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