1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the AF_INET socket handler.
8 *
9 * Version: @(#)sock.h 1.0.4 05/13/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche <flla@stud.uni-sb.de>
15 *
16 * Fixes:
17 * Alan Cox : Volatiles in skbuff pointers. See
18 * skbuff comments. May be overdone,
19 * better to prove they can be removed
20 * than the reverse.
21 * Alan Cox : Added a zapped field for tcp to note
22 * a socket is reset and must stay shut up
23 * Alan Cox : New fields for options
24 * Pauline Middelink : identd support
25 * Alan Cox : Eliminate low level recv/recvfrom
26 * David S. Miller : New socket lookup architecture.
27 * Steve Whitehouse: Default routines for sock_ops
28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
29 * protinfo be just a void pointer, as the
30 * protocol specific parts were moved to
31 * respective headers and ipv4/v6, etc now
32 * use private slabcaches for its socks
33 * Pedro Hortas : New flags field for socket options
34 */
35#ifndef _SOCK_H
36#define _SOCK_H
37
38#include <linux/hardirq.h>
39#include <linux/kernel.h>
40#include <linux/list.h>
41#include <linux/list_nulls.h>
42#include <linux/timer.h>
43#include <linux/cache.h>
44#include <linux/bitops.h>
45#include <linux/lockdep.h>
46#include <linux/netdevice.h>
47#include <linux/skbuff.h> /* struct sk_buff */
48#include <linux/mm.h>
49#include <linux/security.h>
50#include <linux/slab.h>
51#include <linux/uaccess.h>
52#include <linux/page_counter.h>
53#include <linux/memcontrol.h>
54#include <linux/static_key.h>
55#include <linux/sched.h>
56#include <linux/wait.h>
57#include <linux/cgroup-defs.h>
58#include <linux/rbtree.h>
59#include <linux/rculist_nulls.h>
60#include <linux/poll.h>
61#include <linux/sockptr.h>
62#include <linux/indirect_call_wrapper.h>
63#include <linux/atomic.h>
64#include <linux/refcount.h>
65#include <linux/llist.h>
66#include <net/dst.h>
67#include <net/checksum.h>
68#include <net/tcp_states.h>
69#include <linux/net_tstamp.h>
70#include <net/l3mdev.h>
71#include <uapi/linux/socket.h>
72
73/*
74 * This structure really needs to be cleaned up.
75 * Most of it is for TCP, and not used by any of
76 * the other protocols.
77 */
78
79/* This is the per-socket lock. The spinlock provides a synchronization
80 * between user contexts and software interrupt processing, whereas the
81 * mini-semaphore synchronizes multiple users amongst themselves.
82 */
83typedef struct {
84 spinlock_t slock;
85 int owned;
86 wait_queue_head_t wq;
87 /*
88 * We express the mutex-alike socket_lock semantics
89 * to the lock validator by explicitly managing
90 * the slock as a lock variant (in addition to
91 * the slock itself):
92 */
93#ifdef CONFIG_DEBUG_LOCK_ALLOC
94 struct lockdep_map dep_map;
95#endif
96} socket_lock_t;
97
98struct sock;
99struct proto;
100struct net;
101
102typedef __u32 __bitwise __portpair;
103typedef __u64 __bitwise __addrpair;
104
105/**
106 * struct sock_common - minimal network layer representation of sockets
107 * @skc_daddr: Foreign IPv4 addr
108 * @skc_rcv_saddr: Bound local IPv4 addr
109 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
110 * @skc_hash: hash value used with various protocol lookup tables
111 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
112 * @skc_dport: placeholder for inet_dport/tw_dport
113 * @skc_num: placeholder for inet_num/tw_num
114 * @skc_portpair: __u32 union of @skc_dport & @skc_num
115 * @skc_family: network address family
116 * @skc_state: Connection state
117 * @skc_reuse: %SO_REUSEADDR setting
118 * @skc_reuseport: %SO_REUSEPORT setting
119 * @skc_ipv6only: socket is IPV6 only
120 * @skc_net_refcnt: socket is using net ref counting
121 * @skc_bound_dev_if: bound device index if != 0
122 * @skc_bind_node: bind hash linkage for various protocol lookup tables
123 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
124 * @skc_prot: protocol handlers inside a network family
125 * @skc_net: reference to the network namespace of this socket
126 * @skc_v6_daddr: IPV6 destination address
127 * @skc_v6_rcv_saddr: IPV6 source address
128 * @skc_cookie: socket's cookie value
129 * @skc_node: main hash linkage for various protocol lookup tables
130 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
131 * @skc_tx_queue_mapping: tx queue number for this connection
132 * @skc_rx_queue_mapping: rx queue number for this connection
133 * @skc_flags: place holder for sk_flags
134 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
135 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
136 * @skc_listener: connection request listener socket (aka rsk_listener)
137 * [union with @skc_flags]
138 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
139 * [union with @skc_flags]
140 * @skc_incoming_cpu: record/match cpu processing incoming packets
141 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
142 * [union with @skc_incoming_cpu]
143 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
144 * [union with @skc_incoming_cpu]
145 * @skc_refcnt: reference count
146 *
147 * This is the minimal network layer representation of sockets, the header
148 * for struct sock and struct inet_timewait_sock.
149 */
150struct sock_common {
151 union {
152 __addrpair skc_addrpair;
153 struct {
154 __be32 skc_daddr;
155 __be32 skc_rcv_saddr;
156 };
157 };
158 union {
159 unsigned int skc_hash;
160 __u16 skc_u16hashes[2];
161 };
162 /* skc_dport && skc_num must be grouped as well */
163 union {
164 __portpair skc_portpair;
165 struct {
166 __be16 skc_dport;
167 __u16 skc_num;
168 };
169 };
170
171 unsigned short skc_family;
172 volatile unsigned char skc_state;
173 unsigned char skc_reuse:4;
174 unsigned char skc_reuseport:1;
175 unsigned char skc_ipv6only:1;
176 unsigned char skc_net_refcnt:1;
177 int skc_bound_dev_if;
178 union {
179 struct hlist_node skc_bind_node;
180 struct hlist_node skc_portaddr_node;
181 };
182 struct proto *skc_prot;
183 possible_net_t skc_net;
184
185#if IS_ENABLED(CONFIG_IPV6)
186 struct in6_addr skc_v6_daddr;
187 struct in6_addr skc_v6_rcv_saddr;
188#endif
189
190 atomic64_t skc_cookie;
191
192 /* following fields are padding to force
193 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
194 * assuming IPV6 is enabled. We use this padding differently
195 * for different kind of 'sockets'
196 */
197 union {
198 unsigned long skc_flags;
199 struct sock *skc_listener; /* request_sock */
200 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
201 };
202 /*
203 * fields between dontcopy_begin/dontcopy_end
204 * are not copied in sock_copy()
205 */
206 /* private: */
207 int skc_dontcopy_begin[0];
208 /* public: */
209 union {
210 struct hlist_node skc_node;
211 struct hlist_nulls_node skc_nulls_node;
212 };
213 unsigned short skc_tx_queue_mapping;
214#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
215 unsigned short skc_rx_queue_mapping;
216#endif
217 union {
218 int skc_incoming_cpu;
219 u32 skc_rcv_wnd;
220 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
221 };
222
223 refcount_t skc_refcnt;
224 /* private: */
225 int skc_dontcopy_end[0];
226 union {
227 u32 skc_rxhash;
228 u32 skc_window_clamp;
229 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
230 };
231 /* public: */
232};
233
234struct bpf_local_storage;
235struct sk_filter;
236
237/**
238 * struct sock - network layer representation of sockets
239 * @__sk_common: shared layout with inet_timewait_sock
240 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
241 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
242 * @sk_lock: synchronizer
243 * @sk_kern_sock: True if sock is using kernel lock classes
244 * @sk_rcvbuf: size of receive buffer in bytes
245 * @sk_wq: sock wait queue and async head
246 * @sk_rx_dst: receive input route used by early demux
247 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
248 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
249 * @sk_dst_cache: destination cache
250 * @sk_dst_pending_confirm: need to confirm neighbour
251 * @sk_policy: flow policy
252 * @psp_assoc: PSP association, if socket is PSP-secured
253 * @sk_receive_queue: incoming packets
254 * @sk_wmem_alloc: transmit queue bytes committed
255 * @sk_tsq_flags: TCP Small Queues flags
256 * @sk_write_queue: Packet sending queue
257 * @sk_omem_alloc: "o" is "option" or "other"
258 * @sk_wmem_queued: persistent queue size
259 * @sk_forward_alloc: space allocated forward
260 * @sk_reserved_mem: space reserved and non-reclaimable for the socket
261 * @sk_napi_id: id of the last napi context to receive data for sk
262 * @sk_ll_usec: usecs to busypoll when there is no data
263 * @sk_allocation: allocation mode
264 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
265 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
266 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
267 * @sk_sndbuf: size of send buffer in bytes
268 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
269 * @sk_no_check_rx: allow zero checksum in RX packets
270 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
271 * @sk_gso_disabled: if set, NETIF_F_GSO_MASK is forbidden.
272 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
273 * @sk_gso_max_size: Maximum GSO segment size to build
274 * @sk_gso_max_segs: Maximum number of GSO segments
275 * @sk_pacing_shift: scaling factor for TCP Small Queues
276 * @sk_lingertime: %SO_LINGER l_linger setting
277 * @sk_backlog: always used with the per-socket spinlock held
278 * @sk_callback_lock: used with the callbacks in the end of this struct
279 * @sk_error_queue: rarely used
280 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
281 * IPV6_ADDRFORM for instance)
282 * @sk_err: last error
283 * @sk_err_soft: errors that don't cause failure but are the cause of a
284 * persistent failure not just 'timed out'
285 * @sk_drops: raw/udp drops counter
286 * @sk_drop_counters: optional pointer to numa_drop_counters
287 * @sk_ack_backlog: current listen backlog
288 * @sk_max_ack_backlog: listen backlog set in listen()
289 * @sk_uid: user id of owner
290 * @sk_ino: inode number (zero if orphaned)
291 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
292 * @sk_busy_poll_budget: napi processing budget when busypolling
293 * @sk_priority: %SO_PRIORITY setting
294 * @sk_type: socket type (%SOCK_STREAM, etc)
295 * @sk_protocol: which protocol this socket belongs in this network family
296 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
297 * @sk_peer_pid: &struct pid for this socket's peer
298 * @sk_peer_cred: %SO_PEERCRED setting
299 * @sk_rcvlowat: %SO_RCVLOWAT setting
300 * @sk_rcvtimeo: %SO_RCVTIMEO setting
301 * @sk_sndtimeo: %SO_SNDTIMEO setting
302 * @sk_txhash: computed flow hash for use on transmit
303 * @sk_txrehash: enable TX hash rethink
304 * @sk_filter: socket filtering instructions
305 * @sk_timer: sock cleanup timer
306 * @sk_stamp: time stamp of last packet received
307 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
308 * @sk_tsflags: SO_TIMESTAMPING flags
309 * @sk_bpf_cb_flags: used in bpf_setsockopt()
310 * @sk_use_task_frag: allow sk_page_frag() to use current->task_frag.
311 * Sockets that can be used under memory reclaim should
312 * set this to false.
313 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
314 * for timestamping
315 * @sk_tskey: counter to disambiguate concurrent tstamp requests
316 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
317 * @sk_socket: Identd and reporting IO signals
318 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
319 * @sk_frag: cached page frag
320 * @sk_peek_off: current peek_offset value
321 * @sk_send_head: front of stuff to transmit
322 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
323 * @sk_security: used by security modules
324 * @sk_mark: generic packet mark
325 * @sk_cgrp_data: cgroup data for this cgroup
326 * @sk_memcg: this socket's memory cgroup association
327 * @sk_write_pending: a write to stream socket waits to start
328 * @sk_disconnects: number of disconnect operations performed on this sock
329 * @sk_state_change: callback to indicate change in the state of the sock
330 * @sk_data_ready: callback to indicate there is data to be processed
331 * @sk_write_space: callback to indicate there is bf sending space available
332 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
333 * @sk_backlog_rcv: callback to process the backlog
334 * @sk_validate_xmit_skb: ptr to an optional validate function
335 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
336 * @sk_reuseport_cb: reuseport group container
337 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
338 * @sk_rcu: used during RCU grace period
339 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
340 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
341 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
342 * @sk_txtime_unused: unused txtime flags
343 * @sk_scm_recv_flags: all flags used by scm_recv()
344 * @sk_scm_credentials: flagged by SO_PASSCRED to recv SCM_CREDENTIALS
345 * @sk_scm_security: flagged by SO_PASSSEC to recv SCM_SECURITY
346 * @sk_scm_pidfd: flagged by SO_PASSPIDFD to recv SCM_PIDFD
347 * @sk_scm_rights: flagged by SO_PASSRIGHTS to recv SCM_RIGHTS
348 * @sk_scm_unused: unused flags for scm_recv()
349 * @ns_tracker: tracker for netns reference
350 * @sk_user_frags: xarray of pages the user is holding a reference on.
351 * @sk_owner: reference to the real owner of the socket that calls
352 * sock_lock_init_class_and_name().
353 */
354struct sock {
355 /*
356 * Now struct inet_timewait_sock also uses sock_common, so please just
357 * don't add nothing before this first member (__sk_common) --acme
358 */
359 struct sock_common __sk_common;
360#define sk_node __sk_common.skc_node
361#define sk_nulls_node __sk_common.skc_nulls_node
362#define sk_refcnt __sk_common.skc_refcnt
363#define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
364#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
365#define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
366#endif
367
368#define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
369#define sk_dontcopy_end __sk_common.skc_dontcopy_end
370#define sk_hash __sk_common.skc_hash
371#define sk_portpair __sk_common.skc_portpair
372#define sk_num __sk_common.skc_num
373#define sk_dport __sk_common.skc_dport
374#define sk_addrpair __sk_common.skc_addrpair
375#define sk_daddr __sk_common.skc_daddr
376#define sk_rcv_saddr __sk_common.skc_rcv_saddr
377#define sk_family __sk_common.skc_family
378#define sk_state __sk_common.skc_state
379#define sk_reuse __sk_common.skc_reuse
380#define sk_reuseport __sk_common.skc_reuseport
381#define sk_ipv6only __sk_common.skc_ipv6only
382#define sk_net_refcnt __sk_common.skc_net_refcnt
383#define sk_bound_dev_if __sk_common.skc_bound_dev_if
384#define sk_bind_node __sk_common.skc_bind_node
385#define sk_prot __sk_common.skc_prot
386#define sk_net __sk_common.skc_net
387#define sk_v6_daddr __sk_common.skc_v6_daddr
388#define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
389#define sk_cookie __sk_common.skc_cookie
390#define sk_incoming_cpu __sk_common.skc_incoming_cpu
391#define sk_flags __sk_common.skc_flags
392#define sk_rxhash __sk_common.skc_rxhash
393
394 __cacheline_group_begin(sock_write_rx);
395
396 atomic_t sk_drops;
397 __s32 sk_peek_off;
398 struct sk_buff_head sk_error_queue;
399 struct sk_buff_head sk_receive_queue;
400 /*
401 * The backlog queue is special, it is always used with
402 * the per-socket spinlock held and requires low latency
403 * access. Therefore we special case it's implementation.
404 * Note : rmem_alloc is in this structure to fill a hole
405 * on 64bit arches, not because its logically part of
406 * backlog.
407 */
408 struct {
409 atomic_t rmem_alloc;
410 int len;
411 struct sk_buff *head;
412 struct sk_buff *tail;
413 } sk_backlog;
414#define sk_rmem_alloc sk_backlog.rmem_alloc
415
416 __cacheline_group_end(sock_write_rx);
417
418 __cacheline_group_begin(sock_read_rx);
419 /* early demux fields */
420 struct dst_entry __rcu *sk_rx_dst;
421 int sk_rx_dst_ifindex;
422 u32 sk_rx_dst_cookie;
423
424#ifdef CONFIG_NET_RX_BUSY_POLL
425 unsigned int sk_ll_usec;
426 unsigned int sk_napi_id;
427 u16 sk_busy_poll_budget;
428 u8 sk_prefer_busy_poll;
429#endif
430 u8 sk_userlocks;
431 int sk_rcvbuf;
432
433 struct sk_filter __rcu *sk_filter;
434 union {
435 struct socket_wq __rcu *sk_wq;
436 /* private: */
437 struct socket_wq *sk_wq_raw;
438 /* public: */
439 };
440
441 void (*sk_data_ready)(struct sock *sk);
442 long sk_rcvtimeo;
443 int sk_rcvlowat;
444 __cacheline_group_end(sock_read_rx);
445
446 __cacheline_group_begin(sock_read_rxtx);
447 int sk_err;
448 struct socket *sk_socket;
449#ifdef CONFIG_MEMCG
450 struct mem_cgroup *sk_memcg;
451#endif
452#ifdef CONFIG_XFRM
453 struct xfrm_policy __rcu *sk_policy[2];
454#endif
455#if IS_ENABLED(CONFIG_INET_PSP)
456 struct psp_assoc __rcu *psp_assoc;
457#endif
458 __cacheline_group_end(sock_read_rxtx);
459
460 __cacheline_group_begin(sock_write_rxtx);
461 socket_lock_t sk_lock;
462 u32 sk_reserved_mem;
463 int sk_forward_alloc;
464 u32 sk_tsflags;
465 __cacheline_group_end(sock_write_rxtx);
466
467 __cacheline_group_begin(sock_write_tx);
468 int sk_write_pending;
469 atomic_t sk_omem_alloc;
470 int sk_err_soft;
471
472 int sk_wmem_queued;
473 refcount_t sk_wmem_alloc;
474 unsigned long sk_tsq_flags;
475 union {
476 struct sk_buff *sk_send_head;
477 struct rb_root tcp_rtx_queue;
478 };
479 struct sk_buff_head sk_write_queue;
480 u32 sk_dst_pending_confirm;
481 u32 sk_pacing_status; /* see enum sk_pacing */
482 struct page_frag sk_frag;
483 struct timer_list sk_timer;
484
485 unsigned long sk_pacing_rate; /* bytes per second */
486 atomic_t sk_zckey;
487 atomic_t sk_tskey;
488 __cacheline_group_end(sock_write_tx);
489
490 __cacheline_group_begin(sock_read_tx);
491 unsigned long sk_max_pacing_rate;
492 long sk_sndtimeo;
493 u32 sk_priority;
494 u32 sk_mark;
495 kuid_t sk_uid;
496 u16 sk_protocol;
497 u16 sk_type;
498 struct dst_entry __rcu *sk_dst_cache;
499 netdev_features_t sk_route_caps;
500#ifdef CONFIG_SOCK_VALIDATE_XMIT
501 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
502 struct net_device *dev,
503 struct sk_buff *skb);
504#endif
505 u16 sk_gso_type;
506 u16 sk_gso_max_segs;
507 unsigned int sk_gso_max_size;
508 gfp_t sk_allocation;
509 u32 sk_txhash;
510 int sk_sndbuf;
511 u8 sk_pacing_shift;
512 bool sk_use_task_frag;
513 __cacheline_group_end(sock_read_tx);
514
515 /*
516 * Because of non atomicity rules, all
517 * changes are protected by socket lock.
518 */
519 u8 sk_gso_disabled : 1,
520 sk_kern_sock : 1,
521 sk_no_check_tx : 1,
522 sk_no_check_rx : 1;
523 u8 sk_shutdown;
524 unsigned long sk_lingertime;
525 struct proto *sk_prot_creator;
526 rwlock_t sk_callback_lock;
527 u32 sk_ack_backlog;
528 u32 sk_max_ack_backlog;
529 unsigned long sk_ino;
530 spinlock_t sk_peer_lock;
531 int sk_bind_phc;
532 struct pid *sk_peer_pid;
533 const struct cred *sk_peer_cred;
534
535 ktime_t sk_stamp;
536#if BITS_PER_LONG==32
537 seqlock_t sk_stamp_seq;
538#endif
539 int sk_disconnects;
540
541 union {
542 u8 sk_txrehash;
543 u8 sk_scm_recv_flags;
544 struct {
545 u8 sk_scm_credentials : 1,
546 sk_scm_security : 1,
547 sk_scm_pidfd : 1,
548 sk_scm_rights : 1,
549 sk_scm_unused : 4;
550 };
551 };
552 u8 sk_clockid;
553 u8 sk_txtime_deadline_mode : 1,
554 sk_txtime_report_errors : 1,
555 sk_txtime_unused : 6;
556#define SK_BPF_CB_FLAG_TEST(SK, FLAG) ((SK)->sk_bpf_cb_flags & (FLAG))
557 u8 sk_bpf_cb_flags;
558
559 void *sk_user_data;
560#ifdef CONFIG_SECURITY
561 void *sk_security;
562#endif
563 struct sock_cgroup_data sk_cgrp_data;
564 void (*sk_state_change)(struct sock *sk);
565 void (*sk_write_space)(struct sock *sk);
566 void (*sk_error_report)(struct sock *sk);
567 int (*sk_backlog_rcv)(struct sock *sk,
568 struct sk_buff *skb);
569 void (*sk_destruct)(struct sock *sk);
570 struct sock_reuseport __rcu *sk_reuseport_cb;
571#ifdef CONFIG_BPF_SYSCALL
572 struct bpf_local_storage __rcu *sk_bpf_storage;
573#endif
574 struct numa_drop_counters *sk_drop_counters;
575 struct rcu_head sk_rcu;
576 netns_tracker ns_tracker;
577 struct xarray sk_user_frags;
578
579#if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
580 struct module *sk_owner;
581#endif
582};
583
584struct sock_bh_locked {
585 struct sock *sock;
586 local_lock_t bh_lock;
587};
588
589enum sk_pacing {
590 SK_PACING_NONE = 0,
591 SK_PACING_NEEDED = 1,
592 SK_PACING_FQ = 2,
593};
594
595/* flag bits in sk_user_data
596 *
597 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
598 * not be suitable for copying when cloning the socket. For instance,
599 * it can point to a reference counted object. sk_user_data bottom
600 * bit is set if pointer must not be copied.
601 *
602 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
603 * managed/owned by a BPF reuseport array. This bit should be set
604 * when sk_user_data's sk is added to the bpf's reuseport_array.
605 *
606 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
607 * sk_user_data points to psock type. This bit should be set
608 * when sk_user_data is assigned to a psock object.
609 */
610#define SK_USER_DATA_NOCOPY 1UL
611#define SK_USER_DATA_BPF 2UL
612#define SK_USER_DATA_PSOCK 4UL
613#define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
614 SK_USER_DATA_PSOCK)
615
616/**
617 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
618 * @sk: socket
619 */
620static inline bool sk_user_data_is_nocopy(const struct sock *sk)
621{
622 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
623}
624
625#define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
626
627/**
628 * __locked_read_sk_user_data_with_flags - return the pointer
629 * only if argument flags all has been set in sk_user_data. Otherwise
630 * return NULL
631 *
632 * @sk: socket
633 * @flags: flag bits
634 *
635 * The caller must be holding sk->sk_callback_lock.
636 */
637static inline void *
638__locked_read_sk_user_data_with_flags(const struct sock *sk,
639 uintptr_t flags)
640{
641 uintptr_t sk_user_data =
642 (uintptr_t)rcu_dereference_check(__sk_user_data(sk),
643 lockdep_is_held(&sk->sk_callback_lock));
644
645 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
646
647 if ((sk_user_data & flags) == flags)
648 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
649 return NULL;
650}
651
652/**
653 * __rcu_dereference_sk_user_data_with_flags - return the pointer
654 * only if argument flags all has been set in sk_user_data. Otherwise
655 * return NULL
656 *
657 * @sk: socket
658 * @flags: flag bits
659 */
660static inline void *
661__rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
662 uintptr_t flags)
663{
664 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
665
666 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
667
668 if ((sk_user_data & flags) == flags)
669 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
670 return NULL;
671}
672
673#define rcu_dereference_sk_user_data(sk) \
674 __rcu_dereference_sk_user_data_with_flags(sk, 0)
675#define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
676({ \
677 uintptr_t __tmp1 = (uintptr_t)(ptr), \
678 __tmp2 = (uintptr_t)(flags); \
679 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
680 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
681 rcu_assign_pointer(__sk_user_data((sk)), \
682 __tmp1 | __tmp2); \
683})
684#define rcu_assign_sk_user_data(sk, ptr) \
685 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
686
687static inline
688struct net *sock_net(const struct sock *sk)
689{
690 return read_pnet(pnet: &sk->sk_net);
691}
692
693static inline
694void sock_net_set(struct sock *sk, struct net *net)
695{
696 write_pnet(pnet: &sk->sk_net, net);
697}
698
699/*
700 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
701 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
702 * on a socket means that the socket will reuse everybody else's port
703 * without looking at the other's sk_reuse value.
704 */
705
706#define SK_NO_REUSE 0
707#define SK_CAN_REUSE 1
708#define SK_FORCE_REUSE 2
709
710int sk_set_peek_off(struct sock *sk, int val);
711
712static inline int sk_peek_offset(const struct sock *sk, int flags)
713{
714 if (unlikely(flags & MSG_PEEK)) {
715 return READ_ONCE(sk->sk_peek_off);
716 }
717
718 return 0;
719}
720
721static inline void sk_peek_offset_bwd(struct sock *sk, int val)
722{
723 s32 off = READ_ONCE(sk->sk_peek_off);
724
725 if (unlikely(off >= 0)) {
726 off = max_t(s32, off - val, 0);
727 WRITE_ONCE(sk->sk_peek_off, off);
728 }
729}
730
731static inline void sk_peek_offset_fwd(struct sock *sk, int val)
732{
733 sk_peek_offset_bwd(sk, val: -val);
734}
735
736/*
737 * Hashed lists helper routines
738 */
739static inline struct sock *sk_entry(const struct hlist_node *node)
740{
741 return hlist_entry(node, struct sock, sk_node);
742}
743
744static inline struct sock *__sk_head(const struct hlist_head *head)
745{
746 return hlist_entry(head->first, struct sock, sk_node);
747}
748
749static inline struct sock *sk_head(const struct hlist_head *head)
750{
751 return hlist_empty(h: head) ? NULL : __sk_head(head);
752}
753
754static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
755{
756 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
757}
758
759static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
760{
761 return hlist_nulls_empty(h: head) ? NULL : __sk_nulls_head(head);
762}
763
764static inline struct sock *sk_next(const struct sock *sk)
765{
766 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
767}
768
769static inline struct sock *sk_nulls_next(const struct sock *sk)
770{
771 return (!is_a_nulls(ptr: sk->sk_nulls_node.next)) ?
772 hlist_nulls_entry(sk->sk_nulls_node.next,
773 struct sock, sk_nulls_node) :
774 NULL;
775}
776
777static inline bool sk_unhashed(const struct sock *sk)
778{
779 return hlist_unhashed(h: &sk->sk_node);
780}
781
782static inline bool sk_hashed(const struct sock *sk)
783{
784 return !sk_unhashed(sk);
785}
786
787static inline void sk_node_init(struct hlist_node *node)
788{
789 node->pprev = NULL;
790}
791
792static inline void __sk_del_node(struct sock *sk)
793{
794 __hlist_del(n: &sk->sk_node);
795}
796
797/* NB: equivalent to hlist_del_init_rcu */
798static inline bool __sk_del_node_init(struct sock *sk)
799{
800 if (sk_hashed(sk)) {
801 __sk_del_node(sk);
802 sk_node_init(node: &sk->sk_node);
803 return true;
804 }
805 return false;
806}
807
808/* Grab socket reference count. This operation is valid only
809 when sk is ALREADY grabbed f.e. it is found in hash table
810 or a list and the lookup is made under lock preventing hash table
811 modifications.
812 */
813
814static __always_inline void sock_hold(struct sock *sk)
815{
816 refcount_inc(r: &sk->sk_refcnt);
817}
818
819/* Ungrab socket in the context, which assumes that socket refcnt
820 cannot hit zero, f.e. it is true in context of any socketcall.
821 */
822static __always_inline void __sock_put(struct sock *sk)
823{
824 refcount_dec(r: &sk->sk_refcnt);
825}
826
827static inline bool sk_del_node_init(struct sock *sk)
828{
829 bool rc = __sk_del_node_init(sk);
830
831 if (rc) {
832 /* paranoid for a while -acme */
833 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
834 __sock_put(sk);
835 }
836 return rc;
837}
838#define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
839
840static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
841{
842 if (sk_hashed(sk)) {
843 hlist_nulls_del_init_rcu(n: &sk->sk_nulls_node);
844 return true;
845 }
846 return false;
847}
848
849static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
850{
851 bool rc = __sk_nulls_del_node_init_rcu(sk);
852
853 if (rc) {
854 /* paranoid for a while -acme */
855 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
856 __sock_put(sk);
857 }
858 return rc;
859}
860
861static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
862{
863 hlist_add_head(n: &sk->sk_node, h: list);
864}
865
866static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
867{
868 sock_hold(sk);
869 __sk_add_node(sk, list);
870}
871
872static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
873{
874 sock_hold(sk);
875 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
876 sk->sk_family == AF_INET6)
877 hlist_add_tail_rcu(n: &sk->sk_node, h: list);
878 else
879 hlist_add_head_rcu(n: &sk->sk_node, h: list);
880}
881
882static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
883{
884 sock_hold(sk);
885 hlist_add_tail_rcu(n: &sk->sk_node, h: list);
886}
887
888static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
889{
890 hlist_nulls_add_head_rcu(n: &sk->sk_nulls_node, h: list);
891}
892
893static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
894{
895 hlist_nulls_add_tail_rcu(n: &sk->sk_nulls_node, h: list);
896}
897
898static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
899{
900 sock_hold(sk);
901 __sk_nulls_add_node_rcu(sk, list);
902}
903
904static inline void __sk_del_bind_node(struct sock *sk)
905{
906 __hlist_del(n: &sk->sk_bind_node);
907}
908
909static inline void sk_add_bind_node(struct sock *sk,
910 struct hlist_head *list)
911{
912 hlist_add_head(n: &sk->sk_bind_node, h: list);
913}
914
915#define sk_for_each(__sk, list) \
916 hlist_for_each_entry(__sk, list, sk_node)
917#define sk_for_each_rcu(__sk, list) \
918 hlist_for_each_entry_rcu(__sk, list, sk_node)
919#define sk_nulls_for_each(__sk, node, list) \
920 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
921#define sk_nulls_for_each_rcu(__sk, node, list) \
922 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
923#define sk_for_each_from(__sk) \
924 hlist_for_each_entry_from(__sk, sk_node)
925#define sk_nulls_for_each_from(__sk, node) \
926 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
927 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
928#define sk_for_each_safe(__sk, tmp, list) \
929 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
930#define sk_for_each_bound(__sk, list) \
931 hlist_for_each_entry(__sk, list, sk_bind_node)
932#define sk_for_each_bound_safe(__sk, tmp, list) \
933 hlist_for_each_entry_safe(__sk, tmp, list, sk_bind_node)
934
935/**
936 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
937 * @tpos: the type * to use as a loop cursor.
938 * @pos: the &struct hlist_node to use as a loop cursor.
939 * @head: the head for your list.
940 * @offset: offset of hlist_node within the struct.
941 *
942 */
943#define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
944 for (pos = rcu_dereference(hlist_first_rcu(head)); \
945 pos != NULL && \
946 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
947 pos = rcu_dereference(hlist_next_rcu(pos)))
948
949static inline struct user_namespace *sk_user_ns(const struct sock *sk)
950{
951 /* Careful only use this in a context where these parameters
952 * can not change and must all be valid, such as recvmsg from
953 * userspace.
954 */
955 return sk->sk_socket->file->f_cred->user_ns;
956}
957
958/* Sock flags */
959enum sock_flags {
960 SOCK_DEAD,
961 SOCK_DONE,
962 SOCK_URGINLINE,
963 SOCK_KEEPOPEN,
964 SOCK_LINGER,
965 SOCK_DESTROY,
966 SOCK_BROADCAST,
967 SOCK_TIMESTAMP,
968 SOCK_ZAPPED,
969 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
970 SOCK_DBG, /* %SO_DEBUG setting */
971 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
972 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
973 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
974 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
975 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
976 SOCK_FASYNC, /* fasync() active */
977 SOCK_RXQ_OVFL,
978 SOCK_ZEROCOPY, /* buffers from userspace */
979 SOCK_WIFI_STATUS, /* push wifi status to userspace */
980 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
981 * Will use last 4 bytes of packet sent from
982 * user-space instead.
983 */
984 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
985 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
986 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
987 SOCK_TXTIME,
988 SOCK_XDP, /* XDP is attached */
989 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
990 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */
991 SOCK_RCVPRIORITY, /* Receive SO_PRIORITY ancillary data with packet */
992 SOCK_TIMESTAMPING_ANY, /* Copy of sk_tsflags & TSFLAGS_ANY */
993};
994
995#define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
996/*
997 * The highest bit of sk_tsflags is reserved for kernel-internal
998 * SOCKCM_FLAG_TS_OPT_ID. There is a check in core/sock.c to control that
999 * SOF_TIMESTAMPING* values do not reach this reserved area
1000 */
1001#define SOCKCM_FLAG_TS_OPT_ID BIT(31)
1002
1003static inline void sock_copy_flags(struct sock *nsk, const struct sock *osk)
1004{
1005 nsk->sk_flags = osk->sk_flags;
1006}
1007
1008static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
1009{
1010 __set_bit(flag, &sk->sk_flags);
1011}
1012
1013static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
1014{
1015 __clear_bit(flag, &sk->sk_flags);
1016}
1017
1018static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
1019 int valbool)
1020{
1021 if (valbool)
1022 sock_set_flag(sk, flag: bit);
1023 else
1024 sock_reset_flag(sk, flag: bit);
1025}
1026
1027static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
1028{
1029 return test_bit(flag, &sk->sk_flags);
1030}
1031
1032#ifdef CONFIG_NET
1033DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
1034static inline int sk_memalloc_socks(void)
1035{
1036 return static_branch_unlikely(&memalloc_socks_key);
1037}
1038
1039void __receive_sock(struct file *file);
1040#else
1041
1042static inline int sk_memalloc_socks(void)
1043{
1044 return 0;
1045}
1046
1047static inline void __receive_sock(struct file *file)
1048{ }
1049#endif
1050
1051static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
1052{
1053 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
1054}
1055
1056static inline void sk_acceptq_removed(struct sock *sk)
1057{
1058 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
1059}
1060
1061static inline void sk_acceptq_added(struct sock *sk)
1062{
1063 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
1064}
1065
1066/* Note: If you think the test should be:
1067 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
1068 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
1069 */
1070static inline bool sk_acceptq_is_full(const struct sock *sk)
1071{
1072 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
1073}
1074
1075/*
1076 * Compute minimal free write space needed to queue new packets.
1077 */
1078static inline int sk_stream_min_wspace(const struct sock *sk)
1079{
1080 return READ_ONCE(sk->sk_wmem_queued) >> 1;
1081}
1082
1083static inline int sk_stream_wspace(const struct sock *sk)
1084{
1085 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1086}
1087
1088static inline void sk_wmem_queued_add(struct sock *sk, int val)
1089{
1090 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1091}
1092
1093static inline void sk_forward_alloc_add(struct sock *sk, int val)
1094{
1095 /* Paired with lockless reads of sk->sk_forward_alloc */
1096 WRITE_ONCE(sk->sk_forward_alloc, sk->sk_forward_alloc + val);
1097}
1098
1099void sk_stream_write_space(struct sock *sk);
1100
1101/* OOB backlog add */
1102static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1103{
1104 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1105 skb_dst_force(skb);
1106
1107 if (!sk->sk_backlog.tail)
1108 WRITE_ONCE(sk->sk_backlog.head, skb);
1109 else
1110 sk->sk_backlog.tail->next = skb;
1111
1112 WRITE_ONCE(sk->sk_backlog.tail, skb);
1113 skb->next = NULL;
1114}
1115
1116/*
1117 * Take into account size of receive queue and backlog queue
1118 * Do not take into account this skb truesize,
1119 * to allow even a single big packet to come.
1120 */
1121static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1122{
1123 unsigned int qsize = sk->sk_backlog.len + atomic_read(v: &sk->sk_rmem_alloc);
1124
1125 return qsize > limit;
1126}
1127
1128/* The per-socket spinlock must be held here. */
1129static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1130 unsigned int limit)
1131{
1132 if (sk_rcvqueues_full(sk, limit))
1133 return -ENOBUFS;
1134
1135 /*
1136 * If the skb was allocated from pfmemalloc reserves, only
1137 * allow SOCK_MEMALLOC sockets to use it as this socket is
1138 * helping free memory
1139 */
1140 if (skb_pfmemalloc(skb) && !sock_flag(sk, flag: SOCK_MEMALLOC))
1141 return -ENOMEM;
1142
1143 __sk_add_backlog(sk, skb);
1144 sk->sk_backlog.len += skb->truesize;
1145 return 0;
1146}
1147
1148int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1149
1150INDIRECT_CALLABLE_DECLARE(int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb));
1151INDIRECT_CALLABLE_DECLARE(int tcp_v6_do_rcv(struct sock *sk, struct sk_buff *skb));
1152
1153static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1154{
1155 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1156 return __sk_backlog_rcv(sk, skb);
1157
1158 return INDIRECT_CALL_INET(sk->sk_backlog_rcv,
1159 tcp_v6_do_rcv,
1160 tcp_v4_do_rcv,
1161 sk, skb);
1162}
1163
1164static inline void sk_incoming_cpu_update(struct sock *sk)
1165{
1166 int cpu = raw_smp_processor_id();
1167
1168 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1169 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1170}
1171
1172
1173static inline void sock_rps_save_rxhash(struct sock *sk,
1174 const struct sk_buff *skb)
1175{
1176#ifdef CONFIG_RPS
1177 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1178 * here, and another one in sock_rps_record_flow().
1179 */
1180 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1181 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1182#endif
1183}
1184
1185static inline void sock_rps_reset_rxhash(struct sock *sk)
1186{
1187#ifdef CONFIG_RPS
1188 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1189 WRITE_ONCE(sk->sk_rxhash, 0);
1190#endif
1191}
1192
1193#define sk_wait_event(__sk, __timeo, __condition, __wait) \
1194 ({ int __rc, __dis = __sk->sk_disconnects; \
1195 release_sock(__sk); \
1196 __rc = __condition; \
1197 if (!__rc) { \
1198 *(__timeo) = wait_woken(__wait, \
1199 TASK_INTERRUPTIBLE, \
1200 *(__timeo)); \
1201 } \
1202 sched_annotate_sleep(); \
1203 lock_sock(__sk); \
1204 __rc = __dis == __sk->sk_disconnects ? __condition : -EPIPE; \
1205 __rc; \
1206 })
1207
1208int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1209int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1210void sk_stream_wait_close(struct sock *sk, long timeo_p);
1211int sk_stream_error(struct sock *sk, int flags, int err);
1212void sk_stream_kill_queues(struct sock *sk);
1213void sk_set_memalloc(struct sock *sk);
1214void sk_clear_memalloc(struct sock *sk);
1215
1216void __sk_flush_backlog(struct sock *sk);
1217
1218static inline bool sk_flush_backlog(struct sock *sk)
1219{
1220 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1221 __sk_flush_backlog(sk);
1222 return true;
1223 }
1224 return false;
1225}
1226
1227int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1228
1229struct request_sock_ops;
1230struct timewait_sock_ops;
1231struct inet_hashinfo;
1232struct raw_hashinfo;
1233struct smc_hashinfo;
1234struct module;
1235struct sk_psock;
1236
1237/*
1238 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1239 * un-modified. Special care is taken when initializing object to zero.
1240 */
1241static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1242{
1243 if (offsetof(struct sock, sk_node.next) != 0)
1244 memset(s: sk, c: 0, offsetof(struct sock, sk_node.next));
1245 memset(s: &sk->sk_node.pprev, c: 0,
1246 n: size - offsetof(struct sock, sk_node.pprev));
1247}
1248
1249struct proto_accept_arg {
1250 int flags;
1251 int err;
1252 int is_empty;
1253 bool kern;
1254};
1255
1256/* Networking protocol blocks we attach to sockets.
1257 * socket layer -> transport layer interface
1258 */
1259struct proto {
1260 void (*close)(struct sock *sk,
1261 long timeout);
1262 int (*pre_connect)(struct sock *sk,
1263 struct sockaddr *uaddr,
1264 int addr_len);
1265 int (*connect)(struct sock *sk,
1266 struct sockaddr *uaddr,
1267 int addr_len);
1268 int (*disconnect)(struct sock *sk, int flags);
1269
1270 struct sock * (*accept)(struct sock *sk,
1271 struct proto_accept_arg *arg);
1272
1273 int (*ioctl)(struct sock *sk, int cmd,
1274 int *karg);
1275 int (*init)(struct sock *sk);
1276 void (*destroy)(struct sock *sk);
1277 void (*shutdown)(struct sock *sk, int how);
1278 int (*setsockopt)(struct sock *sk, int level,
1279 int optname, sockptr_t optval,
1280 unsigned int optlen);
1281 int (*getsockopt)(struct sock *sk, int level,
1282 int optname, char __user *optval,
1283 int __user *option);
1284 void (*keepalive)(struct sock *sk, int valbool);
1285#ifdef CONFIG_COMPAT
1286 int (*compat_ioctl)(struct sock *sk,
1287 unsigned int cmd, unsigned long arg);
1288#endif
1289 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1290 size_t len);
1291 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1292 size_t len, int flags, int *addr_len);
1293 void (*splice_eof)(struct socket *sock);
1294 int (*bind)(struct sock *sk,
1295 struct sockaddr *addr, int addr_len);
1296 int (*bind_add)(struct sock *sk,
1297 struct sockaddr *addr, int addr_len);
1298
1299 int (*backlog_rcv) (struct sock *sk,
1300 struct sk_buff *skb);
1301 bool (*bpf_bypass_getsockopt)(int level,
1302 int optname);
1303
1304 void (*release_cb)(struct sock *sk);
1305
1306 /* Keeping track of sk's, looking them up, and port selection methods. */
1307 int (*hash)(struct sock *sk);
1308 void (*unhash)(struct sock *sk);
1309 void (*rehash)(struct sock *sk);
1310 int (*get_port)(struct sock *sk, unsigned short snum);
1311 void (*put_port)(struct sock *sk);
1312#ifdef CONFIG_BPF_SYSCALL
1313 int (*psock_update_sk_prot)(struct sock *sk,
1314 struct sk_psock *psock,
1315 bool restore);
1316#endif
1317
1318 /* Keeping track of sockets in use */
1319#ifdef CONFIG_PROC_FS
1320 unsigned int inuse_idx;
1321#endif
1322
1323 bool (*stream_memory_free)(const struct sock *sk, int wake);
1324 bool (*sock_is_readable)(struct sock *sk);
1325 /* Memory pressure */
1326 void (*enter_memory_pressure)(struct sock *sk);
1327 void (*leave_memory_pressure)(struct sock *sk);
1328 atomic_long_t *memory_allocated; /* Current allocated memory. */
1329 int __percpu *per_cpu_fw_alloc;
1330 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1331
1332 /*
1333 * Pressure flag: try to collapse.
1334 * Technical note: it is used by multiple contexts non atomically.
1335 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1336 * All the __sk_mem_schedule() is of this nature: accounting
1337 * is strict, actions are advisory and have some latency.
1338 */
1339 unsigned long *memory_pressure;
1340 long *sysctl_mem;
1341
1342 int *sysctl_wmem;
1343 int *sysctl_rmem;
1344 u32 sysctl_wmem_offset;
1345 u32 sysctl_rmem_offset;
1346
1347 int max_header;
1348 bool no_autobind;
1349
1350 struct kmem_cache *slab;
1351 unsigned int obj_size;
1352 unsigned int ipv6_pinfo_offset;
1353 slab_flags_t slab_flags;
1354 unsigned int useroffset; /* Usercopy region offset */
1355 unsigned int usersize; /* Usercopy region size */
1356
1357 struct request_sock_ops *rsk_prot;
1358 struct timewait_sock_ops *twsk_prot;
1359
1360 union {
1361 struct inet_hashinfo *hashinfo;
1362 struct udp_table *udp_table;
1363 struct raw_hashinfo *raw_hash;
1364 struct smc_hashinfo *smc_hash;
1365 } h;
1366
1367 struct module *owner;
1368
1369 char name[32];
1370
1371 struct list_head node;
1372 int (*diag_destroy)(struct sock *sk, int err);
1373} __randomize_layout;
1374
1375int proto_register(struct proto *prot, int alloc_slab);
1376void proto_unregister(struct proto *prot);
1377int sock_load_diag_module(int family, int protocol);
1378
1379INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1380
1381static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1382{
1383 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1384 return false;
1385
1386 return sk->sk_prot->stream_memory_free ?
1387 INDIRECT_CALL_INET_1(sk->sk_prot->stream_memory_free,
1388 tcp_stream_memory_free, sk, wake) : true;
1389}
1390
1391static inline bool sk_stream_memory_free(const struct sock *sk)
1392{
1393 return __sk_stream_memory_free(sk, wake: 0);
1394}
1395
1396static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1397{
1398 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1399 __sk_stream_memory_free(sk, wake);
1400}
1401
1402static inline bool sk_stream_is_writeable(const struct sock *sk)
1403{
1404 return __sk_stream_is_writeable(sk, wake: 0);
1405}
1406
1407static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1408 struct cgroup *ancestor)
1409{
1410#ifdef CONFIG_SOCK_CGROUP_DATA
1411 return cgroup_is_descendant(cgrp: sock_cgroup_ptr(skcd: &sk->sk_cgrp_data),
1412 ancestor);
1413#else
1414 return -ENOTSUPP;
1415#endif
1416}
1417
1418#define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1419
1420static inline void sk_sockets_allocated_dec(struct sock *sk)
1421{
1422 percpu_counter_add_batch(fbc: sk->sk_prot->sockets_allocated, amount: -1,
1423 SK_ALLOC_PERCPU_COUNTER_BATCH);
1424}
1425
1426static inline void sk_sockets_allocated_inc(struct sock *sk)
1427{
1428 percpu_counter_add_batch(fbc: sk->sk_prot->sockets_allocated, amount: 1,
1429 SK_ALLOC_PERCPU_COUNTER_BATCH);
1430}
1431
1432static inline u64
1433sk_sockets_allocated_read_positive(struct sock *sk)
1434{
1435 return percpu_counter_read_positive(fbc: sk->sk_prot->sockets_allocated);
1436}
1437
1438static inline int
1439proto_sockets_allocated_sum_positive(struct proto *prot)
1440{
1441 return percpu_counter_sum_positive(fbc: prot->sockets_allocated);
1442}
1443
1444#ifdef CONFIG_PROC_FS
1445#define PROTO_INUSE_NR 64 /* should be enough for the first time */
1446struct prot_inuse {
1447 int all;
1448 int val[PROTO_INUSE_NR];
1449};
1450
1451static inline void sock_prot_inuse_add(const struct net *net,
1452 const struct proto *prot, int val)
1453{
1454 this_cpu_add(net->core.prot_inuse->val[prot->inuse_idx], val);
1455}
1456
1457static inline void sock_inuse_add(const struct net *net, int val)
1458{
1459 this_cpu_add(net->core.prot_inuse->all, val);
1460}
1461
1462int sock_prot_inuse_get(struct net *net, struct proto *proto);
1463int sock_inuse_get(struct net *net);
1464#else
1465static inline void sock_prot_inuse_add(const struct net *net,
1466 const struct proto *prot, int val)
1467{
1468}
1469
1470static inline void sock_inuse_add(const struct net *net, int val)
1471{
1472}
1473#endif
1474
1475
1476/* With per-bucket locks this operation is not-atomic, so that
1477 * this version is not worse.
1478 */
1479static inline int __sk_prot_rehash(struct sock *sk)
1480{
1481 sk->sk_prot->unhash(sk);
1482 return sk->sk_prot->hash(sk);
1483}
1484
1485/* About 10 seconds */
1486#define SOCK_DESTROY_TIME (10*HZ)
1487
1488/* Sockets 0-1023 can't be bound to unless you are superuser */
1489#define PROT_SOCK 1024
1490
1491#define SHUTDOWN_MASK 3
1492#define RCV_SHUTDOWN 1
1493#define SEND_SHUTDOWN 2
1494
1495#define SOCK_BINDADDR_LOCK 4
1496#define SOCK_BINDPORT_LOCK 8
1497/**
1498 * define SOCK_CONNECT_BIND - &sock->sk_userlocks flag for auto-bind at connect() time
1499 */
1500#define SOCK_CONNECT_BIND 16
1501
1502struct socket_alloc {
1503 struct socket socket;
1504 struct inode vfs_inode;
1505};
1506
1507static inline struct socket *SOCKET_I(struct inode *inode)
1508{
1509 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1510}
1511
1512static inline struct inode *SOCK_INODE(struct socket *socket)
1513{
1514 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1515}
1516
1517/*
1518 * Functions for memory accounting
1519 */
1520int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1521int __sk_mem_schedule(struct sock *sk, int size, int kind);
1522void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1523void __sk_mem_reclaim(struct sock *sk, int amount);
1524
1525#define SK_MEM_SEND 0
1526#define SK_MEM_RECV 1
1527
1528/* sysctl_mem values are in pages */
1529static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1530{
1531 return READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1532}
1533
1534static inline int sk_mem_pages(int amt)
1535{
1536 return (amt + PAGE_SIZE - 1) >> PAGE_SHIFT;
1537}
1538
1539static inline bool sk_has_account(struct sock *sk)
1540{
1541 /* return true if protocol supports memory accounting */
1542 return !!sk->sk_prot->memory_allocated;
1543}
1544
1545static inline bool sk_wmem_schedule(struct sock *sk, int size)
1546{
1547 int delta;
1548
1549 if (!sk_has_account(sk))
1550 return true;
1551 delta = size - sk->sk_forward_alloc;
1552 return delta <= 0 || __sk_mem_schedule(sk, size: delta, SK_MEM_SEND);
1553}
1554
1555static inline bool
1556__sk_rmem_schedule(struct sock *sk, int size, bool pfmemalloc)
1557{
1558 int delta;
1559
1560 if (!sk_has_account(sk))
1561 return true;
1562 delta = size - sk->sk_forward_alloc;
1563 return delta <= 0 || __sk_mem_schedule(sk, size: delta, SK_MEM_RECV) ||
1564 pfmemalloc;
1565}
1566
1567static inline bool
1568sk_rmem_schedule(struct sock *sk, const struct sk_buff *skb, int size)
1569{
1570 return __sk_rmem_schedule(sk, size, pfmemalloc: skb_pfmemalloc(skb));
1571}
1572
1573static inline int sk_unused_reserved_mem(const struct sock *sk)
1574{
1575 int unused_mem;
1576
1577 if (likely(!sk->sk_reserved_mem))
1578 return 0;
1579
1580 unused_mem = sk->sk_reserved_mem - sk->sk_wmem_queued -
1581 atomic_read(v: &sk->sk_rmem_alloc);
1582
1583 return unused_mem > 0 ? unused_mem : 0;
1584}
1585
1586static inline void sk_mem_reclaim(struct sock *sk)
1587{
1588 int reclaimable;
1589
1590 if (!sk_has_account(sk))
1591 return;
1592
1593 reclaimable = sk->sk_forward_alloc - sk_unused_reserved_mem(sk);
1594
1595 if (reclaimable >= (int)PAGE_SIZE)
1596 __sk_mem_reclaim(sk, amount: reclaimable);
1597}
1598
1599static inline void sk_mem_reclaim_final(struct sock *sk)
1600{
1601 sk->sk_reserved_mem = 0;
1602 sk_mem_reclaim(sk);
1603}
1604
1605static inline void sk_mem_charge(struct sock *sk, int size)
1606{
1607 if (!sk_has_account(sk))
1608 return;
1609 sk_forward_alloc_add(sk, val: -size);
1610}
1611
1612static inline void sk_mem_uncharge(struct sock *sk, int size)
1613{
1614 if (!sk_has_account(sk))
1615 return;
1616 sk_forward_alloc_add(sk, val: size);
1617 sk_mem_reclaim(sk);
1618}
1619
1620#if IS_ENABLED(CONFIG_PROVE_LOCKING) && IS_ENABLED(CONFIG_MODULES)
1621static inline void sk_owner_set(struct sock *sk, struct module *owner)
1622{
1623 __module_get(owner);
1624 sk->sk_owner = owner;
1625}
1626
1627static inline void sk_owner_clear(struct sock *sk)
1628{
1629 sk->sk_owner = NULL;
1630}
1631
1632static inline void sk_owner_put(struct sock *sk)
1633{
1634 module_put(sk->sk_owner);
1635}
1636#else
1637static inline void sk_owner_set(struct sock *sk, struct module *owner)
1638{
1639}
1640
1641static inline void sk_owner_clear(struct sock *sk)
1642{
1643}
1644
1645static inline void sk_owner_put(struct sock *sk)
1646{
1647}
1648#endif
1649/*
1650 * Macro so as to not evaluate some arguments when
1651 * lockdep is not enabled.
1652 *
1653 * Mark both the sk_lock and the sk_lock.slock as a
1654 * per-address-family lock class.
1655 */
1656#define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1657do { \
1658 sk_owner_set(sk, THIS_MODULE); \
1659 sk->sk_lock.owned = 0; \
1660 init_waitqueue_head(&sk->sk_lock.wq); \
1661 spin_lock_init(&(sk)->sk_lock.slock); \
1662 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1663 sizeof((sk)->sk_lock)); \
1664 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1665 (skey), (sname)); \
1666 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1667} while (0)
1668
1669static inline bool lockdep_sock_is_held(const struct sock *sk)
1670{
1671 return lockdep_is_held(&sk->sk_lock) ||
1672 lockdep_is_held(&sk->sk_lock.slock);
1673}
1674
1675void lock_sock_nested(struct sock *sk, int subclass);
1676
1677static inline void lock_sock(struct sock *sk)
1678{
1679 lock_sock_nested(sk, subclass: 0);
1680}
1681
1682void __lock_sock(struct sock *sk);
1683void __release_sock(struct sock *sk);
1684void release_sock(struct sock *sk);
1685
1686/* BH context may only use the following locking interface. */
1687#define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1688#define bh_lock_sock_nested(__sk) \
1689 spin_lock_nested(&((__sk)->sk_lock.slock), \
1690 SINGLE_DEPTH_NESTING)
1691#define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1692
1693bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1694
1695/**
1696 * lock_sock_fast - fast version of lock_sock
1697 * @sk: socket
1698 *
1699 * This version should be used for very small section, where process won't block
1700 * return false if fast path is taken:
1701 *
1702 * sk_lock.slock locked, owned = 0, BH disabled
1703 *
1704 * return true if slow path is taken:
1705 *
1706 * sk_lock.slock unlocked, owned = 1, BH enabled
1707 */
1708static inline bool lock_sock_fast(struct sock *sk)
1709{
1710 /* The sk_lock has mutex_lock() semantics here. */
1711 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1712
1713 return __lock_sock_fast(sk);
1714}
1715
1716/* fast socket lock variant for caller already holding a [different] socket lock */
1717static inline bool lock_sock_fast_nested(struct sock *sk)
1718{
1719 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1720
1721 return __lock_sock_fast(sk);
1722}
1723
1724/**
1725 * unlock_sock_fast - complement of lock_sock_fast
1726 * @sk: socket
1727 * @slow: slow mode
1728 *
1729 * fast unlock socket for user context.
1730 * If slow mode is on, we call regular release_sock()
1731 */
1732static inline void unlock_sock_fast(struct sock *sk, bool slow)
1733 __releases(&sk->sk_lock.slock)
1734{
1735 if (slow) {
1736 release_sock(sk);
1737 __release(&sk->sk_lock.slock);
1738 } else {
1739 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1740 spin_unlock_bh(lock: &sk->sk_lock.slock);
1741 }
1742}
1743
1744void sockopt_lock_sock(struct sock *sk);
1745void sockopt_release_sock(struct sock *sk);
1746bool sockopt_ns_capable(struct user_namespace *ns, int cap);
1747bool sockopt_capable(int cap);
1748
1749/* Used by processes to "lock" a socket state, so that
1750 * interrupts and bottom half handlers won't change it
1751 * from under us. It essentially blocks any incoming
1752 * packets, so that we won't get any new data or any
1753 * packets that change the state of the socket.
1754 *
1755 * While locked, BH processing will add new packets to
1756 * the backlog queue. This queue is processed by the
1757 * owner of the socket lock right before it is released.
1758 *
1759 * Since ~2.3.5 it is also exclusive sleep lock serializing
1760 * accesses from user process context.
1761 */
1762
1763static inline void sock_owned_by_me(const struct sock *sk)
1764{
1765#ifdef CONFIG_LOCKDEP
1766 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1767#endif
1768}
1769
1770static inline void sock_not_owned_by_me(const struct sock *sk)
1771{
1772#ifdef CONFIG_LOCKDEP
1773 WARN_ON_ONCE(lockdep_sock_is_held(sk) && debug_locks);
1774#endif
1775}
1776
1777static inline bool sock_owned_by_user(const struct sock *sk)
1778{
1779 sock_owned_by_me(sk);
1780 return sk->sk_lock.owned;
1781}
1782
1783static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1784{
1785 return sk->sk_lock.owned;
1786}
1787
1788static inline void sock_release_ownership(struct sock *sk)
1789{
1790 DEBUG_NET_WARN_ON_ONCE(!sock_owned_by_user_nocheck(sk));
1791 sk->sk_lock.owned = 0;
1792
1793 /* The sk_lock has mutex_unlock() semantics: */
1794 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1795}
1796
1797/* no reclassification while locks are held */
1798static inline bool sock_allow_reclassification(const struct sock *csk)
1799{
1800 struct sock *sk = (struct sock *)csk;
1801
1802 return !sock_owned_by_user_nocheck(sk) &&
1803 !spin_is_locked(lock: &sk->sk_lock.slock);
1804}
1805
1806struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1807 struct proto *prot, int kern);
1808void sk_free(struct sock *sk);
1809void sk_net_refcnt_upgrade(struct sock *sk);
1810void sk_destruct(struct sock *sk);
1811struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1812
1813struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1814 gfp_t priority);
1815void __sock_wfree(struct sk_buff *skb);
1816void sock_wfree(struct sk_buff *skb);
1817struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1818 gfp_t priority);
1819void skb_orphan_partial(struct sk_buff *skb);
1820void sock_rfree(struct sk_buff *skb);
1821void sock_efree(struct sk_buff *skb);
1822#ifdef CONFIG_INET
1823void sock_edemux(struct sk_buff *skb);
1824void sock_pfree(struct sk_buff *skb);
1825
1826static inline void skb_set_owner_edemux(struct sk_buff *skb, struct sock *sk)
1827{
1828 skb_orphan(skb);
1829 if (refcount_inc_not_zero(r: &sk->sk_refcnt)) {
1830 skb->sk = sk;
1831 skb->destructor = sock_edemux;
1832 }
1833}
1834#else
1835#define sock_edemux sock_efree
1836#endif
1837
1838int sk_setsockopt(struct sock *sk, int level, int optname,
1839 sockptr_t optval, unsigned int optlen);
1840int sock_setsockopt(struct socket *sock, int level, int op,
1841 sockptr_t optval, unsigned int optlen);
1842int do_sock_setsockopt(struct socket *sock, bool compat, int level,
1843 int optname, sockptr_t optval, int optlen);
1844int do_sock_getsockopt(struct socket *sock, bool compat, int level,
1845 int optname, sockptr_t optval, sockptr_t optlen);
1846
1847int sk_getsockopt(struct sock *sk, int level, int optname,
1848 sockptr_t optval, sockptr_t optlen);
1849int sock_gettstamp(struct socket *sock, void __user *userstamp,
1850 bool timeval, bool time32);
1851struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1852 unsigned long data_len, int noblock,
1853 int *errcode, int max_page_order);
1854
1855static inline struct sk_buff *sock_alloc_send_skb(struct sock *sk,
1856 unsigned long size,
1857 int noblock, int *errcode)
1858{
1859 return sock_alloc_send_pskb(sk, header_len: size, data_len: 0, noblock, errcode, max_page_order: 0);
1860}
1861
1862void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1863void *sock_kmemdup(struct sock *sk, const void *src,
1864 int size, gfp_t priority);
1865void sock_kfree_s(struct sock *sk, void *mem, int size);
1866void sock_kzfree_s(struct sock *sk, void *mem, int size);
1867void sk_send_sigurg(struct sock *sk);
1868
1869static inline void sock_replace_proto(struct sock *sk, struct proto *proto)
1870{
1871 if (sk->sk_socket)
1872 clear_bit(nr: SOCK_SUPPORT_ZC, addr: &sk->sk_socket->flags);
1873 WRITE_ONCE(sk->sk_prot, proto);
1874}
1875
1876struct sockcm_cookie {
1877 u64 transmit_time;
1878 u32 mark;
1879 u32 tsflags;
1880 u32 ts_opt_id;
1881 u32 priority;
1882 u32 dmabuf_id;
1883};
1884
1885static inline void sockcm_init(struct sockcm_cookie *sockc,
1886 const struct sock *sk)
1887{
1888 *sockc = (struct sockcm_cookie) {
1889 .mark = READ_ONCE(sk->sk_mark),
1890 .tsflags = READ_ONCE(sk->sk_tsflags),
1891 .priority = READ_ONCE(sk->sk_priority),
1892 };
1893}
1894
1895int __sock_cmsg_send(struct sock *sk, struct cmsghdr *cmsg,
1896 struct sockcm_cookie *sockc);
1897int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1898 struct sockcm_cookie *sockc);
1899
1900/*
1901 * Functions to fill in entries in struct proto_ops when a protocol
1902 * does not implement a particular function.
1903 */
1904int sock_no_bind(struct socket *, struct sockaddr *, int);
1905int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1906int sock_no_socketpair(struct socket *, struct socket *);
1907int sock_no_accept(struct socket *, struct socket *, struct proto_accept_arg *);
1908int sock_no_getname(struct socket *, struct sockaddr *, int);
1909int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1910int sock_no_listen(struct socket *, int);
1911int sock_no_shutdown(struct socket *, int);
1912int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1913int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1914int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1915int sock_no_mmap(struct file *file, struct socket *sock,
1916 struct vm_area_struct *vma);
1917
1918/*
1919 * Functions to fill in entries in struct proto_ops when a protocol
1920 * uses the inet style.
1921 */
1922int sock_common_getsockopt(struct socket *sock, int level, int optname,
1923 char __user *optval, int __user *optlen);
1924int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1925 int flags);
1926int sock_common_setsockopt(struct socket *sock, int level, int optname,
1927 sockptr_t optval, unsigned int optlen);
1928
1929void sk_common_release(struct sock *sk);
1930
1931/*
1932 * Default socket callbacks and setup code
1933 */
1934
1935/* Initialise core socket variables using an explicit uid. */
1936void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1937
1938/* Initialise core socket variables.
1939 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1940 */
1941void sock_init_data(struct socket *sock, struct sock *sk);
1942
1943/*
1944 * Socket reference counting postulates.
1945 *
1946 * * Each user of socket SHOULD hold a reference count.
1947 * * Each access point to socket (an hash table bucket, reference from a list,
1948 * running timer, skb in flight MUST hold a reference count.
1949 * * When reference count hits 0, it means it will never increase back.
1950 * * When reference count hits 0, it means that no references from
1951 * outside exist to this socket and current process on current CPU
1952 * is last user and may/should destroy this socket.
1953 * * sk_free is called from any context: process, BH, IRQ. When
1954 * it is called, socket has no references from outside -> sk_free
1955 * may release descendant resources allocated by the socket, but
1956 * to the time when it is called, socket is NOT referenced by any
1957 * hash tables, lists etc.
1958 * * Packets, delivered from outside (from network or from another process)
1959 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1960 * when they sit in queue. Otherwise, packets will leak to hole, when
1961 * socket is looked up by one cpu and unhasing is made by another CPU.
1962 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1963 * (leak to backlog). Packet socket does all the processing inside
1964 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1965 * use separate SMP lock, so that they are prone too.
1966 */
1967
1968/* Ungrab socket and destroy it, if it was the last reference. */
1969static inline void sock_put(struct sock *sk)
1970{
1971 if (refcount_dec_and_test(r: &sk->sk_refcnt))
1972 sk_free(sk);
1973}
1974/* Generic version of sock_put(), dealing with all sockets
1975 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1976 */
1977void sock_gen_put(struct sock *sk);
1978
1979int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1980 unsigned int trim_cap, bool refcounted);
1981static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1982 const int nested)
1983{
1984 return __sk_receive_skb(sk, skb, nested, trim_cap: 1, refcounted: true);
1985}
1986
1987static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1988{
1989 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1990 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1991 return;
1992 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1993 * other WRITE_ONCE() because socket lock might be not held.
1994 */
1995 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1996}
1997
1998#define NO_QUEUE_MAPPING USHRT_MAX
1999
2000static inline void sk_tx_queue_clear(struct sock *sk)
2001{
2002 /* Paired with READ_ONCE() in sk_tx_queue_get() and
2003 * other WRITE_ONCE() because socket lock might be not held.
2004 */
2005 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
2006}
2007
2008static inline int sk_tx_queue_get(const struct sock *sk)
2009{
2010 if (sk) {
2011 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
2012 * and sk_tx_queue_set().
2013 */
2014 int val = READ_ONCE(sk->sk_tx_queue_mapping);
2015
2016 if (val != NO_QUEUE_MAPPING)
2017 return val;
2018 }
2019 return -1;
2020}
2021
2022static inline void __sk_rx_queue_set(struct sock *sk,
2023 const struct sk_buff *skb,
2024 bool force_set)
2025{
2026#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2027 if (skb_rx_queue_recorded(skb)) {
2028 u16 rx_queue = skb_get_rx_queue(skb);
2029
2030 if (force_set ||
2031 unlikely(READ_ONCE(sk->sk_rx_queue_mapping) != rx_queue))
2032 WRITE_ONCE(sk->sk_rx_queue_mapping, rx_queue);
2033 }
2034#endif
2035}
2036
2037static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
2038{
2039 __sk_rx_queue_set(sk, skb, force_set: true);
2040}
2041
2042static inline void sk_rx_queue_update(struct sock *sk, const struct sk_buff *skb)
2043{
2044 __sk_rx_queue_set(sk, skb, force_set: false);
2045}
2046
2047static inline void sk_rx_queue_clear(struct sock *sk)
2048{
2049#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2050 WRITE_ONCE(sk->sk_rx_queue_mapping, NO_QUEUE_MAPPING);
2051#endif
2052}
2053
2054static inline int sk_rx_queue_get(const struct sock *sk)
2055{
2056#ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
2057 if (sk) {
2058 int res = READ_ONCE(sk->sk_rx_queue_mapping);
2059
2060 if (res != NO_QUEUE_MAPPING)
2061 return res;
2062 }
2063#endif
2064
2065 return -1;
2066}
2067
2068static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2069{
2070 sk->sk_socket = sock;
2071 if (sock) {
2072 WRITE_ONCE(sk->sk_uid, SOCK_INODE(sock)->i_uid);
2073 WRITE_ONCE(sk->sk_ino, SOCK_INODE(sock)->i_ino);
2074 } else {
2075 /* Note: sk_uid is unchanged. */
2076 WRITE_ONCE(sk->sk_ino, 0);
2077 }
2078}
2079
2080static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2081{
2082 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2083 return &rcu_dereference_raw(sk->sk_wq)->wait;
2084}
2085/* Detach socket from process context.
2086 * Announce socket dead, detach it from wait queue and inode.
2087 * Note that parent inode held reference count on this struct sock,
2088 * we do not release it in this function, because protocol
2089 * probably wants some additional cleanups or even continuing
2090 * to work with this socket (TCP).
2091 */
2092static inline void sock_orphan(struct sock *sk)
2093{
2094 write_lock_bh(&sk->sk_callback_lock);
2095 sock_set_flag(sk, flag: SOCK_DEAD);
2096 sk_set_socket(sk, NULL);
2097 sk->sk_wq = NULL;
2098 write_unlock_bh(&sk->sk_callback_lock);
2099}
2100
2101static inline void sock_graft(struct sock *sk, struct socket *parent)
2102{
2103 WARN_ON(parent->sk);
2104 write_lock_bh(&sk->sk_callback_lock);
2105 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2106 parent->sk = sk;
2107 sk_set_socket(sk, sock: parent);
2108 security_sock_graft(sk, parent);
2109 write_unlock_bh(&sk->sk_callback_lock);
2110}
2111
2112static inline unsigned long sock_i_ino(const struct sock *sk)
2113{
2114 /* Paired with WRITE_ONCE() in sock_graft() and sock_orphan() */
2115 return READ_ONCE(sk->sk_ino);
2116}
2117
2118static inline kuid_t sk_uid(const struct sock *sk)
2119{
2120 /* Paired with WRITE_ONCE() in sockfs_setattr() */
2121 return READ_ONCE(sk->sk_uid);
2122}
2123
2124static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2125{
2126 return sk ? sk_uid(sk) : make_kuid(from: net->user_ns, uid: 0);
2127}
2128
2129static inline u32 net_tx_rndhash(void)
2130{
2131 u32 v = get_random_u32();
2132
2133 return v ?: 1;
2134}
2135
2136static inline void sk_set_txhash(struct sock *sk)
2137{
2138 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2139 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2140}
2141
2142static inline bool sk_rethink_txhash(struct sock *sk)
2143{
2144 if (sk->sk_txhash && sk->sk_txrehash == SOCK_TXREHASH_ENABLED) {
2145 sk_set_txhash(sk);
2146 return true;
2147 }
2148 return false;
2149}
2150
2151static inline struct dst_entry *
2152__sk_dst_get(const struct sock *sk)
2153{
2154 return rcu_dereference_check(sk->sk_dst_cache,
2155 lockdep_sock_is_held(sk));
2156}
2157
2158static inline struct dst_entry *
2159sk_dst_get(const struct sock *sk)
2160{
2161 struct dst_entry *dst;
2162
2163 rcu_read_lock();
2164 dst = rcu_dereference(sk->sk_dst_cache);
2165 if (dst && !rcuref_get(ref: &dst->__rcuref))
2166 dst = NULL;
2167 rcu_read_unlock();
2168 return dst;
2169}
2170
2171static inline void __dst_negative_advice(struct sock *sk)
2172{
2173 struct dst_entry *dst = __sk_dst_get(sk);
2174
2175 if (dst && dst->ops->negative_advice)
2176 dst->ops->negative_advice(sk, dst);
2177}
2178
2179static inline void dst_negative_advice(struct sock *sk)
2180{
2181 sk_rethink_txhash(sk);
2182 __dst_negative_advice(sk);
2183}
2184
2185static inline void
2186__sk_dst_set(struct sock *sk, struct dst_entry *dst)
2187{
2188 struct dst_entry *old_dst;
2189
2190 sk_tx_queue_clear(sk);
2191 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2192 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2193 lockdep_sock_is_held(sk));
2194 rcu_assign_pointer(sk->sk_dst_cache, dst);
2195 dst_release(dst: old_dst);
2196}
2197
2198static inline void
2199sk_dst_set(struct sock *sk, struct dst_entry *dst)
2200{
2201 struct dst_entry *old_dst;
2202
2203 sk_tx_queue_clear(sk);
2204 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2205 old_dst = unrcu_pointer(xchg(&sk->sk_dst_cache, RCU_INITIALIZER(dst)));
2206 dst_release(dst: old_dst);
2207}
2208
2209static inline void
2210__sk_dst_reset(struct sock *sk)
2211{
2212 __sk_dst_set(sk, NULL);
2213}
2214
2215static inline void
2216sk_dst_reset(struct sock *sk)
2217{
2218 sk_dst_set(sk, NULL);
2219}
2220
2221struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2222
2223struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2224
2225static inline void sk_dst_confirm(struct sock *sk)
2226{
2227 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2228 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2229}
2230
2231static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2232{
2233 if (skb_get_dst_pending_confirm(skb)) {
2234 struct sock *sk = skb->sk;
2235
2236 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2237 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2238 neigh_confirm(n);
2239 }
2240}
2241
2242bool sk_mc_loop(const struct sock *sk);
2243
2244static inline bool sk_can_gso(const struct sock *sk)
2245{
2246 return net_gso_ok(features: sk->sk_route_caps, gso_type: sk->sk_gso_type);
2247}
2248
2249void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2250
2251static inline void sk_gso_disable(struct sock *sk)
2252{
2253 sk->sk_gso_disabled = 1;
2254 sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
2255}
2256
2257static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2258 struct iov_iter *from, char *to,
2259 int copy, int offset)
2260{
2261 if (skb->ip_summed == CHECKSUM_NONE) {
2262 __wsum csum = 0;
2263 if (!csum_and_copy_from_iter_full(addr: to, bytes: copy, csum: &csum, i: from))
2264 return -EFAULT;
2265 skb->csum = csum_block_add(csum: skb->csum, csum2: csum, offset);
2266 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2267 if (!copy_from_iter_full_nocache(addr: to, bytes: copy, i: from))
2268 return -EFAULT;
2269 } else if (!copy_from_iter_full(addr: to, bytes: copy, i: from))
2270 return -EFAULT;
2271
2272 return 0;
2273}
2274
2275static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2276 struct iov_iter *from, int copy)
2277{
2278 int err, offset = skb->len;
2279
2280 err = skb_do_copy_data_nocache(sk, skb, from, to: skb_put(skb, len: copy),
2281 copy, offset);
2282 if (err)
2283 __skb_trim(skb, len: offset);
2284
2285 return err;
2286}
2287
2288static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2289 struct sk_buff *skb,
2290 struct page *page,
2291 int off, int copy)
2292{
2293 int err;
2294
2295 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2296 copy, offset: skb->len);
2297 if (err)
2298 return err;
2299
2300 skb_len_add(skb, delta: copy);
2301 sk_wmem_queued_add(sk, val: copy);
2302 sk_mem_charge(sk, size: copy);
2303 return 0;
2304}
2305
2306/**
2307 * sk_wmem_alloc_get - returns write allocations
2308 * @sk: socket
2309 *
2310 * Return: sk_wmem_alloc minus initial offset of one
2311 */
2312static inline int sk_wmem_alloc_get(const struct sock *sk)
2313{
2314 return refcount_read(r: &sk->sk_wmem_alloc) - 1;
2315}
2316
2317/**
2318 * sk_rmem_alloc_get - returns read allocations
2319 * @sk: socket
2320 *
2321 * Return: sk_rmem_alloc
2322 */
2323static inline int sk_rmem_alloc_get(const struct sock *sk)
2324{
2325 return atomic_read(v: &sk->sk_rmem_alloc);
2326}
2327
2328/**
2329 * sk_has_allocations - check if allocations are outstanding
2330 * @sk: socket
2331 *
2332 * Return: true if socket has write or read allocations
2333 */
2334static inline bool sk_has_allocations(const struct sock *sk)
2335{
2336 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2337}
2338
2339/**
2340 * skwq_has_sleeper - check if there are any waiting processes
2341 * @wq: struct socket_wq
2342 *
2343 * Return: true if socket_wq has waiting processes
2344 *
2345 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2346 * barrier call. They were added due to the race found within the tcp code.
2347 *
2348 * Consider following tcp code paths::
2349 *
2350 * CPU1 CPU2
2351 * sys_select receive packet
2352 * ... ...
2353 * __add_wait_queue update tp->rcv_nxt
2354 * ... ...
2355 * tp->rcv_nxt check sock_def_readable
2356 * ... {
2357 * schedule rcu_read_lock();
2358 * wq = rcu_dereference(sk->sk_wq);
2359 * if (wq && waitqueue_active(&wq->wait))
2360 * wake_up_interruptible(&wq->wait)
2361 * ...
2362 * }
2363 *
2364 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2365 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2366 * could then endup calling schedule and sleep forever if there are no more
2367 * data on the socket.
2368 *
2369 */
2370static inline bool skwq_has_sleeper(struct socket_wq *wq)
2371{
2372 return wq && wq_has_sleeper(wq_head: &wq->wait);
2373}
2374
2375/**
2376 * sock_poll_wait - wrapper for the poll_wait call.
2377 * @filp: file
2378 * @sock: socket to wait on
2379 * @p: poll_table
2380 *
2381 * See the comments in the wq_has_sleeper function.
2382 */
2383static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2384 poll_table *p)
2385{
2386 /* Provides a barrier we need to be sure we are in sync
2387 * with the socket flags modification.
2388 *
2389 * This memory barrier is paired in the wq_has_sleeper.
2390 */
2391 poll_wait(filp, wait_address: &sock->wq.wait, p);
2392}
2393
2394static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2395{
2396 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2397 u32 txhash = READ_ONCE(sk->sk_txhash);
2398
2399 if (txhash) {
2400 skb->l4_hash = 1;
2401 skb->hash = txhash;
2402 }
2403}
2404
2405void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2406
2407/*
2408 * Queue a received datagram if it will fit. Stream and sequenced
2409 * protocols can't normally use this as they need to fit buffers in
2410 * and play with them.
2411 *
2412 * Inlined as it's very short and called for pretty much every
2413 * packet ever received.
2414 */
2415static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2416{
2417 skb_orphan(skb);
2418 skb->sk = sk;
2419 skb->destructor = sock_rfree;
2420 atomic_add(i: skb->truesize, v: &sk->sk_rmem_alloc);
2421 sk_mem_charge(sk, size: skb->truesize);
2422}
2423
2424static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2425{
2426 if (sk && refcount_inc_not_zero(r: &sk->sk_refcnt)) {
2427 skb_orphan(skb);
2428 skb->destructor = sock_efree;
2429 skb->sk = sk;
2430 return true;
2431 }
2432 return false;
2433}
2434
2435static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2436{
2437 skb = skb_clone(skb, priority: sk_gfp_mask(sk, GFP_ATOMIC));
2438 if (skb) {
2439 if (sk_rmem_schedule(sk, skb, size: skb->truesize)) {
2440 skb_set_owner_r(skb, sk);
2441 return skb;
2442 }
2443 __kfree_skb(skb);
2444 }
2445 return NULL;
2446}
2447
2448static inline void skb_prepare_for_gro(struct sk_buff *skb)
2449{
2450 if (skb->destructor != sock_wfree) {
2451 skb_orphan(skb);
2452 return;
2453 }
2454 skb->slow_gro = 1;
2455}
2456
2457void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2458 unsigned long expires);
2459
2460void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2461
2462void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2463
2464int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2465 struct sk_buff *skb, unsigned int flags,
2466 void (*destructor)(struct sock *sk,
2467 struct sk_buff *skb));
2468int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2469
2470int sock_queue_rcv_skb_reason(struct sock *sk, struct sk_buff *skb,
2471 enum skb_drop_reason *reason);
2472
2473static inline int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
2474{
2475 return sock_queue_rcv_skb_reason(sk, skb, NULL);
2476}
2477
2478int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2479struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2480
2481/*
2482 * Recover an error report and clear atomically
2483 */
2484
2485static inline int sock_error(struct sock *sk)
2486{
2487 int err;
2488
2489 /* Avoid an atomic operation for the common case.
2490 * This is racy since another cpu/thread can change sk_err under us.
2491 */
2492 if (likely(data_race(!sk->sk_err)))
2493 return 0;
2494
2495 err = xchg(&sk->sk_err, 0);
2496 return -err;
2497}
2498
2499void sk_error_report(struct sock *sk);
2500
2501static inline unsigned long sock_wspace(struct sock *sk)
2502{
2503 int amt = 0;
2504
2505 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2506 amt = sk->sk_sndbuf - refcount_read(r: &sk->sk_wmem_alloc);
2507 if (amt < 0)
2508 amt = 0;
2509 }
2510 return amt;
2511}
2512
2513/* Note:
2514 * We use sk->sk_wq_raw, from contexts knowing this
2515 * pointer is not NULL and cannot disappear/change.
2516 */
2517static inline void sk_set_bit(int nr, struct sock *sk)
2518{
2519 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2520 !sock_flag(sk, flag: SOCK_FASYNC))
2521 return;
2522
2523 set_bit(nr, addr: &sk->sk_wq_raw->flags);
2524}
2525
2526static inline void sk_clear_bit(int nr, struct sock *sk)
2527{
2528 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2529 !sock_flag(sk, flag: SOCK_FASYNC))
2530 return;
2531
2532 clear_bit(nr, addr: &sk->sk_wq_raw->flags);
2533}
2534
2535static inline void sk_wake_async(const struct sock *sk, int how, int band)
2536{
2537 if (sock_flag(sk, flag: SOCK_FASYNC)) {
2538 rcu_read_lock();
2539 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2540 rcu_read_unlock();
2541 }
2542}
2543
2544static inline void sk_wake_async_rcu(const struct sock *sk, int how, int band)
2545{
2546 if (unlikely(sock_flag(sk, SOCK_FASYNC)))
2547 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2548}
2549
2550/* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2551 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2552 * Note: for send buffers, TCP works better if we can build two skbs at
2553 * minimum.
2554 */
2555#define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2556
2557#define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2558#define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2559
2560static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2561{
2562 u32 val;
2563
2564 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2565 return;
2566
2567 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2568 val = max_t(u32, val, sk_unused_reserved_mem(sk));
2569
2570 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2571}
2572
2573/**
2574 * sk_page_frag - return an appropriate page_frag
2575 * @sk: socket
2576 *
2577 * Use the per task page_frag instead of the per socket one for
2578 * optimization when we know that we're in process context and own
2579 * everything that's associated with %current.
2580 *
2581 * Both direct reclaim and page faults can nest inside other
2582 * socket operations and end up recursing into sk_page_frag()
2583 * while it's already in use: explicitly avoid task page_frag
2584 * when users disable sk_use_task_frag.
2585 *
2586 * Return: a per task page_frag if context allows that,
2587 * otherwise a per socket one.
2588 */
2589static inline struct page_frag *sk_page_frag(struct sock *sk)
2590{
2591 if (sk->sk_use_task_frag)
2592 return &current->task_frag;
2593
2594 return &sk->sk_frag;
2595}
2596
2597bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2598
2599/*
2600 * Default write policy as shown to user space via poll/select/SIGIO
2601 */
2602static inline bool sock_writeable(const struct sock *sk)
2603{
2604 return refcount_read(r: &sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2605}
2606
2607static inline gfp_t gfp_any(void)
2608{
2609 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2610}
2611
2612static inline gfp_t gfp_memcg_charge(void)
2613{
2614 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2615}
2616
2617#ifdef CONFIG_MEMCG
2618static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2619{
2620 return sk->sk_memcg;
2621}
2622
2623static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2624{
2625 return mem_cgroup_sockets_enabled && mem_cgroup_from_sk(sk);
2626}
2627
2628static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2629{
2630 struct mem_cgroup *memcg = mem_cgroup_from_sk(sk);
2631
2632#ifdef CONFIG_MEMCG_V1
2633 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys))
2634 return !!memcg->tcpmem_pressure;
2635#endif /* CONFIG_MEMCG_V1 */
2636
2637 do {
2638 if (time_before64(get_jiffies_64(), mem_cgroup_get_socket_pressure(memcg)))
2639 return true;
2640 } while ((memcg = parent_mem_cgroup(memcg)));
2641
2642 return false;
2643}
2644#else
2645static inline struct mem_cgroup *mem_cgroup_from_sk(const struct sock *sk)
2646{
2647 return NULL;
2648}
2649
2650static inline bool mem_cgroup_sk_enabled(const struct sock *sk)
2651{
2652 return false;
2653}
2654
2655static inline bool mem_cgroup_sk_under_memory_pressure(const struct sock *sk)
2656{
2657 return false;
2658}
2659#endif
2660
2661static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2662{
2663 return noblock ? 0 : READ_ONCE(sk->sk_rcvtimeo);
2664}
2665
2666static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2667{
2668 return noblock ? 0 : READ_ONCE(sk->sk_sndtimeo);
2669}
2670
2671static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2672{
2673 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2674
2675 return v ?: 1;
2676}
2677
2678/* Alas, with timeout socket operations are not restartable.
2679 * Compare this to poll().
2680 */
2681static inline int sock_intr_errno(long timeo)
2682{
2683 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2684}
2685
2686struct sock_skb_cb {
2687 u32 dropcount;
2688};
2689
2690/* Store sock_skb_cb at the end of skb->cb[] so protocol families
2691 * using skb->cb[] would keep using it directly and utilize its
2692 * alignment guarantee.
2693 */
2694#define SOCK_SKB_CB_OFFSET (sizeof_field(struct sk_buff, cb) - \
2695 sizeof(struct sock_skb_cb))
2696
2697#define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2698 SOCK_SKB_CB_OFFSET))
2699
2700#define sock_skb_cb_check_size(size) \
2701 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2702
2703static inline void sk_drops_add(struct sock *sk, int segs)
2704{
2705 struct numa_drop_counters *ndc = sk->sk_drop_counters;
2706
2707 if (ndc)
2708 numa_drop_add(ndc, val: segs);
2709 else
2710 atomic_add(i: segs, v: &sk->sk_drops);
2711}
2712
2713static inline void sk_drops_inc(struct sock *sk)
2714{
2715 sk_drops_add(sk, segs: 1);
2716}
2717
2718static inline int sk_drops_read(const struct sock *sk)
2719{
2720 const struct numa_drop_counters *ndc = sk->sk_drop_counters;
2721
2722 if (ndc) {
2723 DEBUG_NET_WARN_ON_ONCE(atomic_read(&sk->sk_drops));
2724 return numa_drop_read(ndc);
2725 }
2726 return atomic_read(v: &sk->sk_drops);
2727}
2728
2729static inline void sk_drops_reset(struct sock *sk)
2730{
2731 struct numa_drop_counters *ndc = sk->sk_drop_counters;
2732
2733 if (ndc)
2734 numa_drop_reset(ndc);
2735 atomic_set(v: &sk->sk_drops, i: 0);
2736}
2737
2738static inline void
2739sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2740{
2741 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, flag: SOCK_RXQ_OVFL) ?
2742 sk_drops_read(sk) : 0;
2743}
2744
2745static inline void sk_drops_skbadd(struct sock *sk, const struct sk_buff *skb)
2746{
2747 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2748
2749 sk_drops_add(sk, segs);
2750}
2751
2752static inline ktime_t sock_read_timestamp(struct sock *sk)
2753{
2754#if BITS_PER_LONG==32
2755 unsigned int seq;
2756 ktime_t kt;
2757
2758 do {
2759 seq = read_seqbegin(&sk->sk_stamp_seq);
2760 kt = sk->sk_stamp;
2761 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2762
2763 return kt;
2764#else
2765 return READ_ONCE(sk->sk_stamp);
2766#endif
2767}
2768
2769static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2770{
2771#if BITS_PER_LONG==32
2772 write_seqlock(&sk->sk_stamp_seq);
2773 sk->sk_stamp = kt;
2774 write_sequnlock(&sk->sk_stamp_seq);
2775#else
2776 WRITE_ONCE(sk->sk_stamp, kt);
2777#endif
2778}
2779
2780void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2781 struct sk_buff *skb);
2782void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2783 struct sk_buff *skb);
2784
2785bool skb_has_tx_timestamp(struct sk_buff *skb, const struct sock *sk);
2786int skb_get_tx_timestamp(struct sk_buff *skb, struct sock *sk,
2787 struct timespec64 *ts);
2788
2789static inline void
2790sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2791{
2792 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2793 u32 tsflags = READ_ONCE(sk->sk_tsflags);
2794 ktime_t kt = skb->tstamp;
2795 /*
2796 * generate control messages if
2797 * - receive time stamping in software requested
2798 * - software time stamp available and wanted
2799 * - hardware time stamps available and wanted
2800 */
2801 if (sock_flag(sk, flag: SOCK_RCVTSTAMP) ||
2802 (tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2803 (kt && tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2804 (hwtstamps->hwtstamp &&
2805 (tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2806 __sock_recv_timestamp(msg, sk, skb);
2807 else
2808 sock_write_timestamp(sk, kt);
2809
2810 if (sock_flag(sk, flag: SOCK_WIFI_STATUS) && skb_wifi_acked_valid(skb))
2811 __sock_recv_wifi_status(msg, sk, skb);
2812}
2813
2814void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2815 struct sk_buff *skb);
2816
2817#define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
2818static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2819 struct sk_buff *skb)
2820{
2821#define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \
2822 (1UL << SOCK_RCVTSTAMP) | \
2823 (1UL << SOCK_RCVMARK) | \
2824 (1UL << SOCK_RCVPRIORITY) | \
2825 (1UL << SOCK_TIMESTAMPING_ANY))
2826#define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2827 SOF_TIMESTAMPING_RAW_HARDWARE)
2828
2829 if (READ_ONCE(sk->sk_flags) & FLAGS_RECV_CMSGS)
2830 __sock_recv_cmsgs(msg, sk, skb);
2831 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2832 sock_write_timestamp(sk, kt: skb->tstamp);
2833 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2834 sock_write_timestamp(sk, kt: 0);
2835}
2836
2837void __sock_tx_timestamp(__u32 tsflags, __u8 *tx_flags);
2838
2839/**
2840 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2841 * @sk: socket sending this packet
2842 * @sockc: pointer to socket cmsg cookie to get timestamping info
2843 * @tx_flags: completed with instructions for time stamping
2844 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2845 *
2846 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2847 */
2848static inline void _sock_tx_timestamp(struct sock *sk,
2849 const struct sockcm_cookie *sockc,
2850 __u8 *tx_flags, __u32 *tskey)
2851{
2852 __u32 tsflags = sockc->tsflags;
2853
2854 if (unlikely(tsflags)) {
2855 __sock_tx_timestamp(tsflags, tx_flags);
2856 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2857 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK) {
2858 if (tsflags & SOCKCM_FLAG_TS_OPT_ID)
2859 *tskey = sockc->ts_opt_id;
2860 else
2861 *tskey = atomic_inc_return(v: &sk->sk_tskey) - 1;
2862 }
2863 }
2864}
2865
2866static inline void sock_tx_timestamp(struct sock *sk,
2867 const struct sockcm_cookie *sockc,
2868 __u8 *tx_flags)
2869{
2870 _sock_tx_timestamp(sk, sockc, tx_flags, NULL);
2871}
2872
2873static inline void skb_setup_tx_timestamp(struct sk_buff *skb,
2874 const struct sockcm_cookie *sockc)
2875{
2876 _sock_tx_timestamp(sk: skb->sk, sockc, tx_flags: &skb_shinfo(skb)->tx_flags,
2877 tskey: &skb_shinfo(skb)->tskey);
2878}
2879
2880static inline bool sk_is_inet(const struct sock *sk)
2881{
2882 int family = READ_ONCE(sk->sk_family);
2883
2884 return family == AF_INET || family == AF_INET6;
2885}
2886
2887static inline bool sk_is_tcp(const struct sock *sk)
2888{
2889 return sk_is_inet(sk) &&
2890 sk->sk_type == SOCK_STREAM &&
2891 sk->sk_protocol == IPPROTO_TCP;
2892}
2893
2894static inline bool sk_is_udp(const struct sock *sk)
2895{
2896 return sk_is_inet(sk) &&
2897 sk->sk_type == SOCK_DGRAM &&
2898 sk->sk_protocol == IPPROTO_UDP;
2899}
2900
2901static inline bool sk_is_unix(const struct sock *sk)
2902{
2903 return sk->sk_family == AF_UNIX;
2904}
2905
2906static inline bool sk_is_stream_unix(const struct sock *sk)
2907{
2908 return sk_is_unix(sk) && sk->sk_type == SOCK_STREAM;
2909}
2910
2911static inline bool sk_is_vsock(const struct sock *sk)
2912{
2913 return sk->sk_family == AF_VSOCK;
2914}
2915
2916static inline bool sk_may_scm_recv(const struct sock *sk)
2917{
2918 return (IS_ENABLED(CONFIG_UNIX) && sk->sk_family == AF_UNIX) ||
2919 sk->sk_family == AF_NETLINK ||
2920 (IS_ENABLED(CONFIG_BT) && sk->sk_family == AF_BLUETOOTH);
2921}
2922
2923/**
2924 * sk_eat_skb - Release a skb if it is no longer needed
2925 * @sk: socket to eat this skb from
2926 * @skb: socket buffer to eat
2927 *
2928 * This routine must be called with interrupts disabled or with the socket
2929 * locked so that the sk_buff queue operation is ok.
2930*/
2931static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2932{
2933 __skb_unlink(skb, list: &sk->sk_receive_queue);
2934 __kfree_skb(skb);
2935}
2936
2937static inline bool
2938skb_sk_is_prefetched(struct sk_buff *skb)
2939{
2940#ifdef CONFIG_INET
2941 return skb->destructor == sock_pfree;
2942#else
2943 return false;
2944#endif /* CONFIG_INET */
2945}
2946
2947/* This helper checks if a socket is a full socket,
2948 * ie _not_ a timewait or request socket.
2949 */
2950static inline bool sk_fullsock(const struct sock *sk)
2951{
2952 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2953}
2954
2955static inline bool
2956sk_is_refcounted(struct sock *sk)
2957{
2958 /* Only full sockets have sk->sk_flags. */
2959 return !sk_fullsock(sk) || !sock_flag(sk, flag: SOCK_RCU_FREE);
2960}
2961
2962static inline bool
2963sk_requests_wifi_status(struct sock *sk)
2964{
2965 return sk && sk_fullsock(sk) && sock_flag(sk, flag: SOCK_WIFI_STATUS);
2966}
2967
2968/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2969 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2970 */
2971static inline bool sk_listener(const struct sock *sk)
2972{
2973 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2974}
2975
2976/* This helper checks if a socket is a LISTEN or NEW_SYN_RECV or TIME_WAIT
2977 * TCP SYNACK messages can be attached to LISTEN or NEW_SYN_RECV (depending on SYNCOOKIE)
2978 * TCP RST and ACK can be attached to TIME_WAIT.
2979 */
2980static inline bool sk_listener_or_tw(const struct sock *sk)
2981{
2982 return (1 << READ_ONCE(sk->sk_state)) &
2983 (TCPF_LISTEN | TCPF_NEW_SYN_RECV | TCPF_TIME_WAIT);
2984}
2985
2986void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2987int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2988 int type);
2989
2990bool sk_ns_capable(const struct sock *sk,
2991 struct user_namespace *user_ns, int cap);
2992bool sk_capable(const struct sock *sk, int cap);
2993bool sk_net_capable(const struct sock *sk, int cap);
2994
2995void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2996
2997/* Take into consideration the size of the struct sk_buff overhead in the
2998 * determination of these values, since that is non-constant across
2999 * platforms. This makes socket queueing behavior and performance
3000 * not depend upon such differences.
3001 */
3002#define _SK_MEM_PACKETS 256
3003#define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
3004#define SK_WMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3005#define SK_RMEM_DEFAULT (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
3006
3007extern __u32 sysctl_wmem_max;
3008extern __u32 sysctl_rmem_max;
3009
3010extern __u32 sysctl_wmem_default;
3011extern __u32 sysctl_rmem_default;
3012
3013#define SKB_FRAG_PAGE_ORDER get_order(32768)
3014DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
3015
3016static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
3017{
3018 /* Does this proto have per netns sysctl_wmem ? */
3019 if (proto->sysctl_wmem_offset)
3020 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
3021
3022 return READ_ONCE(*proto->sysctl_wmem);
3023}
3024
3025static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
3026{
3027 /* Does this proto have per netns sysctl_rmem ? */
3028 if (proto->sysctl_rmem_offset)
3029 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
3030
3031 return READ_ONCE(*proto->sysctl_rmem);
3032}
3033
3034/* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
3035 * Some wifi drivers need to tweak it to get more chunks.
3036 * They can use this helper from their ndo_start_xmit()
3037 */
3038static inline void sk_pacing_shift_update(struct sock *sk, int val)
3039{
3040 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
3041 return;
3042 WRITE_ONCE(sk->sk_pacing_shift, val);
3043}
3044
3045/* if a socket is bound to a device, check that the given device
3046 * index is either the same or that the socket is bound to an L3
3047 * master device and the given device index is also enslaved to
3048 * that L3 master
3049 */
3050static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
3051{
3052 int bound_dev_if = READ_ONCE(sk->sk_bound_dev_if);
3053 int mdif;
3054
3055 if (!bound_dev_if || bound_dev_if == dif)
3056 return true;
3057
3058 mdif = l3mdev_master_ifindex_by_index(net: sock_net(sk), ifindex: dif);
3059 if (mdif && mdif == bound_dev_if)
3060 return true;
3061
3062 return false;
3063}
3064
3065void sock_def_readable(struct sock *sk);
3066
3067int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
3068void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
3069int sock_set_timestamping(struct sock *sk, int optname,
3070 struct so_timestamping timestamping);
3071
3072#if defined(CONFIG_CGROUP_BPF)
3073void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op);
3074#else
3075static inline void bpf_skops_tx_timestamping(struct sock *sk, struct sk_buff *skb, int op)
3076{
3077}
3078#endif
3079void sock_no_linger(struct sock *sk);
3080void sock_set_keepalive(struct sock *sk);
3081void sock_set_priority(struct sock *sk, u32 priority);
3082void sock_set_rcvbuf(struct sock *sk, int val);
3083void sock_set_mark(struct sock *sk, u32 val);
3084void sock_set_reuseaddr(struct sock *sk);
3085void sock_set_reuseport(struct sock *sk);
3086void sock_set_sndtimeo(struct sock *sk, s64 secs);
3087
3088int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
3089
3090int sock_get_timeout(long timeo, void *optval, bool old_timeval);
3091int sock_copy_user_timeval(struct __kernel_sock_timeval *tv,
3092 sockptr_t optval, int optlen, bool old_timeval);
3093
3094int sock_ioctl_inout(struct sock *sk, unsigned int cmd,
3095 void __user *arg, void *karg, size_t size);
3096int sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg);
3097static inline bool sk_is_readable(struct sock *sk)
3098{
3099 const struct proto *prot = READ_ONCE(sk->sk_prot);
3100
3101 if (prot->sock_is_readable)
3102 return prot->sock_is_readable(sk);
3103
3104 return false;
3105}
3106#endif /* _SOCK_H */
3107