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