1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Definitions for the AF_INET socket handler.
8 *
9 * Version: @(#)sock.h 1.0.4 05/13/93
10 *
11 * Authors: Ross Biro
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * Corey Minyard <wf-rch!minyard@relay.EU.net>
14 * Florian La Roche <flla@stud.uni-sb.de>
15 *
16 * Fixes:
17 * Alan Cox : Volatiles in skbuff pointers. See
18 * skbuff comments. May be overdone,
19 * better to prove they can be removed
20 * than the reverse.
21 * Alan Cox : Added a zapped field for tcp to note
22 * a socket is reset and must stay shut up
23 * Alan Cox : New fields for options
24 * Pauline Middelink : identd support
25 * Alan Cox : Eliminate low level recv/recvfrom
26 * David S. Miller : New socket lookup architecture.
27 * Steve Whitehouse: Default routines for sock_ops
28 * Arnaldo C. Melo : removed net_pinfo, tp_pinfo and made
29 * protinfo be just a void pointer, as the
30 * protocol specific parts were moved to
31 * respective headers and ipv4/v6, etc now
32 * use private slabcaches for its socks
33 * Pedro Hortas : New flags field for socket options
34 */
35 #ifndef _SOCK_H
36 #define _SOCK_H
37
38 #include <linux/hardirq.h>
39 #include <linux/kernel.h>
40 #include <linux/list.h>
41 #include <linux/list_nulls.h>
42 #include <linux/timer.h>
43 #include <linux/cache.h>
44 #include <linux/bitops.h>
45 #include <linux/lockdep.h>
46 #include <linux/netdevice.h>
47 #include <linux/skbuff.h> /* struct sk_buff */
48 #include <linux/mm.h>
49 #include <linux/security.h>
50 #include <linux/slab.h>
51 #include <linux/uaccess.h>
52 #include <linux/page_counter.h>
53 #include <linux/memcontrol.h>
54 #include <linux/static_key.h>
55 #include <linux/sched.h>
56 #include <linux/wait.h>
57 #include <linux/cgroup-defs.h>
58 #include <linux/rbtree.h>
59 #include <linux/filter.h>
60 #include <linux/rculist_nulls.h>
61 #include <linux/poll.h>
62 #include <linux/sockptr.h>
63 #include <linux/indirect_call_wrapper.h>
64 #include <linux/atomic.h>
65 #include <linux/refcount.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 #include <linux/android_vendor.h>
73 #include <linux/android_kabi.h>
74
75 /*
76 * This structure really needs to be cleaned up.
77 * Most of it is for TCP, and not used by any of
78 * the other protocols.
79 */
80
81 /* Define this to get the SOCK_DBG debugging facility. */
82 #define SOCK_DEBUGGING
83 #ifdef SOCK_DEBUGGING
84 #define SOCK_DEBUG(sk, msg...) do { if ((sk) && sock_flag((sk), SOCK_DBG)) \
85 printk(KERN_DEBUG msg); } while (0)
86 #else
87 /* Validate arguments and do nothing */
88 static inline __printf(2, 3)
SOCK_DEBUG(const struct sock * sk,const char * msg,...)89 void SOCK_DEBUG(const struct sock *sk, const char *msg, ...)
90 {
91 }
92 #endif
93
94 /* This is the per-socket lock. The spinlock provides a synchronization
95 * between user contexts and software interrupt processing, whereas the
96 * mini-semaphore synchronizes multiple users amongst themselves.
97 */
98 typedef struct {
99 spinlock_t slock;
100 int owned;
101 wait_queue_head_t wq;
102 /*
103 * We express the mutex-alike socket_lock semantics
104 * to the lock validator by explicitly managing
105 * the slock as a lock variant (in addition to
106 * the slock itself):
107 */
108 #ifdef CONFIG_DEBUG_LOCK_ALLOC
109 struct lockdep_map dep_map;
110 #endif
111 } socket_lock_t;
112
113 struct sock;
114 struct proto;
115 struct net;
116
117 typedef __u32 __bitwise __portpair;
118 typedef __u64 __bitwise __addrpair;
119
120 /**
121 * struct sock_common - minimal network layer representation of sockets
122 * @skc_daddr: Foreign IPv4 addr
123 * @skc_rcv_saddr: Bound local IPv4 addr
124 * @skc_addrpair: 8-byte-aligned __u64 union of @skc_daddr & @skc_rcv_saddr
125 * @skc_hash: hash value used with various protocol lookup tables
126 * @skc_u16hashes: two u16 hash values used by UDP lookup tables
127 * @skc_dport: placeholder for inet_dport/tw_dport
128 * @skc_num: placeholder for inet_num/tw_num
129 * @skc_portpair: __u32 union of @skc_dport & @skc_num
130 * @skc_family: network address family
131 * @skc_state: Connection state
132 * @skc_reuse: %SO_REUSEADDR setting
133 * @skc_reuseport: %SO_REUSEPORT setting
134 * @skc_ipv6only: socket is IPV6 only
135 * @skc_net_refcnt: socket is using net ref counting
136 * @skc_bound_dev_if: bound device index if != 0
137 * @skc_bind_node: bind hash linkage for various protocol lookup tables
138 * @skc_portaddr_node: second hash linkage for UDP/UDP-Lite protocol
139 * @skc_prot: protocol handlers inside a network family
140 * @skc_net: reference to the network namespace of this socket
141 * @skc_v6_daddr: IPV6 destination address
142 * @skc_v6_rcv_saddr: IPV6 source address
143 * @skc_cookie: socket's cookie value
144 * @skc_node: main hash linkage for various protocol lookup tables
145 * @skc_nulls_node: main hash linkage for TCP/UDP/UDP-Lite protocol
146 * @skc_tx_queue_mapping: tx queue number for this connection
147 * @skc_rx_queue_mapping: rx queue number for this connection
148 * @skc_flags: place holder for sk_flags
149 * %SO_LINGER (l_onoff), %SO_BROADCAST, %SO_KEEPALIVE,
150 * %SO_OOBINLINE settings, %SO_TIMESTAMPING settings
151 * @skc_listener: connection request listener socket (aka rsk_listener)
152 * [union with @skc_flags]
153 * @skc_tw_dr: (aka tw_dr) ptr to &struct inet_timewait_death_row
154 * [union with @skc_flags]
155 * @skc_incoming_cpu: record/match cpu processing incoming packets
156 * @skc_rcv_wnd: (aka rsk_rcv_wnd) TCP receive window size (possibly scaled)
157 * [union with @skc_incoming_cpu]
158 * @skc_tw_rcv_nxt: (aka tw_rcv_nxt) TCP window next expected seq number
159 * [union with @skc_incoming_cpu]
160 * @skc_refcnt: reference count
161 *
162 * This is the minimal network layer representation of sockets, the header
163 * for struct sock and struct inet_timewait_sock.
164 */
165 struct sock_common {
166 union {
167 __addrpair skc_addrpair;
168 struct {
169 __be32 skc_daddr;
170 __be32 skc_rcv_saddr;
171 };
172 };
173 union {
174 unsigned int skc_hash;
175 __u16 skc_u16hashes[2];
176 };
177 /* skc_dport && skc_num must be grouped as well */
178 union {
179 __portpair skc_portpair;
180 struct {
181 __be16 skc_dport;
182 __u16 skc_num;
183 };
184 };
185
186 unsigned short skc_family;
187 volatile unsigned char skc_state;
188 unsigned char skc_reuse:4;
189 unsigned char skc_reuseport:1;
190 unsigned char skc_ipv6only:1;
191 unsigned char skc_net_refcnt:1;
192 int skc_bound_dev_if;
193 union {
194 struct hlist_node skc_bind_node;
195 struct hlist_node skc_portaddr_node;
196 };
197 struct proto *skc_prot;
198 possible_net_t skc_net;
199
200 #if IS_ENABLED(CONFIG_IPV6)
201 struct in6_addr skc_v6_daddr;
202 struct in6_addr skc_v6_rcv_saddr;
203 #endif
204
205 atomic64_t skc_cookie;
206
207 /* following fields are padding to force
208 * offset(struct sock, sk_refcnt) == 128 on 64bit arches
209 * assuming IPV6 is enabled. We use this padding differently
210 * for different kind of 'sockets'
211 */
212 union {
213 unsigned long skc_flags;
214 struct sock *skc_listener; /* request_sock */
215 struct inet_timewait_death_row *skc_tw_dr; /* inet_timewait_sock */
216 };
217 /*
218 * fields between dontcopy_begin/dontcopy_end
219 * are not copied in sock_copy()
220 */
221 /* private: */
222 int skc_dontcopy_begin[0];
223 /* public: */
224 union {
225 struct hlist_node skc_node;
226 struct hlist_nulls_node skc_nulls_node;
227 };
228 unsigned short skc_tx_queue_mapping;
229 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
230 unsigned short skc_rx_queue_mapping;
231 #endif
232 union {
233 int skc_incoming_cpu;
234 u32 skc_rcv_wnd;
235 u32 skc_tw_rcv_nxt; /* struct tcp_timewait_sock */
236 };
237
238 refcount_t skc_refcnt;
239 /* private: */
240 int skc_dontcopy_end[0];
241 union {
242 u32 skc_rxhash;
243 u32 skc_window_clamp;
244 u32 skc_tw_snd_nxt; /* struct tcp_timewait_sock */
245 };
246 /* public: */
247 };
248
249 struct bpf_local_storage;
250
251 /**
252 * struct sock - network layer representation of sockets
253 * @__sk_common: shared layout with inet_timewait_sock
254 * @sk_shutdown: mask of %SEND_SHUTDOWN and/or %RCV_SHUTDOWN
255 * @sk_userlocks: %SO_SNDBUF and %SO_RCVBUF settings
256 * @sk_lock: synchronizer
257 * @sk_kern_sock: True if sock is using kernel lock classes
258 * @sk_rcvbuf: size of receive buffer in bytes
259 * @sk_wq: sock wait queue and async head
260 * @sk_rx_dst: receive input route used by early demux
261 * @sk_rx_dst_ifindex: ifindex for @sk_rx_dst
262 * @sk_rx_dst_cookie: cookie for @sk_rx_dst
263 * @sk_dst_cache: destination cache
264 * @sk_dst_pending_confirm: need to confirm neighbour
265 * @sk_policy: flow policy
266 * @sk_rx_skb_cache: cache copy of recently accessed RX skb
267 * @sk_receive_queue: incoming packets
268 * @sk_wmem_alloc: transmit queue bytes committed
269 * @sk_tsq_flags: TCP Small Queues flags
270 * @sk_write_queue: Packet sending queue
271 * @sk_omem_alloc: "o" is "option" or "other"
272 * @sk_wmem_queued: persistent queue size
273 * @sk_forward_alloc: space allocated forward
274 * @sk_napi_id: id of the last napi context to receive data for sk
275 * @sk_ll_usec: usecs to busypoll when there is no data
276 * @sk_allocation: allocation mode
277 * @sk_pacing_rate: Pacing rate (if supported by transport/packet scheduler)
278 * @sk_pacing_status: Pacing status (requested, handled by sch_fq)
279 * @sk_max_pacing_rate: Maximum pacing rate (%SO_MAX_PACING_RATE)
280 * @sk_sndbuf: size of send buffer in bytes
281 * @__sk_flags_offset: empty field used to determine location of bitfield
282 * @sk_padding: unused element for alignment
283 * @sk_no_check_tx: %SO_NO_CHECK setting, set checksum in TX packets
284 * @sk_no_check_rx: allow zero checksum in RX packets
285 * @sk_route_caps: route capabilities (e.g. %NETIF_F_TSO)
286 * @sk_route_nocaps: forbidden route capabilities (e.g NETIF_F_GSO_MASK)
287 * @sk_route_forced_caps: static, forced route capabilities
288 * (set in tcp_init_sock())
289 * @sk_gso_type: GSO type (e.g. %SKB_GSO_TCPV4)
290 * @sk_gso_max_size: Maximum GSO segment size to build
291 * @sk_gso_max_segs: Maximum number of GSO segments
292 * @sk_pacing_shift: scaling factor for TCP Small Queues
293 * @sk_lingertime: %SO_LINGER l_linger setting
294 * @sk_backlog: always used with the per-socket spinlock held
295 * @sk_callback_lock: used with the callbacks in the end of this struct
296 * @sk_error_queue: rarely used
297 * @sk_prot_creator: sk_prot of original sock creator (see ipv6_setsockopt,
298 * IPV6_ADDRFORM for instance)
299 * @sk_err: last error
300 * @sk_err_soft: errors that don't cause failure but are the cause of a
301 * persistent failure not just 'timed out'
302 * @sk_drops: raw/udp drops counter
303 * @sk_ack_backlog: current listen backlog
304 * @sk_max_ack_backlog: listen backlog set in listen()
305 * @sk_uid: user id of owner
306 * @sk_prefer_busy_poll: prefer busypolling over softirq processing
307 * @sk_busy_poll_budget: napi processing budget when busypolling
308 * @sk_priority: %SO_PRIORITY setting
309 * @sk_type: socket type (%SOCK_STREAM, etc)
310 * @sk_protocol: which protocol this socket belongs in this network family
311 * @sk_peer_lock: lock protecting @sk_peer_pid and @sk_peer_cred
312 * @sk_peer_pid: &struct pid for this socket's peer
313 * @sk_peer_cred: %SO_PEERCRED setting
314 * @sk_rcvlowat: %SO_RCVLOWAT setting
315 * @sk_rcvtimeo: %SO_RCVTIMEO setting
316 * @sk_sndtimeo: %SO_SNDTIMEO setting
317 * @sk_txhash: computed flow hash for use on transmit
318 * @sk_filter: socket filtering instructions
319 * @sk_timer: sock cleanup timer
320 * @sk_stamp: time stamp of last packet received
321 * @sk_stamp_seq: lock for accessing sk_stamp on 32 bit architectures only
322 * @sk_tsflags: SO_TIMESTAMPING flags
323 * @sk_bind_phc: SO_TIMESTAMPING bind PHC index of PTP virtual clock
324 * for timestamping
325 * @sk_tskey: counter to disambiguate concurrent tstamp requests
326 * @sk_zckey: counter to order MSG_ZEROCOPY notifications
327 * @sk_socket: Identd and reporting IO signals
328 * @sk_user_data: RPC layer private data. Write-protected by @sk_callback_lock.
329 * @sk_frag: cached page frag
330 * @sk_peek_off: current peek_offset value
331 * @sk_send_head: front of stuff to transmit
332 * @tcp_rtx_queue: TCP re-transmit queue [union with @sk_send_head]
333 * @sk_tx_skb_cache: cache copy of recently accessed TX skb
334 * @sk_security: used by security modules
335 * @sk_mark: generic packet mark
336 * @sk_cgrp_data: cgroup data for this cgroup
337 * @sk_memcg: this socket's memory cgroup association
338 * @sk_write_pending: a write to stream socket waits to start
339 * @sk_state_change: callback to indicate change in the state of the sock
340 * @sk_data_ready: callback to indicate there is data to be processed
341 * @sk_write_space: callback to indicate there is bf sending space available
342 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
343 * @sk_backlog_rcv: callback to process the backlog
344 * @sk_validate_xmit_skb: ptr to an optional validate function
345 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
346 * @sk_reuseport_cb: reuseport group container
347 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
348 * @sk_rcu: used during RCU grace period
349 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
350 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
351 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
352 * @sk_txtime_unused: unused txtime flags
353 */
354 struct sock {
355 /*
356 * Now struct inet_timewait_sock also uses sock_common, so please just
357 * don't add nothing before this first member (__sk_common) --acme
358 */
359 struct sock_common __sk_common;
360 #define sk_node __sk_common.skc_node
361 #define sk_nulls_node __sk_common.skc_nulls_node
362 #define sk_refcnt __sk_common.skc_refcnt
363 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
364 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
365 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
366 #endif
367
368 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
369 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
370 #define sk_hash __sk_common.skc_hash
371 #define sk_portpair __sk_common.skc_portpair
372 #define sk_num __sk_common.skc_num
373 #define sk_dport __sk_common.skc_dport
374 #define sk_addrpair __sk_common.skc_addrpair
375 #define sk_daddr __sk_common.skc_daddr
376 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
377 #define sk_family __sk_common.skc_family
378 #define sk_state __sk_common.skc_state
379 #define sk_reuse __sk_common.skc_reuse
380 #define sk_reuseport __sk_common.skc_reuseport
381 #define sk_ipv6only __sk_common.skc_ipv6only
382 #define sk_net_refcnt __sk_common.skc_net_refcnt
383 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
384 #define sk_bind_node __sk_common.skc_bind_node
385 #define sk_prot __sk_common.skc_prot
386 #define sk_net __sk_common.skc_net
387 #define sk_v6_daddr __sk_common.skc_v6_daddr
388 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
389 #define sk_cookie __sk_common.skc_cookie
390 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
391 #define sk_flags __sk_common.skc_flags
392 #define sk_rxhash __sk_common.skc_rxhash
393
394 socket_lock_t sk_lock;
395 atomic_t sk_drops;
396 int sk_rcvlowat;
397 struct sk_buff_head sk_error_queue;
398 struct sk_buff *sk_rx_skb_cache;
399 struct sk_buff_head sk_receive_queue;
400 /*
401 * The backlog queue is special, it is always used with
402 * the per-socket spinlock held and requires low latency
403 * access. Therefore we special case it's implementation.
404 * Note : rmem_alloc is in this structure to fill a hole
405 * on 64bit arches, not because its logically part of
406 * backlog.
407 */
408 struct {
409 atomic_t rmem_alloc;
410 int len;
411 struct sk_buff *head;
412 struct sk_buff *tail;
413 } sk_backlog;
414 #define sk_rmem_alloc sk_backlog.rmem_alloc
415
416 int sk_forward_alloc;
417 #ifdef CONFIG_NET_RX_BUSY_POLL
418 unsigned int sk_ll_usec;
419 /* ===== mostly read cache line ===== */
420 unsigned int sk_napi_id;
421 #endif
422 int sk_rcvbuf;
423
424 struct sk_filter __rcu *sk_filter;
425 union {
426 struct socket_wq __rcu *sk_wq;
427 /* private: */
428 struct socket_wq *sk_wq_raw;
429 /* public: */
430 };
431 #ifdef CONFIG_XFRM
432 struct xfrm_policy __rcu *sk_policy[2];
433 #endif
434 struct dst_entry __rcu *sk_rx_dst;
435 int sk_rx_dst_ifindex;
436 u32 sk_rx_dst_cookie;
437
438 struct dst_entry __rcu *sk_dst_cache;
439 atomic_t sk_omem_alloc;
440 int sk_sndbuf;
441
442 /* ===== cache line for TX ===== */
443 int sk_wmem_queued;
444 refcount_t sk_wmem_alloc;
445 unsigned long sk_tsq_flags;
446 union {
447 struct sk_buff *sk_send_head;
448 struct rb_root tcp_rtx_queue;
449 };
450 struct sk_buff *sk_tx_skb_cache;
451 struct sk_buff_head sk_write_queue;
452 __s32 sk_peek_off;
453 int sk_write_pending;
454 __u32 sk_dst_pending_confirm;
455 u32 sk_pacing_status; /* see enum sk_pacing */
456 long sk_sndtimeo;
457 struct timer_list sk_timer;
458 __u32 sk_priority;
459 __u32 sk_mark;
460 unsigned long sk_pacing_rate; /* bytes per second */
461 unsigned long sk_max_pacing_rate;
462 struct page_frag sk_frag;
463 netdev_features_t sk_route_caps;
464 netdev_features_t sk_route_nocaps;
465 netdev_features_t sk_route_forced_caps;
466 int sk_gso_type;
467 unsigned int sk_gso_max_size;
468 gfp_t sk_allocation;
469 __u32 sk_txhash;
470
471 /*
472 * Because of non atomicity rules, all
473 * changes are protected by socket lock.
474 */
475 u8 sk_padding : 1,
476 sk_kern_sock : 1,
477 sk_no_check_tx : 1,
478 sk_no_check_rx : 1,
479 sk_userlocks : 4;
480 u8 sk_pacing_shift;
481 u16 sk_type;
482 u16 sk_protocol;
483 u16 sk_gso_max_segs;
484 unsigned long sk_lingertime;
485 struct proto *sk_prot_creator;
486 rwlock_t sk_callback_lock;
487 int sk_err,
488 sk_err_soft;
489 u32 sk_ack_backlog;
490 u32 sk_max_ack_backlog;
491 kuid_t sk_uid;
492 #ifdef CONFIG_NET_RX_BUSY_POLL
493 u8 sk_prefer_busy_poll;
494 u16 sk_busy_poll_budget;
495 #endif
496 spinlock_t sk_peer_lock;
497 struct pid *sk_peer_pid;
498 const struct cred *sk_peer_cred;
499
500 long sk_rcvtimeo;
501 ktime_t sk_stamp;
502 #if BITS_PER_LONG==32
503 seqlock_t sk_stamp_seq;
504 #endif
505 u16 sk_tsflags;
506 int sk_bind_phc;
507 u8 sk_shutdown;
508 atomic_t sk_tskey;
509 atomic_t sk_zckey;
510
511 u8 sk_clockid;
512 u8 sk_txtime_deadline_mode : 1,
513 sk_txtime_report_errors : 1,
514 sk_txtime_unused : 6;
515
516 struct socket *sk_socket;
517 void *sk_user_data;
518 #ifdef CONFIG_SECURITY
519 void *sk_security;
520 #endif
521 struct sock_cgroup_data sk_cgrp_data;
522 struct mem_cgroup *sk_memcg;
523 void (*sk_state_change)(struct sock *sk);
524 void (*sk_data_ready)(struct sock *sk);
525 void (*sk_write_space)(struct sock *sk);
526 void (*sk_error_report)(struct sock *sk);
527 int (*sk_backlog_rcv)(struct sock *sk,
528 struct sk_buff *skb);
529 #ifdef CONFIG_SOCK_VALIDATE_XMIT
530 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
531 struct net_device *dev,
532 struct sk_buff *skb);
533 #endif
534 void (*sk_destruct)(struct sock *sk);
535 struct sock_reuseport __rcu *sk_reuseport_cb;
536 #ifdef CONFIG_BPF_SYSCALL
537 struct bpf_local_storage __rcu *sk_bpf_storage;
538 #endif
539 struct rcu_head sk_rcu;
540
541 ANDROID_OEM_DATA(1);
542 ANDROID_KABI_RESERVE(1);
543 ANDROID_KABI_RESERVE(2);
544 ANDROID_KABI_RESERVE(3);
545 ANDROID_KABI_RESERVE(4);
546 ANDROID_KABI_RESERVE(5);
547 ANDROID_KABI_RESERVE(6);
548 ANDROID_KABI_RESERVE(7);
549 ANDROID_KABI_RESERVE(8);
550 };
551
552 enum sk_pacing {
553 SK_PACING_NONE = 0,
554 SK_PACING_NEEDED = 1,
555 SK_PACING_FQ = 2,
556 };
557
558 /* flag bits in sk_user_data
559 *
560 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
561 * not be suitable for copying when cloning the socket. For instance,
562 * it can point to a reference counted object. sk_user_data bottom
563 * bit is set if pointer must not be copied.
564 *
565 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
566 * managed/owned by a BPF reuseport array. This bit should be set
567 * when sk_user_data's sk is added to the bpf's reuseport_array.
568 *
569 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
570 * sk_user_data points to psock type. This bit should be set
571 * when sk_user_data is assigned to a psock object.
572 */
573 #define SK_USER_DATA_NOCOPY 1UL
574 #define SK_USER_DATA_BPF 2UL
575 #define SK_USER_DATA_PSOCK 4UL
576 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
577 SK_USER_DATA_PSOCK)
578
579 /**
580 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
581 * @sk: socket
582 */
sk_user_data_is_nocopy(const struct sock * sk)583 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
584 {
585 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
586 }
587
588 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
589
590 /**
591 * __rcu_dereference_sk_user_data_with_flags - return the pointer
592 * only if argument flags all has been set in sk_user_data. Otherwise
593 * return NULL
594 *
595 * @sk: socket
596 * @flags: flag bits
597 */
598 static inline void *
__rcu_dereference_sk_user_data_with_flags(const struct sock * sk,uintptr_t flags)599 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
600 uintptr_t flags)
601 {
602 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
603
604 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
605
606 if ((sk_user_data & flags) == flags)
607 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
608 return NULL;
609 }
610
611 #define rcu_dereference_sk_user_data(sk) \
612 __rcu_dereference_sk_user_data_with_flags(sk, 0)
613 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
614 ({ \
615 uintptr_t __tmp1 = (uintptr_t)(ptr), \
616 __tmp2 = (uintptr_t)(flags); \
617 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
618 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
619 rcu_assign_pointer(__sk_user_data((sk)), \
620 __tmp1 | __tmp2); \
621 })
622 #define rcu_assign_sk_user_data(sk, ptr) \
623 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
624
625 /*
626 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
627 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
628 * on a socket means that the socket will reuse everybody else's port
629 * without looking at the other's sk_reuse value.
630 */
631
632 #define SK_NO_REUSE 0
633 #define SK_CAN_REUSE 1
634 #define SK_FORCE_REUSE 2
635
636 int sk_set_peek_off(struct sock *sk, int val);
637
sk_peek_offset(struct sock * sk,int flags)638 static inline int sk_peek_offset(struct sock *sk, int flags)
639 {
640 if (unlikely(flags & MSG_PEEK)) {
641 return READ_ONCE(sk->sk_peek_off);
642 }
643
644 return 0;
645 }
646
sk_peek_offset_bwd(struct sock * sk,int val)647 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
648 {
649 s32 off = READ_ONCE(sk->sk_peek_off);
650
651 if (unlikely(off >= 0)) {
652 off = max_t(s32, off - val, 0);
653 WRITE_ONCE(sk->sk_peek_off, off);
654 }
655 }
656
sk_peek_offset_fwd(struct sock * sk,int val)657 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
658 {
659 sk_peek_offset_bwd(sk, -val);
660 }
661
662 /*
663 * Hashed lists helper routines
664 */
sk_entry(const struct hlist_node * node)665 static inline struct sock *sk_entry(const struct hlist_node *node)
666 {
667 return hlist_entry(node, struct sock, sk_node);
668 }
669
__sk_head(const struct hlist_head * head)670 static inline struct sock *__sk_head(const struct hlist_head *head)
671 {
672 return hlist_entry(head->first, struct sock, sk_node);
673 }
674
sk_head(const struct hlist_head * head)675 static inline struct sock *sk_head(const struct hlist_head *head)
676 {
677 return hlist_empty(head) ? NULL : __sk_head(head);
678 }
679
__sk_nulls_head(const struct hlist_nulls_head * head)680 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
681 {
682 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
683 }
684
sk_nulls_head(const struct hlist_nulls_head * head)685 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
686 {
687 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
688 }
689
sk_next(const struct sock * sk)690 static inline struct sock *sk_next(const struct sock *sk)
691 {
692 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
693 }
694
sk_nulls_next(const struct sock * sk)695 static inline struct sock *sk_nulls_next(const struct sock *sk)
696 {
697 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
698 hlist_nulls_entry(sk->sk_nulls_node.next,
699 struct sock, sk_nulls_node) :
700 NULL;
701 }
702
sk_unhashed(const struct sock * sk)703 static inline bool sk_unhashed(const struct sock *sk)
704 {
705 return hlist_unhashed(&sk->sk_node);
706 }
707
sk_hashed(const struct sock * sk)708 static inline bool sk_hashed(const struct sock *sk)
709 {
710 return !sk_unhashed(sk);
711 }
712
sk_node_init(struct hlist_node * node)713 static inline void sk_node_init(struct hlist_node *node)
714 {
715 node->pprev = NULL;
716 }
717
sk_nulls_node_init(struct hlist_nulls_node * node)718 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
719 {
720 node->pprev = NULL;
721 }
722
__sk_del_node(struct sock * sk)723 static inline void __sk_del_node(struct sock *sk)
724 {
725 __hlist_del(&sk->sk_node);
726 }
727
728 /* NB: equivalent to hlist_del_init_rcu */
__sk_del_node_init(struct sock * sk)729 static inline bool __sk_del_node_init(struct sock *sk)
730 {
731 if (sk_hashed(sk)) {
732 __sk_del_node(sk);
733 sk_node_init(&sk->sk_node);
734 return true;
735 }
736 return false;
737 }
738
739 /* Grab socket reference count. This operation is valid only
740 when sk is ALREADY grabbed f.e. it is found in hash table
741 or a list and the lookup is made under lock preventing hash table
742 modifications.
743 */
744
sock_hold(struct sock * sk)745 static __always_inline void sock_hold(struct sock *sk)
746 {
747 refcount_inc(&sk->sk_refcnt);
748 }
749
750 /* Ungrab socket in the context, which assumes that socket refcnt
751 cannot hit zero, f.e. it is true in context of any socketcall.
752 */
__sock_put(struct sock * sk)753 static __always_inline void __sock_put(struct sock *sk)
754 {
755 refcount_dec(&sk->sk_refcnt);
756 }
757
sk_del_node_init(struct sock * sk)758 static inline bool sk_del_node_init(struct sock *sk)
759 {
760 bool rc = __sk_del_node_init(sk);
761
762 if (rc) {
763 /* paranoid for a while -acme */
764 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
765 __sock_put(sk);
766 }
767 return rc;
768 }
769 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
770
__sk_nulls_del_node_init_rcu(struct sock * sk)771 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
772 {
773 if (sk_hashed(sk)) {
774 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
775 return true;
776 }
777 return false;
778 }
779
sk_nulls_del_node_init_rcu(struct sock * sk)780 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
781 {
782 bool rc = __sk_nulls_del_node_init_rcu(sk);
783
784 if (rc) {
785 /* paranoid for a while -acme */
786 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
787 __sock_put(sk);
788 }
789 return rc;
790 }
791
__sk_add_node(struct sock * sk,struct hlist_head * list)792 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
793 {
794 hlist_add_head(&sk->sk_node, list);
795 }
796
sk_add_node(struct sock * sk,struct hlist_head * list)797 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
798 {
799 sock_hold(sk);
800 __sk_add_node(sk, list);
801 }
802
sk_add_node_rcu(struct sock * sk,struct hlist_head * list)803 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
804 {
805 sock_hold(sk);
806 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
807 sk->sk_family == AF_INET6)
808 hlist_add_tail_rcu(&sk->sk_node, list);
809 else
810 hlist_add_head_rcu(&sk->sk_node, list);
811 }
812
sk_add_node_tail_rcu(struct sock * sk,struct hlist_head * list)813 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
814 {
815 sock_hold(sk);
816 hlist_add_tail_rcu(&sk->sk_node, list);
817 }
818
__sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)819 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
820 {
821 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
822 }
823
__sk_nulls_add_node_tail_rcu(struct sock * sk,struct hlist_nulls_head * list)824 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
825 {
826 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
827 }
828
sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)829 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
830 {
831 sock_hold(sk);
832 __sk_nulls_add_node_rcu(sk, list);
833 }
834
__sk_del_bind_node(struct sock * sk)835 static inline void __sk_del_bind_node(struct sock *sk)
836 {
837 __hlist_del(&sk->sk_bind_node);
838 }
839
sk_add_bind_node(struct sock * sk,struct hlist_head * list)840 static inline void sk_add_bind_node(struct sock *sk,
841 struct hlist_head *list)
842 {
843 hlist_add_head(&sk->sk_bind_node, list);
844 }
845
846 #define sk_for_each(__sk, list) \
847 hlist_for_each_entry(__sk, list, sk_node)
848 #define sk_for_each_rcu(__sk, list) \
849 hlist_for_each_entry_rcu(__sk, list, sk_node)
850 #define sk_nulls_for_each(__sk, node, list) \
851 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
852 #define sk_nulls_for_each_rcu(__sk, node, list) \
853 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
854 #define sk_for_each_from(__sk) \
855 hlist_for_each_entry_from(__sk, sk_node)
856 #define sk_nulls_for_each_from(__sk, node) \
857 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
858 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
859 #define sk_for_each_safe(__sk, tmp, list) \
860 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
861 #define sk_for_each_bound(__sk, list) \
862 hlist_for_each_entry(__sk, list, sk_bind_node)
863
864 /**
865 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
866 * @tpos: the type * to use as a loop cursor.
867 * @pos: the &struct hlist_node to use as a loop cursor.
868 * @head: the head for your list.
869 * @offset: offset of hlist_node within the struct.
870 *
871 */
872 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
873 for (pos = rcu_dereference(hlist_first_rcu(head)); \
874 pos != NULL && \
875 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
876 pos = rcu_dereference(hlist_next_rcu(pos)))
877
sk_user_ns(struct sock * sk)878 static inline struct user_namespace *sk_user_ns(struct sock *sk)
879 {
880 /* Careful only use this in a context where these parameters
881 * can not change and must all be valid, such as recvmsg from
882 * userspace.
883 */
884 return sk->sk_socket->file->f_cred->user_ns;
885 }
886
887 /* Sock flags */
888 enum sock_flags {
889 SOCK_DEAD,
890 SOCK_DONE,
891 SOCK_URGINLINE,
892 SOCK_KEEPOPEN,
893 SOCK_LINGER,
894 SOCK_DESTROY,
895 SOCK_BROADCAST,
896 SOCK_TIMESTAMP,
897 SOCK_ZAPPED,
898 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
899 SOCK_DBG, /* %SO_DEBUG setting */
900 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
901 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
902 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
903 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
904 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
905 SOCK_FASYNC, /* fasync() active */
906 SOCK_RXQ_OVFL,
907 SOCK_ZEROCOPY, /* buffers from userspace */
908 SOCK_WIFI_STATUS, /* push wifi status to userspace */
909 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
910 * Will use last 4 bytes of packet sent from
911 * user-space instead.
912 */
913 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
914 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
915 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
916 SOCK_TXTIME,
917 SOCK_XDP, /* XDP is attached */
918 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
919 };
920
921 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
922
sock_copy_flags(struct sock * nsk,struct sock * osk)923 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
924 {
925 nsk->sk_flags = osk->sk_flags;
926 }
927
sock_set_flag(struct sock * sk,enum sock_flags flag)928 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
929 {
930 __set_bit(flag, &sk->sk_flags);
931 }
932
sock_reset_flag(struct sock * sk,enum sock_flags flag)933 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
934 {
935 __clear_bit(flag, &sk->sk_flags);
936 }
937
sock_valbool_flag(struct sock * sk,enum sock_flags bit,int valbool)938 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
939 int valbool)
940 {
941 if (valbool)
942 sock_set_flag(sk, bit);
943 else
944 sock_reset_flag(sk, bit);
945 }
946
sock_flag(const struct sock * sk,enum sock_flags flag)947 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
948 {
949 return test_bit(flag, &sk->sk_flags);
950 }
951
952 #ifdef CONFIG_NET
953 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
sk_memalloc_socks(void)954 static inline int sk_memalloc_socks(void)
955 {
956 return static_branch_unlikely(&memalloc_socks_key);
957 }
958
959 void __receive_sock(struct file *file);
960 #else
961
sk_memalloc_socks(void)962 static inline int sk_memalloc_socks(void)
963 {
964 return 0;
965 }
966
__receive_sock(struct file * file)967 static inline void __receive_sock(struct file *file)
968 { }
969 #endif
970
sk_gfp_mask(const struct sock * sk,gfp_t gfp_mask)971 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
972 {
973 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
974 }
975
sk_acceptq_removed(struct sock * sk)976 static inline void sk_acceptq_removed(struct sock *sk)
977 {
978 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
979 }
980
sk_acceptq_added(struct sock * sk)981 static inline void sk_acceptq_added(struct sock *sk)
982 {
983 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
984 }
985
986 /* Note: If you think the test should be:
987 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
988 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
989 */
sk_acceptq_is_full(const struct sock * sk)990 static inline bool sk_acceptq_is_full(const struct sock *sk)
991 {
992 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
993 }
994
995 /*
996 * Compute minimal free write space needed to queue new packets.
997 */
sk_stream_min_wspace(const struct sock * sk)998 static inline int sk_stream_min_wspace(const struct sock *sk)
999 {
1000 return READ_ONCE(sk->sk_wmem_queued) >> 1;
1001 }
1002
sk_stream_wspace(const struct sock * sk)1003 static inline int sk_stream_wspace(const struct sock *sk)
1004 {
1005 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1006 }
1007
sk_wmem_queued_add(struct sock * sk,int val)1008 static inline void sk_wmem_queued_add(struct sock *sk, int val)
1009 {
1010 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1011 }
1012
1013 void sk_stream_write_space(struct sock *sk);
1014
1015 /* OOB backlog add */
__sk_add_backlog(struct sock * sk,struct sk_buff * skb)1016 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1017 {
1018 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1019 skb_dst_force(skb);
1020
1021 if (!sk->sk_backlog.tail)
1022 WRITE_ONCE(sk->sk_backlog.head, skb);
1023 else
1024 sk->sk_backlog.tail->next = skb;
1025
1026 WRITE_ONCE(sk->sk_backlog.tail, skb);
1027 skb->next = NULL;
1028 }
1029
1030 /*
1031 * Take into account size of receive queue and backlog queue
1032 * Do not take into account this skb truesize,
1033 * to allow even a single big packet to come.
1034 */
sk_rcvqueues_full(const struct sock * sk,unsigned int limit)1035 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1036 {
1037 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1038
1039 return qsize > limit;
1040 }
1041
1042 /* The per-socket spinlock must be held here. */
sk_add_backlog(struct sock * sk,struct sk_buff * skb,unsigned int limit)1043 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1044 unsigned int limit)
1045 {
1046 if (sk_rcvqueues_full(sk, limit))
1047 return -ENOBUFS;
1048
1049 /*
1050 * If the skb was allocated from pfmemalloc reserves, only
1051 * allow SOCK_MEMALLOC sockets to use it as this socket is
1052 * helping free memory
1053 */
1054 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1055 return -ENOMEM;
1056
1057 __sk_add_backlog(sk, skb);
1058 sk->sk_backlog.len += skb->truesize;
1059 return 0;
1060 }
1061
1062 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1063
sk_backlog_rcv(struct sock * sk,struct sk_buff * skb)1064 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1065 {
1066 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1067 return __sk_backlog_rcv(sk, skb);
1068
1069 return sk->sk_backlog_rcv(sk, skb);
1070 }
1071
sk_incoming_cpu_update(struct sock * sk)1072 static inline void sk_incoming_cpu_update(struct sock *sk)
1073 {
1074 int cpu = raw_smp_processor_id();
1075
1076 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1077 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1078 }
1079
sock_rps_record_flow_hash(__u32 hash)1080 static inline void sock_rps_record_flow_hash(__u32 hash)
1081 {
1082 #ifdef CONFIG_RPS
1083 struct rps_sock_flow_table *sock_flow_table;
1084
1085 rcu_read_lock();
1086 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1087 rps_record_sock_flow(sock_flow_table, hash);
1088 rcu_read_unlock();
1089 #endif
1090 }
1091
sock_rps_record_flow(const struct sock * sk)1092 static inline void sock_rps_record_flow(const struct sock *sk)
1093 {
1094 #ifdef CONFIG_RPS
1095 if (static_branch_unlikely(&rfs_needed)) {
1096 /* Reading sk->sk_rxhash might incur an expensive cache line
1097 * miss.
1098 *
1099 * TCP_ESTABLISHED does cover almost all states where RFS
1100 * might be useful, and is cheaper [1] than testing :
1101 * IPv4: inet_sk(sk)->inet_daddr
1102 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1103 * OR an additional socket flag
1104 * [1] : sk_state and sk_prot are in the same cache line.
1105 */
1106 if (sk->sk_state == TCP_ESTABLISHED) {
1107 /* This READ_ONCE() is paired with the WRITE_ONCE()
1108 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
1109 */
1110 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1111 }
1112 }
1113 #endif
1114 }
1115
sock_rps_save_rxhash(struct sock * sk,const struct sk_buff * skb)1116 static inline void sock_rps_save_rxhash(struct sock *sk,
1117 const struct sk_buff *skb)
1118 {
1119 #ifdef CONFIG_RPS
1120 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1121 * here, and another one in sock_rps_record_flow().
1122 */
1123 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1124 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1125 #endif
1126 }
1127
sock_rps_reset_rxhash(struct sock * sk)1128 static inline void sock_rps_reset_rxhash(struct sock *sk)
1129 {
1130 #ifdef CONFIG_RPS
1131 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1132 WRITE_ONCE(sk->sk_rxhash, 0);
1133 #endif
1134 }
1135
1136 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1137 ({ int __rc; \
1138 release_sock(__sk); \
1139 __rc = __condition; \
1140 if (!__rc) { \
1141 *(__timeo) = wait_woken(__wait, \
1142 TASK_INTERRUPTIBLE, \
1143 *(__timeo)); \
1144 } \
1145 sched_annotate_sleep(); \
1146 lock_sock(__sk); \
1147 __rc = __condition; \
1148 __rc; \
1149 })
1150
1151 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1152 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1153 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1154 int sk_stream_error(struct sock *sk, int flags, int err);
1155 void sk_stream_kill_queues(struct sock *sk);
1156 void sk_set_memalloc(struct sock *sk);
1157 void sk_clear_memalloc(struct sock *sk);
1158
1159 void __sk_flush_backlog(struct sock *sk);
1160
sk_flush_backlog(struct sock * sk)1161 static inline bool sk_flush_backlog(struct sock *sk)
1162 {
1163 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1164 __sk_flush_backlog(sk);
1165 return true;
1166 }
1167 return false;
1168 }
1169
1170 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1171
1172 struct request_sock_ops;
1173 struct timewait_sock_ops;
1174 struct inet_hashinfo;
1175 struct raw_hashinfo;
1176 struct smc_hashinfo;
1177 struct module;
1178 struct sk_psock;
1179
1180 /*
1181 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1182 * un-modified. Special care is taken when initializing object to zero.
1183 */
sk_prot_clear_nulls(struct sock * sk,int size)1184 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1185 {
1186 if (offsetof(struct sock, sk_node.next) != 0)
1187 memset(sk, 0, offsetof(struct sock, sk_node.next));
1188 memset(&sk->sk_node.pprev, 0,
1189 size - offsetof(struct sock, sk_node.pprev));
1190 }
1191
1192 /* Networking protocol blocks we attach to sockets.
1193 * socket layer -> transport layer interface
1194 */
1195 struct proto {
1196 void (*close)(struct sock *sk,
1197 long timeout);
1198 int (*pre_connect)(struct sock *sk,
1199 struct sockaddr *uaddr,
1200 int addr_len);
1201 int (*connect)(struct sock *sk,
1202 struct sockaddr *uaddr,
1203 int addr_len);
1204 int (*disconnect)(struct sock *sk, int flags);
1205
1206 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1207 bool kern);
1208
1209 int (*ioctl)(struct sock *sk, int cmd,
1210 unsigned long arg);
1211 int (*init)(struct sock *sk);
1212 void (*destroy)(struct sock *sk);
1213 void (*shutdown)(struct sock *sk, int how);
1214 int (*setsockopt)(struct sock *sk, int level,
1215 int optname, sockptr_t optval,
1216 unsigned int optlen);
1217 int (*getsockopt)(struct sock *sk, int level,
1218 int optname, char __user *optval,
1219 int __user *option);
1220 void (*keepalive)(struct sock *sk, int valbool);
1221 #ifdef CONFIG_COMPAT
1222 int (*compat_ioctl)(struct sock *sk,
1223 unsigned int cmd, unsigned long arg);
1224 #endif
1225 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1226 size_t len);
1227 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1228 size_t len, int noblock, int flags,
1229 int *addr_len);
1230 int (*sendpage)(struct sock *sk, struct page *page,
1231 int offset, size_t size, int flags);
1232 int (*bind)(struct sock *sk,
1233 struct sockaddr *addr, int addr_len);
1234 int (*bind_add)(struct sock *sk,
1235 struct sockaddr *addr, int addr_len);
1236
1237 int (*backlog_rcv) (struct sock *sk,
1238 struct sk_buff *skb);
1239 bool (*bpf_bypass_getsockopt)(int level,
1240 int optname);
1241
1242 void (*release_cb)(struct sock *sk);
1243
1244 /* Keeping track of sk's, looking them up, and port selection methods. */
1245 int (*hash)(struct sock *sk);
1246 void (*unhash)(struct sock *sk);
1247 void (*rehash)(struct sock *sk);
1248 int (*get_port)(struct sock *sk, unsigned short snum);
1249 #ifdef CONFIG_BPF_SYSCALL
1250 int (*psock_update_sk_prot)(struct sock *sk,
1251 struct sk_psock *psock,
1252 bool restore);
1253 #endif
1254
1255 /* Keeping track of sockets in use */
1256 #ifdef CONFIG_PROC_FS
1257 unsigned int inuse_idx;
1258 #endif
1259
1260 bool (*stream_memory_free)(const struct sock *sk, int wake);
1261 bool (*sock_is_readable)(struct sock *sk);
1262 /* Memory pressure */
1263 void (*enter_memory_pressure)(struct sock *sk);
1264 void (*leave_memory_pressure)(struct sock *sk);
1265 atomic_long_t *memory_allocated; /* Current allocated memory. */
1266 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1267 /*
1268 * Pressure flag: try to collapse.
1269 * Technical note: it is used by multiple contexts non atomically.
1270 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1271 * All the __sk_mem_schedule() is of this nature: accounting
1272 * is strict, actions are advisory and have some latency.
1273 */
1274 unsigned long *memory_pressure;
1275 long *sysctl_mem;
1276
1277 int *sysctl_wmem;
1278 int *sysctl_rmem;
1279 u32 sysctl_wmem_offset;
1280 u32 sysctl_rmem_offset;
1281
1282 int max_header;
1283 bool no_autobind;
1284
1285 struct kmem_cache *slab;
1286 unsigned int obj_size;
1287 slab_flags_t slab_flags;
1288 unsigned int useroffset; /* Usercopy region offset */
1289 unsigned int usersize; /* Usercopy region size */
1290
1291 unsigned int __percpu *orphan_count;
1292
1293 struct request_sock_ops *rsk_prot;
1294 struct timewait_sock_ops *twsk_prot;
1295
1296 union {
1297 struct inet_hashinfo *hashinfo;
1298 struct udp_table *udp_table;
1299 struct raw_hashinfo *raw_hash;
1300 struct smc_hashinfo *smc_hash;
1301 } h;
1302
1303 struct module *owner;
1304
1305 char name[32];
1306
1307 struct list_head node;
1308 #ifdef SOCK_REFCNT_DEBUG
1309 atomic_t socks;
1310 #endif
1311 int (*diag_destroy)(struct sock *sk, int err);
1312 } __randomize_layout;
1313
1314 int proto_register(struct proto *prot, int alloc_slab);
1315 void proto_unregister(struct proto *prot);
1316 int sock_load_diag_module(int family, int protocol);
1317
1318 #ifdef SOCK_REFCNT_DEBUG
sk_refcnt_debug_inc(struct sock * sk)1319 static inline void sk_refcnt_debug_inc(struct sock *sk)
1320 {
1321 atomic_inc(&sk->sk_prot->socks);
1322 }
1323
sk_refcnt_debug_dec(struct sock * sk)1324 static inline void sk_refcnt_debug_dec(struct sock *sk)
1325 {
1326 atomic_dec(&sk->sk_prot->socks);
1327 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1328 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1329 }
1330
sk_refcnt_debug_release(const struct sock * sk)1331 static inline void sk_refcnt_debug_release(const struct sock *sk)
1332 {
1333 if (refcount_read(&sk->sk_refcnt) != 1)
1334 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1335 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1336 }
1337 #else /* SOCK_REFCNT_DEBUG */
1338 #define sk_refcnt_debug_inc(sk) do { } while (0)
1339 #define sk_refcnt_debug_dec(sk) do { } while (0)
1340 #define sk_refcnt_debug_release(sk) do { } while (0)
1341 #endif /* SOCK_REFCNT_DEBUG */
1342
1343 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1344
__sk_stream_memory_free(const struct sock * sk,int wake)1345 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1346 {
1347 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1348 return false;
1349
1350 #ifdef CONFIG_INET
1351 return sk->sk_prot->stream_memory_free ?
1352 INDIRECT_CALL_1(sk->sk_prot->stream_memory_free,
1353 tcp_stream_memory_free,
1354 sk, wake) : true;
1355 #else
1356 return sk->sk_prot->stream_memory_free ?
1357 sk->sk_prot->stream_memory_free(sk, wake) : true;
1358 #endif
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
sk_has_memory_pressure(const struct sock * sk)1388 static inline bool sk_has_memory_pressure(const struct sock *sk)
1389 {
1390 return sk->sk_prot->memory_pressure != NULL;
1391 }
1392
sk_under_global_memory_pressure(const struct sock * sk)1393 static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1394 {
1395 return sk->sk_prot->memory_pressure &&
1396 !!READ_ONCE(*sk->sk_prot->memory_pressure);
1397 }
1398
sk_under_memory_pressure(const struct sock * sk)1399 static inline bool sk_under_memory_pressure(const struct sock *sk)
1400 {
1401 if (!sk->sk_prot->memory_pressure)
1402 return false;
1403
1404 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1405 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1406 return true;
1407
1408 return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1409 }
1410
1411 static inline long
sk_memory_allocated(const struct sock * sk)1412 sk_memory_allocated(const struct sock *sk)
1413 {
1414 return atomic_long_read(sk->sk_prot->memory_allocated);
1415 }
1416
1417 static inline long
sk_memory_allocated_add(struct sock * sk,int amt)1418 sk_memory_allocated_add(struct sock *sk, int amt)
1419 {
1420 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1421 }
1422
1423 static inline void
sk_memory_allocated_sub(struct sock * sk,int amt)1424 sk_memory_allocated_sub(struct sock *sk, int amt)
1425 {
1426 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1427 }
1428
1429 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1430
sk_sockets_allocated_dec(struct sock * sk)1431 static inline void sk_sockets_allocated_dec(struct sock *sk)
1432 {
1433 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1434 SK_ALLOC_PERCPU_COUNTER_BATCH);
1435 }
1436
sk_sockets_allocated_inc(struct sock * sk)1437 static inline void sk_sockets_allocated_inc(struct sock *sk)
1438 {
1439 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1440 SK_ALLOC_PERCPU_COUNTER_BATCH);
1441 }
1442
1443 static inline u64
sk_sockets_allocated_read_positive(struct sock * sk)1444 sk_sockets_allocated_read_positive(struct sock *sk)
1445 {
1446 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1447 }
1448
1449 static inline int
proto_sockets_allocated_sum_positive(struct proto * prot)1450 proto_sockets_allocated_sum_positive(struct proto *prot)
1451 {
1452 return percpu_counter_sum_positive(prot->sockets_allocated);
1453 }
1454
1455 static inline long
proto_memory_allocated(struct proto * prot)1456 proto_memory_allocated(struct proto *prot)
1457 {
1458 return atomic_long_read(prot->memory_allocated);
1459 }
1460
1461 static inline bool
proto_memory_pressure(struct proto * prot)1462 proto_memory_pressure(struct proto *prot)
1463 {
1464 if (!prot->memory_pressure)
1465 return false;
1466 return !!READ_ONCE(*prot->memory_pressure);
1467 }
1468
1469
1470 #ifdef CONFIG_PROC_FS
1471 /* Called with local bh disabled */
1472 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1473 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1474 int sock_inuse_get(struct net *net);
1475 #else
sock_prot_inuse_add(struct net * net,struct proto * prot,int inc)1476 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1477 int inc)
1478 {
1479 }
1480 #endif
1481
1482
1483 /* With per-bucket locks this operation is not-atomic, so that
1484 * this version is not worse.
1485 */
__sk_prot_rehash(struct sock * sk)1486 static inline int __sk_prot_rehash(struct sock *sk)
1487 {
1488 sk->sk_prot->unhash(sk);
1489 return sk->sk_prot->hash(sk);
1490 }
1491
1492 /* About 10 seconds */
1493 #define SOCK_DESTROY_TIME (10*HZ)
1494
1495 /* Sockets 0-1023 can't be bound to unless you are superuser */
1496 #define PROT_SOCK 1024
1497
1498 #define SHUTDOWN_MASK 3
1499 #define RCV_SHUTDOWN 1
1500 #define SEND_SHUTDOWN 2
1501
1502 #define SOCK_BINDADDR_LOCK 4
1503 #define SOCK_BINDPORT_LOCK 8
1504
1505 struct socket_alloc {
1506 struct socket socket;
1507 struct inode vfs_inode;
1508 };
1509
SOCKET_I(struct inode * inode)1510 static inline struct socket *SOCKET_I(struct inode *inode)
1511 {
1512 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1513 }
1514
SOCK_INODE(struct socket * socket)1515 static inline struct inode *SOCK_INODE(struct socket *socket)
1516 {
1517 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1518 }
1519
1520 /*
1521 * Functions for memory accounting
1522 */
1523 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1524 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1525 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1526 void __sk_mem_reclaim(struct sock *sk, int amount);
1527
1528 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1529 * do not necessarily have 16x time more memory than 4KB ones.
1530 */
1531 #define SK_MEM_QUANTUM 4096
1532 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1533 #define SK_MEM_SEND 0
1534 #define SK_MEM_RECV 1
1535
1536 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
sk_prot_mem_limits(const struct sock * sk,int index)1537 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1538 {
1539 long val = READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1540
1541 #if PAGE_SIZE > SK_MEM_QUANTUM
1542 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1543 #elif PAGE_SIZE < SK_MEM_QUANTUM
1544 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1545 #endif
1546 return val;
1547 }
1548
sk_mem_pages(int amt)1549 static inline int sk_mem_pages(int amt)
1550 {
1551 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1552 }
1553
sk_has_account(struct sock * sk)1554 static inline bool sk_has_account(struct sock *sk)
1555 {
1556 /* return true if protocol supports memory accounting */
1557 return !!sk->sk_prot->memory_allocated;
1558 }
1559
sk_wmem_schedule(struct sock * sk,int size)1560 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1561 {
1562 int delta;
1563
1564 if (!sk_has_account(sk))
1565 return true;
1566 delta = size - sk->sk_forward_alloc;
1567 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1568 }
1569
1570 static inline bool
sk_rmem_schedule(struct sock * sk,struct sk_buff * skb,int size)1571 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1572 {
1573 int delta;
1574
1575 if (!sk_has_account(sk))
1576 return true;
1577 delta = size - sk->sk_forward_alloc;
1578 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1579 skb_pfmemalloc(skb);
1580 }
1581
sk_mem_reclaim(struct sock * sk)1582 static inline void sk_mem_reclaim(struct sock *sk)
1583 {
1584 if (!sk_has_account(sk))
1585 return;
1586 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1587 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1588 }
1589
sk_mem_reclaim_partial(struct sock * sk)1590 static inline void sk_mem_reclaim_partial(struct sock *sk)
1591 {
1592 if (!sk_has_account(sk))
1593 return;
1594 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1595 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1596 }
1597
sk_mem_charge(struct sock * sk,int size)1598 static inline void sk_mem_charge(struct sock *sk, int size)
1599 {
1600 if (!sk_has_account(sk))
1601 return;
1602 sk->sk_forward_alloc -= size;
1603 }
1604
sk_mem_uncharge(struct sock * sk,int size)1605 static inline void sk_mem_uncharge(struct sock *sk, int size)
1606 {
1607 if (!sk_has_account(sk))
1608 return;
1609 sk->sk_forward_alloc += size;
1610
1611 /* Avoid a possible overflow.
1612 * TCP send queues can make this happen, if sk_mem_reclaim()
1613 * is not called and more than 2 GBytes are released at once.
1614 *
1615 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1616 * no need to hold that much forward allocation anyway.
1617 */
1618 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1619 __sk_mem_reclaim(sk, 1 << 20);
1620 }
1621
1622 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
sk_wmem_free_skb(struct sock * sk,struct sk_buff * skb)1623 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1624 {
1625 sk_wmem_queued_add(sk, -skb->truesize);
1626 sk_mem_uncharge(sk, skb->truesize);
1627 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1628 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1629 skb_ext_reset(skb);
1630 skb_zcopy_clear(skb, true);
1631 sk->sk_tx_skb_cache = skb;
1632 return;
1633 }
1634 __kfree_skb(skb);
1635 }
1636
sock_release_ownership(struct sock * sk)1637 static inline void sock_release_ownership(struct sock *sk)
1638 {
1639 if (sk->sk_lock.owned) {
1640 sk->sk_lock.owned = 0;
1641
1642 /* The sk_lock has mutex_unlock() semantics: */
1643 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1644 }
1645 }
1646
1647 /*
1648 * Macro so as to not evaluate some arguments when
1649 * lockdep is not enabled.
1650 *
1651 * Mark both the sk_lock and the sk_lock.slock as a
1652 * per-address-family lock class.
1653 */
1654 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1655 do { \
1656 sk->sk_lock.owned = 0; \
1657 init_waitqueue_head(&sk->sk_lock.wq); \
1658 spin_lock_init(&(sk)->sk_lock.slock); \
1659 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1660 sizeof((sk)->sk_lock)); \
1661 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1662 (skey), (sname)); \
1663 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1664 } while (0)
1665
lockdep_sock_is_held(const struct sock * sk)1666 static inline bool lockdep_sock_is_held(const struct sock *sk)
1667 {
1668 return lockdep_is_held(&sk->sk_lock) ||
1669 lockdep_is_held(&sk->sk_lock.slock);
1670 }
1671
1672 void lock_sock_nested(struct sock *sk, int subclass);
1673
lock_sock(struct sock * sk)1674 static inline void lock_sock(struct sock *sk)
1675 {
1676 lock_sock_nested(sk, 0);
1677 }
1678
1679 void __lock_sock(struct sock *sk);
1680 void __release_sock(struct sock *sk);
1681 void release_sock(struct sock *sk);
1682
1683 /* BH context may only use the following locking interface. */
1684 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1685 #define bh_lock_sock_nested(__sk) \
1686 spin_lock_nested(&((__sk)->sk_lock.slock), \
1687 SINGLE_DEPTH_NESTING)
1688 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1689
1690 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1691
1692 /**
1693 * lock_sock_fast - fast version of lock_sock
1694 * @sk: socket
1695 *
1696 * This version should be used for very small section, where process wont block
1697 * return false if fast path is taken:
1698 *
1699 * sk_lock.slock locked, owned = 0, BH disabled
1700 *
1701 * return true if slow path is taken:
1702 *
1703 * sk_lock.slock unlocked, owned = 1, BH enabled
1704 */
lock_sock_fast(struct sock * sk)1705 static inline bool lock_sock_fast(struct sock *sk)
1706 {
1707 /* The sk_lock has mutex_lock() semantics here. */
1708 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1709
1710 return __lock_sock_fast(sk);
1711 }
1712
1713 /* fast socket lock variant for caller already holding a [different] socket lock */
lock_sock_fast_nested(struct sock * sk)1714 static inline bool lock_sock_fast_nested(struct sock *sk)
1715 {
1716 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1717
1718 return __lock_sock_fast(sk);
1719 }
1720
1721 /**
1722 * unlock_sock_fast - complement of lock_sock_fast
1723 * @sk: socket
1724 * @slow: slow mode
1725 *
1726 * fast unlock socket for user context.
1727 * If slow mode is on, we call regular release_sock()
1728 */
unlock_sock_fast(struct sock * sk,bool slow)1729 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1730 __releases(&sk->sk_lock.slock)
1731 {
1732 if (slow) {
1733 release_sock(sk);
1734 __release(&sk->sk_lock.slock);
1735 } else {
1736 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1737 spin_unlock_bh(&sk->sk_lock.slock);
1738 }
1739 }
1740
1741 /* Used by processes to "lock" a socket state, so that
1742 * interrupts and bottom half handlers won't change it
1743 * from under us. It essentially blocks any incoming
1744 * packets, so that we won't get any new data or any
1745 * packets that change the state of the socket.
1746 *
1747 * While locked, BH processing will add new packets to
1748 * the backlog queue. This queue is processed by the
1749 * owner of the socket lock right before it is released.
1750 *
1751 * Since ~2.3.5 it is also exclusive sleep lock serializing
1752 * accesses from user process context.
1753 */
1754
sock_owned_by_me(const struct sock * sk)1755 static inline void sock_owned_by_me(const struct sock *sk)
1756 {
1757 #ifdef CONFIG_LOCKDEP
1758 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1759 #endif
1760 }
1761
sock_owned_by_user(const struct sock * sk)1762 static inline bool sock_owned_by_user(const struct sock *sk)
1763 {
1764 sock_owned_by_me(sk);
1765 return sk->sk_lock.owned;
1766 }
1767
sock_owned_by_user_nocheck(const struct sock * sk)1768 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1769 {
1770 return sk->sk_lock.owned;
1771 }
1772
1773 /* no reclassification while locks are held */
sock_allow_reclassification(const struct sock * csk)1774 static inline bool sock_allow_reclassification(const struct sock *csk)
1775 {
1776 struct sock *sk = (struct sock *)csk;
1777
1778 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1779 }
1780
1781 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1782 struct proto *prot, int kern);
1783 void sk_free(struct sock *sk);
1784 void sk_destruct(struct sock *sk);
1785 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1786 void sk_free_unlock_clone(struct sock *sk);
1787
1788 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1789 gfp_t priority);
1790 void __sock_wfree(struct sk_buff *skb);
1791 void sock_wfree(struct sk_buff *skb);
1792 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1793 gfp_t priority);
1794 void skb_orphan_partial(struct sk_buff *skb);
1795 void sock_rfree(struct sk_buff *skb);
1796 void sock_efree(struct sk_buff *skb);
1797 #ifdef CONFIG_INET
1798 void sock_edemux(struct sk_buff *skb);
1799 void sock_pfree(struct sk_buff *skb);
1800 #else
1801 #define sock_edemux sock_efree
1802 #endif
1803
1804 int sock_setsockopt(struct socket *sock, int level, int op,
1805 sockptr_t optval, unsigned int optlen);
1806
1807 int sock_getsockopt(struct socket *sock, int level, int op,
1808 char __user *optval, int __user *optlen);
1809 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1810 bool timeval, bool time32);
1811 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1812 int noblock, int *errcode);
1813 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1814 unsigned long data_len, int noblock,
1815 int *errcode, int max_page_order);
1816 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1817 void sock_kfree_s(struct sock *sk, void *mem, int size);
1818 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1819 void sk_send_sigurg(struct sock *sk);
1820
1821 struct sockcm_cookie {
1822 u64 transmit_time;
1823 u32 mark;
1824 u16 tsflags;
1825 };
1826
sockcm_init(struct sockcm_cookie * sockc,const struct sock * sk)1827 static inline void sockcm_init(struct sockcm_cookie *sockc,
1828 const struct sock *sk)
1829 {
1830 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1831 }
1832
1833 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1834 struct sockcm_cookie *sockc);
1835 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1836 struct sockcm_cookie *sockc);
1837
1838 /*
1839 * Functions to fill in entries in struct proto_ops when a protocol
1840 * does not implement a particular function.
1841 */
1842 int sock_no_bind(struct socket *, struct sockaddr *, int);
1843 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1844 int sock_no_socketpair(struct socket *, struct socket *);
1845 int sock_no_accept(struct socket *, struct socket *, int, bool);
1846 int sock_no_getname(struct socket *, struct sockaddr *, int);
1847 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1848 int sock_no_listen(struct socket *, int);
1849 int sock_no_shutdown(struct socket *, int);
1850 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1851 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1852 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1853 int sock_no_mmap(struct file *file, struct socket *sock,
1854 struct vm_area_struct *vma);
1855 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1856 size_t size, int flags);
1857 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1858 int offset, size_t size, int flags);
1859
1860 /*
1861 * Functions to fill in entries in struct proto_ops when a protocol
1862 * uses the inet style.
1863 */
1864 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1865 char __user *optval, int __user *optlen);
1866 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1867 int flags);
1868 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1869 sockptr_t optval, unsigned int optlen);
1870
1871 void sk_common_release(struct sock *sk);
1872
1873 /*
1874 * Default socket callbacks and setup code
1875 */
1876
1877 /* Initialise core socket variables using an explicit uid. */
1878 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1879
1880 /* Initialise core socket variables.
1881 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1882 */
1883 void sock_init_data(struct socket *sock, struct sock *sk);
1884
1885 /*
1886 * Socket reference counting postulates.
1887 *
1888 * * Each user of socket SHOULD hold a reference count.
1889 * * Each access point to socket (an hash table bucket, reference from a list,
1890 * running timer, skb in flight MUST hold a reference count.
1891 * * When reference count hits 0, it means it will never increase back.
1892 * * When reference count hits 0, it means that no references from
1893 * outside exist to this socket and current process on current CPU
1894 * is last user and may/should destroy this socket.
1895 * * sk_free is called from any context: process, BH, IRQ. When
1896 * it is called, socket has no references from outside -> sk_free
1897 * may release descendant resources allocated by the socket, but
1898 * to the time when it is called, socket is NOT referenced by any
1899 * hash tables, lists etc.
1900 * * Packets, delivered from outside (from network or from another process)
1901 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1902 * when they sit in queue. Otherwise, packets will leak to hole, when
1903 * socket is looked up by one cpu and unhasing is made by another CPU.
1904 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1905 * (leak to backlog). Packet socket does all the processing inside
1906 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1907 * use separate SMP lock, so that they are prone too.
1908 */
1909
1910 /* Ungrab socket and destroy it, if it was the last reference. */
sock_put(struct sock * sk)1911 static inline void sock_put(struct sock *sk)
1912 {
1913 if (refcount_dec_and_test(&sk->sk_refcnt))
1914 sk_free(sk);
1915 }
1916 /* Generic version of sock_put(), dealing with all sockets
1917 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1918 */
1919 void sock_gen_put(struct sock *sk);
1920
1921 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1922 unsigned int trim_cap, bool refcounted);
sk_receive_skb(struct sock * sk,struct sk_buff * skb,const int nested)1923 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1924 const int nested)
1925 {
1926 return __sk_receive_skb(sk, skb, nested, 1, true);
1927 }
1928
sk_tx_queue_set(struct sock * sk,int tx_queue)1929 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1930 {
1931 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1932 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1933 return;
1934 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1935 * other WRITE_ONCE() because socket lock might be not held.
1936 */
1937 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1938 }
1939
1940 #define NO_QUEUE_MAPPING USHRT_MAX
1941
sk_tx_queue_clear(struct sock * sk)1942 static inline void sk_tx_queue_clear(struct sock *sk)
1943 {
1944 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1945 * other WRITE_ONCE() because socket lock might be not held.
1946 */
1947 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1948 }
1949
sk_tx_queue_get(const struct sock * sk)1950 static inline int sk_tx_queue_get(const struct sock *sk)
1951 {
1952 if (sk) {
1953 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
1954 * and sk_tx_queue_set().
1955 */
1956 int val = READ_ONCE(sk->sk_tx_queue_mapping);
1957
1958 if (val != NO_QUEUE_MAPPING)
1959 return val;
1960 }
1961 return -1;
1962 }
1963
sk_rx_queue_set(struct sock * sk,const struct sk_buff * skb)1964 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1965 {
1966 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1967 if (skb_rx_queue_recorded(skb)) {
1968 u16 rx_queue = skb_get_rx_queue(skb);
1969
1970 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1971 return;
1972
1973 sk->sk_rx_queue_mapping = rx_queue;
1974 }
1975 #endif
1976 }
1977
sk_rx_queue_clear(struct sock * sk)1978 static inline void sk_rx_queue_clear(struct sock *sk)
1979 {
1980 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1981 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1982 #endif
1983 }
1984
sk_rx_queue_get(const struct sock * sk)1985 static inline int sk_rx_queue_get(const struct sock *sk)
1986 {
1987 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1988 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1989 return sk->sk_rx_queue_mapping;
1990 #endif
1991
1992 return -1;
1993 }
1994
sk_set_socket(struct sock * sk,struct socket * sock)1995 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
1996 {
1997 sk->sk_socket = sock;
1998 }
1999
sk_sleep(struct sock * sk)2000 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2001 {
2002 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2003 return &rcu_dereference_raw(sk->sk_wq)->wait;
2004 }
2005 /* Detach socket from process context.
2006 * Announce socket dead, detach it from wait queue and inode.
2007 * Note that parent inode held reference count on this struct sock,
2008 * we do not release it in this function, because protocol
2009 * probably wants some additional cleanups or even continuing
2010 * to work with this socket (TCP).
2011 */
sock_orphan(struct sock * sk)2012 static inline void sock_orphan(struct sock *sk)
2013 {
2014 write_lock_bh(&sk->sk_callback_lock);
2015 sock_set_flag(sk, SOCK_DEAD);
2016 sk_set_socket(sk, NULL);
2017 sk->sk_wq = NULL;
2018 write_unlock_bh(&sk->sk_callback_lock);
2019 }
2020
sock_graft(struct sock * sk,struct socket * parent)2021 static inline void sock_graft(struct sock *sk, struct socket *parent)
2022 {
2023 WARN_ON(parent->sk);
2024 write_lock_bh(&sk->sk_callback_lock);
2025 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2026 parent->sk = sk;
2027 sk_set_socket(sk, parent);
2028 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2029 security_sock_graft(sk, parent);
2030 write_unlock_bh(&sk->sk_callback_lock);
2031 }
2032
2033 kuid_t sock_i_uid(struct sock *sk);
2034 unsigned long __sock_i_ino(struct sock *sk);
2035 unsigned long sock_i_ino(struct sock *sk);
2036
sock_net_uid(const struct net * net,const struct sock * sk)2037 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2038 {
2039 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2040 }
2041
net_tx_rndhash(void)2042 static inline u32 net_tx_rndhash(void)
2043 {
2044 u32 v = prandom_u32();
2045
2046 return v ?: 1;
2047 }
2048
sk_set_txhash(struct sock * sk)2049 static inline void sk_set_txhash(struct sock *sk)
2050 {
2051 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2052 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2053 }
2054
sk_rethink_txhash(struct sock * sk)2055 static inline bool sk_rethink_txhash(struct sock *sk)
2056 {
2057 if (sk->sk_txhash) {
2058 sk_set_txhash(sk);
2059 return true;
2060 }
2061 return false;
2062 }
2063
2064 static inline struct dst_entry *
__sk_dst_get(struct sock * sk)2065 __sk_dst_get(struct sock *sk)
2066 {
2067 return rcu_dereference_check(sk->sk_dst_cache,
2068 lockdep_sock_is_held(sk));
2069 }
2070
2071 static inline struct dst_entry *
sk_dst_get(struct sock * sk)2072 sk_dst_get(struct sock *sk)
2073 {
2074 struct dst_entry *dst;
2075
2076 rcu_read_lock();
2077 dst = rcu_dereference(sk->sk_dst_cache);
2078 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
2079 dst = NULL;
2080 rcu_read_unlock();
2081 return dst;
2082 }
2083
__dst_negative_advice(struct sock * sk)2084 static inline void __dst_negative_advice(struct sock *sk)
2085 {
2086 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2087
2088 if (dst && dst->ops->negative_advice) {
2089 ndst = dst->ops->negative_advice(dst);
2090
2091 if (ndst != dst) {
2092 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2093 sk_tx_queue_clear(sk);
2094 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2095 }
2096 }
2097 }
2098
dst_negative_advice(struct sock * sk)2099 static inline void dst_negative_advice(struct sock *sk)
2100 {
2101 sk_rethink_txhash(sk);
2102 __dst_negative_advice(sk);
2103 }
2104
2105 static inline void
__sk_dst_set(struct sock * sk,struct dst_entry * dst)2106 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2107 {
2108 struct dst_entry *old_dst;
2109
2110 sk_tx_queue_clear(sk);
2111 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2112 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2113 lockdep_sock_is_held(sk));
2114 rcu_assign_pointer(sk->sk_dst_cache, dst);
2115 dst_release(old_dst);
2116 }
2117
2118 static inline void
sk_dst_set(struct sock * sk,struct dst_entry * dst)2119 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2120 {
2121 struct dst_entry *old_dst;
2122
2123 sk_tx_queue_clear(sk);
2124 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2125 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2126 dst_release(old_dst);
2127 }
2128
2129 static inline void
__sk_dst_reset(struct sock * sk)2130 __sk_dst_reset(struct sock *sk)
2131 {
2132 __sk_dst_set(sk, NULL);
2133 }
2134
2135 static inline void
sk_dst_reset(struct sock * sk)2136 sk_dst_reset(struct sock *sk)
2137 {
2138 sk_dst_set(sk, NULL);
2139 }
2140
2141 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2142
2143 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2144
sk_dst_confirm(struct sock * sk)2145 static inline void sk_dst_confirm(struct sock *sk)
2146 {
2147 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2148 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2149 }
2150
sock_confirm_neigh(struct sk_buff * skb,struct neighbour * n)2151 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2152 {
2153 if (skb_get_dst_pending_confirm(skb)) {
2154 struct sock *sk = skb->sk;
2155 unsigned long now = jiffies;
2156
2157 /* avoid dirtying neighbour */
2158 if (READ_ONCE(n->confirmed) != now)
2159 WRITE_ONCE(n->confirmed, now);
2160 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2161 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2162 }
2163 }
2164
2165 bool sk_mc_loop(struct sock *sk);
2166
sk_can_gso(const struct sock * sk)2167 static inline bool sk_can_gso(const struct sock *sk)
2168 {
2169 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2170 }
2171
2172 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2173
sk_nocaps_add(struct sock * sk,netdev_features_t flags)2174 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2175 {
2176 sk->sk_route_nocaps |= flags;
2177 sk->sk_route_caps &= ~flags;
2178 }
2179
skb_do_copy_data_nocache(struct sock * sk,struct sk_buff * skb,struct iov_iter * from,char * to,int copy,int offset)2180 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2181 struct iov_iter *from, char *to,
2182 int copy, int offset)
2183 {
2184 if (skb->ip_summed == CHECKSUM_NONE) {
2185 __wsum csum = 0;
2186 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2187 return -EFAULT;
2188 skb->csum = csum_block_add(skb->csum, csum, offset);
2189 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2190 if (!copy_from_iter_full_nocache(to, copy, from))
2191 return -EFAULT;
2192 } else if (!copy_from_iter_full(to, copy, from))
2193 return -EFAULT;
2194
2195 return 0;
2196 }
2197
skb_add_data_nocache(struct sock * sk,struct sk_buff * skb,struct iov_iter * from,int copy)2198 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2199 struct iov_iter *from, int copy)
2200 {
2201 int err, offset = skb->len;
2202
2203 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2204 copy, offset);
2205 if (err)
2206 __skb_trim(skb, offset);
2207
2208 return err;
2209 }
2210
skb_copy_to_page_nocache(struct sock * sk,struct iov_iter * from,struct sk_buff * skb,struct page * page,int off,int copy)2211 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2212 struct sk_buff *skb,
2213 struct page *page,
2214 int off, int copy)
2215 {
2216 int err;
2217
2218 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2219 copy, skb->len);
2220 if (err)
2221 return err;
2222
2223 skb->len += copy;
2224 skb->data_len += copy;
2225 skb->truesize += copy;
2226 sk_wmem_queued_add(sk, copy);
2227 sk_mem_charge(sk, copy);
2228 return 0;
2229 }
2230
2231 /**
2232 * sk_wmem_alloc_get - returns write allocations
2233 * @sk: socket
2234 *
2235 * Return: sk_wmem_alloc minus initial offset of one
2236 */
sk_wmem_alloc_get(const struct sock * sk)2237 static inline int sk_wmem_alloc_get(const struct sock *sk)
2238 {
2239 return refcount_read(&sk->sk_wmem_alloc) - 1;
2240 }
2241
2242 /**
2243 * sk_rmem_alloc_get - returns read allocations
2244 * @sk: socket
2245 *
2246 * Return: sk_rmem_alloc
2247 */
sk_rmem_alloc_get(const struct sock * sk)2248 static inline int sk_rmem_alloc_get(const struct sock *sk)
2249 {
2250 return atomic_read(&sk->sk_rmem_alloc);
2251 }
2252
2253 /**
2254 * sk_has_allocations - check if allocations are outstanding
2255 * @sk: socket
2256 *
2257 * Return: true if socket has write or read allocations
2258 */
sk_has_allocations(const struct sock * sk)2259 static inline bool sk_has_allocations(const struct sock *sk)
2260 {
2261 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2262 }
2263
2264 /**
2265 * skwq_has_sleeper - check if there are any waiting processes
2266 * @wq: struct socket_wq
2267 *
2268 * Return: true if socket_wq has waiting processes
2269 *
2270 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2271 * barrier call. They were added due to the race found within the tcp code.
2272 *
2273 * Consider following tcp code paths::
2274 *
2275 * CPU1 CPU2
2276 * sys_select receive packet
2277 * ... ...
2278 * __add_wait_queue update tp->rcv_nxt
2279 * ... ...
2280 * tp->rcv_nxt check sock_def_readable
2281 * ... {
2282 * schedule rcu_read_lock();
2283 * wq = rcu_dereference(sk->sk_wq);
2284 * if (wq && waitqueue_active(&wq->wait))
2285 * wake_up_interruptible(&wq->wait)
2286 * ...
2287 * }
2288 *
2289 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2290 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2291 * could then endup calling schedule and sleep forever if there are no more
2292 * data on the socket.
2293 *
2294 */
skwq_has_sleeper(struct socket_wq * wq)2295 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2296 {
2297 return wq && wq_has_sleeper(&wq->wait);
2298 }
2299
2300 /**
2301 * sock_poll_wait - place memory barrier behind the poll_wait call.
2302 * @filp: file
2303 * @sock: socket to wait on
2304 * @p: poll_table
2305 *
2306 * See the comments in the wq_has_sleeper function.
2307 */
sock_poll_wait(struct file * filp,struct socket * sock,poll_table * p)2308 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2309 poll_table *p)
2310 {
2311 if (!poll_does_not_wait(p)) {
2312 poll_wait(filp, &sock->wq.wait, p);
2313 /* We need to be sure we are in sync with the
2314 * socket flags modification.
2315 *
2316 * This memory barrier is paired in the wq_has_sleeper.
2317 */
2318 smp_mb();
2319 }
2320 }
2321
skb_set_hash_from_sk(struct sk_buff * skb,struct sock * sk)2322 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2323 {
2324 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2325 u32 txhash = READ_ONCE(sk->sk_txhash);
2326
2327 if (txhash) {
2328 skb->l4_hash = 1;
2329 skb->hash = txhash;
2330 }
2331 }
2332
2333 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2334
2335 /*
2336 * Queue a received datagram if it will fit. Stream and sequenced
2337 * protocols can't normally use this as they need to fit buffers in
2338 * and play with them.
2339 *
2340 * Inlined as it's very short and called for pretty much every
2341 * packet ever received.
2342 */
skb_set_owner_r(struct sk_buff * skb,struct sock * sk)2343 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2344 {
2345 skb_orphan(skb);
2346 skb->sk = sk;
2347 skb->destructor = sock_rfree;
2348 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2349 sk_mem_charge(sk, skb->truesize);
2350 }
2351
skb_set_owner_sk_safe(struct sk_buff * skb,struct sock * sk)2352 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2353 {
2354 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2355 skb_orphan(skb);
2356 skb->destructor = sock_efree;
2357 skb->sk = sk;
2358 return true;
2359 }
2360 return false;
2361 }
2362
skb_clone_and_charge_r(struct sk_buff * skb,struct sock * sk)2363 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2364 {
2365 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2366 if (skb) {
2367 if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2368 skb_set_owner_r(skb, sk);
2369 return skb;
2370 }
2371 __kfree_skb(skb);
2372 }
2373 return NULL;
2374 }
2375
skb_prepare_for_gro(struct sk_buff * skb)2376 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2377 {
2378 if (skb->destructor != sock_wfree) {
2379 skb_orphan(skb);
2380 return;
2381 }
2382 skb->slow_gro = 1;
2383 }
2384
2385 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2386 unsigned long expires);
2387
2388 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2389
2390 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2391
2392 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2393 struct sk_buff *skb, unsigned int flags,
2394 void (*destructor)(struct sock *sk,
2395 struct sk_buff *skb));
2396 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2397 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2398
2399 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2400 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2401
2402 /*
2403 * Recover an error report and clear atomically
2404 */
2405
sock_error(struct sock * sk)2406 static inline int sock_error(struct sock *sk)
2407 {
2408 int err;
2409
2410 /* Avoid an atomic operation for the common case.
2411 * This is racy since another cpu/thread can change sk_err under us.
2412 */
2413 if (likely(data_race(!sk->sk_err)))
2414 return 0;
2415
2416 err = xchg(&sk->sk_err, 0);
2417 return -err;
2418 }
2419
2420 void sk_error_report(struct sock *sk);
2421
sock_wspace(struct sock * sk)2422 static inline unsigned long sock_wspace(struct sock *sk)
2423 {
2424 int amt = 0;
2425
2426 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2427 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2428 if (amt < 0)
2429 amt = 0;
2430 }
2431 return amt;
2432 }
2433
2434 /* Note:
2435 * We use sk->sk_wq_raw, from contexts knowing this
2436 * pointer is not NULL and cannot disappear/change.
2437 */
sk_set_bit(int nr,struct sock * sk)2438 static inline void sk_set_bit(int nr, struct sock *sk)
2439 {
2440 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2441 !sock_flag(sk, SOCK_FASYNC))
2442 return;
2443
2444 set_bit(nr, &sk->sk_wq_raw->flags);
2445 }
2446
sk_clear_bit(int nr,struct sock * sk)2447 static inline void sk_clear_bit(int nr, struct sock *sk)
2448 {
2449 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2450 !sock_flag(sk, SOCK_FASYNC))
2451 return;
2452
2453 clear_bit(nr, &sk->sk_wq_raw->flags);
2454 }
2455
sk_wake_async(const struct sock * sk,int how,int band)2456 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2457 {
2458 if (sock_flag(sk, SOCK_FASYNC)) {
2459 rcu_read_lock();
2460 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2461 rcu_read_unlock();
2462 }
2463 }
2464
2465 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2466 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2467 * Note: for send buffers, TCP works better if we can build two skbs at
2468 * minimum.
2469 */
2470 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2471
2472 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2473 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2474
sk_stream_moderate_sndbuf(struct sock * sk)2475 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2476 {
2477 u32 val;
2478
2479 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2480 return;
2481
2482 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2483
2484 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2485 }
2486
2487 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2488 bool force_schedule);
2489
2490 /**
2491 * sk_page_frag - return an appropriate page_frag
2492 * @sk: socket
2493 *
2494 * Use the per task page_frag instead of the per socket one for
2495 * optimization when we know that we're in process context and own
2496 * everything that's associated with %current.
2497 *
2498 * Both direct reclaim and page faults can nest inside other
2499 * socket operations and end up recursing into sk_page_frag()
2500 * while it's already in use: explicitly avoid task page_frag
2501 * usage if the caller is potentially doing any of them.
2502 * This assumes that page fault handlers use the GFP_NOFS flags.
2503 *
2504 * Return: a per task page_frag if context allows that,
2505 * otherwise a per socket one.
2506 */
sk_page_frag(struct sock * sk)2507 static inline struct page_frag *sk_page_frag(struct sock *sk)
2508 {
2509 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) ==
2510 (__GFP_DIRECT_RECLAIM | __GFP_FS))
2511 return ¤t->task_frag;
2512
2513 return &sk->sk_frag;
2514 }
2515
2516 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2517
2518 /*
2519 * Default write policy as shown to user space via poll/select/SIGIO
2520 */
sock_writeable(const struct sock * sk)2521 static inline bool sock_writeable(const struct sock *sk)
2522 {
2523 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2524 }
2525
gfp_any(void)2526 static inline gfp_t gfp_any(void)
2527 {
2528 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2529 }
2530
gfp_memcg_charge(void)2531 static inline gfp_t gfp_memcg_charge(void)
2532 {
2533 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2534 }
2535
sock_rcvtimeo(const struct sock * sk,bool noblock)2536 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2537 {
2538 return noblock ? 0 : sk->sk_rcvtimeo;
2539 }
2540
sock_sndtimeo(const struct sock * sk,bool noblock)2541 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2542 {
2543 return noblock ? 0 : sk->sk_sndtimeo;
2544 }
2545
sock_rcvlowat(const struct sock * sk,int waitall,int len)2546 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2547 {
2548 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2549
2550 return v ?: 1;
2551 }
2552
2553 /* Alas, with timeout socket operations are not restartable.
2554 * Compare this to poll().
2555 */
sock_intr_errno(long timeo)2556 static inline int sock_intr_errno(long timeo)
2557 {
2558 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2559 }
2560
2561 struct sock_skb_cb {
2562 u32 dropcount;
2563 };
2564
2565 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2566 * using skb->cb[] would keep using it directly and utilize its
2567 * alignement guarantee.
2568 */
2569 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2570 sizeof(struct sock_skb_cb)))
2571
2572 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2573 SOCK_SKB_CB_OFFSET))
2574
2575 #define sock_skb_cb_check_size(size) \
2576 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2577
2578 static inline void
sock_skb_set_dropcount(const struct sock * sk,struct sk_buff * skb)2579 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2580 {
2581 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2582 atomic_read(&sk->sk_drops) : 0;
2583 }
2584
sk_drops_add(struct sock * sk,const struct sk_buff * skb)2585 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2586 {
2587 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2588
2589 atomic_add(segs, &sk->sk_drops);
2590 }
2591
sock_read_timestamp(struct sock * sk)2592 static inline ktime_t sock_read_timestamp(struct sock *sk)
2593 {
2594 #if BITS_PER_LONG==32
2595 unsigned int seq;
2596 ktime_t kt;
2597
2598 do {
2599 seq = read_seqbegin(&sk->sk_stamp_seq);
2600 kt = sk->sk_stamp;
2601 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2602
2603 return kt;
2604 #else
2605 return READ_ONCE(sk->sk_stamp);
2606 #endif
2607 }
2608
sock_write_timestamp(struct sock * sk,ktime_t kt)2609 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2610 {
2611 #if BITS_PER_LONG==32
2612 write_seqlock(&sk->sk_stamp_seq);
2613 sk->sk_stamp = kt;
2614 write_sequnlock(&sk->sk_stamp_seq);
2615 #else
2616 WRITE_ONCE(sk->sk_stamp, kt);
2617 #endif
2618 }
2619
2620 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2621 struct sk_buff *skb);
2622 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2623 struct sk_buff *skb);
2624
2625 static inline void
sock_recv_timestamp(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2626 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2627 {
2628 ktime_t kt = skb->tstamp;
2629 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2630
2631 /*
2632 * generate control messages if
2633 * - receive time stamping in software requested
2634 * - software time stamp available and wanted
2635 * - hardware time stamps available and wanted
2636 */
2637 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2638 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2639 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2640 (hwtstamps->hwtstamp &&
2641 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2642 __sock_recv_timestamp(msg, sk, skb);
2643 else
2644 sock_write_timestamp(sk, kt);
2645
2646 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2647 __sock_recv_wifi_status(msg, sk, skb);
2648 }
2649
2650 void __sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2651 struct sk_buff *skb);
2652
2653 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
sock_recv_ts_and_drops(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2654 static inline void sock_recv_ts_and_drops(struct msghdr *msg, struct sock *sk,
2655 struct sk_buff *skb)
2656 {
2657 #define FLAGS_TS_OR_DROPS ((1UL << SOCK_RXQ_OVFL) | \
2658 (1UL << SOCK_RCVTSTAMP))
2659 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2660 SOF_TIMESTAMPING_RAW_HARDWARE)
2661
2662 if (sk->sk_flags & FLAGS_TS_OR_DROPS || sk->sk_tsflags & TSFLAGS_ANY)
2663 __sock_recv_ts_and_drops(msg, sk, skb);
2664 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2665 sock_write_timestamp(sk, skb->tstamp);
2666 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2667 sock_write_timestamp(sk, 0);
2668 }
2669
2670 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2671
2672 /**
2673 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2674 * @sk: socket sending this packet
2675 * @tsflags: timestamping flags to use
2676 * @tx_flags: completed with instructions for time stamping
2677 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2678 *
2679 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2680 */
_sock_tx_timestamp(struct sock * sk,__u16 tsflags,__u8 * tx_flags,__u32 * tskey)2681 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2682 __u8 *tx_flags, __u32 *tskey)
2683 {
2684 if (unlikely(tsflags)) {
2685 __sock_tx_timestamp(tsflags, tx_flags);
2686 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2687 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2688 *tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2689 }
2690 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2691 *tx_flags |= SKBTX_WIFI_STATUS;
2692 }
2693
sock_tx_timestamp(struct sock * sk,__u16 tsflags,__u8 * tx_flags)2694 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2695 __u8 *tx_flags)
2696 {
2697 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2698 }
2699
skb_setup_tx_timestamp(struct sk_buff * skb,__u16 tsflags)2700 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2701 {
2702 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2703 &skb_shinfo(skb)->tskey);
2704 }
2705
2706 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2707 /**
2708 * sk_eat_skb - Release a skb if it is no longer needed
2709 * @sk: socket to eat this skb from
2710 * @skb: socket buffer to eat
2711 *
2712 * This routine must be called with interrupts disabled or with the socket
2713 * locked so that the sk_buff queue operation is ok.
2714 */
sk_eat_skb(struct sock * sk,struct sk_buff * skb)2715 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2716 {
2717 __skb_unlink(skb, &sk->sk_receive_queue);
2718 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2719 !sk->sk_rx_skb_cache) {
2720 sk->sk_rx_skb_cache = skb;
2721 skb_orphan(skb);
2722 return;
2723 }
2724 __kfree_skb(skb);
2725 }
2726
2727 static inline
sock_net(const struct sock * sk)2728 struct net *sock_net(const struct sock *sk)
2729 {
2730 return read_pnet(&sk->sk_net);
2731 }
2732
2733 static inline
sock_net_set(struct sock * sk,struct net * net)2734 void sock_net_set(struct sock *sk, struct net *net)
2735 {
2736 write_pnet(&sk->sk_net, net);
2737 }
2738
2739 static inline bool
skb_sk_is_prefetched(struct sk_buff * skb)2740 skb_sk_is_prefetched(struct sk_buff *skb)
2741 {
2742 #ifdef CONFIG_INET
2743 return skb->destructor == sock_pfree;
2744 #else
2745 return false;
2746 #endif /* CONFIG_INET */
2747 }
2748
2749 /* This helper checks if a socket is a full socket,
2750 * ie _not_ a timewait or request socket.
2751 */
sk_fullsock(const struct sock * sk)2752 static inline bool sk_fullsock(const struct sock *sk)
2753 {
2754 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2755 }
2756
2757 static inline bool
sk_is_refcounted(struct sock * sk)2758 sk_is_refcounted(struct sock *sk)
2759 {
2760 /* Only full sockets have sk->sk_flags. */
2761 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2762 }
2763
2764 /**
2765 * skb_steal_sock - steal a socket from an sk_buff
2766 * @skb: sk_buff to steal the socket from
2767 * @refcounted: is set to true if the socket is reference-counted
2768 */
2769 static inline struct sock *
skb_steal_sock(struct sk_buff * skb,bool * refcounted)2770 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2771 {
2772 if (skb->sk) {
2773 struct sock *sk = skb->sk;
2774
2775 *refcounted = true;
2776 if (skb_sk_is_prefetched(skb))
2777 *refcounted = sk_is_refcounted(sk);
2778 skb->destructor = NULL;
2779 skb->sk = NULL;
2780 return sk;
2781 }
2782 *refcounted = false;
2783 return NULL;
2784 }
2785
2786 /* Checks if this SKB belongs to an HW offloaded socket
2787 * and whether any SW fallbacks are required based on dev.
2788 * Check decrypted mark in case skb_orphan() cleared socket.
2789 */
sk_validate_xmit_skb(struct sk_buff * skb,struct net_device * dev)2790 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2791 struct net_device *dev)
2792 {
2793 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2794 struct sock *sk = skb->sk;
2795
2796 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2797 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2798 #ifdef CONFIG_TLS_DEVICE
2799 } else if (unlikely(skb->decrypted)) {
2800 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2801 kfree_skb(skb);
2802 skb = NULL;
2803 #endif
2804 }
2805 #endif
2806
2807 return skb;
2808 }
2809
2810 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2811 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2812 */
sk_listener(const struct sock * sk)2813 static inline bool sk_listener(const struct sock *sk)
2814 {
2815 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2816 }
2817
2818 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2819 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2820 int type);
2821
2822 bool sk_ns_capable(const struct sock *sk,
2823 struct user_namespace *user_ns, int cap);
2824 bool sk_capable(const struct sock *sk, int cap);
2825 bool sk_net_capable(const struct sock *sk, int cap);
2826
2827 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2828
2829 /* Take into consideration the size of the struct sk_buff overhead in the
2830 * determination of these values, since that is non-constant across
2831 * platforms. This makes socket queueing behavior and performance
2832 * not depend upon such differences.
2833 */
2834 #define _SK_MEM_PACKETS 256
2835 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2836 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2837 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2838
2839 extern __u32 sysctl_wmem_max;
2840 extern __u32 sysctl_rmem_max;
2841
2842 extern int sysctl_tstamp_allow_data;
2843 extern int sysctl_optmem_max;
2844
2845 extern __u32 sysctl_wmem_default;
2846 extern __u32 sysctl_rmem_default;
2847
2848 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2849 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2850
sk_get_wmem0(const struct sock * sk,const struct proto * proto)2851 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2852 {
2853 /* Does this proto have per netns sysctl_wmem ? */
2854 if (proto->sysctl_wmem_offset)
2855 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2856
2857 return READ_ONCE(*proto->sysctl_wmem);
2858 }
2859
sk_get_rmem0(const struct sock * sk,const struct proto * proto)2860 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2861 {
2862 /* Does this proto have per netns sysctl_rmem ? */
2863 if (proto->sysctl_rmem_offset)
2864 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2865
2866 return READ_ONCE(*proto->sysctl_rmem);
2867 }
2868
2869 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2870 * Some wifi drivers need to tweak it to get more chunks.
2871 * They can use this helper from their ndo_start_xmit()
2872 */
sk_pacing_shift_update(struct sock * sk,int val)2873 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2874 {
2875 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2876 return;
2877 WRITE_ONCE(sk->sk_pacing_shift, val);
2878 }
2879
2880 /* if a socket is bound to a device, check that the given device
2881 * index is either the same or that the socket is bound to an L3
2882 * master device and the given device index is also enslaved to
2883 * that L3 master
2884 */
sk_dev_equal_l3scope(struct sock * sk,int dif)2885 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2886 {
2887 int mdif;
2888
2889 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2890 return true;
2891
2892 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2893 if (mdif && mdif == sk->sk_bound_dev_if)
2894 return true;
2895
2896 return false;
2897 }
2898
2899 void sock_def_readable(struct sock *sk);
2900
2901 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2902 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2903 int sock_set_timestamping(struct sock *sk, int optname,
2904 struct so_timestamping timestamping);
2905
2906 void sock_enable_timestamps(struct sock *sk);
2907 void sock_no_linger(struct sock *sk);
2908 void sock_set_keepalive(struct sock *sk);
2909 void sock_set_priority(struct sock *sk, u32 priority);
2910 void sock_set_rcvbuf(struct sock *sk, int val);
2911 void sock_set_mark(struct sock *sk, u32 val);
2912 void sock_set_reuseaddr(struct sock *sk);
2913 void sock_set_reuseport(struct sock *sk);
2914 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2915
2916 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2917
sk_is_readable(struct sock * sk)2918 static inline bool sk_is_readable(struct sock *sk)
2919 {
2920 if (sk->sk_prot->sock_is_readable)
2921 return sk->sk_prot->sock_is_readable(sk);
2922 return false;
2923 }
2924 #endif /* _SOCK_H */
2925