Loading...
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#include <linux/wait.h>
62#include <linux/cgroup-defs.h>
63#include <linux/rbtree.h>
64#include <linux/filter.h>
65#include <linux/rculist_nulls.h>
66#include <linux/poll.h>
67
68#include <linux/atomic.h>
69#include <linux/refcount.h>
70#include <net/dst.h>
71#include <net/checksum.h>
72#include <net/tcp_states.h>
73#include <linux/net_tstamp.h>
74#include <net/smc.h>
75#include <net/l3mdev.h>
76
77/*
78 * This structure really needs to be cleaned up.
79 * Most of it is for TCP, and not used by any of
80 * the other protocols.
81 */
82
83/* Define this to get the SOCK_DBG debugging facility. */
84#define SOCK_DEBUGGING
85#ifdef SOCK_DEBUGGING
86#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
87 printk(KERN_DEBUG msg); } while (0)
88#else
89/* Validate arguments and do nothing */
90static inline __printf(2, 3)
91void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
92{
93}
94#endif
95
96/* This is the per-socket lock. The spinlock provides a synchronization
97 * between user contexts and software interrupt processing, whereas the
98 * mini-semaphore synchronizes multiple users amongst themselves.
99 */
100typedef struct {
101 spinlock_t slock;
102 int owned;
103 wait_queue_head_t wq;
104 /*
105 * We express the mutex-alike socket_lock semantics
106 * to the lock validator by explicitly managing
107 * the slock as a lock variant (in addition to
108 * the slock itself):
109 */
110#ifdef CONFIG_DEBUG_LOCK_ALLOC
111 struct lockdep_map dep_map;
112#endif
113} socket_lock_t;
114
115struct sock;
116struct proto;
117struct net;
118
119typedef __u32 __bitwise __portpair;
120typedef __u64 __bitwise __addrpair;
121
122/**
123 * struct sock_common - minimal network layer representation of sockets
124 * @skc_daddr: Foreign IPv4 addr
125 * @skc_rcv_saddr: Bound local IPv4 addr
126 * @skc_hash: hash value used with various protocol lookup tables
127 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
128 * @skc_dport: placeholder for inet_dport/tw_dport
129 * @skc_num: placeholder for inet_num/tw_num
130 * @skc_family: network address family
131 * @skc_state: Connection state
132 * @skc_reuse: %SO_REUSEADDR setting
133 * @skc_reuseport: %SO_REUSEPORT setting
134 * @skc_bound_dev_if: bound device index if != 0
135 * @skc_bind_node: bind hash linkage for various protocol lookup tables
136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137 * @skc_prot: protocol handlers inside a network family
138 * @skc_net: reference to the network namespace of this socket
139 * @skc_node: main hash linkage for various protocol lookup tables
140 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
141 * @skc_tx_queue_mapping: tx queue number for this connection
142 * @skc_flags: place holder for sk_flags
143 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
144 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
145 * @skc_incoming_cpu: record/match cpu processing incoming packets
146 * @skc_refcnt: reference count
147 *
148 * This is the minimal network layer representation of sockets, the header
149 * for struct sock and struct inet_timewait_sock.
150 */
151struct sock_common {
152 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
153 * address on 64bit arches : cf INET_MATCH()
154 */
155 union {
156 __addrpair skc_addrpair;
157 struct {
158 __be32 skc_daddr;
159 __be32 skc_rcv_saddr;
160 };
161 };
162 union {
163 unsigned int skc_hash;
164 __u16 skc_u16hashes[2];
165 };
166 /* skc_dport && skc_num must be grouped as well */
167 union {
168 __portpair skc_portpair;
169 struct {
170 __be16 skc_dport;
171 __u16 skc_num;
172 };
173 };
174
175 unsigned short skc_family;
176 volatile unsigned char skc_state;
177 unsigned char skc_reuse:4;
178 unsigned char skc_reuseport:1;
179 unsigned char skc_ipv6only:1;
180 unsigned char skc_net_refcnt:1;
181 int skc_bound_dev_if;
182 union {
183 struct hlist_node skc_bind_node;
184 struct hlist_node skc_portaddr_node;
185 };
186 struct proto *skc_prot;
187 possible_net_t skc_net;
188
189#if IS_ENABLED(CONFIG_IPV6)
190 struct in6_addr skc_v6_daddr;
191 struct in6_addr skc_v6_rcv_saddr;
192#endif
193
194 atomic64_t skc_cookie;
195
196 /* following fields are padding to force
197 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
198 * assuming IPV6 is enabled. We use this padding differently
199 * for different kind of 'sockets'
200 */
201 union {
202 unsigned long skc_flags;
203 struct sock *skc_listener; /* request_sock */
204 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
205 };
206 /*
207 * fields between dontcopy_begin/dontcopy_end
208 * are not copied in sock_copy()
209 */
210 /* private: */
211 int skc_dontcopy_begin[0];
212 /* public: */
213 union {
214 struct hlist_node skc_node;
215 struct hlist_nulls_node skc_nulls_node;
216 };
217 int skc_tx_queue_mapping;
218 union {
219 int skc_incoming_cpu;
220 u32 skc_rcv_wnd;
221 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
222 };
223
224 refcount_t skc_refcnt;
225 /* private: */
226 int skc_dontcopy_end[0];
227 union {
228 u32 skc_rxhash;
229 u32 skc_window_clamp;
230 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
231 };
232 /* public: */
233};
234
235/**
236 * struct sock - network layer representation of sockets
237 * @__sk_common: shared layout with inet_timewait_sock
238 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
239 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
240 * @sk_lock: synchronizer
241 * @sk_kern_sock: True if sock is using kernel lock classes
242 * @sk_rcvbuf: size of receive buffer in bytes
243 * @sk_wq: sock wait queue and async head
244 * @sk_rx_dst: receive input route used by early demux
245 * @sk_dst_cache: destination cache
246 * @sk_dst_pending_confirm: need to confirm neighbour
247 * @sk_policy: flow policy
248 * @sk_receive_queue: incoming packets
249 * @sk_wmem_alloc: transmit queue bytes committed
250 * @sk_tsq_flags: TCP Small Queues flags
251 * @sk_write_queue: Packet sending queue
252 * @sk_omem_alloc: "o" is "option" or "other"
253 * @sk_wmem_queued: persistent queue size
254 * @sk_forward_alloc: space allocated forward
255 * @sk_napi_id: id of the last napi context to receive data for sk
256 * @sk_ll_usec: usecs to busypoll when there is no data
257 * @sk_allocation: allocation mode
258 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
259 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
260 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
261 * @sk_sndbuf: size of send buffer in bytes
262 * @__sk_flags_offset: empty field used to determine location of bitfield
263 * @sk_padding: unused element for alignment
264 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
265 * @sk_no_check_rx: allow zero checksum in RX packets
266 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
267 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
268 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
269 * @sk_gso_max_size: Maximum GSO segment size to build
270 * @sk_gso_max_segs: Maximum number of GSO segments
271 * @sk_pacing_shift: scaling factor for TCP Small Queues
272 * @sk_lingertime: %SO_LINGER l_linger setting
273 * @sk_backlog: always used with the per-socket spinlock held
274 * @sk_callback_lock: used with the callbacks in the end of this struct
275 * @sk_error_queue: rarely used
276 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
277 * IPV6_ADDRFORM for instance)
278 * @sk_err: last error
279 * @sk_err_soft: errors that don't cause failure but are the cause of a
280 * persistent failure not just 'timed out'
281 * @sk_drops: raw/udp drops counter
282 * @sk_ack_backlog: current listen backlog
283 * @sk_max_ack_backlog: listen backlog set in listen()
284 * @sk_uid: user id of owner
285 * @sk_priority: %SO_PRIORITY setting
286 * @sk_type: socket type (%SOCK_STREAM, etc)
287 * @sk_protocol: which protocol this socket belongs in this network family
288 * @sk_peer_pid: &struct pid for this socket's peer
289 * @sk_peer_cred: %SO_PEERCRED setting
290 * @sk_rcvlowat: %SO_RCVLOWAT setting
291 * @sk_rcvtimeo: %SO_RCVTIMEO setting
292 * @sk_sndtimeo: %SO_SNDTIMEO setting
293 * @sk_txhash: computed flow hash for use on transmit
294 * @sk_filter: socket filtering instructions
295 * @sk_timer: sock cleanup timer
296 * @sk_stamp: time stamp of last packet received
297 * @sk_tsflags: SO_TIMESTAMPING socket options
298 * @sk_tskey: counter to disambiguate concurrent tstamp requests
299 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
300 * @sk_socket: Identd and reporting IO signals
301 * @sk_user_data: RPC layer private data
302 * @sk_frag: cached page frag
303 * @sk_peek_off: current peek_offset value
304 * @sk_send_head: front of stuff to transmit
305 * @sk_security: used by security modules
306 * @sk_mark: generic packet mark
307 * @sk_cgrp_data: cgroup data for this cgroup
308 * @sk_memcg: this socket's memory cgroup association
309 * @sk_write_pending: a write to stream socket waits to start
310 * @sk_state_change: callback to indicate change in the state of the sock
311 * @sk_data_ready: callback to indicate there is data to be processed
312 * @sk_write_space: callback to indicate there is bf sending space available
313 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
314 * @sk_backlog_rcv: callback to process the backlog
315 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
316 * @sk_reuseport_cb: reuseport group container
317 * @sk_rcu: used during RCU grace period
318 */
319struct sock {
320 /*
321 * Now struct inet_timewait_sock also uses sock_common, so please just
322 * don't add nothing before this first member (__sk_common) --acme
323 */
324 struct sock_common __sk_common;
325#define sk_node __sk_common.skc_node
326#define sk_nulls_node __sk_common.skc_nulls_node
327#define sk_refcnt __sk_common.skc_refcnt
328#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
329
330#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
331#define sk_dontcopy_end __sk_common.skc_dontcopy_end
332#define sk_hash __sk_common.skc_hash
333#define sk_portpair __sk_common.skc_portpair
334#define sk_num __sk_common.skc_num
335#define sk_dport __sk_common.skc_dport
336#define sk_addrpair __sk_common.skc_addrpair
337#define sk_daddr __sk_common.skc_daddr
338#define sk_rcv_saddr __sk_common.skc_rcv_saddr
339#define sk_family __sk_common.skc_family
340#define sk_state __sk_common.skc_state
341#define sk_reuse __sk_common.skc_reuse
342#define sk_reuseport __sk_common.skc_reuseport
343#define sk_ipv6only __sk_common.skc_ipv6only
344#define sk_net_refcnt __sk_common.skc_net_refcnt
345#define sk_bound_dev_if __sk_common.skc_bound_dev_if
346#define sk_bind_node __sk_common.skc_bind_node
347#define sk_prot __sk_common.skc_prot
348#define sk_net __sk_common.skc_net
349#define sk_v6_daddr __sk_common.skc_v6_daddr
350#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
351#define sk_cookie __sk_common.skc_cookie
352#define sk_incoming_cpu __sk_common.skc_incoming_cpu
353#define sk_flags __sk_common.skc_flags
354#define sk_rxhash __sk_common.skc_rxhash
355
356 socket_lock_t sk_lock;
357 atomic_t sk_drops;
358 int sk_rcvlowat;
359 struct sk_buff_head sk_error_queue;
360 struct sk_buff_head sk_receive_queue;
361 /*
362 * The backlog queue is special, it is always used with
363 * the per-socket spinlock held and requires low latency
364 * access. Therefore we special case it's implementation.
365 * Note : rmem_alloc is in this structure to fill a hole
366 * on 64bit arches, not because its logically part of
367 * backlog.
368 */
369 struct {
370 atomic_t rmem_alloc;
371 int len;
372 struct sk_buff *head;
373 struct sk_buff *tail;
374 } sk_backlog;
375#define sk_rmem_alloc sk_backlog.rmem_alloc
376
377 int sk_forward_alloc;
378#ifdef CONFIG_NET_RX_BUSY_POLL
379 unsigned int sk_ll_usec;
380 /* ===== mostly read cache line ===== */
381 unsigned int sk_napi_id;
382#endif
383 int sk_rcvbuf;
384
385 struct sk_filter __rcu *sk_filter;
386 union {
387 struct socket_wq __rcu *sk_wq;
388 struct socket_wq *sk_wq_raw;
389 };
390#ifdef CONFIG_XFRM
391 struct xfrm_policy __rcu *sk_policy[2];
392#endif
393 struct dst_entry *sk_rx_dst;
394 struct dst_entry __rcu *sk_dst_cache;
395 atomic_t sk_omem_alloc;
396 int sk_sndbuf;
397
398 /* ===== cache line for TX ===== */
399 int sk_wmem_queued;
400 refcount_t sk_wmem_alloc;
401 unsigned long sk_tsq_flags;
402 union {
403 struct sk_buff *sk_send_head;
404 struct rb_root tcp_rtx_queue;
405 };
406 struct sk_buff_head sk_write_queue;
407 __s32 sk_peek_off;
408 int sk_write_pending;
409 __u32 sk_dst_pending_confirm;
410 u32 sk_pacing_status; /* see enum sk_pacing */
411 long sk_sndtimeo;
412 struct timer_list sk_timer;
413 __u32 sk_priority;
414 __u32 sk_mark;
415 u32 sk_pacing_rate; /* bytes per second */
416 u32 sk_max_pacing_rate;
417 struct page_frag sk_frag;
418 netdev_features_t sk_route_caps;
419 netdev_features_t sk_route_nocaps;
420 netdev_features_t sk_route_forced_caps;
421 int sk_gso_type;
422 unsigned int sk_gso_max_size;
423 gfp_t sk_allocation;
424 __u32 sk_txhash;
425
426 /*
427 * Because of non atomicity rules, all
428 * changes are protected by socket lock.
429 */
430 unsigned int __sk_flags_offset[0];
431#ifdef __BIG_ENDIAN_BITFIELD
432#define SK_FL_PROTO_SHIFT 16
433#define SK_FL_PROTO_MASK 0x00ff0000
434
435#define SK_FL_TYPE_SHIFT 0
436#define SK_FL_TYPE_MASK 0x0000ffff
437#else
438#define SK_FL_PROTO_SHIFT 8
439#define SK_FL_PROTO_MASK 0x0000ff00
440
441#define SK_FL_TYPE_SHIFT 16
442#define SK_FL_TYPE_MASK 0xffff0000
443#endif
444
445 unsigned int sk_padding : 1,
446 sk_kern_sock : 1,
447 sk_no_check_tx : 1,
448 sk_no_check_rx : 1,
449 sk_userlocks : 4,
450 sk_protocol : 8,
451 sk_type : 16;
452#define SK_PROTOCOL_MAX U8_MAX
453 u16 sk_gso_max_segs;
454 u8 sk_pacing_shift;
455 unsigned long sk_lingertime;
456 struct proto *sk_prot_creator;
457 rwlock_t sk_callback_lock;
458 int sk_err,
459 sk_err_soft;
460 u32 sk_ack_backlog;
461 u32 sk_max_ack_backlog;
462 kuid_t sk_uid;
463 struct pid *sk_peer_pid;
464 const struct cred *sk_peer_cred;
465 long sk_rcvtimeo;
466 ktime_t sk_stamp;
467 u16 sk_tsflags;
468 u8 sk_shutdown;
469 u32 sk_tskey;
470 atomic_t sk_zckey;
471 struct socket *sk_socket;
472 void *sk_user_data;
473#ifdef CONFIG_SECURITY
474 void *sk_security;
475#endif
476 struct sock_cgroup_data sk_cgrp_data;
477 struct mem_cgroup *sk_memcg;
478 void (*sk_state_change)(struct sock *sk);
479 void (*sk_data_ready)(struct sock *sk);
480 void (*sk_write_space)(struct sock *sk);
481 void (*sk_error_report)(struct sock *sk);
482 int (*sk_backlog_rcv)(struct sock *sk,
483 struct sk_buff *skb);
484 void (*sk_destruct)(struct sock *sk);
485 struct sock_reuseport __rcu *sk_reuseport_cb;
486 struct rcu_head sk_rcu;
487};
488
489enum sk_pacing {
490 SK_PACING_NONE = 0,
491 SK_PACING_NEEDED = 1,
492 SK_PACING_FQ = 2,
493};
494
495#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
496
497#define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
498#define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
499
500/*
501 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
502 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
503 * on a socket means that the socket will reuse everybody else's port
504 * without looking at the other's sk_reuse value.
505 */
506
507#define SK_NO_REUSE 0
508#define SK_CAN_REUSE 1
509#define SK_FORCE_REUSE 2
510
511int sk_set_peek_off(struct sock *sk, int val);
512
513static inline int sk_peek_offset(struct sock *sk, int flags)
514{
515 if (unlikely(flags & MSG_PEEK)) {
516 return READ_ONCE(sk->sk_peek_off);
517 }
518
519 return 0;
520}
521
522static inline void sk_peek_offset_bwd(struct sock *sk, int val)
523{
524 s32 off = READ_ONCE(sk->sk_peek_off);
525
526 if (unlikely(off >= 0)) {
527 off = max_t(s32, off - val, 0);
528 WRITE_ONCE(sk->sk_peek_off, off);
529 }
530}
531
532static inline void sk_peek_offset_fwd(struct sock *sk, int val)
533{
534 sk_peek_offset_bwd(sk, -val);
535}
536
537/*
538 * Hashed lists helper routines
539 */
540static inline struct sock *sk_entry(const struct hlist_node *node)
541{
542 return hlist_entry(node, struct sock, sk_node);
543}
544
545static inline struct sock *__sk_head(const struct hlist_head *head)
546{
547 return hlist_entry(head->first, struct sock, sk_node);
548}
549
550static inline struct sock *sk_head(const struct hlist_head *head)
551{
552 return hlist_empty(head) ? NULL : __sk_head(head);
553}
554
555static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
556{
557 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
558}
559
560static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
561{
562 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
563}
564
565static inline struct sock *sk_next(const struct sock *sk)
566{
567 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
568}
569
570static inline struct sock *sk_nulls_next(const struct sock *sk)
571{
572 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
573 hlist_nulls_entry(sk->sk_nulls_node.next,
574 struct sock, sk_nulls_node) :
575 NULL;
576}
577
578static inline bool sk_unhashed(const struct sock *sk)
579{
580 return hlist_unhashed(&sk->sk_node);
581}
582
583static inline bool sk_hashed(const struct sock *sk)
584{
585 return !sk_unhashed(sk);
586}
587
588static inline void sk_node_init(struct hlist_node *node)
589{
590 node->pprev = NULL;
591}
592
593static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
594{
595 node->pprev = NULL;
596}
597
598static inline void __sk_del_node(struct sock *sk)
599{
600 __hlist_del(&sk->sk_node);
601}
602
603/* NB: equivalent to hlist_del_init_rcu */
604static inline bool __sk_del_node_init(struct sock *sk)
605{
606 if (sk_hashed(sk)) {
607 __sk_del_node(sk);
608 sk_node_init(&sk->sk_node);
609 return true;
610 }
611 return false;
612}
613
614/* Grab socket reference count. This operation is valid only
615 when sk is ALREADY grabbed f.e. it is found in hash table
616 or a list and the lookup is made under lock preventing hash table
617 modifications.
618 */
619
620static __always_inline void sock_hold(struct sock *sk)
621{
622 refcount_inc(&sk->sk_refcnt);
623}
624
625/* Ungrab socket in the context, which assumes that socket refcnt
626 cannot hit zero, f.e. it is true in context of any socketcall.
627 */
628static __always_inline void __sock_put(struct sock *sk)
629{
630 refcount_dec(&sk->sk_refcnt);
631}
632
633static inline bool sk_del_node_init(struct sock *sk)
634{
635 bool rc = __sk_del_node_init(sk);
636
637 if (rc) {
638 /* paranoid for a while -acme */
639 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
640 __sock_put(sk);
641 }
642 return rc;
643}
644#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
645
646static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
647{
648 if (sk_hashed(sk)) {
649 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
650 return true;
651 }
652 return false;
653}
654
655static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
656{
657 bool rc = __sk_nulls_del_node_init_rcu(sk);
658
659 if (rc) {
660 /* paranoid for a while -acme */
661 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
662 __sock_put(sk);
663 }
664 return rc;
665}
666
667static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
668{
669 hlist_add_head(&sk->sk_node, list);
670}
671
672static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
673{
674 sock_hold(sk);
675 __sk_add_node(sk, list);
676}
677
678static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
679{
680 sock_hold(sk);
681 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
682 sk->sk_family == AF_INET6)
683 hlist_add_tail_rcu(&sk->sk_node, list);
684 else
685 hlist_add_head_rcu(&sk->sk_node, list);
686}
687
688static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
689{
690 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
691}
692
693static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
694{
695 sock_hold(sk);
696 __sk_nulls_add_node_rcu(sk, list);
697}
698
699static inline void __sk_del_bind_node(struct sock *sk)
700{
701 __hlist_del(&sk->sk_bind_node);
702}
703
704static inline void sk_add_bind_node(struct sock *sk,
705 struct hlist_head *list)
706{
707 hlist_add_head(&sk->sk_bind_node, list);
708}
709
710#define sk_for_each(__sk, list) \
711 hlist_for_each_entry(__sk, list, sk_node)
712#define sk_for_each_rcu(__sk, list) \
713 hlist_for_each_entry_rcu(__sk, list, sk_node)
714#define sk_nulls_for_each(__sk, node, list) \
715 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
716#define sk_nulls_for_each_rcu(__sk, node, list) \
717 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
718#define sk_for_each_from(__sk) \
719 hlist_for_each_entry_from(__sk, sk_node)
720#define sk_nulls_for_each_from(__sk, node) \
721 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
722 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
723#define sk_for_each_safe(__sk, tmp, list) \
724 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
725#define sk_for_each_bound(__sk, list) \
726 hlist_for_each_entry(__sk, list, sk_bind_node)
727
728/**
729 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
730 * @tpos: the type * to use as a loop cursor.
731 * @pos: the &struct hlist_node to use as a loop cursor.
732 * @head: the head for your list.
733 * @offset: offset of hlist_node within the struct.
734 *
735 */
736#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
737 for (pos = rcu_dereference(hlist_first_rcu(head)); \
738 pos != NULL && \
739 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
740 pos = rcu_dereference(hlist_next_rcu(pos)))
741
742static inline struct user_namespace *sk_user_ns(struct sock *sk)
743{
744 /* Careful only use this in a context where these parameters
745 * can not change and must all be valid, such as recvmsg from
746 * userspace.
747 */
748 return sk->sk_socket->file->f_cred->user_ns;
749}
750
751/* Sock flags */
752enum sock_flags {
753 SOCK_DEAD,
754 SOCK_DONE,
755 SOCK_URGINLINE,
756 SOCK_KEEPOPEN,
757 SOCK_LINGER,
758 SOCK_DESTROY,
759 SOCK_BROADCAST,
760 SOCK_TIMESTAMP,
761 SOCK_ZAPPED,
762 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
763 SOCK_DBG, /* %SO_DEBUG setting */
764 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
765 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
766 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
767 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
768 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
769 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
770 SOCK_FASYNC, /* fasync() active */
771 SOCK_RXQ_OVFL,
772 SOCK_ZEROCOPY, /* buffers from userspace */
773 SOCK_WIFI_STATUS, /* push wifi status to userspace */
774 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
775 * Will use last 4 bytes of packet sent from
776 * user-space instead.
777 */
778 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
779 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
780 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
781};
782
783#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
784
785static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
786{
787 nsk->sk_flags = osk->sk_flags;
788}
789
790static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
791{
792 __set_bit(flag, &sk->sk_flags);
793}
794
795static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
796{
797 __clear_bit(flag, &sk->sk_flags);
798}
799
800static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
801{
802 return test_bit(flag, &sk->sk_flags);
803}
804
805#ifdef CONFIG_NET
806extern struct static_key memalloc_socks;
807static inline int sk_memalloc_socks(void)
808{
809 return static_key_false(&memalloc_socks);
810}
811#else
812
813static inline int sk_memalloc_socks(void)
814{
815 return 0;
816}
817
818#endif
819
820static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
821{
822 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
823}
824
825static inline void sk_acceptq_removed(struct sock *sk)
826{
827 sk->sk_ack_backlog--;
828}
829
830static inline void sk_acceptq_added(struct sock *sk)
831{
832 sk->sk_ack_backlog++;
833}
834
835static inline bool sk_acceptq_is_full(const struct sock *sk)
836{
837 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
838}
839
840/*
841 * Compute minimal free write space needed to queue new packets.
842 */
843static inline int sk_stream_min_wspace(const struct sock *sk)
844{
845 return sk->sk_wmem_queued >> 1;
846}
847
848static inline int sk_stream_wspace(const struct sock *sk)
849{
850 return sk->sk_sndbuf - sk->sk_wmem_queued;
851}
852
853void sk_stream_write_space(struct sock *sk);
854
855/* OOB backlog add */
856static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
857{
858 /* dont let skb dst not refcounted, we are going to leave rcu lock */
859 skb_dst_force(skb);
860
861 if (!sk->sk_backlog.tail)
862 sk->sk_backlog.head = skb;
863 else
864 sk->sk_backlog.tail->next = skb;
865
866 sk->sk_backlog.tail = skb;
867 skb->next = NULL;
868}
869
870/*
871 * Take into account size of receive queue and backlog queue
872 * Do not take into account this skb truesize,
873 * to allow even a single big packet to come.
874 */
875static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
876{
877 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
878
879 return qsize > limit;
880}
881
882/* The per-socket spinlock must be held here. */
883static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
884 unsigned int limit)
885{
886 if (sk_rcvqueues_full(sk, limit))
887 return -ENOBUFS;
888
889 /*
890 * If the skb was allocated from pfmemalloc reserves, only
891 * allow SOCK_MEMALLOC sockets to use it as this socket is
892 * helping free memory
893 */
894 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
895 return -ENOMEM;
896
897 __sk_add_backlog(sk, skb);
898 sk->sk_backlog.len += skb->truesize;
899 return 0;
900}
901
902int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
903
904static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
905{
906 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
907 return __sk_backlog_rcv(sk, skb);
908
909 return sk->sk_backlog_rcv(sk, skb);
910}
911
912static inline void sk_incoming_cpu_update(struct sock *sk)
913{
914 int cpu = raw_smp_processor_id();
915
916 if (unlikely(sk->sk_incoming_cpu != cpu))
917 sk->sk_incoming_cpu = cpu;
918}
919
920static inline void sock_rps_record_flow_hash(__u32 hash)
921{
922#ifdef CONFIG_RPS
923 struct rps_sock_flow_table *sock_flow_table;
924
925 rcu_read_lock();
926 sock_flow_table = rcu_dereference(rps_sock_flow_table);
927 rps_record_sock_flow(sock_flow_table, hash);
928 rcu_read_unlock();
929#endif
930}
931
932static inline void sock_rps_record_flow(const struct sock *sk)
933{
934#ifdef CONFIG_RPS
935 if (static_key_false(&rfs_needed)) {
936 /* Reading sk->sk_rxhash might incur an expensive cache line
937 * miss.
938 *
939 * TCP_ESTABLISHED does cover almost all states where RFS
940 * might be useful, and is cheaper [1] than testing :
941 * IPv4: inet_sk(sk)->inet_daddr
942 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
943 * OR an additional socket flag
944 * [1] : sk_state and sk_prot are in the same cache line.
945 */
946 if (sk->sk_state == TCP_ESTABLISHED)
947 sock_rps_record_flow_hash(sk->sk_rxhash);
948 }
949#endif
950}
951
952static inline void sock_rps_save_rxhash(struct sock *sk,
953 const struct sk_buff *skb)
954{
955#ifdef CONFIG_RPS
956 if (unlikely(sk->sk_rxhash != skb->hash))
957 sk->sk_rxhash = skb->hash;
958#endif
959}
960
961static inline void sock_rps_reset_rxhash(struct sock *sk)
962{
963#ifdef CONFIG_RPS
964 sk->sk_rxhash = 0;
965#endif
966}
967
968#define sk_wait_event(__sk, __timeo, __condition, __wait) \
969 ({ int __rc; \
970 release_sock(__sk); \
971 __rc = __condition; \
972 if (!__rc) { \
973 *(__timeo) = wait_woken(__wait, \
974 TASK_INTERRUPTIBLE, \
975 *(__timeo)); \
976 } \
977 sched_annotate_sleep(); \
978 lock_sock(__sk); \
979 __rc = __condition; \
980 __rc; \
981 })
982
983int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
984int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
985void sk_stream_wait_close(struct sock *sk, long timeo_p);
986int sk_stream_error(struct sock *sk, int flags, int err);
987void sk_stream_kill_queues(struct sock *sk);
988void sk_set_memalloc(struct sock *sk);
989void sk_clear_memalloc(struct sock *sk);
990
991void __sk_flush_backlog(struct sock *sk);
992
993static inline bool sk_flush_backlog(struct sock *sk)
994{
995 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
996 __sk_flush_backlog(sk);
997 return true;
998 }
999 return false;
1000}
1001
1002int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1003
1004struct request_sock_ops;
1005struct timewait_sock_ops;
1006struct inet_hashinfo;
1007struct raw_hashinfo;
1008struct smc_hashinfo;
1009struct module;
1010
1011/*
1012 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1013 * un-modified. Special care is taken when initializing object to zero.
1014 */
1015static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1016{
1017 if (offsetof(struct sock, sk_node.next) != 0)
1018 memset(sk, 0, offsetof(struct sock, sk_node.next));
1019 memset(&sk->sk_node.pprev, 0,
1020 size - offsetof(struct sock, sk_node.pprev));
1021}
1022
1023/* Networking protocol blocks we attach to sockets.
1024 * socket layer -> transport layer interface
1025 */
1026struct proto {
1027 void (*close)(struct sock *sk,
1028 long timeout);
1029 int (*pre_connect)(struct sock *sk,
1030 struct sockaddr *uaddr,
1031 int addr_len);
1032 int (*connect)(struct sock *sk,
1033 struct sockaddr *uaddr,
1034 int addr_len);
1035 int (*disconnect)(struct sock *sk, int flags);
1036
1037 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1038 bool kern);
1039
1040 int (*ioctl)(struct sock *sk, int cmd,
1041 unsigned long arg);
1042 int (*init)(struct sock *sk);
1043 void (*destroy)(struct sock *sk);
1044 void (*shutdown)(struct sock *sk, int how);
1045 int (*setsockopt)(struct sock *sk, int level,
1046 int optname, char __user *optval,
1047 unsigned int optlen);
1048 int (*getsockopt)(struct sock *sk, int level,
1049 int optname, char __user *optval,
1050 int __user *option);
1051 void (*keepalive)(struct sock *sk, int valbool);
1052#ifdef CONFIG_COMPAT
1053 int (*compat_setsockopt)(struct sock *sk,
1054 int level,
1055 int optname, char __user *optval,
1056 unsigned int optlen);
1057 int (*compat_getsockopt)(struct sock *sk,
1058 int level,
1059 int optname, char __user *optval,
1060 int __user *option);
1061 int (*compat_ioctl)(struct sock *sk,
1062 unsigned int cmd, unsigned long arg);
1063#endif
1064 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1065 size_t len);
1066 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1067 size_t len, int noblock, int flags,
1068 int *addr_len);
1069 int (*sendpage)(struct sock *sk, struct page *page,
1070 int offset, size_t size, int flags);
1071 int (*bind)(struct sock *sk,
1072 struct sockaddr *uaddr, int addr_len);
1073
1074 int (*backlog_rcv) (struct sock *sk,
1075 struct sk_buff *skb);
1076
1077 void (*release_cb)(struct sock *sk);
1078
1079 /* Keeping track of sk's, looking them up, and port selection methods. */
1080 int (*hash)(struct sock *sk);
1081 void (*unhash)(struct sock *sk);
1082 void (*rehash)(struct sock *sk);
1083 int (*get_port)(struct sock *sk, unsigned short snum);
1084
1085 /* Keeping track of sockets in use */
1086#ifdef CONFIG_PROC_FS
1087 unsigned int inuse_idx;
1088#endif
1089
1090 bool (*stream_memory_free)(const struct sock *sk);
1091 bool (*stream_memory_read)(const struct sock *sk);
1092 /* Memory pressure */
1093 void (*enter_memory_pressure)(struct sock *sk);
1094 void (*leave_memory_pressure)(struct sock *sk);
1095 atomic_long_t *memory_allocated; /* Current allocated memory. */
1096 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1097 /*
1098 * Pressure flag: try to collapse.
1099 * Technical note: it is used by multiple contexts non atomically.
1100 * All the __sk_mem_schedule() is of this nature: accounting
1101 * is strict, actions are advisory and have some latency.
1102 */
1103 unsigned long *memory_pressure;
1104 long *sysctl_mem;
1105
1106 int *sysctl_wmem;
1107 int *sysctl_rmem;
1108 u32 sysctl_wmem_offset;
1109 u32 sysctl_rmem_offset;
1110
1111 int max_header;
1112 bool no_autobind;
1113
1114 struct kmem_cache *slab;
1115 unsigned int obj_size;
1116 slab_flags_t slab_flags;
1117 unsigned int useroffset; /* Usercopy region offset */
1118 unsigned int usersize; /* Usercopy region size */
1119
1120 struct percpu_counter *orphan_count;
1121
1122 struct request_sock_ops *rsk_prot;
1123 struct timewait_sock_ops *twsk_prot;
1124
1125 union {
1126 struct inet_hashinfo *hashinfo;
1127 struct udp_table *udp_table;
1128 struct raw_hashinfo *raw_hash;
1129 struct smc_hashinfo *smc_hash;
1130 } h;
1131
1132 struct module *owner;
1133
1134 char name[32];
1135
1136 struct list_head node;
1137#ifdef SOCK_REFCNT_DEBUG
1138 atomic_t socks;
1139#endif
1140 int (*diag_destroy)(struct sock *sk, int err);
1141} __randomize_layout;
1142
1143int proto_register(struct proto *prot, int alloc_slab);
1144void proto_unregister(struct proto *prot);
1145int sock_load_diag_module(int family, int protocol);
1146
1147#ifdef SOCK_REFCNT_DEBUG
1148static inline void sk_refcnt_debug_inc(struct sock *sk)
1149{
1150 atomic_inc(&sk->sk_prot->socks);
1151}
1152
1153static inline void sk_refcnt_debug_dec(struct sock *sk)
1154{
1155 atomic_dec(&sk->sk_prot->socks);
1156 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1157 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1158}
1159
1160static inline void sk_refcnt_debug_release(const struct sock *sk)
1161{
1162 if (refcount_read(&sk->sk_refcnt) != 1)
1163 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1164 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1165}
1166#else /* SOCK_REFCNT_DEBUG */
1167#define sk_refcnt_debug_inc(sk) do { } while (0)
1168#define sk_refcnt_debug_dec(sk) do { } while (0)
1169#define sk_refcnt_debug_release(sk) do { } while (0)
1170#endif /* SOCK_REFCNT_DEBUG */
1171
1172static inline bool sk_stream_memory_free(const struct sock *sk)
1173{
1174 if (sk->sk_wmem_queued >= sk->sk_sndbuf)
1175 return false;
1176
1177 return sk->sk_prot->stream_memory_free ?
1178 sk->sk_prot->stream_memory_free(sk) : true;
1179}
1180
1181static inline bool sk_stream_is_writeable(const struct sock *sk)
1182{
1183 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1184 sk_stream_memory_free(sk);
1185}
1186
1187static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1188 struct cgroup *ancestor)
1189{
1190#ifdef CONFIG_SOCK_CGROUP_DATA
1191 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1192 ancestor);
1193#else
1194 return -ENOTSUPP;
1195#endif
1196}
1197
1198static inline bool sk_has_memory_pressure(const struct sock *sk)
1199{
1200 return sk->sk_prot->memory_pressure != NULL;
1201}
1202
1203static inline bool sk_under_memory_pressure(const struct sock *sk)
1204{
1205 if (!sk->sk_prot->memory_pressure)
1206 return false;
1207
1208 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1209 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1210 return true;
1211
1212 return !!*sk->sk_prot->memory_pressure;
1213}
1214
1215static inline long
1216sk_memory_allocated(const struct sock *sk)
1217{
1218 return atomic_long_read(sk->sk_prot->memory_allocated);
1219}
1220
1221static inline long
1222sk_memory_allocated_add(struct sock *sk, int amt)
1223{
1224 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1225}
1226
1227static inline void
1228sk_memory_allocated_sub(struct sock *sk, int amt)
1229{
1230 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1231}
1232
1233static inline void sk_sockets_allocated_dec(struct sock *sk)
1234{
1235 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1236}
1237
1238static inline void sk_sockets_allocated_inc(struct sock *sk)
1239{
1240 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1241}
1242
1243static inline int
1244sk_sockets_allocated_read_positive(struct sock *sk)
1245{
1246 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1247}
1248
1249static inline int
1250proto_sockets_allocated_sum_positive(struct proto *prot)
1251{
1252 return percpu_counter_sum_positive(prot->sockets_allocated);
1253}
1254
1255static inline long
1256proto_memory_allocated(struct proto *prot)
1257{
1258 return atomic_long_read(prot->memory_allocated);
1259}
1260
1261static inline bool
1262proto_memory_pressure(struct proto *prot)
1263{
1264 if (!prot->memory_pressure)
1265 return false;
1266 return !!*prot->memory_pressure;
1267}
1268
1269
1270#ifdef CONFIG_PROC_FS
1271/* Called with local bh disabled */
1272void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1273int sock_prot_inuse_get(struct net *net, struct proto *proto);
1274int sock_inuse_get(struct net *net);
1275#else
1276static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1277 int inc)
1278{
1279}
1280#endif
1281
1282
1283/* With per-bucket locks this operation is not-atomic, so that
1284 * this version is not worse.
1285 */
1286static inline int __sk_prot_rehash(struct sock *sk)
1287{
1288 sk->sk_prot->unhash(sk);
1289 return sk->sk_prot->hash(sk);
1290}
1291
1292/* About 10 seconds */
1293#define SOCK_DESTROY_TIME (10*HZ)
1294
1295/* Sockets 0-1023 can't be bound to unless you are superuser */
1296#define PROT_SOCK 1024
1297
1298#define SHUTDOWN_MASK 3
1299#define RCV_SHUTDOWN 1
1300#define SEND_SHUTDOWN 2
1301
1302#define SOCK_SNDBUF_LOCK 1
1303#define SOCK_RCVBUF_LOCK 2
1304#define SOCK_BINDADDR_LOCK 4
1305#define SOCK_BINDPORT_LOCK 8
1306
1307struct socket_alloc {
1308 struct socket socket;
1309 struct inode vfs_inode;
1310};
1311
1312static inline struct socket *SOCKET_I(struct inode *inode)
1313{
1314 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1315}
1316
1317static inline struct inode *SOCK_INODE(struct socket *socket)
1318{
1319 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1320}
1321
1322/*
1323 * Functions for memory accounting
1324 */
1325int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1326int __sk_mem_schedule(struct sock *sk, int size, int kind);
1327void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1328void __sk_mem_reclaim(struct sock *sk, int amount);
1329
1330/* We used to have PAGE_SIZE here, but systems with 64KB pages
1331 * do not necessarily have 16x time more memory than 4KB ones.
1332 */
1333#define SK_MEM_QUANTUM 4096
1334#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1335#define SK_MEM_SEND 0
1336#define SK_MEM_RECV 1
1337
1338/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1339static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1340{
1341 long val = sk->sk_prot->sysctl_mem[index];
1342
1343#if PAGE_SIZE > SK_MEM_QUANTUM
1344 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1345#elif PAGE_SIZE < SK_MEM_QUANTUM
1346 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1347#endif
1348 return val;
1349}
1350
1351static inline int sk_mem_pages(int amt)
1352{
1353 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1354}
1355
1356static inline bool sk_has_account(struct sock *sk)
1357{
1358 /* return true if protocol supports memory accounting */
1359 return !!sk->sk_prot->memory_allocated;
1360}
1361
1362static inline bool sk_wmem_schedule(struct sock *sk, int size)
1363{
1364 if (!sk_has_account(sk))
1365 return true;
1366 return size <= sk->sk_forward_alloc ||
1367 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1368}
1369
1370static inline bool
1371sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1372{
1373 if (!sk_has_account(sk))
1374 return true;
1375 return size<= sk->sk_forward_alloc ||
1376 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1377 skb_pfmemalloc(skb);
1378}
1379
1380static inline void sk_mem_reclaim(struct sock *sk)
1381{
1382 if (!sk_has_account(sk))
1383 return;
1384 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1385 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1386}
1387
1388static inline void sk_mem_reclaim_partial(struct sock *sk)
1389{
1390 if (!sk_has_account(sk))
1391 return;
1392 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1393 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1394}
1395
1396static inline void sk_mem_charge(struct sock *sk, int size)
1397{
1398 if (!sk_has_account(sk))
1399 return;
1400 sk->sk_forward_alloc -= size;
1401}
1402
1403static inline void sk_mem_uncharge(struct sock *sk, int size)
1404{
1405 if (!sk_has_account(sk))
1406 return;
1407 sk->sk_forward_alloc += size;
1408
1409 /* Avoid a possible overflow.
1410 * TCP send queues can make this happen, if sk_mem_reclaim()
1411 * is not called and more than 2 GBytes are released at once.
1412 *
1413 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1414 * no need to hold that much forward allocation anyway.
1415 */
1416 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1417 __sk_mem_reclaim(sk, 1 << 20);
1418}
1419
1420static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1421{
1422 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1423 sk->sk_wmem_queued -= skb->truesize;
1424 sk_mem_uncharge(sk, skb->truesize);
1425 __kfree_skb(skb);
1426}
1427
1428static inline void sock_release_ownership(struct sock *sk)
1429{
1430 if (sk->sk_lock.owned) {
1431 sk->sk_lock.owned = 0;
1432
1433 /* The sk_lock has mutex_unlock() semantics: */
1434 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1435 }
1436}
1437
1438/*
1439 * Macro so as to not evaluate some arguments when
1440 * lockdep is not enabled.
1441 *
1442 * Mark both the sk_lock and the sk_lock.slock as a
1443 * per-address-family lock class.
1444 */
1445#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1446do { \
1447 sk->sk_lock.owned = 0; \
1448 init_waitqueue_head(&sk->sk_lock.wq); \
1449 spin_lock_init(&(sk)->sk_lock.slock); \
1450 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1451 sizeof((sk)->sk_lock)); \
1452 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1453 (skey), (sname)); \
1454 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1455} while (0)
1456
1457#ifdef CONFIG_LOCKDEP
1458static inline bool lockdep_sock_is_held(const struct sock *sk)
1459{
1460 return lockdep_is_held(&sk->sk_lock) ||
1461 lockdep_is_held(&sk->sk_lock.slock);
1462}
1463#endif
1464
1465void lock_sock_nested(struct sock *sk, int subclass);
1466
1467static inline void lock_sock(struct sock *sk)
1468{
1469 lock_sock_nested(sk, 0);
1470}
1471
1472void release_sock(struct sock *sk);
1473
1474/* BH context may only use the following locking interface. */
1475#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1476#define bh_lock_sock_nested(__sk) \
1477 spin_lock_nested(&((__sk)->sk_lock.slock), \
1478 SINGLE_DEPTH_NESTING)
1479#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1480
1481bool lock_sock_fast(struct sock *sk);
1482/**
1483 * unlock_sock_fast - complement of lock_sock_fast
1484 * @sk: socket
1485 * @slow: slow mode
1486 *
1487 * fast unlock socket for user context.
1488 * If slow mode is on, we call regular release_sock()
1489 */
1490static inline void unlock_sock_fast(struct sock *sk, bool slow)
1491{
1492 if (slow)
1493 release_sock(sk);
1494 else
1495 spin_unlock_bh(&sk->sk_lock.slock);
1496}
1497
1498/* Used by processes to "lock" a socket state, so that
1499 * interrupts and bottom half handlers won't change it
1500 * from under us. It essentially blocks any incoming
1501 * packets, so that we won't get any new data or any
1502 * packets that change the state of the socket.
1503 *
1504 * While locked, BH processing will add new packets to
1505 * the backlog queue. This queue is processed by the
1506 * owner of the socket lock right before it is released.
1507 *
1508 * Since ~2.3.5 it is also exclusive sleep lock serializing
1509 * accesses from user process context.
1510 */
1511
1512static inline void sock_owned_by_me(const struct sock *sk)
1513{
1514#ifdef CONFIG_LOCKDEP
1515 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1516#endif
1517}
1518
1519static inline bool sock_owned_by_user(const struct sock *sk)
1520{
1521 sock_owned_by_me(sk);
1522 return sk->sk_lock.owned;
1523}
1524
1525static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1526{
1527 return sk->sk_lock.owned;
1528}
1529
1530/* no reclassification while locks are held */
1531static inline bool sock_allow_reclassification(const struct sock *csk)
1532{
1533 struct sock *sk = (struct sock *)csk;
1534
1535 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1536}
1537
1538struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1539 struct proto *prot, int kern);
1540void sk_free(struct sock *sk);
1541void sk_destruct(struct sock *sk);
1542struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1543void sk_free_unlock_clone(struct sock *sk);
1544
1545struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1546 gfp_t priority);
1547void __sock_wfree(struct sk_buff *skb);
1548void sock_wfree(struct sk_buff *skb);
1549struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1550 gfp_t priority);
1551void skb_orphan_partial(struct sk_buff *skb);
1552void sock_rfree(struct sk_buff *skb);
1553void sock_efree(struct sk_buff *skb);
1554#ifdef CONFIG_INET
1555void sock_edemux(struct sk_buff *skb);
1556#else
1557#define sock_edemux sock_efree
1558#endif
1559
1560int sock_setsockopt(struct socket *sock, int level, int op,
1561 char __user *optval, unsigned int optlen);
1562
1563int sock_getsockopt(struct socket *sock, int level, int op,
1564 char __user *optval, int __user *optlen);
1565struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1566 int noblock, int *errcode);
1567struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1568 unsigned long data_len, int noblock,
1569 int *errcode, int max_page_order);
1570void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1571void sock_kfree_s(struct sock *sk, void *mem, int size);
1572void sock_kzfree_s(struct sock *sk, void *mem, int size);
1573void sk_send_sigurg(struct sock *sk);
1574
1575struct sockcm_cookie {
1576 u32 mark;
1577 u16 tsflags;
1578};
1579
1580int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1581 struct sockcm_cookie *sockc);
1582int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1583 struct sockcm_cookie *sockc);
1584
1585/*
1586 * Functions to fill in entries in struct proto_ops when a protocol
1587 * does not implement a particular function.
1588 */
1589int sock_no_bind(struct socket *, struct sockaddr *, int);
1590int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1591int sock_no_socketpair(struct socket *, struct socket *);
1592int sock_no_accept(struct socket *, struct socket *, int, bool);
1593int sock_no_getname(struct socket *, struct sockaddr *, int);
1594__poll_t sock_no_poll(struct file *, struct socket *,
1595 struct poll_table_struct *);
1596int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1597int sock_no_listen(struct socket *, int);
1598int sock_no_shutdown(struct socket *, int);
1599int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1600int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1601int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1602int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1603int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1604int sock_no_mmap(struct file *file, struct socket *sock,
1605 struct vm_area_struct *vma);
1606ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1607 size_t size, int flags);
1608ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1609 int offset, size_t size, int flags);
1610
1611/*
1612 * Functions to fill in entries in struct proto_ops when a protocol
1613 * uses the inet style.
1614 */
1615int sock_common_getsockopt(struct socket *sock, int level, int optname,
1616 char __user *optval, int __user *optlen);
1617int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1618 int flags);
1619int sock_common_setsockopt(struct socket *sock, int level, int optname,
1620 char __user *optval, unsigned int optlen);
1621int compat_sock_common_getsockopt(struct socket *sock, int level,
1622 int optname, char __user *optval, int __user *optlen);
1623int compat_sock_common_setsockopt(struct socket *sock, int level,
1624 int optname, char __user *optval, unsigned int optlen);
1625
1626void sk_common_release(struct sock *sk);
1627
1628/*
1629 * Default socket callbacks and setup code
1630 */
1631
1632/* Initialise core socket variables */
1633void sock_init_data(struct socket *sock, struct sock *sk);
1634
1635/*
1636 * Socket reference counting postulates.
1637 *
1638 * * Each user of socket SHOULD hold a reference count.
1639 * * Each access point to socket (an hash table bucket, reference from a list,
1640 * running timer, skb in flight MUST hold a reference count.
1641 * * When reference count hits 0, it means it will never increase back.
1642 * * When reference count hits 0, it means that no references from
1643 * outside exist to this socket and current process on current CPU
1644 * is last user and may/should destroy this socket.
1645 * * sk_free is called from any context: process, BH, IRQ. When
1646 * it is called, socket has no references from outside -> sk_free
1647 * may release descendant resources allocated by the socket, but
1648 * to the time when it is called, socket is NOT referenced by any
1649 * hash tables, lists etc.
1650 * * Packets, delivered from outside (from network or from another process)
1651 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1652 * when they sit in queue. Otherwise, packets will leak to hole, when
1653 * socket is looked up by one cpu and unhasing is made by another CPU.
1654 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1655 * (leak to backlog). Packet socket does all the processing inside
1656 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1657 * use separate SMP lock, so that they are prone too.
1658 */
1659
1660/* Ungrab socket and destroy it, if it was the last reference. */
1661static inline void sock_put(struct sock *sk)
1662{
1663 if (refcount_dec_and_test(&sk->sk_refcnt))
1664 sk_free(sk);
1665}
1666/* Generic version of sock_put(), dealing with all sockets
1667 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1668 */
1669void sock_gen_put(struct sock *sk);
1670
1671int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1672 unsigned int trim_cap, bool refcounted);
1673static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1674 const int nested)
1675{
1676 return __sk_receive_skb(sk, skb, nested, 1, true);
1677}
1678
1679static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1680{
1681 sk->sk_tx_queue_mapping = tx_queue;
1682}
1683
1684static inline void sk_tx_queue_clear(struct sock *sk)
1685{
1686 sk->sk_tx_queue_mapping = -1;
1687}
1688
1689static inline int sk_tx_queue_get(const struct sock *sk)
1690{
1691 return sk ? sk->sk_tx_queue_mapping : -1;
1692}
1693
1694static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1695{
1696 sk_tx_queue_clear(sk);
1697 sk->sk_socket = sock;
1698}
1699
1700static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1701{
1702 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1703 return &rcu_dereference_raw(sk->sk_wq)->wait;
1704}
1705/* Detach socket from process context.
1706 * Announce socket dead, detach it from wait queue and inode.
1707 * Note that parent inode held reference count on this struct sock,
1708 * we do not release it in this function, because protocol
1709 * probably wants some additional cleanups or even continuing
1710 * to work with this socket (TCP).
1711 */
1712static inline void sock_orphan(struct sock *sk)
1713{
1714 write_lock_bh(&sk->sk_callback_lock);
1715 sock_set_flag(sk, SOCK_DEAD);
1716 sk_set_socket(sk, NULL);
1717 sk->sk_wq = NULL;
1718 write_unlock_bh(&sk->sk_callback_lock);
1719}
1720
1721static inline void sock_graft(struct sock *sk, struct socket *parent)
1722{
1723 WARN_ON(parent->sk);
1724 write_lock_bh(&sk->sk_callback_lock);
1725 sk->sk_wq = parent->wq;
1726 parent->sk = sk;
1727 sk_set_socket(sk, parent);
1728 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1729 security_sock_graft(sk, parent);
1730 write_unlock_bh(&sk->sk_callback_lock);
1731}
1732
1733kuid_t sock_i_uid(struct sock *sk);
1734unsigned long sock_i_ino(struct sock *sk);
1735
1736static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1737{
1738 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1739}
1740
1741static inline u32 net_tx_rndhash(void)
1742{
1743 u32 v = prandom_u32();
1744
1745 return v ?: 1;
1746}
1747
1748static inline void sk_set_txhash(struct sock *sk)
1749{
1750 sk->sk_txhash = net_tx_rndhash();
1751}
1752
1753static inline void sk_rethink_txhash(struct sock *sk)
1754{
1755 if (sk->sk_txhash)
1756 sk_set_txhash(sk);
1757}
1758
1759static inline struct dst_entry *
1760__sk_dst_get(struct sock *sk)
1761{
1762 return rcu_dereference_check(sk->sk_dst_cache,
1763 lockdep_sock_is_held(sk));
1764}
1765
1766static inline struct dst_entry *
1767sk_dst_get(struct sock *sk)
1768{
1769 struct dst_entry *dst;
1770
1771 rcu_read_lock();
1772 dst = rcu_dereference(sk->sk_dst_cache);
1773 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1774 dst = NULL;
1775 rcu_read_unlock();
1776 return dst;
1777}
1778
1779static inline void dst_negative_advice(struct sock *sk)
1780{
1781 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1782
1783 sk_rethink_txhash(sk);
1784
1785 if (dst && dst->ops->negative_advice) {
1786 ndst = dst->ops->negative_advice(dst);
1787
1788 if (ndst != dst) {
1789 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1790 sk_tx_queue_clear(sk);
1791 sk->sk_dst_pending_confirm = 0;
1792 }
1793 }
1794}
1795
1796static inline void
1797__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1798{
1799 struct dst_entry *old_dst;
1800
1801 sk_tx_queue_clear(sk);
1802 sk->sk_dst_pending_confirm = 0;
1803 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1804 lockdep_sock_is_held(sk));
1805 rcu_assign_pointer(sk->sk_dst_cache, dst);
1806 dst_release(old_dst);
1807}
1808
1809static inline void
1810sk_dst_set(struct sock *sk, struct dst_entry *dst)
1811{
1812 struct dst_entry *old_dst;
1813
1814 sk_tx_queue_clear(sk);
1815 sk->sk_dst_pending_confirm = 0;
1816 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1817 dst_release(old_dst);
1818}
1819
1820static inline void
1821__sk_dst_reset(struct sock *sk)
1822{
1823 __sk_dst_set(sk, NULL);
1824}
1825
1826static inline void
1827sk_dst_reset(struct sock *sk)
1828{
1829 sk_dst_set(sk, NULL);
1830}
1831
1832struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1833
1834struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1835
1836static inline void sk_dst_confirm(struct sock *sk)
1837{
1838 if (!sk->sk_dst_pending_confirm)
1839 sk->sk_dst_pending_confirm = 1;
1840}
1841
1842static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1843{
1844 if (skb_get_dst_pending_confirm(skb)) {
1845 struct sock *sk = skb->sk;
1846 unsigned long now = jiffies;
1847
1848 /* avoid dirtying neighbour */
1849 if (n->confirmed != now)
1850 n->confirmed = now;
1851 if (sk && sk->sk_dst_pending_confirm)
1852 sk->sk_dst_pending_confirm = 0;
1853 }
1854}
1855
1856bool sk_mc_loop(struct sock *sk);
1857
1858static inline bool sk_can_gso(const struct sock *sk)
1859{
1860 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1861}
1862
1863void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1864
1865static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1866{
1867 sk->sk_route_nocaps |= flags;
1868 sk->sk_route_caps &= ~flags;
1869}
1870
1871static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1872 struct iov_iter *from, char *to,
1873 int copy, int offset)
1874{
1875 if (skb->ip_summed == CHECKSUM_NONE) {
1876 __wsum csum = 0;
1877 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1878 return -EFAULT;
1879 skb->csum = csum_block_add(skb->csum, csum, offset);
1880 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1881 if (!copy_from_iter_full_nocache(to, copy, from))
1882 return -EFAULT;
1883 } else if (!copy_from_iter_full(to, copy, from))
1884 return -EFAULT;
1885
1886 return 0;
1887}
1888
1889static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1890 struct iov_iter *from, int copy)
1891{
1892 int err, offset = skb->len;
1893
1894 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
1895 copy, offset);
1896 if (err)
1897 __skb_trim(skb, offset);
1898
1899 return err;
1900}
1901
1902static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
1903 struct sk_buff *skb,
1904 struct page *page,
1905 int off, int copy)
1906{
1907 int err;
1908
1909 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
1910 copy, skb->len);
1911 if (err)
1912 return err;
1913
1914 skb->len += copy;
1915 skb->data_len += copy;
1916 skb->truesize += copy;
1917 sk->sk_wmem_queued += copy;
1918 sk_mem_charge(sk, copy);
1919 return 0;
1920}
1921
1922/**
1923 * sk_wmem_alloc_get - returns write allocations
1924 * @sk: socket
1925 *
1926 * Returns sk_wmem_alloc minus initial offset of one
1927 */
1928static inline int sk_wmem_alloc_get(const struct sock *sk)
1929{
1930 return refcount_read(&sk->sk_wmem_alloc) - 1;
1931}
1932
1933/**
1934 * sk_rmem_alloc_get - returns read allocations
1935 * @sk: socket
1936 *
1937 * Returns sk_rmem_alloc
1938 */
1939static inline int sk_rmem_alloc_get(const struct sock *sk)
1940{
1941 return atomic_read(&sk->sk_rmem_alloc);
1942}
1943
1944/**
1945 * sk_has_allocations - check if allocations are outstanding
1946 * @sk: socket
1947 *
1948 * Returns true if socket has write or read allocations
1949 */
1950static inline bool sk_has_allocations(const struct sock *sk)
1951{
1952 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
1953}
1954
1955/**
1956 * skwq_has_sleeper - check if there are any waiting processes
1957 * @wq: struct socket_wq
1958 *
1959 * Returns true if socket_wq has waiting processes
1960 *
1961 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
1962 * barrier call. They were added due to the race found within the tcp code.
1963 *
1964 * Consider following tcp code paths::
1965 *
1966 * CPU1 CPU2
1967 * sys_select receive packet
1968 * ... ...
1969 * __add_wait_queue update tp->rcv_nxt
1970 * ... ...
1971 * tp->rcv_nxt check sock_def_readable
1972 * ... {
1973 * schedule rcu_read_lock();
1974 * wq = rcu_dereference(sk->sk_wq);
1975 * if (wq && waitqueue_active(&wq->wait))
1976 * wake_up_interruptible(&wq->wait)
1977 * ...
1978 * }
1979 *
1980 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
1981 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
1982 * could then endup calling schedule and sleep forever if there are no more
1983 * data on the socket.
1984 *
1985 */
1986static inline bool skwq_has_sleeper(struct socket_wq *wq)
1987{
1988 return wq && wq_has_sleeper(&wq->wait);
1989}
1990
1991/**
1992 * sock_poll_wait - place memory barrier behind the poll_wait call.
1993 * @filp: file
1994 * @wait_address: socket wait queue
1995 * @p: poll_table
1996 *
1997 * See the comments in the wq_has_sleeper function.
1998 */
1999static inline void sock_poll_wait(struct file *filp,
2000 wait_queue_head_t *wait_address, poll_table *p)
2001{
2002 if (!poll_does_not_wait(p) && wait_address) {
2003 poll_wait(filp, wait_address, p);
2004 /* We need to be sure we are in sync with the
2005 * socket flags modification.
2006 *
2007 * This memory barrier is paired in the wq_has_sleeper.
2008 */
2009 smp_mb();
2010 }
2011}
2012
2013static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2014{
2015 if (sk->sk_txhash) {
2016 skb->l4_hash = 1;
2017 skb->hash = sk->sk_txhash;
2018 }
2019}
2020
2021void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2022
2023/*
2024 * Queue a received datagram if it will fit. Stream and sequenced
2025 * protocols can't normally use this as they need to fit buffers in
2026 * and play with them.
2027 *
2028 * Inlined as it's very short and called for pretty much every
2029 * packet ever received.
2030 */
2031static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2032{
2033 skb_orphan(skb);
2034 skb->sk = sk;
2035 skb->destructor = sock_rfree;
2036 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2037 sk_mem_charge(sk, skb->truesize);
2038}
2039
2040void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2041 unsigned long expires);
2042
2043void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2044
2045int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2046 struct sk_buff *skb, unsigned int flags,
2047 void (*destructor)(struct sock *sk,
2048 struct sk_buff *skb));
2049int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2050int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2051
2052int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2053struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2054
2055/*
2056 * Recover an error report and clear atomically
2057 */
2058
2059static inline int sock_error(struct sock *sk)
2060{
2061 int err;
2062 if (likely(!sk->sk_err))
2063 return 0;
2064 err = xchg(&sk->sk_err, 0);
2065 return -err;
2066}
2067
2068static inline unsigned long sock_wspace(struct sock *sk)
2069{
2070 int amt = 0;
2071
2072 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2073 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2074 if (amt < 0)
2075 amt = 0;
2076 }
2077 return amt;
2078}
2079
2080/* Note:
2081 * We use sk->sk_wq_raw, from contexts knowing this
2082 * pointer is not NULL and cannot disappear/change.
2083 */
2084static inline void sk_set_bit(int nr, struct sock *sk)
2085{
2086 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2087 !sock_flag(sk, SOCK_FASYNC))
2088 return;
2089
2090 set_bit(nr, &sk->sk_wq_raw->flags);
2091}
2092
2093static inline void sk_clear_bit(int nr, struct sock *sk)
2094{
2095 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2096 !sock_flag(sk, SOCK_FASYNC))
2097 return;
2098
2099 clear_bit(nr, &sk->sk_wq_raw->flags);
2100}
2101
2102static inline void sk_wake_async(const struct sock *sk, int how, int band)
2103{
2104 if (sock_flag(sk, SOCK_FASYNC)) {
2105 rcu_read_lock();
2106 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2107 rcu_read_unlock();
2108 }
2109}
2110
2111/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2112 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2113 * Note: for send buffers, TCP works better if we can build two skbs at
2114 * minimum.
2115 */
2116#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2117
2118#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2119#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2120
2121static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2122{
2123 if (!(sk->sk_userlocks & SOCK_SNDBUF_LOCK)) {
2124 sk->sk_sndbuf = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2125 sk->sk_sndbuf = max_t(u32, sk->sk_sndbuf, SOCK_MIN_SNDBUF);
2126 }
2127}
2128
2129struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2130 bool force_schedule);
2131
2132/**
2133 * sk_page_frag - return an appropriate page_frag
2134 * @sk: socket
2135 *
2136 * If socket allocation mode allows current thread to sleep, it means its
2137 * safe to use the per task page_frag instead of the per socket one.
2138 */
2139static inline struct page_frag *sk_page_frag(struct sock *sk)
2140{
2141 if (gfpflags_allow_blocking(sk->sk_allocation))
2142 return ¤t->task_frag;
2143
2144 return &sk->sk_frag;
2145}
2146
2147bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2148
2149int sk_alloc_sg(struct sock *sk, int len, struct scatterlist *sg,
2150 int sg_start, int *sg_curr, unsigned int *sg_size,
2151 int first_coalesce);
2152
2153/*
2154 * Default write policy as shown to user space via poll/select/SIGIO
2155 */
2156static inline bool sock_writeable(const struct sock *sk)
2157{
2158 return refcount_read(&sk->sk_wmem_alloc) < (sk->sk_sndbuf >> 1);
2159}
2160
2161static inline gfp_t gfp_any(void)
2162{
2163 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2164}
2165
2166static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2167{
2168 return noblock ? 0 : sk->sk_rcvtimeo;
2169}
2170
2171static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2172{
2173 return noblock ? 0 : sk->sk_sndtimeo;
2174}
2175
2176static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2177{
2178 return (waitall ? len : min_t(int, sk->sk_rcvlowat, len)) ? : 1;
2179}
2180
2181/* Alas, with timeout socket operations are not restartable.
2182 * Compare this to poll().
2183 */
2184static inline int sock_intr_errno(long timeo)
2185{
2186 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2187}
2188
2189struct sock_skb_cb {
2190 u32 dropcount;
2191};
2192
2193/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2194 * using skb->cb[] would keep using it directly and utilize its
2195 * alignement guarantee.
2196 */
2197#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2198 sizeof(struct sock_skb_cb)))
2199
2200#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2201 SOCK_SKB_CB_OFFSET))
2202
2203#define sock_skb_cb_check_size(size) \
2204 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2205
2206static inline void
2207sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2208{
2209 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2210 atomic_read(&sk->sk_drops) : 0;
2211}
2212
2213static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2214{
2215 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2216
2217 atomic_add(segs, &sk->sk_drops);
2218}
2219
2220void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2221 struct sk_buff *skb);
2222void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2223 struct sk_buff *skb);
2224
2225static inline void
2226sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2227{
2228 ktime_t kt = skb->tstamp;
2229 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2230
2231 /*
2232 * generate control messages if
2233 * - receive time stamping in software requested
2234 * - software time stamp available and wanted
2235 * - hardware time stamps available and wanted
2236 */
2237 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2238 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2239 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2240 (hwtstamps->hwtstamp &&
2241 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2242 __sock_recv_timestamp(msg, sk, skb);
2243 else
2244 sk->sk_stamp = kt;
2245
2246 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2247 __sock_recv_wifi_status(msg, sk, skb);
2248}
2249
2250void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2251 struct sk_buff *skb);
2252
2253#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2254static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2255 struct sk_buff *skb)
2256{
2257#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2258 (1UL << SOCK_RCVTSTAMP))
2259#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2260 SOF_TIMESTAMPING_RAW_HARDWARE)
2261
2262 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2263 __sock_recv_ts_and_drops(msg, sk, skb);
2264 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2265 sk->sk_stamp = skb->tstamp;
2266 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2267 sk->sk_stamp = 0;
2268}
2269
2270void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2271
2272/**
2273 * sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2274 * @sk: socket sending this packet
2275 * @tsflags: timestamping flags to use
2276 * @tx_flags: completed with instructions for time stamping
2277 *
2278 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2279 */
2280static inline void sock_tx_timestamp(const struct sock *sk, __u16 tsflags,
2281 __u8 *tx_flags)
2282{
2283 if (unlikely(tsflags))
2284 __sock_tx_timestamp(tsflags, tx_flags);
2285 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2286 *tx_flags |= SKBTX_WIFI_STATUS;
2287}
2288
2289/**
2290 * sk_eat_skb - Release a skb if it is no longer needed
2291 * @sk: socket to eat this skb from
2292 * @skb: socket buffer to eat
2293 *
2294 * This routine must be called with interrupts disabled or with the socket
2295 * locked so that the sk_buff queue operation is ok.
2296*/
2297static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2298{
2299 __skb_unlink(skb, &sk->sk_receive_queue);
2300 __kfree_skb(skb);
2301}
2302
2303static inline
2304struct net *sock_net(const struct sock *sk)
2305{
2306 return read_pnet(&sk->sk_net);
2307}
2308
2309static inline
2310void sock_net_set(struct sock *sk, struct net *net)
2311{
2312 write_pnet(&sk->sk_net, net);
2313}
2314
2315static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2316{
2317 if (skb->sk) {
2318 struct sock *sk = skb->sk;
2319
2320 skb->destructor = NULL;
2321 skb->sk = NULL;
2322 return sk;
2323 }
2324 return NULL;
2325}
2326
2327/* This helper checks if a socket is a full socket,
2328 * ie _not_ a timewait or request socket.
2329 */
2330static inline bool sk_fullsock(const struct sock *sk)
2331{
2332 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2333}
2334
2335/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2336 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2337 */
2338static inline bool sk_listener(const struct sock *sk)
2339{
2340 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2341}
2342
2343void sock_enable_timestamp(struct sock *sk, int flag);
2344int sock_get_timestamp(struct sock *, struct timeval __user *);
2345int sock_get_timestampns(struct sock *, struct timespec __user *);
2346int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2347 int type);
2348
2349bool sk_ns_capable(const struct sock *sk,
2350 struct user_namespace *user_ns, int cap);
2351bool sk_capable(const struct sock *sk, int cap);
2352bool sk_net_capable(const struct sock *sk, int cap);
2353
2354void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2355
2356/* Take into consideration the size of the struct sk_buff overhead in the
2357 * determination of these values, since that is non-constant across
2358 * platforms. This makes socket queueing behavior and performance
2359 * not depend upon such differences.
2360 */
2361#define _SK_MEM_PACKETS 256
2362#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2363#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2364#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2365
2366extern __u32 sysctl_wmem_max;
2367extern __u32 sysctl_rmem_max;
2368
2369extern int sysctl_tstamp_allow_data;
2370extern int sysctl_optmem_max;
2371
2372extern __u32 sysctl_wmem_default;
2373extern __u32 sysctl_rmem_default;
2374
2375static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2376{
2377 /* Does this proto have per netns sysctl_wmem ? */
2378 if (proto->sysctl_wmem_offset)
2379 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2380
2381 return *proto->sysctl_wmem;
2382}
2383
2384static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2385{
2386 /* Does this proto have per netns sysctl_rmem ? */
2387 if (proto->sysctl_rmem_offset)
2388 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2389
2390 return *proto->sysctl_rmem;
2391}
2392
2393/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2394 * Some wifi drivers need to tweak it to get more chunks.
2395 * They can use this helper from their ndo_start_xmit()
2396 */
2397static inline void sk_pacing_shift_update(struct sock *sk, int val)
2398{
2399 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2400 return;
2401 sk->sk_pacing_shift = val;
2402}
2403
2404/* if a socket is bound to a device, check that the given device
2405 * index is either the same or that the socket is bound to an L3
2406 * master device and the given device index is also enslaved to
2407 * that L3 master
2408 */
2409static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2410{
2411 int mdif;
2412
2413 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2414 return true;
2415
2416 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2417 if (mdif && mdif == sk->sk_bound_dev_if)
2418 return true;
2419
2420 return false;
2421}
2422
2423#endif /* _SOCK_H */
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the AF_INET socket handler.
8 *
9 * Version: @(#)sock.h 1.0.4 05/13/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche <flla@stud.uni-sb.de>
15 *
16 * Fixes:
17 * Alan Cox : Volatiles in skbuff pointers. See
18 * skbuff comments. May be overdone,
19 * better to prove they can be removed
20 * than the reverse.
21 * Alan Cox : Added a zapped field for tcp to note
22 * a socket is reset and must stay shut up
23 * Alan Cox : New fields for options
24 * Pauline Middelink : identd support
25 * Alan Cox : Eliminate low level recv/recvfrom
26 * David S. Miller : New socket lookup architecture.
27 * Steve Whitehouse: Default routines for sock_ops
28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
29 * protinfo be just a void pointer, as the
30 * protocol specific parts were moved to
31 * respective headers and ipv4/v6, etc now
32 * use private slabcaches for its socks
33 * Pedro Hortas : New flags field for socket options
34 */
35#ifndef _SOCK_H
36#define _SOCK_H
37
38#include <linux/hardirq.h>
39#include <linux/kernel.h>
40#include <linux/list.h>
41#include <linux/list_nulls.h>
42#include <linux/timer.h>
43#include <linux/cache.h>
44#include <linux/bitops.h>
45#include <linux/lockdep.h>
46#include <linux/netdevice.h>
47#include <linux/skbuff.h> /* struct sk_buff */
48#include <linux/mm.h>
49#include <linux/security.h>
50#include <linux/slab.h>
51#include <linux/uaccess.h>
52#include <linux/page_counter.h>
53#include <linux/memcontrol.h>
54#include <linux/static_key.h>
55#include <linux/sched.h>
56#include <linux/wait.h>
57#include <linux/cgroup-defs.h>
58#include <linux/rbtree.h>
59#include <linux/rculist_nulls.h>
60#include <linux/poll.h>
61#include <linux/sockptr.h>
62#include <linux/indirect_call_wrapper.h>
63#include <linux/atomic.h>
64#include <linux/refcount.h>
65#include <linux/llist.h>
66#include <net/dst.h>
67#include <net/checksum.h>
68#include <net/tcp_states.h>
69#include <linux/net_tstamp.h>
70#include <net/l3mdev.h>
71#include <uapi/linux/socket.h>
72
73/*
74 * This structure really needs to be cleaned up.
75 * Most of it is for TCP, and not used by any of
76 * the other protocols.
77 */
78
79/* Define this to get the SOCK_DBG debugging facility. */
80#define SOCK_DEBUGGING
81#ifdef SOCK_DEBUGGING
82#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
83 printk(KERN_DEBUG msg); } while (0)
84#else
85/* Validate arguments and do nothing */
86static inline __printf(2, 3)
87void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
88{
89}
90#endif
91
92/* This is the per-socket lock. The spinlock provides a synchronization
93 * between user contexts and software interrupt processing, whereas the
94 * mini-semaphore synchronizes multiple users amongst themselves.
95 */
96typedef struct {
97 spinlock_t slock;
98 int owned;
99 wait_queue_head_t wq;
100 /*
101 * We express the mutex-alike socket_lock semantics
102 * to the lock validator by explicitly managing
103 * the slock as a lock variant (in addition to
104 * the slock itself):
105 */
106#ifdef CONFIG_DEBUG_LOCK_ALLOC
107 struct lockdep_map dep_map;
108#endif
109} socket_lock_t;
110
111struct sock;
112struct proto;
113struct net;
114
115typedef __u32 __bitwise __portpair;
116typedef __u64 __bitwise __addrpair;
117
118/**
119 * struct sock_common - minimal network layer representation of sockets
120 * @skc_daddr: Foreign IPv4 addr
121 * @skc_rcv_saddr: Bound local IPv4 addr
122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
123 * @skc_hash: hash value used with various protocol lookup tables
124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
125 * @skc_dport: placeholder for inet_dport/tw_dport
126 * @skc_num: placeholder for inet_num/tw_num
127 * @skc_portpair: __u32 union of @skc_dport & @skc_num
128 * @skc_family: network address family
129 * @skc_state: Connection state
130 * @skc_reuse: %SO_REUSEADDR setting
131 * @skc_reuseport: %SO_REUSEPORT setting
132 * @skc_ipv6only: socket is IPV6 only
133 * @skc_net_refcnt: socket is using net ref counting
134 * @skc_bound_dev_if: bound device index if != 0
135 * @skc_bind_node: bind hash linkage for various protocol lookup tables
136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137 * @skc_prot: protocol handlers inside a network family
138 * @skc_net: reference to the network namespace of this socket
139 * @skc_v6_daddr: IPV6 destination address
140 * @skc_v6_rcv_saddr: IPV6 source address
141 * @skc_cookie: socket's cookie value
142 * @skc_node: main hash linkage for various protocol lookup tables
143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
144 * @skc_tx_queue_mapping: tx queue number for this connection
145 * @skc_rx_queue_mapping: rx queue number for this connection
146 * @skc_flags: place holder for sk_flags
147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
149 * @skc_listener: connection request listener socket (aka rsk_listener)
150 * [union with @skc_flags]
151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
152 * [union with @skc_flags]
153 * @skc_incoming_cpu: record/match cpu processing incoming packets
154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
155 * [union with @skc_incoming_cpu]
156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
157 * [union with @skc_incoming_cpu]
158 * @skc_refcnt: reference count
159 *
160 * This is the minimal network layer representation of sockets, the header
161 * for struct sock and struct inet_timewait_sock.
162 */
163struct sock_common {
164 union {
165 __addrpair skc_addrpair;
166 struct {
167 __be32 skc_daddr;
168 __be32 skc_rcv_saddr;
169 };
170 };
171 union {
172 unsigned int skc_hash;
173 __u16 skc_u16hashes[2];
174 };
175 /* skc_dport && skc_num must be grouped as well */
176 union {
177 __portpair skc_portpair;
178 struct {
179 __be16 skc_dport;
180 __u16 skc_num;
181 };
182 };
183
184 unsigned short skc_family;
185 volatile unsigned char skc_state;
186 unsigned char skc_reuse:4;
187 unsigned char skc_reuseport:1;
188 unsigned char skc_ipv6only:1;
189 unsigned char skc_net_refcnt:1;
190 int skc_bound_dev_if;
191 union {
192 struct hlist_node skc_bind_node;
193 struct hlist_node skc_portaddr_node;
194 };
195 struct proto *skc_prot;
196 possible_net_t skc_net;
197
198#if IS_ENABLED(CONFIG_IPV6)
199 struct in6_addr skc_v6_daddr;
200 struct in6_addr skc_v6_rcv_saddr;
201#endif
202
203 atomic64_t skc_cookie;
204
205 /* following fields are padding to force
206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
207 * assuming IPV6 is enabled. We use this padding differently
208 * for different kind of 'sockets'
209 */
210 union {
211 unsigned long skc_flags;
212 struct sock *skc_listener; /* request_sock */
213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
214 };
215 /*
216 * fields between dontcopy_begin/dontcopy_end
217 * are not copied in sock_copy()
218 */
219 /* private: */
220 int skc_dontcopy_begin[0];
221 /* public: */
222 union {
223 struct hlist_node skc_node;
224 struct hlist_nulls_node skc_nulls_node;
225 };
226 unsigned short skc_tx_queue_mapping;
227#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
228 unsigned short skc_rx_queue_mapping;
229#endif
230 union {
231 int skc_incoming_cpu;
232 u32 skc_rcv_wnd;
233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
234 };
235
236 refcount_t skc_refcnt;
237 /* private: */
238 int skc_dontcopy_end[0];
239 union {
240 u32 skc_rxhash;
241 u32 skc_window_clamp;
242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
243 };
244 /* public: */
245};
246
247struct bpf_local_storage;
248struct sk_filter;
249
250/**
251 * struct sock - network layer representation of sockets
252 * @__sk_common: shared layout with inet_timewait_sock
253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
255 * @sk_lock: synchronizer
256 * @sk_kern_sock: True if sock is using kernel lock classes
257 * @sk_rcvbuf: size of receive buffer in bytes
258 * @sk_wq: sock wait queue and async head
259 * @sk_rx_dst: receive input route used by early demux
260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
262 * @sk_dst_cache: destination cache
263 * @sk_dst_pending_confirm: need to confirm neighbour
264 * @sk_policy: flow policy
265 * @sk_receive_queue: incoming packets
266 * @sk_wmem_alloc: transmit queue bytes committed
267 * @sk_tsq_flags: TCP Small Queues flags
268 * @sk_write_queue: Packet sending queue
269 * @sk_omem_alloc: "o" is "option" or "other"
270 * @sk_wmem_queued: persistent queue size
271 * @sk_forward_alloc: space allocated forward
272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket
273 * @sk_napi_id: id of the last napi context to receive data for sk
274 * @sk_ll_usec: usecs to busypoll when there is no data
275 * @sk_allocation: allocation mode
276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
279 * @sk_sndbuf: size of send buffer in bytes
280 * @__sk_flags_offset: empty field used to determine location of bitfield
281 * @sk_padding: unused element for alignment
282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
283 * @sk_no_check_rx: allow zero checksum in RX packets
284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
287 * @sk_gso_max_size: Maximum GSO segment size to build
288 * @sk_gso_max_segs: Maximum number of GSO segments
289 * @sk_pacing_shift: scaling factor for TCP Small Queues
290 * @sk_lingertime: %SO_LINGER l_linger setting
291 * @sk_backlog: always used with the per-socket spinlock held
292 * @sk_callback_lock: used with the callbacks in the end of this struct
293 * @sk_error_queue: rarely used
294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
295 * IPV6_ADDRFORM for instance)
296 * @sk_err: last error
297 * @sk_err_soft: errors that don't cause failure but are the cause of a
298 * persistent failure not just 'timed out'
299 * @sk_drops: raw/udp drops counter
300 * @sk_ack_backlog: current listen backlog
301 * @sk_max_ack_backlog: listen backlog set in listen()
302 * @sk_uid: user id of owner
303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
304 * @sk_busy_poll_budget: napi processing budget when busypolling
305 * @sk_priority: %SO_PRIORITY setting
306 * @sk_type: socket type (%SOCK_STREAM, etc)
307 * @sk_protocol: which protocol this socket belongs in this network family
308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
309 * @sk_peer_pid: &struct pid for this socket's peer
310 * @sk_peer_cred: %SO_PEERCRED setting
311 * @sk_rcvlowat: %SO_RCVLOWAT setting
312 * @sk_rcvtimeo: %SO_RCVTIMEO setting
313 * @sk_sndtimeo: %SO_SNDTIMEO setting
314 * @sk_txhash: computed flow hash for use on transmit
315 * @sk_txrehash: enable TX hash rethink
316 * @sk_filter: socket filtering instructions
317 * @sk_timer: sock cleanup timer
318 * @sk_stamp: time stamp of last packet received
319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
320 * @sk_tsflags: SO_TIMESTAMPING flags
321 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
322 * Sockets that can be used under memory reclaim should
323 * set this to false.
324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
325 * for timestamping
326 * @sk_tskey: counter to disambiguate concurrent tstamp requests
327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
328 * @sk_socket: Identd and reporting IO signals
329 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
330 * @sk_frag: cached page frag
331 * @sk_peek_off: current peek_offset value
332 * @sk_send_head: front of stuff to transmit
333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
334 * @sk_security: used by security modules
335 * @sk_mark: generic packet mark
336 * @sk_cgrp_data: cgroup data for this cgroup
337 * @sk_memcg: this socket's memory cgroup association
338 * @sk_write_pending: a write to stream socket waits to start
339 * @sk_state_change: callback to indicate change in the state of the sock
340 * @sk_data_ready: callback to indicate there is data to be processed
341 * @sk_write_space: callback to indicate there is bf sending space available
342 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
343 * @sk_backlog_rcv: callback to process the backlog
344 * @sk_validate_xmit_skb: ptr to an optional validate function
345 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
346 * @sk_reuseport_cb: reuseport group container
347 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
348 * @sk_rcu: used during RCU grace period
349 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
350 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
351 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
352 * @sk_txtime_unused: unused txtime flags
353 * @ns_tracker: tracker for netns reference
354 * @sk_bind2_node: bind node in the bhash2 table
355 */
356struct sock {
357 /*
358 * Now struct inet_timewait_sock also uses sock_common, so please just
359 * don't add nothing before this first member (__sk_common) --acme
360 */
361 struct sock_common __sk_common;
362#define sk_node __sk_common.skc_node
363#define sk_nulls_node __sk_common.skc_nulls_node
364#define sk_refcnt __sk_common.skc_refcnt
365#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
366#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
367#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
368#endif
369
370#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
371#define sk_dontcopy_end __sk_common.skc_dontcopy_end
372#define sk_hash __sk_common.skc_hash
373#define sk_portpair __sk_common.skc_portpair
374#define sk_num __sk_common.skc_num
375#define sk_dport __sk_common.skc_dport
376#define sk_addrpair __sk_common.skc_addrpair
377#define sk_daddr __sk_common.skc_daddr
378#define sk_rcv_saddr __sk_common.skc_rcv_saddr
379#define sk_family __sk_common.skc_family
380#define sk_state __sk_common.skc_state
381#define sk_reuse __sk_common.skc_reuse
382#define sk_reuseport __sk_common.skc_reuseport
383#define sk_ipv6only __sk_common.skc_ipv6only
384#define sk_net_refcnt __sk_common.skc_net_refcnt
385#define sk_bound_dev_if __sk_common.skc_bound_dev_if
386#define sk_bind_node __sk_common.skc_bind_node
387#define sk_prot __sk_common.skc_prot
388#define sk_net __sk_common.skc_net
389#define sk_v6_daddr __sk_common.skc_v6_daddr
390#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
391#define sk_cookie __sk_common.skc_cookie
392#define sk_incoming_cpu __sk_common.skc_incoming_cpu
393#define sk_flags __sk_common.skc_flags
394#define sk_rxhash __sk_common.skc_rxhash
395
396 /* early demux fields */
397 struct dst_entry __rcu *sk_rx_dst;
398 int sk_rx_dst_ifindex;
399 u32 sk_rx_dst_cookie;
400
401 socket_lock_t sk_lock;
402 atomic_t sk_drops;
403 int sk_rcvlowat;
404 struct sk_buff_head sk_error_queue;
405 struct sk_buff_head sk_receive_queue;
406 /*
407 * The backlog queue is special, it is always used with
408 * the per-socket spinlock held and requires low latency
409 * access. Therefore we special case it's implementation.
410 * Note : rmem_alloc is in this structure to fill a hole
411 * on 64bit arches, not because its logically part of
412 * backlog.
413 */
414 struct {
415 atomic_t rmem_alloc;
416 int len;
417 struct sk_buff *head;
418 struct sk_buff *tail;
419 } sk_backlog;
420
421#define sk_rmem_alloc sk_backlog.rmem_alloc
422
423 int sk_forward_alloc;
424 u32 sk_reserved_mem;
425#ifdef CONFIG_NET_RX_BUSY_POLL
426 unsigned int sk_ll_usec;
427 /* ===== mostly read cache line ===== */
428 unsigned int sk_napi_id;
429#endif
430 int sk_rcvbuf;
431
432 struct sk_filter __rcu *sk_filter;
433 union {
434 struct socket_wq __rcu *sk_wq;
435 /* private: */
436 struct socket_wq *sk_wq_raw;
437 /* public: */
438 };
439#ifdef CONFIG_XFRM
440 struct xfrm_policy __rcu *sk_policy[2];
441#endif
442
443 struct dst_entry __rcu *sk_dst_cache;
444 atomic_t sk_omem_alloc;
445 int sk_sndbuf;
446
447 /* ===== cache line for TX ===== */
448 int sk_wmem_queued;
449 refcount_t sk_wmem_alloc;
450 unsigned long sk_tsq_flags;
451 union {
452 struct sk_buff *sk_send_head;
453 struct rb_root tcp_rtx_queue;
454 };
455 struct sk_buff_head sk_write_queue;
456 __s32 sk_peek_off;
457 int sk_write_pending;
458 __u32 sk_dst_pending_confirm;
459 u32 sk_pacing_status; /* see enum sk_pacing */
460 long sk_sndtimeo;
461 struct timer_list sk_timer;
462 __u32 sk_priority;
463 __u32 sk_mark;
464 unsigned long sk_pacing_rate; /* bytes per second */
465 unsigned long sk_max_pacing_rate;
466 struct page_frag sk_frag;
467 netdev_features_t sk_route_caps;
468 int sk_gso_type;
469 unsigned int sk_gso_max_size;
470 gfp_t sk_allocation;
471 __u32 sk_txhash;
472
473 /*
474 * Because of non atomicity rules, all
475 * changes are protected by socket lock.
476 */
477 u8 sk_gso_disabled : 1,
478 sk_kern_sock : 1,
479 sk_no_check_tx : 1,
480 sk_no_check_rx : 1,
481 sk_userlocks : 4;
482 u8 sk_pacing_shift;
483 u16 sk_type;
484 u16 sk_protocol;
485 u16 sk_gso_max_segs;
486 unsigned long sk_lingertime;
487 struct proto *sk_prot_creator;
488 rwlock_t sk_callback_lock;
489 int sk_err,
490 sk_err_soft;
491 u32 sk_ack_backlog;
492 u32 sk_max_ack_backlog;
493 kuid_t sk_uid;
494 u8 sk_txrehash;
495#ifdef CONFIG_NET_RX_BUSY_POLL
496 u8 sk_prefer_busy_poll;
497 u16 sk_busy_poll_budget;
498#endif
499 spinlock_t sk_peer_lock;
500 int sk_bind_phc;
501 struct pid *sk_peer_pid;
502 const struct cred *sk_peer_cred;
503
504 long sk_rcvtimeo;
505 ktime_t sk_stamp;
506#if BITS_PER_LONG==32
507 seqlock_t sk_stamp_seq;
508#endif
509 atomic_t sk_tskey;
510 atomic_t sk_zckey;
511 u32 sk_tsflags;
512 u8 sk_shutdown;
513
514 u8 sk_clockid;
515 u8 sk_txtime_deadline_mode : 1,
516 sk_txtime_report_errors : 1,
517 sk_txtime_unused : 6;
518 bool sk_use_task_frag;
519
520 struct socket *sk_socket;
521 void *sk_user_data;
522#ifdef CONFIG_SECURITY
523 void *sk_security;
524#endif
525 struct sock_cgroup_data sk_cgrp_data;
526 struct mem_cgroup *sk_memcg;
527 void (*sk_state_change)(struct sock *sk);
528 void (*sk_data_ready)(struct sock *sk);
529 void (*sk_write_space)(struct sock *sk);
530 void (*sk_error_report)(struct sock *sk);
531 int (*sk_backlog_rcv)(struct sock *sk,
532 struct sk_buff *skb);
533#ifdef CONFIG_SOCK_VALIDATE_XMIT
534 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
535 struct net_device *dev,
536 struct sk_buff *skb);
537#endif
538 void (*sk_destruct)(struct sock *sk);
539 struct sock_reuseport __rcu *sk_reuseport_cb;
540#ifdef CONFIG_BPF_SYSCALL
541 struct bpf_local_storage __rcu *sk_bpf_storage;
542#endif
543 struct rcu_head sk_rcu;
544 netns_tracker ns_tracker;
545 struct hlist_node sk_bind2_node;
546};
547
548enum sk_pacing {
549 SK_PACING_NONE = 0,
550 SK_PACING_NEEDED = 1,
551 SK_PACING_FQ = 2,
552};
553
554/* flag bits in sk_user_data
555 *
556 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
557 * not be suitable for copying when cloning the socket. For instance,
558 * it can point to a reference counted object. sk_user_data bottom
559 * bit is set if pointer must not be copied.
560 *
561 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
562 * managed/owned by a BPF reuseport array. This bit should be set
563 * when sk_user_data's sk is added to the bpf's reuseport_array.
564 *
565 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
566 * sk_user_data points to psock type. This bit should be set
567 * when sk_user_data is assigned to a psock object.
568 */
569#define SK_USER_DATA_NOCOPY 1UL
570#define SK_USER_DATA_BPF 2UL
571#define SK_USER_DATA_PSOCK 4UL
572#define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
573 SK_USER_DATA_PSOCK)
574
575/**
576 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
577 * @sk: socket
578 */
579static inline bool sk_user_data_is_nocopy(const struct sock *sk)
580{
581 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
582}
583
584#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
585
586/**
587 * __locked_read_sk_user_data_with_flags - return the pointer
588 * only if argument flags all has been set in sk_user_data. Otherwise
589 * return NULL
590 *
591 * @sk: socket
592 * @flags: flag bits
593 *
594 * The caller must be holding sk->sk_callback_lock.
595 */
596static inline void *
597__locked_read_sk_user_data_with_flags(const struct sock *sk,
598 uintptr_t flags)
599{
600 uintptr_t sk_user_data =
601 (uintptr_t)rcu_dereference_check(__sk_user_data(sk),
602 lockdep_is_held(&sk->sk_callback_lock));
603
604 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
605
606 if ((sk_user_data & flags) == flags)
607 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
608 return NULL;
609}
610
611/**
612 * __rcu_dereference_sk_user_data_with_flags - return the pointer
613 * only if argument flags all has been set in sk_user_data. Otherwise
614 * return NULL
615 *
616 * @sk: socket
617 * @flags: flag bits
618 */
619static inline void *
620__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
621 uintptr_t flags)
622{
623 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
624
625 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
626
627 if ((sk_user_data & flags) == flags)
628 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
629 return NULL;
630}
631
632#define rcu_dereference_sk_user_data(sk) \
633 __rcu_dereference_sk_user_data_with_flags(sk, 0)
634#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
635({ \
636 uintptr_t __tmp1 = (uintptr_t)(ptr), \
637 __tmp2 = (uintptr_t)(flags); \
638 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
639 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
640 rcu_assign_pointer(__sk_user_data((sk)), \
641 __tmp1 | __tmp2); \
642})
643#define rcu_assign_sk_user_data(sk, ptr) \
644 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
645
646static inline
647struct net *sock_net(const struct sock *sk)
648{
649 return read_pnet(&sk->sk_net);
650}
651
652static inline
653void sock_net_set(struct sock *sk, struct net *net)
654{
655 write_pnet(&sk->sk_net, net);
656}
657
658/*
659 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
660 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
661 * on a socket means that the socket will reuse everybody else's port
662 * without looking at the other's sk_reuse value.
663 */
664
665#define SK_NO_REUSE 0
666#define SK_CAN_REUSE 1
667#define SK_FORCE_REUSE 2
668
669int sk_set_peek_off(struct sock *sk, int val);
670
671static inline int sk_peek_offset(const struct sock *sk, int flags)
672{
673 if (unlikely(flags & MSG_PEEK)) {
674 return READ_ONCE(sk->sk_peek_off);
675 }
676
677 return 0;
678}
679
680static inline void sk_peek_offset_bwd(struct sock *sk, int val)
681{
682 s32 off = READ_ONCE(sk->sk_peek_off);
683
684 if (unlikely(off >= 0)) {
685 off = max_t(s32, off - val, 0);
686 WRITE_ONCE(sk->sk_peek_off, off);
687 }
688}
689
690static inline void sk_peek_offset_fwd(struct sock *sk, int val)
691{
692 sk_peek_offset_bwd(sk, -val);
693}
694
695/*
696 * Hashed lists helper routines
697 */
698static inline struct sock *sk_entry(const struct hlist_node *node)
699{
700 return hlist_entry(node, struct sock, sk_node);
701}
702
703static inline struct sock *__sk_head(const struct hlist_head *head)
704{
705 return hlist_entry(head->first, struct sock, sk_node);
706}
707
708static inline struct sock *sk_head(const struct hlist_head *head)
709{
710 return hlist_empty(head) ? NULL : __sk_head(head);
711}
712
713static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
714{
715 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
716}
717
718static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
719{
720 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
721}
722
723static inline struct sock *sk_next(const struct sock *sk)
724{
725 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
726}
727
728static inline struct sock *sk_nulls_next(const struct sock *sk)
729{
730 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
731 hlist_nulls_entry(sk->sk_nulls_node.next,
732 struct sock, sk_nulls_node) :
733 NULL;
734}
735
736static inline bool sk_unhashed(const struct sock *sk)
737{
738 return hlist_unhashed(&sk->sk_node);
739}
740
741static inline bool sk_hashed(const struct sock *sk)
742{
743 return !sk_unhashed(sk);
744}
745
746static inline void sk_node_init(struct hlist_node *node)
747{
748 node->pprev = NULL;
749}
750
751static inline void __sk_del_node(struct sock *sk)
752{
753 __hlist_del(&sk->sk_node);
754}
755
756/* NB: equivalent to hlist_del_init_rcu */
757static inline bool __sk_del_node_init(struct sock *sk)
758{
759 if (sk_hashed(sk)) {
760 __sk_del_node(sk);
761 sk_node_init(&sk->sk_node);
762 return true;
763 }
764 return false;
765}
766
767/* Grab socket reference count. This operation is valid only
768 when sk is ALREADY grabbed f.e. it is found in hash table
769 or a list and the lookup is made under lock preventing hash table
770 modifications.
771 */
772
773static __always_inline void sock_hold(struct sock *sk)
774{
775 refcount_inc(&sk->sk_refcnt);
776}
777
778/* Ungrab socket in the context, which assumes that socket refcnt
779 cannot hit zero, f.e. it is true in context of any socketcall.
780 */
781static __always_inline void __sock_put(struct sock *sk)
782{
783 refcount_dec(&sk->sk_refcnt);
784}
785
786static inline bool sk_del_node_init(struct sock *sk)
787{
788 bool rc = __sk_del_node_init(sk);
789
790 if (rc) {
791 /* paranoid for a while -acme */
792 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
793 __sock_put(sk);
794 }
795 return rc;
796}
797#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
798
799static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
800{
801 if (sk_hashed(sk)) {
802 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
803 return true;
804 }
805 return false;
806}
807
808static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
809{
810 bool rc = __sk_nulls_del_node_init_rcu(sk);
811
812 if (rc) {
813 /* paranoid for a while -acme */
814 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
815 __sock_put(sk);
816 }
817 return rc;
818}
819
820static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
821{
822 hlist_add_head(&sk->sk_node, list);
823}
824
825static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
826{
827 sock_hold(sk);
828 __sk_add_node(sk, list);
829}
830
831static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
832{
833 sock_hold(sk);
834 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
835 sk->sk_family == AF_INET6)
836 hlist_add_tail_rcu(&sk->sk_node, list);
837 else
838 hlist_add_head_rcu(&sk->sk_node, list);
839}
840
841static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
842{
843 sock_hold(sk);
844 hlist_add_tail_rcu(&sk->sk_node, list);
845}
846
847static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
848{
849 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
850}
851
852static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
853{
854 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
855}
856
857static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
858{
859 sock_hold(sk);
860 __sk_nulls_add_node_rcu(sk, list);
861}
862
863static inline void __sk_del_bind_node(struct sock *sk)
864{
865 __hlist_del(&sk->sk_bind_node);
866}
867
868static inline void sk_add_bind_node(struct sock *sk,
869 struct hlist_head *list)
870{
871 hlist_add_head(&sk->sk_bind_node, list);
872}
873
874static inline void __sk_del_bind2_node(struct sock *sk)
875{
876 __hlist_del(&sk->sk_bind2_node);
877}
878
879static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list)
880{
881 hlist_add_head(&sk->sk_bind2_node, list);
882}
883
884#define sk_for_each(__sk, list) \
885 hlist_for_each_entry(__sk, list, sk_node)
886#define sk_for_each_rcu(__sk, list) \
887 hlist_for_each_entry_rcu(__sk, list, sk_node)
888#define sk_nulls_for_each(__sk, node, list) \
889 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
890#define sk_nulls_for_each_rcu(__sk, node, list) \
891 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
892#define sk_for_each_from(__sk) \
893 hlist_for_each_entry_from(__sk, sk_node)
894#define sk_nulls_for_each_from(__sk, node) \
895 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
896 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
897#define sk_for_each_safe(__sk, tmp, list) \
898 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
899#define sk_for_each_bound(__sk, list) \
900 hlist_for_each_entry(__sk, list, sk_bind_node)
901#define sk_for_each_bound_bhash2(__sk, list) \
902 hlist_for_each_entry(__sk, list, sk_bind2_node)
903
904/**
905 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
906 * @tpos: the type * to use as a loop cursor.
907 * @pos: the &struct hlist_node to use as a loop cursor.
908 * @head: the head for your list.
909 * @offset: offset of hlist_node within the struct.
910 *
911 */
912#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
913 for (pos = rcu_dereference(hlist_first_rcu(head)); \
914 pos != NULL && \
915 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
916 pos = rcu_dereference(hlist_next_rcu(pos)))
917
918static inline struct user_namespace *sk_user_ns(const struct sock *sk)
919{
920 /* Careful only use this in a context where these parameters
921 * can not change and must all be valid, such as recvmsg from
922 * userspace.
923 */
924 return sk->sk_socket->file->f_cred->user_ns;
925}
926
927/* Sock flags */
928enum sock_flags {
929 SOCK_DEAD,
930 SOCK_DONE,
931 SOCK_URGINLINE,
932 SOCK_KEEPOPEN,
933 SOCK_LINGER,
934 SOCK_DESTROY,
935 SOCK_BROADCAST,
936 SOCK_TIMESTAMP,
937 SOCK_ZAPPED,
938 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
939 SOCK_DBG, /* %SO_DEBUG setting */
940 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
941 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
942 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
943 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
944 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
945 SOCK_FASYNC, /* fasync() active */
946 SOCK_RXQ_OVFL,
947 SOCK_ZEROCOPY, /* buffers from userspace */
948 SOCK_WIFI_STATUS, /* push wifi status to userspace */
949 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
950 * Will use last 4 bytes of packet sent from
951 * user-space instead.
952 */
953 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
954 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
955 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
956 SOCK_TXTIME,
957 SOCK_XDP, /* XDP is attached */
958 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
959 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */
960};
961
962#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
963
964static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
965{
966 nsk->sk_flags = osk->sk_flags;
967}
968
969static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
970{
971 __set_bit(flag, &sk->sk_flags);
972}
973
974static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
975{
976 __clear_bit(flag, &sk->sk_flags);
977}
978
979static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
980 int valbool)
981{
982 if (valbool)
983 sock_set_flag(sk, bit);
984 else
985 sock_reset_flag(sk, bit);
986}
987
988static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
989{
990 return test_bit(flag, &sk->sk_flags);
991}
992
993#ifdef CONFIG_NET
994DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
995static inline int sk_memalloc_socks(void)
996{
997 return static_branch_unlikely(&memalloc_socks_key);
998}
999
1000void __receive_sock(struct file *file);
1001#else
1002
1003static inline int sk_memalloc_socks(void)
1004{
1005 return 0;
1006}
1007
1008static inline void __receive_sock(struct file *file)
1009{ }
1010#endif
1011
1012static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1013{
1014 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1015}
1016
1017static inline void sk_acceptq_removed(struct sock *sk)
1018{
1019 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1020}
1021
1022static inline void sk_acceptq_added(struct sock *sk)
1023{
1024 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1025}
1026
1027/* Note: If you think the test should be:
1028 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1029 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1030 */
1031static inline bool sk_acceptq_is_full(const struct sock *sk)
1032{
1033 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1034}
1035
1036/*
1037 * Compute minimal free write space needed to queue new packets.
1038 */
1039static inline int sk_stream_min_wspace(const struct sock *sk)
1040{
1041 return READ_ONCE(sk->sk_wmem_queued) >> 1;
1042}
1043
1044static inline int sk_stream_wspace(const struct sock *sk)
1045{
1046 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1047}
1048
1049static inline void sk_wmem_queued_add(struct sock *sk, int val)
1050{
1051 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1052}
1053
1054void sk_stream_write_space(struct sock *sk);
1055
1056/* OOB backlog add */
1057static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1058{
1059 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1060 skb_dst_force(skb);
1061
1062 if (!sk->sk_backlog.tail)
1063 WRITE_ONCE(sk->sk_backlog.head, skb);
1064 else
1065 sk->sk_backlog.tail->next = skb;
1066
1067 WRITE_ONCE(sk->sk_backlog.tail, skb);
1068 skb->next = NULL;
1069}
1070
1071/*
1072 * Take into account size of receive queue and backlog queue
1073 * Do not take into account this skb truesize,
1074 * to allow even a single big packet to come.
1075 */
1076static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1077{
1078 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1079
1080 return qsize > limit;
1081}
1082
1083/* The per-socket spinlock must be held here. */
1084static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1085 unsigned int limit)
1086{
1087 if (sk_rcvqueues_full(sk, limit))
1088 return -ENOBUFS;
1089
1090 /*
1091 * If the skb was allocated from pfmemalloc reserves, only
1092 * allow SOCK_MEMALLOC sockets to use it as this socket is
1093 * helping free memory
1094 */
1095 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1096 return -ENOMEM;
1097
1098 __sk_add_backlog(sk, skb);
1099 sk->sk_backlog.len += skb->truesize;
1100 return 0;
1101}
1102
1103int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1104
1105INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1106INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1107
1108static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1109{
1110 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1111 return __sk_backlog_rcv(sk, skb);
1112
1113 return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1114 tcp_v6_do_rcv,
1115 tcp_v4_do_rcv,
1116 sk, skb);
1117}
1118
1119static inline void sk_incoming_cpu_update(struct sock *sk)
1120{
1121 int cpu = raw_smp_processor_id();
1122
1123 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1124 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1125}
1126
1127static inline void sock_rps_record_flow_hash(__u32 hash)
1128{
1129#ifdef CONFIG_RPS
1130 struct rps_sock_flow_table *sock_flow_table;
1131
1132 rcu_read_lock();
1133 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1134 rps_record_sock_flow(sock_flow_table, hash);
1135 rcu_read_unlock();
1136#endif
1137}
1138
1139static inline void sock_rps_record_flow(const struct sock *sk)
1140{
1141#ifdef CONFIG_RPS
1142 if (static_branch_unlikely(&rfs_needed)) {
1143 /* Reading sk->sk_rxhash might incur an expensive cache line
1144 * miss.
1145 *
1146 * TCP_ESTABLISHED does cover almost all states where RFS
1147 * might be useful, and is cheaper [1] than testing :
1148 * IPv4: inet_sk(sk)->inet_daddr
1149 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1150 * OR an additional socket flag
1151 * [1] : sk_state and sk_prot are in the same cache line.
1152 */
1153 if (sk->sk_state == TCP_ESTABLISHED)
1154 sock_rps_record_flow_hash(sk->sk_rxhash);
1155 }
1156#endif
1157}
1158
1159static inline void sock_rps_save_rxhash(struct sock *sk,
1160 const struct sk_buff *skb)
1161{
1162#ifdef CONFIG_RPS
1163 if (unlikely(sk->sk_rxhash != skb->hash))
1164 sk->sk_rxhash = skb->hash;
1165#endif
1166}
1167
1168static inline void sock_rps_reset_rxhash(struct sock *sk)
1169{
1170#ifdef CONFIG_RPS
1171 sk->sk_rxhash = 0;
1172#endif
1173}
1174
1175#define sk_wait_event(__sk, __timeo, __condition, __wait) \
1176 ({ int __rc; \
1177 release_sock(__sk); \
1178 __rc = __condition; \
1179 if (!__rc) { \
1180 *(__timeo) = wait_woken(__wait, \
1181 TASK_INTERRUPTIBLE, \
1182 *(__timeo)); \
1183 } \
1184 sched_annotate_sleep(); \
1185 lock_sock(__sk); \
1186 __rc = __condition; \
1187 __rc; \
1188 })
1189
1190int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1191int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1192void sk_stream_wait_close(struct sock *sk, long timeo_p);
1193int sk_stream_error(struct sock *sk, int flags, int err);
1194void sk_stream_kill_queues(struct sock *sk);
1195void sk_set_memalloc(struct sock *sk);
1196void sk_clear_memalloc(struct sock *sk);
1197
1198void __sk_flush_backlog(struct sock *sk);
1199
1200static inline bool sk_flush_backlog(struct sock *sk)
1201{
1202 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1203 __sk_flush_backlog(sk);
1204 return true;
1205 }
1206 return false;
1207}
1208
1209int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1210
1211struct request_sock_ops;
1212struct timewait_sock_ops;
1213struct inet_hashinfo;
1214struct raw_hashinfo;
1215struct smc_hashinfo;
1216struct module;
1217struct sk_psock;
1218
1219/*
1220 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1221 * un-modified. Special care is taken when initializing object to zero.
1222 */
1223static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1224{
1225 if (offsetof(struct sock, sk_node.next) != 0)
1226 memset(sk, 0, offsetof(struct sock, sk_node.next));
1227 memset(&sk->sk_node.pprev, 0,
1228 size - offsetof(struct sock, sk_node.pprev));
1229}
1230
1231/* Networking protocol blocks we attach to sockets.
1232 * socket layer -> transport layer interface
1233 */
1234struct proto {
1235 void (*close)(struct sock *sk,
1236 long timeout);
1237 int (*pre_connect)(struct sock *sk,
1238 struct sockaddr *uaddr,
1239 int addr_len);
1240 int (*connect)(struct sock *sk,
1241 struct sockaddr *uaddr,
1242 int addr_len);
1243 int (*disconnect)(struct sock *sk, int flags);
1244
1245 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1246 bool kern);
1247
1248 int (*ioctl)(struct sock *sk, int cmd,
1249 unsigned long arg);
1250 int (*init)(struct sock *sk);
1251 void (*destroy)(struct sock *sk);
1252 void (*shutdown)(struct sock *sk, int how);
1253 int (*setsockopt)(struct sock *sk, int level,
1254 int optname, sockptr_t optval,
1255 unsigned int optlen);
1256 int (*getsockopt)(struct sock *sk, int level,
1257 int optname, char __user *optval,
1258 int __user *option);
1259 void (*keepalive)(struct sock *sk, int valbool);
1260#ifdef CONFIG_COMPAT
1261 int (*compat_ioctl)(struct sock *sk,
1262 unsigned int cmd, unsigned long arg);
1263#endif
1264 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1265 size_t len);
1266 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1267 size_t len, int flags, int *addr_len);
1268 int (*sendpage)(struct sock *sk, struct page *page,
1269 int offset, size_t size, int flags);
1270 int (*bind)(struct sock *sk,
1271 struct sockaddr *addr, int addr_len);
1272 int (*bind_add)(struct sock *sk,
1273 struct sockaddr *addr, int addr_len);
1274
1275 int (*backlog_rcv) (struct sock *sk,
1276 struct sk_buff *skb);
1277 bool (*bpf_bypass_getsockopt)(int level,
1278 int optname);
1279
1280 void (*release_cb)(struct sock *sk);
1281
1282 /* Keeping track of sk's, looking them up, and port selection methods. */
1283 int (*hash)(struct sock *sk);
1284 void (*unhash)(struct sock *sk);
1285 void (*rehash)(struct sock *sk);
1286 int (*get_port)(struct sock *sk, unsigned short snum);
1287 void (*put_port)(struct sock *sk);
1288#ifdef CONFIG_BPF_SYSCALL
1289 int (*psock_update_sk_prot)(struct sock *sk,
1290 struct sk_psock *psock,
1291 bool restore);
1292#endif
1293
1294 /* Keeping track of sockets in use */
1295#ifdef CONFIG_PROC_FS
1296 unsigned int inuse_idx;
1297#endif
1298
1299#if IS_ENABLED(CONFIG_MPTCP)
1300 int (*forward_alloc_get)(const struct sock *sk);
1301#endif
1302
1303 bool (*stream_memory_free)(const struct sock *sk, int wake);
1304 bool (*sock_is_readable)(struct sock *sk);
1305 /* Memory pressure */
1306 void (*enter_memory_pressure)(struct sock *sk);
1307 void (*leave_memory_pressure)(struct sock *sk);
1308 atomic_long_t *memory_allocated; /* Current allocated memory. */
1309 int __percpu *per_cpu_fw_alloc;
1310 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1311
1312 /*
1313 * Pressure flag: try to collapse.
1314 * Technical note: it is used by multiple contexts non atomically.
1315 * All the __sk_mem_schedule() is of this nature: accounting
1316 * is strict, actions are advisory and have some latency.
1317 */
1318 unsigned long *memory_pressure;
1319 long *sysctl_mem;
1320
1321 int *sysctl_wmem;
1322 int *sysctl_rmem;
1323 u32 sysctl_wmem_offset;
1324 u32 sysctl_rmem_offset;
1325
1326 int max_header;
1327 bool no_autobind;
1328
1329 struct kmem_cache *slab;
1330 unsigned int obj_size;
1331 slab_flags_t slab_flags;
1332 unsigned int useroffset; /* Usercopy region offset */
1333 unsigned int usersize; /* Usercopy region size */
1334
1335 unsigned int __percpu *orphan_count;
1336
1337 struct request_sock_ops *rsk_prot;
1338 struct timewait_sock_ops *twsk_prot;
1339
1340 union {
1341 struct inet_hashinfo *hashinfo;
1342 struct udp_table *udp_table;
1343 struct raw_hashinfo *raw_hash;
1344 struct smc_hashinfo *smc_hash;
1345 } h;
1346
1347 struct module *owner;
1348
1349 char name[32];
1350
1351 struct list_head node;
1352#ifdef SOCK_REFCNT_DEBUG
1353 atomic_t socks;
1354#endif
1355 int (*diag_destroy)(struct sock *sk, int err);
1356} __randomize_layout;
1357
1358int proto_register(struct proto *prot, int alloc_slab);
1359void proto_unregister(struct proto *prot);
1360int sock_load_diag_module(int family, int protocol);
1361
1362#ifdef SOCK_REFCNT_DEBUG
1363static inline void sk_refcnt_debug_inc(struct sock *sk)
1364{
1365 atomic_inc(&sk->sk_prot->socks);
1366}
1367
1368static inline void sk_refcnt_debug_dec(struct sock *sk)
1369{
1370 atomic_dec(&sk->sk_prot->socks);
1371 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1372 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1373}
1374
1375static inline void sk_refcnt_debug_release(const struct sock *sk)
1376{
1377 if (refcount_read(&sk->sk_refcnt) != 1)
1378 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1379 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1380}
1381#else /* SOCK_REFCNT_DEBUG */
1382#define sk_refcnt_debug_inc(sk) do { } while (0)
1383#define sk_refcnt_debug_dec(sk) do { } while (0)
1384#define sk_refcnt_debug_release(sk) do { } while (0)
1385#endif /* SOCK_REFCNT_DEBUG */
1386
1387INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1388
1389static inline int sk_forward_alloc_get(const struct sock *sk)
1390{
1391#if IS_ENABLED(CONFIG_MPTCP)
1392 if (sk->sk_prot->forward_alloc_get)
1393 return sk->sk_prot->forward_alloc_get(sk);
1394#endif
1395 return sk->sk_forward_alloc;
1396}
1397
1398static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1399{
1400 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1401 return false;
1402
1403 return sk->sk_prot->stream_memory_free ?
1404 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1405 tcp_stream_memory_free, sk, wake) : true;
1406}
1407
1408static inline bool sk_stream_memory_free(const struct sock *sk)
1409{
1410 return __sk_stream_memory_free(sk, 0);
1411}
1412
1413static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1414{
1415 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1416 __sk_stream_memory_free(sk, wake);
1417}
1418
1419static inline bool sk_stream_is_writeable(const struct sock *sk)
1420{
1421 return __sk_stream_is_writeable(sk, 0);
1422}
1423
1424static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1425 struct cgroup *ancestor)
1426{
1427#ifdef CONFIG_SOCK_CGROUP_DATA
1428 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1429 ancestor);
1430#else
1431 return -ENOTSUPP;
1432#endif
1433}
1434
1435static inline bool sk_has_memory_pressure(const struct sock *sk)
1436{
1437 return sk->sk_prot->memory_pressure != NULL;
1438}
1439
1440static inline bool sk_under_memory_pressure(const struct sock *sk)
1441{
1442 if (!sk->sk_prot->memory_pressure)
1443 return false;
1444
1445 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1446 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1447 return true;
1448
1449 return !!*sk->sk_prot->memory_pressure;
1450}
1451
1452static inline long
1453proto_memory_allocated(const struct proto *prot)
1454{
1455 return max(0L, atomic_long_read(prot->memory_allocated));
1456}
1457
1458static inline long
1459sk_memory_allocated(const struct sock *sk)
1460{
1461 return proto_memory_allocated(sk->sk_prot);
1462}
1463
1464/* 1 MB per cpu, in page units */
1465#define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT))
1466
1467static inline void
1468sk_memory_allocated_add(struct sock *sk, int amt)
1469{
1470 int local_reserve;
1471
1472 preempt_disable();
1473 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1474 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) {
1475 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1476 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1477 }
1478 preempt_enable();
1479}
1480
1481static inline void
1482sk_memory_allocated_sub(struct sock *sk, int amt)
1483{
1484 int local_reserve;
1485
1486 preempt_disable();
1487 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1488 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) {
1489 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1490 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1491 }
1492 preempt_enable();
1493}
1494
1495#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1496
1497static inline void sk_sockets_allocated_dec(struct sock *sk)
1498{
1499 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1500 SK_ALLOC_PERCPU_COUNTER_BATCH);
1501}
1502
1503static inline void sk_sockets_allocated_inc(struct sock *sk)
1504{
1505 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1506 SK_ALLOC_PERCPU_COUNTER_BATCH);
1507}
1508
1509static inline u64
1510sk_sockets_allocated_read_positive(struct sock *sk)
1511{
1512 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1513}
1514
1515static inline int
1516proto_sockets_allocated_sum_positive(struct proto *prot)
1517{
1518 return percpu_counter_sum_positive(prot->sockets_allocated);
1519}
1520
1521static inline bool
1522proto_memory_pressure(struct proto *prot)
1523{
1524 if (!prot->memory_pressure)
1525 return false;
1526 return !!*prot->memory_pressure;
1527}
1528
1529
1530#ifdef CONFIG_PROC_FS
1531#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1532struct prot_inuse {
1533 int all;
1534 int val[PROTO_INUSE_NR];
1535};
1536
1537static inline void sock_prot_inuse_add(const struct net *net,
1538 const struct proto *prot, int val)
1539{
1540 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1541}
1542
1543static inline void sock_inuse_add(const struct net *net, int val)
1544{
1545 this_cpu_add(net->core.prot_inuse->all, val);
1546}
1547
1548int sock_prot_inuse_get(struct net *net, struct proto *proto);
1549int sock_inuse_get(struct net *net);
1550#else
1551static inline void sock_prot_inuse_add(const struct net *net,
1552 const struct proto *prot, int val)
1553{
1554}
1555
1556static inline void sock_inuse_add(const struct net *net, int val)
1557{
1558}
1559#endif
1560
1561
1562/* With per-bucket locks this operation is not-atomic, so that
1563 * this version is not worse.
1564 */
1565static inline int __sk_prot_rehash(struct sock *sk)
1566{
1567 sk->sk_prot->unhash(sk);
1568 return sk->sk_prot->hash(sk);
1569}
1570
1571/* About 10 seconds */
1572#define SOCK_DESTROY_TIME (10*HZ)
1573
1574/* Sockets 0-1023 can't be bound to unless you are superuser */
1575#define PROT_SOCK 1024
1576
1577#define SHUTDOWN_MASK 3
1578#define RCV_SHUTDOWN 1
1579#define SEND_SHUTDOWN 2
1580
1581#define SOCK_BINDADDR_LOCK 4
1582#define SOCK_BINDPORT_LOCK 8
1583
1584struct socket_alloc {
1585 struct socket socket;
1586 struct inode vfs_inode;
1587};
1588
1589static inline struct socket *SOCKET_I(struct inode *inode)
1590{
1591 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1592}
1593
1594static inline struct inode *SOCK_INODE(struct socket *socket)
1595{
1596 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1597}
1598
1599/*
1600 * Functions for memory accounting
1601 */
1602int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1603int __sk_mem_schedule(struct sock *sk, int size, int kind);
1604void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1605void __sk_mem_reclaim(struct sock *sk, int amount);
1606
1607#define SK_MEM_SEND 0
1608#define SK_MEM_RECV 1
1609
1610/* sysctl_mem values are in pages */
1611static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1612{
1613 return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1614}
1615
1616static inline int sk_mem_pages(int amt)
1617{
1618 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1619}
1620
1621static inline bool sk_has_account(struct sock *sk)
1622{
1623 /* return true if protocol supports memory accounting */
1624 return !!sk->sk_prot->memory_allocated;
1625}
1626
1627static inline bool sk_wmem_schedule(struct sock *sk, int size)
1628{
1629 int delta;
1630
1631 if (!sk_has_account(sk))
1632 return true;
1633 delta = size - sk->sk_forward_alloc;
1634 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1635}
1636
1637static inline bool
1638sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1639{
1640 int delta;
1641
1642 if (!sk_has_account(sk))
1643 return true;
1644 delta = size - sk->sk_forward_alloc;
1645 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1646 skb_pfmemalloc(skb);
1647}
1648
1649static inline int sk_unused_reserved_mem(const struct sock *sk)
1650{
1651 int unused_mem;
1652
1653 if (likely(!sk->sk_reserved_mem))
1654 return 0;
1655
1656 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1657 atomic_read(&sk->sk_rmem_alloc);
1658
1659 return unused_mem > 0 ? unused_mem : 0;
1660}
1661
1662static inline void sk_mem_reclaim(struct sock *sk)
1663{
1664 int reclaimable;
1665
1666 if (!sk_has_account(sk))
1667 return;
1668
1669 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1670
1671 if (reclaimable >= (int)PAGE_SIZE)
1672 __sk_mem_reclaim(sk, reclaimable);
1673}
1674
1675static inline void sk_mem_reclaim_final(struct sock *sk)
1676{
1677 sk->sk_reserved_mem = 0;
1678 sk_mem_reclaim(sk);
1679}
1680
1681static inline void sk_mem_charge(struct sock *sk, int size)
1682{
1683 if (!sk_has_account(sk))
1684 return;
1685 sk->sk_forward_alloc -= size;
1686}
1687
1688static inline void sk_mem_uncharge(struct sock *sk, int size)
1689{
1690 if (!sk_has_account(sk))
1691 return;
1692 sk->sk_forward_alloc += size;
1693 sk_mem_reclaim(sk);
1694}
1695
1696/*
1697 * Macro so as to not evaluate some arguments when
1698 * lockdep is not enabled.
1699 *
1700 * Mark both the sk_lock and the sk_lock.slock as a
1701 * per-address-family lock class.
1702 */
1703#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1704do { \
1705 sk->sk_lock.owned = 0; \
1706 init_waitqueue_head(&sk->sk_lock.wq); \
1707 spin_lock_init(&(sk)->sk_lock.slock); \
1708 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1709 sizeof((sk)->sk_lock)); \
1710 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1711 (skey), (sname)); \
1712 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1713} while (0)
1714
1715static inline bool lockdep_sock_is_held(const struct sock *sk)
1716{
1717 return lockdep_is_held(&sk->sk_lock) ||
1718 lockdep_is_held(&sk->sk_lock.slock);
1719}
1720
1721void lock_sock_nested(struct sock *sk, int subclass);
1722
1723static inline void lock_sock(struct sock *sk)
1724{
1725 lock_sock_nested(sk, 0);
1726}
1727
1728void __lock_sock(struct sock *sk);
1729void __release_sock(struct sock *sk);
1730void release_sock(struct sock *sk);
1731
1732/* BH context may only use the following locking interface. */
1733#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1734#define bh_lock_sock_nested(__sk) \
1735 spin_lock_nested(&((__sk)->sk_lock.slock), \
1736 SINGLE_DEPTH_NESTING)
1737#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1738
1739bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1740
1741/**
1742 * lock_sock_fast - fast version of lock_sock
1743 * @sk: socket
1744 *
1745 * This version should be used for very small section, where process wont block
1746 * return false if fast path is taken:
1747 *
1748 * sk_lock.slock locked, owned = 0, BH disabled
1749 *
1750 * return true if slow path is taken:
1751 *
1752 * sk_lock.slock unlocked, owned = 1, BH enabled
1753 */
1754static inline bool lock_sock_fast(struct sock *sk)
1755{
1756 /* The sk_lock has mutex_lock() semantics here. */
1757 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1758
1759 return __lock_sock_fast(sk);
1760}
1761
1762/* fast socket lock variant for caller already holding a [different] socket lock */
1763static inline bool lock_sock_fast_nested(struct sock *sk)
1764{
1765 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1766
1767 return __lock_sock_fast(sk);
1768}
1769
1770/**
1771 * unlock_sock_fast - complement of lock_sock_fast
1772 * @sk: socket
1773 * @slow: slow mode
1774 *
1775 * fast unlock socket for user context.
1776 * If slow mode is on, we call regular release_sock()
1777 */
1778static inline void unlock_sock_fast(struct sock *sk, bool slow)
1779 __releases(&sk->sk_lock.slock)
1780{
1781 if (slow) {
1782 release_sock(sk);
1783 __release(&sk->sk_lock.slock);
1784 } else {
1785 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1786 spin_unlock_bh(&sk->sk_lock.slock);
1787 }
1788}
1789
1790void sockopt_lock_sock(struct sock *sk);
1791void sockopt_release_sock(struct sock *sk);
1792bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1793bool sockopt_capable(int cap);
1794
1795/* Used by processes to "lock" a socket state, so that
1796 * interrupts and bottom half handlers won't change it
1797 * from under us. It essentially blocks any incoming
1798 * packets, so that we won't get any new data or any
1799 * packets that change the state of the socket.
1800 *
1801 * While locked, BH processing will add new packets to
1802 * the backlog queue. This queue is processed by the
1803 * owner of the socket lock right before it is released.
1804 *
1805 * Since ~2.3.5 it is also exclusive sleep lock serializing
1806 * accesses from user process context.
1807 */
1808
1809static inline void sock_owned_by_me(const struct sock *sk)
1810{
1811#ifdef CONFIG_LOCKDEP
1812 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1813#endif
1814}
1815
1816static inline bool sock_owned_by_user(const struct sock *sk)
1817{
1818 sock_owned_by_me(sk);
1819 return sk->sk_lock.owned;
1820}
1821
1822static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1823{
1824 return sk->sk_lock.owned;
1825}
1826
1827static inline void sock_release_ownership(struct sock *sk)
1828{
1829 if (sock_owned_by_user_nocheck(sk)) {
1830 sk->sk_lock.owned = 0;
1831
1832 /* The sk_lock has mutex_unlock() semantics: */
1833 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1834 }
1835}
1836
1837/* no reclassification while locks are held */
1838static inline bool sock_allow_reclassification(const struct sock *csk)
1839{
1840 struct sock *sk = (struct sock *)csk;
1841
1842 return !sock_owned_by_user_nocheck(sk) &&
1843 !spin_is_locked(&sk->sk_lock.slock);
1844}
1845
1846struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1847 struct proto *prot, int kern);
1848void sk_free(struct sock *sk);
1849void sk_destruct(struct sock *sk);
1850struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1851void sk_free_unlock_clone(struct sock *sk);
1852
1853struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1854 gfp_t priority);
1855void __sock_wfree(struct sk_buff *skb);
1856void sock_wfree(struct sk_buff *skb);
1857struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1858 gfp_t priority);
1859void skb_orphan_partial(struct sk_buff *skb);
1860void sock_rfree(struct sk_buff *skb);
1861void sock_efree(struct sk_buff *skb);
1862#ifdef CONFIG_INET
1863void sock_edemux(struct sk_buff *skb);
1864void sock_pfree(struct sk_buff *skb);
1865#else
1866#define sock_edemux sock_efree
1867#endif
1868
1869int sk_setsockopt(struct sock *sk, int level, int optname,
1870 sockptr_t optval, unsigned int optlen);
1871int sock_setsockopt(struct socket *sock, int level, int op,
1872 sockptr_t optval, unsigned int optlen);
1873
1874int sk_getsockopt(struct sock *sk, int level, int optname,
1875 sockptr_t optval, sockptr_t optlen);
1876int sock_getsockopt(struct socket *sock, int level, int op,
1877 char __user *optval, int __user *optlen);
1878int sock_gettstamp(struct socket *sock, void __user *userstamp,
1879 bool timeval, bool time32);
1880struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1881 unsigned long data_len, int noblock,
1882 int *errcode, int max_page_order);
1883
1884static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1885 unsigned long size,
1886 int noblock, int *errcode)
1887{
1888 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1889}
1890
1891void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1892void sock_kfree_s(struct sock *sk, void *mem, int size);
1893void sock_kzfree_s(struct sock *sk, void *mem, int size);
1894void sk_send_sigurg(struct sock *sk);
1895
1896static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1897{
1898 if (sk->sk_socket)
1899 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1900 WRITE_ONCE(sk->sk_prot, proto);
1901}
1902
1903struct sockcm_cookie {
1904 u64 transmit_time;
1905 u32 mark;
1906 u32 tsflags;
1907};
1908
1909static inline void sockcm_init(struct sockcm_cookie *sockc,
1910 const struct sock *sk)
1911{
1912 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1913}
1914
1915int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1916 struct sockcm_cookie *sockc);
1917int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1918 struct sockcm_cookie *sockc);
1919
1920/*
1921 * Functions to fill in entries in struct proto_ops when a protocol
1922 * does not implement a particular function.
1923 */
1924int sock_no_bind(struct socket *, struct sockaddr *, int);
1925int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1926int sock_no_socketpair(struct socket *, struct socket *);
1927int sock_no_accept(struct socket *, struct socket *, int, bool);
1928int sock_no_getname(struct socket *, struct sockaddr *, int);
1929int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1930int sock_no_listen(struct socket *, int);
1931int sock_no_shutdown(struct socket *, int);
1932int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1933int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1934int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1935int sock_no_mmap(struct file *file, struct socket *sock,
1936 struct vm_area_struct *vma);
1937ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1938 size_t size, int flags);
1939ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1940 int offset, size_t size, int flags);
1941
1942/*
1943 * Functions to fill in entries in struct proto_ops when a protocol
1944 * uses the inet style.
1945 */
1946int sock_common_getsockopt(struct socket *sock, int level, int optname,
1947 char __user *optval, int __user *optlen);
1948int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1949 int flags);
1950int sock_common_setsockopt(struct socket *sock, int level, int optname,
1951 sockptr_t optval, unsigned int optlen);
1952
1953void sk_common_release(struct sock *sk);
1954
1955/*
1956 * Default socket callbacks and setup code
1957 */
1958
1959/* Initialise core socket variables */
1960void sock_init_data(struct socket *sock, struct sock *sk);
1961
1962/*
1963 * Socket reference counting postulates.
1964 *
1965 * * Each user of socket SHOULD hold a reference count.
1966 * * Each access point to socket (an hash table bucket, reference from a list,
1967 * running timer, skb in flight MUST hold a reference count.
1968 * * When reference count hits 0, it means it will never increase back.
1969 * * When reference count hits 0, it means that no references from
1970 * outside exist to this socket and current process on current CPU
1971 * is last user and may/should destroy this socket.
1972 * * sk_free is called from any context: process, BH, IRQ. When
1973 * it is called, socket has no references from outside -> sk_free
1974 * may release descendant resources allocated by the socket, but
1975 * to the time when it is called, socket is NOT referenced by any
1976 * hash tables, lists etc.
1977 * * Packets, delivered from outside (from network or from another process)
1978 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1979 * when they sit in queue. Otherwise, packets will leak to hole, when
1980 * socket is looked up by one cpu and unhasing is made by another CPU.
1981 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1982 * (leak to backlog). Packet socket does all the processing inside
1983 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1984 * use separate SMP lock, so that they are prone too.
1985 */
1986
1987/* Ungrab socket and destroy it, if it was the last reference. */
1988static inline void sock_put(struct sock *sk)
1989{
1990 if (refcount_dec_and_test(&sk->sk_refcnt))
1991 sk_free(sk);
1992}
1993/* Generic version of sock_put(), dealing with all sockets
1994 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1995 */
1996void sock_gen_put(struct sock *sk);
1997
1998int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1999 unsigned int trim_cap, bool refcounted);
2000static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
2001 const int nested)
2002{
2003 return __sk_receive_skb(sk, skb, nested, 1, true);
2004}
2005
2006static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
2007{
2008 /* sk_tx_queue_mapping accept only upto a 16-bit value */
2009 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
2010 return;
2011 sk->sk_tx_queue_mapping = tx_queue;
2012}
2013
2014#define NO_QUEUE_MAPPING USHRT_MAX
2015
2016static inline void sk_tx_queue_clear(struct sock *sk)
2017{
2018 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
2019}
2020
2021static inline int sk_tx_queue_get(const struct sock *sk)
2022{
2023 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
2024 return sk->sk_tx_queue_mapping;
2025
2026 return -1;
2027}
2028
2029static inline void __sk_rx_queue_set(struct sock *sk,
2030 const struct sk_buff *skb,
2031 bool force_set)
2032{
2033#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2034 if (skb_rx_queue_recorded(skb)) {
2035 u16 rx_queue = skb_get_rx_queue(skb);
2036
2037 if (force_set ||
2038 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2039 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2040 }
2041#endif
2042}
2043
2044static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2045{
2046 __sk_rx_queue_set(sk, skb, true);
2047}
2048
2049static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2050{
2051 __sk_rx_queue_set(sk, skb, false);
2052}
2053
2054static inline void sk_rx_queue_clear(struct sock *sk)
2055{
2056#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2057 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2058#endif
2059}
2060
2061static inline int sk_rx_queue_get(const struct sock *sk)
2062{
2063#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2064 if (sk) {
2065 int res = READ_ONCE(sk->sk_rx_queue_mapping);
2066
2067 if (res != NO_QUEUE_MAPPING)
2068 return res;
2069 }
2070#endif
2071
2072 return -1;
2073}
2074
2075static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2076{
2077 sk->sk_socket = sock;
2078}
2079
2080static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2081{
2082 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2083 return &rcu_dereference_raw(sk->sk_wq)->wait;
2084}
2085/* Detach socket from process context.
2086 * Announce socket dead, detach it from wait queue and inode.
2087 * Note that parent inode held reference count on this struct sock,
2088 * we do not release it in this function, because protocol
2089 * probably wants some additional cleanups or even continuing
2090 * to work with this socket (TCP).
2091 */
2092static inline void sock_orphan(struct sock *sk)
2093{
2094 write_lock_bh(&sk->sk_callback_lock);
2095 sock_set_flag(sk, SOCK_DEAD);
2096 sk_set_socket(sk, NULL);
2097 sk->sk_wq = NULL;
2098 write_unlock_bh(&sk->sk_callback_lock);
2099}
2100
2101static inline void sock_graft(struct sock *sk, struct socket *parent)
2102{
2103 WARN_ON(parent->sk);
2104 write_lock_bh(&sk->sk_callback_lock);
2105 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2106 parent->sk = sk;
2107 sk_set_socket(sk, parent);
2108 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2109 security_sock_graft(sk, parent);
2110 write_unlock_bh(&sk->sk_callback_lock);
2111}
2112
2113kuid_t sock_i_uid(struct sock *sk);
2114unsigned long sock_i_ino(struct sock *sk);
2115
2116static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2117{
2118 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2119}
2120
2121static inline u32 net_tx_rndhash(void)
2122{
2123 u32 v = get_random_u32();
2124
2125 return v ?: 1;
2126}
2127
2128static inline void sk_set_txhash(struct sock *sk)
2129{
2130 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2131 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2132}
2133
2134static inline bool sk_rethink_txhash(struct sock *sk)
2135{
2136 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2137 sk_set_txhash(sk);
2138 return true;
2139 }
2140 return false;
2141}
2142
2143static inline struct dst_entry *
2144__sk_dst_get(struct sock *sk)
2145{
2146 return rcu_dereference_check(sk->sk_dst_cache,
2147 lockdep_sock_is_held(sk));
2148}
2149
2150static inline struct dst_entry *
2151sk_dst_get(struct sock *sk)
2152{
2153 struct dst_entry *dst;
2154
2155 rcu_read_lock();
2156 dst = rcu_dereference(sk->sk_dst_cache);
2157 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
2158 dst = NULL;
2159 rcu_read_unlock();
2160 return dst;
2161}
2162
2163static inline void __dst_negative_advice(struct sock *sk)
2164{
2165 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2166
2167 if (dst && dst->ops->negative_advice) {
2168 ndst = dst->ops->negative_advice(dst);
2169
2170 if (ndst != dst) {
2171 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2172 sk_tx_queue_clear(sk);
2173 sk->sk_dst_pending_confirm = 0;
2174 }
2175 }
2176}
2177
2178static inline void dst_negative_advice(struct sock *sk)
2179{
2180 sk_rethink_txhash(sk);
2181 __dst_negative_advice(sk);
2182}
2183
2184static inline void
2185__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2186{
2187 struct dst_entry *old_dst;
2188
2189 sk_tx_queue_clear(sk);
2190 sk->sk_dst_pending_confirm = 0;
2191 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2192 lockdep_sock_is_held(sk));
2193 rcu_assign_pointer(sk->sk_dst_cache, dst);
2194 dst_release(old_dst);
2195}
2196
2197static inline void
2198sk_dst_set(struct sock *sk, struct dst_entry *dst)
2199{
2200 struct dst_entry *old_dst;
2201
2202 sk_tx_queue_clear(sk);
2203 sk->sk_dst_pending_confirm = 0;
2204 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2205 dst_release(old_dst);
2206}
2207
2208static inline void
2209__sk_dst_reset(struct sock *sk)
2210{
2211 __sk_dst_set(sk, NULL);
2212}
2213
2214static inline void
2215sk_dst_reset(struct sock *sk)
2216{
2217 sk_dst_set(sk, NULL);
2218}
2219
2220struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2221
2222struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2223
2224static inline void sk_dst_confirm(struct sock *sk)
2225{
2226 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2227 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2228}
2229
2230static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2231{
2232 if (skb_get_dst_pending_confirm(skb)) {
2233 struct sock *sk = skb->sk;
2234
2235 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2236 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2237 neigh_confirm(n);
2238 }
2239}
2240
2241bool sk_mc_loop(struct sock *sk);
2242
2243static inline bool sk_can_gso(const struct sock *sk)
2244{
2245 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2246}
2247
2248void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2249
2250static inline void sk_gso_disable(struct sock *sk)
2251{
2252 sk->sk_gso_disabled = 1;
2253 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2254}
2255
2256static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2257 struct iov_iter *from, char *to,
2258 int copy, int offset)
2259{
2260 if (skb->ip_summed == CHECKSUM_NONE) {
2261 __wsum csum = 0;
2262 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2263 return -EFAULT;
2264 skb->csum = csum_block_add(skb->csum, csum, offset);
2265 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2266 if (!copy_from_iter_full_nocache(to, copy, from))
2267 return -EFAULT;
2268 } else if (!copy_from_iter_full(to, copy, from))
2269 return -EFAULT;
2270
2271 return 0;
2272}
2273
2274static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2275 struct iov_iter *from, int copy)
2276{
2277 int err, offset = skb->len;
2278
2279 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2280 copy, offset);
2281 if (err)
2282 __skb_trim(skb, offset);
2283
2284 return err;
2285}
2286
2287static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2288 struct sk_buff *skb,
2289 struct page *page,
2290 int off, int copy)
2291{
2292 int err;
2293
2294 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2295 copy, skb->len);
2296 if (err)
2297 return err;
2298
2299 skb_len_add(skb, copy);
2300 sk_wmem_queued_add(sk, copy);
2301 sk_mem_charge(sk, copy);
2302 return 0;
2303}
2304
2305/**
2306 * sk_wmem_alloc_get - returns write allocations
2307 * @sk: socket
2308 *
2309 * Return: sk_wmem_alloc minus initial offset of one
2310 */
2311static inline int sk_wmem_alloc_get(const struct sock *sk)
2312{
2313 return refcount_read(&sk->sk_wmem_alloc) - 1;
2314}
2315
2316/**
2317 * sk_rmem_alloc_get - returns read allocations
2318 * @sk: socket
2319 *
2320 * Return: sk_rmem_alloc
2321 */
2322static inline int sk_rmem_alloc_get(const struct sock *sk)
2323{
2324 return atomic_read(&sk->sk_rmem_alloc);
2325}
2326
2327/**
2328 * sk_has_allocations - check if allocations are outstanding
2329 * @sk: socket
2330 *
2331 * Return: true if socket has write or read allocations
2332 */
2333static inline bool sk_has_allocations(const struct sock *sk)
2334{
2335 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2336}
2337
2338/**
2339 * skwq_has_sleeper - check if there are any waiting processes
2340 * @wq: struct socket_wq
2341 *
2342 * Return: true if socket_wq has waiting processes
2343 *
2344 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2345 * barrier call. They were added due to the race found within the tcp code.
2346 *
2347 * Consider following tcp code paths::
2348 *
2349 * CPU1 CPU2
2350 * sys_select receive packet
2351 * ... ...
2352 * __add_wait_queue update tp->rcv_nxt
2353 * ... ...
2354 * tp->rcv_nxt check sock_def_readable
2355 * ... {
2356 * schedule rcu_read_lock();
2357 * wq = rcu_dereference(sk->sk_wq);
2358 * if (wq && waitqueue_active(&wq->wait))
2359 * wake_up_interruptible(&wq->wait)
2360 * ...
2361 * }
2362 *
2363 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2364 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2365 * could then endup calling schedule and sleep forever if there are no more
2366 * data on the socket.
2367 *
2368 */
2369static inline bool skwq_has_sleeper(struct socket_wq *wq)
2370{
2371 return wq && wq_has_sleeper(&wq->wait);
2372}
2373
2374/**
2375 * sock_poll_wait - place memory barrier behind the poll_wait call.
2376 * @filp: file
2377 * @sock: socket to wait on
2378 * @p: poll_table
2379 *
2380 * See the comments in the wq_has_sleeper function.
2381 */
2382static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2383 poll_table *p)
2384{
2385 if (!poll_does_not_wait(p)) {
2386 poll_wait(filp, &sock->wq.wait, p);
2387 /* We need to be sure we are in sync with the
2388 * socket flags modification.
2389 *
2390 * This memory barrier is paired in the wq_has_sleeper.
2391 */
2392 smp_mb();
2393 }
2394}
2395
2396static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2397{
2398 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2399 u32 txhash = READ_ONCE(sk->sk_txhash);
2400
2401 if (txhash) {
2402 skb->l4_hash = 1;
2403 skb->hash = txhash;
2404 }
2405}
2406
2407void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2408
2409/*
2410 * Queue a received datagram if it will fit. Stream and sequenced
2411 * protocols can't normally use this as they need to fit buffers in
2412 * and play with them.
2413 *
2414 * Inlined as it's very short and called for pretty much every
2415 * packet ever received.
2416 */
2417static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2418{
2419 skb_orphan(skb);
2420 skb->sk = sk;
2421 skb->destructor = sock_rfree;
2422 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2423 sk_mem_charge(sk, skb->truesize);
2424}
2425
2426static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2427{
2428 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2429 skb_orphan(skb);
2430 skb->destructor = sock_efree;
2431 skb->sk = sk;
2432 return true;
2433 }
2434 return false;
2435}
2436
2437static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2438{
2439 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2440 if (skb) {
2441 if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2442 skb_set_owner_r(skb, sk);
2443 return skb;
2444 }
2445 __kfree_skb(skb);
2446 }
2447 return NULL;
2448}
2449
2450static inline void skb_prepare_for_gro(struct sk_buff *skb)
2451{
2452 if (skb->destructor != sock_wfree) {
2453 skb_orphan(skb);
2454 return;
2455 }
2456 skb->slow_gro = 1;
2457}
2458
2459void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2460 unsigned long expires);
2461
2462void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2463
2464void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2465
2466int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2467 struct sk_buff *skb, unsigned int flags,
2468 void (*destructor)(struct sock *sk,
2469 struct sk_buff *skb));
2470int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2471
2472int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2473 enum skb_drop_reason *reason);
2474
2475static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2476{
2477 return sock_queue_rcv_skb_reason(sk, skb, NULL);
2478}
2479
2480int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2481struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2482
2483/*
2484 * Recover an error report and clear atomically
2485 */
2486
2487static inline int sock_error(struct sock *sk)
2488{
2489 int err;
2490
2491 /* Avoid an atomic operation for the common case.
2492 * This is racy since another cpu/thread can change sk_err under us.
2493 */
2494 if (likely(data_race(!sk->sk_err)))
2495 return 0;
2496
2497 err = xchg(&sk->sk_err, 0);
2498 return -err;
2499}
2500
2501void sk_error_report(struct sock *sk);
2502
2503static inline unsigned long sock_wspace(struct sock *sk)
2504{
2505 int amt = 0;
2506
2507 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2508 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2509 if (amt < 0)
2510 amt = 0;
2511 }
2512 return amt;
2513}
2514
2515/* Note:
2516 * We use sk->sk_wq_raw, from contexts knowing this
2517 * pointer is not NULL and cannot disappear/change.
2518 */
2519static inline void sk_set_bit(int nr, struct sock *sk)
2520{
2521 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2522 !sock_flag(sk, SOCK_FASYNC))
2523 return;
2524
2525 set_bit(nr, &sk->sk_wq_raw->flags);
2526}
2527
2528static inline void sk_clear_bit(int nr, struct sock *sk)
2529{
2530 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2531 !sock_flag(sk, SOCK_FASYNC))
2532 return;
2533
2534 clear_bit(nr, &sk->sk_wq_raw->flags);
2535}
2536
2537static inline void sk_wake_async(const struct sock *sk, int how, int band)
2538{
2539 if (sock_flag(sk, SOCK_FASYNC)) {
2540 rcu_read_lock();
2541 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2542 rcu_read_unlock();
2543 }
2544}
2545
2546/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2547 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2548 * Note: for send buffers, TCP works better if we can build two skbs at
2549 * minimum.
2550 */
2551#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2552
2553#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2554#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2555
2556static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2557{
2558 u32 val;
2559
2560 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2561 return;
2562
2563 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2564 val = max_t(u32, val, sk_unused_reserved_mem(sk));
2565
2566 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2567}
2568
2569/**
2570 * sk_page_frag - return an appropriate page_frag
2571 * @sk: socket
2572 *
2573 * Use the per task page_frag instead of the per socket one for
2574 * optimization when we know that we're in process context and own
2575 * everything that's associated with %current.
2576 *
2577 * Both direct reclaim and page faults can nest inside other
2578 * socket operations and end up recursing into sk_page_frag()
2579 * while it's already in use: explicitly avoid task page_frag
2580 * when users disable sk_use_task_frag.
2581 *
2582 * Return: a per task page_frag if context allows that,
2583 * otherwise a per socket one.
2584 */
2585static inline struct page_frag *sk_page_frag(struct sock *sk)
2586{
2587 if (sk->sk_use_task_frag)
2588 return ¤t->task_frag;
2589
2590 return &sk->sk_frag;
2591}
2592
2593bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2594
2595/*
2596 * Default write policy as shown to user space via poll/select/SIGIO
2597 */
2598static inline bool sock_writeable(const struct sock *sk)
2599{
2600 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2601}
2602
2603static inline gfp_t gfp_any(void)
2604{
2605 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2606}
2607
2608static inline gfp_t gfp_memcg_charge(void)
2609{
2610 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2611}
2612
2613static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2614{
2615 return noblock ? 0 : sk->sk_rcvtimeo;
2616}
2617
2618static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2619{
2620 return noblock ? 0 : sk->sk_sndtimeo;
2621}
2622
2623static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2624{
2625 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2626
2627 return v ?: 1;
2628}
2629
2630/* Alas, with timeout socket operations are not restartable.
2631 * Compare this to poll().
2632 */
2633static inline int sock_intr_errno(long timeo)
2634{
2635 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2636}
2637
2638struct sock_skb_cb {
2639 u32 dropcount;
2640};
2641
2642/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2643 * using skb->cb[] would keep using it directly and utilize its
2644 * alignement guarantee.
2645 */
2646#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2647 sizeof(struct sock_skb_cb)))
2648
2649#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2650 SOCK_SKB_CB_OFFSET))
2651
2652#define sock_skb_cb_check_size(size) \
2653 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2654
2655static inline void
2656sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2657{
2658 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2659 atomic_read(&sk->sk_drops) : 0;
2660}
2661
2662static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2663{
2664 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2665
2666 atomic_add(segs, &sk->sk_drops);
2667}
2668
2669static inline ktime_t sock_read_timestamp(struct sock *sk)
2670{
2671#if BITS_PER_LONG==32
2672 unsigned int seq;
2673 ktime_t kt;
2674
2675 do {
2676 seq = read_seqbegin(&sk->sk_stamp_seq);
2677 kt = sk->sk_stamp;
2678 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2679
2680 return kt;
2681#else
2682 return READ_ONCE(sk->sk_stamp);
2683#endif
2684}
2685
2686static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2687{
2688#if BITS_PER_LONG==32
2689 write_seqlock(&sk->sk_stamp_seq);
2690 sk->sk_stamp = kt;
2691 write_sequnlock(&sk->sk_stamp_seq);
2692#else
2693 WRITE_ONCE(sk->sk_stamp, kt);
2694#endif
2695}
2696
2697void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2698 struct sk_buff *skb);
2699void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2700 struct sk_buff *skb);
2701
2702static inline void
2703sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2704{
2705 ktime_t kt = skb->tstamp;
2706 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2707
2708 /*
2709 * generate control messages if
2710 * - receive time stamping in software requested
2711 * - software time stamp available and wanted
2712 * - hardware time stamps available and wanted
2713 */
2714 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2715 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2716 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2717 (hwtstamps->hwtstamp &&
2718 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2719 __sock_recv_timestamp(msg, sk, skb);
2720 else
2721 sock_write_timestamp(sk, kt);
2722
2723 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2724 __sock_recv_wifi_status(msg, sk, skb);
2725}
2726
2727void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2728 struct sk_buff *skb);
2729
2730#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2731static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2732 struct sk_buff *skb)
2733{
2734#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \
2735 (1UL << SOCK_RCVTSTAMP) | \
2736 (1UL << SOCK_RCVMARK))
2737#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2738 SOF_TIMESTAMPING_RAW_HARDWARE)
2739
2740 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY)
2741 __sock_recv_cmsgs(msg, sk, skb);
2742 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2743 sock_write_timestamp(sk, skb->tstamp);
2744 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2745 sock_write_timestamp(sk, 0);
2746}
2747
2748void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2749
2750/**
2751 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2752 * @sk: socket sending this packet
2753 * @tsflags: timestamping flags to use
2754 * @tx_flags: completed with instructions for time stamping
2755 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2756 *
2757 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2758 */
2759static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2760 __u8 *tx_flags, __u32 *tskey)
2761{
2762 if (unlikely(tsflags)) {
2763 __sock_tx_timestamp(tsflags, tx_flags);
2764 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2765 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2766 *tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2767 }
2768 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2769 *tx_flags |= SKBTX_WIFI_STATUS;
2770}
2771
2772static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2773 __u8 *tx_flags)
2774{
2775 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2776}
2777
2778static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2779{
2780 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2781 &skb_shinfo(skb)->tskey);
2782}
2783
2784static inline bool sk_is_tcp(const struct sock *sk)
2785{
2786 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP;
2787}
2788
2789/**
2790 * sk_eat_skb - Release a skb if it is no longer needed
2791 * @sk: socket to eat this skb from
2792 * @skb: socket buffer to eat
2793 *
2794 * This routine must be called with interrupts disabled or with the socket
2795 * locked so that the sk_buff queue operation is ok.
2796*/
2797static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2798{
2799 __skb_unlink(skb, &sk->sk_receive_queue);
2800 __kfree_skb(skb);
2801}
2802
2803static inline bool
2804skb_sk_is_prefetched(struct sk_buff *skb)
2805{
2806#ifdef CONFIG_INET
2807 return skb->destructor == sock_pfree;
2808#else
2809 return false;
2810#endif /* CONFIG_INET */
2811}
2812
2813/* This helper checks if a socket is a full socket,
2814 * ie _not_ a timewait or request socket.
2815 */
2816static inline bool sk_fullsock(const struct sock *sk)
2817{
2818 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2819}
2820
2821static inline bool
2822sk_is_refcounted(struct sock *sk)
2823{
2824 /* Only full sockets have sk->sk_flags. */
2825 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2826}
2827
2828/**
2829 * skb_steal_sock - steal a socket from an sk_buff
2830 * @skb: sk_buff to steal the socket from
2831 * @refcounted: is set to true if the socket is reference-counted
2832 */
2833static inline struct sock *
2834skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2835{
2836 if (skb->sk) {
2837 struct sock *sk = skb->sk;
2838
2839 *refcounted = true;
2840 if (skb_sk_is_prefetched(skb))
2841 *refcounted = sk_is_refcounted(sk);
2842 skb->destructor = NULL;
2843 skb->sk = NULL;
2844 return sk;
2845 }
2846 *refcounted = false;
2847 return NULL;
2848}
2849
2850/* Checks if this SKB belongs to an HW offloaded socket
2851 * and whether any SW fallbacks are required based on dev.
2852 * Check decrypted mark in case skb_orphan() cleared socket.
2853 */
2854static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2855 struct net_device *dev)
2856{
2857#ifdef CONFIG_SOCK_VALIDATE_XMIT
2858 struct sock *sk = skb->sk;
2859
2860 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2861 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2862#ifdef CONFIG_TLS_DEVICE
2863 } else if (unlikely(skb->decrypted)) {
2864 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2865 kfree_skb(skb);
2866 skb = NULL;
2867#endif
2868 }
2869#endif
2870
2871 return skb;
2872}
2873
2874/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2875 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2876 */
2877static inline bool sk_listener(const struct sock *sk)
2878{
2879 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2880}
2881
2882void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2883int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2884 int type);
2885
2886bool sk_ns_capable(const struct sock *sk,
2887 struct user_namespace *user_ns, int cap);
2888bool sk_capable(const struct sock *sk, int cap);
2889bool sk_net_capable(const struct sock *sk, int cap);
2890
2891void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2892
2893/* Take into consideration the size of the struct sk_buff overhead in the
2894 * determination of these values, since that is non-constant across
2895 * platforms. This makes socket queueing behavior and performance
2896 * not depend upon such differences.
2897 */
2898#define _SK_MEM_PACKETS 256
2899#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2900#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2901#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2902
2903extern __u32 sysctl_wmem_max;
2904extern __u32 sysctl_rmem_max;
2905
2906extern int sysctl_tstamp_allow_data;
2907extern int sysctl_optmem_max;
2908
2909extern __u32 sysctl_wmem_default;
2910extern __u32 sysctl_rmem_default;
2911
2912#define SKB_FRAG_PAGE_ORDER get_order(32768)
2913DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2914
2915static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2916{
2917 /* Does this proto have per netns sysctl_wmem ? */
2918 if (proto->sysctl_wmem_offset)
2919 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2920
2921 return READ_ONCE(*proto->sysctl_wmem);
2922}
2923
2924static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2925{
2926 /* Does this proto have per netns sysctl_rmem ? */
2927 if (proto->sysctl_rmem_offset)
2928 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2929
2930 return READ_ONCE(*proto->sysctl_rmem);
2931}
2932
2933/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2934 * Some wifi drivers need to tweak it to get more chunks.
2935 * They can use this helper from their ndo_start_xmit()
2936 */
2937static inline void sk_pacing_shift_update(struct sock *sk, int val)
2938{
2939 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2940 return;
2941 WRITE_ONCE(sk->sk_pacing_shift, val);
2942}
2943
2944/* if a socket is bound to a device, check that the given device
2945 * index is either the same or that the socket is bound to an L3
2946 * master device and the given device index is also enslaved to
2947 * that L3 master
2948 */
2949static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2950{
2951 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2952 int mdif;
2953
2954 if (!bound_dev_if || bound_dev_if == dif)
2955 return true;
2956
2957 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2958 if (mdif && mdif == bound_dev_if)
2959 return true;
2960
2961 return false;
2962}
2963
2964void sock_def_readable(struct sock *sk);
2965
2966int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2967void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2968int sock_set_timestamping(struct sock *sk, int optname,
2969 struct so_timestamping timestamping);
2970
2971void sock_enable_timestamps(struct sock *sk);
2972void sock_no_linger(struct sock *sk);
2973void sock_set_keepalive(struct sock *sk);
2974void sock_set_priority(struct sock *sk, u32 priority);
2975void sock_set_rcvbuf(struct sock *sk, int val);
2976void sock_set_mark(struct sock *sk, u32 val);
2977void sock_set_reuseaddr(struct sock *sk);
2978void sock_set_reuseport(struct sock *sk);
2979void sock_set_sndtimeo(struct sock *sk, s64 secs);
2980
2981int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2982
2983int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2984int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2985 sockptr_t optval, int optlen, bool old_timeval);
2986
2987static inline bool sk_is_readable(struct sock *sk)
2988{
2989 if (sk->sk_prot->sock_is_readable)
2990 return sk->sk_prot->sock_is_readable(sk);
2991 return false;
2992}
2993#endif /* _SOCK_H */