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_wait_pending: number of threads blocked on this socket
340 * @sk_state_change: callback to indicate change in the state of the sock
341 * @sk_data_ready: callback to indicate there is data to be processed
342 * @sk_write_space: callback to indicate there is bf sending space available
343 * @sk_error_report: callback to indicate errors (e.g. %MSG_ERRQUEUE)
344 * @sk_backlog_rcv: callback to process the backlog
345 * @sk_validate_xmit_skb: ptr to an optional validate function
346 * @sk_destruct: called at sock freeing time, i.e. when all refcnt == 0
347 * @sk_reuseport_cb: reuseport group container
348 * @sk_bpf_storage: ptr to cache and control for bpf_sk_storage
349 * @sk_rcu: used during RCU grace period
350 * @sk_clockid: clockid used by time-based scheduling (SO_TXTIME)
351 * @sk_txtime_deadline_mode: set deadline mode for SO_TXTIME
352 * @sk_txtime_report_errors: set report errors mode for SO_TXTIME
353 * @sk_txtime_unused: unused txtime flags
354 */
355 struct sock {
356 /*
357 * Now struct inet_timewait_sock also uses sock_common, so please just
358 * don't add nothing before this first member (__sk_common) --acme
359 */
360 struct sock_common __sk_common;
361 #define sk_node __sk_common.skc_node
362 #define sk_nulls_node __sk_common.skc_nulls_node
363 #define sk_refcnt __sk_common.skc_refcnt
364 #define sk_tx_queue_mapping __sk_common.skc_tx_queue_mapping
365 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
366 #define sk_rx_queue_mapping __sk_common.skc_rx_queue_mapping
367 #endif
368
369 #define sk_dontcopy_begin __sk_common.skc_dontcopy_begin
370 #define sk_dontcopy_end __sk_common.skc_dontcopy_end
371 #define sk_hash __sk_common.skc_hash
372 #define sk_portpair __sk_common.skc_portpair
373 #define sk_num __sk_common.skc_num
374 #define sk_dport __sk_common.skc_dport
375 #define sk_addrpair __sk_common.skc_addrpair
376 #define sk_daddr __sk_common.skc_daddr
377 #define sk_rcv_saddr __sk_common.skc_rcv_saddr
378 #define sk_family __sk_common.skc_family
379 #define sk_state __sk_common.skc_state
380 #define sk_reuse __sk_common.skc_reuse
381 #define sk_reuseport __sk_common.skc_reuseport
382 #define sk_ipv6only __sk_common.skc_ipv6only
383 #define sk_net_refcnt __sk_common.skc_net_refcnt
384 #define sk_bound_dev_if __sk_common.skc_bound_dev_if
385 #define sk_bind_node __sk_common.skc_bind_node
386 #define sk_prot __sk_common.skc_prot
387 #define sk_net __sk_common.skc_net
388 #define sk_v6_daddr __sk_common.skc_v6_daddr
389 #define sk_v6_rcv_saddr __sk_common.skc_v6_rcv_saddr
390 #define sk_cookie __sk_common.skc_cookie
391 #define sk_incoming_cpu __sk_common.skc_incoming_cpu
392 #define sk_flags __sk_common.skc_flags
393 #define sk_rxhash __sk_common.skc_rxhash
394
395 socket_lock_t sk_lock;
396 atomic_t sk_drops;
397 int sk_rcvlowat;
398 struct sk_buff_head sk_error_queue;
399 struct sk_buff *sk_rx_skb_cache;
400 struct sk_buff_head sk_receive_queue;
401 /*
402 * The backlog queue is special, it is always used with
403 * the per-socket spinlock held and requires low latency
404 * access. Therefore we special case it's implementation.
405 * Note : rmem_alloc is in this structure to fill a hole
406 * on 64bit arches, not because its logically part of
407 * backlog.
408 */
409 struct {
410 atomic_t rmem_alloc;
411 int len;
412 struct sk_buff *head;
413 struct sk_buff *tail;
414 } sk_backlog;
415 #define sk_rmem_alloc sk_backlog.rmem_alloc
416
417 int sk_forward_alloc;
418 #ifdef CONFIG_NET_RX_BUSY_POLL
419 unsigned int sk_ll_usec;
420 /* ===== mostly read cache line ===== */
421 unsigned int sk_napi_id;
422 #endif
423 int sk_rcvbuf;
424 int sk_wait_pending;
425
426 struct sk_filter __rcu *sk_filter;
427 union {
428 struct socket_wq __rcu *sk_wq;
429 /* private: */
430 struct socket_wq *sk_wq_raw;
431 /* public: */
432 };
433 #ifdef CONFIG_XFRM
434 struct xfrm_policy __rcu *sk_policy[2];
435 #endif
436 struct dst_entry __rcu *sk_rx_dst;
437 int sk_rx_dst_ifindex;
438 u32 sk_rx_dst_cookie;
439
440 struct dst_entry __rcu *sk_dst_cache;
441 atomic_t sk_omem_alloc;
442 int sk_sndbuf;
443
444 /* ===== cache line for TX ===== */
445 int sk_wmem_queued;
446 refcount_t sk_wmem_alloc;
447 unsigned long sk_tsq_flags;
448 union {
449 struct sk_buff *sk_send_head;
450 struct rb_root tcp_rtx_queue;
451 };
452 struct sk_buff *sk_tx_skb_cache;
453 struct sk_buff_head sk_write_queue;
454 __s32 sk_peek_off;
455 int sk_write_pending;
456 __u32 sk_dst_pending_confirm;
457 u32 sk_pacing_status; /* see enum sk_pacing */
458 long sk_sndtimeo;
459 struct timer_list sk_timer;
460 __u32 sk_priority;
461 __u32 sk_mark;
462 unsigned long sk_pacing_rate; /* bytes per second */
463 unsigned long sk_max_pacing_rate;
464 struct page_frag sk_frag;
465 netdev_features_t sk_route_caps;
466 netdev_features_t sk_route_nocaps;
467 netdev_features_t sk_route_forced_caps;
468 int sk_gso_type;
469 unsigned int sk_gso_max_size;
470 gfp_t sk_allocation;
471 __u32 sk_txhash;
472
473 /*
474 * Because of non atomicity rules, all
475 * changes are protected by socket lock.
476 */
477 u8 sk_padding : 1,
478 sk_kern_sock : 1,
479 sk_no_check_tx : 1,
480 sk_no_check_rx : 1,
481 sk_userlocks : 4;
482 u8 sk_pacing_shift;
483 u16 sk_type;
484 u16 sk_protocol;
485 u16 sk_gso_max_segs;
486 unsigned long sk_lingertime;
487 struct proto *sk_prot_creator;
488 rwlock_t sk_callback_lock;
489 int sk_err,
490 sk_err_soft;
491 u32 sk_ack_backlog;
492 u32 sk_max_ack_backlog;
493 kuid_t sk_uid;
494 #ifdef CONFIG_NET_RX_BUSY_POLL
495 u8 sk_prefer_busy_poll;
496 u16 sk_busy_poll_budget;
497 #endif
498 spinlock_t sk_peer_lock;
499 struct pid *sk_peer_pid;
500 const struct cred *sk_peer_cred;
501
502 long sk_rcvtimeo;
503 ktime_t sk_stamp;
504 #if BITS_PER_LONG==32
505 seqlock_t sk_stamp_seq;
506 #endif
507 u16 sk_tsflags;
508 int sk_bind_phc;
509 u8 sk_shutdown;
510 atomic_t sk_tskey;
511 atomic_t sk_zckey;
512
513 u8 sk_clockid;
514 u8 sk_txtime_deadline_mode : 1,
515 sk_txtime_report_errors : 1,
516 sk_txtime_unused : 6;
517
518 struct socket *sk_socket;
519 void *sk_user_data;
520 #ifdef CONFIG_SECURITY
521 void *sk_security;
522 #endif
523 struct sock_cgroup_data sk_cgrp_data;
524 struct mem_cgroup *sk_memcg;
525 void (*sk_state_change)(struct sock *sk);
526 void (*sk_data_ready)(struct sock *sk);
527 void (*sk_write_space)(struct sock *sk);
528 void (*sk_error_report)(struct sock *sk);
529 int (*sk_backlog_rcv)(struct sock *sk,
530 struct sk_buff *skb);
531 #ifdef CONFIG_SOCK_VALIDATE_XMIT
532 struct sk_buff* (*sk_validate_xmit_skb)(struct sock *sk,
533 struct net_device *dev,
534 struct sk_buff *skb);
535 #endif
536 void (*sk_destruct)(struct sock *sk);
537 struct sock_reuseport __rcu *sk_reuseport_cb;
538 #ifdef CONFIG_BPF_SYSCALL
539 struct bpf_local_storage __rcu *sk_bpf_storage;
540 #endif
541 struct rcu_head sk_rcu;
542
543 ANDROID_OEM_DATA(1);
544 ANDROID_KABI_RESERVE(1);
545 ANDROID_KABI_RESERVE(2);
546 ANDROID_KABI_RESERVE(3);
547 ANDROID_KABI_RESERVE(4);
548 ANDROID_KABI_RESERVE(5);
549 ANDROID_KABI_RESERVE(6);
550 ANDROID_KABI_RESERVE(7);
551 ANDROID_KABI_RESERVE(8);
552 };
553
554 enum sk_pacing {
555 SK_PACING_NONE = 0,
556 SK_PACING_NEEDED = 1,
557 SK_PACING_FQ = 2,
558 };
559
560 /* flag bits in sk_user_data
561 *
562 * - SK_USER_DATA_NOCOPY: Pointer stored in sk_user_data might
563 * not be suitable for copying when cloning the socket. For instance,
564 * it can point to a reference counted object. sk_user_data bottom
565 * bit is set if pointer must not be copied.
566 *
567 * - SK_USER_DATA_BPF: Mark whether sk_user_data field is
568 * managed/owned by a BPF reuseport array. This bit should be set
569 * when sk_user_data's sk is added to the bpf's reuseport_array.
570 *
571 * - SK_USER_DATA_PSOCK: Mark whether pointer stored in
572 * sk_user_data points to psock type. This bit should be set
573 * when sk_user_data is assigned to a psock object.
574 */
575 #define SK_USER_DATA_NOCOPY 1UL
576 #define SK_USER_DATA_BPF 2UL
577 #define SK_USER_DATA_PSOCK 4UL
578 #define SK_USER_DATA_PTRMASK ~(SK_USER_DATA_NOCOPY | SK_USER_DATA_BPF |\
579 SK_USER_DATA_PSOCK)
580
581 /**
582 * sk_user_data_is_nocopy - Test if sk_user_data pointer must not be copied
583 * @sk: socket
584 */
sk_user_data_is_nocopy(const struct sock * sk)585 static inline bool sk_user_data_is_nocopy(const struct sock *sk)
586 {
587 return ((uintptr_t)sk->sk_user_data & SK_USER_DATA_NOCOPY);
588 }
589
590 #define __sk_user_data(sk) ((*((void __rcu **)&(sk)->sk_user_data)))
591
592 /**
593 * __rcu_dereference_sk_user_data_with_flags - return the pointer
594 * only if argument flags all has been set in sk_user_data. Otherwise
595 * return NULL
596 *
597 * @sk: socket
598 * @flags: flag bits
599 */
600 static inline void *
__rcu_dereference_sk_user_data_with_flags(const struct sock * sk,uintptr_t flags)601 __rcu_dereference_sk_user_data_with_flags(const struct sock *sk,
602 uintptr_t flags)
603 {
604 uintptr_t sk_user_data = (uintptr_t)rcu_dereference(__sk_user_data(sk));
605
606 WARN_ON_ONCE(flags & SK_USER_DATA_PTRMASK);
607
608 if ((sk_user_data & flags) == flags)
609 return (void *)(sk_user_data & SK_USER_DATA_PTRMASK);
610 return NULL;
611 }
612
613 #define rcu_dereference_sk_user_data(sk) \
614 __rcu_dereference_sk_user_data_with_flags(sk, 0)
615 #define __rcu_assign_sk_user_data_with_flags(sk, ptr, flags) \
616 ({ \
617 uintptr_t __tmp1 = (uintptr_t)(ptr), \
618 __tmp2 = (uintptr_t)(flags); \
619 WARN_ON_ONCE(__tmp1 & ~SK_USER_DATA_PTRMASK); \
620 WARN_ON_ONCE(__tmp2 & SK_USER_DATA_PTRMASK); \
621 rcu_assign_pointer(__sk_user_data((sk)), \
622 __tmp1 | __tmp2); \
623 })
624 #define rcu_assign_sk_user_data(sk, ptr) \
625 __rcu_assign_sk_user_data_with_flags(sk, ptr, 0)
626
627 /*
628 * SK_CAN_REUSE and SK_NO_REUSE on a socket mean that the socket is OK
629 * or not whether his port will be reused by someone else. SK_FORCE_REUSE
630 * on a socket means that the socket will reuse everybody else's port
631 * without looking at the other's sk_reuse value.
632 */
633
634 #define SK_NO_REUSE 0
635 #define SK_CAN_REUSE 1
636 #define SK_FORCE_REUSE 2
637
638 int sk_set_peek_off(struct sock *sk, int val);
639
sk_peek_offset(struct sock * sk,int flags)640 static inline int sk_peek_offset(struct sock *sk, int flags)
641 {
642 if (unlikely(flags & MSG_PEEK)) {
643 return READ_ONCE(sk->sk_peek_off);
644 }
645
646 return 0;
647 }
648
sk_peek_offset_bwd(struct sock * sk,int val)649 static inline void sk_peek_offset_bwd(struct sock *sk, int val)
650 {
651 s32 off = READ_ONCE(sk->sk_peek_off);
652
653 if (unlikely(off >= 0)) {
654 off = max_t(s32, off - val, 0);
655 WRITE_ONCE(sk->sk_peek_off, off);
656 }
657 }
658
sk_peek_offset_fwd(struct sock * sk,int val)659 static inline void sk_peek_offset_fwd(struct sock *sk, int val)
660 {
661 sk_peek_offset_bwd(sk, -val);
662 }
663
664 /*
665 * Hashed lists helper routines
666 */
sk_entry(const struct hlist_node * node)667 static inline struct sock *sk_entry(const struct hlist_node *node)
668 {
669 return hlist_entry(node, struct sock, sk_node);
670 }
671
__sk_head(const struct hlist_head * head)672 static inline struct sock *__sk_head(const struct hlist_head *head)
673 {
674 return hlist_entry(head->first, struct sock, sk_node);
675 }
676
sk_head(const struct hlist_head * head)677 static inline struct sock *sk_head(const struct hlist_head *head)
678 {
679 return hlist_empty(head) ? NULL : __sk_head(head);
680 }
681
__sk_nulls_head(const struct hlist_nulls_head * head)682 static inline struct sock *__sk_nulls_head(const struct hlist_nulls_head *head)
683 {
684 return hlist_nulls_entry(head->first, struct sock, sk_nulls_node);
685 }
686
sk_nulls_head(const struct hlist_nulls_head * head)687 static inline struct sock *sk_nulls_head(const struct hlist_nulls_head *head)
688 {
689 return hlist_nulls_empty(head) ? NULL : __sk_nulls_head(head);
690 }
691
sk_next(const struct sock * sk)692 static inline struct sock *sk_next(const struct sock *sk)
693 {
694 return hlist_entry_safe(sk->sk_node.next, struct sock, sk_node);
695 }
696
sk_nulls_next(const struct sock * sk)697 static inline struct sock *sk_nulls_next(const struct sock *sk)
698 {
699 return (!is_a_nulls(sk->sk_nulls_node.next)) ?
700 hlist_nulls_entry(sk->sk_nulls_node.next,
701 struct sock, sk_nulls_node) :
702 NULL;
703 }
704
sk_unhashed(const struct sock * sk)705 static inline bool sk_unhashed(const struct sock *sk)
706 {
707 return hlist_unhashed(&sk->sk_node);
708 }
709
sk_hashed(const struct sock * sk)710 static inline bool sk_hashed(const struct sock *sk)
711 {
712 return !sk_unhashed(sk);
713 }
714
sk_node_init(struct hlist_node * node)715 static inline void sk_node_init(struct hlist_node *node)
716 {
717 node->pprev = NULL;
718 }
719
sk_nulls_node_init(struct hlist_nulls_node * node)720 static inline void sk_nulls_node_init(struct hlist_nulls_node *node)
721 {
722 node->pprev = NULL;
723 }
724
__sk_del_node(struct sock * sk)725 static inline void __sk_del_node(struct sock *sk)
726 {
727 __hlist_del(&sk->sk_node);
728 }
729
730 /* NB: equivalent to hlist_del_init_rcu */
__sk_del_node_init(struct sock * sk)731 static inline bool __sk_del_node_init(struct sock *sk)
732 {
733 if (sk_hashed(sk)) {
734 __sk_del_node(sk);
735 sk_node_init(&sk->sk_node);
736 return true;
737 }
738 return false;
739 }
740
741 /* Grab socket reference count. This operation is valid only
742 when sk is ALREADY grabbed f.e. it is found in hash table
743 or a list and the lookup is made under lock preventing hash table
744 modifications.
745 */
746
sock_hold(struct sock * sk)747 static __always_inline void sock_hold(struct sock *sk)
748 {
749 refcount_inc(&sk->sk_refcnt);
750 }
751
752 /* Ungrab socket in the context, which assumes that socket refcnt
753 cannot hit zero, f.e. it is true in context of any socketcall.
754 */
__sock_put(struct sock * sk)755 static __always_inline void __sock_put(struct sock *sk)
756 {
757 refcount_dec(&sk->sk_refcnt);
758 }
759
sk_del_node_init(struct sock * sk)760 static inline bool sk_del_node_init(struct sock *sk)
761 {
762 bool rc = __sk_del_node_init(sk);
763
764 if (rc) {
765 /* paranoid for a while -acme */
766 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
767 __sock_put(sk);
768 }
769 return rc;
770 }
771 #define sk_del_node_init_rcu(sk) sk_del_node_init(sk)
772
__sk_nulls_del_node_init_rcu(struct sock * sk)773 static inline bool __sk_nulls_del_node_init_rcu(struct sock *sk)
774 {
775 if (sk_hashed(sk)) {
776 hlist_nulls_del_init_rcu(&sk->sk_nulls_node);
777 return true;
778 }
779 return false;
780 }
781
sk_nulls_del_node_init_rcu(struct sock * sk)782 static inline bool sk_nulls_del_node_init_rcu(struct sock *sk)
783 {
784 bool rc = __sk_nulls_del_node_init_rcu(sk);
785
786 if (rc) {
787 /* paranoid for a while -acme */
788 WARN_ON(refcount_read(&sk->sk_refcnt) == 1);
789 __sock_put(sk);
790 }
791 return rc;
792 }
793
__sk_add_node(struct sock * sk,struct hlist_head * list)794 static inline void __sk_add_node(struct sock *sk, struct hlist_head *list)
795 {
796 hlist_add_head(&sk->sk_node, list);
797 }
798
sk_add_node(struct sock * sk,struct hlist_head * list)799 static inline void sk_add_node(struct sock *sk, struct hlist_head *list)
800 {
801 sock_hold(sk);
802 __sk_add_node(sk, list);
803 }
804
sk_add_node_rcu(struct sock * sk,struct hlist_head * list)805 static inline void sk_add_node_rcu(struct sock *sk, struct hlist_head *list)
806 {
807 sock_hold(sk);
808 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
809 sk->sk_family == AF_INET6)
810 hlist_add_tail_rcu(&sk->sk_node, list);
811 else
812 hlist_add_head_rcu(&sk->sk_node, list);
813 }
814
sk_add_node_tail_rcu(struct sock * sk,struct hlist_head * list)815 static inline void sk_add_node_tail_rcu(struct sock *sk, struct hlist_head *list)
816 {
817 sock_hold(sk);
818 hlist_add_tail_rcu(&sk->sk_node, list);
819 }
820
__sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)821 static inline void __sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
822 {
823 hlist_nulls_add_head_rcu(&sk->sk_nulls_node, list);
824 }
825
__sk_nulls_add_node_tail_rcu(struct sock * sk,struct hlist_nulls_head * list)826 static inline void __sk_nulls_add_node_tail_rcu(struct sock *sk, struct hlist_nulls_head *list)
827 {
828 hlist_nulls_add_tail_rcu(&sk->sk_nulls_node, list);
829 }
830
sk_nulls_add_node_rcu(struct sock * sk,struct hlist_nulls_head * list)831 static inline void sk_nulls_add_node_rcu(struct sock *sk, struct hlist_nulls_head *list)
832 {
833 sock_hold(sk);
834 __sk_nulls_add_node_rcu(sk, list);
835 }
836
__sk_del_bind_node(struct sock * sk)837 static inline void __sk_del_bind_node(struct sock *sk)
838 {
839 __hlist_del(&sk->sk_bind_node);
840 }
841
sk_add_bind_node(struct sock * sk,struct hlist_head * list)842 static inline void sk_add_bind_node(struct sock *sk,
843 struct hlist_head *list)
844 {
845 hlist_add_head(&sk->sk_bind_node, list);
846 }
847
848 #define sk_for_each(__sk, list) \
849 hlist_for_each_entry(__sk, list, sk_node)
850 #define sk_for_each_rcu(__sk, list) \
851 hlist_for_each_entry_rcu(__sk, list, sk_node)
852 #define sk_nulls_for_each(__sk, node, list) \
853 hlist_nulls_for_each_entry(__sk, node, list, sk_nulls_node)
854 #define sk_nulls_for_each_rcu(__sk, node, list) \
855 hlist_nulls_for_each_entry_rcu(__sk, node, list, sk_nulls_node)
856 #define sk_for_each_from(__sk) \
857 hlist_for_each_entry_from(__sk, sk_node)
858 #define sk_nulls_for_each_from(__sk, node) \
859 if (__sk && ({ node = &(__sk)->sk_nulls_node; 1; })) \
860 hlist_nulls_for_each_entry_from(__sk, node, sk_nulls_node)
861 #define sk_for_each_safe(__sk, tmp, list) \
862 hlist_for_each_entry_safe(__sk, tmp, list, sk_node)
863 #define sk_for_each_bound(__sk, list) \
864 hlist_for_each_entry(__sk, list, sk_bind_node)
865
866 /**
867 * sk_for_each_entry_offset_rcu - iterate over a list at a given struct offset
868 * @tpos: the type * to use as a loop cursor.
869 * @pos: the &struct hlist_node to use as a loop cursor.
870 * @head: the head for your list.
871 * @offset: offset of hlist_node within the struct.
872 *
873 */
874 #define sk_for_each_entry_offset_rcu(tpos, pos, head, offset) \
875 for (pos = rcu_dereference(hlist_first_rcu(head)); \
876 pos != NULL && \
877 ({ tpos = (typeof(*tpos) *)((void *)pos - offset); 1;}); \
878 pos = rcu_dereference(hlist_next_rcu(pos)))
879
sk_user_ns(struct sock * sk)880 static inline struct user_namespace *sk_user_ns(struct sock *sk)
881 {
882 /* Careful only use this in a context where these parameters
883 * can not change and must all be valid, such as recvmsg from
884 * userspace.
885 */
886 return sk->sk_socket->file->f_cred->user_ns;
887 }
888
889 /* Sock flags */
890 enum sock_flags {
891 SOCK_DEAD,
892 SOCK_DONE,
893 SOCK_URGINLINE,
894 SOCK_KEEPOPEN,
895 SOCK_LINGER,
896 SOCK_DESTROY,
897 SOCK_BROADCAST,
898 SOCK_TIMESTAMP,
899 SOCK_ZAPPED,
900 SOCK_USE_WRITE_QUEUE, /* whether to call sk->sk_write_space in sock_wfree */
901 SOCK_DBG, /* %SO_DEBUG setting */
902 SOCK_RCVTSTAMP, /* %SO_TIMESTAMP setting */
903 SOCK_RCVTSTAMPNS, /* %SO_TIMESTAMPNS setting */
904 SOCK_LOCALROUTE, /* route locally only, %SO_DONTROUTE setting */
905 SOCK_MEMALLOC, /* VM depends on this socket for swapping */
906 SOCK_TIMESTAMPING_RX_SOFTWARE, /* %SOF_TIMESTAMPING_RX_SOFTWARE */
907 SOCK_FASYNC, /* fasync() active */
908 SOCK_RXQ_OVFL,
909 SOCK_ZEROCOPY, /* buffers from userspace */
910 SOCK_WIFI_STATUS, /* push wifi status to userspace */
911 SOCK_NOFCS, /* Tell NIC not to do the Ethernet FCS.
912 * Will use last 4 bytes of packet sent from
913 * user-space instead.
914 */
915 SOCK_FILTER_LOCKED, /* Filter cannot be changed anymore */
916 SOCK_SELECT_ERR_QUEUE, /* Wake select on error queue */
917 SOCK_RCU_FREE, /* wait rcu grace period in sk_destruct() */
918 SOCK_TXTIME,
919 SOCK_XDP, /* XDP is attached */
920 SOCK_TSTAMP_NEW, /* Indicates 64 bit timestamps always */
921 SOCK_RCVMARK, /* Receive SO_MARK ancillary data with packet */
922 };
923
924 #define SK_FLAGS_TIMESTAMP ((1UL << SOCK_TIMESTAMP) | (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE))
925
sock_copy_flags(struct sock * nsk,struct sock * osk)926 static inline void sock_copy_flags(struct sock *nsk, struct sock *osk)
927 {
928 nsk->sk_flags = osk->sk_flags;
929 }
930
sock_set_flag(struct sock * sk,enum sock_flags flag)931 static inline void sock_set_flag(struct sock *sk, enum sock_flags flag)
932 {
933 __set_bit(flag, &sk->sk_flags);
934 }
935
sock_reset_flag(struct sock * sk,enum sock_flags flag)936 static inline void sock_reset_flag(struct sock *sk, enum sock_flags flag)
937 {
938 __clear_bit(flag, &sk->sk_flags);
939 }
940
sock_valbool_flag(struct sock * sk,enum sock_flags bit,int valbool)941 static inline void sock_valbool_flag(struct sock *sk, enum sock_flags bit,
942 int valbool)
943 {
944 if (valbool)
945 sock_set_flag(sk, bit);
946 else
947 sock_reset_flag(sk, bit);
948 }
949
sock_flag(const struct sock * sk,enum sock_flags flag)950 static inline bool sock_flag(const struct sock *sk, enum sock_flags flag)
951 {
952 return test_bit(flag, &sk->sk_flags);
953 }
954
955 #ifdef CONFIG_NET
956 DECLARE_STATIC_KEY_FALSE(memalloc_socks_key);
sk_memalloc_socks(void)957 static inline int sk_memalloc_socks(void)
958 {
959 return static_branch_unlikely(&memalloc_socks_key);
960 }
961
962 void __receive_sock(struct file *file);
963 #else
964
sk_memalloc_socks(void)965 static inline int sk_memalloc_socks(void)
966 {
967 return 0;
968 }
969
__receive_sock(struct file * file)970 static inline void __receive_sock(struct file *file)
971 { }
972 #endif
973
sk_gfp_mask(const struct sock * sk,gfp_t gfp_mask)974 static inline gfp_t sk_gfp_mask(const struct sock *sk, gfp_t gfp_mask)
975 {
976 return gfp_mask | (sk->sk_allocation & __GFP_MEMALLOC);
977 }
978
sk_acceptq_removed(struct sock * sk)979 static inline void sk_acceptq_removed(struct sock *sk)
980 {
981 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog - 1);
982 }
983
sk_acceptq_added(struct sock * sk)984 static inline void sk_acceptq_added(struct sock *sk)
985 {
986 WRITE_ONCE(sk->sk_ack_backlog, sk->sk_ack_backlog + 1);
987 }
988
989 /* Note: If you think the test should be:
990 * return READ_ONCE(sk->sk_ack_backlog) >= READ_ONCE(sk->sk_max_ack_backlog);
991 * Then please take a look at commit 64a146513f8f ("[NET]: Revert incorrect accept queue backlog changes.")
992 */
sk_acceptq_is_full(const struct sock * sk)993 static inline bool sk_acceptq_is_full(const struct sock *sk)
994 {
995 return READ_ONCE(sk->sk_ack_backlog) > READ_ONCE(sk->sk_max_ack_backlog);
996 }
997
998 /*
999 * Compute minimal free write space needed to queue new packets.
1000 */
sk_stream_min_wspace(const struct sock * sk)1001 static inline int sk_stream_min_wspace(const struct sock *sk)
1002 {
1003 return READ_ONCE(sk->sk_wmem_queued) >> 1;
1004 }
1005
sk_stream_wspace(const struct sock * sk)1006 static inline int sk_stream_wspace(const struct sock *sk)
1007 {
1008 return READ_ONCE(sk->sk_sndbuf) - READ_ONCE(sk->sk_wmem_queued);
1009 }
1010
sk_wmem_queued_add(struct sock * sk,int val)1011 static inline void sk_wmem_queued_add(struct sock *sk, int val)
1012 {
1013 WRITE_ONCE(sk->sk_wmem_queued, sk->sk_wmem_queued + val);
1014 }
1015
1016 void sk_stream_write_space(struct sock *sk);
1017
1018 /* OOB backlog add */
__sk_add_backlog(struct sock * sk,struct sk_buff * skb)1019 static inline void __sk_add_backlog(struct sock *sk, struct sk_buff *skb)
1020 {
1021 /* dont let skb dst not refcounted, we are going to leave rcu lock */
1022 skb_dst_force(skb);
1023
1024 if (!sk->sk_backlog.tail)
1025 WRITE_ONCE(sk->sk_backlog.head, skb);
1026 else
1027 sk->sk_backlog.tail->next = skb;
1028
1029 WRITE_ONCE(sk->sk_backlog.tail, skb);
1030 skb->next = NULL;
1031 }
1032
1033 /*
1034 * Take into account size of receive queue and backlog queue
1035 * Do not take into account this skb truesize,
1036 * to allow even a single big packet to come.
1037 */
sk_rcvqueues_full(const struct sock * sk,unsigned int limit)1038 static inline bool sk_rcvqueues_full(const struct sock *sk, unsigned int limit)
1039 {
1040 unsigned int qsize = sk->sk_backlog.len + atomic_read(&sk->sk_rmem_alloc);
1041
1042 return qsize > limit;
1043 }
1044
1045 /* The per-socket spinlock must be held here. */
sk_add_backlog(struct sock * sk,struct sk_buff * skb,unsigned int limit)1046 static inline __must_check int sk_add_backlog(struct sock *sk, struct sk_buff *skb,
1047 unsigned int limit)
1048 {
1049 if (sk_rcvqueues_full(sk, limit))
1050 return -ENOBUFS;
1051
1052 /*
1053 * If the skb was allocated from pfmemalloc reserves, only
1054 * allow SOCK_MEMALLOC sockets to use it as this socket is
1055 * helping free memory
1056 */
1057 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC))
1058 return -ENOMEM;
1059
1060 __sk_add_backlog(sk, skb);
1061 sk->sk_backlog.len += skb->truesize;
1062 return 0;
1063 }
1064
1065 int __sk_backlog_rcv(struct sock *sk, struct sk_buff *skb);
1066
sk_backlog_rcv(struct sock * sk,struct sk_buff * skb)1067 static inline int sk_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1068 {
1069 if (sk_memalloc_socks() && skb_pfmemalloc(skb))
1070 return __sk_backlog_rcv(sk, skb);
1071
1072 return sk->sk_backlog_rcv(sk, skb);
1073 }
1074
sk_incoming_cpu_update(struct sock * sk)1075 static inline void sk_incoming_cpu_update(struct sock *sk)
1076 {
1077 int cpu = raw_smp_processor_id();
1078
1079 if (unlikely(READ_ONCE(sk->sk_incoming_cpu) != cpu))
1080 WRITE_ONCE(sk->sk_incoming_cpu, cpu);
1081 }
1082
sock_rps_record_flow_hash(__u32 hash)1083 static inline void sock_rps_record_flow_hash(__u32 hash)
1084 {
1085 #ifdef CONFIG_RPS
1086 struct rps_sock_flow_table *sock_flow_table;
1087
1088 rcu_read_lock();
1089 sock_flow_table = rcu_dereference(rps_sock_flow_table);
1090 rps_record_sock_flow(sock_flow_table, hash);
1091 rcu_read_unlock();
1092 #endif
1093 }
1094
sock_rps_record_flow(const struct sock * sk)1095 static inline void sock_rps_record_flow(const struct sock *sk)
1096 {
1097 #ifdef CONFIG_RPS
1098 if (static_branch_unlikely(&rfs_needed)) {
1099 /* Reading sk->sk_rxhash might incur an expensive cache line
1100 * miss.
1101 *
1102 * TCP_ESTABLISHED does cover almost all states where RFS
1103 * might be useful, and is cheaper [1] than testing :
1104 * IPv4: inet_sk(sk)->inet_daddr
1105 * IPv6: ipv6_addr_any(&sk->sk_v6_daddr)
1106 * OR an additional socket flag
1107 * [1] : sk_state and sk_prot are in the same cache line.
1108 */
1109 if (sk->sk_state == TCP_ESTABLISHED) {
1110 /* This READ_ONCE() is paired with the WRITE_ONCE()
1111 * from sock_rps_save_rxhash() and sock_rps_reset_rxhash().
1112 */
1113 sock_rps_record_flow_hash(READ_ONCE(sk->sk_rxhash));
1114 }
1115 }
1116 #endif
1117 }
1118
sock_rps_save_rxhash(struct sock * sk,const struct sk_buff * skb)1119 static inline void sock_rps_save_rxhash(struct sock *sk,
1120 const struct sk_buff *skb)
1121 {
1122 #ifdef CONFIG_RPS
1123 /* The following WRITE_ONCE() is paired with the READ_ONCE()
1124 * here, and another one in sock_rps_record_flow().
1125 */
1126 if (unlikely(READ_ONCE(sk->sk_rxhash) != skb->hash))
1127 WRITE_ONCE(sk->sk_rxhash, skb->hash);
1128 #endif
1129 }
1130
sock_rps_reset_rxhash(struct sock * sk)1131 static inline void sock_rps_reset_rxhash(struct sock *sk)
1132 {
1133 #ifdef CONFIG_RPS
1134 /* Paired with READ_ONCE() in sock_rps_record_flow() */
1135 WRITE_ONCE(sk->sk_rxhash, 0);
1136 #endif
1137 }
1138
1139 #define sk_wait_event(__sk, __timeo, __condition, __wait) \
1140 ({ int __rc; \
1141 __sk->sk_wait_pending++; \
1142 release_sock(__sk); \
1143 __rc = __condition; \
1144 if (!__rc) { \
1145 *(__timeo) = wait_woken(__wait, \
1146 TASK_INTERRUPTIBLE, \
1147 *(__timeo)); \
1148 } \
1149 sched_annotate_sleep(); \
1150 lock_sock(__sk); \
1151 __sk->sk_wait_pending--; \
1152 __rc = __condition; \
1153 __rc; \
1154 })
1155
1156 int sk_stream_wait_connect(struct sock *sk, long *timeo_p);
1157 int sk_stream_wait_memory(struct sock *sk, long *timeo_p);
1158 void sk_stream_wait_close(struct sock *sk, long timeo_p);
1159 int sk_stream_error(struct sock *sk, int flags, int err);
1160 void sk_stream_kill_queues(struct sock *sk);
1161 void sk_set_memalloc(struct sock *sk);
1162 void sk_clear_memalloc(struct sock *sk);
1163
1164 void __sk_flush_backlog(struct sock *sk);
1165
sk_flush_backlog(struct sock * sk)1166 static inline bool sk_flush_backlog(struct sock *sk)
1167 {
1168 if (unlikely(READ_ONCE(sk->sk_backlog.tail))) {
1169 __sk_flush_backlog(sk);
1170 return true;
1171 }
1172 return false;
1173 }
1174
1175 int sk_wait_data(struct sock *sk, long *timeo, const struct sk_buff *skb);
1176
1177 struct request_sock_ops;
1178 struct timewait_sock_ops;
1179 struct inet_hashinfo;
1180 struct raw_hashinfo;
1181 struct smc_hashinfo;
1182 struct module;
1183 struct sk_psock;
1184
1185 /*
1186 * caches using SLAB_TYPESAFE_BY_RCU should let .next pointer from nulls nodes
1187 * un-modified. Special care is taken when initializing object to zero.
1188 */
sk_prot_clear_nulls(struct sock * sk,int size)1189 static inline void sk_prot_clear_nulls(struct sock *sk, int size)
1190 {
1191 if (offsetof(struct sock, sk_node.next) != 0)
1192 memset(sk, 0, offsetof(struct sock, sk_node.next));
1193 memset(&sk->sk_node.pprev, 0,
1194 size - offsetof(struct sock, sk_node.pprev));
1195 }
1196
1197 /* Networking protocol blocks we attach to sockets.
1198 * socket layer -> transport layer interface
1199 */
1200 struct proto {
1201 void (*close)(struct sock *sk,
1202 long timeout);
1203 int (*pre_connect)(struct sock *sk,
1204 struct sockaddr *uaddr,
1205 int addr_len);
1206 int (*connect)(struct sock *sk,
1207 struct sockaddr *uaddr,
1208 int addr_len);
1209 int (*disconnect)(struct sock *sk, int flags);
1210
1211 struct sock * (*accept)(struct sock *sk, int flags, int *err,
1212 bool kern);
1213
1214 int (*ioctl)(struct sock *sk, int cmd,
1215 unsigned long arg);
1216 int (*init)(struct sock *sk);
1217 void (*destroy)(struct sock *sk);
1218 void (*shutdown)(struct sock *sk, int how);
1219 int (*setsockopt)(struct sock *sk, int level,
1220 int optname, sockptr_t optval,
1221 unsigned int optlen);
1222 int (*getsockopt)(struct sock *sk, int level,
1223 int optname, char __user *optval,
1224 int __user *option);
1225 void (*keepalive)(struct sock *sk, int valbool);
1226 #ifdef CONFIG_COMPAT
1227 int (*compat_ioctl)(struct sock *sk,
1228 unsigned int cmd, unsigned long arg);
1229 #endif
1230 int (*sendmsg)(struct sock *sk, struct msghdr *msg,
1231 size_t len);
1232 int (*recvmsg)(struct sock *sk, struct msghdr *msg,
1233 size_t len, int noblock, int flags,
1234 int *addr_len);
1235 int (*sendpage)(struct sock *sk, struct page *page,
1236 int offset, size_t size, int flags);
1237 int (*bind)(struct sock *sk,
1238 struct sockaddr *addr, int addr_len);
1239 int (*bind_add)(struct sock *sk,
1240 struct sockaddr *addr, int addr_len);
1241
1242 int (*backlog_rcv) (struct sock *sk,
1243 struct sk_buff *skb);
1244 bool (*bpf_bypass_getsockopt)(int level,
1245 int optname);
1246
1247 void (*release_cb)(struct sock *sk);
1248
1249 /* Keeping track of sk's, looking them up, and port selection methods. */
1250 int (*hash)(struct sock *sk);
1251 void (*unhash)(struct sock *sk);
1252 void (*rehash)(struct sock *sk);
1253 int (*get_port)(struct sock *sk, unsigned short snum);
1254 #ifdef CONFIG_BPF_SYSCALL
1255 int (*psock_update_sk_prot)(struct sock *sk,
1256 struct sk_psock *psock,
1257 bool restore);
1258 #endif
1259
1260 /* Keeping track of sockets in use */
1261 #ifdef CONFIG_PROC_FS
1262 unsigned int inuse_idx;
1263 #endif
1264
1265 bool (*stream_memory_free)(const struct sock *sk, int wake);
1266 bool (*sock_is_readable)(struct sock *sk);
1267 /* Memory pressure */
1268 void (*enter_memory_pressure)(struct sock *sk);
1269 void (*leave_memory_pressure)(struct sock *sk);
1270 atomic_long_t *memory_allocated; /* Current allocated memory. */
1271 struct percpu_counter *sockets_allocated; /* Current number of sockets. */
1272 /*
1273 * Pressure flag: try to collapse.
1274 * Technical note: it is used by multiple contexts non atomically.
1275 * Make sure to use READ_ONCE()/WRITE_ONCE() for all reads/writes.
1276 * All the __sk_mem_schedule() is of this nature: accounting
1277 * is strict, actions are advisory and have some latency.
1278 */
1279 unsigned long *memory_pressure;
1280 long *sysctl_mem;
1281
1282 int *sysctl_wmem;
1283 int *sysctl_rmem;
1284 u32 sysctl_wmem_offset;
1285 u32 sysctl_rmem_offset;
1286
1287 int max_header;
1288 bool no_autobind;
1289
1290 struct kmem_cache *slab;
1291 unsigned int obj_size;
1292 slab_flags_t slab_flags;
1293 unsigned int useroffset; /* Usercopy region offset */
1294 unsigned int usersize; /* Usercopy region size */
1295
1296 unsigned int __percpu *orphan_count;
1297
1298 struct request_sock_ops *rsk_prot;
1299 struct timewait_sock_ops *twsk_prot;
1300
1301 union {
1302 struct inet_hashinfo *hashinfo;
1303 struct udp_table *udp_table;
1304 struct raw_hashinfo *raw_hash;
1305 struct smc_hashinfo *smc_hash;
1306 } h;
1307
1308 struct module *owner;
1309
1310 char name[32];
1311
1312 struct list_head node;
1313 #ifdef SOCK_REFCNT_DEBUG
1314 atomic_t socks;
1315 #endif
1316 int (*diag_destroy)(struct sock *sk, int err);
1317 } __randomize_layout;
1318
1319 int proto_register(struct proto *prot, int alloc_slab);
1320 void proto_unregister(struct proto *prot);
1321 int sock_load_diag_module(int family, int protocol);
1322
1323 #ifdef SOCK_REFCNT_DEBUG
sk_refcnt_debug_inc(struct sock * sk)1324 static inline void sk_refcnt_debug_inc(struct sock *sk)
1325 {
1326 atomic_inc(&sk->sk_prot->socks);
1327 }
1328
sk_refcnt_debug_dec(struct sock * sk)1329 static inline void sk_refcnt_debug_dec(struct sock *sk)
1330 {
1331 atomic_dec(&sk->sk_prot->socks);
1332 printk(KERN_DEBUG "%s socket %p released, %d are still alive\n",
1333 sk->sk_prot->name, sk, atomic_read(&sk->sk_prot->socks));
1334 }
1335
sk_refcnt_debug_release(const struct sock * sk)1336 static inline void sk_refcnt_debug_release(const struct sock *sk)
1337 {
1338 if (refcount_read(&sk->sk_refcnt) != 1)
1339 printk(KERN_DEBUG "Destruction of the %s socket %p delayed, refcnt=%d\n",
1340 sk->sk_prot->name, sk, refcount_read(&sk->sk_refcnt));
1341 }
1342 #else /* SOCK_REFCNT_DEBUG */
1343 #define sk_refcnt_debug_inc(sk) do { } while (0)
1344 #define sk_refcnt_debug_dec(sk) do { } while (0)
1345 #define sk_refcnt_debug_release(sk) do { } while (0)
1346 #endif /* SOCK_REFCNT_DEBUG */
1347
1348 INDIRECT_CALLABLE_DECLARE(bool tcp_stream_memory_free(const struct sock *sk, int wake));
1349
__sk_stream_memory_free(const struct sock * sk,int wake)1350 static inline bool __sk_stream_memory_free(const struct sock *sk, int wake)
1351 {
1352 if (READ_ONCE(sk->sk_wmem_queued) >= READ_ONCE(sk->sk_sndbuf))
1353 return false;
1354
1355 #ifdef CONFIG_INET
1356 return sk->sk_prot->stream_memory_free ?
1357 INDIRECT_CALL_1(sk->sk_prot->stream_memory_free,
1358 tcp_stream_memory_free,
1359 sk, wake) : true;
1360 #else
1361 return sk->sk_prot->stream_memory_free ?
1362 sk->sk_prot->stream_memory_free(sk, wake) : true;
1363 #endif
1364 }
1365
sk_stream_memory_free(const struct sock * sk)1366 static inline bool sk_stream_memory_free(const struct sock *sk)
1367 {
1368 return __sk_stream_memory_free(sk, 0);
1369 }
1370
__sk_stream_is_writeable(const struct sock * sk,int wake)1371 static inline bool __sk_stream_is_writeable(const struct sock *sk, int wake)
1372 {
1373 return sk_stream_wspace(sk) >= sk_stream_min_wspace(sk) &&
1374 __sk_stream_memory_free(sk, wake);
1375 }
1376
sk_stream_is_writeable(const struct sock * sk)1377 static inline bool sk_stream_is_writeable(const struct sock *sk)
1378 {
1379 return __sk_stream_is_writeable(sk, 0);
1380 }
1381
sk_under_cgroup_hierarchy(struct sock * sk,struct cgroup * ancestor)1382 static inline int sk_under_cgroup_hierarchy(struct sock *sk,
1383 struct cgroup *ancestor)
1384 {
1385 #ifdef CONFIG_SOCK_CGROUP_DATA
1386 return cgroup_is_descendant(sock_cgroup_ptr(&sk->sk_cgrp_data),
1387 ancestor);
1388 #else
1389 return -ENOTSUPP;
1390 #endif
1391 }
1392
sk_has_memory_pressure(const struct sock * sk)1393 static inline bool sk_has_memory_pressure(const struct sock *sk)
1394 {
1395 return sk->sk_prot->memory_pressure != NULL;
1396 }
1397
sk_under_global_memory_pressure(const struct sock * sk)1398 static inline bool sk_under_global_memory_pressure(const struct sock *sk)
1399 {
1400 return sk->sk_prot->memory_pressure &&
1401 !!READ_ONCE(*sk->sk_prot->memory_pressure);
1402 }
1403
sk_under_memory_pressure(const struct sock * sk)1404 static inline bool sk_under_memory_pressure(const struct sock *sk)
1405 {
1406 if (!sk->sk_prot->memory_pressure)
1407 return false;
1408
1409 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
1410 mem_cgroup_under_socket_pressure(sk->sk_memcg))
1411 return true;
1412
1413 return !!READ_ONCE(*sk->sk_prot->memory_pressure);
1414 }
1415
1416 static inline long
sk_memory_allocated(const struct sock * sk)1417 sk_memory_allocated(const struct sock *sk)
1418 {
1419 return atomic_long_read(sk->sk_prot->memory_allocated);
1420 }
1421
1422 static inline long
sk_memory_allocated_add(struct sock * sk,int amt)1423 sk_memory_allocated_add(struct sock *sk, int amt)
1424 {
1425 return atomic_long_add_return(amt, sk->sk_prot->memory_allocated);
1426 }
1427
1428 static inline void
sk_memory_allocated_sub(struct sock * sk,int amt)1429 sk_memory_allocated_sub(struct sock *sk, int amt)
1430 {
1431 atomic_long_sub(amt, sk->sk_prot->memory_allocated);
1432 }
1433
1434 #define SK_ALLOC_PERCPU_COUNTER_BATCH 16
1435
sk_sockets_allocated_dec(struct sock * sk)1436 static inline void sk_sockets_allocated_dec(struct sock *sk)
1437 {
1438 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, -1,
1439 SK_ALLOC_PERCPU_COUNTER_BATCH);
1440 }
1441
sk_sockets_allocated_inc(struct sock * sk)1442 static inline void sk_sockets_allocated_inc(struct sock *sk)
1443 {
1444 percpu_counter_add_batch(sk->sk_prot->sockets_allocated, 1,
1445 SK_ALLOC_PERCPU_COUNTER_BATCH);
1446 }
1447
1448 static inline u64
sk_sockets_allocated_read_positive(struct sock * sk)1449 sk_sockets_allocated_read_positive(struct sock *sk)
1450 {
1451 return percpu_counter_read_positive(sk->sk_prot->sockets_allocated);
1452 }
1453
1454 static inline int
proto_sockets_allocated_sum_positive(struct proto * prot)1455 proto_sockets_allocated_sum_positive(struct proto *prot)
1456 {
1457 return percpu_counter_sum_positive(prot->sockets_allocated);
1458 }
1459
1460 static inline long
proto_memory_allocated(struct proto * prot)1461 proto_memory_allocated(struct proto *prot)
1462 {
1463 return atomic_long_read(prot->memory_allocated);
1464 }
1465
1466 static inline bool
proto_memory_pressure(struct proto * prot)1467 proto_memory_pressure(struct proto *prot)
1468 {
1469 if (!prot->memory_pressure)
1470 return false;
1471 return !!READ_ONCE(*prot->memory_pressure);
1472 }
1473
1474
1475 #ifdef CONFIG_PROC_FS
1476 /* Called with local bh disabled */
1477 void sock_prot_inuse_add(struct net *net, struct proto *prot, int inc);
1478 int sock_prot_inuse_get(struct net *net, struct proto *proto);
1479 int sock_inuse_get(struct net *net);
1480 #else
sock_prot_inuse_add(struct net * net,struct proto * prot,int inc)1481 static inline void sock_prot_inuse_add(struct net *net, struct proto *prot,
1482 int inc)
1483 {
1484 }
1485 #endif
1486
1487
1488 /* With per-bucket locks this operation is not-atomic, so that
1489 * this version is not worse.
1490 */
__sk_prot_rehash(struct sock * sk)1491 static inline int __sk_prot_rehash(struct sock *sk)
1492 {
1493 sk->sk_prot->unhash(sk);
1494 return sk->sk_prot->hash(sk);
1495 }
1496
1497 /* About 10 seconds */
1498 #define SOCK_DESTROY_TIME (10*HZ)
1499
1500 /* Sockets 0-1023 can't be bound to unless you are superuser */
1501 #define PROT_SOCK 1024
1502
1503 #define SHUTDOWN_MASK 3
1504 #define RCV_SHUTDOWN 1
1505 #define SEND_SHUTDOWN 2
1506
1507 #define SOCK_BINDADDR_LOCK 4
1508 #define SOCK_BINDPORT_LOCK 8
1509
1510 struct socket_alloc {
1511 struct socket socket;
1512 struct inode vfs_inode;
1513 };
1514
SOCKET_I(struct inode * inode)1515 static inline struct socket *SOCKET_I(struct inode *inode)
1516 {
1517 return &container_of(inode, struct socket_alloc, vfs_inode)->socket;
1518 }
1519
SOCK_INODE(struct socket * socket)1520 static inline struct inode *SOCK_INODE(struct socket *socket)
1521 {
1522 return &container_of(socket, struct socket_alloc, socket)->vfs_inode;
1523 }
1524
1525 /*
1526 * Functions for memory accounting
1527 */
1528 int __sk_mem_raise_allocated(struct sock *sk, int size, int amt, int kind);
1529 int __sk_mem_schedule(struct sock *sk, int size, int kind);
1530 void __sk_mem_reduce_allocated(struct sock *sk, int amount);
1531 void __sk_mem_reclaim(struct sock *sk, int amount);
1532
1533 /* We used to have PAGE_SIZE here, but systems with 64KB pages
1534 * do not necessarily have 16x time more memory than 4KB ones.
1535 */
1536 #define SK_MEM_QUANTUM 4096
1537 #define SK_MEM_QUANTUM_SHIFT ilog2(SK_MEM_QUANTUM)
1538 #define SK_MEM_SEND 0
1539 #define SK_MEM_RECV 1
1540
1541 /* sysctl_mem values are in pages, we convert them in SK_MEM_QUANTUM units */
sk_prot_mem_limits(const struct sock * sk,int index)1542 static inline long sk_prot_mem_limits(const struct sock *sk, int index)
1543 {
1544 long val = READ_ONCE(sk->sk_prot->sysctl_mem[index]);
1545
1546 #if PAGE_SIZE > SK_MEM_QUANTUM
1547 val <<= PAGE_SHIFT - SK_MEM_QUANTUM_SHIFT;
1548 #elif PAGE_SIZE < SK_MEM_QUANTUM
1549 val >>= SK_MEM_QUANTUM_SHIFT - PAGE_SHIFT;
1550 #endif
1551 return val;
1552 }
1553
sk_mem_pages(int amt)1554 static inline int sk_mem_pages(int amt)
1555 {
1556 return (amt + SK_MEM_QUANTUM - 1) >> SK_MEM_QUANTUM_SHIFT;
1557 }
1558
sk_has_account(struct sock * sk)1559 static inline bool sk_has_account(struct sock *sk)
1560 {
1561 /* return true if protocol supports memory accounting */
1562 return !!sk->sk_prot->memory_allocated;
1563 }
1564
sk_wmem_schedule(struct sock * sk,int size)1565 static inline bool sk_wmem_schedule(struct sock *sk, int size)
1566 {
1567 int delta;
1568
1569 if (!sk_has_account(sk))
1570 return true;
1571 delta = size - sk->sk_forward_alloc;
1572 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_SEND);
1573 }
1574
1575 static inline bool
sk_rmem_schedule(struct sock * sk,struct sk_buff * skb,int size)1576 sk_rmem_schedule(struct sock *sk, struct sk_buff *skb, int size)
1577 {
1578 int delta;
1579
1580 if (!sk_has_account(sk))
1581 return true;
1582 delta = size - sk->sk_forward_alloc;
1583 return delta <= 0 || __sk_mem_schedule(sk, delta, SK_MEM_RECV) ||
1584 skb_pfmemalloc(skb);
1585 }
1586
sk_mem_reclaim(struct sock * sk)1587 static inline void sk_mem_reclaim(struct sock *sk)
1588 {
1589 if (!sk_has_account(sk))
1590 return;
1591 if (sk->sk_forward_alloc >= SK_MEM_QUANTUM)
1592 __sk_mem_reclaim(sk, sk->sk_forward_alloc);
1593 }
1594
sk_mem_reclaim_partial(struct sock * sk)1595 static inline void sk_mem_reclaim_partial(struct sock *sk)
1596 {
1597 if (!sk_has_account(sk))
1598 return;
1599 if (sk->sk_forward_alloc > SK_MEM_QUANTUM)
1600 __sk_mem_reclaim(sk, sk->sk_forward_alloc - 1);
1601 }
1602
sk_mem_charge(struct sock * sk,int size)1603 static inline void sk_mem_charge(struct sock *sk, int size)
1604 {
1605 if (!sk_has_account(sk))
1606 return;
1607 sk->sk_forward_alloc -= size;
1608 }
1609
sk_mem_uncharge(struct sock * sk,int size)1610 static inline void sk_mem_uncharge(struct sock *sk, int size)
1611 {
1612 if (!sk_has_account(sk))
1613 return;
1614 sk->sk_forward_alloc += size;
1615
1616 /* Avoid a possible overflow.
1617 * TCP send queues can make this happen, if sk_mem_reclaim()
1618 * is not called and more than 2 GBytes are released at once.
1619 *
1620 * If we reach 2 MBytes, reclaim 1 MBytes right now, there is
1621 * no need to hold that much forward allocation anyway.
1622 */
1623 if (unlikely(sk->sk_forward_alloc >= 1 << 21))
1624 __sk_mem_reclaim(sk, 1 << 20);
1625 }
1626
1627 DECLARE_STATIC_KEY_FALSE(tcp_tx_skb_cache_key);
sk_wmem_free_skb(struct sock * sk,struct sk_buff * skb)1628 static inline void sk_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
1629 {
1630 sk_wmem_queued_add(sk, -skb->truesize);
1631 sk_mem_uncharge(sk, skb->truesize);
1632 if (static_branch_unlikely(&tcp_tx_skb_cache_key) &&
1633 !sk->sk_tx_skb_cache && !skb_cloned(skb)) {
1634 skb_ext_reset(skb);
1635 skb_zcopy_clear(skb, true);
1636 sk->sk_tx_skb_cache = skb;
1637 return;
1638 }
1639 __kfree_skb(skb);
1640 }
1641
sock_release_ownership(struct sock * sk)1642 static inline void sock_release_ownership(struct sock *sk)
1643 {
1644 if (sk->sk_lock.owned) {
1645 sk->sk_lock.owned = 0;
1646
1647 /* The sk_lock has mutex_unlock() semantics: */
1648 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1649 }
1650 }
1651
1652 /*
1653 * Macro so as to not evaluate some arguments when
1654 * lockdep is not enabled.
1655 *
1656 * Mark both the sk_lock and the sk_lock.slock as a
1657 * per-address-family lock class.
1658 */
1659 #define sock_lock_init_class_and_name(sk, sname, skey, name, key) \
1660 do { \
1661 sk->sk_lock.owned = 0; \
1662 init_waitqueue_head(&sk->sk_lock.wq); \
1663 spin_lock_init(&(sk)->sk_lock.slock); \
1664 debug_check_no_locks_freed((void *)&(sk)->sk_lock, \
1665 sizeof((sk)->sk_lock)); \
1666 lockdep_set_class_and_name(&(sk)->sk_lock.slock, \
1667 (skey), (sname)); \
1668 lockdep_init_map(&(sk)->sk_lock.dep_map, (name), (key), 0); \
1669 } while (0)
1670
lockdep_sock_is_held(const struct sock * sk)1671 static inline bool lockdep_sock_is_held(const struct sock *sk)
1672 {
1673 return lockdep_is_held(&sk->sk_lock) ||
1674 lockdep_is_held(&sk->sk_lock.slock);
1675 }
1676
1677 void lock_sock_nested(struct sock *sk, int subclass);
1678
lock_sock(struct sock * sk)1679 static inline void lock_sock(struct sock *sk)
1680 {
1681 lock_sock_nested(sk, 0);
1682 }
1683
1684 void __lock_sock(struct sock *sk);
1685 void __release_sock(struct sock *sk);
1686 void release_sock(struct sock *sk);
1687
1688 /* BH context may only use the following locking interface. */
1689 #define bh_lock_sock(__sk) spin_lock(&((__sk)->sk_lock.slock))
1690 #define bh_lock_sock_nested(__sk) \
1691 spin_lock_nested(&((__sk)->sk_lock.slock), \
1692 SINGLE_DEPTH_NESTING)
1693 #define bh_unlock_sock(__sk) spin_unlock(&((__sk)->sk_lock.slock))
1694
1695 bool __lock_sock_fast(struct sock *sk) __acquires(&sk->sk_lock.slock);
1696
1697 /**
1698 * lock_sock_fast - fast version of lock_sock
1699 * @sk: socket
1700 *
1701 * This version should be used for very small section, where process wont block
1702 * return false if fast path is taken:
1703 *
1704 * sk_lock.slock locked, owned = 0, BH disabled
1705 *
1706 * return true if slow path is taken:
1707 *
1708 * sk_lock.slock unlocked, owned = 1, BH enabled
1709 */
lock_sock_fast(struct sock * sk)1710 static inline bool lock_sock_fast(struct sock *sk)
1711 {
1712 /* The sk_lock has mutex_lock() semantics here. */
1713 mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
1714
1715 return __lock_sock_fast(sk);
1716 }
1717
1718 /* fast socket lock variant for caller already holding a [different] socket lock */
lock_sock_fast_nested(struct sock * sk)1719 static inline bool lock_sock_fast_nested(struct sock *sk)
1720 {
1721 mutex_acquire(&sk->sk_lock.dep_map, SINGLE_DEPTH_NESTING, 0, _RET_IP_);
1722
1723 return __lock_sock_fast(sk);
1724 }
1725
1726 /**
1727 * unlock_sock_fast - complement of lock_sock_fast
1728 * @sk: socket
1729 * @slow: slow mode
1730 *
1731 * fast unlock socket for user context.
1732 * If slow mode is on, we call regular release_sock()
1733 */
unlock_sock_fast(struct sock * sk,bool slow)1734 static inline void unlock_sock_fast(struct sock *sk, bool slow)
1735 __releases(&sk->sk_lock.slock)
1736 {
1737 if (slow) {
1738 release_sock(sk);
1739 __release(&sk->sk_lock.slock);
1740 } else {
1741 mutex_release(&sk->sk_lock.dep_map, _RET_IP_);
1742 spin_unlock_bh(&sk->sk_lock.slock);
1743 }
1744 }
1745
1746 /* Used by processes to "lock" a socket state, so that
1747 * interrupts and bottom half handlers won't change it
1748 * from under us. It essentially blocks any incoming
1749 * packets, so that we won't get any new data or any
1750 * packets that change the state of the socket.
1751 *
1752 * While locked, BH processing will add new packets to
1753 * the backlog queue. This queue is processed by the
1754 * owner of the socket lock right before it is released.
1755 *
1756 * Since ~2.3.5 it is also exclusive sleep lock serializing
1757 * accesses from user process context.
1758 */
1759
sock_owned_by_me(const struct sock * sk)1760 static inline void sock_owned_by_me(const struct sock *sk)
1761 {
1762 #ifdef CONFIG_LOCKDEP
1763 WARN_ON_ONCE(!lockdep_sock_is_held(sk) && debug_locks);
1764 #endif
1765 }
1766
sock_owned_by_user(const struct sock * sk)1767 static inline bool sock_owned_by_user(const struct sock *sk)
1768 {
1769 sock_owned_by_me(sk);
1770 return sk->sk_lock.owned;
1771 }
1772
sock_owned_by_user_nocheck(const struct sock * sk)1773 static inline bool sock_owned_by_user_nocheck(const struct sock *sk)
1774 {
1775 return sk->sk_lock.owned;
1776 }
1777
1778 /* no reclassification while locks are held */
sock_allow_reclassification(const struct sock * csk)1779 static inline bool sock_allow_reclassification(const struct sock *csk)
1780 {
1781 struct sock *sk = (struct sock *)csk;
1782
1783 return !sk->sk_lock.owned && !spin_is_locked(&sk->sk_lock.slock);
1784 }
1785
1786 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1787 struct proto *prot, int kern);
1788 void sk_free(struct sock *sk);
1789 void sk_destruct(struct sock *sk);
1790 struct sock *sk_clone_lock(const struct sock *sk, const gfp_t priority);
1791 void sk_free_unlock_clone(struct sock *sk);
1792
1793 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1794 gfp_t priority);
1795 void __sock_wfree(struct sk_buff *skb);
1796 void sock_wfree(struct sk_buff *skb);
1797 struct sk_buff *sock_omalloc(struct sock *sk, unsigned long size,
1798 gfp_t priority);
1799 void skb_orphan_partial(struct sk_buff *skb);
1800 void sock_rfree(struct sk_buff *skb);
1801 void sock_efree(struct sk_buff *skb);
1802 #ifdef CONFIG_INET
1803 void sock_edemux(struct sk_buff *skb);
1804 void sock_pfree(struct sk_buff *skb);
1805 #else
1806 #define sock_edemux sock_efree
1807 #endif
1808
1809 int sock_setsockopt(struct socket *sock, int level, int op,
1810 sockptr_t optval, unsigned int optlen);
1811
1812 int sock_getsockopt(struct socket *sock, int level, int op,
1813 char __user *optval, int __user *optlen);
1814 int sock_gettstamp(struct socket *sock, void __user *userstamp,
1815 bool timeval, bool time32);
1816 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1817 int noblock, int *errcode);
1818 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1819 unsigned long data_len, int noblock,
1820 int *errcode, int max_page_order);
1821 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority);
1822 void sock_kfree_s(struct sock *sk, void *mem, int size);
1823 void sock_kzfree_s(struct sock *sk, void *mem, int size);
1824 void sk_send_sigurg(struct sock *sk);
1825
1826 struct sockcm_cookie {
1827 u64 transmit_time;
1828 u32 mark;
1829 u16 tsflags;
1830 };
1831
sockcm_init(struct sockcm_cookie * sockc,const struct sock * sk)1832 static inline void sockcm_init(struct sockcm_cookie *sockc,
1833 const struct sock *sk)
1834 {
1835 *sockc = (struct sockcm_cookie) { .tsflags = sk->sk_tsflags };
1836 }
1837
1838 int __sock_cmsg_send(struct sock *sk, struct msghdr *msg, struct cmsghdr *cmsg,
1839 struct sockcm_cookie *sockc);
1840 int sock_cmsg_send(struct sock *sk, struct msghdr *msg,
1841 struct sockcm_cookie *sockc);
1842
1843 /*
1844 * Functions to fill in entries in struct proto_ops when a protocol
1845 * does not implement a particular function.
1846 */
1847 int sock_no_bind(struct socket *, struct sockaddr *, int);
1848 int sock_no_connect(struct socket *, struct sockaddr *, int, int);
1849 int sock_no_socketpair(struct socket *, struct socket *);
1850 int sock_no_accept(struct socket *, struct socket *, int, bool);
1851 int sock_no_getname(struct socket *, struct sockaddr *, int);
1852 int sock_no_ioctl(struct socket *, unsigned int, unsigned long);
1853 int sock_no_listen(struct socket *, int);
1854 int sock_no_shutdown(struct socket *, int);
1855 int sock_no_sendmsg(struct socket *, struct msghdr *, size_t);
1856 int sock_no_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t len);
1857 int sock_no_recvmsg(struct socket *, struct msghdr *, size_t, int);
1858 int sock_no_mmap(struct file *file, struct socket *sock,
1859 struct vm_area_struct *vma);
1860 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset,
1861 size_t size, int flags);
1862 ssize_t sock_no_sendpage_locked(struct sock *sk, struct page *page,
1863 int offset, size_t size, int flags);
1864
1865 /*
1866 * Functions to fill in entries in struct proto_ops when a protocol
1867 * uses the inet style.
1868 */
1869 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1870 char __user *optval, int __user *optlen);
1871 int sock_common_recvmsg(struct socket *sock, struct msghdr *msg, size_t size,
1872 int flags);
1873 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1874 sockptr_t optval, unsigned int optlen);
1875
1876 void sk_common_release(struct sock *sk);
1877
1878 /*
1879 * Default socket callbacks and setup code
1880 */
1881
1882 /* Initialise core socket variables using an explicit uid. */
1883 void sock_init_data_uid(struct socket *sock, struct sock *sk, kuid_t uid);
1884
1885 /* Initialise core socket variables.
1886 * Assumes struct socket *sock is embedded in a struct socket_alloc.
1887 */
1888 void sock_init_data(struct socket *sock, struct sock *sk);
1889
1890 /*
1891 * Socket reference counting postulates.
1892 *
1893 * * Each user of socket SHOULD hold a reference count.
1894 * * Each access point to socket (an hash table bucket, reference from a list,
1895 * running timer, skb in flight MUST hold a reference count.
1896 * * When reference count hits 0, it means it will never increase back.
1897 * * When reference count hits 0, it means that no references from
1898 * outside exist to this socket and current process on current CPU
1899 * is last user and may/should destroy this socket.
1900 * * sk_free is called from any context: process, BH, IRQ. When
1901 * it is called, socket has no references from outside -> sk_free
1902 * may release descendant resources allocated by the socket, but
1903 * to the time when it is called, socket is NOT referenced by any
1904 * hash tables, lists etc.
1905 * * Packets, delivered from outside (from network or from another process)
1906 * and enqueued on receive/error queues SHOULD NOT grab reference count,
1907 * when they sit in queue. Otherwise, packets will leak to hole, when
1908 * socket is looked up by one cpu and unhasing is made by another CPU.
1909 * It is true for udp/raw, netlink (leak to receive and error queues), tcp
1910 * (leak to backlog). Packet socket does all the processing inside
1911 * BR_NETPROTO_LOCK, so that it has not this race condition. UNIX sockets
1912 * use separate SMP lock, so that they are prone too.
1913 */
1914
1915 /* Ungrab socket and destroy it, if it was the last reference. */
sock_put(struct sock * sk)1916 static inline void sock_put(struct sock *sk)
1917 {
1918 if (refcount_dec_and_test(&sk->sk_refcnt))
1919 sk_free(sk);
1920 }
1921 /* Generic version of sock_put(), dealing with all sockets
1922 * (TCP_TIMEWAIT, TCP_NEW_SYN_RECV, ESTABLISHED...)
1923 */
1924 void sock_gen_put(struct sock *sk);
1925
1926 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested,
1927 unsigned int trim_cap, bool refcounted);
sk_receive_skb(struct sock * sk,struct sk_buff * skb,const int nested)1928 static inline int sk_receive_skb(struct sock *sk, struct sk_buff *skb,
1929 const int nested)
1930 {
1931 return __sk_receive_skb(sk, skb, nested, 1, true);
1932 }
1933
sk_tx_queue_set(struct sock * sk,int tx_queue)1934 static inline void sk_tx_queue_set(struct sock *sk, int tx_queue)
1935 {
1936 /* sk_tx_queue_mapping accept only upto a 16-bit value */
1937 if (WARN_ON_ONCE((unsigned short)tx_queue >= USHRT_MAX))
1938 return;
1939 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1940 * other WRITE_ONCE() because socket lock might be not held.
1941 */
1942 WRITE_ONCE(sk->sk_tx_queue_mapping, tx_queue);
1943 }
1944
1945 #define NO_QUEUE_MAPPING USHRT_MAX
1946
sk_tx_queue_clear(struct sock * sk)1947 static inline void sk_tx_queue_clear(struct sock *sk)
1948 {
1949 /* Paired with READ_ONCE() in sk_tx_queue_get() and
1950 * other WRITE_ONCE() because socket lock might be not held.
1951 */
1952 WRITE_ONCE(sk->sk_tx_queue_mapping, NO_QUEUE_MAPPING);
1953 }
1954
sk_tx_queue_get(const struct sock * sk)1955 static inline int sk_tx_queue_get(const struct sock *sk)
1956 {
1957 if (sk) {
1958 /* Paired with WRITE_ONCE() in sk_tx_queue_clear()
1959 * and sk_tx_queue_set().
1960 */
1961 int val = READ_ONCE(sk->sk_tx_queue_mapping);
1962
1963 if (val != NO_QUEUE_MAPPING)
1964 return val;
1965 }
1966 return -1;
1967 }
1968
sk_rx_queue_set(struct sock * sk,const struct sk_buff * skb)1969 static inline void sk_rx_queue_set(struct sock *sk, const struct sk_buff *skb)
1970 {
1971 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1972 if (skb_rx_queue_recorded(skb)) {
1973 u16 rx_queue = skb_get_rx_queue(skb);
1974
1975 if (WARN_ON_ONCE(rx_queue == NO_QUEUE_MAPPING))
1976 return;
1977
1978 sk->sk_rx_queue_mapping = rx_queue;
1979 }
1980 #endif
1981 }
1982
sk_rx_queue_clear(struct sock * sk)1983 static inline void sk_rx_queue_clear(struct sock *sk)
1984 {
1985 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1986 sk->sk_rx_queue_mapping = NO_QUEUE_MAPPING;
1987 #endif
1988 }
1989
sk_rx_queue_get(const struct sock * sk)1990 static inline int sk_rx_queue_get(const struct sock *sk)
1991 {
1992 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1993 if (sk && sk->sk_rx_queue_mapping != NO_QUEUE_MAPPING)
1994 return sk->sk_rx_queue_mapping;
1995 #endif
1996
1997 return -1;
1998 }
1999
sk_set_socket(struct sock * sk,struct socket * sock)2000 static inline void sk_set_socket(struct sock *sk, struct socket *sock)
2001 {
2002 sk->sk_socket = sock;
2003 }
2004
sk_sleep(struct sock * sk)2005 static inline wait_queue_head_t *sk_sleep(struct sock *sk)
2006 {
2007 BUILD_BUG_ON(offsetof(struct socket_wq, wait) != 0);
2008 return &rcu_dereference_raw(sk->sk_wq)->wait;
2009 }
2010 /* Detach socket from process context.
2011 * Announce socket dead, detach it from wait queue and inode.
2012 * Note that parent inode held reference count on this struct sock,
2013 * we do not release it in this function, because protocol
2014 * probably wants some additional cleanups or even continuing
2015 * to work with this socket (TCP).
2016 */
sock_orphan(struct sock * sk)2017 static inline void sock_orphan(struct sock *sk)
2018 {
2019 write_lock_bh(&sk->sk_callback_lock);
2020 sock_set_flag(sk, SOCK_DEAD);
2021 sk_set_socket(sk, NULL);
2022 sk->sk_wq = NULL;
2023 write_unlock_bh(&sk->sk_callback_lock);
2024 }
2025
sock_graft(struct sock * sk,struct socket * parent)2026 static inline void sock_graft(struct sock *sk, struct socket *parent)
2027 {
2028 WARN_ON(parent->sk);
2029 write_lock_bh(&sk->sk_callback_lock);
2030 rcu_assign_pointer(sk->sk_wq, &parent->wq);
2031 parent->sk = sk;
2032 sk_set_socket(sk, parent);
2033 sk->sk_uid = SOCK_INODE(parent)->i_uid;
2034 security_sock_graft(sk, parent);
2035 write_unlock_bh(&sk->sk_callback_lock);
2036 }
2037
2038 kuid_t sock_i_uid(struct sock *sk);
2039 unsigned long __sock_i_ino(struct sock *sk);
2040 unsigned long sock_i_ino(struct sock *sk);
2041
sock_net_uid(const struct net * net,const struct sock * sk)2042 static inline kuid_t sock_net_uid(const struct net *net, const struct sock *sk)
2043 {
2044 return sk ? sk->sk_uid : make_kuid(net->user_ns, 0);
2045 }
2046
net_tx_rndhash(void)2047 static inline u32 net_tx_rndhash(void)
2048 {
2049 u32 v = prandom_u32();
2050
2051 return v ?: 1;
2052 }
2053
sk_set_txhash(struct sock * sk)2054 static inline void sk_set_txhash(struct sock *sk)
2055 {
2056 /* This pairs with READ_ONCE() in skb_set_hash_from_sk() */
2057 WRITE_ONCE(sk->sk_txhash, net_tx_rndhash());
2058 }
2059
sk_rethink_txhash(struct sock * sk)2060 static inline bool sk_rethink_txhash(struct sock *sk)
2061 {
2062 if (sk->sk_txhash) {
2063 sk_set_txhash(sk);
2064 return true;
2065 }
2066 return false;
2067 }
2068
2069 static inline struct dst_entry *
__sk_dst_get(struct sock * sk)2070 __sk_dst_get(struct sock *sk)
2071 {
2072 return rcu_dereference_check(sk->sk_dst_cache,
2073 lockdep_sock_is_held(sk));
2074 }
2075
2076 static inline struct dst_entry *
sk_dst_get(struct sock * sk)2077 sk_dst_get(struct sock *sk)
2078 {
2079 struct dst_entry *dst;
2080
2081 rcu_read_lock();
2082 dst = rcu_dereference(sk->sk_dst_cache);
2083 if (dst && !atomic_inc_not_zero(&dst->__refcnt))
2084 dst = NULL;
2085 rcu_read_unlock();
2086 return dst;
2087 }
2088
__dst_negative_advice(struct sock * sk)2089 static inline void __dst_negative_advice(struct sock *sk)
2090 {
2091 struct dst_entry *ndst, *dst = __sk_dst_get(sk);
2092
2093 if (dst && dst->ops->negative_advice) {
2094 ndst = dst->ops->negative_advice(dst);
2095
2096 if (ndst != dst) {
2097 rcu_assign_pointer(sk->sk_dst_cache, ndst);
2098 sk_tx_queue_clear(sk);
2099 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2100 }
2101 }
2102 }
2103
dst_negative_advice(struct sock * sk)2104 static inline void dst_negative_advice(struct sock *sk)
2105 {
2106 sk_rethink_txhash(sk);
2107 __dst_negative_advice(sk);
2108 }
2109
2110 static inline void
__sk_dst_set(struct sock * sk,struct dst_entry * dst)2111 __sk_dst_set(struct sock *sk, struct dst_entry *dst)
2112 {
2113 struct dst_entry *old_dst;
2114
2115 sk_tx_queue_clear(sk);
2116 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2117 old_dst = rcu_dereference_protected(sk->sk_dst_cache,
2118 lockdep_sock_is_held(sk));
2119 rcu_assign_pointer(sk->sk_dst_cache, dst);
2120 dst_release(old_dst);
2121 }
2122
2123 static inline void
sk_dst_set(struct sock * sk,struct dst_entry * dst)2124 sk_dst_set(struct sock *sk, struct dst_entry *dst)
2125 {
2126 struct dst_entry *old_dst;
2127
2128 sk_tx_queue_clear(sk);
2129 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2130 old_dst = xchg((__force struct dst_entry **)&sk->sk_dst_cache, dst);
2131 dst_release(old_dst);
2132 }
2133
2134 static inline void
__sk_dst_reset(struct sock * sk)2135 __sk_dst_reset(struct sock *sk)
2136 {
2137 __sk_dst_set(sk, NULL);
2138 }
2139
2140 static inline void
sk_dst_reset(struct sock * sk)2141 sk_dst_reset(struct sock *sk)
2142 {
2143 sk_dst_set(sk, NULL);
2144 }
2145
2146 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie);
2147
2148 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie);
2149
sk_dst_confirm(struct sock * sk)2150 static inline void sk_dst_confirm(struct sock *sk)
2151 {
2152 if (!READ_ONCE(sk->sk_dst_pending_confirm))
2153 WRITE_ONCE(sk->sk_dst_pending_confirm, 1);
2154 }
2155
sock_confirm_neigh(struct sk_buff * skb,struct neighbour * n)2156 static inline void sock_confirm_neigh(struct sk_buff *skb, struct neighbour *n)
2157 {
2158 if (skb_get_dst_pending_confirm(skb)) {
2159 struct sock *sk = skb->sk;
2160 unsigned long now = jiffies;
2161
2162 /* avoid dirtying neighbour */
2163 if (READ_ONCE(n->confirmed) != now)
2164 WRITE_ONCE(n->confirmed, now);
2165 if (sk && READ_ONCE(sk->sk_dst_pending_confirm))
2166 WRITE_ONCE(sk->sk_dst_pending_confirm, 0);
2167 }
2168 }
2169
2170 bool sk_mc_loop(struct sock *sk);
2171
sk_can_gso(const struct sock * sk)2172 static inline bool sk_can_gso(const struct sock *sk)
2173 {
2174 return net_gso_ok(sk->sk_route_caps, sk->sk_gso_type);
2175 }
2176
2177 void sk_setup_caps(struct sock *sk, struct dst_entry *dst);
2178
sk_nocaps_add(struct sock * sk,netdev_features_t flags)2179 static inline void sk_nocaps_add(struct sock *sk, netdev_features_t flags)
2180 {
2181 sk->sk_route_nocaps |= flags;
2182 sk->sk_route_caps &= ~flags;
2183 }
2184
skb_do_copy_data_nocache(struct sock * sk,struct sk_buff * skb,struct iov_iter * from,char * to,int copy,int offset)2185 static inline int skb_do_copy_data_nocache(struct sock *sk, struct sk_buff *skb,
2186 struct iov_iter *from, char *to,
2187 int copy, int offset)
2188 {
2189 if (skb->ip_summed == CHECKSUM_NONE) {
2190 __wsum csum = 0;
2191 if (!csum_and_copy_from_iter_full(to, copy, &csum, from))
2192 return -EFAULT;
2193 skb->csum = csum_block_add(skb->csum, csum, offset);
2194 } else if (sk->sk_route_caps & NETIF_F_NOCACHE_COPY) {
2195 if (!copy_from_iter_full_nocache(to, copy, from))
2196 return -EFAULT;
2197 } else if (!copy_from_iter_full(to, copy, from))
2198 return -EFAULT;
2199
2200 return 0;
2201 }
2202
skb_add_data_nocache(struct sock * sk,struct sk_buff * skb,struct iov_iter * from,int copy)2203 static inline int skb_add_data_nocache(struct sock *sk, struct sk_buff *skb,
2204 struct iov_iter *from, int copy)
2205 {
2206 int err, offset = skb->len;
2207
2208 err = skb_do_copy_data_nocache(sk, skb, from, skb_put(skb, copy),
2209 copy, offset);
2210 if (err)
2211 __skb_trim(skb, offset);
2212
2213 return err;
2214 }
2215
skb_copy_to_page_nocache(struct sock * sk,struct iov_iter * from,struct sk_buff * skb,struct page * page,int off,int copy)2216 static inline int skb_copy_to_page_nocache(struct sock *sk, struct iov_iter *from,
2217 struct sk_buff *skb,
2218 struct page *page,
2219 int off, int copy)
2220 {
2221 int err;
2222
2223 err = skb_do_copy_data_nocache(sk, skb, from, page_address(page) + off,
2224 copy, skb->len);
2225 if (err)
2226 return err;
2227
2228 skb->len += copy;
2229 skb->data_len += copy;
2230 skb->truesize += copy;
2231 sk_wmem_queued_add(sk, copy);
2232 sk_mem_charge(sk, copy);
2233 return 0;
2234 }
2235
2236 /**
2237 * sk_wmem_alloc_get - returns write allocations
2238 * @sk: socket
2239 *
2240 * Return: sk_wmem_alloc minus initial offset of one
2241 */
sk_wmem_alloc_get(const struct sock * sk)2242 static inline int sk_wmem_alloc_get(const struct sock *sk)
2243 {
2244 return refcount_read(&sk->sk_wmem_alloc) - 1;
2245 }
2246
2247 /**
2248 * sk_rmem_alloc_get - returns read allocations
2249 * @sk: socket
2250 *
2251 * Return: sk_rmem_alloc
2252 */
sk_rmem_alloc_get(const struct sock * sk)2253 static inline int sk_rmem_alloc_get(const struct sock *sk)
2254 {
2255 return atomic_read(&sk->sk_rmem_alloc);
2256 }
2257
2258 /**
2259 * sk_has_allocations - check if allocations are outstanding
2260 * @sk: socket
2261 *
2262 * Return: true if socket has write or read allocations
2263 */
sk_has_allocations(const struct sock * sk)2264 static inline bool sk_has_allocations(const struct sock *sk)
2265 {
2266 return sk_wmem_alloc_get(sk) || sk_rmem_alloc_get(sk);
2267 }
2268
2269 /**
2270 * skwq_has_sleeper - check if there are any waiting processes
2271 * @wq: struct socket_wq
2272 *
2273 * Return: true if socket_wq has waiting processes
2274 *
2275 * The purpose of the skwq_has_sleeper and sock_poll_wait is to wrap the memory
2276 * barrier call. They were added due to the race found within the tcp code.
2277 *
2278 * Consider following tcp code paths::
2279 *
2280 * CPU1 CPU2
2281 * sys_select receive packet
2282 * ... ...
2283 * __add_wait_queue update tp->rcv_nxt
2284 * ... ...
2285 * tp->rcv_nxt check sock_def_readable
2286 * ... {
2287 * schedule rcu_read_lock();
2288 * wq = rcu_dereference(sk->sk_wq);
2289 * if (wq && waitqueue_active(&wq->wait))
2290 * wake_up_interruptible(&wq->wait)
2291 * ...
2292 * }
2293 *
2294 * The race for tcp fires when the __add_wait_queue changes done by CPU1 stay
2295 * in its cache, and so does the tp->rcv_nxt update on CPU2 side. The CPU1
2296 * could then endup calling schedule and sleep forever if there are no more
2297 * data on the socket.
2298 *
2299 */
skwq_has_sleeper(struct socket_wq * wq)2300 static inline bool skwq_has_sleeper(struct socket_wq *wq)
2301 {
2302 return wq && wq_has_sleeper(&wq->wait);
2303 }
2304
2305 /**
2306 * sock_poll_wait - place memory barrier behind the poll_wait call.
2307 * @filp: file
2308 * @sock: socket to wait on
2309 * @p: poll_table
2310 *
2311 * See the comments in the wq_has_sleeper function.
2312 */
sock_poll_wait(struct file * filp,struct socket * sock,poll_table * p)2313 static inline void sock_poll_wait(struct file *filp, struct socket *sock,
2314 poll_table *p)
2315 {
2316 if (!poll_does_not_wait(p)) {
2317 poll_wait(filp, &sock->wq.wait, p);
2318 /* We need to be sure we are in sync with the
2319 * socket flags modification.
2320 *
2321 * This memory barrier is paired in the wq_has_sleeper.
2322 */
2323 smp_mb();
2324 }
2325 }
2326
skb_set_hash_from_sk(struct sk_buff * skb,struct sock * sk)2327 static inline void skb_set_hash_from_sk(struct sk_buff *skb, struct sock *sk)
2328 {
2329 /* This pairs with WRITE_ONCE() in sk_set_txhash() */
2330 u32 txhash = READ_ONCE(sk->sk_txhash);
2331
2332 if (txhash) {
2333 skb->l4_hash = 1;
2334 skb->hash = txhash;
2335 }
2336 }
2337
2338 void skb_set_owner_w(struct sk_buff *skb, struct sock *sk);
2339
2340 /*
2341 * Queue a received datagram if it will fit. Stream and sequenced
2342 * protocols can't normally use this as they need to fit buffers in
2343 * and play with them.
2344 *
2345 * Inlined as it's very short and called for pretty much every
2346 * packet ever received.
2347 */
skb_set_owner_r(struct sk_buff * skb,struct sock * sk)2348 static inline void skb_set_owner_r(struct sk_buff *skb, struct sock *sk)
2349 {
2350 skb_orphan(skb);
2351 skb->sk = sk;
2352 skb->destructor = sock_rfree;
2353 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
2354 sk_mem_charge(sk, skb->truesize);
2355 }
2356
skb_set_owner_sk_safe(struct sk_buff * skb,struct sock * sk)2357 static inline __must_check bool skb_set_owner_sk_safe(struct sk_buff *skb, struct sock *sk)
2358 {
2359 if (sk && refcount_inc_not_zero(&sk->sk_refcnt)) {
2360 skb_orphan(skb);
2361 skb->destructor = sock_efree;
2362 skb->sk = sk;
2363 return true;
2364 }
2365 return false;
2366 }
2367
skb_clone_and_charge_r(struct sk_buff * skb,struct sock * sk)2368 static inline struct sk_buff *skb_clone_and_charge_r(struct sk_buff *skb, struct sock *sk)
2369 {
2370 skb = skb_clone(skb, sk_gfp_mask(sk, GFP_ATOMIC));
2371 if (skb) {
2372 if (sk_rmem_schedule(sk, skb, skb->truesize)) {
2373 skb_set_owner_r(skb, sk);
2374 return skb;
2375 }
2376 __kfree_skb(skb);
2377 }
2378 return NULL;
2379 }
2380
skb_prepare_for_gro(struct sk_buff * skb)2381 static inline void skb_prepare_for_gro(struct sk_buff *skb)
2382 {
2383 if (skb->destructor != sock_wfree) {
2384 skb_orphan(skb);
2385 return;
2386 }
2387 skb->slow_gro = 1;
2388 }
2389
2390 void sk_reset_timer(struct sock *sk, struct timer_list *timer,
2391 unsigned long expires);
2392
2393 void sk_stop_timer(struct sock *sk, struct timer_list *timer);
2394
2395 void sk_stop_timer_sync(struct sock *sk, struct timer_list *timer);
2396
2397 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
2398 struct sk_buff *skb, unsigned int flags,
2399 void (*destructor)(struct sock *sk,
2400 struct sk_buff *skb));
2401 int __sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2402 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb);
2403
2404 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb);
2405 struct sk_buff *sock_dequeue_err_skb(struct sock *sk);
2406
2407 /*
2408 * Recover an error report and clear atomically
2409 */
2410
sock_error(struct sock * sk)2411 static inline int sock_error(struct sock *sk)
2412 {
2413 int err;
2414
2415 /* Avoid an atomic operation for the common case.
2416 * This is racy since another cpu/thread can change sk_err under us.
2417 */
2418 if (likely(data_race(!sk->sk_err)))
2419 return 0;
2420
2421 err = xchg(&sk->sk_err, 0);
2422 return -err;
2423 }
2424
2425 void sk_error_report(struct sock *sk);
2426
sock_wspace(struct sock * sk)2427 static inline unsigned long sock_wspace(struct sock *sk)
2428 {
2429 int amt = 0;
2430
2431 if (!(sk->sk_shutdown & SEND_SHUTDOWN)) {
2432 amt = sk->sk_sndbuf - refcount_read(&sk->sk_wmem_alloc);
2433 if (amt < 0)
2434 amt = 0;
2435 }
2436 return amt;
2437 }
2438
2439 /* Note:
2440 * We use sk->sk_wq_raw, from contexts knowing this
2441 * pointer is not NULL and cannot disappear/change.
2442 */
sk_set_bit(int nr,struct sock * sk)2443 static inline void sk_set_bit(int nr, struct sock *sk)
2444 {
2445 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2446 !sock_flag(sk, SOCK_FASYNC))
2447 return;
2448
2449 set_bit(nr, &sk->sk_wq_raw->flags);
2450 }
2451
sk_clear_bit(int nr,struct sock * sk)2452 static inline void sk_clear_bit(int nr, struct sock *sk)
2453 {
2454 if ((nr == SOCKWQ_ASYNC_NOSPACE || nr == SOCKWQ_ASYNC_WAITDATA) &&
2455 !sock_flag(sk, SOCK_FASYNC))
2456 return;
2457
2458 clear_bit(nr, &sk->sk_wq_raw->flags);
2459 }
2460
sk_wake_async(const struct sock * sk,int how,int band)2461 static inline void sk_wake_async(const struct sock *sk, int how, int band)
2462 {
2463 if (sock_flag(sk, SOCK_FASYNC)) {
2464 rcu_read_lock();
2465 sock_wake_async(rcu_dereference(sk->sk_wq), how, band);
2466 rcu_read_unlock();
2467 }
2468 }
2469
2470 /* Since sk_{r,w}mem_alloc sums skb->truesize, even a small frame might
2471 * need sizeof(sk_buff) + MTU + padding, unless net driver perform copybreak.
2472 * Note: for send buffers, TCP works better if we can build two skbs at
2473 * minimum.
2474 */
2475 #define TCP_SKB_MIN_TRUESIZE (2048 + SKB_DATA_ALIGN(sizeof(struct sk_buff)))
2476
2477 #define SOCK_MIN_SNDBUF (TCP_SKB_MIN_TRUESIZE * 2)
2478 #define SOCK_MIN_RCVBUF TCP_SKB_MIN_TRUESIZE
2479
sk_stream_moderate_sndbuf(struct sock * sk)2480 static inline void sk_stream_moderate_sndbuf(struct sock *sk)
2481 {
2482 u32 val;
2483
2484 if (sk->sk_userlocks & SOCK_SNDBUF_LOCK)
2485 return;
2486
2487 val = min(sk->sk_sndbuf, sk->sk_wmem_queued >> 1);
2488
2489 WRITE_ONCE(sk->sk_sndbuf, max_t(u32, val, SOCK_MIN_SNDBUF));
2490 }
2491
2492 struct sk_buff *sk_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
2493 bool force_schedule);
2494
2495 /**
2496 * sk_page_frag - return an appropriate page_frag
2497 * @sk: socket
2498 *
2499 * Use the per task page_frag instead of the per socket one for
2500 * optimization when we know that we're in process context and own
2501 * everything that's associated with %current.
2502 *
2503 * Both direct reclaim and page faults can nest inside other
2504 * socket operations and end up recursing into sk_page_frag()
2505 * while it's already in use: explicitly avoid task page_frag
2506 * usage if the caller is potentially doing any of them.
2507 * This assumes that page fault handlers use the GFP_NOFS flags.
2508 *
2509 * Return: a per task page_frag if context allows that,
2510 * otherwise a per socket one.
2511 */
sk_page_frag(struct sock * sk)2512 static inline struct page_frag *sk_page_frag(struct sock *sk)
2513 {
2514 if ((sk->sk_allocation & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC | __GFP_FS)) ==
2515 (__GFP_DIRECT_RECLAIM | __GFP_FS))
2516 return ¤t->task_frag;
2517
2518 return &sk->sk_frag;
2519 }
2520
2521 bool sk_page_frag_refill(struct sock *sk, struct page_frag *pfrag);
2522
2523 /*
2524 * Default write policy as shown to user space via poll/select/SIGIO
2525 */
sock_writeable(const struct sock * sk)2526 static inline bool sock_writeable(const struct sock *sk)
2527 {
2528 return refcount_read(&sk->sk_wmem_alloc) < (READ_ONCE(sk->sk_sndbuf) >> 1);
2529 }
2530
gfp_any(void)2531 static inline gfp_t gfp_any(void)
2532 {
2533 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2534 }
2535
gfp_memcg_charge(void)2536 static inline gfp_t gfp_memcg_charge(void)
2537 {
2538 return in_softirq() ? GFP_ATOMIC : GFP_KERNEL;
2539 }
2540
sock_rcvtimeo(const struct sock * sk,bool noblock)2541 static inline long sock_rcvtimeo(const struct sock *sk, bool noblock)
2542 {
2543 return noblock ? 0 : sk->sk_rcvtimeo;
2544 }
2545
sock_sndtimeo(const struct sock * sk,bool noblock)2546 static inline long sock_sndtimeo(const struct sock *sk, bool noblock)
2547 {
2548 return noblock ? 0 : sk->sk_sndtimeo;
2549 }
2550
sock_rcvlowat(const struct sock * sk,int waitall,int len)2551 static inline int sock_rcvlowat(const struct sock *sk, int waitall, int len)
2552 {
2553 int v = waitall ? len : min_t(int, READ_ONCE(sk->sk_rcvlowat), len);
2554
2555 return v ?: 1;
2556 }
2557
2558 /* Alas, with timeout socket operations are not restartable.
2559 * Compare this to poll().
2560 */
sock_intr_errno(long timeo)2561 static inline int sock_intr_errno(long timeo)
2562 {
2563 return timeo == MAX_SCHEDULE_TIMEOUT ? -ERESTARTSYS : -EINTR;
2564 }
2565
2566 struct sock_skb_cb {
2567 u32 dropcount;
2568 };
2569
2570 /* Store sock_skb_cb at the end of skb->cb[] so protocol families
2571 * using skb->cb[] would keep using it directly and utilize its
2572 * alignement guarantee.
2573 */
2574 #define SOCK_SKB_CB_OFFSET ((sizeof_field(struct sk_buff, cb) - \
2575 sizeof(struct sock_skb_cb)))
2576
2577 #define SOCK_SKB_CB(__skb) ((struct sock_skb_cb *)((__skb)->cb + \
2578 SOCK_SKB_CB_OFFSET))
2579
2580 #define sock_skb_cb_check_size(size) \
2581 BUILD_BUG_ON((size) > SOCK_SKB_CB_OFFSET)
2582
2583 static inline void
sock_skb_set_dropcount(const struct sock * sk,struct sk_buff * skb)2584 sock_skb_set_dropcount(const struct sock *sk, struct sk_buff *skb)
2585 {
2586 SOCK_SKB_CB(skb)->dropcount = sock_flag(sk, SOCK_RXQ_OVFL) ?
2587 atomic_read(&sk->sk_drops) : 0;
2588 }
2589
sk_drops_add(struct sock * sk,const struct sk_buff * skb)2590 static inline void sk_drops_add(struct sock *sk, const struct sk_buff *skb)
2591 {
2592 int segs = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2593
2594 atomic_add(segs, &sk->sk_drops);
2595 }
2596
sock_read_timestamp(struct sock * sk)2597 static inline ktime_t sock_read_timestamp(struct sock *sk)
2598 {
2599 #if BITS_PER_LONG==32
2600 unsigned int seq;
2601 ktime_t kt;
2602
2603 do {
2604 seq = read_seqbegin(&sk->sk_stamp_seq);
2605 kt = sk->sk_stamp;
2606 } while (read_seqretry(&sk->sk_stamp_seq, seq));
2607
2608 return kt;
2609 #else
2610 return READ_ONCE(sk->sk_stamp);
2611 #endif
2612 }
2613
sock_write_timestamp(struct sock * sk,ktime_t kt)2614 static inline void sock_write_timestamp(struct sock *sk, ktime_t kt)
2615 {
2616 #if BITS_PER_LONG==32
2617 write_seqlock(&sk->sk_stamp_seq);
2618 sk->sk_stamp = kt;
2619 write_sequnlock(&sk->sk_stamp_seq);
2620 #else
2621 WRITE_ONCE(sk->sk_stamp, kt);
2622 #endif
2623 }
2624
2625 void __sock_recv_timestamp(struct msghdr *msg, struct sock *sk,
2626 struct sk_buff *skb);
2627 void __sock_recv_wifi_status(struct msghdr *msg, struct sock *sk,
2628 struct sk_buff *skb);
2629
2630 static inline void
sock_recv_timestamp(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2631 sock_recv_timestamp(struct msghdr *msg, struct sock *sk, struct sk_buff *skb)
2632 {
2633 ktime_t kt = skb->tstamp;
2634 struct skb_shared_hwtstamps *hwtstamps = skb_hwtstamps(skb);
2635
2636 /*
2637 * generate control messages if
2638 * - receive time stamping in software requested
2639 * - software time stamp available and wanted
2640 * - hardware time stamps available and wanted
2641 */
2642 if (sock_flag(sk, SOCK_RCVTSTAMP) ||
2643 (sk->sk_tsflags & SOF_TIMESTAMPING_RX_SOFTWARE) ||
2644 (kt && sk->sk_tsflags & SOF_TIMESTAMPING_SOFTWARE) ||
2645 (hwtstamps->hwtstamp &&
2646 (sk->sk_tsflags & SOF_TIMESTAMPING_RAW_HARDWARE)))
2647 __sock_recv_timestamp(msg, sk, skb);
2648 else
2649 sock_write_timestamp(sk, kt);
2650
2651 if (sock_flag(sk, SOCK_WIFI_STATUS) && skb->wifi_acked_valid)
2652 __sock_recv_wifi_status(msg, sk, skb);
2653 }
2654
2655 void __sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2656 struct sk_buff *skb);
2657
2658 #define SK_DEFAULT_STAMP (-1L * NSEC_PER_SEC)
sock_recv_cmsgs(struct msghdr * msg,struct sock * sk,struct sk_buff * skb)2659 static inline void sock_recv_cmsgs(struct msghdr *msg, struct sock *sk,
2660 struct sk_buff *skb)
2661 {
2662 #define FLAGS_RECV_CMSGS ((1UL << SOCK_RXQ_OVFL) | \
2663 (1UL << SOCK_RCVTSTAMP) | \
2664 (1UL << SOCK_RCVMARK))
2665 #define TSFLAGS_ANY (SOF_TIMESTAMPING_SOFTWARE | \
2666 SOF_TIMESTAMPING_RAW_HARDWARE)
2667
2668 if (sk->sk_flags & FLAGS_RECV_CMSGS || sk->sk_tsflags & TSFLAGS_ANY)
2669 __sock_recv_cmsgs(msg, sk, skb);
2670 else if (unlikely(sock_flag(sk, SOCK_TIMESTAMP)))
2671 sock_write_timestamp(sk, skb->tstamp);
2672 else if (unlikely(sock_read_timestamp(sk) == SK_DEFAULT_STAMP))
2673 sock_write_timestamp(sk, 0);
2674 }
2675
2676 void __sock_tx_timestamp(__u16 tsflags, __u8 *tx_flags);
2677
2678 /**
2679 * _sock_tx_timestamp - checks whether the outgoing packet is to be time stamped
2680 * @sk: socket sending this packet
2681 * @tsflags: timestamping flags to use
2682 * @tx_flags: completed with instructions for time stamping
2683 * @tskey: filled in with next sk_tskey (not for TCP, which uses seqno)
2684 *
2685 * Note: callers should take care of initial ``*tx_flags`` value (usually 0)
2686 */
_sock_tx_timestamp(struct sock * sk,__u16 tsflags,__u8 * tx_flags,__u32 * tskey)2687 static inline void _sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2688 __u8 *tx_flags, __u32 *tskey)
2689 {
2690 if (unlikely(tsflags)) {
2691 __sock_tx_timestamp(tsflags, tx_flags);
2692 if (tsflags & SOF_TIMESTAMPING_OPT_ID && tskey &&
2693 tsflags & SOF_TIMESTAMPING_TX_RECORD_MASK)
2694 *tskey = atomic_inc_return(&sk->sk_tskey) - 1;
2695 }
2696 if (unlikely(sock_flag(sk, SOCK_WIFI_STATUS)))
2697 *tx_flags |= SKBTX_WIFI_STATUS;
2698 }
2699
sock_tx_timestamp(struct sock * sk,__u16 tsflags,__u8 * tx_flags)2700 static inline void sock_tx_timestamp(struct sock *sk, __u16 tsflags,
2701 __u8 *tx_flags)
2702 {
2703 _sock_tx_timestamp(sk, tsflags, tx_flags, NULL);
2704 }
2705
skb_setup_tx_timestamp(struct sk_buff * skb,__u16 tsflags)2706 static inline void skb_setup_tx_timestamp(struct sk_buff *skb, __u16 tsflags)
2707 {
2708 _sock_tx_timestamp(skb->sk, tsflags, &skb_shinfo(skb)->tx_flags,
2709 &skb_shinfo(skb)->tskey);
2710 }
2711
2712 DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
2713 /**
2714 * sk_eat_skb - Release a skb if it is no longer needed
2715 * @sk: socket to eat this skb from
2716 * @skb: socket buffer to eat
2717 *
2718 * This routine must be called with interrupts disabled or with the socket
2719 * locked so that the sk_buff queue operation is ok.
2720 */
sk_eat_skb(struct sock * sk,struct sk_buff * skb)2721 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
2722 {
2723 __skb_unlink(skb, &sk->sk_receive_queue);
2724 if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
2725 !sk->sk_rx_skb_cache) {
2726 sk->sk_rx_skb_cache = skb;
2727 skb_orphan(skb);
2728 return;
2729 }
2730 __kfree_skb(skb);
2731 }
2732
2733 static inline
sock_net(const struct sock * sk)2734 struct net *sock_net(const struct sock *sk)
2735 {
2736 return read_pnet(&sk->sk_net);
2737 }
2738
2739 static inline
sock_net_set(struct sock * sk,struct net * net)2740 void sock_net_set(struct sock *sk, struct net *net)
2741 {
2742 write_pnet(&sk->sk_net, net);
2743 }
2744
2745 static inline bool
skb_sk_is_prefetched(struct sk_buff * skb)2746 skb_sk_is_prefetched(struct sk_buff *skb)
2747 {
2748 #ifdef CONFIG_INET
2749 return skb->destructor == sock_pfree;
2750 #else
2751 return false;
2752 #endif /* CONFIG_INET */
2753 }
2754
2755 /* This helper checks if a socket is a full socket,
2756 * ie _not_ a timewait or request socket.
2757 */
sk_fullsock(const struct sock * sk)2758 static inline bool sk_fullsock(const struct sock *sk)
2759 {
2760 return (1 << sk->sk_state) & ~(TCPF_TIME_WAIT | TCPF_NEW_SYN_RECV);
2761 }
2762
2763 static inline bool
sk_is_refcounted(struct sock * sk)2764 sk_is_refcounted(struct sock *sk)
2765 {
2766 /* Only full sockets have sk->sk_flags. */
2767 return !sk_fullsock(sk) || !sock_flag(sk, SOCK_RCU_FREE);
2768 }
2769
2770 /**
2771 * skb_steal_sock - steal a socket from an sk_buff
2772 * @skb: sk_buff to steal the socket from
2773 * @refcounted: is set to true if the socket is reference-counted
2774 */
2775 static inline struct sock *
skb_steal_sock(struct sk_buff * skb,bool * refcounted)2776 skb_steal_sock(struct sk_buff *skb, bool *refcounted)
2777 {
2778 if (skb->sk) {
2779 struct sock *sk = skb->sk;
2780
2781 *refcounted = true;
2782 if (skb_sk_is_prefetched(skb))
2783 *refcounted = sk_is_refcounted(sk);
2784 skb->destructor = NULL;
2785 skb->sk = NULL;
2786 return sk;
2787 }
2788 *refcounted = false;
2789 return NULL;
2790 }
2791
2792 /* Checks if this SKB belongs to an HW offloaded socket
2793 * and whether any SW fallbacks are required based on dev.
2794 * Check decrypted mark in case skb_orphan() cleared socket.
2795 */
sk_validate_xmit_skb(struct sk_buff * skb,struct net_device * dev)2796 static inline struct sk_buff *sk_validate_xmit_skb(struct sk_buff *skb,
2797 struct net_device *dev)
2798 {
2799 #ifdef CONFIG_SOCK_VALIDATE_XMIT
2800 struct sock *sk = skb->sk;
2801
2802 if (sk && sk_fullsock(sk) && sk->sk_validate_xmit_skb) {
2803 skb = sk->sk_validate_xmit_skb(sk, dev, skb);
2804 #ifdef CONFIG_TLS_DEVICE
2805 } else if (unlikely(skb->decrypted)) {
2806 pr_warn_ratelimited("unencrypted skb with no associated socket - dropping\n");
2807 kfree_skb(skb);
2808 skb = NULL;
2809 #endif
2810 }
2811 #endif
2812
2813 return skb;
2814 }
2815
2816 /* This helper checks if a socket is a LISTEN or NEW_SYN_RECV
2817 * SYNACK messages can be attached to either ones (depending on SYNCOOKIE)
2818 */
sk_listener(const struct sock * sk)2819 static inline bool sk_listener(const struct sock *sk)
2820 {
2821 return (1 << sk->sk_state) & (TCPF_LISTEN | TCPF_NEW_SYN_RECV);
2822 }
2823
2824 void sock_enable_timestamp(struct sock *sk, enum sock_flags flag);
2825 int sock_recv_errqueue(struct sock *sk, struct msghdr *msg, int len, int level,
2826 int type);
2827
2828 bool sk_ns_capable(const struct sock *sk,
2829 struct user_namespace *user_ns, int cap);
2830 bool sk_capable(const struct sock *sk, int cap);
2831 bool sk_net_capable(const struct sock *sk, int cap);
2832
2833 void sk_get_meminfo(const struct sock *sk, u32 *meminfo);
2834
2835 /* Take into consideration the size of the struct sk_buff overhead in the
2836 * determination of these values, since that is non-constant across
2837 * platforms. This makes socket queueing behavior and performance
2838 * not depend upon such differences.
2839 */
2840 #define _SK_MEM_PACKETS 256
2841 #define _SK_MEM_OVERHEAD SKB_TRUESIZE(256)
2842 #define SK_WMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2843 #define SK_RMEM_MAX (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
2844
2845 extern __u32 sysctl_wmem_max;
2846 extern __u32 sysctl_rmem_max;
2847
2848 extern int sysctl_tstamp_allow_data;
2849 extern int sysctl_optmem_max;
2850
2851 extern __u32 sysctl_wmem_default;
2852 extern __u32 sysctl_rmem_default;
2853
2854 #define SKB_FRAG_PAGE_ORDER get_order(32768)
2855 DECLARE_STATIC_KEY_FALSE(net_high_order_alloc_disable_key);
2856
sk_get_wmem0(const struct sock * sk,const struct proto * proto)2857 static inline int sk_get_wmem0(const struct sock *sk, const struct proto *proto)
2858 {
2859 /* Does this proto have per netns sysctl_wmem ? */
2860 if (proto->sysctl_wmem_offset)
2861 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_wmem_offset));
2862
2863 return READ_ONCE(*proto->sysctl_wmem);
2864 }
2865
sk_get_rmem0(const struct sock * sk,const struct proto * proto)2866 static inline int sk_get_rmem0(const struct sock *sk, const struct proto *proto)
2867 {
2868 /* Does this proto have per netns sysctl_rmem ? */
2869 if (proto->sysctl_rmem_offset)
2870 return READ_ONCE(*(int *)((void *)sock_net(sk) + proto->sysctl_rmem_offset));
2871
2872 return READ_ONCE(*proto->sysctl_rmem);
2873 }
2874
2875 /* Default TCP Small queue budget is ~1 ms of data (1sec >> 10)
2876 * Some wifi drivers need to tweak it to get more chunks.
2877 * They can use this helper from their ndo_start_xmit()
2878 */
sk_pacing_shift_update(struct sock * sk,int val)2879 static inline void sk_pacing_shift_update(struct sock *sk, int val)
2880 {
2881 if (!sk || !sk_fullsock(sk) || READ_ONCE(sk->sk_pacing_shift) == val)
2882 return;
2883 WRITE_ONCE(sk->sk_pacing_shift, val);
2884 }
2885
2886 /* if a socket is bound to a device, check that the given device
2887 * index is either the same or that the socket is bound to an L3
2888 * master device and the given device index is also enslaved to
2889 * that L3 master
2890 */
sk_dev_equal_l3scope(struct sock * sk,int dif)2891 static inline bool sk_dev_equal_l3scope(struct sock *sk, int dif)
2892 {
2893 int mdif;
2894
2895 if (!sk->sk_bound_dev_if || sk->sk_bound_dev_if == dif)
2896 return true;
2897
2898 mdif = l3mdev_master_ifindex_by_index(sock_net(sk), dif);
2899 if (mdif && mdif == sk->sk_bound_dev_if)
2900 return true;
2901
2902 return false;
2903 }
2904
2905 void sock_def_readable(struct sock *sk);
2906
2907 int sock_bindtoindex(struct sock *sk, int ifindex, bool lock_sk);
2908 void sock_set_timestamp(struct sock *sk, int optname, bool valbool);
2909 int sock_set_timestamping(struct sock *sk, int optname,
2910 struct so_timestamping timestamping);
2911
2912 void sock_enable_timestamps(struct sock *sk);
2913 void sock_no_linger(struct sock *sk);
2914 void sock_set_keepalive(struct sock *sk);
2915 void sock_set_priority(struct sock *sk, u32 priority);
2916 void sock_set_rcvbuf(struct sock *sk, int val);
2917 void sock_set_mark(struct sock *sk, u32 val);
2918 void sock_set_reuseaddr(struct sock *sk);
2919 void sock_set_reuseport(struct sock *sk);
2920 void sock_set_sndtimeo(struct sock *sk, s64 secs);
2921
2922 int sock_bind_add(struct sock *sk, struct sockaddr *addr, int addr_len);
2923
sk_is_readable(struct sock * sk)2924 static inline bool sk_is_readable(struct sock *sk)
2925 {
2926 if (sk->sk_prot->sock_is_readable)
2927 return sk->sk_prot->sock_is_readable(sk);
2928 return false;
2929 }
2930 #endif /* _SOCK_H */
2931