Loading...
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/filter.h>
60#include <linux/rculist_nulls.h>
61#include <linux/poll.h>
62
63#include <linux/atomic.h>
64#include <linux/refcount.h>
65#include <net/dst.h>
66#include <net/checksum.h>
67#include <net/tcp_states.h>
68#include <linux/net_tstamp.h>
69#include <net/smc.h>
70#include <net/l3mdev.h>
71
72/*
73 * This structure really needs to be cleaned up.
74 * Most of it is for TCP, and not used by any of
75 * the other protocols.
76 */
77
78/* Define this to get the SOCK_DBG debugging facility. */
79#define SOCK_DEBUGGING
80#ifdef SOCK_DEBUGGING
81#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
82 printk(KERN_DEBUG msg); } while (0)
83#else
84/* Validate arguments and do nothing */
85static inline __printf(2, 3)
86void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
87{
88}
89#endif
90
91/* This is the per-socket lock. The spinlock provides a synchronization
92 * between user contexts and software interrupt processing, whereas the
93 * mini-semaphore synchronizes multiple users amongst themselves.
94 */
95typedef struct {
96 spinlock_t slock;
97 int owned;
98 wait_queue_head_t wq;
99 /*
100 * We express the mutex-alike socket_lock semantics
101 * to the lock validator by explicitly managing
102 * the slock as a lock variant (in addition to
103 * the slock itself):
104 */
105#ifdef CONFIG_DEBUG_LOCK_ALLOC
106 struct lockdep_map dep_map;
107#endif
108} socket_lock_t;
109
110struct sock;
111struct proto;
112struct net;
113
114typedef __u32 __bitwise __portpair;
115typedef __u64 __bitwise __addrpair;
116
117/**
118 * struct sock_common - minimal network layer representation of sockets
119 * @skc_daddr: Foreign IPv4 addr
120 * @skc_rcv_saddr: Bound local IPv4 addr
121 * @skc_hash: hash value used with various protocol lookup tables
122 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
123 * @skc_dport: placeholder for inet_dport/tw_dport
124 * @skc_num: placeholder for inet_num/tw_num
125 * @skc_family: network address family
126 * @skc_state: Connection state
127 * @skc_reuse: %SO_REUSEADDR setting
128 * @skc_reuseport: %SO_REUSEPORT setting
129 * @skc_bound_dev_if: bound device index if != 0
130 * @skc_bind_node: bind hash linkage for various protocol lookup tables
131 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
132 * @skc_prot: protocol handlers inside a network family
133 * @skc_net: reference to the network namespace of this socket
134 * @skc_node: main hash linkage for various protocol lookup tables
135 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
136 * @skc_tx_queue_mapping: tx queue number for this connection
137 * @skc_rx_queue_mapping: rx queue number for this connection
138 * @skc_flags: place holder for sk_flags
139 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
140 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
141 * @skc_incoming_cpu: record/match cpu processing incoming packets
142 * @skc_refcnt: reference count
143 *
144 * This is the minimal network layer representation of sockets, the header
145 * for struct sock and struct inet_timewait_sock.
146 */
147struct sock_common {
148 /* skc_daddr and skc_rcv_saddr must be grouped on a 8 bytes aligned
149 * address on 64bit arches : cf INET_MATCH()
150 */
151 union {
152 __addrpair skc_addrpair;
153 struct {
154 __be32 skc_daddr;
155 __be32 skc_rcv_saddr;
156 };
157 };
158 union {
159 unsigned int skc_hash;
160 __u16 skc_u16hashes[2];
161 };
162 /* skc_dport && skc_num must be grouped as well */
163 union {
164 __portpair skc_portpair;
165 struct {
166 __be16 skc_dport;
167 __u16 skc_num;
168 };
169 };
170
171 unsigned short skc_family;
172 volatile unsigned char skc_state;
173 unsigned char skc_reuse:4;
174 unsigned char skc_reuseport:1;
175 unsigned char skc_ipv6only:1;
176 unsigned char skc_net_refcnt:1;
177 int skc_bound_dev_if;
178 union {
179 struct hlist_node skc_bind_node;
180 struct hlist_node skc_portaddr_node;
181 };
182 struct proto *skc_prot;
183 possible_net_t skc_net;
184
185#if IS_ENABLED(CONFIG_IPV6)
186 struct in6_addr skc_v6_daddr;
187 struct in6_addr skc_v6_rcv_saddr;
188#endif
189
190 atomic64_t skc_cookie;
191
192 /* following fields are padding to force
193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
194 * assuming IPV6 is enabled. We use this padding differently
195 * for different kind of 'sockets'
196 */
197 union {
198 unsigned long skc_flags;
199 struct sock *skc_listener; /* request_sock */
200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
201 };
202 /*
203 * fields between dontcopy_begin/dontcopy_end
204 * are not copied in sock_copy()
205 */
206 /* private: */
207 int skc_dontcopy_begin[0];
208 /* public: */
209 union {
210 struct hlist_node skc_node;
211 struct hlist_nulls_node skc_nulls_node;
212 };
213 unsigned short skc_tx_queue_mapping;
214#ifdef CONFIG_XPS
215 unsigned short skc_rx_queue_mapping;
216#endif
217 union {
218 int skc_incoming_cpu;
219 u32 skc_rcv_wnd;
220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
221 };
222
223 refcount_t skc_refcnt;
224 /* private: */
225 int skc_dontcopy_end[0];
226 union {
227 u32 skc_rxhash;
228 u32 skc_window_clamp;
229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
230 };
231 /* public: */
232};
233
234struct bpf_sk_storage;
235
236/**
237 * struct sock - network layer representation of sockets
238 * @__sk_common: shared layout with inet_timewait_sock
239 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
240 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
241 * @sk_lock: synchronizer
242 * @sk_kern_sock: True if sock is using kernel lock classes
243 * @sk_rcvbuf: size of receive buffer in bytes
244 * @sk_wq: sock wait queue and async head
245 * @sk_rx_dst: receive input route used by early demux
246 * @sk_dst_cache: destination cache
247 * @sk_dst_pending_confirm: need to confirm neighbour
248 * @sk_policy: flow policy
249 * @sk_receive_queue: incoming packets
250 * @sk_wmem_alloc: transmit queue bytes committed
251 * @sk_tsq_flags: TCP Small Queues flags
252 * @sk_write_queue: Packet sending queue
253 * @sk_omem_alloc: "o" is "option" or "other"
254 * @sk_wmem_queued: persistent queue size
255 * @sk_forward_alloc: space allocated forward
256 * @sk_napi_id: id of the last napi context to receive data for sk
257 * @sk_ll_usec: usecs to busypoll when there is no data
258 * @sk_allocation: allocation mode
259 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
260 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
261 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
262 * @sk_sndbuf: size of send buffer in bytes
263 * @__sk_flags_offset: empty field used to determine location of bitfield
264 * @sk_padding: unused element for alignment
265 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
266 * @sk_no_check_rx: allow zero checksum in RX packets
267 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
268 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
269 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
270 * @sk_gso_max_size: Maximum GSO segment size to build
271 * @sk_gso_max_segs: Maximum number of GSO segments
272 * @sk_pacing_shift: scaling factor for TCP Small Queues
273 * @sk_lingertime: %SO_LINGER l_linger setting
274 * @sk_backlog: always used with the per-socket spinlock held
275 * @sk_callback_lock: used with the callbacks in the end of this struct
276 * @sk_error_queue: rarely used
277 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
278 * IPV6_ADDRFORM for instance)
279 * @sk_err: last error
280 * @sk_err_soft: errors that don't cause failure but are the cause of a
281 * persistent failure not just 'timed out'
282 * @sk_drops: raw/udp drops counter
283 * @sk_ack_backlog: current listen backlog
284 * @sk_max_ack_backlog: listen backlog set in listen()
285 * @sk_uid: user id of owner
286 * @sk_priority: %SO_PRIORITY setting
287 * @sk_type: socket type (%SOCK_STREAM, etc)
288 * @sk_protocol: which protocol this socket belongs in this network family
289 * @sk_peer_pid: &struct pid for this socket's peer
290 * @sk_peer_cred: %SO_PEERCRED setting
291 * @sk_rcvlowat: %SO_RCVLOWAT setting
292 * @sk_rcvtimeo: %SO_RCVTIMEO setting
293 * @sk_sndtimeo: %SO_SNDTIMEO setting
294 * @sk_txhash: computed flow hash for use on transmit
295 * @sk_filter: socket filtering instructions
296 * @sk_timer: sock cleanup timer
297 * @sk_stamp: time stamp of last packet received
298 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
299 * @sk_tsflags: SO_TIMESTAMPING socket options
300 * @sk_tskey: counter to disambiguate concurrent tstamp requests
301 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
302 * @sk_socket: Identd and reporting IO signals
303 * @sk_user_data: RPC layer private data
304 * @sk_frag: cached page frag
305 * @sk_peek_off: current peek_offset value
306 * @sk_send_head: front of stuff to transmit
307 * @sk_security: used by security modules
308 * @sk_mark: generic packet mark
309 * @sk_cgrp_data: cgroup data for this cgroup
310 * @sk_memcg: this socket's memory cgroup association
311 * @sk_write_pending: a write to stream socket waits to start
312 * @sk_state_change: callback to indicate change in the state of the sock
313 * @sk_data_ready: callback to indicate there is data to be processed
314 * @sk_write_space: callback to indicate there is bf sending space available
315 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
316 * @sk_backlog_rcv: callback to process the backlog
317 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
318 * @sk_reuseport_cb: reuseport group container
319 * @sk_rcu: used during RCU grace period
320 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
321 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
322 * @sk_txtime_unused: unused txtime flags
323 */
324struct sock {
325 /*
326 * Now struct inet_timewait_sock also uses sock_common, so please just
327 * don't add nothing before this first member (__sk_common) --acme
328 */
329 struct sock_common __sk_common;
330#define sk_node __sk_common.skc_node
331#define sk_nulls_node __sk_common.skc_nulls_node
332#define sk_refcnt __sk_common.skc_refcnt
333#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
334#ifdef CONFIG_XPS
335#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
336#endif
337
338#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
339#define sk_dontcopy_end __sk_common.skc_dontcopy_end
340#define sk_hash __sk_common.skc_hash
341#define sk_portpair __sk_common.skc_portpair
342#define sk_num __sk_common.skc_num
343#define sk_dport __sk_common.skc_dport
344#define sk_addrpair __sk_common.skc_addrpair
345#define sk_daddr __sk_common.skc_daddr
346#define sk_rcv_saddr __sk_common.skc_rcv_saddr
347#define sk_family __sk_common.skc_family
348#define sk_state __sk_common.skc_state
349#define sk_reuse __sk_common.skc_reuse
350#define sk_reuseport __sk_common.skc_reuseport
351#define sk_ipv6only __sk_common.skc_ipv6only
352#define sk_net_refcnt __sk_common.skc_net_refcnt
353#define sk_bound_dev_if __sk_common.skc_bound_dev_if
354#define sk_bind_node __sk_common.skc_bind_node
355#define sk_prot __sk_common.skc_prot
356#define sk_net __sk_common.skc_net
357#define sk_v6_daddr __sk_common.skc_v6_daddr
358#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
359#define sk_cookie __sk_common.skc_cookie
360#define sk_incoming_cpu __sk_common.skc_incoming_cpu
361#define sk_flags __sk_common.skc_flags
362#define sk_rxhash __sk_common.skc_rxhash
363
364 socket_lock_t sk_lock;
365 atomic_t sk_drops;
366 int sk_rcvlowat;
367 struct sk_buff_head sk_error_queue;
368 struct sk_buff *sk_rx_skb_cache;
369 struct sk_buff_head sk_receive_queue;
370 /*
371 * The backlog queue is special, it is always used with
372 * the per-socket spinlock held and requires low latency
373 * access. Therefore we special case it's implementation.
374 * Note : rmem_alloc is in this structure to fill a hole
375 * on 64bit arches, not because its logically part of
376 * backlog.
377 */
378 struct {
379 atomic_t rmem_alloc;
380 int len;
381 struct sk_buff *head;
382 struct sk_buff *tail;
383 } sk_backlog;
384#define sk_rmem_alloc sk_backlog.rmem_alloc
385
386 int sk_forward_alloc;
387#ifdef CONFIG_NET_RX_BUSY_POLL
388 unsigned int sk_ll_usec;
389 /* ===== mostly read cache line ===== */
390 unsigned int sk_napi_id;
391#endif
392 int sk_rcvbuf;
393
394 struct sk_filter __rcu *sk_filter;
395 union {
396 struct socket_wq __rcu *sk_wq;
397 struct socket_wq *sk_wq_raw;
398 };
399#ifdef CONFIG_XFRM
400 struct xfrm_policy __rcu *sk_policy[2];
401#endif
402 struct dst_entry *sk_rx_dst;
403 struct dst_entry __rcu *sk_dst_cache;
404 atomic_t sk_omem_alloc;
405 int sk_sndbuf;
406
407 /* ===== cache line for TX ===== */
408 int sk_wmem_queued;
409 refcount_t sk_wmem_alloc;
410 unsigned long sk_tsq_flags;
411 union {
412 struct sk_buff *sk_send_head;
413 struct rb_root tcp_rtx_queue;
414 };
415 struct sk_buff *sk_tx_skb_cache;
416 struct sk_buff_head sk_write_queue;
417 __s32 sk_peek_off;
418 int sk_write_pending;
419 __u32 sk_dst_pending_confirm;
420 u32 sk_pacing_status; /* see enum sk_pacing */
421 long sk_sndtimeo;
422 struct timer_list sk_timer;
423 __u32 sk_priority;
424 __u32 sk_mark;
425 unsigned long sk_pacing_rate; /* bytes per second */
426 unsigned long sk_max_pacing_rate;
427 struct page_frag sk_frag;
428 netdev_features_t sk_route_caps;
429 netdev_features_t sk_route_nocaps;
430 netdev_features_t sk_route_forced_caps;
431 int sk_gso_type;
432 unsigned int sk_gso_max_size;
433 gfp_t sk_allocation;
434 __u32 sk_txhash;
435
436 /*
437 * Because of non atomicity rules, all
438 * changes are protected by socket lock.
439 */
440 unsigned int __sk_flags_offset[0];
441#ifdef __BIG_ENDIAN_BITFIELD
442#define SK_FL_PROTO_SHIFT 16
443#define SK_FL_PROTO_MASK 0x00ff0000
444
445#define SK_FL_TYPE_SHIFT 0
446#define SK_FL_TYPE_MASK 0x0000ffff
447#else
448#define SK_FL_PROTO_SHIFT 8
449#define SK_FL_PROTO_MASK 0x0000ff00
450
451#define SK_FL_TYPE_SHIFT 16
452#define SK_FL_TYPE_MASK 0xffff0000
453#endif
454
455 unsigned int sk_padding : 1,
456 sk_kern_sock : 1,
457 sk_no_check_tx : 1,
458 sk_no_check_rx : 1,
459 sk_userlocks : 4,
460 sk_protocol : 8,
461 sk_type : 16;
462#define SK_PROTOCOL_MAX U8_MAX
463 u16 sk_gso_max_segs;
464 u8 sk_pacing_shift;
465 unsigned long sk_lingertime;
466 struct proto *sk_prot_creator;
467 rwlock_t sk_callback_lock;
468 int sk_err,
469 sk_err_soft;
470 u32 sk_ack_backlog;
471 u32 sk_max_ack_backlog;
472 kuid_t sk_uid;
473 struct pid *sk_peer_pid;
474 const struct cred *sk_peer_cred;
475 long sk_rcvtimeo;
476 ktime_t sk_stamp;
477#if BITS_PER_LONG==32
478 seqlock_t sk_stamp_seq;
479#endif
480 u16 sk_tsflags;
481 u8 sk_shutdown;
482 u32 sk_tskey;
483 atomic_t sk_zckey;
484
485 u8 sk_clockid;
486 u8 sk_txtime_deadline_mode : 1,
487 sk_txtime_report_errors : 1,
488 sk_txtime_unused : 6;
489
490 struct socket *sk_socket;
491 void *sk_user_data;
492#ifdef CONFIG_SECURITY
493 void *sk_security;
494#endif
495 struct sock_cgroup_data sk_cgrp_data;
496 struct mem_cgroup *sk_memcg;
497 void (*sk_state_change)(struct sock *sk);
498 void (*sk_data_ready)(struct sock *sk);
499 void (*sk_write_space)(struct sock *sk);
500 void (*sk_error_report)(struct sock *sk);
501 int (*sk_backlog_rcv)(struct sock *sk,
502 struct sk_buff *skb);
503#ifdef CONFIG_SOCK_VALIDATE_XMIT
504 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
505 struct net_device *dev,
506 struct sk_buff *skb);
507#endif
508 void (*sk_destruct)(struct sock *sk);
509 struct sock_reuseport __rcu *sk_reuseport_cb;
510#ifdef CONFIG_BPF_SYSCALL
511 struct bpf_sk_storage __rcu *sk_bpf_storage;
512#endif
513 struct rcu_head sk_rcu;
514};
515
516enum sk_pacing {
517 SK_PACING_NONE = 0,
518 SK_PACING_NEEDED = 1,
519 SK_PACING_FQ = 2,
520};
521
522#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
523
524#define rcu_dereference_sk_user_data(sk) rcu_dereference(__sk_user_data((sk)))
525#define rcu_assign_sk_user_data(sk, ptr) rcu_assign_pointer(__sk_user_data((sk)), ptr)
526
527/*
528 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
529 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
530 * on a socket means that the socket will reuse everybody else's port
531 * without looking at the other's sk_reuse value.
532 */
533
534#define SK_NO_REUSE 0
535#define SK_CAN_REUSE 1
536#define SK_FORCE_REUSE 2
537
538int sk_set_peek_off(struct sock *sk, int val);
539
540static inline int sk_peek_offset(struct sock *sk, int flags)
541{
542 if (unlikely(flags & MSG_PEEK)) {
543 return READ_ONCE(sk->sk_peek_off);
544 }
545
546 return 0;
547}
548
549static inline void sk_peek_offset_bwd(struct sock *sk, int val)
550{
551 s32 off = READ_ONCE(sk->sk_peek_off);
552
553 if (unlikely(off >= 0)) {
554 off = max_t(s32, off - val, 0);
555 WRITE_ONCE(sk->sk_peek_off, off);
556 }
557}
558
559static inline void sk_peek_offset_fwd(struct sock *sk, int val)
560{
561 sk_peek_offset_bwd(sk, -val);
562}
563
564/*
565 * Hashed lists helper routines
566 */
567static inline struct sock *sk_entry(const struct hlist_node *node)
568{
569 return hlist_entry(node, struct sock, sk_node);
570}
571
572static inline struct sock *__sk_head(const struct hlist_head *head)
573{
574 return hlist_entry(head->first, struct sock, sk_node);
575}
576
577static inline struct sock *sk_head(const struct hlist_head *head)
578{
579 return hlist_empty(head) ? NULL : __sk_head(head);
580}
581
582static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
583{
584 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
585}
586
587static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
588{
589 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
590}
591
592static inline struct sock *sk_next(const struct sock *sk)
593{
594 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
595}
596
597static inline struct sock *sk_nulls_next(const struct sock *sk)
598{
599 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
600 hlist_nulls_entry(sk->sk_nulls_node.next,
601 struct sock, sk_nulls_node) :
602 NULL;
603}
604
605static inline bool sk_unhashed(const struct sock *sk)
606{
607 return hlist_unhashed(&sk->sk_node);
608}
609
610static inline bool sk_hashed(const struct sock *sk)
611{
612 return !sk_unhashed(sk);
613}
614
615static inline void sk_node_init(struct hlist_node *node)
616{
617 node->pprev = NULL;
618}
619
620static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
621{
622 node->pprev = NULL;
623}
624
625static inline void __sk_del_node(struct sock *sk)
626{
627 __hlist_del(&sk->sk_node);
628}
629
630/* NB: equivalent to hlist_del_init_rcu */
631static inline bool __sk_del_node_init(struct sock *sk)
632{
633 if (sk_hashed(sk)) {
634 __sk_del_node(sk);
635 sk_node_init(&sk->sk_node);
636 return true;
637 }
638 return false;
639}
640
641/* Grab socket reference count. This operation is valid only
642 when sk is ALREADY grabbed f.e. it is found in hash table
643 or a list and the lookup is made under lock preventing hash table
644 modifications.
645 */
646
647static __always_inline void sock_hold(struct sock *sk)
648{
649 refcount_inc(&sk->sk_refcnt);
650}
651
652/* Ungrab socket in the context, which assumes that socket refcnt
653 cannot hit zero, f.e. it is true in context of any socketcall.
654 */
655static __always_inline void __sock_put(struct sock *sk)
656{
657 refcount_dec(&sk->sk_refcnt);
658}
659
660static inline bool sk_del_node_init(struct sock *sk)
661{
662 bool rc = __sk_del_node_init(sk);
663
664 if (rc) {
665 /* paranoid for a while -acme */
666 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
667 __sock_put(sk);
668 }
669 return rc;
670}
671#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
672
673static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
674{
675 if (sk_hashed(sk)) {
676 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
677 return true;
678 }
679 return false;
680}
681
682static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
683{
684 bool rc = __sk_nulls_del_node_init_rcu(sk);
685
686 if (rc) {
687 /* paranoid for a while -acme */
688 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
689 __sock_put(sk);
690 }
691 return rc;
692}
693
694static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
695{
696 hlist_add_head(&sk->sk_node, list);
697}
698
699static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
700{
701 sock_hold(sk);
702 __sk_add_node(sk, list);
703}
704
705static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
706{
707 sock_hold(sk);
708 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
709 sk->sk_family == AF_INET6)
710 hlist_add_tail_rcu(&sk->sk_node, list);
711 else
712 hlist_add_head_rcu(&sk->sk_node, list);
713}
714
715static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
716{
717 sock_hold(sk);
718 hlist_add_tail_rcu(&sk->sk_node, list);
719}
720
721static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
722{
723 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
724}
725
726static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
727{
728 sock_hold(sk);
729 __sk_nulls_add_node_rcu(sk, list);
730}
731
732static inline void __sk_del_bind_node(struct sock *sk)
733{
734 __hlist_del(&sk->sk_bind_node);
735}
736
737static inline void sk_add_bind_node(struct sock *sk,
738 struct hlist_head *list)
739{
740 hlist_add_head(&sk->sk_bind_node, list);
741}
742
743#define sk_for_each(__sk, list) \
744 hlist_for_each_entry(__sk, list, sk_node)
745#define sk_for_each_rcu(__sk, list) \
746 hlist_for_each_entry_rcu(__sk, list, sk_node)
747#define sk_nulls_for_each(__sk, node, list) \
748 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
749#define sk_nulls_for_each_rcu(__sk, node, list) \
750 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
751#define sk_for_each_from(__sk) \
752 hlist_for_each_entry_from(__sk, sk_node)
753#define sk_nulls_for_each_from(__sk, node) \
754 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
755 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
756#define sk_for_each_safe(__sk, tmp, list) \
757 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
758#define sk_for_each_bound(__sk, list) \
759 hlist_for_each_entry(__sk, list, sk_bind_node)
760
761/**
762 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
763 * @tpos: the type * to use as a loop cursor.
764 * @pos: the &struct hlist_node to use as a loop cursor.
765 * @head: the head for your list.
766 * @offset: offset of hlist_node within the struct.
767 *
768 */
769#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
770 for (pos = rcu_dereference(hlist_first_rcu(head)); \
771 pos != NULL && \
772 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
773 pos = rcu_dereference(hlist_next_rcu(pos)))
774
775static inline struct user_namespace *sk_user_ns(struct sock *sk)
776{
777 /* Careful only use this in a context where these parameters
778 * can not change and must all be valid, such as recvmsg from
779 * userspace.
780 */
781 return sk->sk_socket->file->f_cred->user_ns;
782}
783
784/* Sock flags */
785enum sock_flags {
786 SOCK_DEAD,
787 SOCK_DONE,
788 SOCK_URGINLINE,
789 SOCK_KEEPOPEN,
790 SOCK_LINGER,
791 SOCK_DESTROY,
792 SOCK_BROADCAST,
793 SOCK_TIMESTAMP,
794 SOCK_ZAPPED,
795 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
796 SOCK_DBG, /* %SO_DEBUG setting */
797 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
798 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
799 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
800 SOCK_QUEUE_SHRUNK, /* write queue has been shrunk recently */
801 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
802 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
803 SOCK_FASYNC, /* fasync() active */
804 SOCK_RXQ_OVFL,
805 SOCK_ZEROCOPY, /* buffers from userspace */
806 SOCK_WIFI_STATUS, /* push wifi status to userspace */
807 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
808 * Will use last 4 bytes of packet sent from
809 * user-space instead.
810 */
811 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
812 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
813 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
814 SOCK_TXTIME,
815 SOCK_XDP, /* XDP is attached */
816 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
817};
818
819#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
820
821static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
822{
823 nsk->sk_flags = osk->sk_flags;
824}
825
826static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
827{
828 __set_bit(flag, &sk->sk_flags);
829}
830
831static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
832{
833 __clear_bit(flag, &sk->sk_flags);
834}
835
836static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
837{
838 return test_bit(flag, &sk->sk_flags);
839}
840
841#ifdef CONFIG_NET
842DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
843static inline int sk_memalloc_socks(void)
844{
845 return static_branch_unlikely(&memalloc_socks_key);
846}
847#else
848
849static inline int sk_memalloc_socks(void)
850{
851 return 0;
852}
853
854#endif
855
856static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
857{
858 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
859}
860
861static inline void sk_acceptq_removed(struct sock *sk)
862{
863 sk->sk_ack_backlog--;
864}
865
866static inline void sk_acceptq_added(struct sock *sk)
867{
868 sk->sk_ack_backlog++;
869}
870
871static inline bool sk_acceptq_is_full(const struct sock *sk)
872{
873 return sk->sk_ack_backlog > sk->sk_max_ack_backlog;
874}
875
876/*
877 * Compute minimal free write space needed to queue new packets.
878 */
879static inline int sk_stream_min_wspace(const struct sock *sk)
880{
881 return READ_ONCE(sk->sk_wmem_queued) >> 1;
882}
883
884static inline int sk_stream_wspace(const struct sock *sk)
885{
886 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
887}
888
889static inline void sk_wmem_queued_add(struct sock *sk, int val)
890{
891 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
892}
893
894void sk_stream_write_space(struct sock *sk);
895
896/* OOB backlog add */
897static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
898{
899 /* dont let skb dst not refcounted, we are going to leave rcu lock */
900 skb_dst_force(skb);
901
902 if (!sk->sk_backlog.tail)
903 sk->sk_backlog.head = skb;
904 else
905 sk->sk_backlog.tail->next = skb;
906
907 sk->sk_backlog.tail = skb;
908 skb->next = NULL;
909}
910
911/*
912 * Take into account size of receive queue and backlog queue
913 * Do not take into account this skb truesize,
914 * to allow even a single big packet to come.
915 */
916static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
917{
918 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
919
920 return qsize > limit;
921}
922
923/* The per-socket spinlock must be held here. */
924static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
925 unsigned int limit)
926{
927 if (sk_rcvqueues_full(sk, limit))
928 return -ENOBUFS;
929
930 /*
931 * If the skb was allocated from pfmemalloc reserves, only
932 * allow SOCK_MEMALLOC sockets to use it as this socket is
933 * helping free memory
934 */
935 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
936 return -ENOMEM;
937
938 __sk_add_backlog(sk, skb);
939 sk->sk_backlog.len += skb->truesize;
940 return 0;
941}
942
943int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
944
945static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
946{
947 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
948 return __sk_backlog_rcv(sk, skb);
949
950 return sk->sk_backlog_rcv(sk, skb);
951}
952
953static inline void sk_incoming_cpu_update(struct sock *sk)
954{
955 int cpu = raw_smp_processor_id();
956
957 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
958 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
959}
960
961static inline void sock_rps_record_flow_hash(__u32 hash)
962{
963#ifdef CONFIG_RPS
964 struct rps_sock_flow_table *sock_flow_table;
965
966 rcu_read_lock();
967 sock_flow_table = rcu_dereference(rps_sock_flow_table);
968 rps_record_sock_flow(sock_flow_table, hash);
969 rcu_read_unlock();
970#endif
971}
972
973static inline void sock_rps_record_flow(const struct sock *sk)
974{
975#ifdef CONFIG_RPS
976 if (static_branch_unlikely(&rfs_needed)) {
977 /* Reading sk->sk_rxhash might incur an expensive cache line
978 * miss.
979 *
980 * TCP_ESTABLISHED does cover almost all states where RFS
981 * might be useful, and is cheaper [1] than testing :
982 * IPv4: inet_sk(sk)->inet_daddr
983 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
984 * OR an additional socket flag
985 * [1] : sk_state and sk_prot are in the same cache line.
986 */
987 if (sk->sk_state == TCP_ESTABLISHED)
988 sock_rps_record_flow_hash(sk->sk_rxhash);
989 }
990#endif
991}
992
993static inline void sock_rps_save_rxhash(struct sock *sk,
994 const struct sk_buff *skb)
995{
996#ifdef CONFIG_RPS
997 if (unlikely(sk->sk_rxhash != skb->hash))
998 sk->sk_rxhash = skb->hash;
999#endif
1000}
1001
1002static inline void sock_rps_reset_rxhash(struct sock *sk)
1003{
1004#ifdef CONFIG_RPS
1005 sk->sk_rxhash = 0;
1006#endif
1007}
1008
1009#define sk_wait_event(__sk, __timeo, __condition, __wait) \
1010 ({ int __rc; \
1011 release_sock(__sk); \
1012 __rc = __condition; \
1013 if (!__rc) { \
1014 *(__timeo) = wait_woken(__wait, \
1015 TASK_INTERRUPTIBLE, \
1016 *(__timeo)); \
1017 } \
1018 sched_annotate_sleep(); \
1019 lock_sock(__sk); \
1020 __rc = __condition; \
1021 __rc; \
1022 })
1023
1024int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1025int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1026void sk_stream_wait_close(struct sock *sk, long timeo_p);
1027int sk_stream_error(struct sock *sk, int flags, int err);
1028void sk_stream_kill_queues(struct sock *sk);
1029void sk_set_memalloc(struct sock *sk);
1030void sk_clear_memalloc(struct sock *sk);
1031
1032void __sk_flush_backlog(struct sock *sk);
1033
1034static inline bool sk_flush_backlog(struct sock *sk)
1035{
1036 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1037 __sk_flush_backlog(sk);
1038 return true;
1039 }
1040 return false;
1041}
1042
1043int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1044
1045struct request_sock_ops;
1046struct timewait_sock_ops;
1047struct inet_hashinfo;
1048struct raw_hashinfo;
1049struct smc_hashinfo;
1050struct module;
1051
1052/*
1053 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1054 * un-modified. Special care is taken when initializing object to zero.
1055 */
1056static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1057{
1058 if (offsetof(struct sock, sk_node.next) != 0)
1059 memset(sk, 0, offsetof(struct sock, sk_node.next));
1060 memset(&sk->sk_node.pprev, 0,
1061 size - offsetof(struct sock, sk_node.pprev));
1062}
1063
1064/* Networking protocol blocks we attach to sockets.
1065 * socket layer -> transport layer interface
1066 */
1067struct proto {
1068 void (*close)(struct sock *sk,
1069 long timeout);
1070 int (*pre_connect)(struct sock *sk,
1071 struct sockaddr *uaddr,
1072 int addr_len);
1073 int (*connect)(struct sock *sk,
1074 struct sockaddr *uaddr,
1075 int addr_len);
1076 int (*disconnect)(struct sock *sk, int flags);
1077
1078 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1079 bool kern);
1080
1081 int (*ioctl)(struct sock *sk, int cmd,
1082 unsigned long arg);
1083 int (*init)(struct sock *sk);
1084 void (*destroy)(struct sock *sk);
1085 void (*shutdown)(struct sock *sk, int how);
1086 int (*setsockopt)(struct sock *sk, int level,
1087 int optname, char __user *optval,
1088 unsigned int optlen);
1089 int (*getsockopt)(struct sock *sk, int level,
1090 int optname, char __user *optval,
1091 int __user *option);
1092 void (*keepalive)(struct sock *sk, int valbool);
1093#ifdef CONFIG_COMPAT
1094 int (*compat_setsockopt)(struct sock *sk,
1095 int level,
1096 int optname, char __user *optval,
1097 unsigned int optlen);
1098 int (*compat_getsockopt)(struct sock *sk,
1099 int level,
1100 int optname, char __user *optval,
1101 int __user *option);
1102 int (*compat_ioctl)(struct sock *sk,
1103 unsigned int cmd, unsigned long arg);
1104#endif
1105 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1106 size_t len);
1107 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1108 size_t len, int noblock, int flags,
1109 int *addr_len);
1110 int (*sendpage)(struct sock *sk, struct page *page,
1111 int offset, size_t size, int flags);
1112 int (*bind)(struct sock *sk,
1113 struct sockaddr *uaddr, int addr_len);
1114
1115 int (*backlog_rcv) (struct sock *sk,
1116 struct sk_buff *skb);
1117
1118 void (*release_cb)(struct sock *sk);
1119
1120 /* Keeping track of sk's, looking them up, and port selection methods. */
1121 int (*hash)(struct sock *sk);
1122 void (*unhash)(struct sock *sk);
1123 void (*rehash)(struct sock *sk);
1124 int (*get_port)(struct sock *sk, unsigned short snum);
1125
1126 /* Keeping track of sockets in use */
1127#ifdef CONFIG_PROC_FS
1128 unsigned int inuse_idx;
1129#endif
1130
1131 bool (*stream_memory_free)(const struct sock *sk, int wake);
1132 bool (*stream_memory_read)(const struct sock *sk);
1133 /* Memory pressure */
1134 void (*enter_memory_pressure)(struct sock *sk);
1135 void (*leave_memory_pressure)(struct sock *sk);
1136 atomic_long_t *memory_allocated; /* Current allocated memory. */
1137 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1138 /*
1139 * Pressure flag: try to collapse.
1140 * Technical note: it is used by multiple contexts non atomically.
1141 * All the __sk_mem_schedule() is of this nature: accounting
1142 * is strict, actions are advisory and have some latency.
1143 */
1144 unsigned long *memory_pressure;
1145 long *sysctl_mem;
1146
1147 int *sysctl_wmem;
1148 int *sysctl_rmem;
1149 u32 sysctl_wmem_offset;
1150 u32 sysctl_rmem_offset;
1151
1152 int max_header;
1153 bool no_autobind;
1154
1155 struct kmem_cache *slab;
1156 unsigned int obj_size;
1157 slab_flags_t slab_flags;
1158 unsigned int useroffset; /* Usercopy region offset */
1159 unsigned int usersize; /* Usercopy region size */
1160
1161 struct percpu_counter *orphan_count;
1162
1163 struct request_sock_ops *rsk_prot;
1164 struct timewait_sock_ops *twsk_prot;
1165
1166 union {
1167 struct inet_hashinfo *hashinfo;
1168 struct udp_table *udp_table;
1169 struct raw_hashinfo *raw_hash;
1170 struct smc_hashinfo *smc_hash;
1171 } h;
1172
1173 struct module *owner;
1174
1175 char name[32];
1176
1177 struct list_head node;
1178#ifdef SOCK_REFCNT_DEBUG
1179 atomic_t socks;
1180#endif
1181 int (*diag_destroy)(struct sock *sk, int err);
1182} __randomize_layout;
1183
1184int proto_register(struct proto *prot, int alloc_slab);
1185void proto_unregister(struct proto *prot);
1186int sock_load_diag_module(int family, int protocol);
1187
1188#ifdef SOCK_REFCNT_DEBUG
1189static inline void sk_refcnt_debug_inc(struct sock *sk)
1190{
1191 atomic_inc(&sk->sk_prot->socks);
1192}
1193
1194static inline void sk_refcnt_debug_dec(struct sock *sk)
1195{
1196 atomic_dec(&sk->sk_prot->socks);
1197 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1198 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1199}
1200
1201static inline void sk_refcnt_debug_release(const struct sock *sk)
1202{
1203 if (refcount_read(&sk->sk_refcnt) != 1)
1204 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1205 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1206}
1207#else /* SOCK_REFCNT_DEBUG */
1208#define sk_refcnt_debug_inc(sk) do { } while (0)
1209#define sk_refcnt_debug_dec(sk) do { } while (0)
1210#define sk_refcnt_debug_release(sk) do { } while (0)
1211#endif /* SOCK_REFCNT_DEBUG */
1212
1213static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1214{
1215 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1216 return false;
1217
1218 return sk->sk_prot->stream_memory_free ?
1219 sk->sk_prot->stream_memory_free(sk, wake) : true;
1220}
1221
1222static inline bool sk_stream_memory_free(const struct sock *sk)
1223{
1224 return __sk_stream_memory_free(sk, 0);
1225}
1226
1227static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1228{
1229 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1230 __sk_stream_memory_free(sk, wake);
1231}
1232
1233static inline bool sk_stream_is_writeable(const struct sock *sk)
1234{
1235 return __sk_stream_is_writeable(sk, 0);
1236}
1237
1238static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1239 struct cgroup *ancestor)
1240{
1241#ifdef CONFIG_SOCK_CGROUP_DATA
1242 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1243 ancestor);
1244#else
1245 return -ENOTSUPP;
1246#endif
1247}
1248
1249static inline bool sk_has_memory_pressure(const struct sock *sk)
1250{
1251 return sk->sk_prot->memory_pressure != NULL;
1252}
1253
1254static inline bool sk_under_memory_pressure(const struct sock *sk)
1255{
1256 if (!sk->sk_prot->memory_pressure)
1257 return false;
1258
1259 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1260 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1261 return true;
1262
1263 return !!*sk->sk_prot->memory_pressure;
1264}
1265
1266static inline long
1267sk_memory_allocated(const struct sock *sk)
1268{
1269 return atomic_long_read(sk->sk_prot->memory_allocated);
1270}
1271
1272static inline long
1273sk_memory_allocated_add(struct sock *sk, int amt)
1274{
1275 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1276}
1277
1278static inline void
1279sk_memory_allocated_sub(struct sock *sk, int amt)
1280{
1281 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1282}
1283
1284static inline void sk_sockets_allocated_dec(struct sock *sk)
1285{
1286 percpu_counter_dec(sk->sk_prot->sockets_allocated);
1287}
1288
1289static inline void sk_sockets_allocated_inc(struct sock *sk)
1290{
1291 percpu_counter_inc(sk->sk_prot->sockets_allocated);
1292}
1293
1294static inline u64
1295sk_sockets_allocated_read_positive(struct sock *sk)
1296{
1297 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1298}
1299
1300static inline int
1301proto_sockets_allocated_sum_positive(struct proto *prot)
1302{
1303 return percpu_counter_sum_positive(prot->sockets_allocated);
1304}
1305
1306static inline long
1307proto_memory_allocated(struct proto *prot)
1308{
1309 return atomic_long_read(prot->memory_allocated);
1310}
1311
1312static inline bool
1313proto_memory_pressure(struct proto *prot)
1314{
1315 if (!prot->memory_pressure)
1316 return false;
1317 return !!*prot->memory_pressure;
1318}
1319
1320
1321#ifdef CONFIG_PROC_FS
1322/* Called with local bh disabled */
1323void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1324int sock_prot_inuse_get(struct net *net, struct proto *proto);
1325int sock_inuse_get(struct net *net);
1326#else
1327static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1328 int inc)
1329{
1330}
1331#endif
1332
1333
1334/* With per-bucket locks this operation is not-atomic, so that
1335 * this version is not worse.
1336 */
1337static inline int __sk_prot_rehash(struct sock *sk)
1338{
1339 sk->sk_prot->unhash(sk);
1340 return sk->sk_prot->hash(sk);
1341}
1342
1343/* About 10 seconds */
1344#define SOCK_DESTROY_TIME (10*HZ)
1345
1346/* Sockets 0-1023 can't be bound to unless you are superuser */
1347#define PROT_SOCK 1024
1348
1349#define SHUTDOWN_MASK 3
1350#define RCV_SHUTDOWN 1
1351#define SEND_SHUTDOWN 2
1352
1353#define SOCK_SNDBUF_LOCK 1
1354#define SOCK_RCVBUF_LOCK 2
1355#define SOCK_BINDADDR_LOCK 4
1356#define SOCK_BINDPORT_LOCK 8
1357
1358struct socket_alloc {
1359 struct socket socket;
1360 struct inode vfs_inode;
1361};
1362
1363static inline struct socket *SOCKET_I(struct inode *inode)
1364{
1365 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1366}
1367
1368static inline struct inode *SOCK_INODE(struct socket *socket)
1369{
1370 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1371}
1372
1373/*
1374 * Functions for memory accounting
1375 */
1376int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1377int __sk_mem_schedule(struct sock *sk, int size, int kind);
1378void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1379void __sk_mem_reclaim(struct sock *sk, int amount);
1380
1381/* We used to have PAGE_SIZE here, but systems with 64KB pages
1382 * do not necessarily have 16x time more memory than 4KB ones.
1383 */
1384#define SK_MEM_QUANTUM 4096
1385#define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1386#define SK_MEM_SEND 0
1387#define SK_MEM_RECV 1
1388
1389/* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
1390static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1391{
1392 long val = sk->sk_prot->sysctl_mem[index];
1393
1394#if PAGE_SIZE > SK_MEM_QUANTUM
1395 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1396#elif PAGE_SIZE < SK_MEM_QUANTUM
1397 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1398#endif
1399 return val;
1400}
1401
1402static inline int sk_mem_pages(int amt)
1403{
1404 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1405}
1406
1407static inline bool sk_has_account(struct sock *sk)
1408{
1409 /* return true if protocol supports memory accounting */
1410 return !!sk->sk_prot->memory_allocated;
1411}
1412
1413static inline bool sk_wmem_schedule(struct sock *sk, int size)
1414{
1415 if (!sk_has_account(sk))
1416 return true;
1417 return size <= sk->sk_forward_alloc ||
1418 __sk_mem_schedule(sk, size, SK_MEM_SEND);
1419}
1420
1421static inline bool
1422sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1423{
1424 if (!sk_has_account(sk))
1425 return true;
1426 return size<= sk->sk_forward_alloc ||
1427 __sk_mem_schedule(sk, size, SK_MEM_RECV) ||
1428 skb_pfmemalloc(skb);
1429}
1430
1431static inline void sk_mem_reclaim(struct sock *sk)
1432{
1433 if (!sk_has_account(sk))
1434 return;
1435 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1436 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1437}
1438
1439static inline void sk_mem_reclaim_partial(struct sock *sk)
1440{
1441 if (!sk_has_account(sk))
1442 return;
1443 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1444 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1445}
1446
1447static inline void sk_mem_charge(struct sock *sk, int size)
1448{
1449 if (!sk_has_account(sk))
1450 return;
1451 sk->sk_forward_alloc -= size;
1452}
1453
1454static inline void sk_mem_uncharge(struct sock *sk, int size)
1455{
1456 if (!sk_has_account(sk))
1457 return;
1458 sk->sk_forward_alloc += size;
1459
1460 /* Avoid a possible overflow.
1461 * TCP send queues can make this happen, if sk_mem_reclaim()
1462 * is not called and more than 2 GBytes are released at once.
1463 *
1464 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1465 * no need to hold that much forward allocation anyway.
1466 */
1467 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1468 __sk_mem_reclaim(sk, 1 << 20);
1469}
1470
1471DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
1472static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1473{
1474 sock_set_flag(sk, SOCK_QUEUE_SHRUNK);
1475 sk_wmem_queued_add(sk, -skb->truesize);
1476 sk_mem_uncharge(sk, skb->truesize);
1477 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1478 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1479 skb_zcopy_clear(skb, true);
1480 sk->sk_tx_skb_cache = skb;
1481 return;
1482 }
1483 __kfree_skb(skb);
1484}
1485
1486static inline void sock_release_ownership(struct sock *sk)
1487{
1488 if (sk->sk_lock.owned) {
1489 sk->sk_lock.owned = 0;
1490
1491 /* The sk_lock has mutex_unlock() semantics: */
1492 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1493 }
1494}
1495
1496/*
1497 * Macro so as to not evaluate some arguments when
1498 * lockdep is not enabled.
1499 *
1500 * Mark both the sk_lock and the sk_lock.slock as a
1501 * per-address-family lock class.
1502 */
1503#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1504do { \
1505 sk->sk_lock.owned = 0; \
1506 init_waitqueue_head(&sk->sk_lock.wq); \
1507 spin_lock_init(&(sk)->sk_lock.slock); \
1508 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1509 sizeof((sk)->sk_lock)); \
1510 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1511 (skey), (sname)); \
1512 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1513} while (0)
1514
1515#ifdef CONFIG_LOCKDEP
1516static inline bool lockdep_sock_is_held(const struct sock *sk)
1517{
1518 return lockdep_is_held(&sk->sk_lock) ||
1519 lockdep_is_held(&sk->sk_lock.slock);
1520}
1521#endif
1522
1523void lock_sock_nested(struct sock *sk, int subclass);
1524
1525static inline void lock_sock(struct sock *sk)
1526{
1527 lock_sock_nested(sk, 0);
1528}
1529
1530void __release_sock(struct sock *sk);
1531void release_sock(struct sock *sk);
1532
1533/* BH context may only use the following locking interface. */
1534#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1535#define bh_lock_sock_nested(__sk) \
1536 spin_lock_nested(&((__sk)->sk_lock.slock), \
1537 SINGLE_DEPTH_NESTING)
1538#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1539
1540bool lock_sock_fast(struct sock *sk);
1541/**
1542 * unlock_sock_fast - complement of lock_sock_fast
1543 * @sk: socket
1544 * @slow: slow mode
1545 *
1546 * fast unlock socket for user context.
1547 * If slow mode is on, we call regular release_sock()
1548 */
1549static inline void unlock_sock_fast(struct sock *sk, bool slow)
1550{
1551 if (slow)
1552 release_sock(sk);
1553 else
1554 spin_unlock_bh(&sk->sk_lock.slock);
1555}
1556
1557/* Used by processes to "lock" a socket state, so that
1558 * interrupts and bottom half handlers won't change it
1559 * from under us. It essentially blocks any incoming
1560 * packets, so that we won't get any new data or any
1561 * packets that change the state of the socket.
1562 *
1563 * While locked, BH processing will add new packets to
1564 * the backlog queue. This queue is processed by the
1565 * owner of the socket lock right before it is released.
1566 *
1567 * Since ~2.3.5 it is also exclusive sleep lock serializing
1568 * accesses from user process context.
1569 */
1570
1571static inline void sock_owned_by_me(const struct sock *sk)
1572{
1573#ifdef CONFIG_LOCKDEP
1574 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1575#endif
1576}
1577
1578static inline bool sock_owned_by_user(const struct sock *sk)
1579{
1580 sock_owned_by_me(sk);
1581 return sk->sk_lock.owned;
1582}
1583
1584static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1585{
1586 return sk->sk_lock.owned;
1587}
1588
1589/* no reclassification while locks are held */
1590static inline bool sock_allow_reclassification(const struct sock *csk)
1591{
1592 struct sock *sk = (struct sock *)csk;
1593
1594 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1595}
1596
1597struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1598 struct proto *prot, int kern);
1599void sk_free(struct sock *sk);
1600void sk_destruct(struct sock *sk);
1601struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1602void sk_free_unlock_clone(struct sock *sk);
1603
1604struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1605 gfp_t priority);
1606void __sock_wfree(struct sk_buff *skb);
1607void sock_wfree(struct sk_buff *skb);
1608struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1609 gfp_t priority);
1610void skb_orphan_partial(struct sk_buff *skb);
1611void sock_rfree(struct sk_buff *skb);
1612void sock_efree(struct sk_buff *skb);
1613#ifdef CONFIG_INET
1614void sock_edemux(struct sk_buff *skb);
1615#else
1616#define sock_edemux sock_efree
1617#endif
1618
1619int sock_setsockopt(struct socket *sock, int level, int op,
1620 char __user *optval, unsigned int optlen);
1621
1622int sock_getsockopt(struct socket *sock, int level, int op,
1623 char __user *optval, int __user *optlen);
1624int sock_gettstamp(struct socket *sock, void __user *userstamp,
1625 bool timeval, bool time32);
1626struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1627 int noblock, int *errcode);
1628struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1629 unsigned long data_len, int noblock,
1630 int *errcode, int max_page_order);
1631void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1632void sock_kfree_s(struct sock *sk, void *mem, int size);
1633void sock_kzfree_s(struct sock *sk, void *mem, int size);
1634void sk_send_sigurg(struct sock *sk);
1635
1636struct sockcm_cookie {
1637 u64 transmit_time;
1638 u32 mark;
1639 u16 tsflags;
1640};
1641
1642static inline void sockcm_init(struct sockcm_cookie *sockc,
1643 const struct sock *sk)
1644{
1645 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1646}
1647
1648int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1649 struct sockcm_cookie *sockc);
1650int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1651 struct sockcm_cookie *sockc);
1652
1653/*
1654 * Functions to fill in entries in struct proto_ops when a protocol
1655 * does not implement a particular function.
1656 */
1657int sock_no_bind(struct socket *, struct sockaddr *, int);
1658int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1659int sock_no_socketpair(struct socket *, struct socket *);
1660int sock_no_accept(struct socket *, struct socket *, int, bool);
1661int sock_no_getname(struct socket *, struct sockaddr *, int);
1662int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1663int sock_no_listen(struct socket *, int);
1664int sock_no_shutdown(struct socket *, int);
1665int sock_no_getsockopt(struct socket *, int , int, char __user *, int __user *);
1666int sock_no_setsockopt(struct socket *, int, int, char __user *, unsigned int);
1667int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1668int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1669int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1670int sock_no_mmap(struct file *file, struct socket *sock,
1671 struct vm_area_struct *vma);
1672ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1673 size_t size, int flags);
1674ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1675 int offset, size_t size, int flags);
1676
1677/*
1678 * Functions to fill in entries in struct proto_ops when a protocol
1679 * uses the inet style.
1680 */
1681int sock_common_getsockopt(struct socket *sock, int level, int optname,
1682 char __user *optval, int __user *optlen);
1683int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1684 int flags);
1685int sock_common_setsockopt(struct socket *sock, int level, int optname,
1686 char __user *optval, unsigned int optlen);
1687int compat_sock_common_getsockopt(struct socket *sock, int level,
1688 int optname, char __user *optval, int __user *optlen);
1689int compat_sock_common_setsockopt(struct socket *sock, int level,
1690 int optname, char __user *optval, unsigned int optlen);
1691
1692void sk_common_release(struct sock *sk);
1693
1694/*
1695 * Default socket callbacks and setup code
1696 */
1697
1698/* Initialise core socket variables */
1699void sock_init_data(struct socket *sock, struct sock *sk);
1700
1701/*
1702 * Socket reference counting postulates.
1703 *
1704 * * Each user of socket SHOULD hold a reference count.
1705 * * Each access point to socket (an hash table bucket, reference from a list,
1706 * running timer, skb in flight MUST hold a reference count.
1707 * * When reference count hits 0, it means it will never increase back.
1708 * * When reference count hits 0, it means that no references from
1709 * outside exist to this socket and current process on current CPU
1710 * is last user and may/should destroy this socket.
1711 * * sk_free is called from any context: process, BH, IRQ. When
1712 * it is called, socket has no references from outside -> sk_free
1713 * may release descendant resources allocated by the socket, but
1714 * to the time when it is called, socket is NOT referenced by any
1715 * hash tables, lists etc.
1716 * * Packets, delivered from outside (from network or from another process)
1717 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1718 * when they sit in queue. Otherwise, packets will leak to hole, when
1719 * socket is looked up by one cpu and unhasing is made by another CPU.
1720 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1721 * (leak to backlog). Packet socket does all the processing inside
1722 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1723 * use separate SMP lock, so that they are prone too.
1724 */
1725
1726/* Ungrab socket and destroy it, if it was the last reference. */
1727static inline void sock_put(struct sock *sk)
1728{
1729 if (refcount_dec_and_test(&sk->sk_refcnt))
1730 sk_free(sk);
1731}
1732/* Generic version of sock_put(), dealing with all sockets
1733 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1734 */
1735void sock_gen_put(struct sock *sk);
1736
1737int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1738 unsigned int trim_cap, bool refcounted);
1739static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1740 const int nested)
1741{
1742 return __sk_receive_skb(sk, skb, nested, 1, true);
1743}
1744
1745static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1746{
1747 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1748 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1749 return;
1750 sk->sk_tx_queue_mapping = tx_queue;
1751}
1752
1753#define NO_QUEUE_MAPPING USHRT_MAX
1754
1755static inline void sk_tx_queue_clear(struct sock *sk)
1756{
1757 sk->sk_tx_queue_mapping = NO_QUEUE_MAPPING;
1758}
1759
1760static inline int sk_tx_queue_get(const struct sock *sk)
1761{
1762 if (sk && sk->sk_tx_queue_mapping != NO_QUEUE_MAPPING)
1763 return sk->sk_tx_queue_mapping;
1764
1765 return -1;
1766}
1767
1768static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1769{
1770#ifdef CONFIG_XPS
1771 if (skb_rx_queue_recorded(skb)) {
1772 u16 rx_queue = skb_get_rx_queue(skb);
1773
1774 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1775 return;
1776
1777 sk->sk_rx_queue_mapping = rx_queue;
1778 }
1779#endif
1780}
1781
1782static inline void sk_rx_queue_clear(struct sock *sk)
1783{
1784#ifdef CONFIG_XPS
1785 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1786#endif
1787}
1788
1789#ifdef CONFIG_XPS
1790static inline int sk_rx_queue_get(const struct sock *sk)
1791{
1792 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1793 return sk->sk_rx_queue_mapping;
1794
1795 return -1;
1796}
1797#endif
1798
1799static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1800{
1801 sk_tx_queue_clear(sk);
1802 sk->sk_socket = sock;
1803}
1804
1805static inline wait_queue_head_t *sk_sleep(struct sock *sk)
1806{
1807 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
1808 return &rcu_dereference_raw(sk->sk_wq)->wait;
1809}
1810/* Detach socket from process context.
1811 * Announce socket dead, detach it from wait queue and inode.
1812 * Note that parent inode held reference count on this struct sock,
1813 * we do not release it in this function, because protocol
1814 * probably wants some additional cleanups or even continuing
1815 * to work with this socket (TCP).
1816 */
1817static inline void sock_orphan(struct sock *sk)
1818{
1819 write_lock_bh(&sk->sk_callback_lock);
1820 sock_set_flag(sk, SOCK_DEAD);
1821 sk_set_socket(sk, NULL);
1822 sk->sk_wq = NULL;
1823 write_unlock_bh(&sk->sk_callback_lock);
1824}
1825
1826static inline void sock_graft(struct sock *sk, struct socket *parent)
1827{
1828 WARN_ON(parent->sk);
1829 write_lock_bh(&sk->sk_callback_lock);
1830 rcu_assign_pointer(sk->sk_wq, &parent->wq);
1831 parent->sk = sk;
1832 sk_set_socket(sk, parent);
1833 sk->sk_uid = SOCK_INODE(parent)->i_uid;
1834 security_sock_graft(sk, parent);
1835 write_unlock_bh(&sk->sk_callback_lock);
1836}
1837
1838kuid_t sock_i_uid(struct sock *sk);
1839unsigned long sock_i_ino(struct sock *sk);
1840
1841static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
1842{
1843 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
1844}
1845
1846static inline u32 net_tx_rndhash(void)
1847{
1848 u32 v = prandom_u32();
1849
1850 return v ?: 1;
1851}
1852
1853static inline void sk_set_txhash(struct sock *sk)
1854{
1855 sk->sk_txhash = net_tx_rndhash();
1856}
1857
1858static inline void sk_rethink_txhash(struct sock *sk)
1859{
1860 if (sk->sk_txhash)
1861 sk_set_txhash(sk);
1862}
1863
1864static inline struct dst_entry *
1865__sk_dst_get(struct sock *sk)
1866{
1867 return rcu_dereference_check(sk->sk_dst_cache,
1868 lockdep_sock_is_held(sk));
1869}
1870
1871static inline struct dst_entry *
1872sk_dst_get(struct sock *sk)
1873{
1874 struct dst_entry *dst;
1875
1876 rcu_read_lock();
1877 dst = rcu_dereference(sk->sk_dst_cache);
1878 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
1879 dst = NULL;
1880 rcu_read_unlock();
1881 return dst;
1882}
1883
1884static inline void dst_negative_advice(struct sock *sk)
1885{
1886 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
1887
1888 sk_rethink_txhash(sk);
1889
1890 if (dst && dst->ops->negative_advice) {
1891 ndst = dst->ops->negative_advice(dst);
1892
1893 if (ndst != dst) {
1894 rcu_assign_pointer(sk->sk_dst_cache, ndst);
1895 sk_tx_queue_clear(sk);
1896 sk->sk_dst_pending_confirm = 0;
1897 }
1898 }
1899}
1900
1901static inline void
1902__sk_dst_set(struct sock *sk, struct dst_entry *dst)
1903{
1904 struct dst_entry *old_dst;
1905
1906 sk_tx_queue_clear(sk);
1907 sk->sk_dst_pending_confirm = 0;
1908 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
1909 lockdep_sock_is_held(sk));
1910 rcu_assign_pointer(sk->sk_dst_cache, dst);
1911 dst_release(old_dst);
1912}
1913
1914static inline void
1915sk_dst_set(struct sock *sk, struct dst_entry *dst)
1916{
1917 struct dst_entry *old_dst;
1918
1919 sk_tx_queue_clear(sk);
1920 sk->sk_dst_pending_confirm = 0;
1921 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
1922 dst_release(old_dst);
1923}
1924
1925static inline void
1926__sk_dst_reset(struct sock *sk)
1927{
1928 __sk_dst_set(sk, NULL);
1929}
1930
1931static inline void
1932sk_dst_reset(struct sock *sk)
1933{
1934 sk_dst_set(sk, NULL);
1935}
1936
1937struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
1938
1939struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
1940
1941static inline void sk_dst_confirm(struct sock *sk)
1942{
1943 if (!sk->sk_dst_pending_confirm)
1944 sk->sk_dst_pending_confirm = 1;
1945}
1946
1947static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
1948{
1949 if (skb_get_dst_pending_confirm(skb)) {
1950 struct sock *sk = skb->sk;
1951 unsigned long now = jiffies;
1952
1953 /* avoid dirtying neighbour */
1954 if (n->confirmed != now)
1955 n->confirmed = now;
1956 if (sk && sk->sk_dst_pending_confirm)
1957 sk->sk_dst_pending_confirm = 0;
1958 }
1959}
1960
1961bool sk_mc_loop(struct sock *sk);
1962
1963static inline bool sk_can_gso(const struct sock *sk)
1964{
1965 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
1966}
1967
1968void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
1969
1970static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
1971{
1972 sk->sk_route_nocaps |= flags;
1973 sk->sk_route_caps &= ~flags;
1974}
1975
1976static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
1977 struct iov_iter *from, char *to,
1978 int copy, int offset)
1979{
1980 if (skb->ip_summed == CHECKSUM_NONE) {
1981 __wsum csum = 0;
1982 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
1983 return -EFAULT;
1984 skb->csum = csum_block_add(skb->csum, csum, offset);
1985 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
1986 if (!copy_from_iter_full_nocache(to, copy, from))
1987 return -EFAULT;
1988 } else if (!copy_from_iter_full(to, copy, from))
1989 return -EFAULT;
1990
1991 return 0;
1992}
1993
1994static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
1995 struct iov_iter *from, int copy)
1996{
1997 int err, offset = skb->len;
1998
1999 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2000 copy, offset);
2001 if (err)
2002 __skb_trim(skb, offset);
2003
2004 return err;
2005}
2006
2007static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2008 struct sk_buff *skb,
2009 struct page *page,
2010 int off, int copy)
2011{
2012 int err;
2013
2014 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2015 copy, skb->len);
2016 if (err)
2017 return err;
2018
2019 skb->len += copy;
2020 skb->data_len += copy;
2021 skb->truesize += copy;
2022 sk_wmem_queued_add(sk, copy);
2023 sk_mem_charge(sk, copy);
2024 return 0;
2025}
2026
2027/**
2028 * sk_wmem_alloc_get - returns write allocations
2029 * @sk: socket
2030 *
2031 * Returns sk_wmem_alloc minus initial offset of one
2032 */
2033static inline int sk_wmem_alloc_get(const struct sock *sk)
2034{
2035 return refcount_read(&sk->sk_wmem_alloc) - 1;
2036}
2037
2038/**
2039 * sk_rmem_alloc_get - returns read allocations
2040 * @sk: socket
2041 *
2042 * Returns sk_rmem_alloc
2043 */
2044static inline int sk_rmem_alloc_get(const struct sock *sk)
2045{
2046 return atomic_read(&sk->sk_rmem_alloc);
2047}
2048
2049/**
2050 * sk_has_allocations - check if allocations are outstanding
2051 * @sk: socket
2052 *
2053 * Returns true if socket has write or read allocations
2054 */
2055static inline bool sk_has_allocations(const struct sock *sk)
2056{
2057 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2058}
2059
2060/**
2061 * skwq_has_sleeper - check if there are any waiting processes
2062 * @wq: struct socket_wq
2063 *
2064 * Returns true if socket_wq has waiting processes
2065 *
2066 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2067 * barrier call. They were added due to the race found within the tcp code.
2068 *
2069 * Consider following tcp code paths::
2070 *
2071 * CPU1 CPU2
2072 * sys_select receive packet
2073 * ... ...
2074 * __add_wait_queue update tp->rcv_nxt
2075 * ... ...
2076 * tp->rcv_nxt check sock_def_readable
2077 * ... {
2078 * schedule rcu_read_lock();
2079 * wq = rcu_dereference(sk->sk_wq);
2080 * if (wq && waitqueue_active(&wq->wait))
2081 * wake_up_interruptible(&wq->wait)
2082 * ...
2083 * }
2084 *
2085 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2086 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2087 * could then endup calling schedule and sleep forever if there are no more
2088 * data on the socket.
2089 *
2090 */
2091static inline bool skwq_has_sleeper(struct socket_wq *wq)
2092{
2093 return wq && wq_has_sleeper(&wq->wait);
2094}
2095
2096/**
2097 * sock_poll_wait - place memory barrier behind the poll_wait call.
2098 * @filp: file
2099 * @sock: socket to wait on
2100 * @p: poll_table
2101 *
2102 * See the comments in the wq_has_sleeper function.
2103 */
2104static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2105 poll_table *p)
2106{
2107 if (!poll_does_not_wait(p)) {
2108 poll_wait(filp, &sock->wq.wait, p);
2109 /* We need to be sure we are in sync with the
2110 * socket flags modification.
2111 *
2112 * This memory barrier is paired in the wq_has_sleeper.
2113 */
2114 smp_mb();
2115 }
2116}
2117
2118static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2119{
2120 if (sk->sk_txhash) {
2121 skb->l4_hash = 1;
2122 skb->hash = sk->sk_txhash;
2123 }
2124}
2125
2126void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2127
2128/*
2129 * Queue a received datagram if it will fit. Stream and sequenced
2130 * protocols can't normally use this as they need to fit buffers in
2131 * and play with them.
2132 *
2133 * Inlined as it's very short and called for pretty much every
2134 * packet ever received.
2135 */
2136static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2137{
2138 skb_orphan(skb);
2139 skb->sk = sk;
2140 skb->destructor = sock_rfree;
2141 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2142 sk_mem_charge(sk, skb->truesize);
2143}
2144
2145void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2146 unsigned long expires);
2147
2148void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2149
2150int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2151 struct sk_buff *skb, unsigned int flags,
2152 void (*destructor)(struct sock *sk,
2153 struct sk_buff *skb));
2154int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2155int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2156
2157int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2158struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2159
2160/*
2161 * Recover an error report and clear atomically
2162 */
2163
2164static inline int sock_error(struct sock *sk)
2165{
2166 int err;
2167 if (likely(!sk->sk_err))
2168 return 0;
2169 err = xchg(&sk->sk_err, 0);
2170 return -err;
2171}
2172
2173static inline unsigned long sock_wspace(struct sock *sk)
2174{
2175 int amt = 0;
2176
2177 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2178 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2179 if (amt < 0)
2180 amt = 0;
2181 }
2182 return amt;
2183}
2184
2185/* Note:
2186 * We use sk->sk_wq_raw, from contexts knowing this
2187 * pointer is not NULL and cannot disappear/change.
2188 */
2189static inline void sk_set_bit(int nr, struct sock *sk)
2190{
2191 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2192 !sock_flag(sk, SOCK_FASYNC))
2193 return;
2194
2195 set_bit(nr, &sk->sk_wq_raw->flags);
2196}
2197
2198static inline void sk_clear_bit(int nr, struct sock *sk)
2199{
2200 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2201 !sock_flag(sk, SOCK_FASYNC))
2202 return;
2203
2204 clear_bit(nr, &sk->sk_wq_raw->flags);
2205}
2206
2207static inline void sk_wake_async(const struct sock *sk, int how, int band)
2208{
2209 if (sock_flag(sk, SOCK_FASYNC)) {
2210 rcu_read_lock();
2211 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2212 rcu_read_unlock();
2213 }
2214}
2215
2216/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2217 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2218 * Note: for send buffers, TCP works better if we can build two skbs at
2219 * minimum.
2220 */
2221#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2222
2223#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2224#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2225
2226static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2227{
2228 u32 val;
2229
2230 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2231 return;
2232
2233 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2234
2235 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2236}
2237
2238struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2239 bool force_schedule);
2240
2241/**
2242 * sk_page_frag - return an appropriate page_frag
2243 * @sk: socket
2244 *
2245 * Use the per task page_frag instead of the per socket one for
2246 * optimization when we know that we're in the normal context and owns
2247 * everything that's associated with %current.
2248 *
2249 * gfpflags_allow_blocking() isn't enough here as direct reclaim may nest
2250 * inside other socket operations and end up recursing into sk_page_frag()
2251 * while it's already in use.
2252 */
2253static inline struct page_frag *sk_page_frag(struct sock *sk)
2254{
2255 if (gfpflags_normal_context(sk->sk_allocation))
2256 return ¤t->task_frag;
2257
2258 return &sk->sk_frag;
2259}
2260
2261bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2262
2263/*
2264 * Default write policy as shown to user space via poll/select/SIGIO
2265 */
2266static inline bool sock_writeable(const struct sock *sk)
2267{
2268 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2269}
2270
2271static inline gfp_t gfp_any(void)
2272{
2273 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2274}
2275
2276static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2277{
2278 return noblock ? 0 : sk->sk_rcvtimeo;
2279}
2280
2281static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2282{
2283 return noblock ? 0 : sk->sk_sndtimeo;
2284}
2285
2286static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2287{
2288 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2289
2290 return v ?: 1;
2291}
2292
2293/* Alas, with timeout socket operations are not restartable.
2294 * Compare this to poll().
2295 */
2296static inline int sock_intr_errno(long timeo)
2297{
2298 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2299}
2300
2301struct sock_skb_cb {
2302 u32 dropcount;
2303};
2304
2305/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2306 * using skb->cb[] would keep using it directly and utilize its
2307 * alignement guarantee.
2308 */
2309#define SOCK_SKB_CB_OFFSET ((FIELD_SIZEOF(struct sk_buff, cb) - \
2310 sizeof(struct sock_skb_cb)))
2311
2312#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2313 SOCK_SKB_CB_OFFSET))
2314
2315#define sock_skb_cb_check_size(size) \
2316 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2317
2318static inline void
2319sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2320{
2321 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2322 atomic_read(&sk->sk_drops) : 0;
2323}
2324
2325static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2326{
2327 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2328
2329 atomic_add(segs, &sk->sk_drops);
2330}
2331
2332static inline ktime_t sock_read_timestamp(struct sock *sk)
2333{
2334#if BITS_PER_LONG==32
2335 unsigned int seq;
2336 ktime_t kt;
2337
2338 do {
2339 seq = read_seqbegin(&sk->sk_stamp_seq);
2340 kt = sk->sk_stamp;
2341 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2342
2343 return kt;
2344#else
2345 return READ_ONCE(sk->sk_stamp);
2346#endif
2347}
2348
2349static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2350{
2351#if BITS_PER_LONG==32
2352 write_seqlock(&sk->sk_stamp_seq);
2353 sk->sk_stamp = kt;
2354 write_sequnlock(&sk->sk_stamp_seq);
2355#else
2356 WRITE_ONCE(sk->sk_stamp, kt);
2357#endif
2358}
2359
2360void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2361 struct sk_buff *skb);
2362void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2363 struct sk_buff *skb);
2364
2365static inline void
2366sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2367{
2368 ktime_t kt = skb->tstamp;
2369 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2370
2371 /*
2372 * generate control messages if
2373 * - receive time stamping in software requested
2374 * - software time stamp available and wanted
2375 * - hardware time stamps available and wanted
2376 */
2377 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2378 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2379 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2380 (hwtstamps->hwtstamp &&
2381 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2382 __sock_recv_timestamp(msg, sk, skb);
2383 else
2384 sock_write_timestamp(sk, kt);
2385
2386 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2387 __sock_recv_wifi_status(msg, sk, skb);
2388}
2389
2390void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2391 struct sk_buff *skb);
2392
2393#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2394static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2395 struct sk_buff *skb)
2396{
2397#define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2398 (1UL << SOCK_RCVTSTAMP))
2399#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2400 SOF_TIMESTAMPING_RAW_HARDWARE)
2401
2402 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2403 __sock_recv_ts_and_drops(msg, sk, skb);
2404 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2405 sock_write_timestamp(sk, skb->tstamp);
2406 else if (unlikely(sk->sk_stamp == SK_DEFAULT_STAMP))
2407 sock_write_timestamp(sk, 0);
2408}
2409
2410void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2411
2412/**
2413 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2414 * @sk: socket sending this packet
2415 * @tsflags: timestamping flags to use
2416 * @tx_flags: completed with instructions for time stamping
2417 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2418 *
2419 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2420 */
2421static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2422 __u8 *tx_flags, __u32 *tskey)
2423{
2424 if (unlikely(tsflags)) {
2425 __sock_tx_timestamp(tsflags, tx_flags);
2426 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2427 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2428 *tskey = sk->sk_tskey++;
2429 }
2430 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2431 *tx_flags |= SKBTX_WIFI_STATUS;
2432}
2433
2434static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2435 __u8 *tx_flags)
2436{
2437 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2438}
2439
2440static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2441{
2442 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2443 &skb_shinfo(skb)->tskey);
2444}
2445
2446/**
2447 * sk_eat_skb - Release a skb if it is no longer needed
2448 * @sk: socket to eat this skb from
2449 * @skb: socket buffer to eat
2450 *
2451 * This routine must be called with interrupts disabled or with the socket
2452 * locked so that the sk_buff queue operation is ok.
2453*/
2454DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2455static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2456{
2457 __skb_unlink(skb, &sk->sk_receive_queue);
2458 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2459 !sk->sk_rx_skb_cache) {
2460 sk->sk_rx_skb_cache = skb;
2461 skb_orphan(skb);
2462 return;
2463 }
2464 __kfree_skb(skb);
2465}
2466
2467static inline
2468struct net *sock_net(const struct sock *sk)
2469{
2470 return read_pnet(&sk->sk_net);
2471}
2472
2473static inline
2474void sock_net_set(struct sock *sk, struct net *net)
2475{
2476 write_pnet(&sk->sk_net, net);
2477}
2478
2479static inline struct sock *skb_steal_sock(struct sk_buff *skb)
2480{
2481 if (skb->sk) {
2482 struct sock *sk = skb->sk;
2483
2484 skb->destructor = NULL;
2485 skb->sk = NULL;
2486 return sk;
2487 }
2488 return NULL;
2489}
2490
2491/* This helper checks if a socket is a full socket,
2492 * ie _not_ a timewait or request socket.
2493 */
2494static inline bool sk_fullsock(const struct sock *sk)
2495{
2496 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2497}
2498
2499/* Checks if this SKB belongs to an HW offloaded socket
2500 * and whether any SW fallbacks are required based on dev.
2501 * Check decrypted mark in case skb_orphan() cleared socket.
2502 */
2503static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2504 struct net_device *dev)
2505{
2506#ifdef CONFIG_SOCK_VALIDATE_XMIT
2507 struct sock *sk = skb->sk;
2508
2509 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2510 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2511#ifdef CONFIG_TLS_DEVICE
2512 } else if (unlikely(skb->decrypted)) {
2513 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2514 kfree_skb(skb);
2515 skb = NULL;
2516#endif
2517 }
2518#endif
2519
2520 return skb;
2521}
2522
2523/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2524 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2525 */
2526static inline bool sk_listener(const struct sock *sk)
2527{
2528 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2529}
2530
2531void sock_enable_timestamp(struct sock *sk, int flag);
2532int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2533 int type);
2534
2535bool sk_ns_capable(const struct sock *sk,
2536 struct user_namespace *user_ns, int cap);
2537bool sk_capable(const struct sock *sk, int cap);
2538bool sk_net_capable(const struct sock *sk, int cap);
2539
2540void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2541
2542/* Take into consideration the size of the struct sk_buff overhead in the
2543 * determination of these values, since that is non-constant across
2544 * platforms. This makes socket queueing behavior and performance
2545 * not depend upon such differences.
2546 */
2547#define _SK_MEM_PACKETS 256
2548#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2549#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2550#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2551
2552extern __u32 sysctl_wmem_max;
2553extern __u32 sysctl_rmem_max;
2554
2555extern int sysctl_tstamp_allow_data;
2556extern int sysctl_optmem_max;
2557
2558extern __u32 sysctl_wmem_default;
2559extern __u32 sysctl_rmem_default;
2560
2561DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2562
2563static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2564{
2565 /* Does this proto have per netns sysctl_wmem ? */
2566 if (proto->sysctl_wmem_offset)
2567 return *(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset);
2568
2569 return *proto->sysctl_wmem;
2570}
2571
2572static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2573{
2574 /* Does this proto have per netns sysctl_rmem ? */
2575 if (proto->sysctl_rmem_offset)
2576 return *(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset);
2577
2578 return *proto->sysctl_rmem;
2579}
2580
2581/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2582 * Some wifi drivers need to tweak it to get more chunks.
2583 * They can use this helper from their ndo_start_xmit()
2584 */
2585static inline void sk_pacing_shift_update(struct sock *sk, int val)
2586{
2587 if (!sk || !sk_fullsock(sk) || sk->sk_pacing_shift == val)
2588 return;
2589 sk->sk_pacing_shift = val;
2590}
2591
2592/* if a socket is bound to a device, check that the given device
2593 * index is either the same or that the socket is bound to an L3
2594 * master device and the given device index is also enslaved to
2595 * that L3 master
2596 */
2597static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2598{
2599 int mdif;
2600
2601 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2602 return true;
2603
2604 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2605 if (mdif && mdif == sk->sk_bound_dev_if)
2606 return true;
2607
2608 return false;
2609}
2610
2611#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/* This is the per-socket lock. The spinlock provides a synchronization
80 * between user contexts and software interrupt processing, whereas the
81 * mini-semaphore synchronizes multiple users amongst themselves.
82 */
83typedef struct {
84 spinlock_t slock;
85 int owned;
86 wait_queue_head_t wq;
87 /*
88 * We express the mutex-alike socket_lock semantics
89 * to the lock validator by explicitly managing
90 * the slock as a lock variant (in addition to
91 * the slock itself):
92 */
93#ifdef CONFIG_DEBUG_LOCK_ALLOC
94 struct lockdep_map dep_map;
95#endif
96} socket_lock_t;
97
98struct sock;
99struct proto;
100struct net;
101
102typedef __u32 __bitwise __portpair;
103typedef __u64 __bitwise __addrpair;
104
105/**
106 * struct sock_common - minimal network layer representation of sockets
107 * @skc_daddr: Foreign IPv4 addr
108 * @skc_rcv_saddr: Bound local IPv4 addr
109 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
110 * @skc_hash: hash value used with various protocol lookup tables
111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
112 * @skc_dport: placeholder for inet_dport/tw_dport
113 * @skc_num: placeholder for inet_num/tw_num
114 * @skc_portpair: __u32 union of @skc_dport & @skc_num
115 * @skc_family: network address family
116 * @skc_state: Connection state
117 * @skc_reuse: %SO_REUSEADDR setting
118 * @skc_reuseport: %SO_REUSEPORT setting
119 * @skc_ipv6only: socket is IPV6 only
120 * @skc_net_refcnt: socket is using net ref counting
121 * @skc_bound_dev_if: bound device index if != 0
122 * @skc_bind_node: bind hash linkage for various protocol lookup tables
123 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
124 * @skc_prot: protocol handlers inside a network family
125 * @skc_net: reference to the network namespace of this socket
126 * @skc_v6_daddr: IPV6 destination address
127 * @skc_v6_rcv_saddr: IPV6 source address
128 * @skc_cookie: socket's cookie value
129 * @skc_node: main hash linkage for various protocol lookup tables
130 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
131 * @skc_tx_queue_mapping: tx queue number for this connection
132 * @skc_rx_queue_mapping: rx queue number for this connection
133 * @skc_flags: place holder for sk_flags
134 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
135 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
136 * @skc_listener: connection request listener socket (aka rsk_listener)
137 * [union with @skc_flags]
138 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
139 * [union with @skc_flags]
140 * @skc_incoming_cpu: record/match cpu processing incoming packets
141 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
142 * [union with @skc_incoming_cpu]
143 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
144 * [union with @skc_incoming_cpu]
145 * @skc_refcnt: reference count
146 *
147 * This is the minimal network layer representation of sockets, the header
148 * for struct sock and struct inet_timewait_sock.
149 */
150struct sock_common {
151 union {
152 __addrpair skc_addrpair;
153 struct {
154 __be32 skc_daddr;
155 __be32 skc_rcv_saddr;
156 };
157 };
158 union {
159 unsigned int skc_hash;
160 __u16 skc_u16hashes[2];
161 };
162 /* skc_dport && skc_num must be grouped as well */
163 union {
164 __portpair skc_portpair;
165 struct {
166 __be16 skc_dport;
167 __u16 skc_num;
168 };
169 };
170
171 unsigned short skc_family;
172 volatile unsigned char skc_state;
173 unsigned char skc_reuse:4;
174 unsigned char skc_reuseport:1;
175 unsigned char skc_ipv6only:1;
176 unsigned char skc_net_refcnt:1;
177 int skc_bound_dev_if;
178 union {
179 struct hlist_node skc_bind_node;
180 struct hlist_node skc_portaddr_node;
181 };
182 struct proto *skc_prot;
183 possible_net_t skc_net;
184
185#if IS_ENABLED(CONFIG_IPV6)
186 struct in6_addr skc_v6_daddr;
187 struct in6_addr skc_v6_rcv_saddr;
188#endif
189
190 atomic64_t skc_cookie;
191
192 /* following fields are padding to force
193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
194 * assuming IPV6 is enabled. We use this padding differently
195 * for different kind of 'sockets'
196 */
197 union {
198 unsigned long skc_flags;
199 struct sock *skc_listener; /* request_sock */
200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
201 };
202 /*
203 * fields between dontcopy_begin/dontcopy_end
204 * are not copied in sock_copy()
205 */
206 /* private: */
207 int skc_dontcopy_begin[0];
208 /* public: */
209 union {
210 struct hlist_node skc_node;
211 struct hlist_nulls_node skc_nulls_node;
212 };
213 unsigned short skc_tx_queue_mapping;
214#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
215 unsigned short skc_rx_queue_mapping;
216#endif
217 union {
218 int skc_incoming_cpu;
219 u32 skc_rcv_wnd;
220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
221 };
222
223 refcount_t skc_refcnt;
224 /* private: */
225 int skc_dontcopy_end[0];
226 union {
227 u32 skc_rxhash;
228 u32 skc_window_clamp;
229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
230 };
231 /* public: */
232};
233
234struct bpf_local_storage;
235struct sk_filter;
236
237/**
238 * struct sock - network layer representation of sockets
239 * @__sk_common: shared layout with inet_timewait_sock
240 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
241 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
242 * @sk_lock: synchronizer
243 * @sk_kern_sock: True if sock is using kernel lock classes
244 * @sk_rcvbuf: size of receive buffer in bytes
245 * @sk_wq: sock wait queue and async head
246 * @sk_rx_dst: receive input route used by early demux
247 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
248 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
249 * @sk_dst_cache: destination cache
250 * @sk_dst_pending_confirm: need to confirm neighbour
251 * @sk_policy: flow policy
252 * @sk_receive_queue: incoming packets
253 * @sk_wmem_alloc: transmit queue bytes committed
254 * @sk_tsq_flags: TCP Small Queues flags
255 * @sk_write_queue: Packet sending queue
256 * @sk_omem_alloc: "o" is "option" or "other"
257 * @sk_wmem_queued: persistent queue size
258 * @sk_forward_alloc: space allocated forward
259 * @sk_reserved_mem: space reserved and non-reclaimable for the socket
260 * @sk_napi_id: id of the last napi context to receive data for sk
261 * @sk_ll_usec: usecs to busypoll when there is no data
262 * @sk_allocation: allocation mode
263 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
264 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
265 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
266 * @sk_sndbuf: size of send buffer in bytes
267 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
268 * @sk_no_check_rx: allow zero checksum in RX packets
269 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
270 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
271 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
272 * @sk_gso_max_size: Maximum GSO segment size to build
273 * @sk_gso_max_segs: Maximum number of GSO segments
274 * @sk_pacing_shift: scaling factor for TCP Small Queues
275 * @sk_lingertime: %SO_LINGER l_linger setting
276 * @sk_backlog: always used with the per-socket spinlock held
277 * @sk_callback_lock: used with the callbacks in the end of this struct
278 * @sk_error_queue: rarely used
279 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
280 * IPV6_ADDRFORM for instance)
281 * @sk_err: last error
282 * @sk_err_soft: errors that don't cause failure but are the cause of a
283 * persistent failure not just 'timed out'
284 * @sk_drops: raw/udp drops counter
285 * @sk_ack_backlog: current listen backlog
286 * @sk_max_ack_backlog: listen backlog set in listen()
287 * @sk_uid: user id of owner
288 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
289 * @sk_busy_poll_budget: napi processing budget when busypolling
290 * @sk_priority: %SO_PRIORITY setting
291 * @sk_type: socket type (%SOCK_STREAM, etc)
292 * @sk_protocol: which protocol this socket belongs in this network family
293 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
294 * @sk_peer_pid: &struct pid for this socket's peer
295 * @sk_peer_cred: %SO_PEERCRED setting
296 * @sk_rcvlowat: %SO_RCVLOWAT setting
297 * @sk_rcvtimeo: %SO_RCVTIMEO setting
298 * @sk_sndtimeo: %SO_SNDTIMEO setting
299 * @sk_txhash: computed flow hash for use on transmit
300 * @sk_txrehash: enable TX hash rethink
301 * @sk_filter: socket filtering instructions
302 * @sk_timer: sock cleanup timer
303 * @sk_stamp: time stamp of last packet received
304 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
305 * @sk_tsflags: SO_TIMESTAMPING flags
306 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
307 * Sockets that can be used under memory reclaim should
308 * set this to false.
309 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
310 * for timestamping
311 * @sk_tskey: counter to disambiguate concurrent tstamp requests
312 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
313 * @sk_socket: Identd and reporting IO signals
314 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
315 * @sk_frag: cached page frag
316 * @sk_peek_off: current peek_offset value
317 * @sk_send_head: front of stuff to transmit
318 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
319 * @sk_security: used by security modules
320 * @sk_mark: generic packet mark
321 * @sk_cgrp_data: cgroup data for this cgroup
322 * @sk_memcg: this socket's memory cgroup association
323 * @sk_write_pending: a write to stream socket waits to start
324 * @sk_disconnects: number of disconnect operations performed on this sock
325 * @sk_state_change: callback to indicate change in the state of the sock
326 * @sk_data_ready: callback to indicate there is data to be processed
327 * @sk_write_space: callback to indicate there is bf sending space available
328 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
329 * @sk_backlog_rcv: callback to process the backlog
330 * @sk_validate_xmit_skb: ptr to an optional validate function
331 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
332 * @sk_reuseport_cb: reuseport group container
333 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
334 * @sk_rcu: used during RCU grace period
335 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
336 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
337 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
338 * @sk_txtime_unused: unused txtime flags
339 * @ns_tracker: tracker for netns reference
340 */
341struct sock {
342 /*
343 * Now struct inet_timewait_sock also uses sock_common, so please just
344 * don't add nothing before this first member (__sk_common) --acme
345 */
346 struct sock_common __sk_common;
347#define sk_node __sk_common.skc_node
348#define sk_nulls_node __sk_common.skc_nulls_node
349#define sk_refcnt __sk_common.skc_refcnt
350#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
351#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
352#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
353#endif
354
355#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
356#define sk_dontcopy_end __sk_common.skc_dontcopy_end
357#define sk_hash __sk_common.skc_hash
358#define sk_portpair __sk_common.skc_portpair
359#define sk_num __sk_common.skc_num
360#define sk_dport __sk_common.skc_dport
361#define sk_addrpair __sk_common.skc_addrpair
362#define sk_daddr __sk_common.skc_daddr
363#define sk_rcv_saddr __sk_common.skc_rcv_saddr
364#define sk_family __sk_common.skc_family
365#define sk_state __sk_common.skc_state
366#define sk_reuse __sk_common.skc_reuse
367#define sk_reuseport __sk_common.skc_reuseport
368#define sk_ipv6only __sk_common.skc_ipv6only
369#define sk_net_refcnt __sk_common.skc_net_refcnt
370#define sk_bound_dev_if __sk_common.skc_bound_dev_if
371#define sk_bind_node __sk_common.skc_bind_node
372#define sk_prot __sk_common.skc_prot
373#define sk_net __sk_common.skc_net
374#define sk_v6_daddr __sk_common.skc_v6_daddr
375#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
376#define sk_cookie __sk_common.skc_cookie
377#define sk_incoming_cpu __sk_common.skc_incoming_cpu
378#define sk_flags __sk_common.skc_flags
379#define sk_rxhash __sk_common.skc_rxhash
380
381 /* early demux fields */
382 struct dst_entry __rcu *sk_rx_dst;
383 int sk_rx_dst_ifindex;
384 u32 sk_rx_dst_cookie;
385
386 socket_lock_t sk_lock;
387 atomic_t sk_drops;
388 int sk_rcvlowat;
389 struct sk_buff_head sk_error_queue;
390 struct sk_buff_head sk_receive_queue;
391 /*
392 * The backlog queue is special, it is always used with
393 * the per-socket spinlock held and requires low latency
394 * access. Therefore we special case it's implementation.
395 * Note : rmem_alloc is in this structure to fill a hole
396 * on 64bit arches, not because its logically part of
397 * backlog.
398 */
399 struct {
400 atomic_t rmem_alloc;
401 int len;
402 struct sk_buff *head;
403 struct sk_buff *tail;
404 } sk_backlog;
405
406#define sk_rmem_alloc sk_backlog.rmem_alloc
407
408 int sk_forward_alloc;
409 u32 sk_reserved_mem;
410#ifdef CONFIG_NET_RX_BUSY_POLL
411 unsigned int sk_ll_usec;
412 /* ===== mostly read cache line ===== */
413 unsigned int sk_napi_id;
414#endif
415 int sk_rcvbuf;
416 int sk_disconnects;
417
418 struct sk_filter __rcu *sk_filter;
419 union {
420 struct socket_wq __rcu *sk_wq;
421 /* private: */
422 struct socket_wq *sk_wq_raw;
423 /* public: */
424 };
425#ifdef CONFIG_XFRM
426 struct xfrm_policy __rcu *sk_policy[2];
427#endif
428
429 struct dst_entry __rcu *sk_dst_cache;
430 atomic_t sk_omem_alloc;
431 int sk_sndbuf;
432
433 /* ===== cache line for TX ===== */
434 int sk_wmem_queued;
435 refcount_t sk_wmem_alloc;
436 unsigned long sk_tsq_flags;
437 union {
438 struct sk_buff *sk_send_head;
439 struct rb_root tcp_rtx_queue;
440 };
441 struct sk_buff_head sk_write_queue;
442 __s32 sk_peek_off;
443 int sk_write_pending;
444 __u32 sk_dst_pending_confirm;
445 u32 sk_pacing_status; /* see enum sk_pacing */
446 long sk_sndtimeo;
447 struct timer_list sk_timer;
448 __u32 sk_priority;
449 __u32 sk_mark;
450 unsigned long sk_pacing_rate; /* bytes per second */
451 unsigned long sk_max_pacing_rate;
452 struct page_frag sk_frag;
453 netdev_features_t sk_route_caps;
454 int sk_gso_type;
455 unsigned int sk_gso_max_size;
456 gfp_t sk_allocation;
457 __u32 sk_txhash;
458
459 /*
460 * Because of non atomicity rules, all
461 * changes are protected by socket lock.
462 */
463 u8 sk_gso_disabled : 1,
464 sk_kern_sock : 1,
465 sk_no_check_tx : 1,
466 sk_no_check_rx : 1,
467 sk_userlocks : 4;
468 u8 sk_pacing_shift;
469 u16 sk_type;
470 u16 sk_protocol;
471 u16 sk_gso_max_segs;
472 unsigned long sk_lingertime;
473 struct proto *sk_prot_creator;
474 rwlock_t sk_callback_lock;
475 int sk_err,
476 sk_err_soft;
477 u32 sk_ack_backlog;
478 u32 sk_max_ack_backlog;
479 kuid_t sk_uid;
480 u8 sk_txrehash;
481#ifdef CONFIG_NET_RX_BUSY_POLL
482 u8 sk_prefer_busy_poll;
483 u16 sk_busy_poll_budget;
484#endif
485 spinlock_t sk_peer_lock;
486 int sk_bind_phc;
487 struct pid *sk_peer_pid;
488 const struct cred *sk_peer_cred;
489
490 long sk_rcvtimeo;
491 ktime_t sk_stamp;
492#if BITS_PER_LONG==32
493 seqlock_t sk_stamp_seq;
494#endif
495 atomic_t sk_tskey;
496 atomic_t sk_zckey;
497 u32 sk_tsflags;
498 u8 sk_shutdown;
499
500 u8 sk_clockid;
501 u8 sk_txtime_deadline_mode : 1,
502 sk_txtime_report_errors : 1,
503 sk_txtime_unused : 6;
504 bool sk_use_task_frag;
505
506 struct socket *sk_socket;
507 void *sk_user_data;
508#ifdef CONFIG_SECURITY
509 void *sk_security;
510#endif
511 struct sock_cgroup_data sk_cgrp_data;
512 struct mem_cgroup *sk_memcg;
513 void (*sk_state_change)(struct sock *sk);
514 void (*sk_data_ready)(struct sock *sk);
515 void (*sk_write_space)(struct sock *sk);
516 void (*sk_error_report)(struct sock *sk);
517 int (*sk_backlog_rcv)(struct sock *sk,
518 struct sk_buff *skb);
519#ifdef CONFIG_SOCK_VALIDATE_XMIT
520 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
521 struct net_device *dev,
522 struct sk_buff *skb);
523#endif
524 void (*sk_destruct)(struct sock *sk);
525 struct sock_reuseport __rcu *sk_reuseport_cb;
526#ifdef CONFIG_BPF_SYSCALL
527 struct bpf_local_storage __rcu *sk_bpf_storage;
528#endif
529 struct rcu_head sk_rcu;
530 netns_tracker ns_tracker;
531};
532
533enum sk_pacing {
534 SK_PACING_NONE = 0,
535 SK_PACING_NEEDED = 1,
536 SK_PACING_FQ = 2,
537};
538
539/* flag bits in sk_user_data
540 *
541 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
542 * not be suitable for copying when cloning the socket. For instance,
543 * it can point to a reference counted object. sk_user_data bottom
544 * bit is set if pointer must not be copied.
545 *
546 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
547 * managed/owned by a BPF reuseport array. This bit should be set
548 * when sk_user_data's sk is added to the bpf's reuseport_array.
549 *
550 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
551 * sk_user_data points to psock type. This bit should be set
552 * when sk_user_data is assigned to a psock object.
553 */
554#define SK_USER_DATA_NOCOPY 1UL
555#define SK_USER_DATA_BPF 2UL
556#define SK_USER_DATA_PSOCK 4UL
557#define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
558 SK_USER_DATA_PSOCK)
559
560/**
561 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
562 * @sk: socket
563 */
564static inline bool sk_user_data_is_nocopy(const struct sock *sk)
565{
566 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
567}
568
569#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
570
571/**
572 * __locked_read_sk_user_data_with_flags - return the pointer
573 * only if argument flags all has been set in sk_user_data. Otherwise
574 * return NULL
575 *
576 * @sk: socket
577 * @flags: flag bits
578 *
579 * The caller must be holding sk->sk_callback_lock.
580 */
581static inline void *
582__locked_read_sk_user_data_with_flags(const struct sock *sk,
583 uintptr_t flags)
584{
585 uintptr_t sk_user_data =
586 (uintptr_t)rcu_dereference_check(__sk_user_data(sk),
587 lockdep_is_held(&sk->sk_callback_lock));
588
589 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
590
591 if ((sk_user_data & flags) == flags)
592 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
593 return NULL;
594}
595
596/**
597 * __rcu_dereference_sk_user_data_with_flags - return the pointer
598 * only if argument flags all has been set in sk_user_data. Otherwise
599 * return NULL
600 *
601 * @sk: socket
602 * @flags: flag bits
603 */
604static inline void *
605__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
606 uintptr_t flags)
607{
608 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
609
610 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
611
612 if ((sk_user_data & flags) == flags)
613 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
614 return NULL;
615}
616
617#define rcu_dereference_sk_user_data(sk) \
618 __rcu_dereference_sk_user_data_with_flags(sk, 0)
619#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
620({ \
621 uintptr_t __tmp1 = (uintptr_t)(ptr), \
622 __tmp2 = (uintptr_t)(flags); \
623 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
624 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
625 rcu_assign_pointer(__sk_user_data((sk)), \
626 __tmp1 | __tmp2); \
627})
628#define rcu_assign_sk_user_data(sk, ptr) \
629 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
630
631static inline
632struct net *sock_net(const struct sock *sk)
633{
634 return read_pnet(&sk->sk_net);
635}
636
637static inline
638void sock_net_set(struct sock *sk, struct net *net)
639{
640 write_pnet(&sk->sk_net, net);
641}
642
643/*
644 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
645 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
646 * on a socket means that the socket will reuse everybody else's port
647 * without looking at the other's sk_reuse value.
648 */
649
650#define SK_NO_REUSE 0
651#define SK_CAN_REUSE 1
652#define SK_FORCE_REUSE 2
653
654int sk_set_peek_off(struct sock *sk, int val);
655
656static inline int sk_peek_offset(const struct sock *sk, int flags)
657{
658 if (unlikely(flags & MSG_PEEK)) {
659 return READ_ONCE(sk->sk_peek_off);
660 }
661
662 return 0;
663}
664
665static inline void sk_peek_offset_bwd(struct sock *sk, int val)
666{
667 s32 off = READ_ONCE(sk->sk_peek_off);
668
669 if (unlikely(off >= 0)) {
670 off = max_t(s32, off - val, 0);
671 WRITE_ONCE(sk->sk_peek_off, off);
672 }
673}
674
675static inline void sk_peek_offset_fwd(struct sock *sk, int val)
676{
677 sk_peek_offset_bwd(sk, -val);
678}
679
680/*
681 * Hashed lists helper routines
682 */
683static inline struct sock *sk_entry(const struct hlist_node *node)
684{
685 return hlist_entry(node, struct sock, sk_node);
686}
687
688static inline struct sock *__sk_head(const struct hlist_head *head)
689{
690 return hlist_entry(head->first, struct sock, sk_node);
691}
692
693static inline struct sock *sk_head(const struct hlist_head *head)
694{
695 return hlist_empty(head) ? NULL : __sk_head(head);
696}
697
698static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
699{
700 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
701}
702
703static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
704{
705 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
706}
707
708static inline struct sock *sk_next(const struct sock *sk)
709{
710 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
711}
712
713static inline struct sock *sk_nulls_next(const struct sock *sk)
714{
715 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
716 hlist_nulls_entry(sk->sk_nulls_node.next,
717 struct sock, sk_nulls_node) :
718 NULL;
719}
720
721static inline bool sk_unhashed(const struct sock *sk)
722{
723 return hlist_unhashed(&sk->sk_node);
724}
725
726static inline bool sk_hashed(const struct sock *sk)
727{
728 return !sk_unhashed(sk);
729}
730
731static inline void sk_node_init(struct hlist_node *node)
732{
733 node->pprev = NULL;
734}
735
736static inline void __sk_del_node(struct sock *sk)
737{
738 __hlist_del(&sk->sk_node);
739}
740
741/* NB: equivalent to hlist_del_init_rcu */
742static inline bool __sk_del_node_init(struct sock *sk)
743{
744 if (sk_hashed(sk)) {
745 __sk_del_node(sk);
746 sk_node_init(&sk->sk_node);
747 return true;
748 }
749 return false;
750}
751
752/* Grab socket reference count. This operation is valid only
753 when sk is ALREADY grabbed f.e. it is found in hash table
754 or a list and the lookup is made under lock preventing hash table
755 modifications.
756 */
757
758static __always_inline void sock_hold(struct sock *sk)
759{
760 refcount_inc(&sk->sk_refcnt);
761}
762
763/* Ungrab socket in the context, which assumes that socket refcnt
764 cannot hit zero, f.e. it is true in context of any socketcall.
765 */
766static __always_inline void __sock_put(struct sock *sk)
767{
768 refcount_dec(&sk->sk_refcnt);
769}
770
771static inline bool sk_del_node_init(struct sock *sk)
772{
773 bool rc = __sk_del_node_init(sk);
774
775 if (rc) {
776 /* paranoid for a while -acme */
777 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
778 __sock_put(sk);
779 }
780 return rc;
781}
782#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
783
784static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
785{
786 if (sk_hashed(sk)) {
787 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
788 return true;
789 }
790 return false;
791}
792
793static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
794{
795 bool rc = __sk_nulls_del_node_init_rcu(sk);
796
797 if (rc) {
798 /* paranoid for a while -acme */
799 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
800 __sock_put(sk);
801 }
802 return rc;
803}
804
805static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
806{
807 hlist_add_head(&sk->sk_node, list);
808}
809
810static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
811{
812 sock_hold(sk);
813 __sk_add_node(sk, list);
814}
815
816static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
817{
818 sock_hold(sk);
819 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
820 sk->sk_family == AF_INET6)
821 hlist_add_tail_rcu(&sk->sk_node, list);
822 else
823 hlist_add_head_rcu(&sk->sk_node, list);
824}
825
826static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
827{
828 sock_hold(sk);
829 hlist_add_tail_rcu(&sk->sk_node, list);
830}
831
832static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
833{
834 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
835}
836
837static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
838{
839 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
840}
841
842static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
843{
844 sock_hold(sk);
845 __sk_nulls_add_node_rcu(sk, list);
846}
847
848static inline void __sk_del_bind_node(struct sock *sk)
849{
850 __hlist_del(&sk->sk_bind_node);
851}
852
853static inline void sk_add_bind_node(struct sock *sk,
854 struct hlist_head *list)
855{
856 hlist_add_head(&sk->sk_bind_node, list);
857}
858
859#define sk_for_each(__sk, list) \
860 hlist_for_each_entry(__sk, list, sk_node)
861#define sk_for_each_rcu(__sk, list) \
862 hlist_for_each_entry_rcu(__sk, list, sk_node)
863#define sk_nulls_for_each(__sk, node, list) \
864 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
865#define sk_nulls_for_each_rcu(__sk, node, list) \
866 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
867#define sk_for_each_from(__sk) \
868 hlist_for_each_entry_from(__sk, sk_node)
869#define sk_nulls_for_each_from(__sk, node) \
870 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
871 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
872#define sk_for_each_safe(__sk, tmp, list) \
873 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
874#define sk_for_each_bound(__sk, list) \
875 hlist_for_each_entry(__sk, list, sk_bind_node)
876
877/**
878 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
879 * @tpos: the type * to use as a loop cursor.
880 * @pos: the &struct hlist_node to use as a loop cursor.
881 * @head: the head for your list.
882 * @offset: offset of hlist_node within the struct.
883 *
884 */
885#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
886 for (pos = rcu_dereference(hlist_first_rcu(head)); \
887 pos != NULL && \
888 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
889 pos = rcu_dereference(hlist_next_rcu(pos)))
890
891static inline struct user_namespace *sk_user_ns(const struct sock *sk)
892{
893 /* Careful only use this in a context where these parameters
894 * can not change and must all be valid, such as recvmsg from
895 * userspace.
896 */
897 return sk->sk_socket->file->f_cred->user_ns;
898}
899
900/* Sock flags */
901enum sock_flags {
902 SOCK_DEAD,
903 SOCK_DONE,
904 SOCK_URGINLINE,
905 SOCK_KEEPOPEN,
906 SOCK_LINGER,
907 SOCK_DESTROY,
908 SOCK_BROADCAST,
909 SOCK_TIMESTAMP,
910 SOCK_ZAPPED,
911 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
912 SOCK_DBG, /* %SO_DEBUG setting */
913 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
914 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
915 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
916 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
917 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
918 SOCK_FASYNC, /* fasync() active */
919 SOCK_RXQ_OVFL,
920 SOCK_ZEROCOPY, /* buffers from userspace */
921 SOCK_WIFI_STATUS, /* push wifi status to userspace */
922 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
923 * Will use last 4 bytes of packet sent from
924 * user-space instead.
925 */
926 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
927 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
928 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
929 SOCK_TXTIME,
930 SOCK_XDP, /* XDP is attached */
931 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
932 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */
933};
934
935#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
936
937static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
938{
939 nsk->sk_flags = osk->sk_flags;
940}
941
942static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
943{
944 __set_bit(flag, &sk->sk_flags);
945}
946
947static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
948{
949 __clear_bit(flag, &sk->sk_flags);
950}
951
952static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
953 int valbool)
954{
955 if (valbool)
956 sock_set_flag(sk, bit);
957 else
958 sock_reset_flag(sk, bit);
959}
960
961static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
962{
963 return test_bit(flag, &sk->sk_flags);
964}
965
966#ifdef CONFIG_NET
967DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
968static inline int sk_memalloc_socks(void)
969{
970 return static_branch_unlikely(&memalloc_socks_key);
971}
972
973void __receive_sock(struct file *file);
974#else
975
976static inline int sk_memalloc_socks(void)
977{
978 return 0;
979}
980
981static inline void __receive_sock(struct file *file)
982{ }
983#endif
984
985static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
986{
987 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
988}
989
990static inline void sk_acceptq_removed(struct sock *sk)
991{
992 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
993}
994
995static inline void sk_acceptq_added(struct sock *sk)
996{
997 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
998}
999
1000/* Note: If you think the test should be:
1001 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1002 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1003 */
1004static inline bool sk_acceptq_is_full(const struct sock *sk)
1005{
1006 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1007}
1008
1009/*
1010 * Compute minimal free write space needed to queue new packets.
1011 */
1012static inline int sk_stream_min_wspace(const struct sock *sk)
1013{
1014 return READ_ONCE(sk->sk_wmem_queued) >> 1;
1015}
1016
1017static inline int sk_stream_wspace(const struct sock *sk)
1018{
1019 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1020}
1021
1022static inline void sk_wmem_queued_add(struct sock *sk, int val)
1023{
1024 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1025}
1026
1027static inline void sk_forward_alloc_add(struct sock *sk, int val)
1028{
1029 /* Paired with lockless reads of sk->sk_forward_alloc */
1030 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1031}
1032
1033void sk_stream_write_space(struct sock *sk);
1034
1035/* OOB backlog add */
1036static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1037{
1038 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1039 skb_dst_force(skb);
1040
1041 if (!sk->sk_backlog.tail)
1042 WRITE_ONCE(sk->sk_backlog.head, skb);
1043 else
1044 sk->sk_backlog.tail->next = skb;
1045
1046 WRITE_ONCE(sk->sk_backlog.tail, skb);
1047 skb->next = NULL;
1048}
1049
1050/*
1051 * Take into account size of receive queue and backlog queue
1052 * Do not take into account this skb truesize,
1053 * to allow even a single big packet to come.
1054 */
1055static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1056{
1057 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1058
1059 return qsize > limit;
1060}
1061
1062/* The per-socket spinlock must be held here. */
1063static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1064 unsigned int limit)
1065{
1066 if (sk_rcvqueues_full(sk, limit))
1067 return -ENOBUFS;
1068
1069 /*
1070 * If the skb was allocated from pfmemalloc reserves, only
1071 * allow SOCK_MEMALLOC sockets to use it as this socket is
1072 * helping free memory
1073 */
1074 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1075 return -ENOMEM;
1076
1077 __sk_add_backlog(sk, skb);
1078 sk->sk_backlog.len += skb->truesize;
1079 return 0;
1080}
1081
1082int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1083
1084INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1085INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1086
1087static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1088{
1089 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1090 return __sk_backlog_rcv(sk, skb);
1091
1092 return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1093 tcp_v6_do_rcv,
1094 tcp_v4_do_rcv,
1095 sk, skb);
1096}
1097
1098static inline void sk_incoming_cpu_update(struct sock *sk)
1099{
1100 int cpu = raw_smp_processor_id();
1101
1102 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1103 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1104}
1105
1106static inline void sock_rps_record_flow_hash(__u32 hash)
1107{
1108#ifdef CONFIG_RPS
1109 struct rps_sock_flow_table *sock_flow_table;
1110
1111 rcu_read_lock();
1112 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1113 rps_record_sock_flow(sock_flow_table, hash);
1114 rcu_read_unlock();
1115#endif
1116}
1117
1118static inline void sock_rps_record_flow(const struct sock *sk)
1119{
1120#ifdef CONFIG_RPS
1121 if (static_branch_unlikely(&rfs_needed)) {
1122 /* Reading sk->sk_rxhash might incur an expensive cache line
1123 * miss.
1124 *
1125 * TCP_ESTABLISHED does cover almost all states where RFS
1126 * might be useful, and is cheaper [1] than testing :
1127 * IPv4: inet_sk(sk)->inet_daddr
1128 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1129 * OR an additional socket flag
1130 * [1] : sk_state and sk_prot are in the same cache line.
1131 */
1132 if (sk->sk_state == TCP_ESTABLISHED) {
1133 /* This READ_ONCE() is paired with the WRITE_ONCE()
1134 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
1135 */
1136 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1137 }
1138 }
1139#endif
1140}
1141
1142static inline void sock_rps_save_rxhash(struct sock *sk,
1143 const struct sk_buff *skb)
1144{
1145#ifdef CONFIG_RPS
1146 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1147 * here, and another one in sock_rps_record_flow().
1148 */
1149 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1150 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1151#endif
1152}
1153
1154static inline void sock_rps_reset_rxhash(struct sock *sk)
1155{
1156#ifdef CONFIG_RPS
1157 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1158 WRITE_ONCE(sk->sk_rxhash, 0);
1159#endif
1160}
1161
1162#define sk_wait_event(__sk, __timeo, __condition, __wait) \
1163 ({ int __rc, __dis = __sk->sk_disconnects; \
1164 release_sock(__sk); \
1165 __rc = __condition; \
1166 if (!__rc) { \
1167 *(__timeo) = wait_woken(__wait, \
1168 TASK_INTERRUPTIBLE, \
1169 *(__timeo)); \
1170 } \
1171 sched_annotate_sleep(); \
1172 lock_sock(__sk); \
1173 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1174 __rc; \
1175 })
1176
1177int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1178int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1179void sk_stream_wait_close(struct sock *sk, long timeo_p);
1180int sk_stream_error(struct sock *sk, int flags, int err);
1181void sk_stream_kill_queues(struct sock *sk);
1182void sk_set_memalloc(struct sock *sk);
1183void sk_clear_memalloc(struct sock *sk);
1184
1185void __sk_flush_backlog(struct sock *sk);
1186
1187static inline bool sk_flush_backlog(struct sock *sk)
1188{
1189 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1190 __sk_flush_backlog(sk);
1191 return true;
1192 }
1193 return false;
1194}
1195
1196int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1197
1198struct request_sock_ops;
1199struct timewait_sock_ops;
1200struct inet_hashinfo;
1201struct raw_hashinfo;
1202struct smc_hashinfo;
1203struct module;
1204struct sk_psock;
1205
1206/*
1207 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1208 * un-modified. Special care is taken when initializing object to zero.
1209 */
1210static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1211{
1212 if (offsetof(struct sock, sk_node.next) != 0)
1213 memset(sk, 0, offsetof(struct sock, sk_node.next));
1214 memset(&sk->sk_node.pprev, 0,
1215 size - offsetof(struct sock, sk_node.pprev));
1216}
1217
1218/* Networking protocol blocks we attach to sockets.
1219 * socket layer -> transport layer interface
1220 */
1221struct proto {
1222 void (*close)(struct sock *sk,
1223 long timeout);
1224 int (*pre_connect)(struct sock *sk,
1225 struct sockaddr *uaddr,
1226 int addr_len);
1227 int (*connect)(struct sock *sk,
1228 struct sockaddr *uaddr,
1229 int addr_len);
1230 int (*disconnect)(struct sock *sk, int flags);
1231
1232 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1233 bool kern);
1234
1235 int (*ioctl)(struct sock *sk, int cmd,
1236 int *karg);
1237 int (*init)(struct sock *sk);
1238 void (*destroy)(struct sock *sk);
1239 void (*shutdown)(struct sock *sk, int how);
1240 int (*setsockopt)(struct sock *sk, int level,
1241 int optname, sockptr_t optval,
1242 unsigned int optlen);
1243 int (*getsockopt)(struct sock *sk, int level,
1244 int optname, char __user *optval,
1245 int __user *option);
1246 void (*keepalive)(struct sock *sk, int valbool);
1247#ifdef CONFIG_COMPAT
1248 int (*compat_ioctl)(struct sock *sk,
1249 unsigned int cmd, unsigned long arg);
1250#endif
1251 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1252 size_t len);
1253 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1254 size_t len, int flags, int *addr_len);
1255 void (*splice_eof)(struct socket *sock);
1256 int (*bind)(struct sock *sk,
1257 struct sockaddr *addr, int addr_len);
1258 int (*bind_add)(struct sock *sk,
1259 struct sockaddr *addr, int addr_len);
1260
1261 int (*backlog_rcv) (struct sock *sk,
1262 struct sk_buff *skb);
1263 bool (*bpf_bypass_getsockopt)(int level,
1264 int optname);
1265
1266 void (*release_cb)(struct sock *sk);
1267
1268 /* Keeping track of sk's, looking them up, and port selection methods. */
1269 int (*hash)(struct sock *sk);
1270 void (*unhash)(struct sock *sk);
1271 void (*rehash)(struct sock *sk);
1272 int (*get_port)(struct sock *sk, unsigned short snum);
1273 void (*put_port)(struct sock *sk);
1274#ifdef CONFIG_BPF_SYSCALL
1275 int (*psock_update_sk_prot)(struct sock *sk,
1276 struct sk_psock *psock,
1277 bool restore);
1278#endif
1279
1280 /* Keeping track of sockets in use */
1281#ifdef CONFIG_PROC_FS
1282 unsigned int inuse_idx;
1283#endif
1284
1285#if IS_ENABLED(CONFIG_MPTCP)
1286 int (*forward_alloc_get)(const struct sock *sk);
1287#endif
1288
1289 bool (*stream_memory_free)(const struct sock *sk, int wake);
1290 bool (*sock_is_readable)(struct sock *sk);
1291 /* Memory pressure */
1292 void (*enter_memory_pressure)(struct sock *sk);
1293 void (*leave_memory_pressure)(struct sock *sk);
1294 atomic_long_t *memory_allocated; /* Current allocated memory. */
1295 int __percpu *per_cpu_fw_alloc;
1296 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1297
1298 /*
1299 * Pressure flag: try to collapse.
1300 * Technical note: it is used by multiple contexts non atomically.
1301 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1302 * All the __sk_mem_schedule() is of this nature: accounting
1303 * is strict, actions are advisory and have some latency.
1304 */
1305 unsigned long *memory_pressure;
1306 long *sysctl_mem;
1307
1308 int *sysctl_wmem;
1309 int *sysctl_rmem;
1310 u32 sysctl_wmem_offset;
1311 u32 sysctl_rmem_offset;
1312
1313 int max_header;
1314 bool no_autobind;
1315
1316 struct kmem_cache *slab;
1317 unsigned int obj_size;
1318 unsigned int ipv6_pinfo_offset;
1319 slab_flags_t slab_flags;
1320 unsigned int useroffset; /* Usercopy region offset */
1321 unsigned int usersize; /* Usercopy region size */
1322
1323 unsigned int __percpu *orphan_count;
1324
1325 struct request_sock_ops *rsk_prot;
1326 struct timewait_sock_ops *twsk_prot;
1327
1328 union {
1329 struct inet_hashinfo *hashinfo;
1330 struct udp_table *udp_table;
1331 struct raw_hashinfo *raw_hash;
1332 struct smc_hashinfo *smc_hash;
1333 } h;
1334
1335 struct module *owner;
1336
1337 char name[32];
1338
1339 struct list_head node;
1340 int (*diag_destroy)(struct sock *sk, int err);
1341} __randomize_layout;
1342
1343int proto_register(struct proto *prot, int alloc_slab);
1344void proto_unregister(struct proto *prot);
1345int sock_load_diag_module(int family, int protocol);
1346
1347INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1348
1349static inline int sk_forward_alloc_get(const struct sock *sk)
1350{
1351#if IS_ENABLED(CONFIG_MPTCP)
1352 if (sk->sk_prot->forward_alloc_get)
1353 return sk->sk_prot->forward_alloc_get(sk);
1354#endif
1355 return READ_ONCE(sk->sk_forward_alloc);
1356}
1357
1358static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1359{
1360 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1361 return false;
1362
1363 return sk->sk_prot->stream_memory_free ?
1364 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1365 tcp_stream_memory_free, sk, wake) : true;
1366}
1367
1368static inline bool sk_stream_memory_free(const struct sock *sk)
1369{
1370 return __sk_stream_memory_free(sk, 0);
1371}
1372
1373static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1374{
1375 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1376 __sk_stream_memory_free(sk, wake);
1377}
1378
1379static inline bool sk_stream_is_writeable(const struct sock *sk)
1380{
1381 return __sk_stream_is_writeable(sk, 0);
1382}
1383
1384static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1385 struct cgroup *ancestor)
1386{
1387#ifdef CONFIG_SOCK_CGROUP_DATA
1388 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1389 ancestor);
1390#else
1391 return -ENOTSUPP;
1392#endif
1393}
1394
1395static inline bool sk_has_memory_pressure(const struct sock *sk)
1396{
1397 return sk->sk_prot->memory_pressure != NULL;
1398}
1399
1400static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1401{
1402 return sk->sk_prot->memory_pressure &&
1403 !!READ_ONCE(*sk->sk_prot->memory_pressure);
1404}
1405
1406static inline bool sk_under_memory_pressure(const struct sock *sk)
1407{
1408 if (!sk->sk_prot->memory_pressure)
1409 return false;
1410
1411 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1412 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1413 return true;
1414
1415 return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1416}
1417
1418static inline long
1419proto_memory_allocated(const struct proto *prot)
1420{
1421 return max(0L, atomic_long_read(prot->memory_allocated));
1422}
1423
1424static inline long
1425sk_memory_allocated(const struct sock *sk)
1426{
1427 return proto_memory_allocated(sk->sk_prot);
1428}
1429
1430/* 1 MB per cpu, in page units */
1431#define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT))
1432
1433static inline void
1434sk_memory_allocated_add(struct sock *sk, int amt)
1435{
1436 int local_reserve;
1437
1438 preempt_disable();
1439 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1440 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) {
1441 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1442 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1443 }
1444 preempt_enable();
1445}
1446
1447static inline void
1448sk_memory_allocated_sub(struct sock *sk, int amt)
1449{
1450 int local_reserve;
1451
1452 preempt_disable();
1453 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1454 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) {
1455 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1456 atomic_long_add(local_reserve, sk->sk_prot->memory_allocated);
1457 }
1458 preempt_enable();
1459}
1460
1461#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1462
1463static inline void sk_sockets_allocated_dec(struct sock *sk)
1464{
1465 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1466 SK_ALLOC_PERCPU_COUNTER_BATCH);
1467}
1468
1469static inline void sk_sockets_allocated_inc(struct sock *sk)
1470{
1471 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1472 SK_ALLOC_PERCPU_COUNTER_BATCH);
1473}
1474
1475static inline u64
1476sk_sockets_allocated_read_positive(struct sock *sk)
1477{
1478 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1479}
1480
1481static inline int
1482proto_sockets_allocated_sum_positive(struct proto *prot)
1483{
1484 return percpu_counter_sum_positive(prot->sockets_allocated);
1485}
1486
1487static inline bool
1488proto_memory_pressure(struct proto *prot)
1489{
1490 if (!prot->memory_pressure)
1491 return false;
1492 return !!READ_ONCE(*prot->memory_pressure);
1493}
1494
1495
1496#ifdef CONFIG_PROC_FS
1497#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1498struct prot_inuse {
1499 int all;
1500 int val[PROTO_INUSE_NR];
1501};
1502
1503static inline void sock_prot_inuse_add(const struct net *net,
1504 const struct proto *prot, int val)
1505{
1506 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1507}
1508
1509static inline void sock_inuse_add(const struct net *net, int val)
1510{
1511 this_cpu_add(net->core.prot_inuse->all, val);
1512}
1513
1514int sock_prot_inuse_get(struct net *net, struct proto *proto);
1515int sock_inuse_get(struct net *net);
1516#else
1517static inline void sock_prot_inuse_add(const struct net *net,
1518 const struct proto *prot, int val)
1519{
1520}
1521
1522static inline void sock_inuse_add(const struct net *net, int val)
1523{
1524}
1525#endif
1526
1527
1528/* With per-bucket locks this operation is not-atomic, so that
1529 * this version is not worse.
1530 */
1531static inline int __sk_prot_rehash(struct sock *sk)
1532{
1533 sk->sk_prot->unhash(sk);
1534 return sk->sk_prot->hash(sk);
1535}
1536
1537/* About 10 seconds */
1538#define SOCK_DESTROY_TIME (10*HZ)
1539
1540/* Sockets 0-1023 can't be bound to unless you are superuser */
1541#define PROT_SOCK 1024
1542
1543#define SHUTDOWN_MASK 3
1544#define RCV_SHUTDOWN 1
1545#define SEND_SHUTDOWN 2
1546
1547#define SOCK_BINDADDR_LOCK 4
1548#define SOCK_BINDPORT_LOCK 8
1549
1550struct socket_alloc {
1551 struct socket socket;
1552 struct inode vfs_inode;
1553};
1554
1555static inline struct socket *SOCKET_I(struct inode *inode)
1556{
1557 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1558}
1559
1560static inline struct inode *SOCK_INODE(struct socket *socket)
1561{
1562 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1563}
1564
1565/*
1566 * Functions for memory accounting
1567 */
1568int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1569int __sk_mem_schedule(struct sock *sk, int size, int kind);
1570void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1571void __sk_mem_reclaim(struct sock *sk, int amount);
1572
1573#define SK_MEM_SEND 0
1574#define SK_MEM_RECV 1
1575
1576/* sysctl_mem values are in pages */
1577static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1578{
1579 return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1580}
1581
1582static inline int sk_mem_pages(int amt)
1583{
1584 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1585}
1586
1587static inline bool sk_has_account(struct sock *sk)
1588{
1589 /* return true if protocol supports memory accounting */
1590 return !!sk->sk_prot->memory_allocated;
1591}
1592
1593static inline bool sk_wmem_schedule(struct sock *sk, int size)
1594{
1595 int delta;
1596
1597 if (!sk_has_account(sk))
1598 return true;
1599 delta = size - sk->sk_forward_alloc;
1600 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1601}
1602
1603static inline bool
1604sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1605{
1606 int delta;
1607
1608 if (!sk_has_account(sk))
1609 return true;
1610 delta = size - sk->sk_forward_alloc;
1611 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1612 skb_pfmemalloc(skb);
1613}
1614
1615static inline int sk_unused_reserved_mem(const struct sock *sk)
1616{
1617 int unused_mem;
1618
1619 if (likely(!sk->sk_reserved_mem))
1620 return 0;
1621
1622 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1623 atomic_read(&sk->sk_rmem_alloc);
1624
1625 return unused_mem > 0 ? unused_mem : 0;
1626}
1627
1628static inline void sk_mem_reclaim(struct sock *sk)
1629{
1630 int reclaimable;
1631
1632 if (!sk_has_account(sk))
1633 return;
1634
1635 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1636
1637 if (reclaimable >= (int)PAGE_SIZE)
1638 __sk_mem_reclaim(sk, reclaimable);
1639}
1640
1641static inline void sk_mem_reclaim_final(struct sock *sk)
1642{
1643 sk->sk_reserved_mem = 0;
1644 sk_mem_reclaim(sk);
1645}
1646
1647static inline void sk_mem_charge(struct sock *sk, int size)
1648{
1649 if (!sk_has_account(sk))
1650 return;
1651 sk_forward_alloc_add(sk, -size);
1652}
1653
1654static inline void sk_mem_uncharge(struct sock *sk, int size)
1655{
1656 if (!sk_has_account(sk))
1657 return;
1658 sk_forward_alloc_add(sk, size);
1659 sk_mem_reclaim(sk);
1660}
1661
1662/*
1663 * Macro so as to not evaluate some arguments when
1664 * lockdep is not enabled.
1665 *
1666 * Mark both the sk_lock and the sk_lock.slock as a
1667 * per-address-family lock class.
1668 */
1669#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1670do { \
1671 sk->sk_lock.owned = 0; \
1672 init_waitqueue_head(&sk->sk_lock.wq); \
1673 spin_lock_init(&(sk)->sk_lock.slock); \
1674 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1675 sizeof((sk)->sk_lock)); \
1676 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1677 (skey), (sname)); \
1678 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1679} while (0)
1680
1681static inline bool lockdep_sock_is_held(const struct sock *sk)
1682{
1683 return lockdep_is_held(&sk->sk_lock) ||
1684 lockdep_is_held(&sk->sk_lock.slock);
1685}
1686
1687void lock_sock_nested(struct sock *sk, int subclass);
1688
1689static inline void lock_sock(struct sock *sk)
1690{
1691 lock_sock_nested(sk, 0);
1692}
1693
1694void __lock_sock(struct sock *sk);
1695void __release_sock(struct sock *sk);
1696void release_sock(struct sock *sk);
1697
1698/* BH context may only use the following locking interface. */
1699#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1700#define bh_lock_sock_nested(__sk) \
1701 spin_lock_nested(&((__sk)->sk_lock.slock), \
1702 SINGLE_DEPTH_NESTING)
1703#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1704
1705bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1706
1707/**
1708 * lock_sock_fast - fast version of lock_sock
1709 * @sk: socket
1710 *
1711 * This version should be used for very small section, where process wont block
1712 * return false if fast path is taken:
1713 *
1714 * sk_lock.slock locked, owned = 0, BH disabled
1715 *
1716 * return true if slow path is taken:
1717 *
1718 * sk_lock.slock unlocked, owned = 1, BH enabled
1719 */
1720static inline bool lock_sock_fast(struct sock *sk)
1721{
1722 /* The sk_lock has mutex_lock() semantics here. */
1723 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1724
1725 return __lock_sock_fast(sk);
1726}
1727
1728/* fast socket lock variant for caller already holding a [different] socket lock */
1729static inline bool lock_sock_fast_nested(struct sock *sk)
1730{
1731 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1732
1733 return __lock_sock_fast(sk);
1734}
1735
1736/**
1737 * unlock_sock_fast - complement of lock_sock_fast
1738 * @sk: socket
1739 * @slow: slow mode
1740 *
1741 * fast unlock socket for user context.
1742 * If slow mode is on, we call regular release_sock()
1743 */
1744static inline void unlock_sock_fast(struct sock *sk, bool slow)
1745 __releases(&sk->sk_lock.slock)
1746{
1747 if (slow) {
1748 release_sock(sk);
1749 __release(&sk->sk_lock.slock);
1750 } else {
1751 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1752 spin_unlock_bh(&sk->sk_lock.slock);
1753 }
1754}
1755
1756void sockopt_lock_sock(struct sock *sk);
1757void sockopt_release_sock(struct sock *sk);
1758bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1759bool sockopt_capable(int cap);
1760
1761/* Used by processes to "lock" a socket state, so that
1762 * interrupts and bottom half handlers won't change it
1763 * from under us. It essentially blocks any incoming
1764 * packets, so that we won't get any new data or any
1765 * packets that change the state of the socket.
1766 *
1767 * While locked, BH processing will add new packets to
1768 * the backlog queue. This queue is processed by the
1769 * owner of the socket lock right before it is released.
1770 *
1771 * Since ~2.3.5 it is also exclusive sleep lock serializing
1772 * accesses from user process context.
1773 */
1774
1775static inline void sock_owned_by_me(const struct sock *sk)
1776{
1777#ifdef CONFIG_LOCKDEP
1778 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1779#endif
1780}
1781
1782static inline bool sock_owned_by_user(const struct sock *sk)
1783{
1784 sock_owned_by_me(sk);
1785 return sk->sk_lock.owned;
1786}
1787
1788static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1789{
1790 return sk->sk_lock.owned;
1791}
1792
1793static inline void sock_release_ownership(struct sock *sk)
1794{
1795 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1796 sk->sk_lock.owned = 0;
1797
1798 /* The sk_lock has mutex_unlock() semantics: */
1799 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1800}
1801
1802/* no reclassification while locks are held */
1803static inline bool sock_allow_reclassification(const struct sock *csk)
1804{
1805 struct sock *sk = (struct sock *)csk;
1806
1807 return !sock_owned_by_user_nocheck(sk) &&
1808 !spin_is_locked(&sk->sk_lock.slock);
1809}
1810
1811struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1812 struct proto *prot, int kern);
1813void sk_free(struct sock *sk);
1814void sk_destruct(struct sock *sk);
1815struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1816void sk_free_unlock_clone(struct sock *sk);
1817
1818struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1819 gfp_t priority);
1820void __sock_wfree(struct sk_buff *skb);
1821void sock_wfree(struct sk_buff *skb);
1822struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1823 gfp_t priority);
1824void skb_orphan_partial(struct sk_buff *skb);
1825void sock_rfree(struct sk_buff *skb);
1826void sock_efree(struct sk_buff *skb);
1827#ifdef CONFIG_INET
1828void sock_edemux(struct sk_buff *skb);
1829void sock_pfree(struct sk_buff *skb);
1830#else
1831#define sock_edemux sock_efree
1832#endif
1833
1834int sk_setsockopt(struct sock *sk, int level, int optname,
1835 sockptr_t optval, unsigned int optlen);
1836int sock_setsockopt(struct socket *sock, int level, int op,
1837 sockptr_t optval, unsigned int optlen);
1838int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1839 int optname, sockptr_t optval, int optlen);
1840int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1841 int optname, sockptr_t optval, sockptr_t optlen);
1842
1843int sk_getsockopt(struct sock *sk, int level, int optname,
1844 sockptr_t optval, sockptr_t optlen);
1845int sock_gettstamp(struct socket *sock, void __user *userstamp,
1846 bool timeval, bool time32);
1847struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1848 unsigned long data_len, int noblock,
1849 int *errcode, int max_page_order);
1850
1851static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1852 unsigned long size,
1853 int noblock, int *errcode)
1854{
1855 return sock_alloc_send_pskb(sk, size, 0, noblock, errcode, 0);
1856}
1857
1858void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1859void sock_kfree_s(struct sock *sk, void *mem, int size);
1860void sock_kzfree_s(struct sock *sk, void *mem, int size);
1861void sk_send_sigurg(struct sock *sk);
1862
1863static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1864{
1865 if (sk->sk_socket)
1866 clear_bit(SOCK_SUPPORT_ZC, &sk->sk_socket->flags);
1867 WRITE_ONCE(sk->sk_prot, proto);
1868}
1869
1870struct sockcm_cookie {
1871 u64 transmit_time;
1872 u32 mark;
1873 u32 tsflags;
1874};
1875
1876static inline void sockcm_init(struct sockcm_cookie *sockc,
1877 const struct sock *sk)
1878{
1879 *sockc = (struct sockcm_cookie) {
1880 .tsflags = READ_ONCE(sk->sk_tsflags)
1881 };
1882}
1883
1884int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1885 struct sockcm_cookie *sockc);
1886int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1887 struct sockcm_cookie *sockc);
1888
1889/*
1890 * Functions to fill in entries in struct proto_ops when a protocol
1891 * does not implement a particular function.
1892 */
1893int sock_no_bind(struct socket *, struct sockaddr *, int);
1894int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1895int sock_no_socketpair(struct socket *, struct socket *);
1896int sock_no_accept(struct socket *, struct socket *, int, bool);
1897int sock_no_getname(struct socket *, struct sockaddr *, int);
1898int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1899int sock_no_listen(struct socket *, int);
1900int sock_no_shutdown(struct socket *, int);
1901int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1902int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1903int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1904int sock_no_mmap(struct file *file, struct socket *sock,
1905 struct vm_area_struct *vma);
1906
1907/*
1908 * Functions to fill in entries in struct proto_ops when a protocol
1909 * uses the inet style.
1910 */
1911int sock_common_getsockopt(struct socket *sock, int level, int optname,
1912 char __user *optval, int __user *optlen);
1913int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1914 int flags);
1915int sock_common_setsockopt(struct socket *sock, int level, int optname,
1916 sockptr_t optval, unsigned int optlen);
1917
1918void sk_common_release(struct sock *sk);
1919
1920/*
1921 * Default socket callbacks and setup code
1922 */
1923
1924/* Initialise core socket variables using an explicit uid. */
1925void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1926
1927/* Initialise core socket variables.
1928 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1929 */
1930void sock_init_data(struct socket *sock, struct sock *sk);
1931
1932/*
1933 * Socket reference counting postulates.
1934 *
1935 * * Each user of socket SHOULD hold a reference count.
1936 * * Each access point to socket (an hash table bucket, reference from a list,
1937 * running timer, skb in flight MUST hold a reference count.
1938 * * When reference count hits 0, it means it will never increase back.
1939 * * When reference count hits 0, it means that no references from
1940 * outside exist to this socket and current process on current CPU
1941 * is last user and may/should destroy this socket.
1942 * * sk_free is called from any context: process, BH, IRQ. When
1943 * it is called, socket has no references from outside -> sk_free
1944 * may release descendant resources allocated by the socket, but
1945 * to the time when it is called, socket is NOT referenced by any
1946 * hash tables, lists etc.
1947 * * Packets, delivered from outside (from network or from another process)
1948 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1949 * when they sit in queue. Otherwise, packets will leak to hole, when
1950 * socket is looked up by one cpu and unhasing is made by another CPU.
1951 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1952 * (leak to backlog). Packet socket does all the processing inside
1953 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1954 * use separate SMP lock, so that they are prone too.
1955 */
1956
1957/* Ungrab socket and destroy it, if it was the last reference. */
1958static inline void sock_put(struct sock *sk)
1959{
1960 if (refcount_dec_and_test(&sk->sk_refcnt))
1961 sk_free(sk);
1962}
1963/* Generic version of sock_put(), dealing with all sockets
1964 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1965 */
1966void sock_gen_put(struct sock *sk);
1967
1968int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1969 unsigned int trim_cap, bool refcounted);
1970static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1971 const int nested)
1972{
1973 return __sk_receive_skb(sk, skb, nested, 1, true);
1974}
1975
1976static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1977{
1978 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1979 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1980 return;
1981 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1982 * other WRITE_ONCE() because socket lock might be not held.
1983 */
1984 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1985}
1986
1987#define NO_QUEUE_MAPPING USHRT_MAX
1988
1989static inline void sk_tx_queue_clear(struct sock *sk)
1990{
1991 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1992 * other WRITE_ONCE() because socket lock might be not held.
1993 */
1994 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1995}
1996
1997static inline int sk_tx_queue_get(const struct sock *sk)
1998{
1999 if (sk) {
2000 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
2001 * and sk_tx_queue_set().
2002 */
2003 int val = READ_ONCE(sk->sk_tx_queue_mapping);
2004
2005 if (val != NO_QUEUE_MAPPING)
2006 return val;
2007 }
2008 return -1;
2009}
2010
2011static inline void __sk_rx_queue_set(struct sock *sk,
2012 const struct sk_buff *skb,
2013 bool force_set)
2014{
2015#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2016 if (skb_rx_queue_recorded(skb)) {
2017 u16 rx_queue = skb_get_rx_queue(skb);
2018
2019 if (force_set ||
2020 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2021 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2022 }
2023#endif
2024}
2025
2026static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2027{
2028 __sk_rx_queue_set(sk, skb, true);
2029}
2030
2031static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2032{
2033 __sk_rx_queue_set(sk, skb, false);
2034}
2035
2036static inline void sk_rx_queue_clear(struct sock *sk)
2037{
2038#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2039 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2040#endif
2041}
2042
2043static inline int sk_rx_queue_get(const struct sock *sk)
2044{
2045#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2046 if (sk) {
2047 int res = READ_ONCE(sk->sk_rx_queue_mapping);
2048
2049 if (res != NO_QUEUE_MAPPING)
2050 return res;
2051 }
2052#endif
2053
2054 return -1;
2055}
2056
2057static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2058{
2059 sk->sk_socket = sock;
2060}
2061
2062static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2063{
2064 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2065 return &rcu_dereference_raw(sk->sk_wq)->wait;
2066}
2067/* Detach socket from process context.
2068 * Announce socket dead, detach it from wait queue and inode.
2069 * Note that parent inode held reference count on this struct sock,
2070 * we do not release it in this function, because protocol
2071 * probably wants some additional cleanups or even continuing
2072 * to work with this socket (TCP).
2073 */
2074static inline void sock_orphan(struct sock *sk)
2075{
2076 write_lock_bh(&sk->sk_callback_lock);
2077 sock_set_flag(sk, SOCK_DEAD);
2078 sk_set_socket(sk, NULL);
2079 sk->sk_wq = NULL;
2080 write_unlock_bh(&sk->sk_callback_lock);
2081}
2082
2083static inline void sock_graft(struct sock *sk, struct socket *parent)
2084{
2085 WARN_ON(parent->sk);
2086 write_lock_bh(&sk->sk_callback_lock);
2087 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2088 parent->sk = sk;
2089 sk_set_socket(sk, parent);
2090 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2091 security_sock_graft(sk, parent);
2092 write_unlock_bh(&sk->sk_callback_lock);
2093}
2094
2095kuid_t sock_i_uid(struct sock *sk);
2096unsigned long __sock_i_ino(struct sock *sk);
2097unsigned long sock_i_ino(struct sock *sk);
2098
2099static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2100{
2101 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2102}
2103
2104static inline u32 net_tx_rndhash(void)
2105{
2106 u32 v = get_random_u32();
2107
2108 return v ?: 1;
2109}
2110
2111static inline void sk_set_txhash(struct sock *sk)
2112{
2113 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2114 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2115}
2116
2117static inline bool sk_rethink_txhash(struct sock *sk)
2118{
2119 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2120 sk_set_txhash(sk);
2121 return true;
2122 }
2123 return false;
2124}
2125
2126static inline struct dst_entry *
2127__sk_dst_get(const struct sock *sk)
2128{
2129 return rcu_dereference_check(sk->sk_dst_cache,
2130 lockdep_sock_is_held(sk));
2131}
2132
2133static inline struct dst_entry *
2134sk_dst_get(const struct sock *sk)
2135{
2136 struct dst_entry *dst;
2137
2138 rcu_read_lock();
2139 dst = rcu_dereference(sk->sk_dst_cache);
2140 if (dst && !rcuref_get(&dst->__rcuref))
2141 dst = NULL;
2142 rcu_read_unlock();
2143 return dst;
2144}
2145
2146static inline void __dst_negative_advice(struct sock *sk)
2147{
2148 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2149
2150 if (dst && dst->ops->negative_advice) {
2151 ndst = dst->ops->negative_advice(dst);
2152
2153 if (ndst != dst) {
2154 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2155 sk_tx_queue_clear(sk);
2156 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2157 }
2158 }
2159}
2160
2161static inline void dst_negative_advice(struct sock *sk)
2162{
2163 sk_rethink_txhash(sk);
2164 __dst_negative_advice(sk);
2165}
2166
2167static inline void
2168__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2169{
2170 struct dst_entry *old_dst;
2171
2172 sk_tx_queue_clear(sk);
2173 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2174 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2175 lockdep_sock_is_held(sk));
2176 rcu_assign_pointer(sk->sk_dst_cache, dst);
2177 dst_release(old_dst);
2178}
2179
2180static inline void
2181sk_dst_set(struct sock *sk, struct dst_entry *dst)
2182{
2183 struct dst_entry *old_dst;
2184
2185 sk_tx_queue_clear(sk);
2186 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2187 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2188 dst_release(old_dst);
2189}
2190
2191static inline void
2192__sk_dst_reset(struct sock *sk)
2193{
2194 __sk_dst_set(sk, NULL);
2195}
2196
2197static inline void
2198sk_dst_reset(struct sock *sk)
2199{
2200 sk_dst_set(sk, NULL);
2201}
2202
2203struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2204
2205struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2206
2207static inline void sk_dst_confirm(struct sock *sk)
2208{
2209 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2210 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2211}
2212
2213static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2214{
2215 if (skb_get_dst_pending_confirm(skb)) {
2216 struct sock *sk = skb->sk;
2217
2218 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2219 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2220 neigh_confirm(n);
2221 }
2222}
2223
2224bool sk_mc_loop(const struct sock *sk);
2225
2226static inline bool sk_can_gso(const struct sock *sk)
2227{
2228 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2229}
2230
2231void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2232
2233static inline void sk_gso_disable(struct sock *sk)
2234{
2235 sk->sk_gso_disabled = 1;
2236 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2237}
2238
2239static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2240 struct iov_iter *from, char *to,
2241 int copy, int offset)
2242{
2243 if (skb->ip_summed == CHECKSUM_NONE) {
2244 __wsum csum = 0;
2245 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2246 return -EFAULT;
2247 skb->csum = csum_block_add(skb->csum, csum, offset);
2248 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2249 if (!copy_from_iter_full_nocache(to, copy, from))
2250 return -EFAULT;
2251 } else if (!copy_from_iter_full(to, copy, from))
2252 return -EFAULT;
2253
2254 return 0;
2255}
2256
2257static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2258 struct iov_iter *from, int copy)
2259{
2260 int err, offset = skb->len;
2261
2262 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2263 copy, offset);
2264 if (err)
2265 __skb_trim(skb, offset);
2266
2267 return err;
2268}
2269
2270static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2271 struct sk_buff *skb,
2272 struct page *page,
2273 int off, int copy)
2274{
2275 int err;
2276
2277 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2278 copy, skb->len);
2279 if (err)
2280 return err;
2281
2282 skb_len_add(skb, copy);
2283 sk_wmem_queued_add(sk, copy);
2284 sk_mem_charge(sk, copy);
2285 return 0;
2286}
2287
2288/**
2289 * sk_wmem_alloc_get - returns write allocations
2290 * @sk: socket
2291 *
2292 * Return: sk_wmem_alloc minus initial offset of one
2293 */
2294static inline int sk_wmem_alloc_get(const struct sock *sk)
2295{
2296 return refcount_read(&sk->sk_wmem_alloc) - 1;
2297}
2298
2299/**
2300 * sk_rmem_alloc_get - returns read allocations
2301 * @sk: socket
2302 *
2303 * Return: sk_rmem_alloc
2304 */
2305static inline int sk_rmem_alloc_get(const struct sock *sk)
2306{
2307 return atomic_read(&sk->sk_rmem_alloc);
2308}
2309
2310/**
2311 * sk_has_allocations - check if allocations are outstanding
2312 * @sk: socket
2313 *
2314 * Return: true if socket has write or read allocations
2315 */
2316static inline bool sk_has_allocations(const struct sock *sk)
2317{
2318 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2319}
2320
2321/**
2322 * skwq_has_sleeper - check if there are any waiting processes
2323 * @wq: struct socket_wq
2324 *
2325 * Return: true if socket_wq has waiting processes
2326 *
2327 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2328 * barrier call. They were added due to the race found within the tcp code.
2329 *
2330 * Consider following tcp code paths::
2331 *
2332 * CPU1 CPU2
2333 * sys_select receive packet
2334 * ... ...
2335 * __add_wait_queue update tp->rcv_nxt
2336 * ... ...
2337 * tp->rcv_nxt check sock_def_readable
2338 * ... {
2339 * schedule rcu_read_lock();
2340 * wq = rcu_dereference(sk->sk_wq);
2341 * if (wq && waitqueue_active(&wq->wait))
2342 * wake_up_interruptible(&wq->wait)
2343 * ...
2344 * }
2345 *
2346 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2347 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2348 * could then endup calling schedule and sleep forever if there are no more
2349 * data on the socket.
2350 *
2351 */
2352static inline bool skwq_has_sleeper(struct socket_wq *wq)
2353{
2354 return wq && wq_has_sleeper(&wq->wait);
2355}
2356
2357/**
2358 * sock_poll_wait - place memory barrier behind the poll_wait call.
2359 * @filp: file
2360 * @sock: socket to wait on
2361 * @p: poll_table
2362 *
2363 * See the comments in the wq_has_sleeper function.
2364 */
2365static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2366 poll_table *p)
2367{
2368 if (!poll_does_not_wait(p)) {
2369 poll_wait(filp, &sock->wq.wait, p);
2370 /* We need to be sure we are in sync with the
2371 * socket flags modification.
2372 *
2373 * This memory barrier is paired in the wq_has_sleeper.
2374 */
2375 smp_mb();
2376 }
2377}
2378
2379static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2380{
2381 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2382 u32 txhash = READ_ONCE(sk->sk_txhash);
2383
2384 if (txhash) {
2385 skb->l4_hash = 1;
2386 skb->hash = txhash;
2387 }
2388}
2389
2390void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2391
2392/*
2393 * Queue a received datagram if it will fit. Stream and sequenced
2394 * protocols can't normally use this as they need to fit buffers in
2395 * and play with them.
2396 *
2397 * Inlined as it's very short and called for pretty much every
2398 * packet ever received.
2399 */
2400static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2401{
2402 skb_orphan(skb);
2403 skb->sk = sk;
2404 skb->destructor = sock_rfree;
2405 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2406 sk_mem_charge(sk, skb->truesize);
2407}
2408
2409static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2410{
2411 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2412 skb_orphan(skb);
2413 skb->destructor = sock_efree;
2414 skb->sk = sk;
2415 return true;
2416 }
2417 return false;
2418}
2419
2420static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2421{
2422 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2423 if (skb) {
2424 if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2425 skb_set_owner_r(skb, sk);
2426 return skb;
2427 }
2428 __kfree_skb(skb);
2429 }
2430 return NULL;
2431}
2432
2433static inline void skb_prepare_for_gro(struct sk_buff *skb)
2434{
2435 if (skb->destructor != sock_wfree) {
2436 skb_orphan(skb);
2437 return;
2438 }
2439 skb->slow_gro = 1;
2440}
2441
2442void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2443 unsigned long expires);
2444
2445void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2446
2447void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2448
2449int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2450 struct sk_buff *skb, unsigned int flags,
2451 void (*destructor)(struct sock *sk,
2452 struct sk_buff *skb));
2453int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2454
2455int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2456 enum skb_drop_reason *reason);
2457
2458static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2459{
2460 return sock_queue_rcv_skb_reason(sk, skb, NULL);
2461}
2462
2463int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2464struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2465
2466/*
2467 * Recover an error report and clear atomically
2468 */
2469
2470static inline int sock_error(struct sock *sk)
2471{
2472 int err;
2473
2474 /* Avoid an atomic operation for the common case.
2475 * This is racy since another cpu/thread can change sk_err under us.
2476 */
2477 if (likely(data_race(!sk->sk_err)))
2478 return 0;
2479
2480 err = xchg(&sk->sk_err, 0);
2481 return -err;
2482}
2483
2484void sk_error_report(struct sock *sk);
2485
2486static inline unsigned long sock_wspace(struct sock *sk)
2487{
2488 int amt = 0;
2489
2490 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2491 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2492 if (amt < 0)
2493 amt = 0;
2494 }
2495 return amt;
2496}
2497
2498/* Note:
2499 * We use sk->sk_wq_raw, from contexts knowing this
2500 * pointer is not NULL and cannot disappear/change.
2501 */
2502static inline void sk_set_bit(int nr, struct sock *sk)
2503{
2504 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2505 !sock_flag(sk, SOCK_FASYNC))
2506 return;
2507
2508 set_bit(nr, &sk->sk_wq_raw->flags);
2509}
2510
2511static inline void sk_clear_bit(int nr, struct sock *sk)
2512{
2513 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2514 !sock_flag(sk, SOCK_FASYNC))
2515 return;
2516
2517 clear_bit(nr, &sk->sk_wq_raw->flags);
2518}
2519
2520static inline void sk_wake_async(const struct sock *sk, int how, int band)
2521{
2522 if (sock_flag(sk, SOCK_FASYNC)) {
2523 rcu_read_lock();
2524 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2525 rcu_read_unlock();
2526 }
2527}
2528
2529/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2530 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2531 * Note: for send buffers, TCP works better if we can build two skbs at
2532 * minimum.
2533 */
2534#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2535
2536#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2537#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2538
2539static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2540{
2541 u32 val;
2542
2543 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2544 return;
2545
2546 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2547 val = max_t(u32, val, sk_unused_reserved_mem(sk));
2548
2549 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2550}
2551
2552/**
2553 * sk_page_frag - return an appropriate page_frag
2554 * @sk: socket
2555 *
2556 * Use the per task page_frag instead of the per socket one for
2557 * optimization when we know that we're in process context and own
2558 * everything that's associated with %current.
2559 *
2560 * Both direct reclaim and page faults can nest inside other
2561 * socket operations and end up recursing into sk_page_frag()
2562 * while it's already in use: explicitly avoid task page_frag
2563 * when users disable sk_use_task_frag.
2564 *
2565 * Return: a per task page_frag if context allows that,
2566 * otherwise a per socket one.
2567 */
2568static inline struct page_frag *sk_page_frag(struct sock *sk)
2569{
2570 if (sk->sk_use_task_frag)
2571 return ¤t->task_frag;
2572
2573 return &sk->sk_frag;
2574}
2575
2576bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2577
2578/*
2579 * Default write policy as shown to user space via poll/select/SIGIO
2580 */
2581static inline bool sock_writeable(const struct sock *sk)
2582{
2583 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2584}
2585
2586static inline gfp_t gfp_any(void)
2587{
2588 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2589}
2590
2591static inline gfp_t gfp_memcg_charge(void)
2592{
2593 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2594}
2595
2596static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2597{
2598 return noblock ? 0 : sk->sk_rcvtimeo;
2599}
2600
2601static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2602{
2603 return noblock ? 0 : sk->sk_sndtimeo;
2604}
2605
2606static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2607{
2608 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2609
2610 return v ?: 1;
2611}
2612
2613/* Alas, with timeout socket operations are not restartable.
2614 * Compare this to poll().
2615 */
2616static inline int sock_intr_errno(long timeo)
2617{
2618 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2619}
2620
2621struct sock_skb_cb {
2622 u32 dropcount;
2623};
2624
2625/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2626 * using skb->cb[] would keep using it directly and utilize its
2627 * alignement guarantee.
2628 */
2629#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2630 sizeof(struct sock_skb_cb)))
2631
2632#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2633 SOCK_SKB_CB_OFFSET))
2634
2635#define sock_skb_cb_check_size(size) \
2636 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2637
2638static inline void
2639sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2640{
2641 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2642 atomic_read(&sk->sk_drops) : 0;
2643}
2644
2645static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2646{
2647 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2648
2649 atomic_add(segs, &sk->sk_drops);
2650}
2651
2652static inline ktime_t sock_read_timestamp(struct sock *sk)
2653{
2654#if BITS_PER_LONG==32
2655 unsigned int seq;
2656 ktime_t kt;
2657
2658 do {
2659 seq = read_seqbegin(&sk->sk_stamp_seq);
2660 kt = sk->sk_stamp;
2661 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2662
2663 return kt;
2664#else
2665 return READ_ONCE(sk->sk_stamp);
2666#endif
2667}
2668
2669static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2670{
2671#if BITS_PER_LONG==32
2672 write_seqlock(&sk->sk_stamp_seq);
2673 sk->sk_stamp = kt;
2674 write_sequnlock(&sk->sk_stamp_seq);
2675#else
2676 WRITE_ONCE(sk->sk_stamp, kt);
2677#endif
2678}
2679
2680void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2681 struct sk_buff *skb);
2682void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2683 struct sk_buff *skb);
2684
2685static inline void
2686sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2687{
2688 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2689 u32 tsflags = READ_ONCE(sk->sk_tsflags);
2690 ktime_t kt = skb->tstamp;
2691 /*
2692 * generate control messages if
2693 * - receive time stamping in software requested
2694 * - software time stamp available and wanted
2695 * - hardware time stamps available and wanted
2696 */
2697 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2698 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2699 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2700 (hwtstamps->hwtstamp &&
2701 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2702 __sock_recv_timestamp(msg, sk, skb);
2703 else
2704 sock_write_timestamp(sk, kt);
2705
2706 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2707 __sock_recv_wifi_status(msg, sk, skb);
2708}
2709
2710void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2711 struct sk_buff *skb);
2712
2713#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2714static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2715 struct sk_buff *skb)
2716{
2717#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \
2718 (1UL << SOCK_RCVTSTAMP) | \
2719 (1UL << SOCK_RCVMARK))
2720#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2721 SOF_TIMESTAMPING_RAW_HARDWARE)
2722
2723 if (sk->sk_flags & FLAGS_RECV_CMSGS ||
2724 READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY)
2725 __sock_recv_cmsgs(msg, sk, skb);
2726 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2727 sock_write_timestamp(sk, skb->tstamp);
2728 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2729 sock_write_timestamp(sk, 0);
2730}
2731
2732void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2733
2734/**
2735 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2736 * @sk: socket sending this packet
2737 * @tsflags: timestamping flags to use
2738 * @tx_flags: completed with instructions for time stamping
2739 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2740 *
2741 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2742 */
2743static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2744 __u8 *tx_flags, __u32 *tskey)
2745{
2746 if (unlikely(tsflags)) {
2747 __sock_tx_timestamp(tsflags, tx_flags);
2748 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2749 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2750 *tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2751 }
2752 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2753 *tx_flags |= SKBTX_WIFI_STATUS;
2754}
2755
2756static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2757 __u8 *tx_flags)
2758{
2759 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2760}
2761
2762static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2763{
2764 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2765 &skb_shinfo(skb)->tskey);
2766}
2767
2768static inline bool sk_is_inet(const struct sock *sk)
2769{
2770 int family = READ_ONCE(sk->sk_family);
2771
2772 return family == AF_INET || family == AF_INET6;
2773}
2774
2775static inline bool sk_is_tcp(const struct sock *sk)
2776{
2777 return sk_is_inet(sk) &&
2778 sk->sk_type == SOCK_STREAM &&
2779 sk->sk_protocol == IPPROTO_TCP;
2780}
2781
2782static inline bool sk_is_udp(const struct sock *sk)
2783{
2784 return sk_is_inet(sk) &&
2785 sk->sk_type == SOCK_DGRAM &&
2786 sk->sk_protocol == IPPROTO_UDP;
2787}
2788
2789static inline bool sk_is_stream_unix(const struct sock *sk)
2790{
2791 return sk->sk_family == AF_UNIX && sk->sk_type == SOCK_STREAM;
2792}
2793
2794/**
2795 * sk_eat_skb - Release a skb if it is no longer needed
2796 * @sk: socket to eat this skb from
2797 * @skb: socket buffer to eat
2798 *
2799 * This routine must be called with interrupts disabled or with the socket
2800 * locked so that the sk_buff queue operation is ok.
2801*/
2802static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2803{
2804 __skb_unlink(skb, &sk->sk_receive_queue);
2805 __kfree_skb(skb);
2806}
2807
2808static inline bool
2809skb_sk_is_prefetched(struct sk_buff *skb)
2810{
2811#ifdef CONFIG_INET
2812 return skb->destructor == sock_pfree;
2813#else
2814 return false;
2815#endif /* CONFIG_INET */
2816}
2817
2818/* This helper checks if a socket is a full socket,
2819 * ie _not_ a timewait or request socket.
2820 */
2821static inline bool sk_fullsock(const struct sock *sk)
2822{
2823 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2824}
2825
2826static inline bool
2827sk_is_refcounted(struct sock *sk)
2828{
2829 /* Only full sockets have sk->sk_flags. */
2830 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2831}
2832
2833/**
2834 * skb_steal_sock - steal a socket from an sk_buff
2835 * @skb: sk_buff to steal the socket from
2836 * @refcounted: is set to true if the socket is reference-counted
2837 * @prefetched: is set to true if the socket was assigned from bpf
2838 */
2839static inline struct sock *
2840skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched)
2841{
2842 if (skb->sk) {
2843 struct sock *sk = skb->sk;
2844
2845 *refcounted = true;
2846 *prefetched = skb_sk_is_prefetched(skb);
2847 if (*prefetched)
2848 *refcounted = sk_is_refcounted(sk);
2849 skb->destructor = NULL;
2850 skb->sk = NULL;
2851 return sk;
2852 }
2853 *prefetched = false;
2854 *refcounted = false;
2855 return NULL;
2856}
2857
2858/* Checks if this SKB belongs to an HW offloaded socket
2859 * and whether any SW fallbacks are required based on dev.
2860 * Check decrypted mark in case skb_orphan() cleared socket.
2861 */
2862static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2863 struct net_device *dev)
2864{
2865#ifdef CONFIG_SOCK_VALIDATE_XMIT
2866 struct sock *sk = skb->sk;
2867
2868 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2869 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2870#ifdef CONFIG_TLS_DEVICE
2871 } else if (unlikely(skb->decrypted)) {
2872 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2873 kfree_skb(skb);
2874 skb = NULL;
2875#endif
2876 }
2877#endif
2878
2879 return skb;
2880}
2881
2882/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2883 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2884 */
2885static inline bool sk_listener(const struct sock *sk)
2886{
2887 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2888}
2889
2890void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2891int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2892 int type);
2893
2894bool sk_ns_capable(const struct sock *sk,
2895 struct user_namespace *user_ns, int cap);
2896bool sk_capable(const struct sock *sk, int cap);
2897bool sk_net_capable(const struct sock *sk, int cap);
2898
2899void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2900
2901/* Take into consideration the size of the struct sk_buff overhead in the
2902 * determination of these values, since that is non-constant across
2903 * platforms. This makes socket queueing behavior and performance
2904 * not depend upon such differences.
2905 */
2906#define _SK_MEM_PACKETS 256
2907#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2908#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2909#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2910
2911extern __u32 sysctl_wmem_max;
2912extern __u32 sysctl_rmem_max;
2913
2914extern int sysctl_tstamp_allow_data;
2915
2916extern __u32 sysctl_wmem_default;
2917extern __u32 sysctl_rmem_default;
2918
2919#define SKB_FRAG_PAGE_ORDER get_order(32768)
2920DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2921
2922static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2923{
2924 /* Does this proto have per netns sysctl_wmem ? */
2925 if (proto->sysctl_wmem_offset)
2926 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2927
2928 return READ_ONCE(*proto->sysctl_wmem);
2929}
2930
2931static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2932{
2933 /* Does this proto have per netns sysctl_rmem ? */
2934 if (proto->sysctl_rmem_offset)
2935 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2936
2937 return READ_ONCE(*proto->sysctl_rmem);
2938}
2939
2940/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2941 * Some wifi drivers need to tweak it to get more chunks.
2942 * They can use this helper from their ndo_start_xmit()
2943 */
2944static inline void sk_pacing_shift_update(struct sock *sk, int val)
2945{
2946 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2947 return;
2948 WRITE_ONCE(sk->sk_pacing_shift, val);
2949}
2950
2951/* if a socket is bound to a device, check that the given device
2952 * index is either the same or that the socket is bound to an L3
2953 * master device and the given device index is also enslaved to
2954 * that L3 master
2955 */
2956static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2957{
2958 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2959 int mdif;
2960
2961 if (!bound_dev_if || bound_dev_if == dif)
2962 return true;
2963
2964 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2965 if (mdif && mdif == bound_dev_if)
2966 return true;
2967
2968 return false;
2969}
2970
2971void sock_def_readable(struct sock *sk);
2972
2973int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2974void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2975int sock_set_timestamping(struct sock *sk, int optname,
2976 struct so_timestamping timestamping);
2977
2978void sock_enable_timestamps(struct sock *sk);
2979void sock_no_linger(struct sock *sk);
2980void sock_set_keepalive(struct sock *sk);
2981void sock_set_priority(struct sock *sk, u32 priority);
2982void sock_set_rcvbuf(struct sock *sk, int val);
2983void sock_set_mark(struct sock *sk, u32 val);
2984void sock_set_reuseaddr(struct sock *sk);
2985void sock_set_reuseport(struct sock *sk);
2986void sock_set_sndtimeo(struct sock *sk, s64 secs);
2987
2988int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2989
2990int sock_get_timeout(long timeo, void *optval, bool old_timeval);
2991int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
2992 sockptr_t optval, int optlen, bool old_timeval);
2993
2994int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
2995 void __user *arg, void *karg, size_t size);
2996int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
2997static inline bool sk_is_readable(struct sock *sk)
2998{
2999 if (sk->sk_prot->sock_is_readable)
3000 return sk->sk_prot->sock_is_readable(sk);
3001 return false;
3002}
3003#endif /* _SOCK_H */