1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the AF_INET socket handler.
8 *
9 * Version: @(#)sock.h 1.0.4 05/13/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche <flla@stud.uni-sb.de>
15 *
16 * Fixes:
17 * Alan Cox : Volatiles in skbuff pointers. See
18 * skbuff comments. May be overdone,
19 * better to prove they can be removed
20 * than the reverse.
21 * Alan Cox : Added a zapped field for tcp to note
22 * a socket is reset and must stay shut up
23 * Alan Cox : New fields for options
24 * Pauline Middelink : identd support
25 * Alan Cox : Eliminate low level recv/recvfrom
26 * David S. Miller : New socket lookup architecture.
27 * Steve Whitehouse: Default routines for sock_ops
28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
29 * protinfo be just a void pointer, as the
30 * protocol specific parts were moved to
31 * respective headers and ipv4/v6, etc now
32 * use private slabcaches for its socks
33 * Pedro Hortas : New flags field for socket options
34 */
35#ifndef _SOCK_H
36#define _SOCK_H
37
38#include <linux/hardirq.h>
39#include <linux/kernel.h>
40#include <linux/list.h>
41#include <linux/list_nulls.h>
42#include <linux/timer.h>
43#include <linux/cache.h>
44#include <linux/bitops.h>
45#include <linux/lockdep.h>
46#include <linux/netdevice.h>
47#include <linux/skbuff.h> /* struct sk_buff */
48#include <linux/mm.h>
49#include <linux/security.h>
50#include <linux/slab.h>
51#include <linux/uaccess.h>
52#include <linux/page_counter.h>
53#include <linux/memcontrol.h>
54#include <linux/static_key.h>
55#include <linux/sched.h>
56#include <linux/wait.h>
57#include <linux/cgroup-defs.h>
58#include <linux/rbtree.h>
59#include <linux/rculist_nulls.h>
60#include <linux/poll.h>
61#include <linux/sockptr.h>
62#include <linux/indirect_call_wrapper.h>
63#include <linux/atomic.h>
64#include <linux/refcount.h>
65#include <linux/llist.h>
66#include <net/dst.h>
67#include <net/checksum.h>
68#include <net/tcp_states.h>
69#include <linux/net_tstamp.h>
70#include <net/l3mdev.h>
71#include <uapi/linux/socket.h>
72
73/*
74 * This structure really needs to be cleaned up.
75 * Most of it is for TCP, and not used by any of
76 * the other protocols.
77 */
78
79/* Define this to get the SOCK_DBG debugging facility. */
80#define SOCK_DEBUGGING
81#ifdef SOCK_DEBUGGING
82#define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
83 printk(KERN_DEBUG msg); } while (0)
84#else
85/* Validate arguments and do nothing */
86static inline __printf(2, 3)
87void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
88{
89}
90#endif
91
92/* This is the per-socket lock. The spinlock provides a synchronization
93 * between user contexts and software interrupt processing, whereas the
94 * mini-semaphore synchronizes multiple users amongst themselves.
95 */
96typedef struct {
97 spinlock_t slock;
98 int owned;
99 wait_queue_head_t wq;
100 /*
101 * We express the mutex-alike socket_lock semantics
102 * to the lock validator by explicitly managing
103 * the slock as a lock variant (in addition to
104 * the slock itself):
105 */
106#ifdef CONFIG_DEBUG_LOCK_ALLOC
107 struct lockdep_map dep_map;
108#endif
109} socket_lock_t;
110
111struct sock;
112struct proto;
113struct net;
114
115typedef __u32 __bitwise __portpair;
116typedef __u64 __bitwise __addrpair;
117
118/**
119 * struct sock_common - minimal network layer representation of sockets
120 * @skc_daddr: Foreign IPv4 addr
121 * @skc_rcv_saddr: Bound local IPv4 addr
122 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
123 * @skc_hash: hash value used with various protocol lookup tables
124 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
125 * @skc_dport: placeholder for inet_dport/tw_dport
126 * @skc_num: placeholder for inet_num/tw_num
127 * @skc_portpair: __u32 union of @skc_dport & @skc_num
128 * @skc_family: network address family
129 * @skc_state: Connection state
130 * @skc_reuse: %SO_REUSEADDR setting
131 * @skc_reuseport: %SO_REUSEPORT setting
132 * @skc_ipv6only: socket is IPV6 only
133 * @skc_net_refcnt: socket is using net ref counting
134 * @skc_bound_dev_if: bound device index if != 0
135 * @skc_bind_node: bind hash linkage for various protocol lookup tables
136 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
137 * @skc_prot: protocol handlers inside a network family
138 * @skc_net: reference to the network namespace of this socket
139 * @skc_v6_daddr: IPV6 destination address
140 * @skc_v6_rcv_saddr: IPV6 source address
141 * @skc_cookie: socket's cookie value
142 * @skc_node: main hash linkage for various protocol lookup tables
143 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
144 * @skc_tx_queue_mapping: tx queue number for this connection
145 * @skc_rx_queue_mapping: rx queue number for this connection
146 * @skc_flags: place holder for sk_flags
147 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
148 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
149 * @skc_listener: connection request listener socket (aka rsk_listener)
150 * [union with @skc_flags]
151 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
152 * [union with @skc_flags]
153 * @skc_incoming_cpu: record/match cpu processing incoming packets
154 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
155 * [union with @skc_incoming_cpu]
156 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
157 * [union with @skc_incoming_cpu]
158 * @skc_refcnt: reference count
159 *
160 * This is the minimal network layer representation of sockets, the header
161 * for struct sock and struct inet_timewait_sock.
162 */
163struct sock_common {
164 union {
165 __addrpair skc_addrpair;
166 struct {
167 __be32 skc_daddr;
168 __be32 skc_rcv_saddr;
169 };
170 };
171 union {
172 unsigned int skc_hash;
173 __u16 skc_u16hashes[2];
174 };
175 /* skc_dport && skc_num must be grouped as well */
176 union {
177 __portpair skc_portpair;
178 struct {
179 __be16 skc_dport;
180 __u16 skc_num;
181 };
182 };
183
184 unsigned short skc_family;
185 volatile unsigned char skc_state;
186 unsigned char skc_reuse:4;
187 unsigned char skc_reuseport:1;
188 unsigned char skc_ipv6only:1;
189 unsigned char skc_net_refcnt:1;
190 int skc_bound_dev_if;
191 union {
192 struct hlist_node skc_bind_node;
193 struct hlist_node skc_portaddr_node;
194 };
195 struct proto *skc_prot;
196 possible_net_t skc_net;
197
198#if IS_ENABLED(CONFIG_IPV6)
199 struct in6_addr skc_v6_daddr;
200 struct in6_addr skc_v6_rcv_saddr;
201#endif
202
203 atomic64_t skc_cookie;
204
205 /* following fields are padding to force
206 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
207 * assuming IPV6 is enabled. We use this padding differently
208 * for different kind of 'sockets'
209 */
210 union {
211 unsigned long skc_flags;
212 struct sock *skc_listener; /* request_sock */
213 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
214 };
215 /*
216 * fields between dontcopy_begin/dontcopy_end
217 * are not copied in sock_copy()
218 */
219 /* private: */
220 int skc_dontcopy_begin[0];
221 /* public: */
222 union {
223 struct hlist_node skc_node;
224 struct hlist_nulls_node skc_nulls_node;
225 };
226 unsigned short skc_tx_queue_mapping;
227#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
228 unsigned short skc_rx_queue_mapping;
229#endif
230 union {
231 int skc_incoming_cpu;
232 u32 skc_rcv_wnd;
233 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
234 };
235
236 refcount_t skc_refcnt;
237 /* private: */
238 int skc_dontcopy_end[0];
239 union {
240 u32 skc_rxhash;
241 u32 skc_window_clamp;
242 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
243 };
244 /* public: */
245};
246
247struct bpf_local_storage;
248struct sk_filter;
249
250/**
251 * struct sock - network layer representation of sockets
252 * @__sk_common: shared layout with inet_timewait_sock
253 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
254 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
255 * @sk_lock: synchronizer
256 * @sk_kern_sock: True if sock is using kernel lock classes
257 * @sk_rcvbuf: size of receive buffer in bytes
258 * @sk_wq: sock wait queue and async head
259 * @sk_rx_dst: receive input route used by early demux
260 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
261 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
262 * @sk_dst_cache: destination cache
263 * @sk_dst_pending_confirm: need to confirm neighbour
264 * @sk_policy: flow policy
265 * @sk_receive_queue: incoming packets
266 * @sk_wmem_alloc: transmit queue bytes committed
267 * @sk_tsq_flags: TCP Small Queues flags
268 * @sk_write_queue: Packet sending queue
269 * @sk_omem_alloc: "o" is "option" or "other"
270 * @sk_wmem_queued: persistent queue size
271 * @sk_forward_alloc: space allocated forward
272 * @sk_reserved_mem: space reserved and non-reclaimable for the socket
273 * @sk_napi_id: id of the last napi context to receive data for sk
274 * @sk_ll_usec: usecs to busypoll when there is no data
275 * @sk_allocation: allocation mode
276 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
277 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
278 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
279 * @sk_sndbuf: size of send buffer in bytes
280 * @__sk_flags_offset: empty field used to determine location of bitfield
281 * @sk_padding: unused element for alignment
282 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
283 * @sk_no_check_rx: allow zero checksum in RX packets
284 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
285 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
286 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
287 * @sk_gso_max_size: Maximum GSO segment size to build
288 * @sk_gso_max_segs: Maximum number of GSO segments
289 * @sk_pacing_shift: scaling factor for TCP Small Queues
290 * @sk_lingertime: %SO_LINGER l_linger setting
291 * @sk_backlog: always used with the per-socket spinlock held
292 * @sk_callback_lock: used with the callbacks in the end of this struct
293 * @sk_error_queue: rarely used
294 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
295 * IPV6_ADDRFORM for instance)
296 * @sk_err: last error
297 * @sk_err_soft: errors that don't cause failure but are the cause of a
298 * persistent failure not just 'timed out'
299 * @sk_drops: raw/udp drops counter
300 * @sk_ack_backlog: current listen backlog
301 * @sk_max_ack_backlog: listen backlog set in listen()
302 * @sk_uid: user id of owner
303 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
304 * @sk_busy_poll_budget: napi processing budget when busypolling
305 * @sk_priority: %SO_PRIORITY setting
306 * @sk_type: socket type (%SOCK_STREAM, etc)
307 * @sk_protocol: which protocol this socket belongs in this network family
308 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
309 * @sk_peer_pid: &struct pid for this socket's peer
310 * @sk_peer_cred: %SO_PEERCRED setting
311 * @sk_rcvlowat: %SO_RCVLOWAT setting
312 * @sk_rcvtimeo: %SO_RCVTIMEO setting
313 * @sk_sndtimeo: %SO_SNDTIMEO setting
314 * @sk_txhash: computed flow hash for use on transmit
315 * @sk_txrehash: enable TX hash rethink
316 * @sk_filter: socket filtering instructions
317 * @sk_timer: sock cleanup timer
318 * @sk_stamp: time stamp of last packet received
319 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
320 * @sk_tsflags: SO_TIMESTAMPING flags
321 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
322 * Sockets that can be used under memory reclaim should
323 * set this to false.
324 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
325 * for timestamping
326 * @sk_tskey: counter to disambiguate concurrent tstamp requests
327 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
328 * @sk_socket: Identd and reporting IO signals
329 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
330 * @sk_frag: cached page frag
331 * @sk_peek_off: current peek_offset value
332 * @sk_send_head: front of stuff to transmit
333 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
334 * @sk_security: used by security modules
335 * @sk_mark: generic packet mark
336 * @sk_cgrp_data: cgroup data for this cgroup
337 * @sk_memcg: this socket's memory cgroup association
338 * @sk_write_pending: a write to stream socket waits to start
339 * @sk_disconnects: number of disconnect operations performed on this sock
340 * @sk_state_change: callback to indicate change in the state of the sock
341 * @sk_data_ready: callback to indicate there is data to be processed
342 * @sk_write_space: callback to indicate there is bf sending space available
343 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
344 * @sk_backlog_rcv: callback to process the backlog
345 * @sk_validate_xmit_skb: ptr to an optional validate function
346 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
347 * @sk_reuseport_cb: reuseport group container
348 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
349 * @sk_rcu: used during RCU grace period
350 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
351 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
352 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
353 * @sk_txtime_unused: unused txtime flags
354 * @ns_tracker: tracker for netns reference
355 * @sk_bind2_node: bind node in the bhash2 table
356 */
357struct sock {
358 /*
359 * Now struct inet_timewait_sock also uses sock_common, so please just
360 * don't add nothing before this first member (__sk_common) --acme
361 */
362 struct sock_common __sk_common;
363#define sk_node __sk_common.skc_node
364#define sk_nulls_node __sk_common.skc_nulls_node
365#define sk_refcnt __sk_common.skc_refcnt
366#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
367#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
368#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
369#endif
370
371#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
372#define sk_dontcopy_end __sk_common.skc_dontcopy_end
373#define sk_hash __sk_common.skc_hash
374#define sk_portpair __sk_common.skc_portpair
375#define sk_num __sk_common.skc_num
376#define sk_dport __sk_common.skc_dport
377#define sk_addrpair __sk_common.skc_addrpair
378#define sk_daddr __sk_common.skc_daddr
379#define sk_rcv_saddr __sk_common.skc_rcv_saddr
380#define sk_family __sk_common.skc_family
381#define sk_state __sk_common.skc_state
382#define sk_reuse __sk_common.skc_reuse
383#define sk_reuseport __sk_common.skc_reuseport
384#define sk_ipv6only __sk_common.skc_ipv6only
385#define sk_net_refcnt __sk_common.skc_net_refcnt
386#define sk_bound_dev_if __sk_common.skc_bound_dev_if
387#define sk_bind_node __sk_common.skc_bind_node
388#define sk_prot __sk_common.skc_prot
389#define sk_net __sk_common.skc_net
390#define sk_v6_daddr __sk_common.skc_v6_daddr
391#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
392#define sk_cookie __sk_common.skc_cookie
393#define sk_incoming_cpu __sk_common.skc_incoming_cpu
394#define sk_flags __sk_common.skc_flags
395#define sk_rxhash __sk_common.skc_rxhash
396
397 /* early demux fields */
398 struct dst_entry __rcu *sk_rx_dst;
399 int sk_rx_dst_ifindex;
400 u32 sk_rx_dst_cookie;
401
402 socket_lock_t sk_lock;
403 atomic_t sk_drops;
404 int sk_rcvlowat;
405 struct sk_buff_head sk_error_queue;
406 struct sk_buff_head sk_receive_queue;
407 /*
408 * The backlog queue is special, it is always used with
409 * the per-socket spinlock held and requires low latency
410 * access. Therefore we special case it's implementation.
411 * Note : rmem_alloc is in this structure to fill a hole
412 * on 64bit arches, not because its logically part of
413 * backlog.
414 */
415 struct {
416 atomic_t rmem_alloc;
417 int len;
418 struct sk_buff *head;
419 struct sk_buff *tail;
420 } sk_backlog;
421
422#define sk_rmem_alloc sk_backlog.rmem_alloc
423
424 int sk_forward_alloc;
425 u32 sk_reserved_mem;
426#ifdef CONFIG_NET_RX_BUSY_POLL
427 unsigned int sk_ll_usec;
428 /* ===== mostly read cache line ===== */
429 unsigned int sk_napi_id;
430#endif
431 int sk_rcvbuf;
432 int sk_disconnects;
433
434 struct sk_filter __rcu *sk_filter;
435 union {
436 struct socket_wq __rcu *sk_wq;
437 /* private: */
438 struct socket_wq *sk_wq_raw;
439 /* public: */
440 };
441#ifdef CONFIG_XFRM
442 struct xfrm_policy __rcu *sk_policy[2];
443#endif
444
445 struct dst_entry __rcu *sk_dst_cache;
446 atomic_t sk_omem_alloc;
447 int sk_sndbuf;
448
449 /* ===== cache line for TX ===== */
450 int sk_wmem_queued;
451 refcount_t sk_wmem_alloc;
452 unsigned long sk_tsq_flags;
453 union {
454 struct sk_buff *sk_send_head;
455 struct rb_root tcp_rtx_queue;
456 };
457 struct sk_buff_head sk_write_queue;
458 __s32 sk_peek_off;
459 int sk_write_pending;
460 __u32 sk_dst_pending_confirm;
461 u32 sk_pacing_status; /* see enum sk_pacing */
462 long sk_sndtimeo;
463 struct timer_list sk_timer;
464 __u32 sk_priority;
465 __u32 sk_mark;
466 unsigned long sk_pacing_rate; /* bytes per second */
467 unsigned long sk_max_pacing_rate;
468 struct page_frag sk_frag;
469 netdev_features_t sk_route_caps;
470 int sk_gso_type;
471 unsigned int sk_gso_max_size;
472 gfp_t sk_allocation;
473 __u32 sk_txhash;
474
475 /*
476 * Because of non atomicity rules, all
477 * changes are protected by socket lock.
478 */
479 u8 sk_gso_disabled : 1,
480 sk_kern_sock : 1,
481 sk_no_check_tx : 1,
482 sk_no_check_rx : 1,
483 sk_userlocks : 4;
484 u8 sk_pacing_shift;
485 u16 sk_type;
486 u16 sk_protocol;
487 u16 sk_gso_max_segs;
488 unsigned long sk_lingertime;
489 struct proto *sk_prot_creator;
490 rwlock_t sk_callback_lock;
491 int sk_err,
492 sk_err_soft;
493 u32 sk_ack_backlog;
494 u32 sk_max_ack_backlog;
495 kuid_t sk_uid;
496 u8 sk_txrehash;
497#ifdef CONFIG_NET_RX_BUSY_POLL
498 u8 sk_prefer_busy_poll;
499 u16 sk_busy_poll_budget;
500#endif
501 spinlock_t sk_peer_lock;
502 int sk_bind_phc;
503 struct pid *sk_peer_pid;
504 const struct cred *sk_peer_cred;
505
506 long sk_rcvtimeo;
507 ktime_t sk_stamp;
508#if BITS_PER_LONG==32
509 seqlock_t sk_stamp_seq;
510#endif
511 atomic_t sk_tskey;
512 atomic_t sk_zckey;
513 u32 sk_tsflags;
514 u8 sk_shutdown;
515
516 u8 sk_clockid;
517 u8 sk_txtime_deadline_mode : 1,
518 sk_txtime_report_errors : 1,
519 sk_txtime_unused : 6;
520 bool sk_use_task_frag;
521
522 struct socket *sk_socket;
523 void *sk_user_data;
524#ifdef CONFIG_SECURITY
525 void *sk_security;
526#endif
527 struct sock_cgroup_data sk_cgrp_data;
528 struct mem_cgroup *sk_memcg;
529 void (*sk_state_change)(struct sock *sk);
530 void (*sk_data_ready)(struct sock *sk);
531 void (*sk_write_space)(struct sock *sk);
532 void (*sk_error_report)(struct sock *sk);
533 int (*sk_backlog_rcv)(struct sock *sk,
534 struct sk_buff *skb);
535#ifdef CONFIG_SOCK_VALIDATE_XMIT
536 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
537 struct net_device *dev,
538 struct sk_buff *skb);
539#endif
540 void (*sk_destruct)(struct sock *sk);
541 struct sock_reuseport __rcu *sk_reuseport_cb;
542#ifdef CONFIG_BPF_SYSCALL
543 struct bpf_local_storage __rcu *sk_bpf_storage;
544#endif
545 struct rcu_head sk_rcu;
546 netns_tracker ns_tracker;
547 struct hlist_node sk_bind2_node;
548};
549
550enum sk_pacing {
551 SK_PACING_NONE = 0,
552 SK_PACING_NEEDED = 1,
553 SK_PACING_FQ = 2,
554};
555
556/* flag bits in sk_user_data
557 *
558 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
559 * not be suitable for copying when cloning the socket. For instance,
560 * it can point to a reference counted object. sk_user_data bottom
561 * bit is set if pointer must not be copied.
562 *
563 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
564 * managed/owned by a BPF reuseport array. This bit should be set
565 * when sk_user_data's sk is added to the bpf's reuseport_array.
566 *
567 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
568 * sk_user_data points to psock type. This bit should be set
569 * when sk_user_data is assigned to a psock object.
570 */
571#define SK_USER_DATA_NOCOPY 1UL
572#define SK_USER_DATA_BPF 2UL
573#define SK_USER_DATA_PSOCK 4UL
574#define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
575 SK_USER_DATA_PSOCK)
576
577/**
578 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
579 * @sk: socket
580 */
581static inline bool sk_user_data_is_nocopy(const struct sock *sk)
582{
583 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
584}
585
586#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
587
588/**
589 * __locked_read_sk_user_data_with_flags - return the pointer
590 * only if argument flags all has been set in sk_user_data. Otherwise
591 * return NULL
592 *
593 * @sk: socket
594 * @flags: flag bits
595 *
596 * The caller must be holding sk->sk_callback_lock.
597 */
598static inline void *
599__locked_read_sk_user_data_with_flags(const struct sock *sk,
600 uintptr_t flags)
601{
602 uintptr_t sk_user_data =
603 (uintptr_t)rcu_dereference_check(__sk_user_data(sk),
604 lockdep_is_held(&sk->sk_callback_lock));
605
606 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
607
608 if ((sk_user_data & flags) == flags)
609 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
610 return NULL;
611}
612
613/**
614 * __rcu_dereference_sk_user_data_with_flags - return the pointer
615 * only if argument flags all has been set in sk_user_data. Otherwise
616 * return NULL
617 *
618 * @sk: socket
619 * @flags: flag bits
620 */
621static inline void *
622__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
623 uintptr_t flags)
624{
625 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
626
627 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
628
629 if ((sk_user_data & flags) == flags)
630 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
631 return NULL;
632}
633
634#define rcu_dereference_sk_user_data(sk) \
635 __rcu_dereference_sk_user_data_with_flags(sk, 0)
636#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
637({ \
638 uintptr_t __tmp1 = (uintptr_t)(ptr), \
639 __tmp2 = (uintptr_t)(flags); \
640 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
641 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
642 rcu_assign_pointer(__sk_user_data((sk)), \
643 __tmp1 | __tmp2); \
644})
645#define rcu_assign_sk_user_data(sk, ptr) \
646 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
647
648static inline
649struct net *sock_net(const struct sock *sk)
650{
651 return read_pnet(pnet: &sk->sk_net);
652}
653
654static inline
655void sock_net_set(struct sock *sk, struct net *net)
656{
657 write_pnet(pnet: &sk->sk_net, net);
658}
659
660/*
661 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
662 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
663 * on a socket means that the socket will reuse everybody else's port
664 * without looking at the other's sk_reuse value.
665 */
666
667#define SK_NO_REUSE 0
668#define SK_CAN_REUSE 1
669#define SK_FORCE_REUSE 2
670
671int sk_set_peek_off(struct sock *sk, int val);
672
673static inline int sk_peek_offset(const struct sock *sk, int flags)
674{
675 if (unlikely(flags & MSG_PEEK)) {
676 return READ_ONCE(sk->sk_peek_off);
677 }
678
679 return 0;
680}
681
682static inline void sk_peek_offset_bwd(struct sock *sk, int val)
683{
684 s32 off = READ_ONCE(sk->sk_peek_off);
685
686 if (unlikely(off >= 0)) {
687 off = max_t(s32, off - val, 0);
688 WRITE_ONCE(sk->sk_peek_off, off);
689 }
690}
691
692static inline void sk_peek_offset_fwd(struct sock *sk, int val)
693{
694 sk_peek_offset_bwd(sk, val: -val);
695}
696
697/*
698 * Hashed lists helper routines
699 */
700static inline struct sock *sk_entry(const struct hlist_node *node)
701{
702 return hlist_entry(node, struct sock, sk_node);
703}
704
705static inline struct sock *__sk_head(const struct hlist_head *head)
706{
707 return hlist_entry(head->first, struct sock, sk_node);
708}
709
710static inline struct sock *sk_head(const struct hlist_head *head)
711{
712 return hlist_empty(h: head) ? NULL : __sk_head(head);
713}
714
715static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
716{
717 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
718}
719
720static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
721{
722 return hlist_nulls_empty(h: head) ? NULL : __sk_nulls_head(head);
723}
724
725static inline struct sock *sk_next(const struct sock *sk)
726{
727 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
728}
729
730static inline struct sock *sk_nulls_next(const struct sock *sk)
731{
732 return (!is_a_nulls(ptr: sk->sk_nulls_node.next)) ?
733 hlist_nulls_entry(sk->sk_nulls_node.next,
734 struct sock, sk_nulls_node) :
735 NULL;
736}
737
738static inline bool sk_unhashed(const struct sock *sk)
739{
740 return hlist_unhashed(h: &sk->sk_node);
741}
742
743static inline bool sk_hashed(const struct sock *sk)
744{
745 return !sk_unhashed(sk);
746}
747
748static inline void sk_node_init(struct hlist_node *node)
749{
750 node->pprev = NULL;
751}
752
753static inline void __sk_del_node(struct sock *sk)
754{
755 __hlist_del(n: &sk->sk_node);
756}
757
758/* NB: equivalent to hlist_del_init_rcu */
759static inline bool __sk_del_node_init(struct sock *sk)
760{
761 if (sk_hashed(sk)) {
762 __sk_del_node(sk);
763 sk_node_init(node: &sk->sk_node);
764 return true;
765 }
766 return false;
767}
768
769/* Grab socket reference count. This operation is valid only
770 when sk is ALREADY grabbed f.e. it is found in hash table
771 or a list and the lookup is made under lock preventing hash table
772 modifications.
773 */
774
775static __always_inline void sock_hold(struct sock *sk)
776{
777 refcount_inc(r: &sk->sk_refcnt);
778}
779
780/* Ungrab socket in the context, which assumes that socket refcnt
781 cannot hit zero, f.e. it is true in context of any socketcall.
782 */
783static __always_inline void __sock_put(struct sock *sk)
784{
785 refcount_dec(r: &sk->sk_refcnt);
786}
787
788static inline bool sk_del_node_init(struct sock *sk)
789{
790 bool rc = __sk_del_node_init(sk);
791
792 if (rc) {
793 /* paranoid for a while -acme */
794 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
795 __sock_put(sk);
796 }
797 return rc;
798}
799#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
800
801static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
802{
803 if (sk_hashed(sk)) {
804 hlist_nulls_del_init_rcu(n: &sk->sk_nulls_node);
805 return true;
806 }
807 return false;
808}
809
810static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
811{
812 bool rc = __sk_nulls_del_node_init_rcu(sk);
813
814 if (rc) {
815 /* paranoid for a while -acme */
816 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
817 __sock_put(sk);
818 }
819 return rc;
820}
821
822static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
823{
824 hlist_add_head(n: &sk->sk_node, h: list);
825}
826
827static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
828{
829 sock_hold(sk);
830 __sk_add_node(sk, list);
831}
832
833static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
834{
835 sock_hold(sk);
836 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
837 sk->sk_family == AF_INET6)
838 hlist_add_tail_rcu(n: &sk->sk_node, h: list);
839 else
840 hlist_add_head_rcu(n: &sk->sk_node, h: list);
841}
842
843static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
844{
845 sock_hold(sk);
846 hlist_add_tail_rcu(n: &sk->sk_node, h: list);
847}
848
849static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
850{
851 hlist_nulls_add_head_rcu(n: &sk->sk_nulls_node, h: list);
852}
853
854static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
855{
856 hlist_nulls_add_tail_rcu(n: &sk->sk_nulls_node, h: list);
857}
858
859static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
860{
861 sock_hold(sk);
862 __sk_nulls_add_node_rcu(sk, list);
863}
864
865static inline void __sk_del_bind_node(struct sock *sk)
866{
867 __hlist_del(n: &sk->sk_bind_node);
868}
869
870static inline void sk_add_bind_node(struct sock *sk,
871 struct hlist_head *list)
872{
873 hlist_add_head(n: &sk->sk_bind_node, h: list);
874}
875
876static inline void __sk_del_bind2_node(struct sock *sk)
877{
878 __hlist_del(n: &sk->sk_bind2_node);
879}
880
881static inline void sk_add_bind2_node(struct sock *sk, struct hlist_head *list)
882{
883 hlist_add_head(n: &sk->sk_bind2_node, h: list);
884}
885
886#define sk_for_each(__sk, list) \
887 hlist_for_each_entry(__sk, list, sk_node)
888#define sk_for_each_rcu(__sk, list) \
889 hlist_for_each_entry_rcu(__sk, list, sk_node)
890#define sk_nulls_for_each(__sk, node, list) \
891 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
892#define sk_nulls_for_each_rcu(__sk, node, list) \
893 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
894#define sk_for_each_from(__sk) \
895 hlist_for_each_entry_from(__sk, sk_node)
896#define sk_nulls_for_each_from(__sk, node) \
897 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
898 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
899#define sk_for_each_safe(__sk, tmp, list) \
900 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
901#define sk_for_each_bound(__sk, list) \
902 hlist_for_each_entry(__sk, list, sk_bind_node)
903#define sk_for_each_bound_bhash2(__sk, list) \
904 hlist_for_each_entry(__sk, list, sk_bind2_node)
905
906/**
907 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
908 * @tpos: the type * to use as a loop cursor.
909 * @pos: the &struct hlist_node to use as a loop cursor.
910 * @head: the head for your list.
911 * @offset: offset of hlist_node within the struct.
912 *
913 */
914#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
915 for (pos = rcu_dereference(hlist_first_rcu(head)); \
916 pos != NULL && \
917 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
918 pos = rcu_dereference(hlist_next_rcu(pos)))
919
920static inline struct user_namespace *sk_user_ns(const struct sock *sk)
921{
922 /* Careful only use this in a context where these parameters
923 * can not change and must all be valid, such as recvmsg from
924 * userspace.
925 */
926 return sk->sk_socket->file->f_cred->user_ns;
927}
928
929/* Sock flags */
930enum sock_flags {
931 SOCK_DEAD,
932 SOCK_DONE,
933 SOCK_URGINLINE,
934 SOCK_KEEPOPEN,
935 SOCK_LINGER,
936 SOCK_DESTROY,
937 SOCK_BROADCAST,
938 SOCK_TIMESTAMP,
939 SOCK_ZAPPED,
940 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
941 SOCK_DBG, /* %SO_DEBUG setting */
942 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
943 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
944 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
945 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
946 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
947 SOCK_FASYNC, /* fasync() active */
948 SOCK_RXQ_OVFL,
949 SOCK_ZEROCOPY, /* buffers from userspace */
950 SOCK_WIFI_STATUS, /* push wifi status to userspace */
951 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
952 * Will use last 4 bytes of packet sent from
953 * user-space instead.
954 */
955 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
956 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
957 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
958 SOCK_TXTIME,
959 SOCK_XDP, /* XDP is attached */
960 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
961 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */
962};
963
964#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
965
966static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
967{
968 nsk->sk_flags = osk->sk_flags;
969}
970
971static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
972{
973 __set_bit(flag, &sk->sk_flags);
974}
975
976static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
977{
978 __clear_bit(flag, &sk->sk_flags);
979}
980
981static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
982 int valbool)
983{
984 if (valbool)
985 sock_set_flag(sk, flag: bit);
986 else
987 sock_reset_flag(sk, flag: bit);
988}
989
990static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
991{
992 return test_bit(flag, &sk->sk_flags);
993}
994
995#ifdef CONFIG_NET
996DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
997static inline int sk_memalloc_socks(void)
998{
999 return static_branch_unlikely(&memalloc_socks_key);
1000}
1001
1002void __receive_sock(struct file *file);
1003#else
1004
1005static inline int sk_memalloc_socks(void)
1006{
1007 return 0;
1008}
1009
1010static inline void __receive_sock(struct file *file)
1011{ }
1012#endif
1013
1014static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1015{
1016 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1017}
1018
1019static inline void sk_acceptq_removed(struct sock *sk)
1020{
1021 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1022}
1023
1024static inline void sk_acceptq_added(struct sock *sk)
1025{
1026 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1027}
1028
1029/* Note: If you think the test should be:
1030 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1031 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1032 */
1033static inline bool sk_acceptq_is_full(const struct sock *sk)
1034{
1035 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1036}
1037
1038/*
1039 * Compute minimal free write space needed to queue new packets.
1040 */
1041static inline int sk_stream_min_wspace(const struct sock *sk)
1042{
1043 return READ_ONCE(sk->sk_wmem_queued) >> 1;
1044}
1045
1046static inline int sk_stream_wspace(const struct sock *sk)
1047{
1048 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1049}
1050
1051static inline void sk_wmem_queued_add(struct sock *sk, int val)
1052{
1053 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1054}
1055
1056static inline void sk_forward_alloc_add(struct sock *sk, int val)
1057{
1058 /* Paired with lockless reads of sk->sk_forward_alloc */
1059 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1060}
1061
1062void sk_stream_write_space(struct sock *sk);
1063
1064/* OOB backlog add */
1065static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1066{
1067 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1068 skb_dst_force(skb);
1069
1070 if (!sk->sk_backlog.tail)
1071 WRITE_ONCE(sk->sk_backlog.head, skb);
1072 else
1073 sk->sk_backlog.tail->next = skb;
1074
1075 WRITE_ONCE(sk->sk_backlog.tail, skb);
1076 skb->next = NULL;
1077}
1078
1079/*
1080 * Take into account size of receive queue and backlog queue
1081 * Do not take into account this skb truesize,
1082 * to allow even a single big packet to come.
1083 */
1084static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1085{
1086 unsigned int qsize = sk->sk_backlog.len + atomic_read(v: &sk->sk_rmem_alloc);
1087
1088 return qsize > limit;
1089}
1090
1091/* The per-socket spinlock must be held here. */
1092static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1093 unsigned int limit)
1094{
1095 if (sk_rcvqueues_full(sk, limit))
1096 return -ENOBUFS;
1097
1098 /*
1099 * If the skb was allocated from pfmemalloc reserves, only
1100 * allow SOCK_MEMALLOC sockets to use it as this socket is
1101 * helping free memory
1102 */
1103 if (skb_pfmemalloc(skb) && !sock_flag(sk, flag: SOCK_MEMALLOC))
1104 return -ENOMEM;
1105
1106 __sk_add_backlog(sk, skb);
1107 sk->sk_backlog.len += skb->truesize;
1108 return 0;
1109}
1110
1111int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1112
1113INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1114INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1115
1116static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1117{
1118 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1119 return __sk_backlog_rcv(sk, skb);
1120
1121 return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1122 tcp_v6_do_rcv,
1123 tcp_v4_do_rcv,
1124 sk, skb);
1125}
1126
1127static inline void sk_incoming_cpu_update(struct sock *sk)
1128{
1129 int cpu = raw_smp_processor_id();
1130
1131 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1132 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1133}
1134
1135static inline void sock_rps_record_flow_hash(__u32 hash)
1136{
1137#ifdef CONFIG_RPS
1138 struct rps_sock_flow_table *sock_flow_table;
1139
1140 rcu_read_lock();
1141 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1142 rps_record_sock_flow(table: sock_flow_table, hash);
1143 rcu_read_unlock();
1144#endif
1145}
1146
1147static inline void sock_rps_record_flow(const struct sock *sk)
1148{
1149#ifdef CONFIG_RPS
1150 if (static_branch_unlikely(&rfs_needed)) {
1151 /* Reading sk->sk_rxhash might incur an expensive cache line
1152 * miss.
1153 *
1154 * TCP_ESTABLISHED does cover almost all states where RFS
1155 * might be useful, and is cheaper [1] than testing :
1156 * IPv4: inet_sk(sk)->inet_daddr
1157 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1158 * OR an additional socket flag
1159 * [1] : sk_state and sk_prot are in the same cache line.
1160 */
1161 if (sk->sk_state == TCP_ESTABLISHED) {
1162 /* This READ_ONCE() is paired with the WRITE_ONCE()
1163 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
1164 */
1165 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1166 }
1167 }
1168#endif
1169}
1170
1171static inline void sock_rps_save_rxhash(struct sock *sk,
1172 const struct sk_buff *skb)
1173{
1174#ifdef CONFIG_RPS
1175 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1176 * here, and another one in sock_rps_record_flow().
1177 */
1178 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1179 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1180#endif
1181}
1182
1183static inline void sock_rps_reset_rxhash(struct sock *sk)
1184{
1185#ifdef CONFIG_RPS
1186 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1187 WRITE_ONCE(sk->sk_rxhash, 0);
1188#endif
1189}
1190
1191#define sk_wait_event(__sk, __timeo, __condition, __wait) \
1192 ({ int __rc, __dis = __sk->sk_disconnects; \
1193 release_sock(__sk); \
1194 __rc = __condition; \
1195 if (!__rc) { \
1196 *(__timeo) = wait_woken(__wait, \
1197 TASK_INTERRUPTIBLE, \
1198 *(__timeo)); \
1199 } \
1200 sched_annotate_sleep(); \
1201 lock_sock(__sk); \
1202 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1203 __rc; \
1204 })
1205
1206int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1207int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1208void sk_stream_wait_close(struct sock *sk, long timeo_p);
1209int sk_stream_error(struct sock *sk, int flags, int err);
1210void sk_stream_kill_queues(struct sock *sk);
1211void sk_set_memalloc(struct sock *sk);
1212void sk_clear_memalloc(struct sock *sk);
1213
1214void __sk_flush_backlog(struct sock *sk);
1215
1216static inline bool sk_flush_backlog(struct sock *sk)
1217{
1218 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1219 __sk_flush_backlog(sk);
1220 return true;
1221 }
1222 return false;
1223}
1224
1225int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1226
1227struct request_sock_ops;
1228struct timewait_sock_ops;
1229struct inet_hashinfo;
1230struct raw_hashinfo;
1231struct smc_hashinfo;
1232struct module;
1233struct sk_psock;
1234
1235/*
1236 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1237 * un-modified. Special care is taken when initializing object to zero.
1238 */
1239static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1240{
1241 if (offsetof(struct sock, sk_node.next) != 0)
1242 memset(sk, 0, offsetof(struct sock, sk_node.next));
1243 memset(&sk->sk_node.pprev, 0,
1244 size - offsetof(struct sock, sk_node.pprev));
1245}
1246
1247/* Networking protocol blocks we attach to sockets.
1248 * socket layer -> transport layer interface
1249 */
1250struct proto {
1251 void (*close)(struct sock *sk,
1252 long timeout);
1253 int (*pre_connect)(struct sock *sk,
1254 struct sockaddr *uaddr,
1255 int addr_len);
1256 int (*connect)(struct sock *sk,
1257 struct sockaddr *uaddr,
1258 int addr_len);
1259 int (*disconnect)(struct sock *sk, int flags);
1260
1261 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1262 bool kern);
1263
1264 int (*ioctl)(struct sock *sk, int cmd,
1265 int *karg);
1266 int (*init)(struct sock *sk);
1267 void (*destroy)(struct sock *sk);
1268 void (*shutdown)(struct sock *sk, int how);
1269 int (*setsockopt)(struct sock *sk, int level,
1270 int optname, sockptr_t optval,
1271 unsigned int optlen);
1272 int (*getsockopt)(struct sock *sk, int level,
1273 int optname, char __user *optval,
1274 int __user *option);
1275 void (*keepalive)(struct sock *sk, int valbool);
1276#ifdef CONFIG_COMPAT
1277 int (*compat_ioctl)(struct sock *sk,
1278 unsigned int cmd, unsigned long arg);
1279#endif
1280 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1281 size_t len);
1282 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1283 size_t len, int flags, int *addr_len);
1284 void (*splice_eof)(struct socket *sock);
1285 int (*bind)(struct sock *sk,
1286 struct sockaddr *addr, int addr_len);
1287 int (*bind_add)(struct sock *sk,
1288 struct sockaddr *addr, int addr_len);
1289
1290 int (*backlog_rcv) (struct sock *sk,
1291 struct sk_buff *skb);
1292 bool (*bpf_bypass_getsockopt)(int level,
1293 int optname);
1294
1295 void (*release_cb)(struct sock *sk);
1296
1297 /* Keeping track of sk's, looking them up, and port selection methods. */
1298 int (*hash)(struct sock *sk);
1299 void (*unhash)(struct sock *sk);
1300 void (*rehash)(struct sock *sk);
1301 int (*get_port)(struct sock *sk, unsigned short snum);
1302 void (*put_port)(struct sock *sk);
1303#ifdef CONFIG_BPF_SYSCALL
1304 int (*psock_update_sk_prot)(struct sock *sk,
1305 struct sk_psock *psock,
1306 bool restore);
1307#endif
1308
1309 /* Keeping track of sockets in use */
1310#ifdef CONFIG_PROC_FS
1311 unsigned int inuse_idx;
1312#endif
1313
1314#if IS_ENABLED(CONFIG_MPTCP)
1315 int (*forward_alloc_get)(const struct sock *sk);
1316#endif
1317
1318 bool (*stream_memory_free)(const struct sock *sk, int wake);
1319 bool (*sock_is_readable)(struct sock *sk);
1320 /* Memory pressure */
1321 void (*enter_memory_pressure)(struct sock *sk);
1322 void (*leave_memory_pressure)(struct sock *sk);
1323 atomic_long_t *memory_allocated; /* Current allocated memory. */
1324 int __percpu *per_cpu_fw_alloc;
1325 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1326
1327 /*
1328 * Pressure flag: try to collapse.
1329 * Technical note: it is used by multiple contexts non atomically.
1330 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1331 * All the __sk_mem_schedule() is of this nature: accounting
1332 * is strict, actions are advisory and have some latency.
1333 */
1334 unsigned long *memory_pressure;
1335 long *sysctl_mem;
1336
1337 int *sysctl_wmem;
1338 int *sysctl_rmem;
1339 u32 sysctl_wmem_offset;
1340 u32 sysctl_rmem_offset;
1341
1342 int max_header;
1343 bool no_autobind;
1344
1345 struct kmem_cache *slab;
1346 unsigned int obj_size;
1347 unsigned int ipv6_pinfo_offset;
1348 slab_flags_t slab_flags;
1349 unsigned int useroffset; /* Usercopy region offset */
1350 unsigned int usersize; /* Usercopy region size */
1351
1352 unsigned int __percpu *orphan_count;
1353
1354 struct request_sock_ops *rsk_prot;
1355 struct timewait_sock_ops *twsk_prot;
1356
1357 union {
1358 struct inet_hashinfo *hashinfo;
1359 struct udp_table *udp_table;
1360 struct raw_hashinfo *raw_hash;
1361 struct smc_hashinfo *smc_hash;
1362 } h;
1363
1364 struct module *owner;
1365
1366 char name[32];
1367
1368 struct list_head node;
1369 int (*diag_destroy)(struct sock *sk, int err);
1370} __randomize_layout;
1371
1372int proto_register(struct proto *prot, int alloc_slab);
1373void proto_unregister(struct proto *prot);
1374int sock_load_diag_module(int family, int protocol);
1375
1376INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1377
1378static inline int sk_forward_alloc_get(const struct sock *sk)
1379{
1380#if IS_ENABLED(CONFIG_MPTCP)
1381 if (sk->sk_prot->forward_alloc_get)
1382 return sk->sk_prot->forward_alloc_get(sk);
1383#endif
1384 return READ_ONCE(sk->sk_forward_alloc);
1385}
1386
1387static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1388{
1389 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1390 return false;
1391
1392 return sk->sk_prot->stream_memory_free ?
1393 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1394 tcp_stream_memory_free, sk, wake) : true;
1395}
1396
1397static inline bool sk_stream_memory_free(const struct sock *sk)
1398{
1399 return __sk_stream_memory_free(sk, wake: 0);
1400}
1401
1402static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1403{
1404 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1405 __sk_stream_memory_free(sk, wake);
1406}
1407
1408static inline bool sk_stream_is_writeable(const struct sock *sk)
1409{
1410 return __sk_stream_is_writeable(sk, wake: 0);
1411}
1412
1413static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1414 struct cgroup *ancestor)
1415{
1416#ifdef CONFIG_SOCK_CGROUP_DATA
1417 return cgroup_is_descendant(cgrp: sock_cgroup_ptr(skcd: &sk->sk_cgrp_data),
1418 ancestor);
1419#else
1420 return -ENOTSUPP;
1421#endif
1422}
1423
1424static inline bool sk_has_memory_pressure(const struct sock *sk)
1425{
1426 return sk->sk_prot->memory_pressure != NULL;
1427}
1428
1429static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1430{
1431 return sk->sk_prot->memory_pressure &&
1432 !!READ_ONCE(*sk->sk_prot->memory_pressure);
1433}
1434
1435static inline bool sk_under_memory_pressure(const struct sock *sk)
1436{
1437 if (!sk->sk_prot->memory_pressure)
1438 return false;
1439
1440 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1441 mem_cgroup_under_socket_pressure(memcg: sk->sk_memcg))
1442 return true;
1443
1444 return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1445}
1446
1447static inline long
1448proto_memory_allocated(const struct proto *prot)
1449{
1450 return max(0L, atomic_long_read(prot->memory_allocated));
1451}
1452
1453static inline long
1454sk_memory_allocated(const struct sock *sk)
1455{
1456 return proto_memory_allocated(prot: sk->sk_prot);
1457}
1458
1459/* 1 MB per cpu, in page units */
1460#define SK_MEMORY_PCPU_RESERVE (1 << (20 - PAGE_SHIFT))
1461
1462static inline void
1463sk_memory_allocated_add(struct sock *sk, int amt)
1464{
1465 int local_reserve;
1466
1467 preempt_disable();
1468 local_reserve = __this_cpu_add_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1469 if (local_reserve >= SK_MEMORY_PCPU_RESERVE) {
1470 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1471 atomic_long_add(i: local_reserve, v: sk->sk_prot->memory_allocated);
1472 }
1473 preempt_enable();
1474}
1475
1476static inline void
1477sk_memory_allocated_sub(struct sock *sk, int amt)
1478{
1479 int local_reserve;
1480
1481 preempt_disable();
1482 local_reserve = __this_cpu_sub_return(*sk->sk_prot->per_cpu_fw_alloc, amt);
1483 if (local_reserve <= -SK_MEMORY_PCPU_RESERVE) {
1484 __this_cpu_sub(*sk->sk_prot->per_cpu_fw_alloc, local_reserve);
1485 atomic_long_add(i: local_reserve, v: sk->sk_prot->memory_allocated);
1486 }
1487 preempt_enable();
1488}
1489
1490#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1491
1492static inline void sk_sockets_allocated_dec(struct sock *sk)
1493{
1494 percpu_counter_add_batch(fbc: sk->sk_prot->sockets_allocated, amount: -1,
1495 SK_ALLOC_PERCPU_COUNTER_BATCH);
1496}
1497
1498static inline void sk_sockets_allocated_inc(struct sock *sk)
1499{
1500 percpu_counter_add_batch(fbc: sk->sk_prot->sockets_allocated, amount: 1,
1501 SK_ALLOC_PERCPU_COUNTER_BATCH);
1502}
1503
1504static inline u64
1505sk_sockets_allocated_read_positive(struct sock *sk)
1506{
1507 return percpu_counter_read_positive(fbc: sk->sk_prot->sockets_allocated);
1508}
1509
1510static inline int
1511proto_sockets_allocated_sum_positive(struct proto *prot)
1512{
1513 return percpu_counter_sum_positive(fbc: prot->sockets_allocated);
1514}
1515
1516static inline bool
1517proto_memory_pressure(struct proto *prot)
1518{
1519 if (!prot->memory_pressure)
1520 return false;
1521 return !!READ_ONCE(*prot->memory_pressure);
1522}
1523
1524
1525#ifdef CONFIG_PROC_FS
1526#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1527struct prot_inuse {
1528 int all;
1529 int val[PROTO_INUSE_NR];
1530};
1531
1532static inline void sock_prot_inuse_add(const struct net *net,
1533 const struct proto *prot, int val)
1534{
1535 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1536}
1537
1538static inline void sock_inuse_add(const struct net *net, int val)
1539{
1540 this_cpu_add(net->core.prot_inuse->all, val);
1541}
1542
1543int sock_prot_inuse_get(struct net *net, struct proto *proto);
1544int sock_inuse_get(struct net *net);
1545#else
1546static inline void sock_prot_inuse_add(const struct net *net,
1547 const struct proto *prot, int val)
1548{
1549}
1550
1551static inline void sock_inuse_add(const struct net *net, int val)
1552{
1553}
1554#endif
1555
1556
1557/* With per-bucket locks this operation is not-atomic, so that
1558 * this version is not worse.
1559 */
1560static inline int __sk_prot_rehash(struct sock *sk)
1561{
1562 sk->sk_prot->unhash(sk);
1563 return sk->sk_prot->hash(sk);
1564}
1565
1566/* About 10 seconds */
1567#define SOCK_DESTROY_TIME (10*HZ)
1568
1569/* Sockets 0-1023 can't be bound to unless you are superuser */
1570#define PROT_SOCK 1024
1571
1572#define SHUTDOWN_MASK 3
1573#define RCV_SHUTDOWN 1
1574#define SEND_SHUTDOWN 2
1575
1576#define SOCK_BINDADDR_LOCK 4
1577#define SOCK_BINDPORT_LOCK 8
1578
1579struct socket_alloc {
1580 struct socket socket;
1581 struct inode vfs_inode;
1582};
1583
1584static inline struct socket *SOCKET_I(struct inode *inode)
1585{
1586 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1587}
1588
1589static inline struct inode *SOCK_INODE(struct socket *socket)
1590{
1591 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1592}
1593
1594/*
1595 * Functions for memory accounting
1596 */
1597int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1598int __sk_mem_schedule(struct sock *sk, int size, int kind);
1599void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1600void __sk_mem_reclaim(struct sock *sk, int amount);
1601
1602#define SK_MEM_SEND 0
1603#define SK_MEM_RECV 1
1604
1605/* sysctl_mem values are in pages */
1606static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1607{
1608 return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1609}
1610
1611static inline int sk_mem_pages(int amt)
1612{
1613 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1614}
1615
1616static inline bool sk_has_account(struct sock *sk)
1617{
1618 /* return true if protocol supports memory accounting */
1619 return !!sk->sk_prot->memory_allocated;
1620}
1621
1622static inline bool sk_wmem_schedule(struct sock *sk, int size)
1623{
1624 int delta;
1625
1626 if (!sk_has_account(sk))
1627 return true;
1628 delta = size - sk->sk_forward_alloc;
1629 return delta <= 0 || __sk_mem_schedule(sk, size: delta, SK_MEM_SEND);
1630}
1631
1632static inline bool
1633sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1634{
1635 int delta;
1636
1637 if (!sk_has_account(sk))
1638 return true;
1639 delta = size - sk->sk_forward_alloc;
1640 return delta <= 0 || __sk_mem_schedule(sk, size: delta, SK_MEM_RECV) ||
1641 skb_pfmemalloc(skb);
1642}
1643
1644static inline int sk_unused_reserved_mem(const struct sock *sk)
1645{
1646 int unused_mem;
1647
1648 if (likely(!sk->sk_reserved_mem))
1649 return 0;
1650
1651 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1652 atomic_read(v: &sk->sk_rmem_alloc);
1653
1654 return unused_mem > 0 ? unused_mem : 0;
1655}
1656
1657static inline void sk_mem_reclaim(struct sock *sk)
1658{
1659 int reclaimable;
1660
1661 if (!sk_has_account(sk))
1662 return;
1663
1664 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1665
1666 if (reclaimable >= (int)PAGE_SIZE)
1667 __sk_mem_reclaim(sk, amount: reclaimable);
1668}
1669
1670static inline void sk_mem_reclaim_final(struct sock *sk)
1671{
1672 sk->sk_reserved_mem = 0;
1673 sk_mem_reclaim(sk);
1674}
1675
1676static inline void sk_mem_charge(struct sock *sk, int size)
1677{
1678 if (!sk_has_account(sk))
1679 return;
1680 sk_forward_alloc_add(sk, val: -size);
1681}
1682
1683static inline void sk_mem_uncharge(struct sock *sk, int size)
1684{
1685 if (!sk_has_account(sk))
1686 return;
1687 sk_forward_alloc_add(sk, val: size);
1688 sk_mem_reclaim(sk);
1689}
1690
1691/*
1692 * Macro so as to not evaluate some arguments when
1693 * lockdep is not enabled.
1694 *
1695 * Mark both the sk_lock and the sk_lock.slock as a
1696 * per-address-family lock class.
1697 */
1698#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1699do { \
1700 sk->sk_lock.owned = 0; \
1701 init_waitqueue_head(&sk->sk_lock.wq); \
1702 spin_lock_init(&(sk)->sk_lock.slock); \
1703 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1704 sizeof((sk)->sk_lock)); \
1705 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1706 (skey), (sname)); \
1707 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1708} while (0)
1709
1710static inline bool lockdep_sock_is_held(const struct sock *sk)
1711{
1712 return lockdep_is_held(&sk->sk_lock) ||
1713 lockdep_is_held(&sk->sk_lock.slock);
1714}
1715
1716void lock_sock_nested(struct sock *sk, int subclass);
1717
1718static inline void lock_sock(struct sock *sk)
1719{
1720 lock_sock_nested(sk, subclass: 0);
1721}
1722
1723void __lock_sock(struct sock *sk);
1724void __release_sock(struct sock *sk);
1725void release_sock(struct sock *sk);
1726
1727/* BH context may only use the following locking interface. */
1728#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1729#define bh_lock_sock_nested(__sk) \
1730 spin_lock_nested(&((__sk)->sk_lock.slock), \
1731 SINGLE_DEPTH_NESTING)
1732#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1733
1734bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1735
1736/**
1737 * lock_sock_fast - fast version of lock_sock
1738 * @sk: socket
1739 *
1740 * This version should be used for very small section, where process wont block
1741 * return false if fast path is taken:
1742 *
1743 * sk_lock.slock locked, owned = 0, BH disabled
1744 *
1745 * return true if slow path is taken:
1746 *
1747 * sk_lock.slock unlocked, owned = 1, BH enabled
1748 */
1749static inline bool lock_sock_fast(struct sock *sk)
1750{
1751 /* The sk_lock has mutex_lock() semantics here. */
1752 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1753
1754 return __lock_sock_fast(sk);
1755}
1756
1757/* fast socket lock variant for caller already holding a [different] socket lock */
1758static inline bool lock_sock_fast_nested(struct sock *sk)
1759{
1760 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1761
1762 return __lock_sock_fast(sk);
1763}
1764
1765/**
1766 * unlock_sock_fast - complement of lock_sock_fast
1767 * @sk: socket
1768 * @slow: slow mode
1769 *
1770 * fast unlock socket for user context.
1771 * If slow mode is on, we call regular release_sock()
1772 */
1773static inline void unlock_sock_fast(struct sock *sk, bool slow)
1774 __releases(&sk->sk_lock.slock)
1775{
1776 if (slow) {
1777 release_sock(sk);
1778 __release(&sk->sk_lock.slock);
1779 } else {
1780 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1781 spin_unlock_bh(lock: &sk->sk_lock.slock);
1782 }
1783}
1784
1785void sockopt_lock_sock(struct sock *sk);
1786void sockopt_release_sock(struct sock *sk);
1787bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1788bool sockopt_capable(int cap);
1789
1790/* Used by processes to "lock" a socket state, so that
1791 * interrupts and bottom half handlers won't change it
1792 * from under us. It essentially blocks any incoming
1793 * packets, so that we won't get any new data or any
1794 * packets that change the state of the socket.
1795 *
1796 * While locked, BH processing will add new packets to
1797 * the backlog queue. This queue is processed by the
1798 * owner of the socket lock right before it is released.
1799 *
1800 * Since ~2.3.5 it is also exclusive sleep lock serializing
1801 * accesses from user process context.
1802 */
1803
1804static inline void sock_owned_by_me(const struct sock *sk)
1805{
1806#ifdef CONFIG_LOCKDEP
1807 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1808#endif
1809}
1810
1811static inline bool sock_owned_by_user(const struct sock *sk)
1812{
1813 sock_owned_by_me(sk);
1814 return sk->sk_lock.owned;
1815}
1816
1817static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1818{
1819 return sk->sk_lock.owned;
1820}
1821
1822static inline void sock_release_ownership(struct sock *sk)
1823{
1824 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1825 sk->sk_lock.owned = 0;
1826
1827 /* The sk_lock has mutex_unlock() semantics: */
1828 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1829}
1830
1831/* no reclassification while locks are held */
1832static inline bool sock_allow_reclassification(const struct sock *csk)
1833{
1834 struct sock *sk = (struct sock *)csk;
1835
1836 return !sock_owned_by_user_nocheck(sk) &&
1837 !spin_is_locked(lock: &sk->sk_lock.slock);
1838}
1839
1840struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1841 struct proto *prot, int kern);
1842void sk_free(struct sock *sk);
1843void sk_destruct(struct sock *sk);
1844struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1845void sk_free_unlock_clone(struct sock *sk);
1846
1847struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1848 gfp_t priority);
1849void __sock_wfree(struct sk_buff *skb);
1850void sock_wfree(struct sk_buff *skb);
1851struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1852 gfp_t priority);
1853void skb_orphan_partial(struct sk_buff *skb);
1854void sock_rfree(struct sk_buff *skb);
1855void sock_efree(struct sk_buff *skb);
1856#ifdef CONFIG_INET
1857void sock_edemux(struct sk_buff *skb);
1858void sock_pfree(struct sk_buff *skb);
1859#else
1860#define sock_edemux sock_efree
1861#endif
1862
1863int sk_setsockopt(struct sock *sk, int level, int optname,
1864 sockptr_t optval, unsigned int optlen);
1865int sock_setsockopt(struct socket *sock, int level, int op,
1866 sockptr_t optval, unsigned int optlen);
1867int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1868 int optname, sockptr_t optval, int optlen);
1869int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1870 int optname, sockptr_t optval, sockptr_t optlen);
1871
1872int sk_getsockopt(struct sock *sk, int level, int optname,
1873 sockptr_t optval, sockptr_t optlen);
1874int sock_gettstamp(struct socket *sock, void __user *userstamp,
1875 bool timeval, bool time32);
1876struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1877 unsigned long data_len, int noblock,
1878 int *errcode, int max_page_order);
1879
1880static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1881 unsigned long size,
1882 int noblock, int *errcode)
1883{
1884 return sock_alloc_send_pskb(sk, header_len: size, data_len: 0, noblock, errcode, max_page_order: 0);
1885}
1886
1887void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1888void sock_kfree_s(struct sock *sk, void *mem, int size);
1889void sock_kzfree_s(struct sock *sk, void *mem, int size);
1890void sk_send_sigurg(struct sock *sk);
1891
1892static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1893{
1894 if (sk->sk_socket)
1895 clear_bit(SOCK_SUPPORT_ZC, addr: &sk->sk_socket->flags);
1896 WRITE_ONCE(sk->sk_prot, proto);
1897}
1898
1899struct sockcm_cookie {
1900 u64 transmit_time;
1901 u32 mark;
1902 u32 tsflags;
1903};
1904
1905static inline void sockcm_init(struct sockcm_cookie *sockc,
1906 const struct sock *sk)
1907{
1908 *sockc = (struct sockcm_cookie) {
1909 .tsflags = READ_ONCE(sk->sk_tsflags)
1910 };
1911}
1912
1913int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1914 struct sockcm_cookie *sockc);
1915int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1916 struct sockcm_cookie *sockc);
1917
1918/*
1919 * Functions to fill in entries in struct proto_ops when a protocol
1920 * does not implement a particular function.
1921 */
1922int sock_no_bind(struct socket *, struct sockaddr *, int);
1923int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1924int sock_no_socketpair(struct socket *, struct socket *);
1925int sock_no_accept(struct socket *, struct socket *, int, bool);
1926int sock_no_getname(struct socket *, struct sockaddr *, int);
1927int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1928int sock_no_listen(struct socket *, int);
1929int sock_no_shutdown(struct socket *, int);
1930int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1931int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1932int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1933int sock_no_mmap(struct file *file, struct socket *sock,
1934 struct vm_area_struct *vma);
1935
1936/*
1937 * Functions to fill in entries in struct proto_ops when a protocol
1938 * uses the inet style.
1939 */
1940int sock_common_getsockopt(struct socket *sock, int level, int optname,
1941 char __user *optval, int __user *optlen);
1942int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1943 int flags);
1944int sock_common_setsockopt(struct socket *sock, int level, int optname,
1945 sockptr_t optval, unsigned int optlen);
1946
1947void sk_common_release(struct sock *sk);
1948
1949/*
1950 * Default socket callbacks and setup code
1951 */
1952
1953/* Initialise core socket variables using an explicit uid. */
1954void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1955
1956/* Initialise core socket variables.
1957 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1958 */
1959void sock_init_data(struct socket *sock, struct sock *sk);
1960
1961/*
1962 * Socket reference counting postulates.
1963 *
1964 * * Each user of socket SHOULD hold a reference count.
1965 * * Each access point to socket (an hash table bucket, reference from a list,
1966 * running timer, skb in flight MUST hold a reference count.
1967 * * When reference count hits 0, it means it will never increase back.
1968 * * When reference count hits 0, it means that no references from
1969 * outside exist to this socket and current process on current CPU
1970 * is last user and may/should destroy this socket.
1971 * * sk_free is called from any context: process, BH, IRQ. When
1972 * it is called, socket has no references from outside -> sk_free
1973 * may release descendant resources allocated by the socket, but
1974 * to the time when it is called, socket is NOT referenced by any
1975 * hash tables, lists etc.
1976 * * Packets, delivered from outside (from network or from another process)
1977 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1978 * when they sit in queue. Otherwise, packets will leak to hole, when
1979 * socket is looked up by one cpu and unhasing is made by another CPU.
1980 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1981 * (leak to backlog). Packet socket does all the processing inside
1982 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1983 * use separate SMP lock, so that they are prone too.
1984 */
1985
1986/* Ungrab socket and destroy it, if it was the last reference. */
1987static inline void sock_put(struct sock *sk)
1988{
1989 if (refcount_dec_and_test(r: &sk->sk_refcnt))
1990 sk_free(sk);
1991}
1992/* Generic version of sock_put(), dealing with all sockets
1993 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1994 */
1995void sock_gen_put(struct sock *sk);
1996
1997int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1998 unsigned int trim_cap, bool refcounted);
1999static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
2000 const int nested)
2001{
2002 return __sk_receive_skb(sk, skb, nested, trim_cap: 1, refcounted: true);
2003}
2004
2005static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
2006{
2007 /* sk_tx_queue_mapping accept only upto a 16-bit value */
2008 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
2009 return;
2010 /* Paired with READ_ONCE() in sk_tx_queue_get() and
2011 * other WRITE_ONCE() because socket lock might be not held.
2012 */
2013 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
2014}
2015
2016#define NO_QUEUE_MAPPING USHRT_MAX
2017
2018static inline void sk_tx_queue_clear(struct sock *sk)
2019{
2020 /* Paired with READ_ONCE() in sk_tx_queue_get() and
2021 * other WRITE_ONCE() because socket lock might be not held.
2022 */
2023 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
2024}
2025
2026static inline int sk_tx_queue_get(const struct sock *sk)
2027{
2028 if (sk) {
2029 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
2030 * and sk_tx_queue_set().
2031 */
2032 int val = READ_ONCE(sk->sk_tx_queue_mapping);
2033
2034 if (val != NO_QUEUE_MAPPING)
2035 return val;
2036 }
2037 return -1;
2038}
2039
2040static inline void __sk_rx_queue_set(struct sock *sk,
2041 const struct sk_buff *skb,
2042 bool force_set)
2043{
2044#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2045 if (skb_rx_queue_recorded(skb)) {
2046 u16 rx_queue = skb_get_rx_queue(skb);
2047
2048 if (force_set ||
2049 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2050 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2051 }
2052#endif
2053}
2054
2055static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2056{
2057 __sk_rx_queue_set(sk, skb, force_set: true);
2058}
2059
2060static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2061{
2062 __sk_rx_queue_set(sk, skb, force_set: false);
2063}
2064
2065static inline void sk_rx_queue_clear(struct sock *sk)
2066{
2067#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2068 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2069#endif
2070}
2071
2072static inline int sk_rx_queue_get(const struct sock *sk)
2073{
2074#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2075 if (sk) {
2076 int res = READ_ONCE(sk->sk_rx_queue_mapping);
2077
2078 if (res != NO_QUEUE_MAPPING)
2079 return res;
2080 }
2081#endif
2082
2083 return -1;
2084}
2085
2086static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2087{
2088 sk->sk_socket = sock;
2089}
2090
2091static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2092{
2093 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2094 return &rcu_dereference_raw(sk->sk_wq)->wait;
2095}
2096/* Detach socket from process context.
2097 * Announce socket dead, detach it from wait queue and inode.
2098 * Note that parent inode held reference count on this struct sock,
2099 * we do not release it in this function, because protocol
2100 * probably wants some additional cleanups or even continuing
2101 * to work with this socket (TCP).
2102 */
2103static inline void sock_orphan(struct sock *sk)
2104{
2105 write_lock_bh(&sk->sk_callback_lock);
2106 sock_set_flag(sk, flag: SOCK_DEAD);
2107 sk_set_socket(sk, NULL);
2108 sk->sk_wq = NULL;
2109 write_unlock_bh(&sk->sk_callback_lock);
2110}
2111
2112static inline void sock_graft(struct sock *sk, struct socket *parent)
2113{
2114 WARN_ON(parent->sk);
2115 write_lock_bh(&sk->sk_callback_lock);
2116 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2117 parent->sk = sk;
2118 sk_set_socket(sk, sock: parent);
2119 sk->sk_uid = SOCK_INODE(socket: parent)->i_uid;
2120 security_sock_graft(sk, parent);
2121 write_unlock_bh(&sk->sk_callback_lock);
2122}
2123
2124kuid_t sock_i_uid(struct sock *sk);
2125unsigned long __sock_i_ino(struct sock *sk);
2126unsigned long sock_i_ino(struct sock *sk);
2127
2128static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2129{
2130 return sk ? sk->sk_uid : make_kuid(from: net->user_ns, uid: 0);
2131}
2132
2133static inline u32 net_tx_rndhash(void)
2134{
2135 u32 v = get_random_u32();
2136
2137 return v ?: 1;
2138}
2139
2140static inline void sk_set_txhash(struct sock *sk)
2141{
2142 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2143 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2144}
2145
2146static inline bool sk_rethink_txhash(struct sock *sk)
2147{
2148 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2149 sk_set_txhash(sk);
2150 return true;
2151 }
2152 return false;
2153}
2154
2155static inline struct dst_entry *
2156__sk_dst_get(const struct sock *sk)
2157{
2158 return rcu_dereference_check(sk->sk_dst_cache,
2159 lockdep_sock_is_held(sk));
2160}
2161
2162static inline struct dst_entry *
2163sk_dst_get(const struct sock *sk)
2164{
2165 struct dst_entry *dst;
2166
2167 rcu_read_lock();
2168 dst = rcu_dereference(sk->sk_dst_cache);
2169 if (dst && !rcuref_get(ref: &dst->__rcuref))
2170 dst = NULL;
2171 rcu_read_unlock();
2172 return dst;
2173}
2174
2175static inline void __dst_negative_advice(struct sock *sk)
2176{
2177 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2178
2179 if (dst && dst->ops->negative_advice) {
2180 ndst = dst->ops->negative_advice(dst);
2181
2182 if (ndst != dst) {
2183 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2184 sk_tx_queue_clear(sk);
2185 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2186 }
2187 }
2188}
2189
2190static inline void dst_negative_advice(struct sock *sk)
2191{
2192 sk_rethink_txhash(sk);
2193 __dst_negative_advice(sk);
2194}
2195
2196static inline void
2197__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2198{
2199 struct dst_entry *old_dst;
2200
2201 sk_tx_queue_clear(sk);
2202 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2203 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2204 lockdep_sock_is_held(sk));
2205 rcu_assign_pointer(sk->sk_dst_cache, dst);
2206 dst_release(dst: old_dst);
2207}
2208
2209static inline void
2210sk_dst_set(struct sock *sk, struct dst_entry *dst)
2211{
2212 struct dst_entry *old_dst;
2213
2214 sk_tx_queue_clear(sk);
2215 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2216 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2217 dst_release(dst: old_dst);
2218}
2219
2220static inline void
2221__sk_dst_reset(struct sock *sk)
2222{
2223 __sk_dst_set(sk, NULL);
2224}
2225
2226static inline void
2227sk_dst_reset(struct sock *sk)
2228{
2229 sk_dst_set(sk, NULL);
2230}
2231
2232struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2233
2234struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2235
2236static inline void sk_dst_confirm(struct sock *sk)
2237{
2238 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2239 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2240}
2241
2242static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2243{
2244 if (skb_get_dst_pending_confirm(skb)) {
2245 struct sock *sk = skb->sk;
2246
2247 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2248 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2249 neigh_confirm(n);
2250 }
2251}
2252
2253bool sk_mc_loop(const struct sock *sk);
2254
2255static inline bool sk_can_gso(const struct sock *sk)
2256{
2257 return net_gso_ok(features: sk->sk_route_caps, gso_type: sk->sk_gso_type);
2258}
2259
2260void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2261
2262static inline void sk_gso_disable(struct sock *sk)
2263{
2264 sk->sk_gso_disabled = 1;
2265 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2266}
2267
2268static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2269 struct iov_iter *from, char *to,
2270 int copy, int offset)
2271{
2272 if (skb->ip_summed == CHECKSUM_NONE) {
2273 __wsum csum = 0;
2274 if (!csum_and_copy_from_iter_full(addr: to, bytes: copy, csum: &csum, i: from))
2275 return -EFAULT;
2276 skb->csum = csum_block_add(csum: skb->csum, csum2: csum, offset);
2277 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2278 if (!copy_from_iter_full_nocache(addr: to, bytes: copy, i: from))
2279 return -EFAULT;
2280 } else if (!copy_from_iter_full(addr: to, bytes: copy, i: from))
2281 return -EFAULT;
2282
2283 return 0;
2284}
2285
2286static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2287 struct iov_iter *from, int copy)
2288{
2289 int err, offset = skb->len;
2290
2291 err = skb_do_copy_data_nocache(sk, skb, from, to: skb_put(skb, len: copy),
2292 copy, offset);
2293 if (err)
2294 __skb_trim(skb, len: offset);
2295
2296 return err;
2297}
2298
2299static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2300 struct sk_buff *skb,
2301 struct page *page,
2302 int off, int copy)
2303{
2304 int err;
2305
2306 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2307 copy, offset: skb->len);
2308 if (err)
2309 return err;
2310
2311 skb_len_add(skb, delta: copy);
2312 sk_wmem_queued_add(sk, val: copy);
2313 sk_mem_charge(sk, size: copy);
2314 return 0;
2315}
2316
2317/**
2318 * sk_wmem_alloc_get - returns write allocations
2319 * @sk: socket
2320 *
2321 * Return: sk_wmem_alloc minus initial offset of one
2322 */
2323static inline int sk_wmem_alloc_get(const struct sock *sk)
2324{
2325 return refcount_read(r: &sk->sk_wmem_alloc) - 1;
2326}
2327
2328/**
2329 * sk_rmem_alloc_get - returns read allocations
2330 * @sk: socket
2331 *
2332 * Return: sk_rmem_alloc
2333 */
2334static inline int sk_rmem_alloc_get(const struct sock *sk)
2335{
2336 return atomic_read(v: &sk->sk_rmem_alloc);
2337}
2338
2339/**
2340 * sk_has_allocations - check if allocations are outstanding
2341 * @sk: socket
2342 *
2343 * Return: true if socket has write or read allocations
2344 */
2345static inline bool sk_has_allocations(const struct sock *sk)
2346{
2347 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2348}
2349
2350/**
2351 * skwq_has_sleeper - check if there are any waiting processes
2352 * @wq: struct socket_wq
2353 *
2354 * Return: true if socket_wq has waiting processes
2355 *
2356 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2357 * barrier call. They were added due to the race found within the tcp code.
2358 *
2359 * Consider following tcp code paths::
2360 *
2361 * CPU1 CPU2
2362 * sys_select receive packet
2363 * ... ...
2364 * __add_wait_queue update tp->rcv_nxt
2365 * ... ...
2366 * tp->rcv_nxt check sock_def_readable
2367 * ... {
2368 * schedule rcu_read_lock();
2369 * wq = rcu_dereference(sk->sk_wq);
2370 * if (wq && waitqueue_active(&wq->wait))
2371 * wake_up_interruptible(&wq->wait)
2372 * ...
2373 * }
2374 *
2375 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2376 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2377 * could then endup calling schedule and sleep forever if there are no more
2378 * data on the socket.
2379 *
2380 */
2381static inline bool skwq_has_sleeper(struct socket_wq *wq)
2382{
2383 return wq && wq_has_sleeper(wq_head: &wq->wait);
2384}
2385
2386/**
2387 * sock_poll_wait - place memory barrier behind the poll_wait call.
2388 * @filp: file
2389 * @sock: socket to wait on
2390 * @p: poll_table
2391 *
2392 * See the comments in the wq_has_sleeper function.
2393 */
2394static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2395 poll_table *p)
2396{
2397 if (!poll_does_not_wait(p)) {
2398 poll_wait(filp, wait_address: &sock->wq.wait, p);
2399 /* We need to be sure we are in sync with the
2400 * socket flags modification.
2401 *
2402 * This memory barrier is paired in the wq_has_sleeper.
2403 */
2404 smp_mb();
2405 }
2406}
2407
2408static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2409{
2410 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2411 u32 txhash = READ_ONCE(sk->sk_txhash);
2412
2413 if (txhash) {
2414 skb->l4_hash = 1;
2415 skb->hash = txhash;
2416 }
2417}
2418
2419void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2420
2421/*
2422 * Queue a received datagram if it will fit. Stream and sequenced
2423 * protocols can't normally use this as they need to fit buffers in
2424 * and play with them.
2425 *
2426 * Inlined as it's very short and called for pretty much every
2427 * packet ever received.
2428 */
2429static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2430{
2431 skb_orphan(skb);
2432 skb->sk = sk;
2433 skb->destructor = sock_rfree;
2434 atomic_add(i: skb->truesize, v: &sk->sk_rmem_alloc);
2435 sk_mem_charge(sk, size: skb->truesize);
2436}
2437
2438static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2439{
2440 if (sk && refcount_inc_not_zero(r: &sk->sk_refcnt)) {
2441 skb_orphan(skb);
2442 skb->destructor = sock_efree;
2443 skb->sk = sk;
2444 return true;
2445 }
2446 return false;
2447}
2448
2449static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2450{
2451 skb = skb_clone(skb, priority: sk_gfp_mask(sk, GFP_ATOMIC));
2452 if (skb) {
2453 if (sk_rmem_schedule(sk, skb, size: skb->truesize)) {
2454 skb_set_owner_r(skb, sk);
2455 return skb;
2456 }
2457 __kfree_skb(skb);
2458 }
2459 return NULL;
2460}
2461
2462static inline void skb_prepare_for_gro(struct sk_buff *skb)
2463{
2464 if (skb->destructor != sock_wfree) {
2465 skb_orphan(skb);
2466 return;
2467 }
2468 skb->slow_gro = 1;
2469}
2470
2471void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2472 unsigned long expires);
2473
2474void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2475
2476void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2477
2478int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2479 struct sk_buff *skb, unsigned int flags,
2480 void (*destructor)(struct sock *sk,
2481 struct sk_buff *skb));
2482int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2483
2484int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2485 enum skb_drop_reason *reason);
2486
2487static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2488{
2489 return sock_queue_rcv_skb_reason(sk, skb, NULL);
2490}
2491
2492int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2493struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2494
2495/*
2496 * Recover an error report and clear atomically
2497 */
2498
2499static inline int sock_error(struct sock *sk)
2500{
2501 int err;
2502
2503 /* Avoid an atomic operation for the common case.
2504 * This is racy since another cpu/thread can change sk_err under us.
2505 */
2506 if (likely(data_race(!sk->sk_err)))
2507 return 0;
2508
2509 err = xchg(&sk->sk_err, 0);
2510 return -err;
2511}
2512
2513void sk_error_report(struct sock *sk);
2514
2515static inline unsigned long sock_wspace(struct sock *sk)
2516{
2517 int amt = 0;
2518
2519 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2520 amt = sk->sk_sndbuf - refcount_read(r: &sk->sk_wmem_alloc);
2521 if (amt < 0)
2522 amt = 0;
2523 }
2524 return amt;
2525}
2526
2527/* Note:
2528 * We use sk->sk_wq_raw, from contexts knowing this
2529 * pointer is not NULL and cannot disappear/change.
2530 */
2531static inline void sk_set_bit(int nr, struct sock *sk)
2532{
2533 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2534 !sock_flag(sk, flag: SOCK_FASYNC))
2535 return;
2536
2537 set_bit(nr, addr: &sk->sk_wq_raw->flags);
2538}
2539
2540static inline void sk_clear_bit(int nr, struct sock *sk)
2541{
2542 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2543 !sock_flag(sk, flag: SOCK_FASYNC))
2544 return;
2545
2546 clear_bit(nr, addr: &sk->sk_wq_raw->flags);
2547}
2548
2549static inline void sk_wake_async(const struct sock *sk, int how, int band)
2550{
2551 if (sock_flag(sk, flag: SOCK_FASYNC)) {
2552 rcu_read_lock();
2553 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2554 rcu_read_unlock();
2555 }
2556}
2557
2558/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2559 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2560 * Note: for send buffers, TCP works better if we can build two skbs at
2561 * minimum.
2562 */
2563#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2564
2565#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2566#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2567
2568static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2569{
2570 u32 val;
2571
2572 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2573 return;
2574
2575 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2576 val = max_t(u32, val, sk_unused_reserved_mem(sk));
2577
2578 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2579}
2580
2581/**
2582 * sk_page_frag - return an appropriate page_frag
2583 * @sk: socket
2584 *
2585 * Use the per task page_frag instead of the per socket one for
2586 * optimization when we know that we're in process context and own
2587 * everything that's associated with %current.
2588 *
2589 * Both direct reclaim and page faults can nest inside other
2590 * socket operations and end up recursing into sk_page_frag()
2591 * while it's already in use: explicitly avoid task page_frag
2592 * when users disable sk_use_task_frag.
2593 *
2594 * Return: a per task page_frag if context allows that,
2595 * otherwise a per socket one.
2596 */
2597static inline struct page_frag *sk_page_frag(struct sock *sk)
2598{
2599 if (sk->sk_use_task_frag)
2600 return &current->task_frag;
2601
2602 return &sk->sk_frag;
2603}
2604
2605bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2606
2607/*
2608 * Default write policy as shown to user space via poll/select/SIGIO
2609 */
2610static inline bool sock_writeable(const struct sock *sk)
2611{
2612 return refcount_read(r: &sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2613}
2614
2615static inline gfp_t gfp_any(void)
2616{
2617 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2618}
2619
2620static inline gfp_t gfp_memcg_charge(void)
2621{
2622 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2623}
2624
2625static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2626{
2627 return noblock ? 0 : sk->sk_rcvtimeo;
2628}
2629
2630static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2631{
2632 return noblock ? 0 : sk->sk_sndtimeo;
2633}
2634
2635static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2636{
2637 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2638
2639 return v ?: 1;
2640}
2641
2642/* Alas, with timeout socket operations are not restartable.
2643 * Compare this to poll().
2644 */
2645static inline int sock_intr_errno(long timeo)
2646{
2647 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2648}
2649
2650struct sock_skb_cb {
2651 u32 dropcount;
2652};
2653
2654/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2655 * using skb->cb[] would keep using it directly and utilize its
2656 * alignement guarantee.
2657 */
2658#define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2659 sizeof(struct sock_skb_cb)))
2660
2661#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2662 SOCK_SKB_CB_OFFSET))
2663
2664#define sock_skb_cb_check_size(size) \
2665 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2666
2667static inline void
2668sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2669{
2670 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, flag: SOCK_RXQ_OVFL) ?
2671 atomic_read(v: &sk->sk_drops) : 0;
2672}
2673
2674static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2675{
2676 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2677
2678 atomic_add(i: segs, v: &sk->sk_drops);
2679}
2680
2681static inline ktime_t sock_read_timestamp(struct sock *sk)
2682{
2683#if BITS_PER_LONG==32
2684 unsigned int seq;
2685 ktime_t kt;
2686
2687 do {
2688 seq = read_seqbegin(&sk->sk_stamp_seq);
2689 kt = sk->sk_stamp;
2690 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2691
2692 return kt;
2693#else
2694 return READ_ONCE(sk->sk_stamp);
2695#endif
2696}
2697
2698static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2699{
2700#if BITS_PER_LONG==32
2701 write_seqlock(&sk->sk_stamp_seq);
2702 sk->sk_stamp = kt;
2703 write_sequnlock(&sk->sk_stamp_seq);
2704#else
2705 WRITE_ONCE(sk->sk_stamp, kt);
2706#endif
2707}
2708
2709void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2710 struct sk_buff *skb);
2711void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2712 struct sk_buff *skb);
2713
2714static inline void
2715sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2716{
2717 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2718 u32 tsflags = READ_ONCE(sk->sk_tsflags);
2719 ktime_t kt = skb->tstamp;
2720 /*
2721 * generate control messages if
2722 * - receive time stamping in software requested
2723 * - software time stamp available and wanted
2724 * - hardware time stamps available and wanted
2725 */
2726 if (sock_flag(sk, flag: SOCK_RCVTSTAMP) ||
2727 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2728 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2729 (hwtstamps->hwtstamp &&
2730 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2731 __sock_recv_timestamp(msg, sk, skb);
2732 else
2733 sock_write_timestamp(sk, kt);
2734
2735 if (sock_flag(sk, flag: SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2736 __sock_recv_wifi_status(msg, sk, skb);
2737}
2738
2739void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2740 struct sk_buff *skb);
2741
2742#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2743static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2744 struct sk_buff *skb)
2745{
2746#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \
2747 (1UL << SOCK_RCVTSTAMP) | \
2748 (1UL << SOCK_RCVMARK))
2749#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2750 SOF_TIMESTAMPING_RAW_HARDWARE)
2751
2752 if (sk->sk_flags & FLAGS_RECV_CMSGS ||
2753 READ_ONCE(sk->sk_tsflags) & TSFLAGS_ANY)
2754 __sock_recv_cmsgs(msg, sk, skb);
2755 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2756 sock_write_timestamp(sk, kt: skb->tstamp);
2757 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2758 sock_write_timestamp(sk, kt: 0);
2759}
2760
2761void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2762
2763/**
2764 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2765 * @sk: socket sending this packet
2766 * @tsflags: timestamping flags to use
2767 * @tx_flags: completed with instructions for time stamping
2768 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2769 *
2770 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2771 */
2772static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2773 __u8 *tx_flags, __u32 *tskey)
2774{
2775 if (unlikely(tsflags)) {
2776 __sock_tx_timestamp(tsflags, tx_flags);
2777 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2778 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2779 *tskey = atomic_inc_return(v: &sk->sk_tskey) - 1;
2780 }
2781 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2782 *tx_flags |= SKBTX_WIFI_STATUS;
2783}
2784
2785static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2786 __u8 *tx_flags)
2787{
2788 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2789}
2790
2791static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2792{
2793 _sock_tx_timestamp(sk: skb->sk, tsflags, tx_flags: &skb_shinfo(skb)->tx_flags,
2794 tskey: &skb_shinfo(skb)->tskey);
2795}
2796
2797static inline bool sk_is_tcp(const struct sock *sk)
2798{
2799 return sk->sk_type == SOCK_STREAM && sk->sk_protocol == IPPROTO_TCP;
2800}
2801
2802/**
2803 * sk_eat_skb - Release a skb if it is no longer needed
2804 * @sk: socket to eat this skb from
2805 * @skb: socket buffer to eat
2806 *
2807 * This routine must be called with interrupts disabled or with the socket
2808 * locked so that the sk_buff queue operation is ok.
2809*/
2810static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2811{
2812 __skb_unlink(skb, list: &sk->sk_receive_queue);
2813 __kfree_skb(skb);
2814}
2815
2816static inline bool
2817skb_sk_is_prefetched(struct sk_buff *skb)
2818{
2819#ifdef CONFIG_INET
2820 return skb->destructor == sock_pfree;
2821#else
2822 return false;
2823#endif /* CONFIG_INET */
2824}
2825
2826/* This helper checks if a socket is a full socket,
2827 * ie _not_ a timewait or request socket.
2828 */
2829static inline bool sk_fullsock(const struct sock *sk)
2830{
2831 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2832}
2833
2834static inline bool
2835sk_is_refcounted(struct sock *sk)
2836{
2837 /* Only full sockets have sk->sk_flags. */
2838 return !sk_fullsock(sk) || !sock_flag(sk, flag: SOCK_RCU_FREE);
2839}
2840
2841/**
2842 * skb_steal_sock - steal a socket from an sk_buff
2843 * @skb: sk_buff to steal the socket from
2844 * @refcounted: is set to true if the socket is reference-counted
2845 * @prefetched: is set to true if the socket was assigned from bpf
2846 */
2847static inline struct sock *
2848skb_steal_sock(struct sk_buff *skb, bool *refcounted, bool *prefetched)
2849{
2850 if (skb->sk) {
2851 struct sock *sk = skb->sk;
2852
2853 *refcounted = true;
2854 *prefetched = skb_sk_is_prefetched(skb);
2855 if (*prefetched)
2856 *refcounted = sk_is_refcounted(sk);
2857 skb->destructor = NULL;
2858 skb->sk = NULL;
2859 return sk;
2860 }
2861 *prefetched = false;
2862 *refcounted = false;
2863 return NULL;
2864}
2865
2866/* Checks if this SKB belongs to an HW offloaded socket
2867 * and whether any SW fallbacks are required based on dev.
2868 * Check decrypted mark in case skb_orphan() cleared socket.
2869 */
2870static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2871 struct net_device *dev)
2872{
2873#ifdef CONFIG_SOCK_VALIDATE_XMIT
2874 struct sock *sk = skb->sk;
2875
2876 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2877 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2878#ifdef CONFIG_TLS_DEVICE
2879 } else if (unlikely(skb->decrypted)) {
2880 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2881 kfree_skb(skb);
2882 skb = NULL;
2883#endif
2884 }
2885#endif
2886
2887 return skb;
2888}
2889
2890/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2891 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2892 */
2893static inline bool sk_listener(const struct sock *sk)
2894{
2895 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2896}
2897
2898void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2899int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2900 int type);
2901
2902bool sk_ns_capable(const struct sock *sk,
2903 struct user_namespace *user_ns, int cap);
2904bool sk_capable(const struct sock *sk, int cap);
2905bool sk_net_capable(const struct sock *sk, int cap);
2906
2907void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2908
2909/* Take into consideration the size of the struct sk_buff overhead in the
2910 * determination of these values, since that is non-constant across
2911 * platforms. This makes socket queueing behavior and performance
2912 * not depend upon such differences.
2913 */
2914#define _SK_MEM_PACKETS 256
2915#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2916#define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2917#define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2918
2919extern __u32 sysctl_wmem_max;
2920extern __u32 sysctl_rmem_max;
2921
2922extern int sysctl_tstamp_allow_data;
2923extern int sysctl_optmem_max;
2924
2925extern __u32 sysctl_wmem_default;
2926extern __u32 sysctl_rmem_default;
2927
2928#define SKB_FRAG_PAGE_ORDER get_order(32768)
2929DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2930
2931static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2932{
2933 /* Does this proto have per netns sysctl_wmem ? */
2934 if (proto->sysctl_wmem_offset)
2935 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2936
2937 return READ_ONCE(*proto->sysctl_wmem);
2938}
2939
2940static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2941{
2942 /* Does this proto have per netns sysctl_rmem ? */
2943 if (proto->sysctl_rmem_offset)
2944 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2945
2946 return READ_ONCE(*proto->sysctl_rmem);
2947}
2948
2949/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2950 * Some wifi drivers need to tweak it to get more chunks.
2951 * They can use this helper from their ndo_start_xmit()
2952 */
2953static inline void sk_pacing_shift_update(struct sock *sk, int val)
2954{
2955 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2956 return;
2957 WRITE_ONCE(sk->sk_pacing_shift, val);
2958}
2959
2960/* if a socket is bound to a device, check that the given device
2961 * index is either the same or that the socket is bound to an L3
2962 * master device and the given device index is also enslaved to
2963 * that L3 master
2964 */
2965static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2966{
2967 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
2968 int mdif;
2969
2970 if (!bound_dev_if || bound_dev_if == dif)
2971 return true;
2972
2973 mdif = l3mdev_master_ifindex_by_index(net: sock_net(sk), ifindex: dif);
2974 if (mdif && mdif == bound_dev_if)
2975 return true;
2976
2977 return false;
2978}
2979
2980void sock_def_readable(struct sock *sk);
2981
2982int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2983void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2984int sock_set_timestamping(struct sock *sk, int optname,
2985 struct so_timestamping timestamping);
2986
2987void sock_enable_timestamps(struct sock *sk);
2988void sock_no_linger(struct sock *sk);
2989void sock_set_keepalive(struct sock *sk);
2990void sock_set_priority(struct sock *sk, u32 priority);
2991void sock_set_rcvbuf(struct sock *sk, int val);
2992void sock_set_mark(struct sock *sk, u32 val);
2993void sock_set_reuseaddr(struct sock *sk);
2994void sock_set_reuseport(struct sock *sk);
2995void sock_set_sndtimeo(struct sock *sk, s64 secs);
2996
2997int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2998
2999int sock_get_timeout(long timeo, void *optval, bool old_timeval);
3000int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
3001 sockptr_t optval, int optlen, bool old_timeval);
3002
3003int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
3004 void __user *arg, void *karg, size_t size);
3005int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
3006static inline bool sk_is_readable(struct sock *sk)
3007{
3008 if (sk->sk_prot->sock_is_readable)
3009 return sk->sk_prot->sock_is_readable(sk);
3010 return false;
3011}
3012#endif /* _SOCK_H */
3013

source code of linux/include/net/sock.h