no_getname(struct socket *, struct sockaddr *, int *, int);
1537 unsigned int sock_no_poll(struct file *, struct socket *,
1538 struct poll_table_struct *);
1539 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1540 int sock_no_listen(struct socket *, int);
1541 int sock_no_shutdown(struct socket *, int);
1542 int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1543 int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1544 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1545 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1546 int sock_no_mmap(struct file *file, struct socket *sock,
1547 struct vm_area_struct *vma);
1548 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1549 size_t size, int flags);
1550
1551 /*
1552 * Functions to fill in entries in struct proto_ops when a protocol
1553 * uses the inet style.
1554 */
1555 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1556 char __user *optval, int __user *optlen);
1557 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1558 int flags);
1559 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1560 char __user *optval, unsigned int optlen);
1561 int compat_sock_common_getsockopt(struct socket *sock, int level,
1562 int optname, char __user *optval, int __user *optlen);
1563 int compat_sock_common_setsockopt(struct socket *sock, int level,
1564 int optname, char __user *optval, unsigned int optlen);
1565
1566 void sk_common_release(struct sock *sk);
1567
1568 /*
1569 * Default socket callbacks and setup code
1570 */
1571
1572 /* Initialise core socket variables */
1573 void sock_init_data(struct socket *sock, struct sock *sk);
1574
1575 /*
1576 * Socket reference counting postulates.
1577 *
1578 * * Each user of socket SHOULD hold a reference count.
1579 * * Each access point to socket (an hash table bucket, reference from a list,
1580 * running timer, skb in flight MUST hold a reference count.
1581 * * When reference count hits 0, it means it will never increase back.
1582 * * When reference count hits 0, it means that no references from
1583 * outside exist to this socket and current process on current CPU
1584 * is last user and may/should destroy this socket.
1585 * * sk_free is called from any context: process, BH, IRQ. When
1586 * it is called, socket has no references from outside -> sk_free
1587 * may release descendant resources allocated by the socket, but
1588 * to the time when it is called, socket is NOT referenced by any
1589 * hash tables, lists etc.
1590 * * Packets, delivered from outside (from network or from another process)
1591 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1592 * when they sit in queue. Otherwise, packets will leak to hole, when
1593 * socket is looked up by one cpu and unhasing is made by another CPU.
1594 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1595 * (leak to backlog). Packet socket does all the processing inside
1596 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1597 * use separate SMP lock, so that they are prone too.
1598 */
1599
1600 /* Ungrab socket and destroy it, if it was the last reference. */
1601 static inline void sock_put(struct sock *sk)
1602 {
1603 if (atomic_dec_and_test(&sk->sk_refcnt))
1604 sk_free(sk);
1605 }
1606 /* Generic version of sock_put(), dealing with all sockets
1607 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1608 */
1609 void sock_gen_put(struct sock *sk);
1610
1611 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested);
1612
1613 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1614 {
1615 sk->sk_tx_queue_mapping = tx_queue;
1616 }
1617
1618 static inline void sk_tx_queue_clear(struct sock *sk)
1619 {
1620 sk->sk_tx_queue_mapping = -1;
1621 }
1622
1623 static inline int sk_tx_queue_get(const struct sock *sk)
1624 {
1625 return sk ? sk->sk_tx_queue_mapping : -1;
1626 }
1627
1628 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1629 {
1630 sk_tx_queue_clear(sk);
1631 sk->sk_socket = sock;
1632 }
1633
1634 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1635 {
1636 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1637 return &rcu_dereference_raw(sk->sk_wq)->wait;
1638 }
1639 /* Detach socket from process context.
1640 * Announce socket dead, detach it from wait queue and inode.
1641 * Note that parent inode held reference count on this struct sock,
1642 * we do not release it in this function, because protocol
1643 * probably wants some additional cleanups or even continuing
1644 * to work with this socket (TCP).
1645 */
1646 static inline void sock_orphan(struct sock *sk)
1647 {
1648 write_lock_bh(&sk->sk_callback_lock);
1649 sock_set_flag(sk, SOCK_DEAD);
1650 sk_set_socket(sk, NULL);
1651 sk->sk_wq = NULL;
1652 write_unlock_bh(&sk->sk_callback_lock);
1653 }
1654
1655 static inline void sock_graft(struct sock *sk, struct socket *parent)
1656 {
1657 write_lock_bh(&sk->sk_callback_lock);
1658 sk->sk_wq = parent->wq;
1659 parent->sk = sk;
1660 sk_set_socket(sk, parent);
1661 security_sock_graft(sk, parent);
1662 write_unlock_bh(&sk->sk_callback_lock);
1663 }
1664
1665 kuid_t sock_i_uid(struct sock *sk);
1666 unsigned long sock_i_ino(struct sock *sk);
1667
1668 static inline void sk_set_txhash(struct sock *sk)
1669 {
1670 sk->sk_txhash = prandom_u32();
1671
1672 if (unlikely(!sk->sk_txhash))
1673 sk->sk_txhash = 1;
1674 }
1675
1676 static inline void sk_rethink_txhash(struct sock *sk)
1677 {
1678 if (sk->sk_txhash)
1679 sk_set_txhash(sk);
1680 }
1681
1682 static inline struct dst_entry *
1683 __sk_dst_get(struct sock *sk)
1684 {
1685 return rcu_dereference_check(sk->sk_dst_cache, sock_owned_by_user(sk) ||
1686 lockdep_is_held(&sk->sk_lock.slock));
1687 }
1688
1689 static inline struct dst_entry *
1690 sk_dst_get(struct sock *sk)
1691 {
1692 struct dst_entry *dst;
1693
1694 rcu_read_lock();
1695 dst = rcu_dereference(sk->sk_dst_cache);
1696 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1697 dst = NULL;
1698 rcu_read_unlock();
1699 return dst;
1700 }
1701
1702 static inline void dst_negative_advice(struct sock *sk)
1703 {
1704 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1705
1706 sk_rethink_txhash(sk);
1707
1708 if (dst && dst->ops->negative_advice) {
1709 ndst = dst->ops->negative_advice(dst);
1710
1711 if (ndst != dst) {
1712 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1713 sk_tx_queue_clear(sk);
1714 }
1715 }
1716 }
1717
1718 static inline void
1719 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
1720 {
1721 struct dst_entry *old_dst;
1722
1723 sk_tx_queue_clear(sk);
1724 /*
1725 * This can be called while sk is owned by the caller only,
1726 * with no state that can be checked in a rcu_dereference_check() cond
1727 */
1728 old_dst = rcu_dereference_raw(sk->sk_dst_cache);
1729 rcu_assign_pointer(sk->sk_dst_cache, dst);
1730 dst_release(old_dst);
1731 }
1732
1733 static inline void
1734 sk_dst_set(struct sock *sk, struct dst_entry *dst)
1735 {
1736 struct dst_entry *old_dst;
1737
1738 sk_tx_queue_clear(sk);
1739 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1740 dst_release(old_dst);
1741 }
1742
1743 static inline void
1744 __sk_dst_reset(struct sock *sk)
1745 {
1746 __sk_dst_set(sk, NULL);
1747 }
1748
1749 static inline void
1750 sk_dst_reset(struct sock *sk)
1751 {
1752 sk_dst_set(sk, NULL);
1753 }
1754
1755 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1756
1757 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1758
1759 bool sk_mc_loop(struct sock *sk);
1760
1761 static inline bool sk_can_gso(const struct sock *sk)
1762 {
1763 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1764 }
1765
1766 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1767
1768 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1769 {
1770 sk->sk_route_nocaps |= flags;
1771 sk->sk_route_caps &= ~flags;
1772 }
1773
1774 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1775 struct iov_iter *from, char *to,
1776 int copy, int offset)
1777 {
1778 if (skb->ip_summed == CHECKSUM_NONE) {
1779 __wsum csum = 0;
1780 if (csum_and_copy_from_iter(to, copy, &csum, from) != copy)
1781 return -EFAULT;
1782 skb->csum = csum_block_add(skb->csum, csum, offset);
1783 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1784 if (copy_from_iter_nocache(to, copy, from) != copy)
1785 return -EFAULT;
1786 } else if (copy_from_iter(to, copy, from) != copy)
1787 return -EFAULT;
1788
1789 return 0;
1790 }
1791
1792 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1793 struct iov_iter *from, int copy)
1794 {
1795 int err, offset = skb->len;
1796
1797 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1798 copy, offset);
1799 if (err)
1800 __skb_trim(skb, offset);
1801
1802 return err;
1803 }
1804
1805 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1806 struct sk_buff *skb,
1807 struct page *page,
1808 int off, int copy)
1809 {
1810 int err;
1811
1812 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1813 copy, skb->len);
1814 if (err)
1815 return err;
1816
1817 skb->len += copy;
1818 skb->data_len += copy;
1819 skb->truesize += copy;
1820 sk->sk_wmem_queued += copy;
1821 sk_mem_charge(sk, copy);
1822 return 0;
1823 }
1824
1825 /**
1826 * sk_wmem_alloc_get - returns write allocations
1827 * @sk: socket
1828 *
1829 * Returns sk_wmem_alloc minus initial offset of one
1830 */
1831 static inline int sk_wmem_alloc_get(const struct sock *sk)
1832 {
1833 return atomic_read(&sk->sk_wmem_alloc) - 1;
1834 }
1835
1836 /**
1837 * sk_rmem_alloc_get - returns read allocations
1838 * @sk: socket
1839 *
1840 * Returns sk_rmem_alloc
1841 */
1842 static inline int sk_rmem_alloc_get(const struct sock *sk)
1843 {
1844 return atomic_read(&sk->sk_rmem_alloc);
1845 }
1846
1847 /**
1848 * sk_has_allocations - check if allocations are outstanding
1849 * @sk: socket
1850 *
1851 * Returns true if socket has write or read allocations
1852 */
1853 static inline bool sk_has_allocations(const struct sock *sk)
1854 {
1855 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1856 }
1857
1858 /**
1859 * wq_has_sleeper - check if there are any waiting processes
1860 * @wq: struct socket_wq
1861 *
1862 * Returns true if socket_wq has waiting processes
1863 *
1864 * The purpose of the wq_has_sleeper and sock_poll_wait is to wrap the memory
1865 * barrier call. They were added due to the race found within the tcp code.
1866 *
1867 * Consider following tcp code paths:
1868 *
1869 * CPU1 CPU2
1870 *
1871 * sys_select receive packet
1872 * ... ...
1873 * __add_wait_queue update tp->rcv_nxt
1874 * ... ...
1875 * tp->rcv_nxt check sock_def_readable
1876 * ... {
1877 * schedule rcu_read_lock();
1878 * wq = rcu_dereference(sk->sk_wq);
1879 * if (wq && waitqueue_active(&wq->wait))
1880 * wake_up_interruptible(&wq->wait)
1881 * ...
1882 * }
1883 *
1884 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1885 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1886 * could then endup calling schedule and sleep forever if there are no more
1887 * data on the socket.
1888 *
1889 */
1890 static inline bool wq_has_sleeper(struct socket_wq *wq)
1891 {
1892 /* We need to be sure we are in sync with the
1893 * add_wait_queue modifications to the wait queue.
1894 *
1895 * This memory barrier is paired in the sock_poll_wait.
1896 */
1897 smp_mb();
1898 return wq && waitqueue_active(&wq->wait);
1899 }
1900
1901 /**
1902 * sock_poll_wait - place memory barrier behind the poll_wait call.
1903 * @filp: file
1904 * @wait_address: socket wait queue
1905 * @p: poll_table
1906 *
1907 * See the comments in the wq_has_sleeper function.
1908 */
1909 static inline void sock_poll_wait(struct file *filp,
1910 wait_queue_head_t *wait_address, poll_table *p)
1911 {
1912 if (!poll_does_not_wait(p) && wait_address) {
1913 poll_wait(filp, wait_address, p);
1914 /* We need to be sure we are in sync with the
1915 * socket flags modification.
1916 *
1917 * This memory barrier is paired in the wq_has_sleeper.
1918 */
1919 smp_mb();
1920 }
1921 }
1922
1923 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
1924 {
1925 if (sk->sk_txhash) {
1926 skb->l4_hash = 1;
1927 skb->hash = sk->sk_txhash;
1928 }
1929 }
1930
1931 /*
1932 * Queue a received datagram if it will fit. Stream and sequenced
1933 * protocols can't normally use this as they need to fit buffers in
1934 * and play with them.
1935 *
1936 * Inlined as it's very short and called for pretty much every
1937 * packet ever received.
1938 */
1939
1940 static inline void skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1941 {
1942 skb_orphan(skb);
1943 skb->sk = sk;
1944 skb->destructor = sock_wfree;
1945 skb_set_hash_from_sk(skb, sk);
1946 /*
1947 * We used to take a refcount on sk, but following operation
1948 * is enough to guarantee sk_free() wont free this sock until
1949 * all in-flight packets are completed
1950 */
1951 atomic_add(skb->truesize, &sk->sk_wmem_alloc);
1952 }
1953
1954 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
1955 {
1956 skb_orphan(skb);
1957 skb->sk = sk;
1958 skb->destructor = sock_rfree;
1959 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
1960 sk_mem_charge(sk, skb->truesize);
1961 }
1962
1963 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
1964 unsigned long expires);
1965
1966 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
1967
1968 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
1969
1970 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
1971 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
1972
1973 /*
1974 * Recover an error report and clear atomically
1975 */
1976
1977 static inline int sock_error(struct sock *sk)
1978 {
1979 int err;
1980 if (likely(!sk->sk_err))
1981 return 0;
1982 err = xchg(&sk->sk_err, 0);
1983 return -err;
1984 }
1985
1986 static inline unsigned long sock_wspace(struct sock *sk)
1987 {
1988 int amt = 0;
1989
1990 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
1991 amt = sk->sk_sndbuf - atomic_read(&sk->sk_wmem_alloc);
1992 if (amt < 0)
1993 amt = 0;
1994 }
1995 return amt;
1996 }
1997
1998 static inline void sk_wake_async(struct sock *sk, int how, int band)
1999 {
2000 if (sock_flag(sk, SOCK_FASYNC))
2001 sock_wake_async(sk->sk_socket, how, band);
2002 }
2003
2004 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2005 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2006 * Note: for send buffers, TCP works better if we can build two skbs at
2007 * minimum.
2008 */
2009 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2010
2011 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2012 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2013
2014 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2015 {
2016 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2017 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2018 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2019 }
2020 }
2021
2022 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2023 bool force_schedule);
2024
2025 /**
2026 * sk_page_frag - return an appropriate page_frag
2027 * @sk: socket
2028 *
2029 * If socket allocation mode allows current thread to sleep, it means its
2030 * safe to use the per task page_frag instead of the per socket one.
2031 */
2032 static inline struct page_frag *sk_page_frag(struct sock *sk)
2033 {
2034 if (sk->sk_allocation & __GFP_WAIT)
2035 return ¤t->task_frag;
2036
2037 return &sk->sk_frag;
2038 }
2039
2040 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2041
2042 /*
2043 * Default write policy as shown to user space via poll/select/SIGIO
2044 */
2045 static inline bool sock_writeable(const struct sock *sk)
2046 {
2047 return atomic_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2048 }
2049
2050 static inline gfp_t gfp_any(void)
2051 {
2052 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2053 }
2054
2055 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2056 {
2057 return noblock ? 0 : sk->sk_rcvtimeo;
2058 }
2059
2060 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2061 {
2062 return noblock ? 0 : sk->sk_sndtimeo;
2063 }
2064
2065 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2066 {
2067 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2068 }
2069
2070 /* Alas, with timeout socket operations are not restartable.
2071 * Compare this to poll().
2072 */
2073 static inline int sock_intr_errno(long timeo)
2074 {
2075 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2076 }
2077
2078 struct sock_skb_cb {
2079 u32 dropcount;
2080 };
2081
2082 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2083 * using skb->cb[] would keep using it directly and utilize its
2084 * alignement guarantee.
2085 */
2086 #define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2087 sizeof(struct sock_skb_cb)))
2088
2089 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2090 SOCK_SKB_CB_OFFSET))
2091
2092 #define sock_skb_cb_check_size(size) \
2093 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2094
2095 static inline void
2096 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2097 {
2098 SOCK_SKB_CB(skb)->dropcount = atomic_read(&sk->sk_drops);
2099 }
2100
2101 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2102 struct sk_buff *skb);
2103 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2104 struct sk_buff *skb);
2105
2106 static inline void
2107 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2108 {
2109 ktime_t kt = skb->tstamp;
2110 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2111
2112 /*
2113 * generate control messages if
2114 * - receive time stamping in software requested
2115 * - software time stamp available and wanted
2116 * - hardware time stamps available and wanted
2117 */
2118 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2119 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2120 (kt.tv64 && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2121 (hwtstamps->hwtstamp.tv64 &&
2122 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2123 __sock_recv_timestamp(msg, sk, skb);
2124 else
2125 sk->sk_stamp = kt;
2126
2127 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2128 __sock_recv_wifi_status(msg, sk, skb);
2129 }
2130
2131 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2132 struct sk_buff *skb);
2133
2134 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2135 struct sk_buff *skb)
2136 {
2137 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2138 (1UL << SOCK_RCVTSTAMP))
2139 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2140 SOF_TIMESTAMPING_RAW_HARDWARE)
2141
2142 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2143 __sock_recv_ts_and_drops(msg, sk, skb);
2144 else
2145 sk->sk_stamp = skb->tstamp;
2146 }
2147
2148 void __sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags);
2149
2150 /**
2151 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2152 * @sk: socket sending this packet
2153 * @tx_flags: completed with instructions for time stamping
2154 *
2155 * Note : callers should take care of initial *tx_flags value (usually 0)
2156 */
2157 static inline void sock_tx_timestamp(const struct sock *sk, __u8 *tx_flags)
2158 {
2159 if (unlikely(sk->sk_tsflags))
2160 __sock_tx_timestamp(sk, tx_flags);
2161 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2162 *tx_flags |= SKBTX_WIFI_STATUS;
2163 }
2164
2165 /**
2166 * sk_eat_skb - Release a skb if it is no longer needed
2167 * @sk: socket to eat this skb from
2168 * @skb: socket buffer to eat
2169 *
2170 * This routine must be called with interrupts disabled or with the socket
2171 * locked so that the sk_buff queue operation is ok.
2172 */
2173 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2174 {
2175 __skb_unlink(skb, &sk->sk_receive_queue);
2176 __kfree_skb(skb);
2177 }
2178
2179 static inline
2180 struct net *sock_net(const struct sock *sk)
2181 {
2182 return read_pnet(&sk->sk_net);
2183 }
2184
2185 static inline
2186 void sock_net_set(struct sock *sk, struct net *net)
2187 {
2188 write_pnet(&sk->sk_net, net);
2189 }
2190
2191 static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2192 {
2193 if (skb->sk) {
2194 struct sock *sk = skb->sk;
2195
2196 skb->destructor = NULL;
2197 skb->sk = NULL;
2198 return sk;
2199 }
2200 return NULL;
2201 }
2202
2203 /* This helper checks if a socket is a full socket,
2204 * ie _not_ a timewait or request socket.
2205 */
2206 static inline bool sk_fullsock(const struct sock *sk)
2207 {
2208 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2209 }
2210
2211 void sock_enable_timestamp(struct sock *sk, int flag);
2212 int sock_get_timestamp(struct sock *, struct timeval __user *);
2213 int sock_get_timestampns(struct sock *, struct timespec __user *);
2214 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2215 int type);
2216
2217 bool sk_ns_capable(const struct sock *sk,
2218 struct user_namespace *user_ns, int cap);
2219 bool sk_capable(const struct sock *sk, int cap);
2220 bool sk_net_capable(const struct sock *sk, int cap);
2221
2222 extern __u32 sysctl_wmem_max;
2223 extern __u32 sysctl_rmem_max;
2224
2225 extern int sysctl_tstamp_allow_data;
2226 extern int sysctl_optmem_max;
2227
2228 extern __u32 sysctl_wmem_default;
2229 extern __u32 sysctl_rmem_default;
2230
2231 #endif /* _SOCK_H */
2232
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.